diff --git a/parse/train/7Yhok3vJpU/images/03ece730eef5feb9b8dbfb55c1bbe5689e720b027f75544f852b23e525f594a7.jpg b/parse/train/7Yhok3vJpU/images/03ece730eef5feb9b8dbfb55c1bbe5689e720b027f75544f852b23e525f594a7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1d25b87326f46e42f820fe27e8d34c8078265f21 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/03ece730eef5feb9b8dbfb55c1bbe5689e720b027f75544f852b23e525f594a7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bde4b7a7ca43edfbddec838b8eb2685d2ca04e3cca1d0176b053af650b7574af +size 16961 diff --git a/parse/train/7Yhok3vJpU/images/10339a658a49a0924ac45107f28f41ecfa87419a92e24bf15789f9a88b3a945e.jpg b/parse/train/7Yhok3vJpU/images/10339a658a49a0924ac45107f28f41ecfa87419a92e24bf15789f9a88b3a945e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cfcc8eb276d31e708aa52c878ef8cc57486ff406 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/10339a658a49a0924ac45107f28f41ecfa87419a92e24bf15789f9a88b3a945e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:041d121a3e4614ae97a873da76325669da887f25d2ef60e837ada0748bfcc81d +size 17726 diff --git a/parse/train/7Yhok3vJpU/images/1547d31d312b4846653cad9c1ed96718f659643a6c01358ed503036e8a724ea8.jpg b/parse/train/7Yhok3vJpU/images/1547d31d312b4846653cad9c1ed96718f659643a6c01358ed503036e8a724ea8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3857c78e9c5eba9dbf8b93b7614c7e6a02f622f3 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/1547d31d312b4846653cad9c1ed96718f659643a6c01358ed503036e8a724ea8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2751e0f35ec66954148c346a34fb06e81ab1369dbcffa56960afb38b88a7f87e +size 2779 diff --git a/parse/train/7Yhok3vJpU/images/1b58e85c1b453f87737e889cd7142fed7390780fc0f0c1756bd6f62f3560126b.jpg b/parse/train/7Yhok3vJpU/images/1b58e85c1b453f87737e889cd7142fed7390780fc0f0c1756bd6f62f3560126b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..35b6c4126f1946f007f53fa00695e6bff18b12af --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/1b58e85c1b453f87737e889cd7142fed7390780fc0f0c1756bd6f62f3560126b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d2cac76ac2b06287a13c9a936bab04dc6bc19bcf13f4ba70fff0b4e724a998eb +size 77051 diff --git a/parse/train/7Yhok3vJpU/images/1b862724a1351b53818185d4f52db5796e72b3c622f507dca402c28189dfc6a7.jpg b/parse/train/7Yhok3vJpU/images/1b862724a1351b53818185d4f52db5796e72b3c622f507dca402c28189dfc6a7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..83c7db44c3405829dcce951e0bf05af4005b4e0c --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/1b862724a1351b53818185d4f52db5796e72b3c622f507dca402c28189dfc6a7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:28cc195873eb262e0b82d35fd572f0591ec3c43cc35622035f598ba4073b6a6f +size 64306 diff --git a/parse/train/7Yhok3vJpU/images/1ec4995fc9dbae2463a0ef5c51f676e86d31860224bce44cd7c84c9422234019.jpg b/parse/train/7Yhok3vJpU/images/1ec4995fc9dbae2463a0ef5c51f676e86d31860224bce44cd7c84c9422234019.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8dafddd39734b4fd147002c8a2fd09a1985b9a33 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/1ec4995fc9dbae2463a0ef5c51f676e86d31860224bce44cd7c84c9422234019.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bacbd34d736bab136b0f1d8045d762b715d021e31cf83f7cc35c07037491f85c +size 3575 diff --git a/parse/train/7Yhok3vJpU/images/227fc9a287f5e1eda38168112ec867444e874973f478b5bea85208831d0a5625.jpg b/parse/train/7Yhok3vJpU/images/227fc9a287f5e1eda38168112ec867444e874973f478b5bea85208831d0a5625.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9145fa5474d20723588ca5859d17cfecf92ea482 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/227fc9a287f5e1eda38168112ec867444e874973f478b5bea85208831d0a5625.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1ce9885bafa15e5169818bb25e9ea822b4859fb3f35674cdbde19c5197e1f179 +size 8042 diff --git a/parse/train/7Yhok3vJpU/images/272a974ef7d496f88181e8c74c4b702c525ed37dbfd8442493f5e20269d26187.jpg b/parse/train/7Yhok3vJpU/images/272a974ef7d496f88181e8c74c4b702c525ed37dbfd8442493f5e20269d26187.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fd8affadf9a4ad4ef66230ccb7c331a9949849f4 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/272a974ef7d496f88181e8c74c4b702c525ed37dbfd8442493f5e20269d26187.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3fc7e792896d804652cf1cf57ce15a38210b6c30e7d23e92d5a8d0d07274938f +size 15893 diff --git a/parse/train/7Yhok3vJpU/images/32f660214c7ad25d07c3d5ae87cd4dc650f693ba3dabb563f1766660eeae83d8.jpg b/parse/train/7Yhok3vJpU/images/32f660214c7ad25d07c3d5ae87cd4dc650f693ba3dabb563f1766660eeae83d8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..34ef320692d81deae9700653711402e9047961d6 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/32f660214c7ad25d07c3d5ae87cd4dc650f693ba3dabb563f1766660eeae83d8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:821194574f08b1f63a784771ef0005ca08e4cf5be3cfdbcf759094e53f5a3419 +size 27546 diff --git a/parse/train/7Yhok3vJpU/images/487ccf3e40b2099b5b62af271309a2ef2570fe174e4874859dcacee9f27940d9.jpg b/parse/train/7Yhok3vJpU/images/487ccf3e40b2099b5b62af271309a2ef2570fe174e4874859dcacee9f27940d9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a5efd6d6551ebfa52be1be2e9af6c40946da1643 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/487ccf3e40b2099b5b62af271309a2ef2570fe174e4874859dcacee9f27940d9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89bfd73e1ea53fa322d0c5852965abb1c531dceeb62786a7cd75e21a3426008e +size 7394 diff --git a/parse/train/7Yhok3vJpU/images/4a77a3007324713ae2f486ac47167ea60bca25def88b2de62bb08f9fd4d8e76d.jpg b/parse/train/7Yhok3vJpU/images/4a77a3007324713ae2f486ac47167ea60bca25def88b2de62bb08f9fd4d8e76d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..54cf3fcb2ca9ba1616cd20d9be394c1f6a832c94 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/4a77a3007324713ae2f486ac47167ea60bca25def88b2de62bb08f9fd4d8e76d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e069ef58bc25a3d5dde76610e3efa2c8d287569fc253410569522a78d6d9a146 +size 35017 diff --git a/parse/train/7Yhok3vJpU/images/5cb06c2121d1c46113fbd532d4eae33bc4a4a3ff286676c34cc07525332f354b.jpg b/parse/train/7Yhok3vJpU/images/5cb06c2121d1c46113fbd532d4eae33bc4a4a3ff286676c34cc07525332f354b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4512a46a0be365bffee231ad79f4a0064f8dec5b --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/5cb06c2121d1c46113fbd532d4eae33bc4a4a3ff286676c34cc07525332f354b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a079918f4c36db08d50136ec2d91c93cd04b2acd5b26159ef1b0d70364802504 +size 24221 diff --git a/parse/train/7Yhok3vJpU/images/76a5227aa259e581dbcd6a56fea4a8c78d8433381a26caf960824d7ddb935f1d.jpg b/parse/train/7Yhok3vJpU/images/76a5227aa259e581dbcd6a56fea4a8c78d8433381a26caf960824d7ddb935f1d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b809904c4483f2310a76a38dc181b982a4787a23 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/76a5227aa259e581dbcd6a56fea4a8c78d8433381a26caf960824d7ddb935f1d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9259558d250815aea7997e0e29a125b14ae81e143694ddc67b5ab8bbaa5c6461 +size 7690 diff --git a/parse/train/7Yhok3vJpU/images/797234188419c8764cfd4a209fbe87211a7806af6cbfe1909e4a4e4c65545a86.jpg b/parse/train/7Yhok3vJpU/images/797234188419c8764cfd4a209fbe87211a7806af6cbfe1909e4a4e4c65545a86.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0715676ee1be54493eb041a6d5a550eb96560005 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/797234188419c8764cfd4a209fbe87211a7806af6cbfe1909e4a4e4c65545a86.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aa8bf99a88337e85be0f49e04ef856848460aa4fa2c1e982b961d62b49f87e52 +size 7668 diff --git a/parse/train/7Yhok3vJpU/images/7fda920ffbc46747ea8419d691aacf07bee68040fb14f703107bfde96624ecad.jpg b/parse/train/7Yhok3vJpU/images/7fda920ffbc46747ea8419d691aacf07bee68040fb14f703107bfde96624ecad.jpg new file mode 100644 index 0000000000000000000000000000000000000000..db4374ae8ccbe7d14f66035c6f11a7bca6e78fea --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/7fda920ffbc46747ea8419d691aacf07bee68040fb14f703107bfde96624ecad.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7f08d9513f59efa055ffa2acad1dbac7c97906929fb8d08fa2e4aeff3b3f0119 +size 16850 diff --git a/parse/train/7Yhok3vJpU/images/904fbb92ef6f824eb7bc2389212778896e3cd3ccf5dd9049c5a2126c9a815f81.jpg b/parse/train/7Yhok3vJpU/images/904fbb92ef6f824eb7bc2389212778896e3cd3ccf5dd9049c5a2126c9a815f81.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a10fbd2d2de5c782c67879a20dc0559b65e915fc --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/904fbb92ef6f824eb7bc2389212778896e3cd3ccf5dd9049c5a2126c9a815f81.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c47ce4854ba06fffd7c6f7f2f2cd39eb7678bf74fbb18f23493877043b9c30ed +size 17344 diff --git a/parse/train/7Yhok3vJpU/images/97baf5a47a6cf2838bc6648bbb2bcfb21f0c886170251ff52ef795866c21ab57.jpg b/parse/train/7Yhok3vJpU/images/97baf5a47a6cf2838bc6648bbb2bcfb21f0c886170251ff52ef795866c21ab57.jpg new file mode 100644 index 0000000000000000000000000000000000000000..18b19f45ca1f16c756b795b15aadd17ce1d84968 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/97baf5a47a6cf2838bc6648bbb2bcfb21f0c886170251ff52ef795866c21ab57.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:789224301fad87ac0f2b2fe3214f29ea35ef580beca390bfdaff1a835b8373d7 +size 22991 diff --git a/parse/train/7Yhok3vJpU/images/a2903558b11fe401e2b357241657e05b379ee53b2dac927a774fe8d9332a9677.jpg b/parse/train/7Yhok3vJpU/images/a2903558b11fe401e2b357241657e05b379ee53b2dac927a774fe8d9332a9677.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e63a07d2d9150282afd08c567472cb195606ba10 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/a2903558b11fe401e2b357241657e05b379ee53b2dac927a774fe8d9332a9677.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cf0ce7c0e7fd753bebd9c738ca861a258e90e19485a5311866695a3a3eb5d638 +size 4939 diff --git a/parse/train/7Yhok3vJpU/images/b3239d0f44064e3f1b8190689f36fef09633f4df0b4f663268326b386b2a54b0.jpg b/parse/train/7Yhok3vJpU/images/b3239d0f44064e3f1b8190689f36fef09633f4df0b4f663268326b386b2a54b0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9e0ad37c876088aa52ac851efbf4cf90ffc5c880 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/b3239d0f44064e3f1b8190689f36fef09633f4df0b4f663268326b386b2a54b0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bdd312bc5d4f3507260b5b7bbfee4e12645c0726f86cd34531f996a898ba227a +size 62908 diff --git a/parse/train/7Yhok3vJpU/images/ba34f6a1fa07659aa2dd260834c793fb1bf46d6926d18dfb7931ba102f44f467.jpg b/parse/train/7Yhok3vJpU/images/ba34f6a1fa07659aa2dd260834c793fb1bf46d6926d18dfb7931ba102f44f467.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cd427a254a3a1f98a34ca22303c9f3cfadc38bdb --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/ba34f6a1fa07659aa2dd260834c793fb1bf46d6926d18dfb7931ba102f44f467.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d4962af6b7716c7e04a827e6131c974518d7e9222496e62e0710173a4ab7ce88 +size 2914 diff --git a/parse/train/7Yhok3vJpU/images/bb373d7f711155c47cca77973226284aacdfea6d1937f758b0d61c5480f4f15c.jpg b/parse/train/7Yhok3vJpU/images/bb373d7f711155c47cca77973226284aacdfea6d1937f758b0d61c5480f4f15c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3cc5750e8c92eaa4db4c090f8d8f6a4676bdbd71 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/bb373d7f711155c47cca77973226284aacdfea6d1937f758b0d61c5480f4f15c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4fd3455601d077602f899b51cc3bc12f050777a91e587cc5c8b99fa076e4bc6f +size 3867 diff --git a/parse/train/7Yhok3vJpU/images/c0580e2f94f8e959e0556e1525feeb88f7612e50ace567b789cd3abcf5ad1e51.jpg b/parse/train/7Yhok3vJpU/images/c0580e2f94f8e959e0556e1525feeb88f7612e50ace567b789cd3abcf5ad1e51.jpg new file mode 100644 index 0000000000000000000000000000000000000000..26b821d22c682f198f61bca90123d14f49987e6a --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/c0580e2f94f8e959e0556e1525feeb88f7612e50ace567b789cd3abcf5ad1e51.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:504c155c445592b32aafc51a893a4319917666b2a78a447ff83245b4ac8c3b0f +size 2637 diff --git a/parse/train/7Yhok3vJpU/images/c254aa492ed8850f2cffddcc0f6a8a4f01007d65394d8f7689e35e27e769de33.jpg b/parse/train/7Yhok3vJpU/images/c254aa492ed8850f2cffddcc0f6a8a4f01007d65394d8f7689e35e27e769de33.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6d505e66cf646f47fa96c42a9ccb64c06dfb1412 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/c254aa492ed8850f2cffddcc0f6a8a4f01007d65394d8f7689e35e27e769de33.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8c4f7b5985789e17c361b353bf539fffcaebe22871226e0c0b705eca9193bbc3 +size 17874 diff --git a/parse/train/7Yhok3vJpU/images/c880b2a87bcd6562b13a165093a948bb108d08b26a692590b9d8fcd0fb2aaf0b.jpg b/parse/train/7Yhok3vJpU/images/c880b2a87bcd6562b13a165093a948bb108d08b26a692590b9d8fcd0fb2aaf0b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5e4407033461387456f70f1a938f0df1666731bb --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/c880b2a87bcd6562b13a165093a948bb108d08b26a692590b9d8fcd0fb2aaf0b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe17fbed01409e3da6fba11c5da898bea49e973bf68ed7bce32a7de0b01eff04 +size 22885 diff --git a/parse/train/7Yhok3vJpU/images/cd0e70e58a94f4fcbf8c3fcf6c53a6795fa4f76eaf2bcc99f9da9c7d998655ff.jpg b/parse/train/7Yhok3vJpU/images/cd0e70e58a94f4fcbf8c3fcf6c53a6795fa4f76eaf2bcc99f9da9c7d998655ff.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8cce9b6a75c5f853cbf2ea0c4693a8bfef3aee5c --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/cd0e70e58a94f4fcbf8c3fcf6c53a6795fa4f76eaf2bcc99f9da9c7d998655ff.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d57634fc93cec2d184cd19ab9e3d0c6ca5381ecf71c68ff0e72fe9e9d26873fd +size 56302 diff --git a/parse/train/7Yhok3vJpU/images/d45aa662d4712d72985dd3f41578b8e4a2bd3a9681382face1f1cf211e4c3a33.jpg b/parse/train/7Yhok3vJpU/images/d45aa662d4712d72985dd3f41578b8e4a2bd3a9681382face1f1cf211e4c3a33.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6c4c8a93fe1b47546c1f2ffe281f53d728a2f8c4 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/d45aa662d4712d72985dd3f41578b8e4a2bd3a9681382face1f1cf211e4c3a33.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:71f73e813a89fc361b059675ed04289be1e153ea170743f8429b445c2f502800 +size 15434 diff --git a/parse/train/7Yhok3vJpU/images/da48883d1b3e68f8fde8391d7c425565e69d162597da6b41e44bcbf648f733a1.jpg b/parse/train/7Yhok3vJpU/images/da48883d1b3e68f8fde8391d7c425565e69d162597da6b41e44bcbf648f733a1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..34ac1beed1d376c017d39ca39c6d69dfedbe61d0 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/da48883d1b3e68f8fde8391d7c425565e69d162597da6b41e44bcbf648f733a1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b482b2ae43440bc6c7ca52b087a0ef8c1106e929be5b00aef92d5e90462a5590 +size 22335 diff --git a/parse/train/7Yhok3vJpU/images/ddebb6e382209f81701a3a28cdcff5f46ad0bee0b5b9d65fc3eb50cfb21766b0.jpg b/parse/train/7Yhok3vJpU/images/ddebb6e382209f81701a3a28cdcff5f46ad0bee0b5b9d65fc3eb50cfb21766b0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..09e9617746076b355f666a4c9ddfb96049a79605 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/ddebb6e382209f81701a3a28cdcff5f46ad0bee0b5b9d65fc3eb50cfb21766b0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1f72ddf6a57a145154a096d69658a7b13c95b215356c5bd6c6936934c3b1a3c3 +size 140508 diff --git a/parse/train/7Yhok3vJpU/images/ee34b59cd30d4f57c6ceda7f9a1a8d5f13bcd857a0568a13edf6962d9ea09583.jpg b/parse/train/7Yhok3vJpU/images/ee34b59cd30d4f57c6ceda7f9a1a8d5f13bcd857a0568a13edf6962d9ea09583.jpg new file mode 100644 index 0000000000000000000000000000000000000000..88c823a7c64c94d1bc2b8aa9f16f389060e9e4d6 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/ee34b59cd30d4f57c6ceda7f9a1a8d5f13bcd857a0568a13edf6962d9ea09583.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bcecd44b413b59381c6cabbdbefafdc6985ac875829e61d11cbbc2b5647bdea6 +size 23559 diff --git a/parse/train/7Yhok3vJpU/images/ef206f2119cac6929dfa3268358d5a9ecc44537a7bab01534645c218b4a726d4.jpg b/parse/train/7Yhok3vJpU/images/ef206f2119cac6929dfa3268358d5a9ecc44537a7bab01534645c218b4a726d4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bc6158a8841c15ea05d205192cff73034676b5a7 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/ef206f2119cac6929dfa3268358d5a9ecc44537a7bab01534645c218b4a726d4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2a1aac506bae623254943fdf307b5a34c0ef83110c1341862ca0043b69bc3c5a +size 2515 diff --git a/parse/train/7Yhok3vJpU/images/f8c5d2e85a0b3af334273456a91f9cdd7c9d4a0a5f8f4ec8b83e8e7cd3b1181a.jpg b/parse/train/7Yhok3vJpU/images/f8c5d2e85a0b3af334273456a91f9cdd7c9d4a0a5f8f4ec8b83e8e7cd3b1181a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5585b8731377cdb5369ce75d80ad95623d530dd6 --- /dev/null +++ b/parse/train/7Yhok3vJpU/images/f8c5d2e85a0b3af334273456a91f9cdd7c9d4a0a5f8f4ec8b83e8e7cd3b1181a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d563c532d900767168d79896c05e12f62f60d188c331f5e43765e1527e690da1 +size 8156 diff --git a/parse/train/B1G9tvcgx/images/06cdb8130b72418e484bd88c7a72c3295a6dd43723f58361d1b6fc7254026797.jpg b/parse/train/B1G9tvcgx/images/06cdb8130b72418e484bd88c7a72c3295a6dd43723f58361d1b6fc7254026797.jpg new file mode 100644 index 0000000000000000000000000000000000000000..48f03e6fda9a76a6b8a3ec4eff811cafa18a6349 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/06cdb8130b72418e484bd88c7a72c3295a6dd43723f58361d1b6fc7254026797.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c633b02745d68eccee54c30332889302cb429b550dcd24ef23270ad8364126ed +size 89468 diff --git a/parse/train/B1G9tvcgx/images/14471daa059b8d95974e504c430617cd19cc8a3ebbfa34c05662cdee40253162.jpg b/parse/train/B1G9tvcgx/images/14471daa059b8d95974e504c430617cd19cc8a3ebbfa34c05662cdee40253162.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4bf890b63fbe90b660b5217e5a4bdbcc04fea79b --- /dev/null +++ b/parse/train/B1G9tvcgx/images/14471daa059b8d95974e504c430617cd19cc8a3ebbfa34c05662cdee40253162.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:90a5ae7bff9d1476386b697f31e5fe05f28baebc63b49e81494bb23cb7e041bd +size 50845 diff --git a/parse/train/B1G9tvcgx/images/23012d4624eca34e1c303667d6e14ec10c97268f0b4680d616f88c774e11979f.jpg b/parse/train/B1G9tvcgx/images/23012d4624eca34e1c303667d6e14ec10c97268f0b4680d616f88c774e11979f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cb64329269d5841b29b2062f01d7d8bc9faaa20b --- /dev/null +++ b/parse/train/B1G9tvcgx/images/23012d4624eca34e1c303667d6e14ec10c97268f0b4680d616f88c774e11979f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0c090f7e07f0d6f1565a9cd2b7758b37e3fda66818a4674f424bf5bca2a3d27c +size 86346 diff --git a/parse/train/B1G9tvcgx/images/23f734b0c3a08c3fb01f46f0d47f13e1958a31b641b8ea49c1bb8c45641aeb0b.jpg b/parse/train/B1G9tvcgx/images/23f734b0c3a08c3fb01f46f0d47f13e1958a31b641b8ea49c1bb8c45641aeb0b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..74ed28e3e157e9d3de8e530a0ea8df9ffd13cf01 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/23f734b0c3a08c3fb01f46f0d47f13e1958a31b641b8ea49c1bb8c45641aeb0b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a5c3701fd1d390e2a5cf4f019368c5392eb4b73de2ab76fff8116549cd904a20 +size 14602 diff --git a/parse/train/B1G9tvcgx/images/240f2e29898a1b38f193efebd44fb38557f72522869993c09bbcae7d9178a52b.jpg b/parse/train/B1G9tvcgx/images/240f2e29898a1b38f193efebd44fb38557f72522869993c09bbcae7d9178a52b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..08fb3e0197a4e2ca4563da9c3eb01d80a2667a10 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/240f2e29898a1b38f193efebd44fb38557f72522869993c09bbcae7d9178a52b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b406bfc3cee30f52b1c0bb983b85387ee9048a35df31634a30edabe528945559 +size 51666 diff --git a/parse/train/B1G9tvcgx/images/299dae0bf7b503a86a3de4e96e9f8b69988b20636ff9692e950426a0ab3ffeb3.jpg b/parse/train/B1G9tvcgx/images/299dae0bf7b503a86a3de4e96e9f8b69988b20636ff9692e950426a0ab3ffeb3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e4adbdd42873f79dac57150fe44f50d7a278132c --- /dev/null +++ b/parse/train/B1G9tvcgx/images/299dae0bf7b503a86a3de4e96e9f8b69988b20636ff9692e950426a0ab3ffeb3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f5a1ec312e2956a4e633ab4c617a2d6f4c611cda00373befa405c79e9ad51865 +size 5235 diff --git a/parse/train/B1G9tvcgx/images/2c9cbbbc325b95ee17810529bec3d8c10b1b4be16b57ee667ba3a6348e29d4a5.jpg b/parse/train/B1G9tvcgx/images/2c9cbbbc325b95ee17810529bec3d8c10b1b4be16b57ee667ba3a6348e29d4a5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d3a7534ff0b5f3aff3f3241fd3c0fbc22406fdbd --- /dev/null +++ b/parse/train/B1G9tvcgx/images/2c9cbbbc325b95ee17810529bec3d8c10b1b4be16b57ee667ba3a6348e29d4a5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2268af595df930b3e945e4a98bbdd098e52aea314278e15e4dcb65e2781f83c8 +size 14067 diff --git a/parse/train/B1G9tvcgx/images/336f0b01b64b29847f26f5ecff9a7f50cb3d156bfae6b39739bff0d875544a49.jpg b/parse/train/B1G9tvcgx/images/336f0b01b64b29847f26f5ecff9a7f50cb3d156bfae6b39739bff0d875544a49.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ab4b4d3e9e4c1be83e0daf3b1416c3ea0e66e305 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/336f0b01b64b29847f26f5ecff9a7f50cb3d156bfae6b39739bff0d875544a49.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89db2d502003b0ff4462fb4add30d148970451c3132304250a34ae70483ccbb0 +size 24760 diff --git a/parse/train/B1G9tvcgx/images/3899a0a8465e2ce98deaee871de5b2b920dce614f48ba85d244452b665f39c4a.jpg b/parse/train/B1G9tvcgx/images/3899a0a8465e2ce98deaee871de5b2b920dce614f48ba85d244452b665f39c4a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..67c61d72541de645f0e44ce254975f8cde6816c6 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/3899a0a8465e2ce98deaee871de5b2b920dce614f48ba85d244452b665f39c4a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:75d70f244c5c82eb3d6c51719801342d025e7fec3c983d5dd266db25cd8d520e +size 65675 diff --git a/parse/train/B1G9tvcgx/images/3e6e262a2cb9ed0521cc90b099233c2907e3bb10cdea3ffc73cf509db5803456.jpg b/parse/train/B1G9tvcgx/images/3e6e262a2cb9ed0521cc90b099233c2907e3bb10cdea3ffc73cf509db5803456.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3c909e1cc91753ba7e6bd98f9c978c5305db9e2e --- /dev/null +++ b/parse/train/B1G9tvcgx/images/3e6e262a2cb9ed0521cc90b099233c2907e3bb10cdea3ffc73cf509db5803456.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bed938ecab7d54ab750022087beec59f5bd080a9942b1b7c558201cb90a071b9 +size 53908 diff --git a/parse/train/B1G9tvcgx/images/401402a8fe1a4881873fd8a84766e65eef1b866c373de7db920db7592bf11743.jpg b/parse/train/B1G9tvcgx/images/401402a8fe1a4881873fd8a84766e65eef1b866c373de7db920db7592bf11743.jpg new file mode 100644 index 0000000000000000000000000000000000000000..97926b8956d99cd028b421827b7635632a7f570b --- /dev/null +++ b/parse/train/B1G9tvcgx/images/401402a8fe1a4881873fd8a84766e65eef1b866c373de7db920db7592bf11743.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f0f3df74c4396ee874dab06c9616068cd7c59db14346523ea776f3cad4f483c4 +size 85510 diff --git a/parse/train/B1G9tvcgx/images/40438e3127b0e3dfabe5ef181edd81e6a9767fc1c201406ec2d489eec5be4b9a.jpg b/parse/train/B1G9tvcgx/images/40438e3127b0e3dfabe5ef181edd81e6a9767fc1c201406ec2d489eec5be4b9a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3d7ef776134b26fb28a9818083e7c3d14b22678a --- /dev/null +++ b/parse/train/B1G9tvcgx/images/40438e3127b0e3dfabe5ef181edd81e6a9767fc1c201406ec2d489eec5be4b9a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2b4a9987a4dc2002b94fd800f5ec57fe26237252f0fbfa6b90a2dd89a8394592 +size 31667 diff --git a/parse/train/B1G9tvcgx/images/46806f443a5d1b257d6202fc27ff7119c7a1f533c7430b581a327e357fcab959.jpg b/parse/train/B1G9tvcgx/images/46806f443a5d1b257d6202fc27ff7119c7a1f533c7430b581a327e357fcab959.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a720032a337c25a57773d4c96202938d1c0661e2 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/46806f443a5d1b257d6202fc27ff7119c7a1f533c7430b581a327e357fcab959.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:769c999403a09de83513c6d51f4dab093e9259be95889bd0bd61bd4545042a0f +size 11568 diff --git a/parse/train/B1G9tvcgx/images/4cf686d842327ec69a90e4b59544c93c4bbf7f5db98b7ef11c7a1a405e320bf0.jpg b/parse/train/B1G9tvcgx/images/4cf686d842327ec69a90e4b59544c93c4bbf7f5db98b7ef11c7a1a405e320bf0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5ddd320ae5076f72737a57903a4fa2eae7a4c7f7 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/4cf686d842327ec69a90e4b59544c93c4bbf7f5db98b7ef11c7a1a405e320bf0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a8147c1192a0a1a23f0887d4f145e76225133fb4b3508d6e15068d1e1b4b3956 +size 16138 diff --git a/parse/train/B1G9tvcgx/images/5cbe81cb72340f2ba8e4f579766d3dc674988f34d481a805fadc437a74525605.jpg b/parse/train/B1G9tvcgx/images/5cbe81cb72340f2ba8e4f579766d3dc674988f34d481a805fadc437a74525605.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c89f44f8bc029d88e7e90100404b1d5dda0b33d5 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/5cbe81cb72340f2ba8e4f579766d3dc674988f34d481a805fadc437a74525605.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1156d5c4daf19cd74b83df1dd0fc59370e5c6b71fa2f5c754550fc7d3f182cdc +size 23934 diff --git a/parse/train/B1G9tvcgx/images/6502f6b973ad8ec97a68b43790daf2182aca15ec76644b4f658449b216fde451.jpg b/parse/train/B1G9tvcgx/images/6502f6b973ad8ec97a68b43790daf2182aca15ec76644b4f658449b216fde451.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f7891ff4fa8f5963f7b25da3ee3b617f1a4fae4e --- /dev/null +++ b/parse/train/B1G9tvcgx/images/6502f6b973ad8ec97a68b43790daf2182aca15ec76644b4f658449b216fde451.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b7ed83e16e12c6dd07a10d023827d274e531b2393801bc544d823997d3e73791 +size 7124 diff --git a/parse/train/B1G9tvcgx/images/70259d0e0a7683914c944fb8a1e952ca9352c46ec72e07fbb0adaf84555a36f1.jpg b/parse/train/B1G9tvcgx/images/70259d0e0a7683914c944fb8a1e952ca9352c46ec72e07fbb0adaf84555a36f1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..da81f6cc353d12c877c31ed42bac50382785dfba --- /dev/null +++ b/parse/train/B1G9tvcgx/images/70259d0e0a7683914c944fb8a1e952ca9352c46ec72e07fbb0adaf84555a36f1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec975eeb51028ad7597c4614936e0e73780e61d8e9263510316da11b4317d65f +size 4568 diff --git a/parse/train/B1G9tvcgx/images/7d59f4e9d8e0d16dcda6d9fe5f00e4043acaa4825994a380abfab9cb596829ee.jpg b/parse/train/B1G9tvcgx/images/7d59f4e9d8e0d16dcda6d9fe5f00e4043acaa4825994a380abfab9cb596829ee.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f73bebaefa9804a57d14e34f48b957988d13aa4e --- /dev/null +++ b/parse/train/B1G9tvcgx/images/7d59f4e9d8e0d16dcda6d9fe5f00e4043acaa4825994a380abfab9cb596829ee.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f59b89f07dc68a495ac4020eb550532023a026f4ef35ac1cfde8fd23fd1ac2f0 +size 42225 diff --git a/parse/train/B1G9tvcgx/images/81ce8181818576a9c1ef7b4436af568b765c92f924c961157b1a440ee3e66220.jpg b/parse/train/B1G9tvcgx/images/81ce8181818576a9c1ef7b4436af568b765c92f924c961157b1a440ee3e66220.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f82d7bb21b2830cf255c83235ef2fcd337a5cfdb --- /dev/null +++ b/parse/train/B1G9tvcgx/images/81ce8181818576a9c1ef7b4436af568b765c92f924c961157b1a440ee3e66220.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0757cc7af52552f0a44867d07f535457092b6a9cd62f582247a2a282d643a627 +size 6008 diff --git a/parse/train/B1G9tvcgx/images/82bdaf12150fc3d8e9e717efde53d35c62a587f9dfaba72176b6817c40bdc457.jpg b/parse/train/B1G9tvcgx/images/82bdaf12150fc3d8e9e717efde53d35c62a587f9dfaba72176b6817c40bdc457.jpg new file mode 100644 index 0000000000000000000000000000000000000000..137d9716f790e461d29de6c98945877aa00dac5d --- /dev/null +++ b/parse/train/B1G9tvcgx/images/82bdaf12150fc3d8e9e717efde53d35c62a587f9dfaba72176b6817c40bdc457.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5761baab8df59e52b7f7284cbaff0d9a76b3e5e388e42ce9bd4b13414ccdfc6f +size 15297 diff --git a/parse/train/B1G9tvcgx/images/846bddf000f225251138a3e89465935cc60a3ca957d5dade41a91ee1d2ddf123.jpg b/parse/train/B1G9tvcgx/images/846bddf000f225251138a3e89465935cc60a3ca957d5dade41a91ee1d2ddf123.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b9ba4a876807bf61d1d9d2e32d4b72e3bc8ff648 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/846bddf000f225251138a3e89465935cc60a3ca957d5dade41a91ee1d2ddf123.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b91afb40a42e1747c1855facd60110a5f9be95385ad5c81ecaef3a7fbc639536 +size 3268 diff --git a/parse/train/B1G9tvcgx/images/883b83a46b4563901602a6c1dc431482e25758f8a151cea196a8465ec6b538d2.jpg b/parse/train/B1G9tvcgx/images/883b83a46b4563901602a6c1dc431482e25758f8a151cea196a8465ec6b538d2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0ad37c60203ec5140ea6f27e20b7ac0fa291cad3 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/883b83a46b4563901602a6c1dc431482e25758f8a151cea196a8465ec6b538d2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3cb9a0d9f121d4ee3e952c6ec0cc51d35a4e3ede7d3df1b5a829b4ec6c4c08d3 +size 26295 diff --git a/parse/train/B1G9tvcgx/images/8af5c4d5b27d353d4a326de062e21493febd2eacc9bc99e2fb71b765366ac08a.jpg b/parse/train/B1G9tvcgx/images/8af5c4d5b27d353d4a326de062e21493febd2eacc9bc99e2fb71b765366ac08a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e136177aae4c0b210314f8a0ad2bc3616b479781 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/8af5c4d5b27d353d4a326de062e21493febd2eacc9bc99e2fb71b765366ac08a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f286e390967c12d25c13b443d9fcc2df07fa8a8229669b026bebbbe46251fc84 +size 13450 diff --git a/parse/train/B1G9tvcgx/images/8be7ff761f48d12a81eaaa70bf4c62d437638b12f62d9d8cd9b8e52448d94dc3.jpg b/parse/train/B1G9tvcgx/images/8be7ff761f48d12a81eaaa70bf4c62d437638b12f62d9d8cd9b8e52448d94dc3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..647f135e589b971a45ec9727f451c27d3f92e92f --- /dev/null +++ b/parse/train/B1G9tvcgx/images/8be7ff761f48d12a81eaaa70bf4c62d437638b12f62d9d8cd9b8e52448d94dc3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:69376cd8c4cc5fd9c4e86dfc3b14454bcbe9f3ee6c665b41fb47440cb995ce6b +size 37475 diff --git a/parse/train/B1G9tvcgx/images/a56e14ddb8c057b7c0069709365a555c8048694ea5de00bf35520da2e5743370.jpg b/parse/train/B1G9tvcgx/images/a56e14ddb8c057b7c0069709365a555c8048694ea5de00bf35520da2e5743370.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d4f77359cd104f313c44d1d042c7505469058751 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/a56e14ddb8c057b7c0069709365a555c8048694ea5de00bf35520da2e5743370.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8d5438fef94fed1d045f9a69add8661997eabd1618e42a32994ca96227982c96 +size 7969 diff --git a/parse/train/B1G9tvcgx/images/a9260f0f6290f1c96ecbfc79ab0478644bff0af1a4a394761f746d36cd74f6ab.jpg b/parse/train/B1G9tvcgx/images/a9260f0f6290f1c96ecbfc79ab0478644bff0af1a4a394761f746d36cd74f6ab.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4ebf7df86cadfabb8a9b542fd634f14d4656c3b9 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/a9260f0f6290f1c96ecbfc79ab0478644bff0af1a4a394761f746d36cd74f6ab.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ddd69d205ac8efc175fe747bab356355f3ed0f4802be9ac226fd3816323f1af2 +size 40543 diff --git a/parse/train/B1G9tvcgx/images/a983063495e5006dbabbbc66734d275a48316b65a5141943c2d1c94782a309fe.jpg b/parse/train/B1G9tvcgx/images/a983063495e5006dbabbbc66734d275a48316b65a5141943c2d1c94782a309fe.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f8dc0de92d09d55af372db2cc231a9366e7d4570 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/a983063495e5006dbabbbc66734d275a48316b65a5141943c2d1c94782a309fe.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:facfb720ceb250610fefe6fbc9ba93eeaf53e9768b602eda509774924245cf6c +size 4583 diff --git a/parse/train/B1G9tvcgx/images/acc29606458a5807494c450d7d98cfbf2019552dbf1817b3f5117cb49d155fc3.jpg b/parse/train/B1G9tvcgx/images/acc29606458a5807494c450d7d98cfbf2019552dbf1817b3f5117cb49d155fc3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c04bbc54ded6937101013a155a0b745c6ed6b040 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/acc29606458a5807494c450d7d98cfbf2019552dbf1817b3f5117cb49d155fc3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b6bdbf59c07187184e9c6010a28934aecdbc9b46b77ac89ac6ba2f1cdc4046f4 +size 71290 diff --git a/parse/train/B1G9tvcgx/images/b0308383548bfd1bdb406863a7474d490fee0deafc740e5feabcee5ef1f7bab4.jpg b/parse/train/B1G9tvcgx/images/b0308383548bfd1bdb406863a7474d490fee0deafc740e5feabcee5ef1f7bab4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a0eabaf58dbd580c63a85e29d9dc051de696f601 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/b0308383548bfd1bdb406863a7474d490fee0deafc740e5feabcee5ef1f7bab4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:594496ea1551b3d096186c0ae774fbb656d89209c61aa3385c0a509d28e5e835 +size 39254 diff --git a/parse/train/B1G9tvcgx/images/b9aa0f94d1e754dd559290e8983f30176de9651a391cc7a962dce590875455c5.jpg b/parse/train/B1G9tvcgx/images/b9aa0f94d1e754dd559290e8983f30176de9651a391cc7a962dce590875455c5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1a0a1dabfaf2bc82a4ff920d191618cb1099231f --- /dev/null +++ b/parse/train/B1G9tvcgx/images/b9aa0f94d1e754dd559290e8983f30176de9651a391cc7a962dce590875455c5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0910efca1d1940ae3cdabfc7645b4702d0afc5ea7a91db358a2c8887cfdb4514 +size 71310 diff --git a/parse/train/B1G9tvcgx/images/ba82b395bd30213c57d6b3bf44aafdcbbab8ac2771bc15054d7f9a543da7db27.jpg b/parse/train/B1G9tvcgx/images/ba82b395bd30213c57d6b3bf44aafdcbbab8ac2771bc15054d7f9a543da7db27.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ed5a7a284e0b961f4155146fce0c6bb65c49e681 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/ba82b395bd30213c57d6b3bf44aafdcbbab8ac2771bc15054d7f9a543da7db27.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dc94f7b742d59770681efd871c841f724d5dccf9b8a3ebfd743d850a64935a6c +size 44377 diff --git a/parse/train/B1G9tvcgx/images/bb605f8c5f246dc01410e5a4789cf874f02fe60269a8ccfae2a0e74f7144f6b4.jpg b/parse/train/B1G9tvcgx/images/bb605f8c5f246dc01410e5a4789cf874f02fe60269a8ccfae2a0e74f7144f6b4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f40dc273ec02af238c729ba194b33e3bc4a479b7 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/bb605f8c5f246dc01410e5a4789cf874f02fe60269a8ccfae2a0e74f7144f6b4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b39a1563312be9e6ce2a412412d7032dcfb6b607404e50722a6e09e642fe0ad6 +size 33685 diff --git a/parse/train/B1G9tvcgx/images/be44e2eca20e772743f6f2e21e2702d4272dad150589e6eca8ea1113b2c8c7d9.jpg b/parse/train/B1G9tvcgx/images/be44e2eca20e772743f6f2e21e2702d4272dad150589e6eca8ea1113b2c8c7d9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..046d4a83b6eb12ba2ded35c6b8cc0b940e8b5e3e --- /dev/null +++ b/parse/train/B1G9tvcgx/images/be44e2eca20e772743f6f2e21e2702d4272dad150589e6eca8ea1113b2c8c7d9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c9868ac8ff95fee5b604eee8d38df2f41cfd5632cc7d152ae45882e22f9062f1 +size 9831 diff --git a/parse/train/B1G9tvcgx/images/c456d60b1958b7c4cc5590dafbc840a9ea90f793a80d17524128d3f410acec0f.jpg b/parse/train/B1G9tvcgx/images/c456d60b1958b7c4cc5590dafbc840a9ea90f793a80d17524128d3f410acec0f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5a37896954b09c7c51d849d447bdabae2b28cc77 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/c456d60b1958b7c4cc5590dafbc840a9ea90f793a80d17524128d3f410acec0f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e186c5fc92e96568c6042fc2c120c66de5df021d44937e8a16bc8a4b9f8766cc +size 13091 diff --git a/parse/train/B1G9tvcgx/images/c7a23d3386179254d992cf59a6bfb4684fdde8b205007edc2ffd3f3352dbb994.jpg b/parse/train/B1G9tvcgx/images/c7a23d3386179254d992cf59a6bfb4684fdde8b205007edc2ffd3f3352dbb994.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c50e65639ed29221c82981ef8604aa923fec6966 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/c7a23d3386179254d992cf59a6bfb4684fdde8b205007edc2ffd3f3352dbb994.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c2ba0a28f224e1e1203ee34c49c9f4d20edb135cc98771966e966df1136afa13 +size 12596 diff --git a/parse/train/B1G9tvcgx/images/c7bae09f36a43d6018ae579ac15c03de257556b8181a0a9f38fd44944c05724c.jpg b/parse/train/B1G9tvcgx/images/c7bae09f36a43d6018ae579ac15c03de257556b8181a0a9f38fd44944c05724c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..83cdfc6117894c287cac9c33aa61eaa1b51ad1b3 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/c7bae09f36a43d6018ae579ac15c03de257556b8181a0a9f38fd44944c05724c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3a782aabb6ae1b4da0f4d0f9c9812283cb1b5571b2d674c112dd687ad178f9f8 +size 21892 diff --git a/parse/train/B1G9tvcgx/images/ccfa72a2ec2ff7edca4f4312964c8243f7c282543990f91671c968c41875b71c.jpg b/parse/train/B1G9tvcgx/images/ccfa72a2ec2ff7edca4f4312964c8243f7c282543990f91671c968c41875b71c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ca1078073229f12879968aeb0c82f6d84048fc0c --- /dev/null +++ b/parse/train/B1G9tvcgx/images/ccfa72a2ec2ff7edca4f4312964c8243f7c282543990f91671c968c41875b71c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aa348fa061925d1e59e10f5ba50576f2fa615b52e3908da6c615691b3411d9ca +size 5905 diff --git a/parse/train/B1G9tvcgx/images/d411a02e14a3e6d08312a40b1f51766c1fe322869b0edcd0e47bc3d2e6bd81c1.jpg b/parse/train/B1G9tvcgx/images/d411a02e14a3e6d08312a40b1f51766c1fe322869b0edcd0e47bc3d2e6bd81c1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9af1a297487c3f7b0e8723297010c5d902713878 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/d411a02e14a3e6d08312a40b1f51766c1fe322869b0edcd0e47bc3d2e6bd81c1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2e1f0c21bdfc1794c0b8e8ace3b3e7c1aea947d85a57290ea2a73f65c588621d +size 115786 diff --git a/parse/train/B1G9tvcgx/images/e7151f581c4957dd2df8001ca4e255d512e80383a168a778786ef6cf7af980b9.jpg b/parse/train/B1G9tvcgx/images/e7151f581c4957dd2df8001ca4e255d512e80383a168a778786ef6cf7af980b9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..760066862b1e2546864920bcd8360c6045630cd3 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/e7151f581c4957dd2df8001ca4e255d512e80383a168a778786ef6cf7af980b9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cbb2ce6543993091a34286540e44c370beba612166e91454bf6157d97c70f662 +size 40532 diff --git a/parse/train/B1G9tvcgx/images/e7e1cffb66ec64837e0cac453d10edb9da5dedffdae576f87147bad40a0f7acb.jpg b/parse/train/B1G9tvcgx/images/e7e1cffb66ec64837e0cac453d10edb9da5dedffdae576f87147bad40a0f7acb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..337a7cd3027a19cbcc848559151090dfa1948c05 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/e7e1cffb66ec64837e0cac453d10edb9da5dedffdae576f87147bad40a0f7acb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3eeec0640baa2737bf81939594aef225e118fbf2abdfc12c40fdd8455b280989 +size 8376 diff --git a/parse/train/B1G9tvcgx/images/f3b82a6e19832c2302eff7d41d12b5e9c5026c1c1b6cbe2164bace73bb36d10d.jpg b/parse/train/B1G9tvcgx/images/f3b82a6e19832c2302eff7d41d12b5e9c5026c1c1b6cbe2164bace73bb36d10d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..efdd5138b68c0f8acacd579e487e4e6bca7d44d9 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/f3b82a6e19832c2302eff7d41d12b5e9c5026c1c1b6cbe2164bace73bb36d10d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7faa12569d8b4a1594d885c31456148ee54144a82f7c6a4bc6dea90ccd38893b +size 32058 diff --git a/parse/train/B1G9tvcgx/images/f4a6e36a04bba32d9bb47b3aaf084fd5b564ef6b93f245a779d068b254459b67.jpg b/parse/train/B1G9tvcgx/images/f4a6e36a04bba32d9bb47b3aaf084fd5b564ef6b93f245a779d068b254459b67.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8096518f7aee8e13991b098223c256a6cd95a5eb --- /dev/null +++ b/parse/train/B1G9tvcgx/images/f4a6e36a04bba32d9bb47b3aaf084fd5b564ef6b93f245a779d068b254459b67.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:60f5dd8a1f3f0eadbb4e4f663251031a89afefa9760c4ef73a939ffccf76a7fb +size 58614 diff --git a/parse/train/B1G9tvcgx/images/f5285ea63366850a021ab9f7f8e2cac294f6d2bbde53ffb04b34c0156f44619e.jpg b/parse/train/B1G9tvcgx/images/f5285ea63366850a021ab9f7f8e2cac294f6d2bbde53ffb04b34c0156f44619e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b7ddd6f22a8a65aa3a9dcf8b65cfb646051c89e4 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/f5285ea63366850a021ab9f7f8e2cac294f6d2bbde53ffb04b34c0156f44619e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1f03907b66a8c73b1258b0a17699476edde9d4d782914d1f761e8e7304a59278 +size 31732 diff --git a/parse/train/B1G9tvcgx/images/f968de7ed3c90ef7d20c5eef06c2507c4c5715eb815a39c0c100bb701d4b2774.jpg b/parse/train/B1G9tvcgx/images/f968de7ed3c90ef7d20c5eef06c2507c4c5715eb815a39c0c100bb701d4b2774.jpg new file mode 100644 index 0000000000000000000000000000000000000000..61930da76512c72b8b34de66098d2fa453e00509 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/f968de7ed3c90ef7d20c5eef06c2507c4c5715eb815a39c0c100bb701d4b2774.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cbcf2a9db794b49671329d001c58b980a3bf076a57987ecdc7d775f190f3fa85 +size 8590 diff --git a/parse/train/B1G9tvcgx/images/fcea07378be0eca7db3ccf46a7e91732f1442d9772a1f368e90dce6e0187e23d.jpg b/parse/train/B1G9tvcgx/images/fcea07378be0eca7db3ccf46a7e91732f1442d9772a1f368e90dce6e0187e23d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bf065f27ca171b70a1dcdbb273a3a012eb238534 --- /dev/null +++ b/parse/train/B1G9tvcgx/images/fcea07378be0eca7db3ccf46a7e91732f1442d9772a1f368e90dce6e0187e23d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5ef4f3f741c2e28297bb40bf56b3b1d013057349829c5d6f7a95fc444cdb026a +size 5427 diff --git a/parse/train/B1uvH_gC-/images/056c55e83f05f2f9a5b8d809148ee337b2154b3b58de244e89fd69fb40c3e816.jpg b/parse/train/B1uvH_gC-/images/056c55e83f05f2f9a5b8d809148ee337b2154b3b58de244e89fd69fb40c3e816.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d0bd45b18e9a575f0210cfd8872ebf3203b00e37 --- /dev/null +++ b/parse/train/B1uvH_gC-/images/056c55e83f05f2f9a5b8d809148ee337b2154b3b58de244e89fd69fb40c3e816.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e930521878c5b65fbd692b6f40db3666a8295918fb2945417b4dee48f8689afe +size 4979 diff --git a/parse/train/B1uvH_gC-/images/067f94adf55c46127a90161eb05f3fb40f35eae1e24282393a2e207e5bf0c625.jpg b/parse/train/B1uvH_gC-/images/067f94adf55c46127a90161eb05f3fb40f35eae1e24282393a2e207e5bf0c625.jpg new file mode 100644 index 0000000000000000000000000000000000000000..49dd929c04464cc1c6a195a5983544b84c485173 --- /dev/null +++ b/parse/train/B1uvH_gC-/images/067f94adf55c46127a90161eb05f3fb40f35eae1e24282393a2e207e5bf0c625.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:42be7f871c57fa7083bbfb48b816add5af9cfac3412052156b3642f59b9009e3 +size 1601 diff --git a/parse/train/B1uvH_gC-/images/0bdb60483ad25ce126717db85f6420092f580bcf4a3dea150a9b13298f12ef55.jpg b/parse/train/B1uvH_gC-/images/0bdb60483ad25ce126717db85f6420092f580bcf4a3dea150a9b13298f12ef55.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a54033c30f585c92438780c95de735264f626a5c --- /dev/null +++ b/parse/train/B1uvH_gC-/images/0bdb60483ad25ce126717db85f6420092f580bcf4a3dea150a9b13298f12ef55.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dd9c64366c6ded338b4d8c3b7caf45fa25ad9249de767449deabb73339826b88 +size 8117 diff --git a/parse/train/B1uvH_gC-/images/1192f770f43cf79258d17aecf1b9b90fdc30122acdb142febb67172cda5f83e1.jpg b/parse/train/B1uvH_gC-/images/1192f770f43cf79258d17aecf1b9b90fdc30122acdb142febb67172cda5f83e1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..790953cd7bd2efb207e83a87d9187a5e39d3e421 --- /dev/null +++ b/parse/train/B1uvH_gC-/images/1192f770f43cf79258d17aecf1b9b90fdc30122acdb142febb67172cda5f83e1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:247e6cd688e4bbc3c2d709b96792e1ba5850b9a76d3e93bf1cc7c1f1c7f326ce +size 56914 diff --git a/parse/train/B1uvH_gC-/images/11c4e9f6c7f703ee00e0f45a02cc7c6d25db28a8f359a7e37e7fccc8199fbfd6.jpg b/parse/train/B1uvH_gC-/images/11c4e9f6c7f703ee00e0f45a02cc7c6d25db28a8f359a7e37e7fccc8199fbfd6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..00f308ae42cc21b0c1e6e87b609f74daab60013d --- /dev/null +++ b/parse/train/B1uvH_gC-/images/11c4e9f6c7f703ee00e0f45a02cc7c6d25db28a8f359a7e37e7fccc8199fbfd6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9b675a1c2ae655c1fa4bc67cf4500e504e74abfd67352576aa6b49ba3ff959de +size 63027 diff --git a/parse/train/B1uvH_gC-/images/1e0e72e284db26e93a4bb04264800d80160a28d61248115414e2408bf0ab5e79.jpg b/parse/train/B1uvH_gC-/images/1e0e72e284db26e93a4bb04264800d80160a28d61248115414e2408bf0ab5e79.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a37e3d43e4393dc37408f78ce9848ceda98d5a99 --- /dev/null +++ b/parse/train/B1uvH_gC-/images/1e0e72e284db26e93a4bb04264800d80160a28d61248115414e2408bf0ab5e79.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89fef72b928ac561245b902c19c2c2c7c8898ed79d2281bffd8afa848e758605 +size 3989 diff --git a/parse/train/B1uvH_gC-/images/1ec51c80d80d9979b708950fb6aa3f8e181c5487ae1cb146326260e4ee359779.jpg b/parse/train/B1uvH_gC-/images/1ec51c80d80d9979b708950fb6aa3f8e181c5487ae1cb146326260e4ee359779.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9d67082218e3142548e7827ca03b1ac813728aae --- /dev/null +++ b/parse/train/B1uvH_gC-/images/1ec51c80d80d9979b708950fb6aa3f8e181c5487ae1cb146326260e4ee359779.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6d2a3b85158bc2ee306fcbdfcb4aba90a55598d298ae6df7664e2bb70fec98c0 +size 7115 diff --git a/parse/train/B1uvH_gC-/images/21b58a74e5448145d64324281c028acc2f5278afbafe7ee573e8caef648b3fc3.jpg b/parse/train/B1uvH_gC-/images/21b58a74e5448145d64324281c028acc2f5278afbafe7ee573e8caef648b3fc3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3f44a65499dd6c7791100bda16917d1de4015a12 --- /dev/null +++ b/parse/train/B1uvH_gC-/images/21b58a74e5448145d64324281c028acc2f5278afbafe7ee573e8caef648b3fc3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6af6da95bf671134689a98f6738ab5fbbb9dcd7668b6227f6214a3545f375c67 +size 61516 diff --git a/parse/train/B1uvH_gC-/images/2e204ad6304c9ea2bca00a267d9368dd47bc769f7c21d428f3697498d50fe0da.jpg b/parse/train/B1uvH_gC-/images/2e204ad6304c9ea2bca00a267d9368dd47bc769f7c21d428f3697498d50fe0da.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ded9ee7543b4068bc586d791b3002bdf876e0c55 --- /dev/null +++ b/parse/train/B1uvH_gC-/images/2e204ad6304c9ea2bca00a267d9368dd47bc769f7c21d428f3697498d50fe0da.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:58ba94ea4847eaaf7b211ee1f4858270498bd179b341bd025ee82cb99a334448 +size 27599 diff --git a/parse/train/B1uvH_gC-/images/3159108861cf69b01d621e79a4006689641c5f95a69042c8651c3b791066bd28.jpg b/parse/train/B1uvH_gC-/images/3159108861cf69b01d621e79a4006689641c5f95a69042c8651c3b791066bd28.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9db11fccdaabd2a766c9a06189e5bd372e4a961a --- /dev/null +++ b/parse/train/B1uvH_gC-/images/3159108861cf69b01d621e79a4006689641c5f95a69042c8651c3b791066bd28.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4f969e3f7bb52705b8e30b380b74013a24894f023a1979dc233c74ae6e5479a0 +size 44870 diff --git a/parse/train/B1uvH_gC-/images/3c392512f0fced5edccdf2537d5e50a4dd7b1aced3b5be269dec808b045b5f82.jpg b/parse/train/B1uvH_gC-/images/3c392512f0fced5edccdf2537d5e50a4dd7b1aced3b5be269dec808b045b5f82.jpg new file mode 100644 index 0000000000000000000000000000000000000000..239273a353447abfc1a4c83c5c6c8cee747915e7 --- /dev/null +++ b/parse/train/B1uvH_gC-/images/3c392512f0fced5edccdf2537d5e50a4dd7b1aced3b5be269dec808b045b5f82.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fc1c321357bebd30fda03d4065967bcc4137e82736097e97fe9f1fdcc68e5a03 +size 12883 diff --git a/parse/train/B1uvH_gC-/images/67613dc0591ad6925f8924d6d58c2963fe047e733c35296d13cd92d3df147dd0.jpg b/parse/train/B1uvH_gC-/images/67613dc0591ad6925f8924d6d58c2963fe047e733c35296d13cd92d3df147dd0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cf44af91a49dc16209cdd8f28af0957ee0a868e6 --- /dev/null +++ b/parse/train/B1uvH_gC-/images/67613dc0591ad6925f8924d6d58c2963fe047e733c35296d13cd92d3df147dd0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:86b4d2d75bb4ced048557de0d16f284593cfb218342e41f85f5489ab0f2b14d0 +size 55905 diff --git a/parse/train/B1uvH_gC-/images/70625e1c8d170a331b7b17b153db79441811a3f732f614c93d90325b141700b4.jpg b/parse/train/B1uvH_gC-/images/70625e1c8d170a331b7b17b153db79441811a3f732f614c93d90325b141700b4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..536b148efa21ca3148c367c38cf58f66f6239014 --- /dev/null +++ b/parse/train/B1uvH_gC-/images/70625e1c8d170a331b7b17b153db79441811a3f732f614c93d90325b141700b4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:35e33896ebf880ad9a5313354ff050dc0d20268734a497dc99f83ed7a7d95b97 +size 45089 diff --git a/parse/train/B1uvH_gC-/images/73c08738a754ea4c022aeedda0837ef3e7869e334016c3468b84fcf52d1d9483.jpg b/parse/train/B1uvH_gC-/images/73c08738a754ea4c022aeedda0837ef3e7869e334016c3468b84fcf52d1d9483.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3d4c2525632641495e4f15ae35b2247fc206db51 --- /dev/null +++ b/parse/train/B1uvH_gC-/images/73c08738a754ea4c022aeedda0837ef3e7869e334016c3468b84fcf52d1d9483.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4788cd3ad84aa518ffaa94bc01b020f81446b19572822c134c8f4c6594116aa0 +size 52323 diff --git a/parse/train/B1uvH_gC-/images/a79b8d87b1ff51e41dae1568ed792d8505f689a9468de0f7a6462fd97ee57d77.jpg b/parse/train/B1uvH_gC-/images/a79b8d87b1ff51e41dae1568ed792d8505f689a9468de0f7a6462fd97ee57d77.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5ac31ccd12902fe25abcc4964fd9d2713395e16b --- /dev/null +++ b/parse/train/B1uvH_gC-/images/a79b8d87b1ff51e41dae1568ed792d8505f689a9468de0f7a6462fd97ee57d77.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f3dd7518f062f5a24705043fbd2d54f3b1f973c0ea2c0706913a14e368f2b055 +size 5822 diff --git a/parse/train/B1uvH_gC-/images/b3a79b4142cbc27ded70e8f2769057186f23e49766eedcd8367d541d1ead13a2.jpg b/parse/train/B1uvH_gC-/images/b3a79b4142cbc27ded70e8f2769057186f23e49766eedcd8367d541d1ead13a2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bdc86ce6f1dca6efac0be6f054aaf8afbf40c3d5 --- /dev/null +++ b/parse/train/B1uvH_gC-/images/b3a79b4142cbc27ded70e8f2769057186f23e49766eedcd8367d541d1ead13a2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a8245a51916a5f85bac77f5fe1ab281a24d934e07487b2f3a4b5a40cf5150a75 +size 3641 diff --git a/parse/train/B1uvH_gC-/images/bf1f3693d156037b9601dfce3c9f001d7aa6afe83ffab9f332154499a012667d.jpg b/parse/train/B1uvH_gC-/images/bf1f3693d156037b9601dfce3c9f001d7aa6afe83ffab9f332154499a012667d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a64a2466c874714cb5d07841bf61ffb9d6a851b1 --- /dev/null +++ b/parse/train/B1uvH_gC-/images/bf1f3693d156037b9601dfce3c9f001d7aa6afe83ffab9f332154499a012667d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:11f27be9b5288bde025c91003d7ea874771c8bb528e1805cb3374fbf0ae7802e +size 5256 diff --git a/parse/train/B1uvH_gC-/images/df79e49165d87fb4c5d7086dd30330ff97ccaf5409b8afd18d41a28b479623e7.jpg b/parse/train/B1uvH_gC-/images/df79e49165d87fb4c5d7086dd30330ff97ccaf5409b8afd18d41a28b479623e7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ab15621b9a59723353b29e6d6e1cee667d09a043 --- /dev/null +++ b/parse/train/B1uvH_gC-/images/df79e49165d87fb4c5d7086dd30330ff97ccaf5409b8afd18d41a28b479623e7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:27b05c8e8681bfd66938dc9115047f4ba7866ee6999bc27c05354b3b9036bdd1 +size 87180 diff --git a/parse/train/B1uvH_gC-/images/fccb551294646faa8bba3229463a00918a7f61158c6c45d1db7f36f7ad8d630b.jpg b/parse/train/B1uvH_gC-/images/fccb551294646faa8bba3229463a00918a7f61158c6c45d1db7f36f7ad8d630b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4e33986b6cf9b4067da3b6d565e9a361d122d0be --- /dev/null +++ b/parse/train/B1uvH_gC-/images/fccb551294646faa8bba3229463a00918a7f61158c6c45d1db7f36f7ad8d630b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:df432557d789eaa9efd2b6b89087f43e811c2228b410ed011dc51c54118495d7 +size 43401 diff --git a/parse/train/B1x1ma4tDr/images/1e46d3e88077e81097613373fcc5a8b27c7c453ce169bee2d1882e6fb619eebc.jpg b/parse/train/B1x1ma4tDr/images/1e46d3e88077e81097613373fcc5a8b27c7c453ce169bee2d1882e6fb619eebc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f466cee0714fc1f8c9eec77a42d2dfa5e0cc68a8 --- /dev/null +++ b/parse/train/B1x1ma4tDr/images/1e46d3e88077e81097613373fcc5a8b27c7c453ce169bee2d1882e6fb619eebc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:530e2ba7ffdce218eac8788f554d587a0acbecf1bdb8026cbfa209fe9a5d4300 +size 39749 diff --git a/parse/train/B1x1ma4tDr/images/28fbf2379fdd01d9581cdab52dc4eb847082c1784e65dcbb804b6c4e60115aac.jpg b/parse/train/B1x1ma4tDr/images/28fbf2379fdd01d9581cdab52dc4eb847082c1784e65dcbb804b6c4e60115aac.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ba611a42f73d3742ccabdb8f1389e7395dbf3e3e --- /dev/null +++ b/parse/train/B1x1ma4tDr/images/28fbf2379fdd01d9581cdab52dc4eb847082c1784e65dcbb804b6c4e60115aac.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aad5c511e1fb0e563caa6fbdd16c83d2d2a4342df6b9f16f75e9828526607d0f +size 23732 diff --git a/parse/train/B1x1ma4tDr/images/29b6ce19f9423b3e8f0838181814129dac7a88c32b963d6b698b95a0e570425f.jpg b/parse/train/B1x1ma4tDr/images/29b6ce19f9423b3e8f0838181814129dac7a88c32b963d6b698b95a0e570425f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..aa4bebea4f12e5b860485c563e80803ef33cf6d2 --- /dev/null +++ b/parse/train/B1x1ma4tDr/images/29b6ce19f9423b3e8f0838181814129dac7a88c32b963d6b698b95a0e570425f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5b358e4d35f588a8d84fe98835d16c2956dac8cbcf5c046d92f31b7b1d75351d +size 47474 diff --git a/parse/train/B1x1ma4tDr/images/3581853ac178f33be8f9520b8d3ecfdb7ccd3ddd4adef98cc5b7a6863db827eb.jpg b/parse/train/B1x1ma4tDr/images/3581853ac178f33be8f9520b8d3ecfdb7ccd3ddd4adef98cc5b7a6863db827eb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ef6451c4fed03091793c263487f51ce5d2c140fd --- /dev/null +++ b/parse/train/B1x1ma4tDr/images/3581853ac178f33be8f9520b8d3ecfdb7ccd3ddd4adef98cc5b7a6863db827eb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4975a0310d3af21d117475764fce45fb5c7a127f01b19c94084aad861a24a9a7 +size 136170 diff --git a/parse/train/B1x1ma4tDr/images/3bc8490f22d6c211b143660796b5b285828ea45faf2047dba02b1d9417bbea5e.jpg b/parse/train/B1x1ma4tDr/images/3bc8490f22d6c211b143660796b5b285828ea45faf2047dba02b1d9417bbea5e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bf5461614aa8efb1b825c8f1a3e98e4a1a4ed03b --- /dev/null +++ b/parse/train/B1x1ma4tDr/images/3bc8490f22d6c211b143660796b5b285828ea45faf2047dba02b1d9417bbea5e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0a05e43db5e80a3a7b2769e180801eafe93150e3cbd58788adaca33e29a13bb5 +size 2905 diff --git a/parse/train/B1x1ma4tDr/images/414209bf77fffcbe6e4b5785b534f765efaa40fc9593c11d89c199f57b09dd0d.jpg b/parse/train/B1x1ma4tDr/images/414209bf77fffcbe6e4b5785b534f765efaa40fc9593c11d89c199f57b09dd0d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3bafa7151bb650f76b673620dedfa33f6a2fbb1c --- /dev/null +++ b/parse/train/B1x1ma4tDr/images/414209bf77fffcbe6e4b5785b534f765efaa40fc9593c11d89c199f57b09dd0d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bd8c1993f201e3cc34999303b09e6367c1a40d57622bae488d05004a08d924e3 +size 115358 diff --git a/parse/train/B1x1ma4tDr/images/8189f0a8c83496b77ed6f4363f98ed6c7f40bb280b987e643790218d286552d9.jpg b/parse/train/B1x1ma4tDr/images/8189f0a8c83496b77ed6f4363f98ed6c7f40bb280b987e643790218d286552d9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b25440efd5ae47d36205db54061dbd09d7e5a7ec --- /dev/null +++ b/parse/train/B1x1ma4tDr/images/8189f0a8c83496b77ed6f4363f98ed6c7f40bb280b987e643790218d286552d9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f39b76953e8ec37fc5b1b923e6979be13635465626e782e2e9ef2793f7fc5378 +size 5047 diff --git a/parse/train/B1x1ma4tDr/images/855c51131f1c81de60a033cdab15dea61c347a2213820e4eb8131484c53c7d92.jpg b/parse/train/B1x1ma4tDr/images/855c51131f1c81de60a033cdab15dea61c347a2213820e4eb8131484c53c7d92.jpg new file mode 100644 index 0000000000000000000000000000000000000000..38e4deb49359d9c790952b29285d58517afc7122 --- /dev/null +++ b/parse/train/B1x1ma4tDr/images/855c51131f1c81de60a033cdab15dea61c347a2213820e4eb8131484c53c7d92.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3f2d4be6cbc227ba3891d7fabe17a69f4811df3bcf7bae69b021a60378cbdc88 +size 95236 diff --git a/parse/train/B1x1ma4tDr/images/95409b8633e035a724bc68d02ab2d1992b50edd4cb5bfa6097ab38f71d0db2b2.jpg b/parse/train/B1x1ma4tDr/images/95409b8633e035a724bc68d02ab2d1992b50edd4cb5bfa6097ab38f71d0db2b2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7d3cd49580166ddd2c7a7fc08ce1da3d9c105640 --- /dev/null +++ b/parse/train/B1x1ma4tDr/images/95409b8633e035a724bc68d02ab2d1992b50edd4cb5bfa6097ab38f71d0db2b2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:337c5b10fcfe0240a2354dd04505ec4048e488d468f88f98d9bc0fefd316b0a9 +size 4200 diff --git a/parse/train/B1x1ma4tDr/images/ae69231c516ed09fed6808a0b174b09d4bc195050f2cd9f0a31748d0ba579fdf.jpg b/parse/train/B1x1ma4tDr/images/ae69231c516ed09fed6808a0b174b09d4bc195050f2cd9f0a31748d0ba579fdf.jpg new file mode 100644 index 0000000000000000000000000000000000000000..367224e96a7c4c129b99eb7948cfcb64ccb44275 --- /dev/null +++ b/parse/train/B1x1ma4tDr/images/ae69231c516ed09fed6808a0b174b09d4bc195050f2cd9f0a31748d0ba579fdf.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a2564bf4347237f683d42f4a2f36d669f360fa292d50352910c8176ac2327ad5 +size 167363 diff --git a/parse/train/B1x1ma4tDr/images/b3faeae273ee3875acc36bee50e5284123271a04e2aaa285692bb408b6dac899.jpg b/parse/train/B1x1ma4tDr/images/b3faeae273ee3875acc36bee50e5284123271a04e2aaa285692bb408b6dac899.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9e117a863b5e00407c4e2bba8bc1599988ba502e --- /dev/null +++ b/parse/train/B1x1ma4tDr/images/b3faeae273ee3875acc36bee50e5284123271a04e2aaa285692bb408b6dac899.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3b027c74824bc6affd45d790311a4618d24c2cdb47025c58b2ed01299bd2abef +size 47939 diff --git a/parse/train/B1x1ma4tDr/images/bbcebb94dbb9524aceff36f3d790224d20b0aa25c1c4063781359c32a7a78840.jpg b/parse/train/B1x1ma4tDr/images/bbcebb94dbb9524aceff36f3d790224d20b0aa25c1c4063781359c32a7a78840.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f3c99090e648249a41c7d669b231cc4abcaa12bd --- /dev/null +++ b/parse/train/B1x1ma4tDr/images/bbcebb94dbb9524aceff36f3d790224d20b0aa25c1c4063781359c32a7a78840.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7fe9dc145a05e3dd3dade8679a5d02fc28fcf23376f3bff036f2b4c3a9410088 +size 46246 diff --git a/parse/train/B1x1ma4tDr/images/bd88ff9e47ccff3eb0e42328cb0a8c160fa622e73f91acb33d768d9b8538c122.jpg b/parse/train/B1x1ma4tDr/images/bd88ff9e47ccff3eb0e42328cb0a8c160fa622e73f91acb33d768d9b8538c122.jpg new file mode 100644 index 0000000000000000000000000000000000000000..59058d7151ac63fab64178168cfe02401cf71431 --- /dev/null +++ b/parse/train/B1x1ma4tDr/images/bd88ff9e47ccff3eb0e42328cb0a8c160fa622e73f91acb33d768d9b8538c122.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0f13ae5181dc3019585721fd668cfd4dfbe1ec17bbc6f93e0f0217a7854135ff +size 31700 diff --git a/parse/train/B1x1ma4tDr/images/c92111047211e6624c3c18178c25de728271a9d73876307e26e7198c28563b9e.jpg b/parse/train/B1x1ma4tDr/images/c92111047211e6624c3c18178c25de728271a9d73876307e26e7198c28563b9e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dfedea74e88cce8311fef65effbdcfbb6ca983a1 --- /dev/null +++ b/parse/train/B1x1ma4tDr/images/c92111047211e6624c3c18178c25de728271a9d73876307e26e7198c28563b9e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:51faa6c3d3b1043f62d7e4178b1da8282df539cf16a29b42f8b249c0b33b6604 +size 118859 diff --git a/parse/train/B1x1ma4tDr/images/c95d4ee55979ba19b01ae62dfb5a5d8253c9951d7da9f72c9d3948df4ce205b9.jpg b/parse/train/B1x1ma4tDr/images/c95d4ee55979ba19b01ae62dfb5a5d8253c9951d7da9f72c9d3948df4ce205b9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..08108242196128bfa8930d89e0179d05c8a59a9e --- /dev/null +++ b/parse/train/B1x1ma4tDr/images/c95d4ee55979ba19b01ae62dfb5a5d8253c9951d7da9f72c9d3948df4ce205b9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d2413daa47b9411be6f62b4455ee164efc5aace5252f616e6bd814f3e45af17c +size 22049 diff --git a/parse/train/B1x1ma4tDr/images/ce1f3b59584b9a627b918133ad3040953912b3a4848fa1f6c5f5ee4c06b20c05.jpg b/parse/train/B1x1ma4tDr/images/ce1f3b59584b9a627b918133ad3040953912b3a4848fa1f6c5f5ee4c06b20c05.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4bcf7bc0ff56c6396a05cead7f304923b6713cc8 --- /dev/null +++ b/parse/train/B1x1ma4tDr/images/ce1f3b59584b9a627b918133ad3040953912b3a4848fa1f6c5f5ee4c06b20c05.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7ac9b0f0cb6626dd853f75922029c28332da74171a16c32f046161444877ef7b +size 46152 diff --git a/parse/train/B1x1ma4tDr/images/f2f0d286056bf63409e00e2505c28e1a83f0f8f4bf3231325c3cd798a3824c35.jpg b/parse/train/B1x1ma4tDr/images/f2f0d286056bf63409e00e2505c28e1a83f0f8f4bf3231325c3cd798a3824c35.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7b28ab3904dbc0e3ccbdde4f776bfcc6a95ada2a --- /dev/null +++ b/parse/train/B1x1ma4tDr/images/f2f0d286056bf63409e00e2505c28e1a83f0f8f4bf3231325c3cd798a3824c35.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8ab068db453c43c3e42541cdb3b6a23f0bc362d48947bc4866e199dc8f2f033e +size 33455 diff --git a/parse/train/B1x1ma4tDr/images/fa896847d87570431cc38b9224ca4abe27a8aff48a34596545edbbaff14b57f9.jpg b/parse/train/B1x1ma4tDr/images/fa896847d87570431cc38b9224ca4abe27a8aff48a34596545edbbaff14b57f9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..64a784bd2812d75ea551f5e1f048595ac537077c --- /dev/null +++ b/parse/train/B1x1ma4tDr/images/fa896847d87570431cc38b9224ca4abe27a8aff48a34596545edbbaff14b57f9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:42062ed8cbc1a05ad23dcccfdd4a70279f1416c3341d8489dff965d1318e5316 +size 4805 diff --git a/parse/train/B1x1ma4tDr/images/fc2a75c6922a8612960c274a7c544c93dd9219f3d957c450a976e733f97cde88.jpg b/parse/train/B1x1ma4tDr/images/fc2a75c6922a8612960c274a7c544c93dd9219f3d957c450a976e733f97cde88.jpg new file mode 100644 index 0000000000000000000000000000000000000000..789196e0a1216de9bacdda43ae8725aa9ceb6a1a --- /dev/null +++ b/parse/train/B1x1ma4tDr/images/fc2a75c6922a8612960c274a7c544c93dd9219f3d957c450a976e733f97cde88.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7cb0cb002936f183af1e6388f7bf8e55e9e5133c77c9e66f9657f794115456bc +size 4893 diff --git a/parse/train/BJJLHbb0-/images/00488ae9e10e591305ee3df5338686621163e4481b8afb5515b5d05d0dad1f97.jpg b/parse/train/BJJLHbb0-/images/00488ae9e10e591305ee3df5338686621163e4481b8afb5515b5d05d0dad1f97.jpg new file mode 100644 index 0000000000000000000000000000000000000000..df8c40df7f9b82dfa3fc9e04b2fab85b39ddd2ac --- /dev/null +++ b/parse/train/BJJLHbb0-/images/00488ae9e10e591305ee3df5338686621163e4481b8afb5515b5d05d0dad1f97.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b6ae1f56cb3c6b9b0b4d230f032cac164c250ec1b7c6a8ae630b2f67d4d810ea +size 38294 diff --git a/parse/train/BJJLHbb0-/images/01476dee6fb6173f47e652d610207f68da7a7ad4e9f50b777965753c5e392a46.jpg b/parse/train/BJJLHbb0-/images/01476dee6fb6173f47e652d610207f68da7a7ad4e9f50b777965753c5e392a46.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2a8d2cbeaf3d67b5778ce740fcb120648662ade6 --- /dev/null +++ b/parse/train/BJJLHbb0-/images/01476dee6fb6173f47e652d610207f68da7a7ad4e9f50b777965753c5e392a46.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aecd51730ccbdba2f7d06b56691b5be9f443afea09f2f223c3f6388e9a39837b +size 161978 diff --git a/parse/train/BJJLHbb0-/images/019a6125eb91a43fe564c91298ff6e8c2feb2998e909b36210c60432970b0c8f.jpg b/parse/train/BJJLHbb0-/images/019a6125eb91a43fe564c91298ff6e8c2feb2998e909b36210c60432970b0c8f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..57704d66136fd9ddcc347afb4feb97de2067a881 --- /dev/null +++ b/parse/train/BJJLHbb0-/images/019a6125eb91a43fe564c91298ff6e8c2feb2998e909b36210c60432970b0c8f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:06de2b8efb53720bcba47c60854bda93153cae138bdd63e739d3c39c853119a3 +size 2032 diff --git a/parse/train/BJJLHbb0-/images/054fe8c2780ed7be0d24ccf7a3ddd4397c45d761aa64884e467777ab06daedad.jpg b/parse/train/BJJLHbb0-/images/054fe8c2780ed7be0d24ccf7a3ddd4397c45d761aa64884e467777ab06daedad.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2bbb55b8396ad8ccfe89d5a636f9f77e7557fab1 --- /dev/null +++ b/parse/train/BJJLHbb0-/images/054fe8c2780ed7be0d24ccf7a3ddd4397c45d761aa64884e467777ab06daedad.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ef613077d3860cde510299012a120e6d15b232ede91a1ccbd58ee086250ade6c +size 10698 diff --git a/parse/train/BJJLHbb0-/images/08697fe93b8e5f7b393d77f96dd6f6dcb211a8e98ad85b04ca22cc94b599ba7b.jpg b/parse/train/BJJLHbb0-/images/08697fe93b8e5f7b393d77f96dd6f6dcb211a8e98ad85b04ca22cc94b599ba7b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0d579e0d77d360e074951848dd62ac5cf2fa79aa --- /dev/null +++ b/parse/train/BJJLHbb0-/images/08697fe93b8e5f7b393d77f96dd6f6dcb211a8e98ad85b04ca22cc94b599ba7b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0ced120183ea81926ef38fd0c7e5aebcae325632c6c4d0633779b77a592f3ce3 +size 18491 diff --git a/parse/train/BJJLHbb0-/images/22f8cf84bfc6bd6a16223afffff49372745940ca6eafb1555e754848beb779a1.jpg b/parse/train/BJJLHbb0-/images/22f8cf84bfc6bd6a16223afffff49372745940ca6eafb1555e754848beb779a1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..651f4a7d94a9a5403b9823065261a7abb27d485f --- /dev/null +++ b/parse/train/BJJLHbb0-/images/22f8cf84bfc6bd6a16223afffff49372745940ca6eafb1555e754848beb779a1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:01827e462138328516f8e61f6cf72ac779ed07926bf5a44ee922afb2b839ef15 +size 116558 diff --git a/parse/train/BJJLHbb0-/images/24602db1ad87746646f5e11c6479d1558e34fbaafcfbb9ca2ff92fa2b1e5f647.jpg b/parse/train/BJJLHbb0-/images/24602db1ad87746646f5e11c6479d1558e34fbaafcfbb9ca2ff92fa2b1e5f647.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2469cd1b37d01ce08ac8f637058f3b296bce3c1b --- /dev/null +++ b/parse/train/BJJLHbb0-/images/24602db1ad87746646f5e11c6479d1558e34fbaafcfbb9ca2ff92fa2b1e5f647.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:75cf19325929b88ec9135b6400bb694b8866ac7bc70df97bd7cd6fcfbd0d940f +size 15625 diff --git a/parse/train/BJJLHbb0-/images/52b1337e6de5c48ef16a0847a07f7fe2325af035c184aac61872ee4990f89633.jpg b/parse/train/BJJLHbb0-/images/52b1337e6de5c48ef16a0847a07f7fe2325af035c184aac61872ee4990f89633.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dcc63e287fbca4687b0f53a47d41f154a53938d2 --- /dev/null +++ b/parse/train/BJJLHbb0-/images/52b1337e6de5c48ef16a0847a07f7fe2325af035c184aac61872ee4990f89633.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cb37f3d7688ca8c8c483628282516a00138510d0832d1f4b9529b95db04eb96b +size 5610 diff --git a/parse/train/BJJLHbb0-/images/53e33eb6f1f337bda9595ea9a996a2ff0bf6ad37c8a124f619a170c5e6139145.jpg b/parse/train/BJJLHbb0-/images/53e33eb6f1f337bda9595ea9a996a2ff0bf6ad37c8a124f619a170c5e6139145.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f32cd48e1cdcb11589fe00a21d895bb588e72d60 --- /dev/null +++ b/parse/train/BJJLHbb0-/images/53e33eb6f1f337bda9595ea9a996a2ff0bf6ad37c8a124f619a170c5e6139145.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:74402cf2921ac41944da71b08fc5420ab57c1c0693377dad0241d743be83830e +size 19437 diff --git a/parse/train/BJJLHbb0-/images/7ef60cc47735019dc414f87a92aee0b77d6144586f1c319c2b7a3b3309163a1c.jpg b/parse/train/BJJLHbb0-/images/7ef60cc47735019dc414f87a92aee0b77d6144586f1c319c2b7a3b3309163a1c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5644b06a089d92f1464ff6ac03a6978b379de6ae --- /dev/null +++ b/parse/train/BJJLHbb0-/images/7ef60cc47735019dc414f87a92aee0b77d6144586f1c319c2b7a3b3309163a1c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2645b0b8b086364a9b333e89bfba07a53c28894de19900dfc2320b543bd06ef9 +size 14187 diff --git a/parse/train/BJJLHbb0-/images/8f0d414422daa0d9e21d941997178e29a2a252c31f4275bf47e517796576cbd1.jpg b/parse/train/BJJLHbb0-/images/8f0d414422daa0d9e21d941997178e29a2a252c31f4275bf47e517796576cbd1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c31bcf6abaecbd42062beb92a1f81ac67a3eaab2 --- /dev/null +++ b/parse/train/BJJLHbb0-/images/8f0d414422daa0d9e21d941997178e29a2a252c31f4275bf47e517796576cbd1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:635221fa8e537be1c4ff1ec0fa794a4ecf66ddb0a586f7190398d5577a39208f +size 43288 diff --git a/parse/train/BJJLHbb0-/images/aa416e889e7be576187a66646bc0d2fff197201b3e408dd7f04bdb44df553ddf.jpg b/parse/train/BJJLHbb0-/images/aa416e889e7be576187a66646bc0d2fff197201b3e408dd7f04bdb44df553ddf.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f47aa45fced75927a7d75e804c5412ecb14436c8 --- /dev/null +++ b/parse/train/BJJLHbb0-/images/aa416e889e7be576187a66646bc0d2fff197201b3e408dd7f04bdb44df553ddf.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e80269c35ae5b1c968674dbe18daeaca04a0966823ff51a89f4c307108c4a2ad +size 9677 diff --git a/parse/train/BJJLHbb0-/images/be8b2f11b2cdfe9b18c67497b1ab5676acf79fed9ddfef55e9623d3ef87dc691.jpg b/parse/train/BJJLHbb0-/images/be8b2f11b2cdfe9b18c67497b1ab5676acf79fed9ddfef55e9623d3ef87dc691.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c94a26d44382bf5e946bb23aff9c115def94df11 --- /dev/null +++ b/parse/train/BJJLHbb0-/images/be8b2f11b2cdfe9b18c67497b1ab5676acf79fed9ddfef55e9623d3ef87dc691.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:92146d69e6daaf2ae017a6e31d9044ae9dbcd64af5221a4bddece21aa83ec3cb +size 70410 diff --git a/parse/train/BJJLHbb0-/images/c3786f721b8b530becaeaafbf175cd8bdf6a60622bdd18be81fd17a1095b42b4.jpg b/parse/train/BJJLHbb0-/images/c3786f721b8b530becaeaafbf175cd8bdf6a60622bdd18be81fd17a1095b42b4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4867e89e15cd54154b2ef5b5f7f12941df8e4870 --- /dev/null +++ b/parse/train/BJJLHbb0-/images/c3786f721b8b530becaeaafbf175cd8bdf6a60622bdd18be81fd17a1095b42b4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eda648e26038a2fd039e90f785516cc3dd54030d3058c555351f40a9a258e97c +size 4686 diff --git a/parse/train/BJJLHbb0-/images/c5fa5510c891fc3392030409a6e0ef7d4a5ecfe39b5bbbf70a649f002a32f109.jpg b/parse/train/BJJLHbb0-/images/c5fa5510c891fc3392030409a6e0ef7d4a5ecfe39b5bbbf70a649f002a32f109.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cf992c3d039c0264d20f1724350a375524edfed4 --- /dev/null +++ b/parse/train/BJJLHbb0-/images/c5fa5510c891fc3392030409a6e0ef7d4a5ecfe39b5bbbf70a649f002a32f109.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ebf32decc33191c9d624c8769c8b1da815d69da046bbfb124cd25eec033775d9 +size 24970 diff --git a/parse/train/BJJLHbb0-/images/c6ea57f5c666816a0ba52ddea70b2a392baef9c580bb960f4208ab0723c2c832.jpg b/parse/train/BJJLHbb0-/images/c6ea57f5c666816a0ba52ddea70b2a392baef9c580bb960f4208ab0723c2c832.jpg new file mode 100644 index 0000000000000000000000000000000000000000..066793103b021e442cfc4343eb35037c30b8eee0 --- /dev/null +++ b/parse/train/BJJLHbb0-/images/c6ea57f5c666816a0ba52ddea70b2a392baef9c580bb960f4208ab0723c2c832.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c366ef423fdc89b503beef91e96b8f0216ae6fbe4e08385d9c67a77cb3a9bacb +size 13100 diff --git a/parse/train/BJJLHbb0-/images/e65557047a18c9460e3021ad775b4b67245c9a3f4276976ddf3c6a79da28c20d.jpg b/parse/train/BJJLHbb0-/images/e65557047a18c9460e3021ad775b4b67245c9a3f4276976ddf3c6a79da28c20d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..439dd80b6e0d0f7020ec04f69c7a076ad817fcf0 --- /dev/null +++ b/parse/train/BJJLHbb0-/images/e65557047a18c9460e3021ad775b4b67245c9a3f4276976ddf3c6a79da28c20d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe19713ba85129eb92b5d4bd46e2f9d6304eedede6699f569eb54f5c4b5c224b +size 125608 diff --git a/parse/train/BJJLHbb0-/images/f2ae569af0c77c073ceb2478d742fec9928a3302c6d134a81f5f79f2093db204.jpg b/parse/train/BJJLHbb0-/images/f2ae569af0c77c073ceb2478d742fec9928a3302c6d134a81f5f79f2093db204.jpg new file mode 100644 index 0000000000000000000000000000000000000000..25eb849a47b62d2050611a3824694d2a711ecf2e --- /dev/null +++ b/parse/train/BJJLHbb0-/images/f2ae569af0c77c073ceb2478d742fec9928a3302c6d134a81f5f79f2093db204.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:053c079c5995a805b5927fa0b13097111a2836db398e8ebad5b65f6566013433 +size 39471 diff --git a/parse/train/BJe0Gn0cY7/images/038152f49625e2418b8fda6762b57dcc4b175b0d8e4c65383f95dce344ca6b8b.jpg b/parse/train/BJe0Gn0cY7/images/038152f49625e2418b8fda6762b57dcc4b175b0d8e4c65383f95dce344ca6b8b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..339d645feadca22a7c9764a14b467dfe9b348515 --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/038152f49625e2418b8fda6762b57dcc4b175b0d8e4c65383f95dce344ca6b8b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:44899e1ca2eb36ce3e70ec88f4423df23be9420aa9c13a1c582b5e156498f230 +size 23094 diff --git a/parse/train/BJe0Gn0cY7/images/06219a66a5816c886971d7add498f2f3c5a71074b7ece491b8f0369b43337d97.jpg b/parse/train/BJe0Gn0cY7/images/06219a66a5816c886971d7add498f2f3c5a71074b7ece491b8f0369b43337d97.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e7ef8cd846c6d28aa5b307b195711e453fbe6f03 --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/06219a66a5816c886971d7add498f2f3c5a71074b7ece491b8f0369b43337d97.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6997945bdcf633cc878ce6179530d11625860b1312e43995964996456f8f71c7 +size 54571 diff --git a/parse/train/BJe0Gn0cY7/images/4a0a5a2676f25d88edd39c1b0ccb024824600748a8f6aa7a47b25f270f60fa99.jpg b/parse/train/BJe0Gn0cY7/images/4a0a5a2676f25d88edd39c1b0ccb024824600748a8f6aa7a47b25f270f60fa99.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4705ac56288b9c8cb5c0955a26ce2d51ad3534d5 --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/4a0a5a2676f25d88edd39c1b0ccb024824600748a8f6aa7a47b25f270f60fa99.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c76653151d56b7b055cca9fcbd186db3c9cd001474db6022b2595ef825f540b6 +size 82676 diff --git a/parse/train/BJe0Gn0cY7/images/5348af1b07de5fc5b3a7891e6ad85c035a314d13a82705329e080025fec46b58.jpg b/parse/train/BJe0Gn0cY7/images/5348af1b07de5fc5b3a7891e6ad85c035a314d13a82705329e080025fec46b58.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8fbce1209fc86bf8f0431c22c97601c275cd529d --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/5348af1b07de5fc5b3a7891e6ad85c035a314d13a82705329e080025fec46b58.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d9c61dfc7e54dba00158037faeca872505b91bc06349d87d0301c13c528d9c05 +size 244900 diff --git a/parse/train/BJe0Gn0cY7/images/551d1b269a53173cee4deacd556b51c09c1b43c7419475fc554116cbd7085f2f.jpg b/parse/train/BJe0Gn0cY7/images/551d1b269a53173cee4deacd556b51c09c1b43c7419475fc554116cbd7085f2f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dfbc9dfbce951f6a6f0056fcba5635ccaf17b64d --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/551d1b269a53173cee4deacd556b51c09c1b43c7419475fc554116cbd7085f2f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bae96040d0a69a408d3b93a969f980265ebdc23e9844f93d1854c56785066581 +size 8672 diff --git a/parse/train/BJe0Gn0cY7/images/5ecd201862786c6ad281b48bb97cdce0067c60d5498fe11249b820472fae4d40.jpg b/parse/train/BJe0Gn0cY7/images/5ecd201862786c6ad281b48bb97cdce0067c60d5498fe11249b820472fae4d40.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2160e98bae1e93ad0a5af234028a4c787fc996e8 --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/5ecd201862786c6ad281b48bb97cdce0067c60d5498fe11249b820472fae4d40.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0082afb3177973c4b1f087d77f140b579be0d2b83e04e795249d1a90aeec7f5c +size 209170 diff --git a/parse/train/BJe0Gn0cY7/images/671816185a2d6c36e43a2632b3490befa9c69c1eff615f07fc0c6f0ab425f7a8.jpg b/parse/train/BJe0Gn0cY7/images/671816185a2d6c36e43a2632b3490befa9c69c1eff615f07fc0c6f0ab425f7a8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7f49c221d5de941c79a16ac731ec4afaa32a53a5 --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/671816185a2d6c36e43a2632b3490befa9c69c1eff615f07fc0c6f0ab425f7a8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:051e601c9d839edbc60165495502c0ded6bf1d46930db15841682b680ed7e1ca +size 42233 diff --git a/parse/train/BJe0Gn0cY7/images/71e38d7855d8c1501cf4144a5362be1e10438aff96ece5ada6013e45b01cbc66.jpg b/parse/train/BJe0Gn0cY7/images/71e38d7855d8c1501cf4144a5362be1e10438aff96ece5ada6013e45b01cbc66.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ee6fcced47d1c7b1b242e5067c6d11861827d6a0 --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/71e38d7855d8c1501cf4144a5362be1e10438aff96ece5ada6013e45b01cbc66.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:16c01947145445699dd7ce33f321986d1e733fc64adde3d0ea291667cc7adfd1 +size 33484 diff --git a/parse/train/BJe0Gn0cY7/images/7bcc122ce616e62ecc30a0a9bec96f69fdba31944a989e5a1cca287f6e65eab5.jpg b/parse/train/BJe0Gn0cY7/images/7bcc122ce616e62ecc30a0a9bec96f69fdba31944a989e5a1cca287f6e65eab5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..157148228717e7ad288cd510ebf39976d5f3a990 --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/7bcc122ce616e62ecc30a0a9bec96f69fdba31944a989e5a1cca287f6e65eab5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3668d8fa867c04e8f957a6b9b9556ff34b85ba1b1952623f933c9859762a111d +size 18658 diff --git a/parse/train/BJe0Gn0cY7/images/7c0bb155c3047daf656b65088d01679e6e607736836d4db0e5c121341ded96d7.jpg b/parse/train/BJe0Gn0cY7/images/7c0bb155c3047daf656b65088d01679e6e607736836d4db0e5c121341ded96d7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6ba56e36ff0e9d30d2da98940bf9479103140dde --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/7c0bb155c3047daf656b65088d01679e6e607736836d4db0e5c121341ded96d7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1aa82000b4b33fed3853f344bf5b4d91913cdaf9152939750454a29619df930c +size 9725 diff --git a/parse/train/BJe0Gn0cY7/images/8f19b9bf2d2345d86d40e6cfe46ec016ebe0f2cfa216e2ce377003d34fc26896.jpg b/parse/train/BJe0Gn0cY7/images/8f19b9bf2d2345d86d40e6cfe46ec016ebe0f2cfa216e2ce377003d34fc26896.jpg new file mode 100644 index 0000000000000000000000000000000000000000..db7f345cb0b2fe4a3eee71d7edf7c4090d833a84 --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/8f19b9bf2d2345d86d40e6cfe46ec016ebe0f2cfa216e2ce377003d34fc26896.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a46f6b971434ba8e7ba74b51e27ee5ef8f708f92083e25a0b1c2514991423f62 +size 8548 diff --git a/parse/train/BJe0Gn0cY7/images/94d032f1c3325469b78beccc19909bac153bf0ccb50777b626fcea67e63afe26.jpg b/parse/train/BJe0Gn0cY7/images/94d032f1c3325469b78beccc19909bac153bf0ccb50777b626fcea67e63afe26.jpg new file mode 100644 index 0000000000000000000000000000000000000000..24b92eadd24d18b1bfb79695c3ee50db325344bd --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/94d032f1c3325469b78beccc19909bac153bf0ccb50777b626fcea67e63afe26.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1fced328c50cd3c7f4134df79aa8974f09aaa0c15ce9fec473e322bdd935da1a +size 218765 diff --git a/parse/train/BJe0Gn0cY7/images/96d9addc22e60e168ff5d1e7958bdff93343f2a45d2e61040a28536d05f6e151.jpg b/parse/train/BJe0Gn0cY7/images/96d9addc22e60e168ff5d1e7958bdff93343f2a45d2e61040a28536d05f6e151.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7b16f02ae0d499c979f67b2db945e8b8f00d034b --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/96d9addc22e60e168ff5d1e7958bdff93343f2a45d2e61040a28536d05f6e151.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9ca943a8f83cc726fc8302864087b0f05dea06d93bc5714af6c9829026bf96ca +size 167854 diff --git a/parse/train/BJe0Gn0cY7/images/98acb7250c925f66c1bb295850a0e41c41d98540ea4eb1d98d7e55b616a619cc.jpg b/parse/train/BJe0Gn0cY7/images/98acb7250c925f66c1bb295850a0e41c41d98540ea4eb1d98d7e55b616a619cc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..aacbb67ec069415ea2f8e34561c89e43838b17cb --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/98acb7250c925f66c1bb295850a0e41c41d98540ea4eb1d98d7e55b616a619cc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d3476b1337a6ea6c3c25c4a516129fc9159ceeb8cd6f57de8afe93e6a149fdd3 +size 13440 diff --git a/parse/train/BJe0Gn0cY7/images/9dbe6ec7dfda6a11f6aa94282585f10644166bfa9bbd00bb632b7f44733b7966.jpg b/parse/train/BJe0Gn0cY7/images/9dbe6ec7dfda6a11f6aa94282585f10644166bfa9bbd00bb632b7f44733b7966.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9d54599ec35b7da63edaa7b6559c9528a9dd521e --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/9dbe6ec7dfda6a11f6aa94282585f10644166bfa9bbd00bb632b7f44733b7966.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:767deb10cd6bdda8fad13ae68e4d1f5ea7421b0fccb39cc049c2fd09d210df89 +size 20395 diff --git a/parse/train/BJe0Gn0cY7/images/af605089075918355479017ac36446f1d9f26c86ed4b7fc0ed97b333cebe8394.jpg b/parse/train/BJe0Gn0cY7/images/af605089075918355479017ac36446f1d9f26c86ed4b7fc0ed97b333cebe8394.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8e4ed9403e0a90e8b4cd17a5e164d7fa81d2353f --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/af605089075918355479017ac36446f1d9f26c86ed4b7fc0ed97b333cebe8394.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e9d7b4a95cb4fffd6a75e7bfdb76a9fa6f68f8f3d17ed111fb33ba01ec81d16a +size 9022 diff --git a/parse/train/BJe0Gn0cY7/images/b26f0983da198a70769bb1211ad9b5e0fa2027048a9d6ed4a03ca919a3ba4d3f.jpg b/parse/train/BJe0Gn0cY7/images/b26f0983da198a70769bb1211ad9b5e0fa2027048a9d6ed4a03ca919a3ba4d3f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..12f2244c251dc5e298bdc1277925b37d6b34362f --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/b26f0983da198a70769bb1211ad9b5e0fa2027048a9d6ed4a03ca919a3ba4d3f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:db74e1e492a25f30cdf340f336e19d748b446d042b3f83303c13d4d2bb521d88 +size 3282 diff --git a/parse/train/BJe0Gn0cY7/images/b3b271e5a4791a7f22391abbc00b14a9ae7cc8526e73cf8d3a038b6cdd216136.jpg b/parse/train/BJe0Gn0cY7/images/b3b271e5a4791a7f22391abbc00b14a9ae7cc8526e73cf8d3a038b6cdd216136.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b082ac90db7cc23580c9881eb3316e67f9c4bd19 --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/b3b271e5a4791a7f22391abbc00b14a9ae7cc8526e73cf8d3a038b6cdd216136.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec045dc323b561a2831151654f352b112dc5498a1a3c04df6c178a99fad09df4 +size 81479 diff --git a/parse/train/BJe0Gn0cY7/images/b72e61eeb4c483028eab5b9b3fdca1eaf0f245c20c9835c46ff228876a8ef17f.jpg b/parse/train/BJe0Gn0cY7/images/b72e61eeb4c483028eab5b9b3fdca1eaf0f245c20c9835c46ff228876a8ef17f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..81855331f19ae9ef2b2b128e008335d600ce748b --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/b72e61eeb4c483028eab5b9b3fdca1eaf0f245c20c9835c46ff228876a8ef17f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:752a6110f5261c3eecd9834e69da7905e1bdc2d51b044e7655d511e601bb792d +size 27365 diff --git a/parse/train/BJe0Gn0cY7/images/c43e091d58617209bf9771f5ec9e22de2ca434a685c95e27aa564d701646c30c.jpg b/parse/train/BJe0Gn0cY7/images/c43e091d58617209bf9771f5ec9e22de2ca434a685c95e27aa564d701646c30c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a659b5c2faf0bbae58c20d5a3f8b90f54daf9359 --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/c43e091d58617209bf9771f5ec9e22de2ca434a685c95e27aa564d701646c30c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:85223e47c7a69f71b13efffe6e689b3047cfa056c3d3801b1383b9d90bcd1c36 +size 12407 diff --git a/parse/train/BJe0Gn0cY7/images/cab1e3f29f62a196b4d9af7603b18f8bcae19977160cc6b24a9ae2ecb506a7b3.jpg b/parse/train/BJe0Gn0cY7/images/cab1e3f29f62a196b4d9af7603b18f8bcae19977160cc6b24a9ae2ecb506a7b3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..eda4f260afed72240ab26f8b9613144a4dffa888 --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/cab1e3f29f62a196b4d9af7603b18f8bcae19977160cc6b24a9ae2ecb506a7b3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:db48d7a8dfe51a8400ce84e660711bdf7726fe66448b2ba335663aa829020301 +size 15644 diff --git a/parse/train/BJe0Gn0cY7/images/d90b7a61554fabf3756a2bec321b3c2fe76175f5e55b88f9e4c4a34dcb64b3f6.jpg b/parse/train/BJe0Gn0cY7/images/d90b7a61554fabf3756a2bec321b3c2fe76175f5e55b88f9e4c4a34dcb64b3f6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..07ee2a3ea3f702129e29e0c009b0a0b95866dcca --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/d90b7a61554fabf3756a2bec321b3c2fe76175f5e55b88f9e4c4a34dcb64b3f6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fceae30e54b561148ed44cf2061093539f00656ac1718a21a88fe07d37266f09 +size 253001 diff --git a/parse/train/BJe0Gn0cY7/images/d9ac3eb72cff0ab1952145c446d9c81ab402f04283b477e56ee5561a2cd312f5.jpg b/parse/train/BJe0Gn0cY7/images/d9ac3eb72cff0ab1952145c446d9c81ab402f04283b477e56ee5561a2cd312f5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..63d55a12455741558289ed6be69d9b527b936c2f --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/d9ac3eb72cff0ab1952145c446d9c81ab402f04283b477e56ee5561a2cd312f5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5cece3c8d1418fed707b43a7505a96459d9570916208d3204a920e0221301535 +size 35645 diff --git a/parse/train/BJe0Gn0cY7/images/dbb386a2b44f233f01291f7e774e9642b0cef23096f96e8aa5fc53bec1fe62c8.jpg b/parse/train/BJe0Gn0cY7/images/dbb386a2b44f233f01291f7e774e9642b0cef23096f96e8aa5fc53bec1fe62c8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6484599d424ab4eba3c783bb56d3698ba59230e7 --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/dbb386a2b44f233f01291f7e774e9642b0cef23096f96e8aa5fc53bec1fe62c8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:af6ec08a1fe465804ac22330e680fbbe8e47a0db6d670935c7a514c98e30f634 +size 155813 diff --git a/parse/train/BJe0Gn0cY7/images/dce3025ccce1bacb8b5719487855bf56efc4f4d60f6f5c7b882ba9446dfdc888.jpg b/parse/train/BJe0Gn0cY7/images/dce3025ccce1bacb8b5719487855bf56efc4f4d60f6f5c7b882ba9446dfdc888.jpg new file mode 100644 index 0000000000000000000000000000000000000000..664f2155c1393a94418ca9be98d62e7aee3b5c9e --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/dce3025ccce1bacb8b5719487855bf56efc4f4d60f6f5c7b882ba9446dfdc888.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:098eb5ec42ce602c4ab506e8b162f6c45fa1a2bd6f4725b2f060104a10e2fd96 +size 18448 diff --git a/parse/train/BJe0Gn0cY7/images/e5338177f56634b35fb001fd21122b31e3d521eb39f8ac0e56250f18ec418ac2.jpg b/parse/train/BJe0Gn0cY7/images/e5338177f56634b35fb001fd21122b31e3d521eb39f8ac0e56250f18ec418ac2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cd349cd648abcfb26d5f0af63cbe66df3c893239 --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/e5338177f56634b35fb001fd21122b31e3d521eb39f8ac0e56250f18ec418ac2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0b26cdd801455c61204cc63d5b21033794ccf609357f858644d19793f01f0077 +size 134938 diff --git a/parse/train/BJe0Gn0cY7/images/e5f600e3710989565574d1d29fed7d3994bf9a4e038d6de369a7826c88d58a9c.jpg b/parse/train/BJe0Gn0cY7/images/e5f600e3710989565574d1d29fed7d3994bf9a4e038d6de369a7826c88d58a9c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b217ce68333fe36d5174b7aa0bade2627b3da34a --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/e5f600e3710989565574d1d29fed7d3994bf9a4e038d6de369a7826c88d58a9c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8bfdbb59d4345d296c0650c03bc241f8ee60fab52c26878732410561503ee758 +size 27851 diff --git a/parse/train/BJe0Gn0cY7/images/ea4c6f6e7f7c7594da3267dc9d65daa9b0e1e437c3b0b9046985c9964175169f.jpg b/parse/train/BJe0Gn0cY7/images/ea4c6f6e7f7c7594da3267dc9d65daa9b0e1e437c3b0b9046985c9964175169f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..19914a9a888b64095602fa980d692692c6f8f9a0 --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/ea4c6f6e7f7c7594da3267dc9d65daa9b0e1e437c3b0b9046985c9964175169f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8c8f5a3c0a81f9964520e6223483b2790d5dc40696a85fc474e53b623ca2223d +size 140394 diff --git a/parse/train/BJe0Gn0cY7/images/fe417643a99a0ceda3e0b4be6a03a883620491ab77a54003ab692a9eb399ee73.jpg b/parse/train/BJe0Gn0cY7/images/fe417643a99a0ceda3e0b4be6a03a883620491ab77a54003ab692a9eb399ee73.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4fb50945c225aa6b9ab7bc2f75d1b0570d0327c8 --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/fe417643a99a0ceda3e0b4be6a03a883620491ab77a54003ab692a9eb399ee73.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:37e32993f60cc25eed5d4cb18762be6b97a50e845087473a80300f82e145a317 +size 63630 diff --git a/parse/train/BJe0Gn0cY7/images/fe79f38bcd5ebf6ba95a1205018c0e16f271d4160b1f635c60abb0f90264ec85.jpg b/parse/train/BJe0Gn0cY7/images/fe79f38bcd5ebf6ba95a1205018c0e16f271d4160b1f635c60abb0f90264ec85.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d1d7a06b6e7e2efda1c7ed47c15441c66c559571 --- /dev/null +++ b/parse/train/BJe0Gn0cY7/images/fe79f38bcd5ebf6ba95a1205018c0e16f271d4160b1f635c60abb0f90264ec85.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3cca5c7195a018f654923c0881e0394326d0c801a72b9d8394942c9a1c18377a +size 3634 diff --git a/parse/train/BJfvAoC9YQ/images/1afed3c469f6a2aa9ffe16c64f90472cd893234e18c870f3a25386ef2e62510e.jpg b/parse/train/BJfvAoC9YQ/images/1afed3c469f6a2aa9ffe16c64f90472cd893234e18c870f3a25386ef2e62510e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..db95da267b91bac8da86865c00034655107ce0f2 --- /dev/null +++ b/parse/train/BJfvAoC9YQ/images/1afed3c469f6a2aa9ffe16c64f90472cd893234e18c870f3a25386ef2e62510e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:989f35188e6be23851aa9b2831d70a3bf99814f9ea18462c447b1f236428222c +size 6344 diff --git a/parse/train/BJfvAoC9YQ/images/2bc1fea6dad7e304097258309b2d6b33340b7655e85e0d14fc19c29cba7ff726.jpg b/parse/train/BJfvAoC9YQ/images/2bc1fea6dad7e304097258309b2d6b33340b7655e85e0d14fc19c29cba7ff726.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c1d88dc14820c959393776619013018857e3b800 --- /dev/null +++ b/parse/train/BJfvAoC9YQ/images/2bc1fea6dad7e304097258309b2d6b33340b7655e85e0d14fc19c29cba7ff726.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ef292e0ea6da824ab2826752804bd0ba32685512990cffea416c4de9fff79987 +size 24514 diff --git a/parse/train/BJfvAoC9YQ/images/3511fb12edf5b80024a9081dfa0fad6b2357699e2a5a5e4a522714ddc71373ed.jpg b/parse/train/BJfvAoC9YQ/images/3511fb12edf5b80024a9081dfa0fad6b2357699e2a5a5e4a522714ddc71373ed.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e9de055a376e38fc1b838bb695e66735a2c30305 --- /dev/null +++ b/parse/train/BJfvAoC9YQ/images/3511fb12edf5b80024a9081dfa0fad6b2357699e2a5a5e4a522714ddc71373ed.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:15b03d5d9c22ccfbc4fc1e3592572576cbc7bd7bfb503f44f85d36808ecdbfa6 +size 48892 diff --git a/parse/train/BJfvAoC9YQ/images/3d3fc2b7ae0a2069c6d36ecabf2ad8cf0df7754c23690485d1315b4768000a0d.jpg b/parse/train/BJfvAoC9YQ/images/3d3fc2b7ae0a2069c6d36ecabf2ad8cf0df7754c23690485d1315b4768000a0d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..90bf7a022ad71b893e8434c7f60f567e0dac6ee2 --- /dev/null +++ b/parse/train/BJfvAoC9YQ/images/3d3fc2b7ae0a2069c6d36ecabf2ad8cf0df7754c23690485d1315b4768000a0d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b5ae63050ca6a0e331bbf77e9c36196684e5067f3e785e6fd862f786d3dbcb95 +size 71162 diff --git a/parse/train/BJfvAoC9YQ/images/4162168d34baeb3bcd95a8c308df9930fbcde7ccadbffd0465c37f7ccda9786a.jpg b/parse/train/BJfvAoC9YQ/images/4162168d34baeb3bcd95a8c308df9930fbcde7ccadbffd0465c37f7ccda9786a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6f364de5633d0b6a1439ecef78310c75d1bea449 --- /dev/null +++ b/parse/train/BJfvAoC9YQ/images/4162168d34baeb3bcd95a8c308df9930fbcde7ccadbffd0465c37f7ccda9786a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:733374e71404bf5c4d00d9d2742c45a8954867185c4e2716bc30156fde3f0523 +size 41067 diff --git a/parse/train/BJfvAoC9YQ/images/735a3683e1dc37162cbb0ec6c0c1801b5c67abba7ae0bd0a350ba32955730489.jpg b/parse/train/BJfvAoC9YQ/images/735a3683e1dc37162cbb0ec6c0c1801b5c67abba7ae0bd0a350ba32955730489.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f371771a9ca8d562d4db356a3b8dda2b9558f444 --- /dev/null +++ b/parse/train/BJfvAoC9YQ/images/735a3683e1dc37162cbb0ec6c0c1801b5c67abba7ae0bd0a350ba32955730489.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4621606903d0fdf255183c09214355cebe25180fc97cffdbb9162d8d8afa1947 +size 5028 diff --git a/parse/train/BJfvAoC9YQ/images/7b8f4de1e5d4cdd04aac0b1d1733bb9cdee426a0236cea34c72a01f569a3bb20.jpg b/parse/train/BJfvAoC9YQ/images/7b8f4de1e5d4cdd04aac0b1d1733bb9cdee426a0236cea34c72a01f569a3bb20.jpg new file mode 100644 index 0000000000000000000000000000000000000000..97af42ab05e4d8d4bc58af21709f77553fec8ffe --- /dev/null +++ b/parse/train/BJfvAoC9YQ/images/7b8f4de1e5d4cdd04aac0b1d1733bb9cdee426a0236cea34c72a01f569a3bb20.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b325b00ae7c252c99fa0fb2a113f2a04253c18017be23073b7e01cb33b274ee1 +size 41990 diff --git a/parse/train/BJfvAoC9YQ/images/7b9d3b41b0eea5801cea2ce9fc75b0acb8c21713c77fd1bb01f36f11cbd513b7.jpg b/parse/train/BJfvAoC9YQ/images/7b9d3b41b0eea5801cea2ce9fc75b0acb8c21713c77fd1bb01f36f11cbd513b7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2586c5f972094c90f16b0bf4d8bfc94bb4b3947a --- /dev/null +++ b/parse/train/BJfvAoC9YQ/images/7b9d3b41b0eea5801cea2ce9fc75b0acb8c21713c77fd1bb01f36f11cbd513b7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d1da171d9a59ce8ce0ac64852754ebe9204fb445935c7bd0976f1a710bbd3aed +size 2611 diff --git a/parse/train/BJfvAoC9YQ/images/ae7aad5758c788f6134654dee55eefd87dbdc3c1c547a9c804b0359e7a35a7c9.jpg b/parse/train/BJfvAoC9YQ/images/ae7aad5758c788f6134654dee55eefd87dbdc3c1c547a9c804b0359e7a35a7c9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..811369087a9f9b4edde68871eece8bd4206af6c9 --- /dev/null +++ b/parse/train/BJfvAoC9YQ/images/ae7aad5758c788f6134654dee55eefd87dbdc3c1c547a9c804b0359e7a35a7c9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:17b063ed836ec8acb7e2bd6cc5c6e6cd43e5d9ccc62cbcf0b0da59115f8238b2 +size 24855 diff --git a/parse/train/BJfvAoC9YQ/images/afe65a48a7d026c78cb31a576326d313ae335537c6aebb4687667f91bccfc817.jpg b/parse/train/BJfvAoC9YQ/images/afe65a48a7d026c78cb31a576326d313ae335537c6aebb4687667f91bccfc817.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7d4ee01519d628b420c1d3a8a8c92030dfcd87b1 --- /dev/null +++ b/parse/train/BJfvAoC9YQ/images/afe65a48a7d026c78cb31a576326d313ae335537c6aebb4687667f91bccfc817.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7d8e63869732c7aa94bf8a244512b275528dead207f5123ec28e217ffb168fd4 +size 4588 diff --git a/parse/train/BJfvAoC9YQ/images/d52835a8a94c71dc3d23cf4c178d91f1de02f1faeda123f2125db3fafc2bf9c0.jpg b/parse/train/BJfvAoC9YQ/images/d52835a8a94c71dc3d23cf4c178d91f1de02f1faeda123f2125db3fafc2bf9c0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..388f4f3c67d2235afd40bea9183c894ebea693ef --- /dev/null +++ b/parse/train/BJfvAoC9YQ/images/d52835a8a94c71dc3d23cf4c178d91f1de02f1faeda123f2125db3fafc2bf9c0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ecfcfcf14ae59f4d1d1fb6497d01003cf312981c890cce485e0f530144c80502 +size 12952 diff --git a/parse/train/BJfvAoC9YQ/images/e88b4ee7bab6235c1fe3378164e12dbc1ef6f548f082c88428b8ed9969f4a626.jpg b/parse/train/BJfvAoC9YQ/images/e88b4ee7bab6235c1fe3378164e12dbc1ef6f548f082c88428b8ed9969f4a626.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4540fcd2c6464106e4026a711b129c7b8e8fda64 --- /dev/null +++ b/parse/train/BJfvAoC9YQ/images/e88b4ee7bab6235c1fe3378164e12dbc1ef6f548f082c88428b8ed9969f4a626.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8d900c12678671bba8a2cfc2b139289ed03de4f262017a936f99579728eac304 +size 5697 diff --git a/parse/train/BJfvAoC9YQ/images/e9e4729c16bb88e9211864d8dedd4ae44730c20f3d20205788059608a6c598be.jpg b/parse/train/BJfvAoC9YQ/images/e9e4729c16bb88e9211864d8dedd4ae44730c20f3d20205788059608a6c598be.jpg new file mode 100644 index 0000000000000000000000000000000000000000..88539074d57778adc73d4317119dc84e992053da --- /dev/null +++ b/parse/train/BJfvAoC9YQ/images/e9e4729c16bb88e9211864d8dedd4ae44730c20f3d20205788059608a6c598be.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7efeb59c2f6109d9d3e24eeaad2f1ddea3b49d5636b043ac827f727484c90995 +size 26231 diff --git a/parse/train/BJfvAoC9YQ/images/ed66f02a2b13bb94fb23cfd8e9f354637f53b5d0740ac508396f4608ac541ef6.jpg b/parse/train/BJfvAoC9YQ/images/ed66f02a2b13bb94fb23cfd8e9f354637f53b5d0740ac508396f4608ac541ef6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6278c4e462af3d66b72370ec02c40511bf1c74f3 --- /dev/null +++ b/parse/train/BJfvAoC9YQ/images/ed66f02a2b13bb94fb23cfd8e9f354637f53b5d0740ac508396f4608ac541ef6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b4754cd83781b8d53e60436f0eb4bafe72559a7e84fe43ac489c1f52ec5a35b7 +size 48520 diff --git a/parse/train/BJgza6VtPB/images/03979eb1832a55dff2d4982a15dfaed1b86a25a7cf279b10f11963b5118f22fa.jpg b/parse/train/BJgza6VtPB/images/03979eb1832a55dff2d4982a15dfaed1b86a25a7cf279b10f11963b5118f22fa.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bd0bd62c5792d87403872e59a9aaa4e25e69e63a --- /dev/null +++ b/parse/train/BJgza6VtPB/images/03979eb1832a55dff2d4982a15dfaed1b86a25a7cf279b10f11963b5118f22fa.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:11b9f8361aedd6464c0748e102b21aa39c9c8773df7e06fdcb0b93c4a8bedfa6 +size 40082 diff --git a/parse/train/BJgza6VtPB/images/10914e69247b6ef94c6969048906cf221d6ee9c071284e2da721b3a1bd8f5107.jpg b/parse/train/BJgza6VtPB/images/10914e69247b6ef94c6969048906cf221d6ee9c071284e2da721b3a1bd8f5107.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c3549708069101deea8e6bf9853eaac16a7fa38d --- /dev/null +++ b/parse/train/BJgza6VtPB/images/10914e69247b6ef94c6969048906cf221d6ee9c071284e2da721b3a1bd8f5107.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:13e01bed3b7fcf5455e1408e3c7554da792916b266586c1388d8467ae232acb1 +size 197703 diff --git a/parse/train/BJgza6VtPB/images/1173e4bf4ab67af100500e2503dc2e9212adadab335f4a7bf267867888552e58.jpg b/parse/train/BJgza6VtPB/images/1173e4bf4ab67af100500e2503dc2e9212adadab335f4a7bf267867888552e58.jpg new file mode 100644 index 0000000000000000000000000000000000000000..38348640ce02665acafcdd3947717d4a8cdbb1b9 --- /dev/null +++ b/parse/train/BJgza6VtPB/images/1173e4bf4ab67af100500e2503dc2e9212adadab335f4a7bf267867888552e58.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9f51e98445a71fa4099b35167b33578c52781c400b906be5307ab98ea063e4cd +size 21276 diff --git a/parse/train/BJgza6VtPB/images/1266ec0367a09686c0920c0bbaec46286e697474e8954caf2cf263765da54f5c.jpg b/parse/train/BJgza6VtPB/images/1266ec0367a09686c0920c0bbaec46286e697474e8954caf2cf263765da54f5c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..21ba975e5ff8fb7519230bb4f09cbdd86793aff4 --- /dev/null +++ b/parse/train/BJgza6VtPB/images/1266ec0367a09686c0920c0bbaec46286e697474e8954caf2cf263765da54f5c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ef6be5373e646bab1712b601157399633785866bec18c862f27d6140e0682cc4 +size 22821 diff --git a/parse/train/BJgza6VtPB/images/1f29a04568f394bf533ee075c77fc1d4392323e684e74043cba15e4eda162d36.jpg b/parse/train/BJgza6VtPB/images/1f29a04568f394bf533ee075c77fc1d4392323e684e74043cba15e4eda162d36.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3a6c0ad8be8dcd842739e1d7bb31c4a980a0b7e5 --- /dev/null +++ b/parse/train/BJgza6VtPB/images/1f29a04568f394bf533ee075c77fc1d4392323e684e74043cba15e4eda162d36.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3c47c42f3a1757dc1fead7ee993bb14bf38374eec22d5a218e3c978283a7427c +size 21857 diff --git a/parse/train/BJgza6VtPB/images/46129c38741476da1199214ef20fdc72b1efe38fee8ffa4ff9377da0724eebca.jpg b/parse/train/BJgza6VtPB/images/46129c38741476da1199214ef20fdc72b1efe38fee8ffa4ff9377da0724eebca.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8af0e21d84f9fa365118dbccb27a2421aef07d4d --- /dev/null +++ b/parse/train/BJgza6VtPB/images/46129c38741476da1199214ef20fdc72b1efe38fee8ffa4ff9377da0724eebca.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4c390353a420d9766d0001b0bc5b0adb68fcf09e579070fb76872ec836ce0afd +size 64752 diff --git a/parse/train/BJgza6VtPB/images/5f48d48a79b42b8cd8328a508fb1fd72a229c158d24512ded9834936e2917fa1.jpg b/parse/train/BJgza6VtPB/images/5f48d48a79b42b8cd8328a508fb1fd72a229c158d24512ded9834936e2917fa1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4503b4edf501e7c15f01d24925ac4b6475019969 --- /dev/null +++ b/parse/train/BJgza6VtPB/images/5f48d48a79b42b8cd8328a508fb1fd72a229c158d24512ded9834936e2917fa1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:11403fa3ef6653af170095a8a37532ef9854f5de53c5a2eec1a7d6ba8038398b +size 18481 diff --git a/parse/train/BJgza6VtPB/images/7783ec756cfca81f2f71dfaf99af0fd7cd6e0e60c09c85be046439ff58612863.jpg b/parse/train/BJgza6VtPB/images/7783ec756cfca81f2f71dfaf99af0fd7cd6e0e60c09c85be046439ff58612863.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b3188b21fb98fa1ce49af8b2e6bd622b77b7d4e4 --- /dev/null +++ b/parse/train/BJgza6VtPB/images/7783ec756cfca81f2f71dfaf99af0fd7cd6e0e60c09c85be046439ff58612863.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9819aa926e170932d4d8971a281323bbb725f71f578c6ee55962c0187e78b7f9 +size 118640 diff --git a/parse/train/BJgza6VtPB/images/7bce730181fe5aaafba34bd3ad5e2f996accb734a698be7bbbbd739eb3aeccb8.jpg b/parse/train/BJgza6VtPB/images/7bce730181fe5aaafba34bd3ad5e2f996accb734a698be7bbbbd739eb3aeccb8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a54216d302bf8118bb1c76f091151be4fb9e389e --- /dev/null +++ b/parse/train/BJgza6VtPB/images/7bce730181fe5aaafba34bd3ad5e2f996accb734a698be7bbbbd739eb3aeccb8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3774813b044195b5b8d4c0d77e83453a571efd41ed0afd5c4f270355aba7aa9d +size 27294 diff --git a/parse/train/BJgza6VtPB/images/7df6d3e94db16971ba70d4471797207693e054a0f66b11d6fd2ca5b90ef078dc.jpg b/parse/train/BJgza6VtPB/images/7df6d3e94db16971ba70d4471797207693e054a0f66b11d6fd2ca5b90ef078dc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ecafb6e93ed5b20155573eb5f3193e5cb91f62a1 --- /dev/null +++ b/parse/train/BJgza6VtPB/images/7df6d3e94db16971ba70d4471797207693e054a0f66b11d6fd2ca5b90ef078dc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b913e655343cfd1e6d78eaf80df1daca7af75751ed0a89af3af469a5414ca1d4 +size 119836 diff --git a/parse/train/BJgza6VtPB/images/839e61acf369a8825f1638bcac105d7ee77b6571c9cf07311ec9f339ee16922f.jpg b/parse/train/BJgza6VtPB/images/839e61acf369a8825f1638bcac105d7ee77b6571c9cf07311ec9f339ee16922f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c05c4e50dbddd020ceca47280aec4b9fcbb2d39e --- /dev/null +++ b/parse/train/BJgza6VtPB/images/839e61acf369a8825f1638bcac105d7ee77b6571c9cf07311ec9f339ee16922f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eab91035fc35add0bbdd04b9255bbe674785c8ce30aa7612c3e31f8e2b982aab +size 30942 diff --git a/parse/train/BJgza6VtPB/images/89a710e6ced3625a11e248fec87dfeea71ca3e21b27c78876d09e52ce012345c.jpg b/parse/train/BJgza6VtPB/images/89a710e6ced3625a11e248fec87dfeea71ca3e21b27c78876d09e52ce012345c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3c4cbfe80f84c56f93dc6938c47c4e100a8c5be3 --- /dev/null +++ b/parse/train/BJgza6VtPB/images/89a710e6ced3625a11e248fec87dfeea71ca3e21b27c78876d09e52ce012345c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:080b8b2984d2019477e5635a18c7b496dd20a81e5ea09d2df7aac240a7b3481f +size 25323 diff --git a/parse/train/BJgza6VtPB/images/91b6caa2e7dc167f7da5fd190947e5c3a2f25d7bbaf797c357f76e4c98efcc61.jpg b/parse/train/BJgza6VtPB/images/91b6caa2e7dc167f7da5fd190947e5c3a2f25d7bbaf797c357f76e4c98efcc61.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ee36ef1591e915dc7371efffa9ca4211e6f71398 --- /dev/null +++ b/parse/train/BJgza6VtPB/images/91b6caa2e7dc167f7da5fd190947e5c3a2f25d7bbaf797c357f76e4c98efcc61.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9b975ce57e5abc380f74758bb8b9c6950df1f45b3c191cd8e6c33b92aefb8ea2 +size 24066 diff --git a/parse/train/BJgza6VtPB/images/9a5cef161d9d8d9009c424aa2d14ac6e8e7752528ebe264d3187a33cdde8425d.jpg b/parse/train/BJgza6VtPB/images/9a5cef161d9d8d9009c424aa2d14ac6e8e7752528ebe264d3187a33cdde8425d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4d71130c5fcb9b7c7693fa3ad77531d4b280b6d2 --- /dev/null +++ b/parse/train/BJgza6VtPB/images/9a5cef161d9d8d9009c424aa2d14ac6e8e7752528ebe264d3187a33cdde8425d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89550b8eb04394c890ec0829b1fd19aba1996429696c92e82d8b40292ddcc544 +size 56254 diff --git a/parse/train/BJgza6VtPB/images/a41f8fdd0bb43f4aff4235d9f271b726d2fd4e85e871b25add79430078e4f5f2.jpg b/parse/train/BJgza6VtPB/images/a41f8fdd0bb43f4aff4235d9f271b726d2fd4e85e871b25add79430078e4f5f2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6c0beecbcbd8a2388f1af93e7859072aec989135 --- /dev/null +++ b/parse/train/BJgza6VtPB/images/a41f8fdd0bb43f4aff4235d9f271b726d2fd4e85e871b25add79430078e4f5f2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:23e8cace71df1e9a5c236664899a6ed519b996238d12cbc23a4c2cf5c7f5b04e +size 65448 diff --git a/parse/train/BJgza6VtPB/images/a5f5c1310fb139989b6308747ee70b69487e92be5d687cf13cbcaa9e9b223637.jpg b/parse/train/BJgza6VtPB/images/a5f5c1310fb139989b6308747ee70b69487e92be5d687cf13cbcaa9e9b223637.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6fdefb6c2e91e44a612f0103ecc527a1c5ccb9b7 --- /dev/null +++ b/parse/train/BJgza6VtPB/images/a5f5c1310fb139989b6308747ee70b69487e92be5d687cf13cbcaa9e9b223637.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f4d6cc118307e4c9c2a43763d912c77d486c32ff514808de284d3d9e1ebf2690 +size 204944 diff --git a/parse/train/BJgza6VtPB/images/b7064b3d792aa487cdb86e124aab6b0d4fd2ed0a5011f7737363b8aa1204e761.jpg b/parse/train/BJgza6VtPB/images/b7064b3d792aa487cdb86e124aab6b0d4fd2ed0a5011f7737363b8aa1204e761.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9a07763d30165cce083630813e1f851559b13f45 --- /dev/null +++ b/parse/train/BJgza6VtPB/images/b7064b3d792aa487cdb86e124aab6b0d4fd2ed0a5011f7737363b8aa1204e761.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:76cb10b5e9fd7a67c62c59df1ccc4fd3d34e84469864f1a80a710fe12d542fd1 +size 235364 diff --git a/parse/train/BJgza6VtPB/images/d55c79a1d1f9f898c849cce8e0d43792c67bfe8c58e44f0ab5463aca67fa5bae.jpg b/parse/train/BJgza6VtPB/images/d55c79a1d1f9f898c849cce8e0d43792c67bfe8c58e44f0ab5463aca67fa5bae.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d0fa1160a7a8b4fe4ba96b5b264dc29eac7896f6 --- /dev/null +++ b/parse/train/BJgza6VtPB/images/d55c79a1d1f9f898c849cce8e0d43792c67bfe8c58e44f0ab5463aca67fa5bae.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d9c24ad1640c97542f9db0613a98b065868f1d3403d3fba91ec270e7aa325829 +size 7316 diff --git a/parse/train/BJgza6VtPB/images/e309e212e69aaa8ed47938ec5c4c3ac817921d20d1d13903fecbf9d018e29679.jpg b/parse/train/BJgza6VtPB/images/e309e212e69aaa8ed47938ec5c4c3ac817921d20d1d13903fecbf9d018e29679.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6e873f148c934c782ac071cb27b7d96ec9ad19f7 --- /dev/null +++ b/parse/train/BJgza6VtPB/images/e309e212e69aaa8ed47938ec5c4c3ac817921d20d1d13903fecbf9d018e29679.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:11df3f58977e4150ab02ff6b264024cf53b3c126215ed121c95f0daa39f9e756 +size 27935 diff --git a/parse/train/BJgza6VtPB/images/f993cd4f5146404c29140048f94bea8ae21c0b90b29d10bb16d319d9b644b653.jpg b/parse/train/BJgza6VtPB/images/f993cd4f5146404c29140048f94bea8ae21c0b90b29d10bb16d319d9b644b653.jpg new file mode 100644 index 0000000000000000000000000000000000000000..49f8567680e2d0e795ee6f2f12b6f19a1262ada0 --- /dev/null +++ b/parse/train/BJgza6VtPB/images/f993cd4f5146404c29140048f94bea8ae21c0b90b29d10bb16d319d9b644b653.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f47aa56b18f5cff83e135ba10d51efe2ca150a215160d81ced6338544211fb72 +size 61360 diff --git a/parse/train/BkXmYfbAZ/images/18d0d580e527b38db1746dbbc59e20f9b3e1651f75947f3541ac1a077477e868.jpg b/parse/train/BkXmYfbAZ/images/18d0d580e527b38db1746dbbc59e20f9b3e1651f75947f3541ac1a077477e868.jpg new file mode 100644 index 0000000000000000000000000000000000000000..554a30bf82b522a99d9cd542faaa08362ffdea2c --- /dev/null +++ b/parse/train/BkXmYfbAZ/images/18d0d580e527b38db1746dbbc59e20f9b3e1651f75947f3541ac1a077477e868.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:36fbe68bafc1c0b87b3f8a273b19cc7e51b88d039843e54e56b3fc7ec44948b7 +size 47701 diff --git a/parse/train/BkXmYfbAZ/images/21a767761b1cb6a7e7c0f9310dcc2b1c92637821cee670a78c23cb5e3ca06823.jpg b/parse/train/BkXmYfbAZ/images/21a767761b1cb6a7e7c0f9310dcc2b1c92637821cee670a78c23cb5e3ca06823.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f4c8c0bb15d059ef0b6273945b18f9183d4fb148 --- /dev/null +++ b/parse/train/BkXmYfbAZ/images/21a767761b1cb6a7e7c0f9310dcc2b1c92637821cee670a78c23cb5e3ca06823.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe5103e7acb3733b5ee01b8f7ae4616c5c1a4160dca1502d3edb308f76cd8a66 +size 8041 diff --git a/parse/train/BkXmYfbAZ/images/37be84aa7f5a2e6f13e16f2ad46a4a5144deab0453ab219808c5e18d87231462.jpg b/parse/train/BkXmYfbAZ/images/37be84aa7f5a2e6f13e16f2ad46a4a5144deab0453ab219808c5e18d87231462.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dce6748bbd26fd779c8abc6f0cc24ed9efec3543 --- /dev/null +++ b/parse/train/BkXmYfbAZ/images/37be84aa7f5a2e6f13e16f2ad46a4a5144deab0453ab219808c5e18d87231462.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:25757fa6c5bf02bc4f419d466afc67523542f3dc70de3765b502802c01d865da +size 43886 diff --git a/parse/train/BkXmYfbAZ/images/4d79f4b06728c6f5ab294567efef7b0bf31e337a65184a7903c8677b4f8bc217.jpg b/parse/train/BkXmYfbAZ/images/4d79f4b06728c6f5ab294567efef7b0bf31e337a65184a7903c8677b4f8bc217.jpg new file mode 100644 index 0000000000000000000000000000000000000000..331685a68edd8a786a4d129cfb5e397963d699af --- /dev/null +++ b/parse/train/BkXmYfbAZ/images/4d79f4b06728c6f5ab294567efef7b0bf31e337a65184a7903c8677b4f8bc217.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1c48a04acd8e9b3efa2bda35e13d227582b0ceb8adf84c9b7066bb394c4534cd +size 8026 diff --git a/parse/train/BkXmYfbAZ/images/5500de8ff21b0ce07d1f224a6029810c53d475eb180bc037d711a660e64ecfba.jpg b/parse/train/BkXmYfbAZ/images/5500de8ff21b0ce07d1f224a6029810c53d475eb180bc037d711a660e64ecfba.jpg new file mode 100644 index 0000000000000000000000000000000000000000..599ab55c65c6e5c4469d2bf0c778794f5dff3368 --- /dev/null +++ b/parse/train/BkXmYfbAZ/images/5500de8ff21b0ce07d1f224a6029810c53d475eb180bc037d711a660e64ecfba.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d76091e760d5e5539d4fc010764b42c2e998769c2ffbf6fde70037714af2fc66 +size 16878 diff --git a/parse/train/BkXmYfbAZ/images/5a71757820af176e6415d72244b5e789480cc38d7c7799bc29e599a8f79fd31f.jpg b/parse/train/BkXmYfbAZ/images/5a71757820af176e6415d72244b5e789480cc38d7c7799bc29e599a8f79fd31f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..27661886d887fb1ee217cea94bf090fecec2aa82 --- /dev/null +++ b/parse/train/BkXmYfbAZ/images/5a71757820af176e6415d72244b5e789480cc38d7c7799bc29e599a8f79fd31f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e21bbc1a28dd7730cc7242858cd45c130fb229168d5ae2eaff88812c3ad61ddd +size 25429 diff --git a/parse/train/BkXmYfbAZ/images/61df4c7774e78b99c89495913e5fc23e2051681fa23dac0b88517b400947ce34.jpg b/parse/train/BkXmYfbAZ/images/61df4c7774e78b99c89495913e5fc23e2051681fa23dac0b88517b400947ce34.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a4c2a1e534c797e13376e3dbac2a27e35e61edd8 --- /dev/null +++ b/parse/train/BkXmYfbAZ/images/61df4c7774e78b99c89495913e5fc23e2051681fa23dac0b88517b400947ce34.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:76cb68dc6937f66b98047861e218e4414533b216a994ee2313c503461d0552dc +size 24327 diff --git a/parse/train/BkXmYfbAZ/images/6ba9d8cfcffdb4e2de4c3c575d034021fc80c2859d3f99c642cdfd43223526b2.jpg b/parse/train/BkXmYfbAZ/images/6ba9d8cfcffdb4e2de4c3c575d034021fc80c2859d3f99c642cdfd43223526b2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4949622ead6c0638fb79f7267df531530a80c28c --- /dev/null +++ b/parse/train/BkXmYfbAZ/images/6ba9d8cfcffdb4e2de4c3c575d034021fc80c2859d3f99c642cdfd43223526b2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:61c1e54dd9b60fcfbb053aaab37c6fe2b92e52d78f289833184e04d2a534a30a +size 7137 diff --git a/parse/train/BkXmYfbAZ/images/850e349ffb192a29d5b3083a0953bce9d02e9cbf83bfa99d956e730ace5670ed.jpg b/parse/train/BkXmYfbAZ/images/850e349ffb192a29d5b3083a0953bce9d02e9cbf83bfa99d956e730ace5670ed.jpg new file mode 100644 index 0000000000000000000000000000000000000000..77fe37aaa1f4512691191c619b15a6d3e98ebf4c --- /dev/null +++ b/parse/train/BkXmYfbAZ/images/850e349ffb192a29d5b3083a0953bce9d02e9cbf83bfa99d956e730ace5670ed.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7a73813c0b883f4b844ff897cfc1f80a6e9934a94916cda3c1b81021185c85f1 +size 41726 diff --git a/parse/train/BkXmYfbAZ/images/8a163597b5237975e574600ee9152314f2553070b68566499ba620bf43772a07.jpg b/parse/train/BkXmYfbAZ/images/8a163597b5237975e574600ee9152314f2553070b68566499ba620bf43772a07.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1d9fd323f2e0ca81234647168811cd741a190d5b --- /dev/null +++ b/parse/train/BkXmYfbAZ/images/8a163597b5237975e574600ee9152314f2553070b68566499ba620bf43772a07.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aa8f98368b11d73ff3438cc82dbb2d83e7e81307fea74376848045eee838e493 +size 23395 diff --git a/parse/train/BkXmYfbAZ/images/915028c7df35ad0e86d2d1636dbc3d170d9c746becebdfe2975a8aa48dad0684.jpg b/parse/train/BkXmYfbAZ/images/915028c7df35ad0e86d2d1636dbc3d170d9c746becebdfe2975a8aa48dad0684.jpg new file mode 100644 index 0000000000000000000000000000000000000000..825d4d0d74ef876341cd22adaf3392b4c19c514d --- /dev/null +++ b/parse/train/BkXmYfbAZ/images/915028c7df35ad0e86d2d1636dbc3d170d9c746becebdfe2975a8aa48dad0684.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe93adbafd63c58b1367b11a5458f8672b530874ef4908f463a3a94599bb3968 +size 9137 diff --git a/parse/train/BkXmYfbAZ/images/9b12ec1c0f5ad83779cd8c777dc2b416ea4556beb858d6c3e5488d3d10520472.jpg b/parse/train/BkXmYfbAZ/images/9b12ec1c0f5ad83779cd8c777dc2b416ea4556beb858d6c3e5488d3d10520472.jpg new file mode 100644 index 0000000000000000000000000000000000000000..49c176ef0ba8b6d572b5303e48d654c3c4649342 --- /dev/null +++ b/parse/train/BkXmYfbAZ/images/9b12ec1c0f5ad83779cd8c777dc2b416ea4556beb858d6c3e5488d3d10520472.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9b84e684a26073aac4ecf16137bb1c410b5df017ca53da2c439526f9c7e3549 +size 25532 diff --git a/parse/train/BkXmYfbAZ/images/a7d66cfdbadd41a5efb9572b41501711cad0a97cd95388c302c89c9ae5247552.jpg b/parse/train/BkXmYfbAZ/images/a7d66cfdbadd41a5efb9572b41501711cad0a97cd95388c302c89c9ae5247552.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2b183f8ac0802b139cb7a5b5e9f79edd41f83390 --- /dev/null +++ b/parse/train/BkXmYfbAZ/images/a7d66cfdbadd41a5efb9572b41501711cad0a97cd95388c302c89c9ae5247552.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b5ca4296ce2ba036d44e7cc49e22ef8b5fc4a4441630a7467e02a138d5f2772f +size 10387 diff --git a/parse/train/BkXmYfbAZ/images/be1ea50055988d5088c4103bfcec74edb6832241175098b4dc3e6efbf807bb82.jpg b/parse/train/BkXmYfbAZ/images/be1ea50055988d5088c4103bfcec74edb6832241175098b4dc3e6efbf807bb82.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fe0bbdf0eb68580b299c9a66e93bc39625eb7e3e --- /dev/null +++ b/parse/train/BkXmYfbAZ/images/be1ea50055988d5088c4103bfcec74edb6832241175098b4dc3e6efbf807bb82.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d4621b01a7ce9a5be424a847ee2b156db10ca8a4eb9651de833b1725023c5952 +size 44528 diff --git a/parse/train/BkXmYfbAZ/images/bfd065f8c6423bb009328ca630d4ab1d8a26e919da9bc074084e1621dabdf36f.jpg b/parse/train/BkXmYfbAZ/images/bfd065f8c6423bb009328ca630d4ab1d8a26e919da9bc074084e1621dabdf36f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..143e2df3c04b55a5799f7e22815975e6be3fd554 --- /dev/null +++ b/parse/train/BkXmYfbAZ/images/bfd065f8c6423bb009328ca630d4ab1d8a26e919da9bc074084e1621dabdf36f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:447862f468e75c3d3bac5ad10ca1bd1858b20aae8ec959e21dff2cc49e4e69e9 +size 63211 diff --git a/parse/train/BkXmYfbAZ/images/ca1cd671264823c426fca6d5ec48551748424ab17cb1c843af27c0cfcde350e2.jpg b/parse/train/BkXmYfbAZ/images/ca1cd671264823c426fca6d5ec48551748424ab17cb1c843af27c0cfcde350e2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f374d7d111c5f39dc716c73c0ea148338e3be7e5 --- /dev/null +++ b/parse/train/BkXmYfbAZ/images/ca1cd671264823c426fca6d5ec48551748424ab17cb1c843af27c0cfcde350e2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a05cad578e5f83e8dcfee7cca8b99312ed139251a6fb31b4f7e0d22d3fb41dae +size 3765 diff --git a/parse/train/BkXmYfbAZ/images/dd34428803ba213274e09e0464523d042fa8e980ec3ab038c741699fddd74bb8.jpg b/parse/train/BkXmYfbAZ/images/dd34428803ba213274e09e0464523d042fa8e980ec3ab038c741699fddd74bb8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ed0487647f5c604d064b04fed1ed79651459e7d6 --- /dev/null +++ b/parse/train/BkXmYfbAZ/images/dd34428803ba213274e09e0464523d042fa8e980ec3ab038c741699fddd74bb8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a7486c88e14eee83476d70758907b27b5ae356a5d3074fe84702852466c9610c +size 6688 diff --git a/parse/train/ByOvsIqeg/images/065bfaab774f5e272775601e0ded5149b3d4f916d5211a8b0b9d7bedee67b1a6.jpg b/parse/train/ByOvsIqeg/images/065bfaab774f5e272775601e0ded5149b3d4f916d5211a8b0b9d7bedee67b1a6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..67dd4a37095ca83426bfa8dc1269fde3d6fca8b4 --- /dev/null +++ b/parse/train/ByOvsIqeg/images/065bfaab774f5e272775601e0ded5149b3d4f916d5211a8b0b9d7bedee67b1a6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:75a8083471bf28a6d89c7d1cef145a6871e7fcba1287680710091a613f421eb5 +size 23691 diff --git a/parse/train/ByOvsIqeg/images/17a8de7e2c9668e8569c0870b897cfa109c50e828b27f0112380e7e9f2c698ac.jpg b/parse/train/ByOvsIqeg/images/17a8de7e2c9668e8569c0870b897cfa109c50e828b27f0112380e7e9f2c698ac.jpg new file mode 100644 index 0000000000000000000000000000000000000000..de456c4e6631b5ff6173ed406e253bc4d2b26ced --- /dev/null +++ b/parse/train/ByOvsIqeg/images/17a8de7e2c9668e8569c0870b897cfa109c50e828b27f0112380e7e9f2c698ac.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6b2e85dd881124ff2dfc0a8867c565004977f9f44215601d593979154ccadcf9 +size 14897 diff --git a/parse/train/ByOvsIqeg/images/20b6b25c909fc2eab03fabb2140b9d7206f3e02335eba49cfafba5a5b33360b6.jpg b/parse/train/ByOvsIqeg/images/20b6b25c909fc2eab03fabb2140b9d7206f3e02335eba49cfafba5a5b33360b6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fd3b0372e15ad7bc57b819e9cc142f0d6c451370 --- /dev/null +++ b/parse/train/ByOvsIqeg/images/20b6b25c909fc2eab03fabb2140b9d7206f3e02335eba49cfafba5a5b33360b6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b2845aa7096fdf761f6dfc4ec1d782e00f5889d8f3a425270c70baf9bf511bd1 +size 7288 diff --git a/parse/train/ByOvsIqeg/images/20bec678b951e9e7b03641aaf3c857d7cd8341e4ec34efc02f724b5578fe113b.jpg b/parse/train/ByOvsIqeg/images/20bec678b951e9e7b03641aaf3c857d7cd8341e4ec34efc02f724b5578fe113b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2d88ab35913f4059bbb096105a8cc56e947d2625 --- /dev/null +++ b/parse/train/ByOvsIqeg/images/20bec678b951e9e7b03641aaf3c857d7cd8341e4ec34efc02f724b5578fe113b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1289bc10a11440876904f03e6fbf231e2736ef39437482afc1550797996fea92 +size 113729 diff --git a/parse/train/ByOvsIqeg/images/274103e294a9b385d101077beda43ccb46196b1e4b5f4f46e0a2d67c28c9fbf0.jpg b/parse/train/ByOvsIqeg/images/274103e294a9b385d101077beda43ccb46196b1e4b5f4f46e0a2d67c28c9fbf0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2c0ee090bbc0ccd3ce3cef5dd7eb14670bfe2f8e --- /dev/null +++ b/parse/train/ByOvsIqeg/images/274103e294a9b385d101077beda43ccb46196b1e4b5f4f46e0a2d67c28c9fbf0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3c7eb71755cc4afaa236a48678b2d0b5a5faf35224daddedfd55bc97a690e8eb +size 54730 diff --git a/parse/train/ByOvsIqeg/images/34a3be8804f26a57d411d167237444d8d7958399d4ffb56568305dbab17fe2bf.jpg b/parse/train/ByOvsIqeg/images/34a3be8804f26a57d411d167237444d8d7958399d4ffb56568305dbab17fe2bf.jpg new file mode 100644 index 0000000000000000000000000000000000000000..21dc2670b69f53b4ceef8b536b2e0532eb212fd8 --- /dev/null +++ b/parse/train/ByOvsIqeg/images/34a3be8804f26a57d411d167237444d8d7958399d4ffb56568305dbab17fe2bf.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0a505819f8d4e70224e90a87fb3f264e8b76408fd8fc1efb165b98c704fbf932 +size 54862 diff --git a/parse/train/ByOvsIqeg/images/3e90e297d40d13ac2c2c0f92c4a8912832f7195d585c7a38dba64bb879631e3a.jpg b/parse/train/ByOvsIqeg/images/3e90e297d40d13ac2c2c0f92c4a8912832f7195d585c7a38dba64bb879631e3a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c0d02d1a4cc0b6a519572f4005219c32228b0ce9 --- /dev/null +++ b/parse/train/ByOvsIqeg/images/3e90e297d40d13ac2c2c0f92c4a8912832f7195d585c7a38dba64bb879631e3a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a6050a7f7aecd7724e12f82b52507f4570d66ab588707a41dc08f9f39a0347e8 +size 12310 diff --git a/parse/train/ByOvsIqeg/images/491d5d56bda3cbf9480511a3c0a3e95cb5510759803f8c107a3f374476de40d5.jpg b/parse/train/ByOvsIqeg/images/491d5d56bda3cbf9480511a3c0a3e95cb5510759803f8c107a3f374476de40d5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..93bec6c46049ae82cbef665e10362e0e666d57dd --- /dev/null +++ b/parse/train/ByOvsIqeg/images/491d5d56bda3cbf9480511a3c0a3e95cb5510759803f8c107a3f374476de40d5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7205dfde6abe282205926917bbfd56f38489bf3caeb8c43c1e9616b488377589 +size 33189 diff --git a/parse/train/ByOvsIqeg/images/49a3962fd72029ede29506440d7b9686e912007c445ec6dd483281d28db3427a.jpg b/parse/train/ByOvsIqeg/images/49a3962fd72029ede29506440d7b9686e912007c445ec6dd483281d28db3427a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ea4d6d5d3076b5adb4995a20aac1fbe418999d22 --- /dev/null +++ b/parse/train/ByOvsIqeg/images/49a3962fd72029ede29506440d7b9686e912007c445ec6dd483281d28db3427a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e8700137d1e4007787f81b71571730525fcea9ac0fc0b4b5affaac42d5be1a76 +size 2867 diff --git a/parse/train/ByOvsIqeg/images/532babe973466204da2fb90352d3b6f63c2648dd29dfdae42be577e68034a5cf.jpg b/parse/train/ByOvsIqeg/images/532babe973466204da2fb90352d3b6f63c2648dd29dfdae42be577e68034a5cf.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8a775cb66c570ad82fc635d00ee14806a147cba6 --- /dev/null +++ b/parse/train/ByOvsIqeg/images/532babe973466204da2fb90352d3b6f63c2648dd29dfdae42be577e68034a5cf.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bcc415d95a9861d321296fa8124834c0f66f038de339f5d9dc773bd2d83dbb39 +size 4569 diff --git a/parse/train/ByOvsIqeg/images/5bb071228d1b1a3361886f1d42e183e8d7cdb652a6a568d1501eee2f17856527.jpg b/parse/train/ByOvsIqeg/images/5bb071228d1b1a3361886f1d42e183e8d7cdb652a6a568d1501eee2f17856527.jpg new file mode 100644 index 0000000000000000000000000000000000000000..df4cd35ac7b87492205272b0026c4017d8553824 --- /dev/null +++ b/parse/train/ByOvsIqeg/images/5bb071228d1b1a3361886f1d42e183e8d7cdb652a6a568d1501eee2f17856527.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9498a9ecea7239dc9c7affc1bcfda299d47052396ed1b22acedf4959af3c726f +size 12391 diff --git a/parse/train/ByOvsIqeg/images/70aca7fdc358c77aa9410f4a211e33cdb81f3f50e010e29449d3abfb234d9d2b.jpg b/parse/train/ByOvsIqeg/images/70aca7fdc358c77aa9410f4a211e33cdb81f3f50e010e29449d3abfb234d9d2b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..928ba1e84677c9a6465b517aab152148a204229f --- /dev/null +++ b/parse/train/ByOvsIqeg/images/70aca7fdc358c77aa9410f4a211e33cdb81f3f50e010e29449d3abfb234d9d2b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:131acecf41a13280c8f8e408295d591b0e4fa83c334d72fae8356d0e07efae1f +size 11583 diff --git a/parse/train/ByOvsIqeg/images/797a937a123e416d547a9590449fdc1933d1fd97d9b389860168faf07f3ea464.jpg b/parse/train/ByOvsIqeg/images/797a937a123e416d547a9590449fdc1933d1fd97d9b389860168faf07f3ea464.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1eaac959e8e5051deb2b60e82e05aac5beefa0d2 --- /dev/null +++ b/parse/train/ByOvsIqeg/images/797a937a123e416d547a9590449fdc1933d1fd97d9b389860168faf07f3ea464.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b3ba37c59f56fb9b0e6dbc8ae8f0be34155642703bc600a2ed7a5b0f47c2ae35 +size 43350 diff --git a/parse/train/ByOvsIqeg/images/8167e8f760a41cbdc639d32db33dd260c6f4fafb739369db75799c5e05fa99d1.jpg b/parse/train/ByOvsIqeg/images/8167e8f760a41cbdc639d32db33dd260c6f4fafb739369db75799c5e05fa99d1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..88c4212ed4ebb5389ccd371f706ee8e15f2e27db --- /dev/null +++ b/parse/train/ByOvsIqeg/images/8167e8f760a41cbdc639d32db33dd260c6f4fafb739369db75799c5e05fa99d1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7a1afc7f1fa0e65d0ad8d9eaf1744c21ced7067450a7b3429b52867d3c4c9af4 +size 6123 diff --git a/parse/train/ByOvsIqeg/images/877c7b33e2d9d9821c556d990f6295780dbc89c1b5a332b34e506d15ebfc62df.jpg b/parse/train/ByOvsIqeg/images/877c7b33e2d9d9821c556d990f6295780dbc89c1b5a332b34e506d15ebfc62df.jpg new file mode 100644 index 0000000000000000000000000000000000000000..183d49a0f07146d5c6e4a97a1eeb676365827b1a --- /dev/null +++ b/parse/train/ByOvsIqeg/images/877c7b33e2d9d9821c556d990f6295780dbc89c1b5a332b34e506d15ebfc62df.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:927d6408a4e8b52315fa7f9f70f80d8ed18c5e8b1399ef075fba415fcf50020a +size 7919 diff --git a/parse/train/ByOvsIqeg/images/b80fa05b0669ea58488acd80091c8a2cc7afbf5e245d4e715d465ba566ef1ac6.jpg b/parse/train/ByOvsIqeg/images/b80fa05b0669ea58488acd80091c8a2cc7afbf5e245d4e715d465ba566ef1ac6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cf45cabc1c7c07d6a99c6c75d0b7377c77afbe3b --- /dev/null +++ b/parse/train/ByOvsIqeg/images/b80fa05b0669ea58488acd80091c8a2cc7afbf5e245d4e715d465ba566ef1ac6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fbcfb76cd4520f3527a504c9a6018fe70ba8851a7b1f2ad7f67f2298b3f8d857 +size 21546 diff --git a/parse/train/ByOvsIqeg/images/bd7b23b9c26025a4c3a4beaaaa85011185c26e6e789f3f3cb62ff6e2a71dbbdc.jpg b/parse/train/ByOvsIqeg/images/bd7b23b9c26025a4c3a4beaaaa85011185c26e6e789f3f3cb62ff6e2a71dbbdc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b37915533acb64a21ce9ce505169b528b513d10f --- /dev/null +++ b/parse/train/ByOvsIqeg/images/bd7b23b9c26025a4c3a4beaaaa85011185c26e6e789f3f3cb62ff6e2a71dbbdc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4e6d7d12bfb475057bab2a2d93ba4ac133adab943b1a37acb8914239b73bac03 +size 34108 diff --git a/parse/train/ByOvsIqeg/images/c1da0033dc46d5adef4d5dec2a33c9429bbd293b5beb236f6b474a97c5fb1c7d.jpg b/parse/train/ByOvsIqeg/images/c1da0033dc46d5adef4d5dec2a33c9429bbd293b5beb236f6b474a97c5fb1c7d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2acd096d76da222ca4672ca3e7f545a2b8a6e8bf --- /dev/null +++ b/parse/train/ByOvsIqeg/images/c1da0033dc46d5adef4d5dec2a33c9429bbd293b5beb236f6b474a97c5fb1c7d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:298fc0f749ab449c5e2189fbcf69ab3ed5041a2fc400149885634e5c57a2b067 +size 6247 diff --git a/parse/train/ByOvsIqeg/images/c8412b5ab42884c6f586a1bd08e2e7c84bef80594ea863fcd89db80465ea9b10.jpg b/parse/train/ByOvsIqeg/images/c8412b5ab42884c6f586a1bd08e2e7c84bef80594ea863fcd89db80465ea9b10.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6fa218357a7286baedaadd4519221f953f615c6e --- /dev/null +++ b/parse/train/ByOvsIqeg/images/c8412b5ab42884c6f586a1bd08e2e7c84bef80594ea863fcd89db80465ea9b10.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1bb8e715c5a1954e4cbce2b181d3a391a016ba07eb1eb3422d310fab9858c96a +size 26393 diff --git a/parse/train/ByOvsIqeg/images/d5c89f137ebdad2f4302c5935ff362f628cfc870fd05c573b7a9e6ca4094c997.jpg b/parse/train/ByOvsIqeg/images/d5c89f137ebdad2f4302c5935ff362f628cfc870fd05c573b7a9e6ca4094c997.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8f56cdacf793297e95cb99500ae85fbce1b5e482 --- /dev/null +++ b/parse/train/ByOvsIqeg/images/d5c89f137ebdad2f4302c5935ff362f628cfc870fd05c573b7a9e6ca4094c997.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2cf06cbb80d8f0364a5bc721663c55fe1cf13921ec0bb82dc6d5914e1835c18c +size 61914 diff --git a/parse/train/ByOvsIqeg/images/db865c1c4220222c1b67b84f3fc30d5f890ebdc3f6c3b02c90ac2cc04ef04939.jpg b/parse/train/ByOvsIqeg/images/db865c1c4220222c1b67b84f3fc30d5f890ebdc3f6c3b02c90ac2cc04ef04939.jpg new file mode 100644 index 0000000000000000000000000000000000000000..aa317b9197f3b9cb15c1e8755551880bdfbaf594 --- /dev/null +++ b/parse/train/ByOvsIqeg/images/db865c1c4220222c1b67b84f3fc30d5f890ebdc3f6c3b02c90ac2cc04ef04939.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:de3de6572a3dd0b08ae5c403fcb9e903e6c0a028c259bc38ef5b5a80501a3de3 +size 6088 diff --git a/parse/train/BybtVK9lg/images/0b564eed7a07777407715679da29f016aefb0b4c06830e989bfdd20b645ed04c.jpg b/parse/train/BybtVK9lg/images/0b564eed7a07777407715679da29f016aefb0b4c06830e989bfdd20b645ed04c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..544a8fb16aa7fb748fd9d850f321c838c7729ae2 --- /dev/null +++ b/parse/train/BybtVK9lg/images/0b564eed7a07777407715679da29f016aefb0b4c06830e989bfdd20b645ed04c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:01406c3f49c6505540a3113f603a823f7b39a36b999beaa7117e706f81cdb32b +size 165626 diff --git a/parse/train/BybtVK9lg/images/27e406dfae7c9013aa7a680260f1aaac40ca4ff80858f7b095d08fc0495f5ec9.jpg b/parse/train/BybtVK9lg/images/27e406dfae7c9013aa7a680260f1aaac40ca4ff80858f7b095d08fc0495f5ec9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0a5f4f3b57719971d67f0719ed5a0b0ffea3e883 --- /dev/null +++ b/parse/train/BybtVK9lg/images/27e406dfae7c9013aa7a680260f1aaac40ca4ff80858f7b095d08fc0495f5ec9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d302c184fe59b43a1494a3aaae44aed82db854cb5a2d99b0a5f959a1577740a9 +size 9647 diff --git a/parse/train/BybtVK9lg/images/32f0153706e016ea363eaee8e92f76d86349ba8c9b63fefd7539c67ae3d1c3fd.jpg b/parse/train/BybtVK9lg/images/32f0153706e016ea363eaee8e92f76d86349ba8c9b63fefd7539c67ae3d1c3fd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..422afe803f406845ecb63aeb566a63f85b91b795 --- /dev/null +++ b/parse/train/BybtVK9lg/images/32f0153706e016ea363eaee8e92f76d86349ba8c9b63fefd7539c67ae3d1c3fd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:90b3144cb528a95e46e8247590e22d07da3a098f9eadbf7726f785ad3eab45fc +size 25300 diff --git a/parse/train/BybtVK9lg/images/478ec434801f431d367eac1a83bcdaf4757050663a776ece48964711468074b3.jpg b/parse/train/BybtVK9lg/images/478ec434801f431d367eac1a83bcdaf4757050663a776ece48964711468074b3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..888fe5ae2ffbd1e9cb9612cb6ff8a893db2e430e --- /dev/null +++ b/parse/train/BybtVK9lg/images/478ec434801f431d367eac1a83bcdaf4757050663a776ece48964711468074b3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:62b28e4a5662ee686d8f68f4b2a83f68d3e2310452b6ddf6491c0c9ef02ad074 +size 24057 diff --git a/parse/train/BybtVK9lg/images/588b9bae662bacd8432f36c12f5b2fe78236165827b08830d0cd5411350801fb.jpg b/parse/train/BybtVK9lg/images/588b9bae662bacd8432f36c12f5b2fe78236165827b08830d0cd5411350801fb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ef4b1d5b52ed6dc45ac69cf17ae0ef24d89a0307 --- /dev/null +++ b/parse/train/BybtVK9lg/images/588b9bae662bacd8432f36c12f5b2fe78236165827b08830d0cd5411350801fb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6873c434d8ef1a4c0cb45c5eef32c93a4ce32d87c70488ca6bc20214c6b0b990 +size 24509 diff --git a/parse/train/BybtVK9lg/images/5f9c16482ad7805b66f4bd51ba15b596c3ea8388c1f980b2fca7d11ce53ed0c3.jpg b/parse/train/BybtVK9lg/images/5f9c16482ad7805b66f4bd51ba15b596c3ea8388c1f980b2fca7d11ce53ed0c3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b7511a99417895c90409862ee56134ad27e97ccc --- /dev/null +++ b/parse/train/BybtVK9lg/images/5f9c16482ad7805b66f4bd51ba15b596c3ea8388c1f980b2fca7d11ce53ed0c3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7a5b9f98798b80f10027409bbb47858ceca610ce7ba03fcbb5b83cffd51c6eab +size 23945 diff --git a/parse/train/BybtVK9lg/images/62a74e2e60b6d4d79d2c8f512d79244fc6cc959970f6623a85024f65746023e1.jpg b/parse/train/BybtVK9lg/images/62a74e2e60b6d4d79d2c8f512d79244fc6cc959970f6623a85024f65746023e1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9fc686d5ab513074d14bd4a93767ac2f5562c7ff --- /dev/null +++ b/parse/train/BybtVK9lg/images/62a74e2e60b6d4d79d2c8f512d79244fc6cc959970f6623a85024f65746023e1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d01c7b5c2478aa34153855e8d675423f42b3aa89f1151fe23a93335ec7977219 +size 9461 diff --git a/parse/train/BybtVK9lg/images/739c5c2afeba6028c8dc8a3876cc7e62cf223b081b151b4c71c000e9372c9a48.jpg b/parse/train/BybtVK9lg/images/739c5c2afeba6028c8dc8a3876cc7e62cf223b081b151b4c71c000e9372c9a48.jpg new file mode 100644 index 0000000000000000000000000000000000000000..152c041d35b1230879acbf12505131fdc1d06cd9 --- /dev/null +++ b/parse/train/BybtVK9lg/images/739c5c2afeba6028c8dc8a3876cc7e62cf223b081b151b4c71c000e9372c9a48.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e0fda05f28d7c41a5da6d3204f806e5753aa168e07592d40e53dcf1d8d01518d +size 27116 diff --git a/parse/train/BybtVK9lg/images/781acdad733b25822288b765e852c08b0b23d7bbc25e1ee51a3d0906e234ade5.jpg b/parse/train/BybtVK9lg/images/781acdad733b25822288b765e852c08b0b23d7bbc25e1ee51a3d0906e234ade5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6390590b96292af265733d1ce4d57d3b0f14bdc4 --- /dev/null +++ b/parse/train/BybtVK9lg/images/781acdad733b25822288b765e852c08b0b23d7bbc25e1ee51a3d0906e234ade5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8c342171304aed2e9fae4af813af6bae08483503de3f67a6ecea3d6c0a3679a5 +size 39296 diff --git a/parse/train/BybtVK9lg/images/884bf7a81b00b2e1e0d1f408b65164ecf64d04d823c3f3964eb8a724b7781bc3.jpg b/parse/train/BybtVK9lg/images/884bf7a81b00b2e1e0d1f408b65164ecf64d04d823c3f3964eb8a724b7781bc3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..baf265362821748c5cdd8b525ba845169da1d153 --- /dev/null +++ b/parse/train/BybtVK9lg/images/884bf7a81b00b2e1e0d1f408b65164ecf64d04d823c3f3964eb8a724b7781bc3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:50557d6a3fccf1ddf2ec0ba78fd9aedb6a8e4fd1a121290208410f7a63c821ad +size 9474 diff --git a/parse/train/BybtVK9lg/images/8a99ac3aa617bd8f498a65960710a13a64ff1bae248d3877f646c5d4e2bb514f.jpg b/parse/train/BybtVK9lg/images/8a99ac3aa617bd8f498a65960710a13a64ff1bae248d3877f646c5d4e2bb514f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dbd3aab27f6f1475ca4b923f5cf89f7dc6d2e6b1 --- /dev/null +++ b/parse/train/BybtVK9lg/images/8a99ac3aa617bd8f498a65960710a13a64ff1bae248d3877f646c5d4e2bb514f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6392a0895146977ed636dd515f82cbb25f3b197bbaaedfd7ca9c8f2aad394bd7 +size 7346 diff --git a/parse/train/BybtVK9lg/images/911fb0c748f01e2c6655956acdfd19eb8f0d8d96319048eba29a0f859178078d.jpg b/parse/train/BybtVK9lg/images/911fb0c748f01e2c6655956acdfd19eb8f0d8d96319048eba29a0f859178078d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..604e55cac4338a31a8ffa81d2b3361580916326d --- /dev/null +++ b/parse/train/BybtVK9lg/images/911fb0c748f01e2c6655956acdfd19eb8f0d8d96319048eba29a0f859178078d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9981c9d550b00f669782ff23170a50e4f76e662ec10672189437cac484e9c75e +size 19205 diff --git a/parse/train/BybtVK9lg/images/97d085c8a657535e8bc3d321cc3af25fe230dc71f14cab75c66472fc35dae9e8.jpg b/parse/train/BybtVK9lg/images/97d085c8a657535e8bc3d321cc3af25fe230dc71f14cab75c66472fc35dae9e8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..37bc772071b81bd907e8631a75d2f2fa5b8977ed --- /dev/null +++ b/parse/train/BybtVK9lg/images/97d085c8a657535e8bc3d321cc3af25fe230dc71f14cab75c66472fc35dae9e8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b8fc3ea3ae34f15805f0dbbbfc6d866d4ce5d67e4c1beb008cbc0400f950b99b +size 19180 diff --git a/parse/train/BybtVK9lg/images/b2094ca292cd2e8de3ce313104a3413e9ae589d52d53d728144ce0132732c235.jpg b/parse/train/BybtVK9lg/images/b2094ca292cd2e8de3ce313104a3413e9ae589d52d53d728144ce0132732c235.jpg new file mode 100644 index 0000000000000000000000000000000000000000..316857a14a1929734a412aba6ae4a944222c60e9 --- /dev/null +++ b/parse/train/BybtVK9lg/images/b2094ca292cd2e8de3ce313104a3413e9ae589d52d53d728144ce0132732c235.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:16317f6655b2873f1b4d957770f250cd8e039029b29457b22432a12191381706 +size 46832 diff --git a/parse/train/BybtVK9lg/images/bc79e27465f523217704db7c38625437154c97ebe632d4c497b66cf51b0a0213.jpg b/parse/train/BybtVK9lg/images/bc79e27465f523217704db7c38625437154c97ebe632d4c497b66cf51b0a0213.jpg new file mode 100644 index 0000000000000000000000000000000000000000..50533dd828461d96ef6afd1a184f37d27a280dae --- /dev/null +++ b/parse/train/BybtVK9lg/images/bc79e27465f523217704db7c38625437154c97ebe632d4c497b66cf51b0a0213.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:65e8acfe66c49ea118e7c4a845011b9068dd5b662ee08dec19108f3e617fbb67 +size 9787 diff --git a/parse/train/BybtVK9lg/images/cdb64b776252e4c0dbe9747c33bbd26d6dd89835bfea59a1060c4438c7e90214.jpg b/parse/train/BybtVK9lg/images/cdb64b776252e4c0dbe9747c33bbd26d6dd89835bfea59a1060c4438c7e90214.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d4a981cf373b2e8c17a531e7fc2c1ac13bdff83a --- /dev/null +++ b/parse/train/BybtVK9lg/images/cdb64b776252e4c0dbe9747c33bbd26d6dd89835bfea59a1060c4438c7e90214.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6424f4d3589a5296a7f167d8c30485eaab433e2088f9e816e645ce8a861b301b +size 7395 diff --git a/parse/train/BybtVK9lg/images/cf22b77c6caa001ab7a3a88872abc28b507750a42cb5d4b17d985ef086d69cb7.jpg b/parse/train/BybtVK9lg/images/cf22b77c6caa001ab7a3a88872abc28b507750a42cb5d4b17d985ef086d69cb7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..958c96f024f50e397b0aa4181bfb351144fb5bc8 --- /dev/null +++ b/parse/train/BybtVK9lg/images/cf22b77c6caa001ab7a3a88872abc28b507750a42cb5d4b17d985ef086d69cb7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4180789b026237215d31f317965a05295ca166d76e3e0a74f011015d90d7dbda +size 8134 diff --git a/parse/train/BydrOIcle/BydrOIcle.md b/parse/train/BydrOIcle/BydrOIcle.md new file mode 100644 index 0000000000000000000000000000000000000000..994ea872e21e75da8bfcda503ac9917a9599c6b8 --- /dev/null +++ b/parse/train/BydrOIcle/BydrOIcle.md @@ -0,0 +1,463 @@ +# UNROLLED GENERATIVE ADVERSARIAL NETWORKS + +Luke Metz∗ +Google Brain +lmetz@google.com +Ben Poole† +Stanford University +poole@cs.stanford.edu + +David Pfau Google DeepMind pfau@google.com + +Jascha Sohl-Dickstein Google Brain jaschasd@google.com + +# ABSTRACT + +We introduce a method to stabilize Generative Adversarial Networks (GANs) by defining the generator objective with respect to an unrolled optimization of the discriminator. This allows training to be adjusted between using the optimal discriminator in the generator’s objective, which is ideal but infeasible in practice, and using the current value of the discriminator, which is often unstable and leads to poor solutions. We show how this technique solves the common problem of mode collapse, stabilizes training of GANs with complex recurrent generators, and increases diversity and coverage of the data distribution by the generator. + +# 1 INTRODUCTION + +The use of deep neural networks as generative models for complex data has made great advances in recent years. This success has been achieved through a surprising diversity of training losses and model architectures, including denoising autoencoders (Vincent et al., 2010), variational autoencoders (Kingma & Welling, 2013; Rezende et al., 2014; Gregor et al., 2015; Kulkarni et al., 2015; Burda et al., 2015; Kingma et al., 2016), generative stochastic networks (Alain et al., 2015), diffusion probabilistic models (Sohl-Dickstein et al., 2015), autoregressive models (Theis & Bethge, 2015; van den Oord et al., 2016a;b), real non-volume preserving transformations (Dinh et al., 2014; 2016), Helmholtz machines (Dayan et al., 1995; Bornschein et al., 2015), and Generative Adversarial Networks (GANs) (Goodfellow et al., 2014). + +# 1.1 GENERATIVE ADVERSARIAL NETWORKS + +While most deep generative models are trained by maximizing log likelihood or a lower bound on log likelihood, GANs take a radically different approach that does not require inference or explicit calculation of the data likelihood. Instead, two models are used to solve a minimax game: a generator which samples data, and a discriminator which classifies the data as real or generated. In theory these models are capable of modeling an arbitrarily complex probability distribution. When using the optimal discriminator for a given class of generators, the original GAN proposed by Goodfellow et al. minimizes the Jensen-Shannon divergence between the data distribution and the generator, and extensions generalize this to a wider class of divergences (Nowozin et al., 2016; Sonderby et al., 2016; Poole et al., 2016). + +The ability to train extremely flexible generating functions, without explicitly computing likelihoods or performing inference, and while targeting more mode-seeking divergences as made GANs extremely successful in image generation (Odena et al., 2016; Salimans et al., 2016; Radford et al., 2015), and image super resolution (Ledig et al., 2016). The flexibility of the GAN framework has also enabled a number of successful extensions of the technique, for instance for structured prediction (Reed et al., 2016a;b; Odena et al., 2016), training energy based models (Zhao et al., 2016), and combining the GAN loss with a mutual information loss (Chen et al., 2016). + +In practice, however, GANs suffer from many issues, particularly during training. One common failure mode involves the generator collapsing to produce only a single sample or a small family of very similar samples. Another involves the generator and discriminator oscillating during training, rather than converging to a fixed point. In addition, if one agent becomes much more powerful than the other, the learning signal to the other agent becomes useless, and the system does not learn. To train GANs many tricks must be employed, such as careful selection of architectures (Radford et al., 2015), minibatch discrimination (Salimans et al., 2016), and noise injection (Salimans et al., 2016; Sonderby et al., 2016). Even with these tricks the set of hyperparameters for which training is successful is generally very small in practice. + +Once converged, the generative models produced by the GAN training procedure normally do not cover the whole distribution (Dumoulin et al., 2016; Che et al., 2016), even when targeting a modecovering divergence such as KL. Additionally, because it is intractable to compute the GAN training loss, and because approximate measures of performance such as Parzen window estimates suffer from major flaws (Theis et al., 2016), evaluation of GAN performance is challenging. Currently, human judgement of sample quality is one of the leading metrics for evaluating GANs. In practice this metric does not take into account mode dropping if the number of modes is greater than the number of samples one is visualizing. In fact, the mode dropping problem generally helps visual sample quality as the model can choose to focus on only the most common modes. These common modes correspond, by definition, to more typical samples. Additionally, the generative model is able to allocate more expressive power to the modes it does cover than it would if it attempted to cover all modes. + +# 1.2 DIFFERENTIATING THROUGH OPTIMIZATION + +Many optimization schemes, including SGD, RMSProp (Tieleman & Hinton, 2012), and Adam (Kingma & Ba, 2014), consist of a sequence of differentiable updates to parameters. Gradients can be backpropagated through unrolled optimization updates in a similar fashion to backpropagation through a recurrent neural network. The parameters output by the optimizer can thus be included, in a differentiable way, in another objective (Maclaurin et al., 2015). This idea was first suggested for minimax problems in (Pearlmutter & Siskind, 2008), while (Zhang & Lesser, 2010) provided a theoretical analysis and experimental results on differentiating through a single step of gradient ascent for simple matrix games. Differentiating through unrolled optimization was first scaled to deep networks in (Maclaurin et al., 2015), where it was used for hyperparameter optimization. More recently, (Belanger & McCallum, 2015; Han et al., 2016; Andrychowicz et al., 2016) backpropagate through optimization procedures in contexts unrelated to GANs or minimax games. + +In this work we address the challenges of unstable optimization and mode collapse in GANs by unrolling optimization of the discriminator objective during training. + +# 2 METHOD + +# 2.1 GENERATIVE ADVERSARIAL NETWORKS + +The GAN learning problem is to find the optimal parameters $\theta _ { G } ^ { * }$ for a generator function $G \left( z ; \theta _ { G } \right)$ in a minimax objective, + +$$ +\begin{array} { c } { { \theta _ { G } ^ { * } = \underset { \theta _ { G } } { \mathrm { a r g m i n } } \underset { \theta _ { D } } { \mathrm { m a x } } f \left( \theta _ { G } , \theta _ { D } \right) } } \\ { { \ \mathrm { ~ } } } \\ { { \displaystyle \qquad = \underset { \theta _ { G } } { \mathrm { a r g m i n } } f \left( \theta _ { G } , \theta _ { D } ^ { * } \left( \theta _ { G } \right) \right) } } \\ { { \theta _ { D } ^ { * } \left( \theta _ { G } \right) = \underset { \theta _ { D } } { \mathrm { a r g m a x } } f \left( \theta _ { G } , \theta _ { D } \right) , } } \end{array} +$$ + +where $f$ is commonly chosen to be + +$$ +f \left( \theta _ { G } , \theta _ { D } \right) = \mathbb { E } _ { x \sim p _ { d a t a } } \left[ \log \left( D \left( x ; \theta _ { D } \right) \right) \right] + \mathbb { E } _ { z \sim N ( 0 , I ) } \left[ \log \left( 1 - D \left( G \left( z ; \theta _ { G } \right) ; \theta _ { D } \right) \right) \right] . +$$ + +Here $x \in \mathcal { X }$ is the data variable, $z \in { \mathcal { Z } }$ is the latent variable, $p _ { d a t a }$ is the data distribution, the discriminator $D ( \cdot ; \theta _ { D } ) : \mathcal { X } [ 0 , 1 ]$ outputs the estimated probability that a sample $x$ comes from the data distribution, $\theta _ { D }$ and $\theta _ { G }$ are the discriminator and generator parameters, and the generator function $G ( \cdot ; \theta _ { G } ) : \mathcal { Z } \mathcal { X }$ transforms a sample in the latent space into a sample in the data space. + +For the minimax loss in Eq. 4, the optimal discriminator $D ^ { \ast } \left( x \right)$ is a known smooth function of the generator probability $p _ { G } \left( x \right)$ (Goodfellow et al., 2014), + +$$ +D ^ { * } \left( x \right) = \frac { p _ { d a t a } \left( x \right) } { p _ { d a t a } \left( x \right) + p _ { G } \left( x \right) } . +$$ + +When the generator loss in Eq. 2 is rewritten directly in terms of $p _ { G } \left( x \right)$ and Eq. 5 rather than $\theta _ { G }$ and $\theta _ { D } ^ { * } \left( \theta _ { G } ^ { - } \right)$ , then it is similarly a smooth function of $p _ { G } \left( x \right)$ . These smoothness guarantees are typically lost when $D \left( x ; \theta _ { D } \right)$ and $G \left( z ; \theta _ { G } \right)$ are drawn from parametric families. They nonetheless suggest that the true generator objective in Eq. 2 will often be well behaved, and is a desirable target for direct optimization. + +Explicitly solving for the optimal discriminator parameters $\theta _ { D } ^ { * } \left( \theta _ { G } \right) $ for every update step of the generator $G$ is computationally infeasible for discriminators based on neural networks. Therefore this minimax optimization problem is typically solved by alternating gradient descent on $\theta _ { G }$ and ascent on $\theta _ { D }$ . + +The optimal solution $\theta ^ { * } = \{ \theta _ { G } ^ { * } , \theta _ { D } ^ { * } \}$ is a fixed point of these iterative learning dynamics. Additionally, if $f \left( { \theta } _ { G } , { \theta } _ { D } \right)$ is convex in $\theta _ { G }$ and concave in $\theta _ { D }$ , then alternating gradient descent (ascent) trust region updates are guaranteed to converge to the fixed point, under certain additional weak assumptions (Juditsky et al., 2011). However in practice $f \left( { \theta } _ { G } , { \theta } _ { D } \right)$ is typically very far from convex in $\theta _ { G }$ and concave in $\theta _ { D }$ , and updates are not constrained in an appropriate way. As a result GAN training suffers from mode collapse, undamped oscillations, and other problems detailed in Section 1.1. In order to address these difficulties, we will introduce a surrogate objective function $f _ { K } \left( \theta _ { G } , \theta _ { D } \right)$ for training the generator which more closely resembles the true generator objective $f \left( \theta _ { G } , \theta _ { D } ^ { * } \left( \theta _ { G } \right) \right)$ . + +# 2.2 UNROLLING GANS + +A local optimum of the discriminator parameters $\theta _ { D } ^ { * }$ can be expressed as the fixed point of an iterative optimization procedure, + +$$ +\begin{array} { c } { { \theta _ { D } ^ { 0 } = \theta _ { D } } } \\ { { \displaystyle } } \\ { { \theta _ { D } ^ { k + 1 } = \theta _ { D } ^ { k } + \eta ^ { k } \displaystyle \frac { \mathrm { d } f ( \theta _ { G } , \theta _ { D } ^ { k } ) } { \mathrm { d } \theta _ { D } ^ { k } } } } \\ { { \displaystyle } } \\ { { \theta _ { D } ^ { * } ( \theta _ { G } ) = \displaystyle \operatorname* { l i m } _ { k \infty } \theta _ { D } ^ { k } , } } \end{array} +$$ + +where $\eta ^ { k }$ is the learning rate schedule. For clarity, we have expressed Eq. 7 as a full batch steepest gradient ascent equation. More sophisticated optimizers can be similarly unrolled. In our experiments we unroll Adam (Kingma & Ba, 2014). + +By unrolling for $K$ steps, we create a surrogate objective for the update of the generator, + +$$ +f _ { K } \left( \theta _ { G } , \theta _ { D } \right) = f \left( \theta _ { G } , \theta _ { D } ^ { K } \left( \theta _ { G } , \theta _ { D } \right) \right) . +$$ + +When $K = 0$ this objective corresponds exactly to the standard GAN objective, while as $K \infty$ it corresponds to the true generator objective function $f \left( \theta _ { G } , \theta _ { D } ^ { * } \left( G \right) \right)$ . By adjusting the number of unrolling steps $K$ , we are thus able to interpolate between standard GAN training dynamics with their associated pathologies, and more costly gradient descent on the true generator loss. + +# 2.3 PARAMETER UPDATES + +The generator and discriminator parameter updates using this surrogate loss are + +$$ +\begin{array} { r l } & { \theta _ { G } \theta _ { G } - \eta \frac { \mathrm { d } f _ { K } ( \theta _ { G } , \theta _ { D } ) } { \mathrm { d } \theta _ { G } } } \\ & { \theta _ { D } \theta _ { D } + \eta \frac { \mathrm { d } f ( \theta _ { G } , \theta _ { D } ) } { \mathrm { d } \theta _ { D } } . } \end{array} +$$ + +For clarity we use full batch steepest gradient descent (ascent) with stepsize $\eta$ above, while in experiments we instead use minibatch Adam for both updates. The gradient in Eq. 10 requires backpropagating through the optimization process in Eq. 7. A clear description of differentiation through gradient descent is given as Algorithm 2 in (Maclaurin et al., 2015), though in practice the use of an automatic differentiation package means this step does not need to be programmed explicitly. A pictorial representation of these updates is provided in Figure 1. + +![](images/b8d41d44f95393092b08110ba25482b02bb1bcb0a2930f383731e4e370791ca3.jpg) +Figure 1: An illustration of the computation graph for an unrolled GAN with 3 unrolling steps. The generator update in Equation 10 involves backpropagating the generator gradient (blue arrows) through the unrolled optimization. Each step $k$ in the unrolled optimization uses the gradients of $f _ { k }$ with respect to $\theta _ { D } ^ { k }$ , as described in Equation 7 and indicated by the green arrows. The discriminator update in Equation 11 does not depend on the unrolled optimization (red arrow). + +It is important to distinguish this from an approach suggested in (Goodfellow et al., 2014), that several update steps of the discriminator parameters should be run before each single update step for the generator. In that approach, the update steps for both models are still gradient descent (ascent) with respect to fixed values of the other model parameters, rather than the surrogate loss we describe in Eq. 9. Performing $K$ steps of discriminator update between each single step of generator update corresponds to updating the generator parameters $\theta _ { G }$ using only the first term in Eq. 12 below. + +# 2.4 THE MISSING GRADIENT TERM + +To better understand the behavior of the surrogate loss $f _ { K } \left( \theta _ { G } , \theta _ { D } \right)$ , we examine its gradient with respect to the generator parameters $\theta _ { G }$ , + +$$ +\frac { \mathrm { d } f _ { K } \left( \theta _ { G } , \theta _ { D } \right) } { \mathrm { d } \theta _ { G } } = \frac { \partial f \left( \theta _ { G } , \theta _ { D } ^ { K } \left( \theta _ { G } , \theta _ { D } \right) \right) } { \partial \theta _ { G } } + \frac { \partial f \left( \theta _ { G } , \theta _ { D } ^ { K } \left( \theta _ { G } , \theta _ { D } \right) \right) } { \partial \theta _ { D } ^ { K } \left( \theta _ { G } , \theta _ { D } \right) } \frac { \mathrm { d } \theta _ { D } ^ { K } \left( \theta _ { G } , \theta _ { D } \right) } { \mathrm { d } \theta _ { G } } . +$$ + +Standard GAN training corresponds exactly to updating the generator parameters using only the first term in this gradient, with $\theta _ { D } ^ { K } \left( \theta _ { G } , \theta _ { D } \right) $ being the parameters resulting from the discriminator update step. An optimal generator for any fixed discriminator is a delta function at the $x$ to which the discriminator assigns highest data probability. Therefore, in standard GAN training, each generator update step is a partial collapse towards a delta function. + +The second term captures how the discriminator would react to a change in the generator. It reduces the tendency of the generator to engage in mode collapse. For instance, the second term reflects that as the generator collapses towards a delta function, the discriminator reacts and assigns lower probability to that state, increasing the generator loss. It therefore discourages the generator from collapsing, and may improve stability. + +As $K \infty$ , $\theta _ { D } ^ { K }$ goes to a local optimum of $f$ , where $\frac { \partial f } { \partial \theta _ { D } ^ { K } } = 0$ , and therefore the second term in Eq. 12 goes to 0 (Danskin, 1967). The gradient of the unrolled surrogate loss $f _ { K } \left( \theta _ { G } , \theta _ { D } \right)$ with respect to $\theta _ { G }$ is thus identical to the gradient of the standard GAN loss $f \bar { ( \theta _ { G } , \theta _ { D } ) }$ both when $K = 0$ and when $K \infty$ , where we take $K \infty$ to imply that in the standard GAN the discriminator is also fully optimized between each generator update. Between these two extremes, $f _ { K } \left( \theta _ { G } , \theta _ { D } \right)$ captures additional information about the response of the discriminator to changes in the generator. + +# 2.5 CONSEQUENCES OF THE SURROGATE LOSS + +GANs can be thought of as a game between the discriminator $( D )$ and the generator $( G )$ . The agents take turns taking actions and updating their parameters until a Nash equilibrium is reached. The optimal action for $D$ is to evaluate the probability ratio $\frac { p _ { d a t a } ( x ) } { p _ { G } ( x ) + p _ { d a t a } ( x ) }$ for the generator’s move $x$ (Eq. 5). The optimal generator action is to move its mass to maximize this ratio. + +The initial move for $G$ will be to move as much mass as its parametric family and update step permits to the single point that maximizes the ratio of probability densities. The action $D$ will then take is quite simple. It will track that point, and to the extent allowed by its own parametric family and update step assign low data probability to it, and uniform probability everywhere else. This cycle of $G$ moving and $D$ following will repeat forever or converge depending on the rate of change of the two agents. This is similar to the situation in simple matrix games like rock-paper-scissors and matching pennies, where alternating gradient descent (ascent) with a fixed learning rate is known not to converge (Singh et al., 2000; Bowling & Veloso, 2002). + +In the unrolled case, however, this undesirable behavior no longer occurs. Now $G$ ’s actions take into account how $D$ will respond. In particular, $G$ will try to make steps that $D$ will have a hard time responding to. This extra information helps the generator spread its mass to make the next $D$ step less effective instead of collapsing to a point. + +In principle, a surrogate loss function could be used for both $D$ and $G$ . In the case of 1-step unrolled optimization this is known to lead to convergence for games in which gradient descent (ascent) fails (Zhang & Lesser, 2010). However, the motivation for using the surrogate generator loss in Section 2.2, of unrolling the inner of two nested min and max functions, does not apply to using a surrogate discriminator loss. Additionally, it is more common for the discriminator to overpower the generator than vice-versa when training a GAN. Giving more information to $G$ by allowing it to ‘see into the future’ may thus help the two models be more balanced. + +# 3 EXPERIMENTS + +In this section we demonstrate improved mode coverage and stability by applying this technique to five datasets of increasing complexity. Evaluation of generative models is a notoriously hard problem (Theis et al., 2016). As such the de facto standard in GAN literature has become sample quality as evaluated by a human and/or evaluated by a heuristic (Inception score for example, (Salimans et al., 2016)). While these evaluation metrics do a reasonable job capturing sample quality, they fail to capture sample diversity. In our first 2 experiments diversity is easily evaluated via visual inspection. In our later experiments this is not the case, and we will use a variety of methods to quantify coverage of samples. Our measures are individually strongly suggestive of unrolling reducing mode-collapse and improving stability, but none of them alone are conclusive. We believe that taken together however, they provide extremely compelling evidence for the advantages of unrolling. + +When doing stochastic optimization, we must choose which minibatches to use in the unrolling updates in Eq. 7. We experimented with both a fixed minibatch and re-sampled minibatches for each unrolling step, and found it did not significantly impact the result. We use fixed minibatches for all experiments in this section. + +We provide a reference implementation of this technique at github.com/poolio/unrolled gan. + +# 3.1 MIXTURE OF GAUSSIANS DATASET + +To illustrate the impact of discriminator unrolling, we train a simple GAN architecture on a 2D mixture of 8 Gaussians arranged in a circle. For a detailed list of architecture and hyperparameters see Appendix A. Figure 2 shows the dynamics of this model through time. Without unrolling the generator rotates around the valid modes of the data distribution but is never able to spread out mass. When adding in unrolling steps $\mathbf { G }$ quickly learns to spread probability mass and the system converges to the data distribution. + +In Appendix B we perform further experiments on this toy dataset. We explore how unrolling compares to historical averaging, and compares to using the unrolled discriminator to update the generator, but without backpropagating through the generator. In both cases we find that the unrolled objective performs better. + +![](images/a4459b06fa7e9fcdf667d911bf408b5c5ccfe73af1d14e29428d98326008c54a.jpg) +Figure 2: Unrolling the discriminator stabilizes GAN training on a toy 2D mixture of Gaussians dataset. Columns show a heatmap of the generator distribution after increasing numbers of training steps. The final column shows the data distribution. The top row shows training for a GAN with 10 unrolling steps. Its generator quickly spreads out and converges to the target distribution. The bottom row shows standard GAN training. The generator rotates through the modes of the data distribution. It never converges to a fixed distribution, and only ever assigns significant probability mass to a single data mode at once. + +![](images/ff67f1c079a9b83431ee462581bfb81893ea4eadc9da332a9fff1968becf8c08.jpg) +Figure 3: Unrolled GAN training increases stability for an RNN generator and convolutional discriminator trained on MNIST. The top row was run with 20 unrolling steps. The bottom row is a standard GAN, with 0 unrolling steps. Images are samples from the generator after the indicated number of training steps. + +# 3.2 PATHOLOGICAL MODEL WITH MISMATCHED GENERATOR AND DISCRIMINATOR + +To evaluate the ability of this approach to improve trainability, we look to a traditionally challenging family of models to train – recurrent neural networks (RNNs). In this experiment we try to generate MNIST samples using an LSTM (Hochreiter & Schmidhuber, 1997). MNIST digits are $2 8 \mathbf { x } 2 8$ pixel images. At each timestep of the generator LSTM, it outputs one column of this image, so that after 28 timesteps it has output the entire sample. We use a convolutional neural network as the discriminator. See Appendix C for the full model and training details. Unlike in all previously successful GAN models, there is no symmetry between the generator and the discriminator in this task, resulting in a more complex power balance. Results can be seen in Figure 3. Once again, without unrolling the model quickly collapses, and rotates through a sequence of single modes. Instead of rotating spatially, it cycles through proto-digit like blobs. When running with unrolling steps the generator disperses and appears to cover the whole data distribution, as in the 2D example. + +
Discriminator SizeUnrolling steps01510
1/4 size of D compared to GModes generated30.6± 20.7365.4 ± 34.75236.4 ± 63.30327.2 ± 74.67
KL(model||data)5.99± 0.425.911 ± 0.144.67 ± 0.434.66 ± 0.46
1/2 size of D compared to GModes generated628.0± 140.9523.6± 55.768732.0± 44.98817.4 ± 37.91
KL(model||data)2.58 ±0.7512.44 ±0.261.66 ± 0.0901.43 ± 0.12
+ +Table 1: Unrolled GANs cover more discrete modes when modeling a dataset with 1,000 data modes, corresponding to all combinations of three MNIST digits $[ 1 0 ^ { 3 }$ digit combinations). The number of modes covered is given for different numbers of unrolling steps, and for two different architectures. The reverse KL divergence between model and data is also given. Standard error is provided for both measures. + +# 3.3 MODE AND MANIFOLD COLLAPSE USING AUGMENTED MNIST + +GANs suffer from two different types of model collapse – collapse to a subset of data modes, and collapse to a sub-manifold within the data distribution. In these experiments we isolate both effects using artificially constructed datasets, and demonstrate that unrolling can largely rescue both types of collapse. + +# 3.3.1 DISCRETE MODE COLLAPSE + +To explore the degree to which GANs drop discrete modes in a dataset, we use a technique similar to one from (Che et al., 2016). We construct a dataset by stacking three randomly chosen MNIST digits, so as to construct an RGB image with a different MNIST digit in each color channel. This new dataset has 1,000 distinct modes, corresponding to each combination of the ten MNIST classes in the three channels. + +We train a GAN on this dataset, and generate samples from the trained model (25,600 samples for all experiments). We then compute the predicted class label of each color channel using a pre-trained MNIST classifier. To evaluate performance, we use two metrics: the number of modes for which the generator produced at least one sample, and the KL divergence between the model and the expected data distribution. Within this discrete label space, a KL divergence can be estimated tractably between the generated samples and the data distribution over classes, where the data distribution is a uniform distribution over all 1,000 classes. + +As presented in Table 1, as the number of unrolling steps is increased, both mode coverage and reverse KL divergence improve. Contrary to (Che et al., 2016), we found that reasonably sized models (such as the one used in Section 3.4) covered all 1,000 modes even without unrolling. As such we use smaller convolutional GAN models. Details on the models used are provided in Appendix E. + +We observe an additional interesting effect in this experiment. The benefits of unrolling increase as the discriminator size is reduced. We believe unrolling effectively increases the capacity of the discriminator. The unrolled discriminator can better react to any specific way in which the generator is producing non-data-like samples. When the discriminator is weak, the positive impact of unrolling is thus larger. + +# 3.3.2 MANIFOLD COLLAPSE + +In addition to discrete modes, we examine the effect of unrolling when modeling continuous manifolds. To get at this quantity, we constructed a dataset consisting of colored MNIST digits. Unlike in the previous experiment, a single MNIST digit was chosen, and then assigned a single monochromatic color. With a perfect generator, one should be able to recover the distribution of colors used to generate the digits. We use colored MNIST digits so that the generator also has to model the digits, which makes the task sufficiently complex that the generator is unable to perfectly solve it. The color of each digit is sampled from a 3D normal distribution. Details of this dataset are provided in Appendix F. We will examine the distribution of colors in the samples generated by the trained GAN. As will also be true in the CIFAR10 example in Section 3.4, the lack of diversity in generated colors is almost invisible using only visual inspection of the samples. Samples can be found in Appendix F. + +Table 2: Unrolled GANs better model a continuous distribution. GANs are trained to model randomly colored MNIST digits, where the color is drawn from a Gaussian distribution. The JS divergence between the data and model distributions over digit colors is then reported, along with standard error in the JS divergence. More unrolling steps, and larger models, lead to better JS divergence. + +
Unrolling steps01510
JS divergence with 1/4 layer size0.073 ± 0.00580.142 ± 0.0280.049 ± 0.00210.075 ± 0.012
JS divergence with 1/2 layer size0.095 ± 0.0110.119 ± 0.0100.055 ± 0.00490.074± 0.016
JS divergence with 1/1 layer size0.034 ± 0.00340.050± 0.00260.027 ± 0.00280.025 ± 0.00076
+ +![](images/3c9a4f15816f9475e20e64fe6d880999dbc71b9039d3a86102ace92afbbde1f2.jpg) +Figure 4: Visual perception of sample quality and diversity is very similar for models trained with different numbers of unrolling steps. Actual sample diversity is higher with more unrolling steps. Each pane shows samples generated after training a model on CIFAR10 with 0, 1, 5, and 10 steps of unrolling. + +In order to recover the color the GAN assigned to the digit, we used k-means with 2 clusters, to pick out the foreground color from the background. We then performed this transformation for both the training data and the generated images. Next we fit a Gaussian kernel density estimator to both distributions over digit colors. Finally, we computed the JS divergence between the model and data distributions over colors. Results can be found in Table 2 for several model sizes. Details of the models are provided in Appendix F. + +In general, the best performing models are unrolled for 5-10 steps, and larger models perform better than smaller models. Counter-intuitively, taking 1 unrolling step seems to hurt this measure of diversity. We suspect that this is due to it introducing oscillatory dynamics into training. Taking more unrolling steps however leads to improved performance with unrolling. + +# 3.4 IMAGE MODELING OF CIFAR10 + +Here we test our technique on a more traditional convolutional GAN architecture and task, similar to those used in (Radford et al., 2015; Salimans et al., 2016). In the previous experiments we tested models where the standard GAN training algorithm would not converge. In this section we improve a standard model by reducing its tendency to engage in mode collapse. We ran 4 configurations of this model, varying the number of unrolling steps to be 0, 1, 5, or 10. Each configuration was run 5 times with different random seeds. For full training details see Appendix D. Samples from each of the 4 configurations can be found in Figure 4. There is no obvious difference in visual quality across these model configurations. Visual inspection however provides only a poor measure of sample diversity. + +By training with an unrolled discriminator, we expect to generate more diverse samples which more closely resemble the underlying data distribution. We introduce two techniques to examine sample diversity: inference via optimization, and pairwise distance distributions. + +
Unrolling Steps0 steps1 step5 steps10 steps
Average MSE0.0231± 0.00240.0195 ± 0.00210.0200± 0.00230.0181± 0.0018
PercentBestRank0.63%22.97%15.31%61.09 %
+ +Table 3: GANs trained with unrolling are better able to match images in the training set than standard GANs, likely due to mode dropping by the standard GAN. Results show the MSE between training images and the best reconstruction for a model with the given number of unrolling steps. The fraction of training images best reconstructed by a given model is given in the final column. The best reconstructions is found by optimizing the latent representation $z$ to produce the closest matching pixel output $G \left( z ; \theta _ { G } \right)$ . Results are averaged over all 5 runs of each model with different random seeds. + +# 3.4.1 INFERENCE VIA OPTIMIZATION + +Since likelihood cannot be tractably computed, over-fitting of GANs is typically tested by taking samples and computing the nearest-neighbor images in pixel space from the training data (Goodfellow et al., 2014). We will do the reverse, and measure the ability of the generative model to generate images that look like specific samples from the training data. If we did this by generating random samples from the model, we would need an exponentially large number of samples. We instead treat finding the nearest neighbor $x _ { \mathrm { n e a r e s t } }$ to a target image $x _ { \mathrm { { t a r g e t } } }$ as an optimization task, + +$$ +\begin{array} { r l } & { z _ { \mathrm { n e a r e s t } } = \underset { z } { \mathrm { a r g m i n } } | | G ( z ; \theta _ { G } ) - x _ { \mathrm { t a r g e t } } | \rvert _ { 2 } ^ { 2 } } \\ & { x _ { \mathrm { n e a r e s t } } = G ( z _ { \mathrm { n e a r e s t } } ; \theta _ { G } ) . } \end{array} +$$ + +This concept of backpropagating to generate images has been widely used in visualizing features from discriminative networks (Simonyan et al., 2013; Yosinski et al., 2015; Nguyen et al., 2016) and has been applied to explore the visual manifold of GANs in (Zhu et al., 2016). + +We apply this technique to each of the models trained. We optimize with 3 random starts using LBFGS, which is the optimizer typically used in similar settings such as style transfer (Johnson et al., 2016; Champandard, 2016). Results comparing average mean squared errors between xnearest and $x _ { \mathrm { { t a r g e t } } }$ in pixel space can be found in Table 3. In addition we compute the percent of images for which a certain configuration achieves the lowest loss when compared to the other configurations. + +In the zero step case, there is poor reconstruction and less than $1 \%$ of the time does it obtain the lowest error of the 4 configurations. Taking 1 unrolling step results in a significant improvement in MSE. Taking 10 unrolling steps results in more modest improvement, but continues to reduce the reconstruction MSE. + +To visually see this, we compare the result of the optimization process for 0, 1, 5, and 10 step configurations in Figure 5. To select for images where differences in behavior is most apparent, we sort the data by the absolute value of a fractional difference in MSE between the 0 and 10 step models, $\left| \frac { l _ { 0 s t e p } - l _ { 1 0 s t e p } } { \frac { 1 } { 2 } ( l _ { 0 s t e p } + l _ { 1 0 s t e p } ) } \right|$ This highlights examples where either the 0 or 10 step model cannot accurately fit the data example but the other can. In Appendix G we show the same comparison for models initialized using different random seeds. Many of the zero step images are fuzzy and illdefined suggesting that these images cannot be generated by the standard GAN generative model, and come from a dropped mode. As more unrolling steps are added, the outlines become more clear and well defined – the model covers more of the distribution and thus can recreate these samples. + +# 3.4.2 PAIRWISE DISTANCES + +A second complementary approach is to compare statistics of data samples to the corresponding statistics for samples generated by the various models. One particularly simple and relevant statistic is the distribution over pairwise distances between random pairs of samples. In the case of mode collapse, greater probability mass will be concentrated in smaller volumes, and the distribution over inter-sample distances should be skewed towards smaller distances. We sample random pairs of images from each model, as well as from the training data, and compute histograms of the $\ell _ { 2 }$ distances between those sample pairs. As illustrated in Figure 6, the standard GAN, with zero unrolling steps, has its probability mass skewed towards smaller $\ell _ { 2 }$ intersample distances, compared to real data. As the number of unrolling steps is increased, the histograms over intersample distances increasingly come to resemble that for the data distribution. This is further evidence in support of unrolling decreasing the mode collapse behavior of GANs. + +![](images/3f34af532391d651dea1658c3e0b24559b5a1f84c026a63aecddf941edff8580.jpg) +Figure 5: Training set images are more accurately reconstructed using GANs trained with unrolling than by a standard (0 step) GAN, likely due to mode dropping by the standard GAN. Raw data is on the left, and the optimized images to reach this target follow for 0, 1, 5, and 10 unrolling steps. The reconstruction MSE is listed below each sample. A random 1280 images where selected from the training set, and corresponding best reconstructions for each model were found via optimization. Shown here are the eight images with the largest absolute fractional difference between GANs trained with 0 and 10 unrolling steps. + +# 4 DISCUSSION + +In this work we developed a method to stabilize GAN training and reduce mode collapse by defining the generator objective with respect to unrolled optimization of the discriminator. We then demonstrated the application of this method to several tasks, where it either rescued unstable training, or reduced the tendency of the model to drop regions of the data distribution. + +The main drawback to this method is computational cost of each training step, which increases linearly with the number of unrolling steps. There is a tradeoff between better approximating the true generator loss and the computation required to make this estimate. Depending on the architecture, one unrolling step can be enough. In other more unstable models, such as the RNN case, more are needed to stabilize training. We have some initial positive results suggesting it may be sufficient to further perturb the training gradient in the same direction that a single unrolling step perturbs it. While this is more computationally efficient, further investigation is required. + +The method presented here bridges some of the gap between theoretical and practical results for training of GANs. We believe developing better update rules for the generator and discriminator is an important line of work for GAN training. In this work we have only considered a small fraction of the design space. For instance, the approach could be extended to unroll $G$ when updating $D$ as well – letting the discriminator react to how the generator would move. It is also possible to unroll sequences of $G$ and $D$ updates. This would make updates that are recursive: $G$ could react to maximize performance as if $G$ and $D$ had already updated. + +# ACKNOWLEDGMENTS + +We would like to thank Laurent Dinh, David Dohan, Vincent Dumoulin, Liam Fedus, Ishaan Gulrajani, Julian Ibarz, Eric Jang, Matthew Johnson, Marc Lanctot, Augustus Odena, Gabriel Pereyra, + +![](images/34368461772c640ab500044ba570ada5248c204871939191a4efbed64c4bf4ee.jpg) +Figure 6: As the number of unrolling steps in GAN training is increased, the distribution of pairwise distances between model samples more closely resembles the same distribution for the data. Here we plot histograms of pairwise distances between randomly selected samples. The red line gives pairwise distances in the data, while each of the five blue lines in each plot represents a model trained with a different random seed. The vertical lines are the medians of each distribution. + +Colin Raffel, Sam Schoenholz, Ayush Sekhari, Jon Shlens, and Dale Schuurmans for insightful conversation, as well as the rest of the Google Brain Team. + +# REFERENCES + +Guillaume Alain, Yoshua Bengio, Li Yao, Jason Yosinski, Eric Thibodeau-Laufer, Saizheng Zhang, and Pascal Vincent. Gsns : Generative stochastic networks. arXiv preprint arXiv:1503.05571, 2015. + +Marcin Andrychowicz, Misha Denil, Sergio Gomez, Matthew W Hoffman, David Pfau, Tom Schaul, and Nando de Freitas. Learning to learn by gradient descent by gradient descent. arXiv preprint arXiv:1606.04474, 2016. + +David Belanger and Andrew McCallum. Structured prediction energy networks. arXiv preprint arXiv:1511.06350, 2015. + +Jorg Bornschein, Samira Shabanian, Asja Fischer, and Yoshua Bengio. Bidirectional helmholtz machines. arXiv preprint arXiv:1506.03877, 2015. + +Michael Bowling and Manuela Veloso. Multiagent learning using a variable learning rate. Artificial Intelligence, 136(2):215–250, 2002. + +Yuri Burda, Roger B. Grosse, and Ruslan Salakhutdinov. Importance weighted autoencoders. arXiv preprint arXiv:1509.00519, 2015. + +Alex J. Champandard. Semantic style transfer and turning two-bit doodles into fine artworks. arXiv preprint arXiv:1603.01768, 2016. + +Tong Che, Yanran Li, Athul Paul Jacob, Yoshua Bengio, and Wenjie Li. Mode regularized generative adversarial networks. arXiv preprint arXiv: 1612.02136, 2016. + +Xi Chen, Yan Duan, Rein Houthooft, John Schulman, Ilya Sutskever, and Pieter Abbeel. Infogan: Interpretable representation learning by information maximizing generative adversarial nets. arXiv preprint arXiv:1606.03657, 2016. + +John M Danskin. The theory of max-min and its application to weapons allocation problems, volume 5. Springer Science & Business Media, 1967. + +Peter Dayan, Geoffrey E Hinton, Radford M Neal, and Richard S Zemel. The helmholtz machine. Neural computation, 7(5):889–904, 1995. + +Laurent Dinh, David Krueger, and Yoshua Bengio. NICE: non-linear independent components estimation. arXiv preprint arXiv:1410.8516, 2014. + +Laurent Dinh, Jascha Sohl-Dickstein, and Samy Bengio. Density estimation using real NVP. arXiv preprint arXiv:1605.08803, 2016. + +Vincent Dumoulin, Ishmael Belghazi, Ben Poole, Alex Lamb, Martin Arjovsky, Olivier Mastropietro, and Aaron Courville. Adversarially learned inference. arXiv preprint arXiv:1606.00704, 2016. + +Xavier Glorot and Yoshua Bengio. Understanding the difficulty of training deep feedforward neural networks. In JMLR W&CP: Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics (AISTATS 2010), volume 9, pp. 249–256, May 2010. + +Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Z. Ghahramani, M. Welling, C. Cortes, N. D. Lawrence, and K. Q. Weinberger (eds.), Advances in Neural Information Processing Systems 27, pp. 2672–2680. Curran Associates, Inc., 2014. URL http://papers. nips.cc/paper/5423-generative-adversarial-nets.pdf. + +Karol Gregor, Ivo Danihelka, Alex Graves, and Daan Wierstra. DRAW: A recurrent neural network for image generation. In Proceedings of The 32nd International Conference on Machine Learning, pp. 1462–1471, 2015. URL http://www.jmlr.org/proceedings/papers/v37/ gregor15.html. + +Tian Han, Yang Lu, Song-Chun Zhu, and Ying Nian Wu. Alternating back-propagation for generator network, 2016. URL https://arxiv.org/abs/1606.08571. + +Sepp Hochreiter and Jurgen Schmidhuber. Long short-term memory. ¨ Neural Comput., 9(8):1735– 1780, November 1997. ISSN 0899-7667. doi: 10.1162/neco.1997.9.8.1735. URL http://dx. doi.org/10.1162/neco.1997.9.8.1735. + +Sergey Ioffe and Christian Szegedy. Batch normalization: Accelerating deep network training by reducing internal covariate shift. In Proceedings of the 32nd International Conference on Machine Learning, ICML 2015, Lille, France, 6-11 July 2015, pp. 448–456, 2015. URL http://jmlr. org/proceedings/papers/v37/ioffe15.html. + +Justin Johnson, Alexandre Alahi, and Fei-Fei Li. Perceptual losses for real-time style transfer and super-resolution. arXiv preprint arXiv:1603.08155, 2016. + +Anatoli Juditsky, Arkadi Nemirovski, et al. First order methods for nonsmooth convex large-scale optimization, i: general purpose methods. Optimization for Machine Learning, pp. 121–148, 2011. + +Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014. + +Diederik P Kingma and Max Welling. Auto-encoding variational bayes, 2013. URL https: //arxiv.org/abs/1312.6114. + +Diederik P. Kingma, Tim Salimans, and Max Welling. Improving variational inference with inverse autoregressive flow. 2016. + +Tejas D. Kulkarni, Will Whitney, Pushmeet Kohli, and Joshua B. Tenenbaum. Deep convolutional inverse graphics network. arXiv preprint arXiv:1503.03167, 2015. + +Christian Ledig, Lucas Theis, Ferenc Huszar, Jose Caballero, Andrew Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, and Wenzhe Shi. Photo-realistic single image super-resolution using a generative adversarial network, 2016. URL https://arxiv.org/abs/1609.04802. + +Dougal Maclaurin, David Duvenaud, and Ryan P. Adams. Gradient-based hyperparameter optimization through reversible learning, 2015. + +Anh Nguyen, Alexey Dosovitskiy, Jason Yosinski, Thomas Brox, and Jeff Clune. Synthesizing the preferred inputs for neurons in neural networks via deep generator networks. arXiv preprint arXiv:1605.09304, 2016. + +Sebastian Nowozin, Botond Cseke, and Ryota Tomioka. f-gan: Training generative neural samplers using variational divergence minimization. arXiv preprint arXiv:1606.00709, 2016. + +Augustus Odena, Christopher Olah, and Jonathon Shlens. Conditional image synthesis with auxiliary classifier gans. arXiv preprint arXiv:1610.09585, 2016. + +Barak A. Pearlmutter and Jeffrey Mark Siskind. Reverse-mode ad in a functional framework: Lambda the ultimate backpropagator. ACM Trans. Program. Lang. Syst., 30(2):7:1–7:36, March 2008. ISSN 0164-0925. doi: 10.1145/1330017.1330018. URL http://doi.acm.org/10. 1145/1330017.1330018. + +Ben Poole, Alexander A Alemi, Jascha Sohl-Dickstein, and Anelia Angelova. Improved generator objectives for gans. arXiv preprint arXiv:1612.02780, 2016. + +Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learning with deep convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434, 2015. + +Scott Reed, Zeynep Akata, Santosh Mohan, Samuel Tenka, Bernt Schiele, and Honglak Lee. Learning what and where to draw. In NIPS, 2016a. + +Scott Reed, Zeynep Akata, Xinchen Yan, Lajanugen Logeswaran, Bernt Schiele, and Honglak Lee. Generative adversarial text-to-image synthesis. In Proceedings of The 33rd International Conference on Machine Learning, 2016b. + +Danilo Jimenez Rezende, Shakir Mohamed, and Daan Wierstra. Stochastic backpropagation and variational inference in deep latent gaussian models. In International Conference on Machine Learning. Citeseer, 2014. + +Tim Salimans, Ian J. Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen. Improved techniques for training gans. arXiv preprint arXiv:1606.03498, 2016. + +Karen Simonyan, Andrea Vedaldi, and Andrew Zisserman. Deep inside convolutional networks: Visualising image classification models and saliency maps. arXiv preprint arXiv:1312.6034, 2013. + +Satinder Singh, Michael Kearns, and Yishay Mansour. Nash convergence of gradient dynamics in general-sum games. In Proceedings of the Sixteenth conference on Uncertainty in artificial intelligence, pp. 541–548. Morgan Kaufmann Publishers Inc., 2000. + +Jascha Sohl-Dickstein, Eric A. Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In Proceedings of The 32nd International Conference on Machine Learning, pp. 2256–2265, 2015. URL http://arxiv.org/abs/ 1503.03585. + +Casper Kaae Sonderby, Jose Caballero, Lucas Theis, Wenzhe Shi, and Ferenc Huszar. Amortised map inference for image super-resolution, 2016. URL https://arxiv.org/abs/1610. 04490v1. + +L. Theis and M. Bethge. Generative image modeling using spatial lstms. In Advances in Neural Information Processing Systems 28, Dec 2015. URL http://arxiv.org/abs/1506. $0 3 4 7 8 /$ . + +L. Theis, A. van den Oord, and M. Bethge. A note on the evaluation of generative models. In International Conference on Learning Representations, Apr 2016. URL http://arxiv.org/ abs/1511.01844. + +T. Tieleman and G. Hinton. Lecture 6.5—RmsProp: Divide the gradient by a running average of its recent magnitude. COURSERA: Neural Networks for Machine Learning, 2012. + +Aaron van den Oord, Nal Kalchbrenner, and Koray Kavukcuoglu. Pixel recurrent neural networks.¨ arXiv preprint arXiv:1601.06759, abs/1601.06759, 2016a. URL http://arxiv.org/abs/ 1601.06759. + +Aaron van den Oord, Nal Kalchbrenner, Oriol Vinyals, Lasse Espeholt, Alex Graves, and Ko-¨ ray Kavukcuoglu. Conditional image generation with pixelcnn decoders. arXiv preprint arXiv:1606.05328, 2016b. + +Pascal Vincent, Hugo Larochelle, Isabelle Lajoie, Yoshua Bengio, and Pierre-Antoine Manzagol. Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion. J. Mach. Learn. Res., 11:3371–3408, December 2010. ISSN 1532-4435. URL http://dl.acm.org/citation.cfm?id=1756006.1953039. + +Jason Yosinski, Jeff Clune, Anh Nguyen, Thomas Fuchs, and Hod Lipson. Understanding neural networks through deep visualization. arXiv preprint arXiv:1506.06579, 2015. + +Chongjie Zhang and Victor R Lesser. Multi-agent learning with policy prediction. In Proceedings of the Twenty-Fourth AAAI Conference on Artificial Intelligence, 2010. + +Junbo Zhao, Michael Mathieu, and Yann LeCun. Energy-based generative adversarial network. arXiv preprint arXiv:1609.03126, 2016. + +Jun-Yan Zhu, Philipp Krahenb ¨ uhl, Eli Shechtman, and Alexei A. Efros. Generative visual manipula- ¨ tion on the natural image manifold. In Proceedings of European Conference on Computer Vision (ECCV), 2016. + +# Appendix + +# A 2D GAUSSIAN TRAINING DETAILS + +Network architecture and experimental details for the experiment in Section 3.1 are as follows: + +The dataset is sampled from a mixture of 8 Gaussians of standard deviation 0.02. The means are equally spaced around a circle of radius 2. + +The generator network consists of a fully connected network with 2 hidden layers of size 128 with relu activations followed by a linear projection to 2 dimensions. All weights are initialized to be orthogonal with scaling of 0.8. + +The discriminator network first scales its input down by a factor of 4 (to roughly scale to (-1,1)), followed by 1 layer fully connected network with relu activations to a linear layer to of size 1 to act as the logit. + +The generator minimizes $\mathcal { L } _ { G } = \log ( D ( x ) ) + \log ( 1 - D ( G ( z ) ) )$ and the discriminator minimizes $\mathcal { L } _ { D } \stackrel { = } { = } - \log ( D ( x ) ) - \log ( 1 - D ( \stackrel { . } { G } ( z ) ) )$ where $\mathbf { X }$ is sampled from the data distribution and $z \sim$ $\mathcal { N } ( 0 , I _ { 2 5 6 } )$ . Both networks are optimized using Adam (Kingma & Ba, 2014) with a learning rate of 1e-4 and $\beta _ { 1 } { = } 0 . 5$ . + +The network is trained by alternating updates of the generator and the discriminator. One step consists of either $\mathbf { G }$ or $\mathbf { D }$ updating. + +# B MORE MIXTURE OF GAUSSIAN EXPERIMENTS + +# B.1 EFFECTS OF TIME DELAY / HISTORICAL AVERAGING + +Another comparison we looked at was with regard to historical averaging based approaches. Recently similarly inspired approaches have been used in (Salimans et al., 2016) to stabilize training. For our study, we looked at taking an ensemble of discriminators over time. + +First, we looked at taking an ensemble of the last N steps, as shown in Figure App.1. + +![](images/ca00dc71a894afed5412e7d9c9476916ab9ab1f2648dedea7d529b310373bd67.jpg) +Figure App.1: Historical averaging does not visibly increase stability on the mixture of Gaussians task. Each row corresponds to an ensemble of discriminators which consists of the indicated number of immediately preceding discriminators. The columns correspond to different numbers of training steps. + +To further explore this idea, we ran experiments with an ensemble of 5 discriminators, but with different periods between replacing discriminators in the ensemble. For example, if I sample at a rate of 100, it would take 500 steps to replace all 5 discriminators. Results can be seen in Figure App.2. + +We observe that given longer and longer time delays, the model becomes less and less stable. We hypothesize that this is due to the initial shape of the discriminator loss surface. When training, the discriminator’s estimates of probability densities are only accurate on regions where it was trained. When fixing this discriminator, we are removing the feedback between the generator exploitation and the discriminators ability to move. As a result, the generator is able to exploit these fixed areas of poor performance for older discriminators in the ensemble. New discriminators (over)compensate for this, leading the system to diverge. + +![](images/04adbd77c06eb06d8623be8ab50fe44136a6a0398a13d8fdd2a583c617dbcff0.jpg) +Figure App.2: Introducing longer time delays between the discriminator ensemble results in instability and probability distributions that are not in the window being visualized. The $\mathbf { X }$ axis is the number of weight updates and the y axis is how many steps to skip between discriminator updates when selecting the ensemble of 5 discriminators. + +# B.2 EFFECTS OF THE SECOND GRADIENT + +A second factor we analyzed is the effect of backpropagating the learning signal through the unrolling in Equation 12. We can turn on or off this backpropagation through the unrolling by introducing stop gradient calls into our computation graph between each unrolling step. With the stop gradient in place, the update signal corresponds only to the first term in Equation 12. We looked at 3 configurations: without stop gradients; vanilla unrolled GAN, with stop gradients; and with stop gradients but taking the average over the $k$ unrolling steps instead of taking the final value. Results can be see in Figure App.3. + +We initially observed no difference between unrolling with and without the second gradient, as both required 3 unrolling steps to become stable. When the discriminator is unrolled to convergence, the second gradient term becomes zero. Due to the simplicity of the problem, we suspect that the discriminator nearly converged for every generator step, and the second gradient term was thus irrelevant. + +To test this, we modified the dynamics to perform five generator steps for each discriminator update. Results are shown in Figure App.4. With the discriminator now kept out of equilibrium, successful training can be achieved with half as many unrolling steps when using both terms in the gradient than when only including the first term. + +# C RNN MNIST TRAINING DETAILS + +The network architecture for the experiment in Section 3.2 is as follows: + +The MNIST dataset is scaled to [-1, 1). + +The generator first scales the 256D noise vector through a 256 unit fully connected layer with relu activation. This is then fed into the initial state of a 256D LSTM(Hochreiter & Schmidhuber, 1997) that runs 28 steps corresponding to the number of columns in MNIST. The resulting sequence of activations is projected through a fully connected layer with 28 outputs with a tanh activation function. All weights are initialized via the ”Xavier” initialization (Glorot & Bengio, 2010). The forget bias on the LSTM is initialized to 1. + +The discriminator network feeds the input into a Convolution(16, stride $^ { = 2 }$ ) followed by a Convolution(32, stride $^ { = 2 }$ ) followed by Convolution(32, stride ${ \boldsymbol { \mathbf { \mathit { \varepsilon } } } } = 2 { \boldsymbol { \mathbf { \mathit { \varepsilon } } } }$ ). All convolutions have stride 2. As in (Radford et al., 2015) leaky rectifiers are used with a 0.3 leak. Batch normalization is applied after each layer (Ioffe & Szegedy, 2015). The resulting 4D tensor is then flattened and a linear projection is performed to a single scalar. + +![](images/e082e1fa15fe3c8c65693a9ca8c06d3ede7c2ddcf2d0287c257abe4e8c2ff9b6.jpg) +Figure App.3: If the discriminator remains nearly at its optimum during learning, then performance is nearly identical with and without the second gradient term in Equation 12. As shown in Figure App.4, when the discriminator lags behind the generator, backpropagating through unrolling aids convergence. + +The generator network minimises $\mathcal { L } _ { G } = \log ( D ( G ( z ) ) )$ and the discriminator minimizes $\mathcal { L } _ { D } =$ $\log ( D ( x ) ) + \log ( 1 - D ( G ( z ) ) )$ . Both networks are trained with Adam(Kingma & Ba, 2014) with learning rates of 1e-4 and $\beta _ { 1 } { = } 0 . 5$ . The network is trained alternating updating the generator and the discriminator for 150k steps. One step consists of just 1 network update. + +# D CIFAR10/MNIST TRAINING DETAILS + +The network architectures for the discriminator, generator, and encoder as as follows. All convolutions have a kernel size of 3x3 with batch normalization and leaky ReLU’s with a 0.3 leak. + +The generator network is defined as: + +
number outputsstride
Input: z ~ N(0,I256)Fully connectedReshape to image 4,4,512ConvolutionConvolutionConvolutionConvolution4*4*512256128641or32221
+ +![](images/ddba8731c308e072b5532fd2f2dfbcf72fd2fc3c5d92dcb601c2c03608beee85.jpg) +Unrolled GAN with 5 G Steps per D without second gradient + +Unrolled GAN with 5 G Steps per D + +
number outputsstride
Input: x~ Pdata or G Transposed Convolution Transposed Convolution Transposed Convolution64 128 256222
Flatten
Fully Connected1
+ +![](images/c5b3fa4bd5bb790af426480883819ba983aa6618ec11502817b7cc7fa39e22d5.jpg) +Figure App.4: Backpropagating through the unrolling process aids convergence when the discriminator does not fully converge between generator updates. When taking 5 generator steps per discriminator step unrolling greatly increases stability, requiring only 5 unrolling steps to converge. Without the second gradient it requires 10 unrolling steps. Also see Figure App.3. + +The discriminator network is defined as: + +The generator network minimises $\mathcal { L } _ { G } = \log ( D ( G ( z ) ) )$ and the discriminator minimizes $\mathcal { L } _ { D } =$ $\log ( D ( x ) ) + \log ( 1 - D ( G ( z ) ) )$ . The networks are trained with Adam with a generator learning rate of 1e-4, and a discriminator learning rate of 2e-4. The network is trained alternating updating the generator and the discriminator for $1 0 0 \mathrm { k }$ steps. One step consists of just 1 network update. + +# E 1000 CLASS MNIST + +
number outputsstride
Input: z ~ N(0,I256)Fully connectedReshape to image 4,4,64ConvolutionConvolutionConvolutionConvolution4 *4*643216832221
+ +The discriminator network is parametrized by a size $\mathrm { X }$ and is defined as follows. In our tests, we used X of 1/4 and 1/2. + +
number outputsstride
Input: x ~ Pdata or G Transposed Convolution Transposed Convolution8*X 16*X222
Transposed Convolution Flatten32*X
Fully Connected1
+ +# F COLORED MNIST DATASET + +# F.1 DATASET + +To generate this dataset we first took the mnist digit, $I$ , scaled between 0 and 1. For each image we sample a color, $C$ , normally distributed with mean $\scriptstyle = 0$ and std $\scriptstyle \mathtt { = 0 . 5 }$ . To generate a colored digit between (-1, 1) we do $I * C + \mathsf { \bar { ( } } I - 1 )$ . Finally, we add a small amount of pixel independent noise sampled from a normal distribution with std $= 0 . 2$ , and the resulting values are cliped between (-1, 1). When visualized, this generates images and samples that can be seen in figure App.5. Once again it is very hard to visually see differences in sample diversity when comparing the 128 and the 512 sized models. + +![](images/44d318f018c78dbdaec0c5211309ab172baeb975108ebe77f750982a4193cef6.jpg) +Figure App.5: Right: samples from the data distribution. Middle: Samples from 1/4 size model with 0 look ahead steps (worst diversity). Left: Samples from 1/1 size model with 10 look ahead steps (most diversity). + +# F.2 MODELS + +The models used in this section are parametrized by a variable $\mathrm { X }$ to control capacity. A value of ${ \bf X } { = } 1$ is same architecture used in the cifar10 experiments. We used 1/4, 1/2 and 1 as these values. + +The generator network is defined as: + +
number outputsstride
Input: z~ N(0,I256)Fully connectedReshape to image 4,4,512*XConvolutionConvolutionConvolutionConvolution4*4*512*X256*X128*X64*X32221
+ +The discriminator network is defined as: + +
number outputsstride
Input: x ~ Pdata or G Transposed Convolution Transposed Convolution Transposed Convolution Flatten Fully Connected64*X 128*X 256*X222
+ +# G OPTIMIZATION BASED VISUALIZATIONS + +More examples of model based optimization. We performed 5 runs with different seeds of each of of the unrolling steps configuration. Bellow are comparisons for each run index. Ideally this would be a many to many comparison, but for space efficiency we grouped the runs by the index in which they were run. + +![](images/8d38e597124cc936be0c8530d6fb7e8b1ecc41f5e50351124daf12a2d78e11fa.jpg) +Figure App.6: Samples from 1/5 with different random seeds. + +![](images/be94e2bb107541e788983b4ad2bdc31bebdff50fbbb8529c915238f8cca809e8.jpg) +Figure App.7: Samples from 2/5 with different random seeds. + +![](images/e17957e3a6c6fffe5b489567eb4830a1720a2ae85eadaf27b1b6017a82527e8b.jpg) +Figure App.8: Samples from 3/5 with different random seeds. + +![](images/fbca68637cf3798a1a739daff1548046fba7251170be090a9d1d669d0aa5603e.jpg) +Figure App.9: Samples from 4/5 with different random seeds. + +![](images/c67273dbe4f3a1a6aee81a0d0854850226af640cdd34215ec9e0663f82c11a7d.jpg) +Figure App.10: Samples from 5/5 with different random seeds. \ No newline at end of file diff --git a/parse/train/BydrOIcle/BydrOIcle_content_list.json b/parse/train/BydrOIcle/BydrOIcle_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..946e84320358b54b71d53206c206f42fbeabf0ef --- /dev/null +++ b/parse/train/BydrOIcle/BydrOIcle_content_list.json @@ -0,0 +1,2528 @@ +[ + { + "type": "text", + "text": "UNROLLED GENERATIVE ADVERSARIAL NETWORKS ", + "text_level": 1, + "bbox": [ + 173, + 99, + 807, + 121 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Luke Metz∗ \nGoogle Brain \nlmetz@google.com \nBen Poole† \nStanford University \npoole@cs.stanford.edu ", + "bbox": [ + 184, + 145, + 343, + 188 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "", + "bbox": [ + 398, + 143, + 607, + 188 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "David Pfau Google DeepMind pfau@google.com ", + "bbox": [ + 663, + 145, + 813, + 188 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Jascha Sohl-Dickstein Google Brain jaschasd@google.com ", + "bbox": [ + 183, + 208, + 372, + 251 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "ABSTRACT ", + "text_level": 1, + "bbox": [ + 454, + 287, + 544, + 303 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "We introduce a method to stabilize Generative Adversarial Networks (GANs) by defining the generator objective with respect to an unrolled optimization of the discriminator. This allows training to be adjusted between using the optimal discriminator in the generator’s objective, which is ideal but infeasible in practice, and using the current value of the discriminator, which is often unstable and leads to poor solutions. We show how this technique solves the common problem of mode collapse, stabilizes training of GANs with complex recurrent generators, and increases diversity and coverage of the data distribution by the generator. ", + "bbox": [ + 233, + 318, + 764, + 430 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 INTRODUCTION ", + "text_level": 1, + "bbox": [ + 178, + 457, + 336, + 473 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "The use of deep neural networks as generative models for complex data has made great advances in recent years. This success has been achieved through a surprising diversity of training losses and model architectures, including denoising autoencoders (Vincent et al., 2010), variational autoencoders (Kingma & Welling, 2013; Rezende et al., 2014; Gregor et al., 2015; Kulkarni et al., 2015; Burda et al., 2015; Kingma et al., 2016), generative stochastic networks (Alain et al., 2015), diffusion probabilistic models (Sohl-Dickstein et al., 2015), autoregressive models (Theis & Bethge, 2015; van den Oord et al., 2016a;b), real non-volume preserving transformations (Dinh et al., 2014; 2016), Helmholtz machines (Dayan et al., 1995; Bornschein et al., 2015), and Generative Adversarial Networks (GANs) (Goodfellow et al., 2014). ", + "bbox": [ + 174, + 489, + 825, + 613 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1.1 GENERATIVE ADVERSARIAL NETWORKS ", + "text_level": 1, + "bbox": [ + 178, + 631, + 496, + 645 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "While most deep generative models are trained by maximizing log likelihood or a lower bound on log likelihood, GANs take a radically different approach that does not require inference or explicit calculation of the data likelihood. Instead, two models are used to solve a minimax game: a generator which samples data, and a discriminator which classifies the data as real or generated. In theory these models are capable of modeling an arbitrarily complex probability distribution. When using the optimal discriminator for a given class of generators, the original GAN proposed by Goodfellow et al. minimizes the Jensen-Shannon divergence between the data distribution and the generator, and extensions generalize this to a wider class of divergences (Nowozin et al., 2016; Sonderby et al., 2016; Poole et al., 2016). ", + "bbox": [ + 174, + 657, + 825, + 782 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "The ability to train extremely flexible generating functions, without explicitly computing likelihoods or performing inference, and while targeting more mode-seeking divergences as made GANs extremely successful in image generation (Odena et al., 2016; Salimans et al., 2016; Radford et al., 2015), and image super resolution (Ledig et al., 2016). The flexibility of the GAN framework has also enabled a number of successful extensions of the technique, for instance for structured prediction (Reed et al., 2016a;b; Odena et al., 2016), training energy based models (Zhao et al., 2016), and combining the GAN loss with a mutual information loss (Chen et al., 2016). ", + "bbox": [ + 174, + 790, + 825, + 887 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "In practice, however, GANs suffer from many issues, particularly during training. One common failure mode involves the generator collapsing to produce only a single sample or a small family of very similar samples. Another involves the generator and discriminator oscillating during training, rather than converging to a fixed point. In addition, if one agent becomes much more powerful than the other, the learning signal to the other agent becomes useless, and the system does not learn. To train GANs many tricks must be employed, such as careful selection of architectures (Radford et al., 2015), minibatch discrimination (Salimans et al., 2016), and noise injection (Salimans et al., 2016; Sonderby et al., 2016). Even with these tricks the set of hyperparameters for which training is successful is generally very small in practice. ", + "bbox": [ + 174, + 103, + 825, + 229 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Once converged, the generative models produced by the GAN training procedure normally do not cover the whole distribution (Dumoulin et al., 2016; Che et al., 2016), even when targeting a modecovering divergence such as KL. Additionally, because it is intractable to compute the GAN training loss, and because approximate measures of performance such as Parzen window estimates suffer from major flaws (Theis et al., 2016), evaluation of GAN performance is challenging. Currently, human judgement of sample quality is one of the leading metrics for evaluating GANs. In practice this metric does not take into account mode dropping if the number of modes is greater than the number of samples one is visualizing. In fact, the mode dropping problem generally helps visual sample quality as the model can choose to focus on only the most common modes. These common modes correspond, by definition, to more typical samples. Additionally, the generative model is able to allocate more expressive power to the modes it does cover than it would if it attempted to cover all modes. ", + "bbox": [ + 174, + 236, + 825, + 401 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "1.2 DIFFERENTIATING THROUGH OPTIMIZATION", + "text_level": 1, + "bbox": [ + 176, + 419, + 526, + 433 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Many optimization schemes, including SGD, RMSProp (Tieleman & Hinton, 2012), and Adam (Kingma & Ba, 2014), consist of a sequence of differentiable updates to parameters. Gradients can be backpropagated through unrolled optimization updates in a similar fashion to backpropagation through a recurrent neural network. The parameters output by the optimizer can thus be included, in a differentiable way, in another objective (Maclaurin et al., 2015). This idea was first suggested for minimax problems in (Pearlmutter & Siskind, 2008), while (Zhang & Lesser, 2010) provided a theoretical analysis and experimental results on differentiating through a single step of gradient ascent for simple matrix games. Differentiating through unrolled optimization was first scaled to deep networks in (Maclaurin et al., 2015), where it was used for hyperparameter optimization. More recently, (Belanger & McCallum, 2015; Han et al., 2016; Andrychowicz et al., 2016) backpropagate through optimization procedures in contexts unrelated to GANs or minimax games. ", + "bbox": [ + 174, + 444, + 825, + 598 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "In this work we address the challenges of unstable optimization and mode collapse in GANs by unrolling optimization of the discriminator objective during training. ", + "bbox": [ + 173, + 604, + 821, + 633 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2 METHOD ", + "text_level": 1, + "bbox": [ + 174, + 652, + 282, + 669 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2.1 GENERATIVE ADVERSARIAL NETWORKS ", + "text_level": 1, + "bbox": [ + 174, + 684, + 498, + 698 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "The GAN learning problem is to find the optimal parameters $\\theta _ { G } ^ { * }$ for a generator function $G \\left( z ; \\theta _ { G } \\right)$ in a minimax objective, ", + "bbox": [ + 174, + 709, + 821, + 738 + ], + "page_idx": 1 + }, + { + "type": "equation", + "img_path": "images/b2008d380b8f084085a49f2a632840f8a4144bbae5800b215f3679c79653344b.jpg", + "text": "$$\n\\begin{array} { c } { { \\theta _ { G } ^ { * } = \\underset { \\theta _ { G } } { \\mathrm { a r g m i n } } \\underset { \\theta _ { D } } { \\mathrm { m a x } } f \\left( \\theta _ { G } , \\theta _ { D } \\right) } } \\\\ { { \\ \\mathrm { ~ } } } \\\\ { { \\displaystyle \\qquad = \\underset { \\theta _ { G } } { \\mathrm { a r g m i n } } f \\left( \\theta _ { G } , \\theta _ { D } ^ { * } \\left( \\theta _ { G } \\right) \\right) } } \\\\ { { \\theta _ { D } ^ { * } \\left( \\theta _ { G } \\right) = \\underset { \\theta _ { D } } { \\mathrm { a r g m a x } } f \\left( \\theta _ { G } , \\theta _ { D } \\right) , } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 382, + 741, + 617, + 825 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "where $f$ is commonly chosen to be ", + "bbox": [ + 176, + 828, + 405, + 843 + ], + "page_idx": 1 + }, + { + "type": "equation", + "img_path": "images/9ee6df4b3e18c286fdee617b2aa953c06b25b1f3424af4b30db2853ba740dd9e.jpg", + "text": "$$\nf \\left( \\theta _ { G } , \\theta _ { D } \\right) = \\mathbb { E } _ { x \\sim p _ { d a t a } } \\left[ \\log \\left( D \\left( x ; \\theta _ { D } \\right) \\right) \\right] + \\mathbb { E } _ { z \\sim N ( 0 , I ) } \\left[ \\log \\left( 1 - D \\left( G \\left( z ; \\theta _ { G } \\right) ; \\theta _ { D } \\right) \\right) \\right] .\n$$", + "text_format": "latex", + "bbox": [ + 217, + 847, + 779, + 864 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Here $x \\in \\mathcal { X }$ is the data variable, $z \\in { \\mathcal { Z } }$ is the latent variable, $p _ { d a t a }$ is the data distribution, the discriminator $D ( \\cdot ; \\theta _ { D } ) : \\mathcal { X } [ 0 , 1 ]$ outputs the estimated probability that a sample $x$ comes from the data distribution, $\\theta _ { D }$ and $\\theta _ { G }$ are the discriminator and generator parameters, and the generator function $G ( \\cdot ; \\theta _ { G } ) : \\mathcal { Z } \\mathcal { X }$ transforms a sample in the latent space into a sample in the data space. ", + "bbox": [ + 174, + 867, + 825, + 924 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "For the minimax loss in Eq. 4, the optimal discriminator $D ^ { \\ast } \\left( x \\right)$ is a known smooth function of the generator probability $p _ { G } \\left( x \\right)$ (Goodfellow et al., 2014), ", + "bbox": [ + 171, + 103, + 823, + 132 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/0e2376c104ce7eecf400d8575cb56227acb0e45882302fbde8854a5f2d2b2c6b.jpg", + "text": "$$\nD ^ { * } \\left( x \\right) = \\frac { p _ { d a t a } \\left( x \\right) } { p _ { d a t a } \\left( x \\right) + p _ { G } \\left( x \\right) } .\n$$", + "text_format": "latex", + "bbox": [ + 397, + 137, + 601, + 172 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "When the generator loss in Eq. 2 is rewritten directly in terms of $p _ { G } \\left( x \\right)$ and Eq. 5 rather than $\\theta _ { G }$ and $\\theta _ { D } ^ { * } \\left( \\theta _ { G } ^ { - } \\right)$ , then it is similarly a smooth function of $p _ { G } \\left( x \\right)$ . These smoothness guarantees are typically lost when $D \\left( x ; \\theta _ { D } \\right)$ and $G \\left( z ; \\theta _ { G } \\right)$ are drawn from parametric families. They nonetheless suggest that the true generator objective in Eq. 2 will often be well behaved, and is a desirable target for direct optimization. ", + "bbox": [ + 174, + 178, + 825, + 248 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Explicitly solving for the optimal discriminator parameters $\\theta _ { D } ^ { * } \\left( \\theta _ { G } \\right) $ for every update step of the generator $G$ is computationally infeasible for discriminators based on neural networks. Therefore this minimax optimization problem is typically solved by alternating gradient descent on $\\theta _ { G }$ and ascent on $\\theta _ { D }$ . ", + "bbox": [ + 174, + 255, + 825, + 311 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "The optimal solution $\\theta ^ { * } = \\{ \\theta _ { G } ^ { * } , \\theta _ { D } ^ { * } \\}$ is a fixed point of these iterative learning dynamics. Additionally, if $f \\left( { \\theta } _ { G } , { \\theta } _ { D } \\right)$ is convex in $\\theta _ { G }$ and concave in $\\theta _ { D }$ , then alternating gradient descent (ascent) trust region updates are guaranteed to converge to the fixed point, under certain additional weak assumptions (Juditsky et al., 2011). However in practice $f \\left( { \\theta } _ { G } , { \\theta } _ { D } \\right)$ is typically very far from convex in $\\theta _ { G }$ and concave in $\\theta _ { D }$ , and updates are not constrained in an appropriate way. As a result GAN training suffers from mode collapse, undamped oscillations, and other problems detailed in Section 1.1. In order to address these difficulties, we will introduce a surrogate objective function $f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right)$ for training the generator which more closely resembles the true generator objective $f \\left( \\theta _ { G } , \\theta _ { D } ^ { * } \\left( \\theta _ { G } \\right) \\right)$ . ", + "bbox": [ + 173, + 318, + 825, + 430 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "2.2 UNROLLING GANS ", + "text_level": 1, + "bbox": [ + 176, + 445, + 349, + 460 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "A local optimum of the discriminator parameters $\\theta _ { D } ^ { * }$ can be expressed as the fixed point of an iterative optimization procedure, ", + "bbox": [ + 173, + 472, + 825, + 501 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/f45244ba45a91984f7cab7a13e843b9f8428ca6f8c95d11c332b9e18669248b2.jpg", + "text": "$$\n\\begin{array} { c } { { \\theta _ { D } ^ { 0 } = \\theta _ { D } } } \\\\ { { \\displaystyle } } \\\\ { { \\theta _ { D } ^ { k + 1 } = \\theta _ { D } ^ { k } + \\eta ^ { k } \\displaystyle \\frac { \\mathrm { d } f ( \\theta _ { G } , \\theta _ { D } ^ { k } ) } { \\mathrm { d } \\theta _ { D } ^ { k } } } } \\\\ { { \\displaystyle } } \\\\ { { \\theta _ { D } ^ { * } ( \\theta _ { G } ) = \\displaystyle \\operatorname* { l i m } _ { k \\infty } \\theta _ { D } ^ { k } , } } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 387, + 503, + 609, + 590 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where $\\eta ^ { k }$ is the learning rate schedule. For clarity, we have expressed Eq. 7 as a full batch steepest gradient ascent equation. More sophisticated optimizers can be similarly unrolled. In our experiments we unroll Adam (Kingma & Ba, 2014). ", + "bbox": [ + 174, + 597, + 823, + 638 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "By unrolling for $K$ steps, we create a surrogate objective for the update of the generator, ", + "bbox": [ + 171, + 645, + 751, + 660 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/722de6d2f0bca528a8df9e9213bb3bcba721cff99ce6b981b4c7ab2eda98fd02.jpg", + "text": "$$\nf _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) = f \\left( \\theta _ { G } , \\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) \\right) .\n$$", + "text_format": "latex", + "bbox": [ + 374, + 665, + 622, + 685 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "When $K = 0$ this objective corresponds exactly to the standard GAN objective, while as $K \\infty$ it corresponds to the true generator objective function $f \\left( \\theta _ { G } , \\theta _ { D } ^ { * } \\left( G \\right) \\right)$ . By adjusting the number of unrolling steps $K$ , we are thus able to interpolate between standard GAN training dynamics with their associated pathologies, and more costly gradient descent on the true generator loss. ", + "bbox": [ + 174, + 690, + 823, + 747 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "2.3 PARAMETER UPDATES ", + "text_level": 1, + "bbox": [ + 174, + 763, + 370, + 777 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "The generator and discriminator parameter updates using this surrogate loss are ", + "bbox": [ + 173, + 789, + 694, + 804 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/35895e40c05db7fa3394ccde440178bbdbe1633ad2e976da444caa1f30f35303.jpg", + "text": "$$\n\\begin{array} { r l } & { \\theta _ { G } \\theta _ { G } - \\eta \\frac { \\mathrm { d } f _ { K } ( \\theta _ { G } , \\theta _ { D } ) } { \\mathrm { d } \\theta _ { G } } } \\\\ & { \\theta _ { D } \\theta _ { D } + \\eta \\frac { \\mathrm { d } f ( \\theta _ { G } , \\theta _ { D } ) } { \\mathrm { d } \\theta _ { D } } . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 403, + 809, + 593, + 880 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "For clarity we use full batch steepest gradient descent (ascent) with stepsize $\\eta$ above, while in experiments we instead use minibatch Adam for both updates. The gradient in Eq. 10 requires backpropagating through the optimization process in Eq. 7. A clear description of differentiation through gradient descent is given as Algorithm 2 in (Maclaurin et al., 2015), though in practice the use of an automatic differentiation package means this step does not need to be programmed explicitly. A pictorial representation of these updates is provided in Figure 1. ", + "bbox": [ + 174, + 882, + 823, + 924 + ], + "page_idx": 2 + }, + { + "type": "image", + "img_path": "images/b8d41d44f95393092b08110ba25482b02bb1bcb0a2930f383731e4e370791ca3.jpg", + "image_caption": [ + "Figure 1: An illustration of the computation graph for an unrolled GAN with 3 unrolling steps. The generator update in Equation 10 involves backpropagating the generator gradient (blue arrows) through the unrolled optimization. Each step $k$ in the unrolled optimization uses the gradients of $f _ { k }$ with respect to $\\theta _ { D } ^ { k }$ , as described in Equation 7 and indicated by the green arrows. The discriminator update in Equation 11 does not depend on the unrolled optimization (red arrow). " + ], + "image_footnote": [], + "bbox": [ + 181, + 123, + 821, + 258 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 393, + 825, + 436 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "It is important to distinguish this from an approach suggested in (Goodfellow et al., 2014), that several update steps of the discriminator parameters should be run before each single update step for the generator. In that approach, the update steps for both models are still gradient descent (ascent) with respect to fixed values of the other model parameters, rather than the surrogate loss we describe in Eq. 9. Performing $K$ steps of discriminator update between each single step of generator update corresponds to updating the generator parameters $\\theta _ { G }$ using only the first term in Eq. 12 below. ", + "bbox": [ + 173, + 443, + 825, + 527 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "2.4 THE MISSING GRADIENT TERM ", + "text_level": 1, + "bbox": [ + 176, + 551, + 436, + 565 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "To better understand the behavior of the surrogate loss $f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right)$ , we examine its gradient with respect to the generator parameters $\\theta _ { G }$ , ", + "bbox": [ + 173, + 580, + 823, + 609 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/791d5164930dd3fb042c3335cdcf143313e5660a7175699577db372f61561f22.jpg", + "text": "$$\n\\frac { \\mathrm { d } f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) } { \\mathrm { d } \\theta _ { G } } = \\frac { \\partial f \\left( \\theta _ { G } , \\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) \\right) } { \\partial \\theta _ { G } } + \\frac { \\partial f \\left( \\theta _ { G } , \\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) \\right) } { \\partial \\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) } \\frac { \\mathrm { d } \\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) } { \\mathrm { d } \\theta _ { G } } .\n$$", + "text_format": "latex", + "bbox": [ + 232, + 622, + 766, + 660 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Standard GAN training corresponds exactly to updating the generator parameters using only the first term in this gradient, with $\\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) $ being the parameters resulting from the discriminator update step. An optimal generator for any fixed discriminator is a delta function at the $x$ to which the discriminator assigns highest data probability. Therefore, in standard GAN training, each generator update step is a partial collapse towards a delta function. ", + "bbox": [ + 173, + 679, + 825, + 750 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "The second term captures how the discriminator would react to a change in the generator. It reduces the tendency of the generator to engage in mode collapse. For instance, the second term reflects that as the generator collapses towards a delta function, the discriminator reacts and assigns lower probability to that state, increasing the generator loss. It therefore discourages the generator from collapsing, and may improve stability. ", + "bbox": [ + 173, + 756, + 825, + 827 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "As $K \\infty$ , $\\theta _ { D } ^ { K }$ goes to a local optimum of $f$ , where $\\frac { \\partial f } { \\partial \\theta _ { D } ^ { K } } = 0$ , and therefore the second term in Eq. 12 goes to 0 (Danskin, 1967). The gradient of the unrolled surrogate loss $f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right)$ with respect to $\\theta _ { G }$ is thus identical to the gradient of the standard GAN loss $f \\bar { ( \\theta _ { G } , \\theta _ { D } ) }$ both when $K = 0$ and when $K \\infty$ , where we take $K \\infty$ to imply that in the standard GAN the discriminator is also fully optimized between each generator update. Between these two extremes, $f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right)$ captures additional information about the response of the discriminator to changes in the generator. ", + "bbox": [ + 173, + 832, + 825, + 924 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "2.5 CONSEQUENCES OF THE SURROGATE LOSS", + "text_level": 1, + "bbox": [ + 178, + 103, + 511, + 117 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "GANs can be thought of as a game between the discriminator $( D )$ and the generator $( G )$ . The agents take turns taking actions and updating their parameters until a Nash equilibrium is reached. The optimal action for $D$ is to evaluate the probability ratio $\\frac { p _ { d a t a } ( x ) } { p _ { G } ( x ) + p _ { d a t a } ( x ) }$ for the generator’s move $x$ (Eq. 5). The optimal generator action is to move its mass to maximize this ratio. ", + "bbox": [ + 174, + 131, + 825, + 191 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "The initial move for $G$ will be to move as much mass as its parametric family and update step permits to the single point that maximizes the ratio of probability densities. The action $D$ will then take is quite simple. It will track that point, and to the extent allowed by its own parametric family and update step assign low data probability to it, and uniform probability everywhere else. This cycle of $G$ moving and $D$ following will repeat forever or converge depending on the rate of change of the two agents. This is similar to the situation in simple matrix games like rock-paper-scissors and matching pennies, where alternating gradient descent (ascent) with a fixed learning rate is known not to converge (Singh et al., 2000; Bowling & Veloso, 2002). ", + "bbox": [ + 174, + 199, + 825, + 310 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "In the unrolled case, however, this undesirable behavior no longer occurs. Now $G$ ’s actions take into account how $D$ will respond. In particular, $G$ will try to make steps that $D$ will have a hard time responding to. This extra information helps the generator spread its mass to make the next $D$ step less effective instead of collapsing to a point. ", + "bbox": [ + 174, + 318, + 825, + 373 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "In principle, a surrogate loss function could be used for both $D$ and $G$ . In the case of 1-step unrolled optimization this is known to lead to convergence for games in which gradient descent (ascent) fails (Zhang & Lesser, 2010). However, the motivation for using the surrogate generator loss in Section 2.2, of unrolling the inner of two nested min and max functions, does not apply to using a surrogate discriminator loss. Additionally, it is more common for the discriminator to overpower the generator than vice-versa when training a GAN. Giving more information to $G$ by allowing it to ‘see into the future’ may thus help the two models be more balanced. ", + "bbox": [ + 174, + 380, + 825, + 478 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "3 EXPERIMENTS ", + "text_level": 1, + "bbox": [ + 176, + 501, + 326, + 516 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "In this section we demonstrate improved mode coverage and stability by applying this technique to five datasets of increasing complexity. Evaluation of generative models is a notoriously hard problem (Theis et al., 2016). As such the de facto standard in GAN literature has become sample quality as evaluated by a human and/or evaluated by a heuristic (Inception score for example, (Salimans et al., 2016)). While these evaluation metrics do a reasonable job capturing sample quality, they fail to capture sample diversity. In our first 2 experiments diversity is easily evaluated via visual inspection. In our later experiments this is not the case, and we will use a variety of methods to quantify coverage of samples. Our measures are individually strongly suggestive of unrolling reducing mode-collapse and improving stability, but none of them alone are conclusive. We believe that taken together however, they provide extremely compelling evidence for the advantages of unrolling. ", + "bbox": [ + 174, + 534, + 825, + 674 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "When doing stochastic optimization, we must choose which minibatches to use in the unrolling updates in Eq. 7. We experimented with both a fixed minibatch and re-sampled minibatches for each unrolling step, and found it did not significantly impact the result. We use fixed minibatches for all experiments in this section. ", + "bbox": [ + 176, + 680, + 825, + 736 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "We provide a reference implementation of this technique at github.com/poolio/unrolled gan. ", + "bbox": [ + 173, + 743, + 774, + 757 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "3.1 MIXTURE OF GAUSSIANS DATASET ", + "text_level": 1, + "bbox": [ + 176, + 777, + 459, + 791 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "To illustrate the impact of discriminator unrolling, we train a simple GAN architecture on a 2D mixture of 8 Gaussians arranged in a circle. For a detailed list of architecture and hyperparameters see Appendix A. Figure 2 shows the dynamics of this model through time. Without unrolling the generator rotates around the valid modes of the data distribution but is never able to spread out mass. When adding in unrolling steps $\\mathbf { G }$ quickly learns to spread probability mass and the system converges to the data distribution. ", + "bbox": [ + 174, + 804, + 823, + 888 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "In Appendix B we perform further experiments on this toy dataset. We explore how unrolling compares to historical averaging, and compares to using the unrolled discriminator to update the generator, but without backpropagating through the generator. In both cases we find that the unrolled objective performs better. ", + "bbox": [ + 174, + 895, + 823, + 924 + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/a4459b06fa7e9fcdf667d911bf408b5c5ccfe73af1d14e29428d98326008c54a.jpg", + "image_caption": [ + "Figure 2: Unrolling the discriminator stabilizes GAN training on a toy 2D mixture of Gaussians dataset. Columns show a heatmap of the generator distribution after increasing numbers of training steps. The final column shows the data distribution. The top row shows training for a GAN with 10 unrolling steps. Its generator quickly spreads out and converges to the target distribution. The bottom row shows standard GAN training. The generator rotates through the modes of the data distribution. It never converges to a fixed distribution, and only ever assigns significant probability mass to a single data mode at once. " + ], + "image_footnote": [], + "bbox": [ + 174, + 101, + 823, + 207 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/ff67f1c079a9b83431ee462581bfb81893ea4eadc9da332a9fff1968becf8c08.jpg", + "image_caption": [ + "Figure 3: Unrolled GAN training increases stability for an RNN generator and convolutional discriminator trained on MNIST. The top row was run with 20 unrolling steps. The bottom row is a standard GAN, with 0 unrolling steps. Images are samples from the generator after the indicated number of training steps. " + ], + "image_footnote": [], + "bbox": [ + 173, + 334, + 825, + 599 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 696, + 823, + 726 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "3.2 PATHOLOGICAL MODEL WITH MISMATCHED GENERATOR AND DISCRIMINATOR ", + "text_level": 1, + "bbox": [ + 173, + 744, + 756, + 758 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "To evaluate the ability of this approach to improve trainability, we look to a traditionally challenging family of models to train – recurrent neural networks (RNNs). In this experiment we try to generate MNIST samples using an LSTM (Hochreiter & Schmidhuber, 1997). MNIST digits are $2 8 \\mathbf { x } 2 8$ pixel images. At each timestep of the generator LSTM, it outputs one column of this image, so that after 28 timesteps it has output the entire sample. We use a convolutional neural network as the discriminator. See Appendix C for the full model and training details. Unlike in all previously successful GAN models, there is no symmetry between the generator and the discriminator in this task, resulting in a more complex power balance. Results can be seen in Figure 3. Once again, without unrolling the model quickly collapses, and rotates through a sequence of single modes. Instead of rotating spatially, it cycles through proto-digit like blobs. When running with unrolling steps the generator disperses and appears to cover the whole data distribution, as in the 2D example. ", + "bbox": [ + 174, + 770, + 825, + 924 + ], + "page_idx": 5 + }, + { + "type": "table", + "img_path": "images/e4cb5f019901d91438706efd6a3956d78e353190e6df31aa58c09fecfa15f59b.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
Discriminator SizeUnrolling steps01510
1/4 size of D compared to GModes generated30.6± 20.7365.4 ± 34.75236.4 ± 63.30327.2 ± 74.67
KL(model||data)5.99± 0.425.911 ± 0.144.67 ± 0.434.66 ± 0.46
1/2 size of D compared to GModes generated628.0± 140.9523.6± 55.768732.0± 44.98817.4 ± 37.91
KL(model||data)2.58 ±0.7512.44 ±0.261.66 ± 0.0901.43 ± 0.12
", + "bbox": [ + 96, + 101, + 900, + 175 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Table 1: Unrolled GANs cover more discrete modes when modeling a dataset with 1,000 data modes, corresponding to all combinations of three MNIST digits $[ 1 0 ^ { 3 }$ digit combinations). The number of modes covered is given for different numbers of unrolling steps, and for two different architectures. The reverse KL divergence between model and data is also given. Standard error is provided for both measures. ", + "bbox": [ + 173, + 185, + 825, + 256 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "3.3 MODE AND MANIFOLD COLLAPSE USING AUGMENTED MNIST ", + "text_level": 1, + "bbox": [ + 176, + 284, + 648, + 296 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "GANs suffer from two different types of model collapse – collapse to a subset of data modes, and collapse to a sub-manifold within the data distribution. In these experiments we isolate both effects using artificially constructed datasets, and demonstrate that unrolling can largely rescue both types of collapse. ", + "bbox": [ + 174, + 309, + 825, + 366 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "3.3.1 DISCRETE MODE COLLAPSE ", + "text_level": 1, + "bbox": [ + 176, + 386, + 421, + 400 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "To explore the degree to which GANs drop discrete modes in a dataset, we use a technique similar to one from (Che et al., 2016). We construct a dataset by stacking three randomly chosen MNIST digits, so as to construct an RGB image with a different MNIST digit in each color channel. This new dataset has 1,000 distinct modes, corresponding to each combination of the ten MNIST classes in the three channels. ", + "bbox": [ + 174, + 411, + 823, + 481 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "We train a GAN on this dataset, and generate samples from the trained model (25,600 samples for all experiments). We then compute the predicted class label of each color channel using a pre-trained MNIST classifier. To evaluate performance, we use two metrics: the number of modes for which the generator produced at least one sample, and the KL divergence between the model and the expected data distribution. Within this discrete label space, a KL divergence can be estimated tractably between the generated samples and the data distribution over classes, where the data distribution is a uniform distribution over all 1,000 classes. ", + "bbox": [ + 174, + 488, + 825, + 585 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "As presented in Table 1, as the number of unrolling steps is increased, both mode coverage and reverse KL divergence improve. Contrary to (Che et al., 2016), we found that reasonably sized models (such as the one used in Section 3.4) covered all 1,000 modes even without unrolling. As such we use smaller convolutional GAN models. Details on the models used are provided in Appendix E. ", + "bbox": [ + 174, + 593, + 825, + 648 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "We observe an additional interesting effect in this experiment. The benefits of unrolling increase as the discriminator size is reduced. We believe unrolling effectively increases the capacity of the discriminator. The unrolled discriminator can better react to any specific way in which the generator is producing non-data-like samples. When the discriminator is weak, the positive impact of unrolling is thus larger. ", + "bbox": [ + 174, + 655, + 825, + 726 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "3.3.2 MANIFOLD COLLAPSE ", + "text_level": 1, + "bbox": [ + 174, + 744, + 383, + 758 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "In addition to discrete modes, we examine the effect of unrolling when modeling continuous manifolds. To get at this quantity, we constructed a dataset consisting of colored MNIST digits. Unlike in the previous experiment, a single MNIST digit was chosen, and then assigned a single monochromatic color. With a perfect generator, one should be able to recover the distribution of colors used to generate the digits. We use colored MNIST digits so that the generator also has to model the digits, which makes the task sufficiently complex that the generator is unable to perfectly solve it. The color of each digit is sampled from a 3D normal distribution. Details of this dataset are provided in Appendix F. We will examine the distribution of colors in the samples generated by the trained GAN. As will also be true in the CIFAR10 example in Section 3.4, the lack of diversity in generated colors is almost invisible using only visual inspection of the samples. Samples can be found in Appendix F. ", + "bbox": [ + 174, + 770, + 825, + 924 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/bdf448f52fbc7d7ff1a8589d6a25dd7b7fe2e14cdf8e504055b7f3a333d00a6e.jpg", + "table_caption": [ + "Table 2: Unrolled GANs better model a continuous distribution. GANs are trained to model randomly colored MNIST digits, where the color is drawn from a Gaussian distribution. The JS divergence between the data and model distributions over digit colors is then reported, along with standard error in the JS divergence. More unrolling steps, and larger models, lead to better JS divergence. " + ], + "table_footnote": [], + "table_body": "
Unrolling steps01510
JS divergence with 1/4 layer size0.073 ± 0.00580.142 ± 0.0280.049 ± 0.00210.075 ± 0.012
JS divergence with 1/2 layer size0.095 ± 0.0110.119 ± 0.0100.055 ± 0.00490.074± 0.016
JS divergence with 1/1 layer size0.034 ± 0.00340.050± 0.00260.027 ± 0.00280.025 ± 0.00076
", + "bbox": [ + 133, + 101, + 864, + 161 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/3c9a4f15816f9475e20e64fe6d880999dbc71b9039d3a86102ace92afbbde1f2.jpg", + "image_caption": [ + "Figure 4: Visual perception of sample quality and diversity is very similar for models trained with different numbers of unrolling steps. Actual sample diversity is higher with more unrolling steps. Each pane shows samples generated after training a model on CIFAR10 with 0, 1, 5, and 10 steps of unrolling. " + ], + "image_footnote": [], + "bbox": [ + 184, + 242, + 797, + 439 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "In order to recover the color the GAN assigned to the digit, we used k-means with 2 clusters, to pick out the foreground color from the background. We then performed this transformation for both the training data and the generated images. Next we fit a Gaussian kernel density estimator to both distributions over digit colors. Finally, we computed the JS divergence between the model and data distributions over colors. Results can be found in Table 2 for several model sizes. Details of the models are provided in Appendix F. ", + "bbox": [ + 174, + 554, + 825, + 637 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "In general, the best performing models are unrolled for 5-10 steps, and larger models perform better than smaller models. Counter-intuitively, taking 1 unrolling step seems to hurt this measure of diversity. We suspect that this is due to it introducing oscillatory dynamics into training. Taking more unrolling steps however leads to improved performance with unrolling. ", + "bbox": [ + 174, + 645, + 825, + 700 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "3.4 IMAGE MODELING OF CIFAR10 ", + "text_level": 1, + "bbox": [ + 176, + 722, + 436, + 736 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Here we test our technique on a more traditional convolutional GAN architecture and task, similar to those used in (Radford et al., 2015; Salimans et al., 2016). In the previous experiments we tested models where the standard GAN training algorithm would not converge. In this section we improve a standard model by reducing its tendency to engage in mode collapse. We ran 4 configurations of this model, varying the number of unrolling steps to be 0, 1, 5, or 10. Each configuration was run 5 times with different random seeds. For full training details see Appendix D. Samples from each of the 4 configurations can be found in Figure 4. There is no obvious difference in visual quality across these model configurations. Visual inspection however provides only a poor measure of sample diversity. ", + "bbox": [ + 173, + 750, + 825, + 875 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "By training with an unrolled discriminator, we expect to generate more diverse samples which more closely resemble the underlying data distribution. We introduce two techniques to examine sample diversity: inference via optimization, and pairwise distance distributions. ", + "bbox": [ + 176, + 882, + 823, + 924 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/b50d528ebc1d7e4f4ff566671309621016069f0eaa4be9ec232cad39b3046c6b.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
Unrolling Steps0 steps1 step5 steps10 steps
Average MSE0.0231± 0.00240.0195 ± 0.00210.0200± 0.00230.0181± 0.0018
PercentBestRank0.63%22.97%15.31%61.09 %
", + "bbox": [ + 169, + 102, + 830, + 146 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Table 3: GANs trained with unrolling are better able to match images in the training set than standard GANs, likely due to mode dropping by the standard GAN. Results show the MSE between training images and the best reconstruction for a model with the given number of unrolling steps. The fraction of training images best reconstructed by a given model is given in the final column. The best reconstructions is found by optimizing the latent representation $z$ to produce the closest matching pixel output $G \\left( z ; \\theta _ { G } \\right)$ . Results are averaged over all 5 runs of each model with different random seeds. ", + "bbox": [ + 173, + 156, + 826, + 255 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "3.4.1 INFERENCE VIA OPTIMIZATION ", + "text_level": 1, + "bbox": [ + 176, + 277, + 444, + 291 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Since likelihood cannot be tractably computed, over-fitting of GANs is typically tested by taking samples and computing the nearest-neighbor images in pixel space from the training data (Goodfellow et al., 2014). We will do the reverse, and measure the ability of the generative model to generate images that look like specific samples from the training data. If we did this by generating random samples from the model, we would need an exponentially large number of samples. We instead treat finding the nearest neighbor $x _ { \\mathrm { n e a r e s t } }$ to a target image $x _ { \\mathrm { { t a r g e t } } }$ as an optimization task, ", + "bbox": [ + 174, + 300, + 825, + 386 + ], + "page_idx": 8 + }, + { + "type": "equation", + "img_path": "images/8ebe230db022f5349155b2c7b8f508d347322a1d01e98e07a32852a61e22205d.jpg", + "text": "$$\n\\begin{array} { r l } & { z _ { \\mathrm { n e a r e s t } } = \\underset { z } { \\mathrm { a r g m i n } } | | G ( z ; \\theta _ { G } ) - x _ { \\mathrm { t a r g e t } } | \\rvert _ { 2 } ^ { 2 } } \\\\ & { x _ { \\mathrm { n e a r e s t } } = G ( z _ { \\mathrm { n e a r e s t } } ; \\theta _ { G } ) . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 367, + 391, + 632, + 440 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "This concept of backpropagating to generate images has been widely used in visualizing features from discriminative networks (Simonyan et al., 2013; Yosinski et al., 2015; Nguyen et al., 2016) and has been applied to explore the visual manifold of GANs in (Zhu et al., 2016). ", + "bbox": [ + 174, + 450, + 825, + 492 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "We apply this technique to each of the models trained. We optimize with 3 random starts using LBFGS, which is the optimizer typically used in similar settings such as style transfer (Johnson et al., 2016; Champandard, 2016). Results comparing average mean squared errors between xnearest and $x _ { \\mathrm { { t a r g e t } } }$ in pixel space can be found in Table 3. In addition we compute the percent of images for which a certain configuration achieves the lowest loss when compared to the other configurations. ", + "bbox": [ + 174, + 500, + 825, + 569 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "In the zero step case, there is poor reconstruction and less than $1 \\%$ of the time does it obtain the lowest error of the 4 configurations. Taking 1 unrolling step results in a significant improvement in MSE. Taking 10 unrolling steps results in more modest improvement, but continues to reduce the reconstruction MSE. ", + "bbox": [ + 174, + 575, + 825, + 632 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "To visually see this, we compare the result of the optimization process for 0, 1, 5, and 10 step configurations in Figure 5. To select for images where differences in behavior is most apparent, we sort the data by the absolute value of a fractional difference in MSE between the 0 and 10 step models, $\\left| \\frac { l _ { 0 s t e p } - l _ { 1 0 s t e p } } { \\frac { 1 } { 2 } ( l _ { 0 s t e p } + l _ { 1 0 s t e p } ) } \\right|$ This highlights examples where either the 0 or 10 step model cannot accurately fit the data example but the other can. In Appendix G we show the same comparison for models initialized using different random seeds. Many of the zero step images are fuzzy and illdefined suggesting that these images cannot be generated by the standard GAN generative model, and come from a dropped mode. As more unrolling steps are added, the outlines become more clear and well defined – the model covers more of the distribution and thus can recreate these samples. ", + "bbox": [ + 174, + 638, + 825, + 773 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "3.4.2 PAIRWISE DISTANCES ", + "text_level": 1, + "bbox": [ + 176, + 787, + 380, + 803 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "A second complementary approach is to compare statistics of data samples to the corresponding statistics for samples generated by the various models. One particularly simple and relevant statistic is the distribution over pairwise distances between random pairs of samples. In the case of mode collapse, greater probability mass will be concentrated in smaller volumes, and the distribution over inter-sample distances should be skewed towards smaller distances. We sample random pairs of images from each model, as well as from the training data, and compute histograms of the $\\ell _ { 2 }$ distances between those sample pairs. As illustrated in Figure 6, the standard GAN, with zero unrolling steps, has its probability mass skewed towards smaller $\\ell _ { 2 }$ intersample distances, compared to real data. As the number of unrolling steps is increased, the histograms over intersample distances increasingly come to resemble that for the data distribution. This is further evidence in support of unrolling decreasing the mode collapse behavior of GANs. ", + "bbox": [ + 174, + 811, + 825, + 924 + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/3f34af532391d651dea1658c3e0b24559b5a1f84c026a63aecddf941edff8580.jpg", + "image_caption": [ + "Figure 5: Training set images are more accurately reconstructed using GANs trained with unrolling than by a standard (0 step) GAN, likely due to mode dropping by the standard GAN. Raw data is on the left, and the optimized images to reach this target follow for 0, 1, 5, and 10 unrolling steps. The reconstruction MSE is listed below each sample. A random 1280 images where selected from the training set, and corresponding best reconstructions for each model were found via optimization. Shown here are the eight images with the largest absolute fractional difference between GANs trained with 0 and 10 unrolling steps. " + ], + "image_footnote": [], + "bbox": [ + 176, + 102, + 833, + 352 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 491, + 823, + 534 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "4 DISCUSSION ", + "text_level": 1, + "bbox": [ + 176, + 556, + 310, + 571 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "In this work we developed a method to stabilize GAN training and reduce mode collapse by defining the generator objective with respect to unrolled optimization of the discriminator. We then demonstrated the application of this method to several tasks, where it either rescued unstable training, or reduced the tendency of the model to drop regions of the data distribution. ", + "bbox": [ + 174, + 588, + 825, + 643 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "The main drawback to this method is computational cost of each training step, which increases linearly with the number of unrolling steps. There is a tradeoff between better approximating the true generator loss and the computation required to make this estimate. Depending on the architecture, one unrolling step can be enough. In other more unstable models, such as the RNN case, more are needed to stabilize training. We have some initial positive results suggesting it may be sufficient to further perturb the training gradient in the same direction that a single unrolling step perturbs it. While this is more computationally efficient, further investigation is required. ", + "bbox": [ + 174, + 651, + 825, + 750 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "The method presented here bridges some of the gap between theoretical and practical results for training of GANs. We believe developing better update rules for the generator and discriminator is an important line of work for GAN training. In this work we have only considered a small fraction of the design space. For instance, the approach could be extended to unroll $G$ when updating $D$ as well – letting the discriminator react to how the generator would move. It is also possible to unroll sequences of $G$ and $D$ updates. This would make updates that are recursive: $G$ could react to maximize performance as if $G$ and $D$ had already updated. ", + "bbox": [ + 174, + 756, + 825, + 853 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "ACKNOWLEDGMENTS ", + "text_level": 1, + "bbox": [ + 176, + 871, + 326, + 883 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "We would like to thank Laurent Dinh, David Dohan, Vincent Dumoulin, Liam Fedus, Ishaan Gulrajani, Julian Ibarz, Eric Jang, Matthew Johnson, Marc Lanctot, Augustus Odena, Gabriel Pereyra, ", + "bbox": [ + 174, + 895, + 821, + 924 + ], + "page_idx": 9 + }, + { + "type": "image", + "img_path": "images/34368461772c640ab500044ba570ada5248c204871939191a4efbed64c4bf4ee.jpg", + "image_caption": [ + "Figure 6: As the number of unrolling steps in GAN training is increased, the distribution of pairwise distances between model samples more closely resembles the same distribution for the data. Here we plot histograms of pairwise distances between randomly selected samples. The red line gives pairwise distances in the data, while each of the five blue lines in each plot represents a model trained with a different random seed. The vertical lines are the medians of each distribution. " + ], + "image_footnote": [], + "bbox": [ + 240, + 103, + 754, + 433 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Colin Raffel, Sam Schoenholz, Ayush Sekhari, Jon Shlens, and Dale Schuurmans for insightful conversation, as well as the rest of the Google Brain Team. ", + "bbox": [ + 173, + 544, + 823, + 573 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "REFERENCES ", + "text_level": 1, + "bbox": [ + 174, + 103, + 287, + 117 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Guillaume Alain, Yoshua Bengio, Li Yao, Jason Yosinski, Eric Thibodeau-Laufer, Saizheng Zhang, and Pascal Vincent. Gsns : Generative stochastic networks. arXiv preprint arXiv:1503.05571, 2015. ", + "bbox": [ + 174, + 126, + 823, + 167 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Marcin Andrychowicz, Misha Denil, Sergio Gomez, Matthew W Hoffman, David Pfau, Tom Schaul, and Nando de Freitas. Learning to learn by gradient descent by gradient descent. arXiv preprint arXiv:1606.04474, 2016. ", + "bbox": [ + 173, + 176, + 823, + 219 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "David Belanger and Andrew McCallum. Structured prediction energy networks. arXiv preprint arXiv:1511.06350, 2015. ", + "bbox": [ + 171, + 227, + 823, + 256 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Jorg Bornschein, Samira Shabanian, Asja Fischer, and Yoshua Bengio. Bidirectional helmholtz machines. arXiv preprint arXiv:1506.03877, 2015. ", + "bbox": [ + 169, + 265, + 825, + 294 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Michael Bowling and Manuela Veloso. Multiagent learning using a variable learning rate. Artificial Intelligence, 136(2):215–250, 2002. ", + "bbox": [ + 171, + 301, + 825, + 330 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Yuri Burda, Roger B. Grosse, and Ruslan Salakhutdinov. Importance weighted autoencoders. arXiv preprint arXiv:1509.00519, 2015. ", + "bbox": [ + 171, + 338, + 825, + 367 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Alex J. Champandard. Semantic style transfer and turning two-bit doodles into fine artworks. arXiv preprint arXiv:1603.01768, 2016. ", + "bbox": [ + 171, + 376, + 823, + 405 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Tong Che, Yanran Li, Athul Paul Jacob, Yoshua Bengio, and Wenjie Li. Mode regularized generative adversarial networks. arXiv preprint arXiv: 1612.02136, 2016. ", + "bbox": [ + 173, + 412, + 823, + 443 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Xi Chen, Yan Duan, Rein Houthooft, John Schulman, Ilya Sutskever, and Pieter Abbeel. Infogan: Interpretable representation learning by information maximizing generative adversarial nets. arXiv preprint arXiv:1606.03657, 2016. ", + "bbox": [ + 174, + 450, + 821, + 492 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "John M Danskin. The theory of max-min and its application to weapons allocation problems, volume 5. Springer Science & Business Media, 1967. ", + "bbox": [ + 171, + 501, + 823, + 530 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Peter Dayan, Geoffrey E Hinton, Radford M Neal, and Richard S Zemel. The helmholtz machine. Neural computation, 7(5):889–904, 1995. ", + "bbox": [ + 174, + 537, + 821, + 568 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Laurent Dinh, David Krueger, and Yoshua Bengio. NICE: non-linear independent components estimation. arXiv preprint arXiv:1410.8516, 2014. ", + "bbox": [ + 173, + 575, + 821, + 604 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Laurent Dinh, Jascha Sohl-Dickstein, and Samy Bengio. Density estimation using real NVP. arXiv preprint arXiv:1605.08803, 2016. ", + "bbox": [ + 173, + 612, + 823, + 642 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Vincent Dumoulin, Ishmael Belghazi, Ben Poole, Alex Lamb, Martin Arjovsky, Olivier Mastropietro, and Aaron Courville. Adversarially learned inference. arXiv preprint arXiv:1606.00704, 2016. ", + "bbox": [ + 173, + 648, + 825, + 691 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Xavier Glorot and Yoshua Bengio. Understanding the difficulty of training deep feedforward neural networks. In JMLR W&CP: Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics (AISTATS 2010), volume 9, pp. 249–256, May 2010. ", + "bbox": [ + 174, + 700, + 821, + 744 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Z. Ghahramani, M. Welling, C. Cortes, N. D. Lawrence, and K. Q. Weinberger (eds.), Advances in Neural Information Processing Systems 27, pp. 2672–2680. Curran Associates, Inc., 2014. URL http://papers. nips.cc/paper/5423-generative-adversarial-nets.pdf. ", + "bbox": [ + 173, + 751, + 825, + 821 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Karol Gregor, Ivo Danihelka, Alex Graves, and Daan Wierstra. DRAW: A recurrent neural network for image generation. In Proceedings of The 32nd International Conference on Machine Learning, pp. 1462–1471, 2015. URL http://www.jmlr.org/proceedings/papers/v37/ gregor15.html. ", + "bbox": [ + 174, + 830, + 825, + 887 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Tian Han, Yang Lu, Song-Chun Zhu, and Ying Nian Wu. Alternating back-propagation for generator network, 2016. URL https://arxiv.org/abs/1606.08571. ", + "bbox": [ + 176, + 895, + 820, + 924 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Sepp Hochreiter and Jurgen Schmidhuber. Long short-term memory. ¨ Neural Comput., 9(8):1735– 1780, November 1997. ISSN 0899-7667. doi: 10.1162/neco.1997.9.8.1735. URL http://dx. doi.org/10.1162/neco.1997.9.8.1735. ", + "bbox": [ + 176, + 103, + 820, + 146 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Sergey Ioffe and Christian Szegedy. Batch normalization: Accelerating deep network training by reducing internal covariate shift. In Proceedings of the 32nd International Conference on Machine Learning, ICML 2015, Lille, France, 6-11 July 2015, pp. 448–456, 2015. URL http://jmlr. org/proceedings/papers/v37/ioffe15.html. ", + "bbox": [ + 174, + 156, + 825, + 213 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Justin Johnson, Alexandre Alahi, and Fei-Fei Li. Perceptual losses for real-time style transfer and super-resolution. arXiv preprint arXiv:1603.08155, 2016. ", + "bbox": [ + 173, + 223, + 823, + 252 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Anatoli Juditsky, Arkadi Nemirovski, et al. First order methods for nonsmooth convex large-scale optimization, i: general purpose methods. Optimization for Machine Learning, pp. 121–148, 2011. ", + "bbox": [ + 174, + 262, + 826, + 305 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014. ", + "bbox": [ + 173, + 315, + 823, + 344 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Diederik P Kingma and Max Welling. Auto-encoding variational bayes, 2013. URL https: //arxiv.org/abs/1312.6114. ", + "bbox": [ + 171, + 354, + 821, + 383 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Diederik P. Kingma, Tim Salimans, and Max Welling. Improving variational inference with inverse autoregressive flow. 2016. ", + "bbox": [ + 173, + 393, + 823, + 422 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Tejas D. Kulkarni, Will Whitney, Pushmeet Kohli, and Joshua B. Tenenbaum. Deep convolutional inverse graphics network. arXiv preprint arXiv:1503.03167, 2015. ", + "bbox": [ + 174, + 433, + 821, + 463 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Christian Ledig, Lucas Theis, Ferenc Huszar, Jose Caballero, Andrew Aitken, Alykhan Tejani, Johannes Totz, Zehan Wang, and Wenzhe Shi. Photo-realistic single image super-resolution using a generative adversarial network, 2016. URL https://arxiv.org/abs/1609.04802. ", + "bbox": [ + 174, + 472, + 825, + 515 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Dougal Maclaurin, David Duvenaud, and Ryan P. Adams. Gradient-based hyperparameter optimization through reversible learning, 2015. ", + "bbox": [ + 173, + 525, + 823, + 555 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Anh Nguyen, Alexey Dosovitskiy, Jason Yosinski, Thomas Brox, and Jeff Clune. Synthesizing the preferred inputs for neurons in neural networks via deep generator networks. arXiv preprint arXiv:1605.09304, 2016. ", + "bbox": [ + 173, + 564, + 823, + 608 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Sebastian Nowozin, Botond Cseke, and Ryota Tomioka. f-gan: Training generative neural samplers using variational divergence minimization. arXiv preprint arXiv:1606.00709, 2016. ", + "bbox": [ + 169, + 617, + 823, + 647 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Augustus Odena, Christopher Olah, and Jonathon Shlens. Conditional image synthesis with auxiliary classifier gans. arXiv preprint arXiv:1610.09585, 2016. ", + "bbox": [ + 173, + 657, + 821, + 686 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Barak A. Pearlmutter and Jeffrey Mark Siskind. Reverse-mode ad in a functional framework: Lambda the ultimate backpropagator. ACM Trans. Program. Lang. Syst., 30(2):7:1–7:36, March 2008. ISSN 0164-0925. doi: 10.1145/1330017.1330018. URL http://doi.acm.org/10. 1145/1330017.1330018. ", + "bbox": [ + 173, + 695, + 825, + 753 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Ben Poole, Alexander A Alemi, Jascha Sohl-Dickstein, and Anelia Angelova. Improved generator objectives for gans. arXiv preprint arXiv:1612.02780, 2016. ", + "bbox": [ + 173, + 763, + 823, + 792 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learning with deep convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434, 2015. ", + "bbox": [ + 171, + 803, + 823, + 832 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Scott Reed, Zeynep Akata, Santosh Mohan, Samuel Tenka, Bernt Schiele, and Honglak Lee. Learning what and where to draw. In NIPS, 2016a. ", + "bbox": [ + 173, + 842, + 823, + 871 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Scott Reed, Zeynep Akata, Xinchen Yan, Lajanugen Logeswaran, Bernt Schiele, and Honglak Lee. Generative adversarial text-to-image synthesis. In Proceedings of The 33rd International Conference on Machine Learning, 2016b. ", + "bbox": [ + 174, + 882, + 823, + 924 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Danilo Jimenez Rezende, Shakir Mohamed, and Daan Wierstra. Stochastic backpropagation and variational inference in deep latent gaussian models. In International Conference on Machine Learning. Citeseer, 2014. ", + "bbox": [ + 174, + 103, + 821, + 146 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Tim Salimans, Ian J. Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen. Improved techniques for training gans. arXiv preprint arXiv:1606.03498, 2016. ", + "bbox": [ + 173, + 155, + 821, + 184 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Karen Simonyan, Andrea Vedaldi, and Andrew Zisserman. Deep inside convolutional networks: Visualising image classification models and saliency maps. arXiv preprint arXiv:1312.6034, 2013. ", + "bbox": [ + 171, + 193, + 823, + 222 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Satinder Singh, Michael Kearns, and Yishay Mansour. Nash convergence of gradient dynamics in general-sum games. In Proceedings of the Sixteenth conference on Uncertainty in artificial intelligence, pp. 541–548. Morgan Kaufmann Publishers Inc., 2000. ", + "bbox": [ + 176, + 229, + 823, + 273 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Jascha Sohl-Dickstein, Eric A. Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In Proceedings of The 32nd International Conference on Machine Learning, pp. 2256–2265, 2015. URL http://arxiv.org/abs/ 1503.03585. ", + "bbox": [ + 173, + 281, + 826, + 338 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Casper Kaae Sonderby, Jose Caballero, Lucas Theis, Wenzhe Shi, and Ferenc Huszar. Amortised map inference for image super-resolution, 2016. URL https://arxiv.org/abs/1610. 04490v1. ", + "bbox": [ + 173, + 347, + 825, + 390 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "L. Theis and M. Bethge. Generative image modeling using spatial lstms. In Advances in Neural Information Processing Systems 28, Dec 2015. URL http://arxiv.org/abs/1506. $0 3 4 7 8 /$ . ", + "bbox": [ + 173, + 398, + 825, + 440 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "L. Theis, A. van den Oord, and M. Bethge. A note on the evaluation of generative models. In International Conference on Learning Representations, Apr 2016. URL http://arxiv.org/ abs/1511.01844. ", + "bbox": [ + 174, + 450, + 823, + 493 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "T. Tieleman and G. Hinton. Lecture 6.5—RmsProp: Divide the gradient by a running average of its recent magnitude. COURSERA: Neural Networks for Machine Learning, 2012. ", + "bbox": [ + 169, + 501, + 823, + 531 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Aaron van den Oord, Nal Kalchbrenner, and Koray Kavukcuoglu. Pixel recurrent neural networks.¨ arXiv preprint arXiv:1601.06759, abs/1601.06759, 2016a. URL http://arxiv.org/abs/ 1601.06759. ", + "bbox": [ + 173, + 540, + 823, + 582 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Aaron van den Oord, Nal Kalchbrenner, Oriol Vinyals, Lasse Espeholt, Alex Graves, and Ko-¨ ray Kavukcuoglu. Conditional image generation with pixelcnn decoders. arXiv preprint arXiv:1606.05328, 2016b. ", + "bbox": [ + 173, + 590, + 823, + 633 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Pascal Vincent, Hugo Larochelle, Isabelle Lajoie, Yoshua Bengio, and Pierre-Antoine Manzagol. Stacked denoising autoencoders: Learning useful representations in a deep network with a local denoising criterion. J. Mach. Learn. Res., 11:3371–3408, December 2010. ISSN 1532-4435. URL http://dl.acm.org/citation.cfm?id=1756006.1953039. ", + "bbox": [ + 173, + 642, + 825, + 700 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Jason Yosinski, Jeff Clune, Anh Nguyen, Thomas Fuchs, and Hod Lipson. Understanding neural networks through deep visualization. arXiv preprint arXiv:1506.06579, 2015. ", + "bbox": [ + 173, + 708, + 820, + 738 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Chongjie Zhang and Victor R Lesser. Multi-agent learning with policy prediction. In Proceedings of the Twenty-Fourth AAAI Conference on Artificial Intelligence, 2010. ", + "bbox": [ + 173, + 746, + 823, + 776 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Junbo Zhao, Michael Mathieu, and Yann LeCun. Energy-based generative adversarial network. arXiv preprint arXiv:1609.03126, 2016. ", + "bbox": [ + 173, + 784, + 821, + 814 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Jun-Yan Zhu, Philipp Krahenb ¨ uhl, Eli Shechtman, and Alexei A. Efros. Generative visual manipula- ¨ tion on the natural image manifold. In Proceedings of European Conference on Computer Vision (ECCV), 2016. ", + "bbox": [ + 173, + 821, + 825, + 864 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Appendix ", + "text_level": 1, + "bbox": [ + 176, + 99, + 316, + 127 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "A 2D GAUSSIAN TRAINING DETAILS ", + "text_level": 1, + "bbox": [ + 174, + 148, + 500, + 166 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Network architecture and experimental details for the experiment in Section 3.1 are as follows: ", + "bbox": [ + 173, + 180, + 794, + 195 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "The dataset is sampled from a mixture of 8 Gaussians of standard deviation 0.02. The means are equally spaced around a circle of radius 2. ", + "bbox": [ + 173, + 202, + 823, + 229 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "The generator network consists of a fully connected network with 2 hidden layers of size 128 with relu activations followed by a linear projection to 2 dimensions. All weights are initialized to be orthogonal with scaling of 0.8. ", + "bbox": [ + 174, + 237, + 825, + 279 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "The discriminator network first scales its input down by a factor of 4 (to roughly scale to (-1,1)), followed by 1 layer fully connected network with relu activations to a linear layer to of size 1 to act as the logit. ", + "bbox": [ + 174, + 286, + 825, + 329 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "The generator minimizes $\\mathcal { L } _ { G } = \\log ( D ( x ) ) + \\log ( 1 - D ( G ( z ) ) )$ and the discriminator minimizes $\\mathcal { L } _ { D } \\stackrel { = } { = } - \\log ( D ( x ) ) - \\log ( 1 - D ( \\stackrel { . } { G } ( z ) ) )$ where $\\mathbf { X }$ is sampled from the data distribution and $z \\sim$ $\\mathcal { N } ( 0 , I _ { 2 5 6 } )$ . Both networks are optimized using Adam (Kingma & Ba, 2014) with a learning rate of 1e-4 and $\\beta _ { 1 } { = } 0 . 5$ . ", + "bbox": [ + 174, + 334, + 825, + 391 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "The network is trained by alternating updates of the generator and the discriminator. One step consists of either $\\mathbf { G }$ or $\\mathbf { D }$ updating. ", + "bbox": [ + 174, + 398, + 823, + 426 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "B MORE MIXTURE OF GAUSSIAN EXPERIMENTS ", + "text_level": 1, + "bbox": [ + 176, + 445, + 594, + 462 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "B.1 EFFECTS OF TIME DELAY / HISTORICAL AVERAGING ", + "text_level": 1, + "bbox": [ + 173, + 477, + 576, + 492 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Another comparison we looked at was with regard to historical averaging based approaches. Recently similarly inspired approaches have been used in (Salimans et al., 2016) to stabilize training. For our study, we looked at taking an ensemble of discriminators over time. ", + "bbox": [ + 174, + 502, + 825, + 545 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "First, we looked at taking an ensemble of the last N steps, as shown in Figure App.1. ", + "bbox": [ + 174, + 551, + 725, + 568 + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/ca00dc71a894afed5412e7d9c9476916ab9ab1f2648dedea7d529b310373bd67.jpg", + "image_caption": [ + "Figure App.1: Historical averaging does not visibly increase stability on the mixture of Gaussians task. Each row corresponds to an ensemble of discriminators which consists of the indicated number of immediately preceding discriminators. The columns correspond to different numbers of training steps. " + ], + "image_footnote": [], + "bbox": [ + 174, + 582, + 821, + 720 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "To further explore this idea, we ran experiments with an ensemble of 5 discriminators, but with different periods between replacing discriminators in the ensemble. For example, if I sample at a rate of 100, it would take 500 steps to replace all 5 discriminators. Results can be seen in Figure App.2. ", + "bbox": [ + 174, + 805, + 825, + 861 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "We observe that given longer and longer time delays, the model becomes less and less stable. We hypothesize that this is due to the initial shape of the discriminator loss surface. When training, the discriminator’s estimates of probability densities are only accurate on regions where it was trained. When fixing this discriminator, we are removing the feedback between the generator exploitation and the discriminators ability to move. As a result, the generator is able to exploit these fixed areas of poor performance for older discriminators in the ensemble. New discriminators (over)compensate for this, leading the system to diverge. ", + "bbox": [ + 174, + 867, + 825, + 924 + ], + "page_idx": 14 + }, + { + "type": "image", + "img_path": "images/04adbd77c06eb06d8623be8ab50fe44136a6a0398a13d8fdd2a583c617dbcff0.jpg", + "image_caption": [ + "Figure App.2: Introducing longer time delays between the discriminator ensemble results in instability and probability distributions that are not in the window being visualized. The $\\mathbf { X }$ axis is the number of weight updates and the y axis is how many steps to skip between discriminator updates when selecting the ensemble of 5 discriminators. " + ], + "image_footnote": [], + "bbox": [ + 174, + 102, + 821, + 242 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 342, + 823, + 383 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "B.2 EFFECTS OF THE SECOND GRADIENT ", + "text_level": 1, + "bbox": [ + 176, + 402, + 470, + 416 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "A second factor we analyzed is the effect of backpropagating the learning signal through the unrolling in Equation 12. We can turn on or off this backpropagation through the unrolling by introducing stop gradient calls into our computation graph between each unrolling step. With the stop gradient in place, the update signal corresponds only to the first term in Equation 12. We looked at 3 configurations: without stop gradients; vanilla unrolled GAN, with stop gradients; and with stop gradients but taking the average over the $k$ unrolling steps instead of taking the final value. Results can be see in Figure App.3. ", + "bbox": [ + 174, + 429, + 825, + 526 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "We initially observed no difference between unrolling with and without the second gradient, as both required 3 unrolling steps to become stable. When the discriminator is unrolled to convergence, the second gradient term becomes zero. Due to the simplicity of the problem, we suspect that the discriminator nearly converged for every generator step, and the second gradient term was thus irrelevant. ", + "bbox": [ + 174, + 532, + 825, + 603 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "To test this, we modified the dynamics to perform five generator steps for each discriminator update. Results are shown in Figure App.4. With the discriminator now kept out of equilibrium, successful training can be achieved with half as many unrolling steps when using both terms in the gradient than when only including the first term. ", + "bbox": [ + 174, + 609, + 823, + 665 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "C RNN MNIST TRAINING DETAILS ", + "text_level": 1, + "bbox": [ + 174, + 688, + 493, + 704 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "The network architecture for the experiment in Section 3.2 is as follows: ", + "bbox": [ + 176, + 720, + 647, + 734 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "The MNIST dataset is scaled to [-1, 1). ", + "bbox": [ + 176, + 742, + 429, + 756 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "The generator first scales the 256D noise vector through a 256 unit fully connected layer with relu activation. This is then fed into the initial state of a 256D LSTM(Hochreiter & Schmidhuber, 1997) that runs 28 steps corresponding to the number of columns in MNIST. The resulting sequence of activations is projected through a fully connected layer with 28 outputs with a tanh activation function. All weights are initialized via the ”Xavier” initialization (Glorot & Bengio, 2010). The forget bias on the LSTM is initialized to 1. ", + "bbox": [ + 174, + 762, + 825, + 847 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "The discriminator network feeds the input into a Convolution(16, stride $^ { = 2 }$ ) followed by a Convolution(32, stride $^ { = 2 }$ ) followed by Convolution(32, stride ${ \\boldsymbol { \\mathbf { \\mathit { \\varepsilon } } } } = 2 { \\boldsymbol { \\mathbf { \\mathit { \\varepsilon } } } }$ ). All convolutions have stride 2. As in (Radford et al., 2015) leaky rectifiers are used with a 0.3 leak. Batch normalization is applied after each layer (Ioffe & Szegedy, 2015). The resulting 4D tensor is then flattened and a linear projection is performed to a single scalar. ", + "bbox": [ + 176, + 854, + 823, + 924 + ], + "page_idx": 15 + }, + { + "type": "image", + "img_path": "images/e082e1fa15fe3c8c65693a9ca8c06d3ede7c2ddcf2d0287c257abe4e8c2ff9b6.jpg", + "image_caption": [ + "Figure App.3: If the discriminator remains nearly at its optimum during learning, then performance is nearly identical with and without the second gradient term in Equation 12. As shown in Figure App.4, when the discriminator lags behind the generator, backpropagating through unrolling aids convergence. " + ], + "image_footnote": [], + "bbox": [ + 169, + 97, + 825, + 535 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "The generator network minimises $\\mathcal { L } _ { G } = \\log ( D ( G ( z ) ) )$ and the discriminator minimizes $\\mathcal { L } _ { D } =$ $\\log ( D ( x ) ) + \\log ( 1 - D ( G ( z ) ) )$ . Both networks are trained with Adam(Kingma & Ba, 2014) with learning rates of 1e-4 and $\\beta _ { 1 } { = } 0 . 5$ . The network is trained alternating updating the generator and the discriminator for 150k steps. One step consists of just 1 network update. ", + "bbox": [ + 174, + 637, + 825, + 694 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "D CIFAR10/MNIST TRAINING DETAILS ", + "text_level": 1, + "bbox": [ + 174, + 720, + 534, + 738 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "The network architectures for the discriminator, generator, and encoder as as follows. All convolutions have a kernel size of 3x3 with batch normalization and leaky ReLU’s with a 0.3 leak. ", + "bbox": [ + 176, + 757, + 823, + 786 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "The generator network is defined as: ", + "bbox": [ + 176, + 792, + 413, + 808 + ], + "page_idx": 16 + }, + { + "type": "table", + "img_path": "images/6be1d7b077edb834c2bd6ee68d568dbb574dbb6746300a22e384363f25abe428.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
number outputsstride
Input: z ~ N(0,I256)Fully connectedReshape to image 4,4,512ConvolutionConvolutionConvolutionConvolution4*4*512256128641or32221
", + "bbox": [ + 174, + 814, + 542, + 929 + ], + "page_idx": 16 + }, + { + "type": "image", + "img_path": "images/ddba8731c308e072b5532fd2f2dfbcf72fd2fc3c5d92dcb601c2c03608beee85.jpg", + "image_caption": [ + "Unrolled GAN with 5 G Steps per D without second gradient " + ], + "image_footnote": [], + "bbox": [ + 173, + 117, + 821, + 316 + ], + "page_idx": 17 + }, + { + "type": "table", + "img_path": "images/21b4b254a8e79c21939f50e66bbaf743c06f0c33a79d176b714a375db90c6857.jpg", + "table_caption": [ + "Unrolled GAN with 5 G Steps per D " + ], + "table_footnote": [], + "table_body": "
number outputsstride
Input: x~ Pdata or G Transposed Convolution Transposed Convolution Transposed Convolution64 128 256222
Flatten
Fully Connected1
", + "bbox": [ + 174, + 762, + 531, + 862 + ], + "page_idx": 17 + }, + { + "type": "image", + "img_path": "images/c5b3fa4bd5bb790af426480883819ba983aa6618ec11502817b7cc7fa39e22d5.jpg", + "image_caption": [ + "Figure App.4: Backpropagating through the unrolling process aids convergence when the discriminator does not fully converge between generator updates. When taking 5 generator steps per discriminator step unrolling greatly increases stability, requiring only 5 unrolling steps to converge. Without the second gradient it requires 10 unrolling steps. Also see Figure App.3. " + ], + "image_footnote": [], + "bbox": [ + 173, + 338, + 821, + 536 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "The discriminator network is defined as: ", + "bbox": [ + 174, + 743, + 439, + 757 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "The generator network minimises $\\mathcal { L } _ { G } = \\log ( D ( G ( z ) ) )$ and the discriminator minimizes $\\mathcal { L } _ { D } =$ $\\log ( D ( x ) ) + \\log ( 1 - D ( G ( z ) ) )$ . The networks are trained with Adam with a generator learning rate of 1e-4, and a discriminator learning rate of 2e-4. The network is trained alternating updating the generator and the discriminator for $1 0 0 \\mathrm { k }$ steps. One step consists of just 1 network update. ", + "bbox": [ + 173, + 867, + 825, + 924 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "E 1000 CLASS MNIST ", + "text_level": 1, + "bbox": [ + 174, + 102, + 380, + 118 + ], + "page_idx": 18 + }, + { + "type": "table", + "img_path": "images/6d161339c120af4e827f0777703e338fce4dd94b3a55a7a2c9761991b51c1f31.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
number outputsstride
Input: z ~ N(0,I256)Fully connectedReshape to image 4,4,64ConvolutionConvolutionConvolutionConvolution4 *4*643216832221
", + "bbox": [ + 173, + 136, + 532, + 250 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "The discriminator network is parametrized by a size $\\mathrm { X }$ and is defined as follows. In our tests, we used X of 1/4 and 1/2. ", + "bbox": [ + 171, + 255, + 826, + 284 + ], + "page_idx": 18 + }, + { + "type": "table", + "img_path": "images/d861a766290ac6cab3787aa8042ed8fa20fa08a94a861ed7ebddde0d0efa7d41.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
number outputsstride
Input: x ~ Pdata or G Transposed Convolution Transposed Convolution8*X 16*X222
Transposed Convolution Flatten32*X
Fully Connected1
", + "bbox": [ + 174, + 287, + 531, + 388 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "F COLORED MNIST DATASET ", + "text_level": 1, + "bbox": [ + 174, + 411, + 439, + 429 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "F.1 DATASET ", + "text_level": 1, + "bbox": [ + 174, + 449, + 279, + 463 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "To generate this dataset we first took the mnist digit, $I$ , scaled between 0 and 1. For each image we sample a color, $C$ , normally distributed with mean $\\scriptstyle = 0$ and std $\\scriptstyle \\mathtt { = 0 . 5 }$ . To generate a colored digit between (-1, 1) we do $I * C + \\mathsf { \\bar { ( } } I - 1 )$ . Finally, we add a small amount of pixel independent noise sampled from a normal distribution with std $= 0 . 2$ , and the resulting values are cliped between (-1, 1). When visualized, this generates images and samples that can be seen in figure App.5. Once again it is very hard to visually see differences in sample diversity when comparing the 128 and the 512 sized models. ", + "bbox": [ + 173, + 478, + 825, + 575 + ], + "page_idx": 18 + }, + { + "type": "image", + "img_path": "images/44d318f018c78dbdaec0c5211309ab172baeb975108ebe77f750982a4193cef6.jpg", + "image_caption": [ + "Figure App.5: Right: samples from the data distribution. Middle: Samples from 1/4 size model with 0 look ahead steps (worst diversity). Left: Samples from 1/1 size model with 10 look ahead steps (most diversity). " + ], + "image_footnote": [], + "bbox": [ + 202, + 593, + 795, + 746 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "F.2 MODELS ", + "text_level": 1, + "bbox": [ + 174, + 844, + 276, + 859 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "The models used in this section are parametrized by a variable $\\mathrm { X }$ to control capacity. A value of ${ \\bf X } { = } 1$ is same architecture used in the cifar10 experiments. We used 1/4, 1/2 and 1 as these values. ", + "bbox": [ + 176, + 873, + 825, + 902 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "The generator network is defined as: ", + "bbox": [ + 176, + 909, + 413, + 924 + ], + "page_idx": 18 + }, + { + "type": "table", + "img_path": "images/de1a5ae153f5aca675e1d81f0274ce2bb52042972834a41845eb58fa8a6854f0.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
number outputsstride
Input: z~ N(0,I256)Fully connectedReshape to image 4,4,512*XConvolutionConvolutionConvolutionConvolution4*4*512*X256*X128*X64*X32221
", + "bbox": [ + 174, + 101, + 562, + 215 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "The discriminator network is defined as: ", + "bbox": [ + 176, + 220, + 439, + 234 + ], + "page_idx": 19 + }, + { + "type": "table", + "img_path": "images/8185cc38813b0aac909bf7ec1d0d173e742b2ff60e4a2f788d6a7b905fcd642a.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
number outputsstride
Input: x ~ Pdata or G Transposed Convolution Transposed Convolution Transposed Convolution Flatten Fully Connected64*X 128*X 256*X222
", + "bbox": [ + 174, + 239, + 532, + 340 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "G OPTIMIZATION BASED VISUALIZATIONS ", + "text_level": 1, + "bbox": [ + 174, + 356, + 547, + 372 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "More examples of model based optimization. We performed 5 runs with different seeds of each of of the unrolling steps configuration. Bellow are comparisons for each run index. Ideally this would be a many to many comparison, but for space efficiency we grouped the runs by the index in which they were run. ", + "bbox": [ + 173, + 387, + 825, + 443 + ], + "page_idx": 19 + }, + { + "type": "image", + "img_path": "images/8d38e597124cc936be0c8530d6fb7e8b1ecc41f5e50351124daf12a2d78e11fa.jpg", + "image_caption": [ + "Figure App.6: Samples from 1/5 with different random seeds. " + ], + "image_footnote": [], + "bbox": [ + 189, + 114, + 821, + 883 + ], + "page_idx": 20 + }, + { + "type": "image", + "img_path": "images/be94e2bb107541e788983b4ad2bdc31bebdff50fbbb8529c915238f8cca809e8.jpg", + "image_caption": [ + "Figure App.7: Samples from 2/5 with different random seeds. " + ], + "image_footnote": [], + "bbox": [ + 189, + 113, + 820, + 881 + ], + "page_idx": 21 + }, + { + "type": "image", + "img_path": "images/e17957e3a6c6fffe5b489567eb4830a1720a2ae85eadaf27b1b6017a82527e8b.jpg", + "image_caption": [ + "Figure App.8: Samples from 3/5 with different random seeds. " + ], + "image_footnote": [], + "bbox": [ + 189, + 112, + 820, + 880 + ], + "page_idx": 22 + }, + { + "type": "image", + "img_path": "images/fbca68637cf3798a1a739daff1548046fba7251170be090a9d1d669d0aa5603e.jpg", + "image_caption": [ + "Figure App.9: Samples from 4/5 with different random seeds. " + ], + "image_footnote": [], + "bbox": [ + 189, + 117, + 820, + 880 + ], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/c67273dbe4f3a1a6aee81a0d0854850226af640cdd34215ec9e0663f82c11a7d.jpg", + "image_caption": [ + "Figure App.10: Samples from 5/5 with different random seeds. " + ], + "image_footnote": [], + "bbox": [ + 189, + 112, + 820, + 881 + ], + "page_idx": 24 + } +] \ No newline at end of file diff --git a/parse/train/BydrOIcle/BydrOIcle_middle.json b/parse/train/BydrOIcle/BydrOIcle_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..07177144a4c5bec38051aaf0e358bcfb9b9bcf93 --- /dev/null +++ b/parse/train/BydrOIcle/BydrOIcle_middle.json @@ -0,0 +1,47021 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 106, + 79, + 494, + 96 + ], + "lines": [ + { + "bbox": [ + 106, + 78, + 496, + 99 + ], + "spans": [ + { + "bbox": [ + 106, + 78, + 496, + 99 + ], + "score": 1.0, + "content": "UNROLLED GENERATIVE ADVERSARIAL NETWORKS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 113, + 115, + 210, + 149 + ], + "lines": [ + { + "bbox": [ + 111, + 114, + 166, + 127 + ], + "spans": [ + { + "bbox": [ + 111, + 114, + 166, + 127 + ], + "score": 1.0, + "content": "Luke Metz∗", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 111, + 125, + 169, + 139 + ], + "spans": [ + { + "bbox": [ + 111, + 125, + 169, + 139 + ], + "score": 1.0, + "content": "Google Brain", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 112, + 137, + 211, + 150 + ], + "spans": [ + { + "bbox": [ + 112, + 137, + 211, + 150 + ], + "score": 1.0, + "content": "lmetz@google.com", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 244, + 114, + 372, + 149 + ], + "lines": [ + { + "bbox": [ + 244, + 114, + 293, + 127 + ], + "spans": [ + { + "bbox": [ + 244, + 114, + 293, + 127 + ], + "score": 1.0, + "content": "Ben Poole†", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 243, + 125, + 325, + 139 + ], + "spans": [ + { + "bbox": [ + 243, + 125, + 325, + 139 + ], + "score": 1.0, + "content": "Stanford University", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 244, + 138, + 372, + 149 + ], + "spans": [ + { + "bbox": [ + 244, + 138, + 372, + 149 + ], + "score": 1.0, + "content": "poole@cs.stanford.edu", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 406, + 115, + 498, + 149 + ], + "lines": [ + { + "bbox": [ + 406, + 115, + 457, + 127 + ], + "spans": [ + { + "bbox": [ + 406, + 115, + 457, + 127 + ], + "score": 1.0, + "content": "David Pfau", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 406, + 126, + 484, + 139 + ], + "spans": [ + { + "bbox": [ + 406, + 126, + 484, + 139 + ], + "score": 1.0, + "content": "Google DeepMind", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 405, + 138, + 499, + 149 + ], + "spans": [ + { + "bbox": [ + 405, + 138, + 499, + 149 + ], + "score": 1.0, + "content": "pfau@google.com", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 112, + 165, + 228, + 199 + ], + "lines": [ + { + "bbox": [ + 111, + 164, + 208, + 177 + ], + "spans": [ + { + "bbox": [ + 111, + 164, + 208, + 177 + ], + "score": 1.0, + "content": "Jascha Sohl-Dickstein", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 112, + 176, + 169, + 188 + ], + "spans": [ + { + "bbox": [ + 112, + 176, + 169, + 188 + ], + "score": 1.0, + "content": "Google Brain", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 113, + 188, + 228, + 200 + ], + "spans": [ + { + "bbox": [ + 113, + 188, + 228, + 200 + ], + "score": 1.0, + "content": "jaschasd@google.com", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11 + }, + { + "type": "title", + "bbox": [ + 278, + 228, + 333, + 240 + ], + "lines": [ + { + "bbox": [ + 276, + 226, + 336, + 242 + ], + "spans": [ + { + "bbox": [ + 276, + 226, + 336, + 242 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 143, + 252, + 468, + 341 + ], + "lines": [ + { + "bbox": [ + 141, + 252, + 469, + 266 + ], + "spans": [ + { + "bbox": [ + 141, + 252, + 469, + 266 + ], + "score": 1.0, + "content": "We introduce a method to stabilize Generative Adversarial Networks (GANs) by", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 142, + 264, + 469, + 276 + ], + "spans": [ + { + "bbox": [ + 142, + 264, + 469, + 276 + ], + "score": 1.0, + "content": "defining the generator objective with respect to an unrolled optimization of the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 274, + 469, + 287 + ], + "spans": [ + { + "bbox": [ + 141, + 274, + 469, + 287 + ], + "score": 1.0, + "content": "discriminator. This allows training to be adjusted between using the optimal dis-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 142, + 286, + 469, + 298 + ], + "spans": [ + { + "bbox": [ + 142, + 286, + 469, + 298 + ], + "score": 1.0, + "content": "criminator in the generator’s objective, which is ideal but infeasible in practice,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 142, + 297, + 470, + 309 + ], + "spans": [ + { + "bbox": [ + 142, + 297, + 470, + 309 + ], + "score": 1.0, + "content": "and using the current value of the discriminator, which is often unstable and leads", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 308, + 470, + 320 + ], + "spans": [ + { + "bbox": [ + 141, + 308, + 470, + 320 + ], + "score": 1.0, + "content": "to poor solutions. We show how this technique solves the common problem of", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 318, + 470, + 332 + ], + "spans": [ + { + "bbox": [ + 141, + 318, + 470, + 332 + ], + "score": 1.0, + "content": "mode collapse, stabilizes training of GANs with complex recurrent generators,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 329, + 452, + 343 + ], + "spans": [ + { + "bbox": [ + 141, + 329, + 452, + 343 + ], + "score": 1.0, + "content": "and increases diversity and coverage of the data distribution by the generator.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 17.5 + }, + { + "type": "title", + "bbox": [ + 109, + 362, + 206, + 375 + ], + "lines": [ + { + "bbox": [ + 105, + 361, + 208, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 208, + 378 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 388, + 505, + 486 + ], + "lines": [ + { + "bbox": [ + 106, + 387, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 387, + 506, + 401 + ], + "score": 1.0, + "content": "The use of deep neural networks as generative models for complex data has made great advances", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 398, + 506, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 506, + 412 + ], + "score": 1.0, + "content": "in recent years. This success has been achieved through a surprising diversity of training losses", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 410, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 106, + 410, + 505, + 421 + ], + "score": 1.0, + "content": "and model architectures, including denoising autoencoders (Vincent et al., 2010), variational au-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 420, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 506, + 433 + ], + "score": 1.0, + "content": "toencoders (Kingma & Welling, 2013; Rezende et al., 2014; Gregor et al., 2015; Kulkarni et al.,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 430, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 505, + 444 + ], + "score": 1.0, + "content": "2015; Burda et al., 2015; Kingma et al., 2016), generative stochastic networks (Alain et al., 2015),", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 442, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 505, + 456 + ], + "score": 1.0, + "content": "diffusion probabilistic models (Sohl-Dickstein et al., 2015), autoregressive models (Theis & Bethge,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 452, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 452, + 505, + 466 + ], + "score": 1.0, + "content": "2015; van den Oord et al., 2016a;b), real non-volume preserving transformations (Dinh et al., 2014;", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 464, + 505, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 505, + 477 + ], + "score": 1.0, + "content": "2016), Helmholtz machines (Dayan et al., 1995; Bornschein et al., 2015), and Generative Adversar-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 475, + 299, + 487 + ], + "spans": [ + { + "bbox": [ + 106, + 475, + 299, + 487 + ], + "score": 1.0, + "content": "ial Networks (GANs) (Goodfellow et al., 2014).", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 27 + }, + { + "type": "title", + "bbox": [ + 109, + 500, + 304, + 511 + ], + "lines": [ + { + "bbox": [ + 106, + 500, + 306, + 513 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 306, + 513 + ], + "score": 1.0, + "content": "1.1 GENERATIVE ADVERSARIAL NETWORKS", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 521, + 505, + 620 + ], + "lines": [ + { + "bbox": [ + 106, + 521, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 521, + 505, + 534 + ], + "score": 1.0, + "content": "While most deep generative models are trained by maximizing log likelihood or a lower bound on", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 532, + 505, + 544 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 505, + 544 + ], + "score": 1.0, + "content": "log likelihood, GANs take a radically different approach that does not require inference or explicit", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 542, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 505, + 555 + ], + "score": 1.0, + "content": "calculation of the data likelihood. Instead, two models are used to solve a minimax game: a genera-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 552, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 505, + 568 + ], + "score": 1.0, + "content": "tor which samples data, and a discriminator which classifies the data as real or generated. In theory", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 563, + 506, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 506, + 579 + ], + "score": 1.0, + "content": "these models are capable of modeling an arbitrarily complex probability distribution. When using", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 577, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 505, + 588 + ], + "score": 1.0, + "content": "the optimal discriminator for a given class of generators, the original GAN proposed by Goodfellow", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 586, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 505, + 600 + ], + "score": 1.0, + "content": "et al. minimizes the Jensen-Shannon divergence between the data distribution and the generator,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 598, + 505, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 505, + 611 + ], + "score": 1.0, + "content": "and extensions generalize this to a wider class of divergences (Nowozin et al., 2016; Sonderby et al.,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 609, + 209, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 209, + 621 + ], + "score": 1.0, + "content": "2016; Poole et al., 2016).", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 107, + 626, + 505, + 703 + ], + "lines": [ + { + "bbox": [ + 105, + 624, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 505, + 639 + ], + "score": 1.0, + "content": "The ability to train extremely flexible generating functions, without explicitly computing likeli-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 636, + 506, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 506, + 649 + ], + "score": 1.0, + "content": "hoods or performing inference, and while targeting more mode-seeking divergences as made GANs", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 647, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 647, + 506, + 661 + ], + "score": 1.0, + "content": "extremely successful in image generation (Odena et al., 2016; Salimans et al., 2016; Radford et al.,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 658, + 506, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 506, + 672 + ], + "score": 1.0, + "content": "2015), and image super resolution (Ledig et al., 2016). The flexibility of the GAN framework has", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 670, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 106, + 670, + 505, + 681 + ], + "score": 1.0, + "content": "also enabled a number of successful extensions of the technique, for instance for structured predic-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 681, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 681, + 505, + 693 + ], + "score": 1.0, + "content": "tion (Reed et al., 2016a;b; Odena et al., 2016), training energy based models (Zhao et al., 2016), and", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 692, + 412, + 704 + ], + "spans": [ + { + "bbox": [ + 106, + 692, + 412, + 704 + ], + "score": 1.0, + "content": "combining the GAN loss with a mutual information loss (Chen et al., 2016).", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 45 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 118, + 710, + 437, + 732 + ], + "lines": [ + { + "bbox": [ + 118, + 709, + 436, + 722 + ], + "spans": [ + { + "bbox": [ + 118, + 709, + 436, + 722 + ], + "score": 1.0, + "content": "∗Work done as a member of the Google Brain Residency program (g.co/brainresidency)", + "type": "text" + } + ] + }, + { + "bbox": [ + 118, + 720, + 315, + 734 + ], + "spans": [ + { + "bbox": [ + 118, + 720, + 315, + 734 + ], + "score": 1.0, + "content": "†Work completed as part of a Google Brain internship", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 303, + 752, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 106, + 79, + 494, + 96 + ], + "lines": [ + { + "bbox": [ + 106, + 78, + 496, + 99 + ], + "spans": [ + { + "bbox": [ + 106, + 78, + 496, + 99 + ], + "score": 1.0, + "content": "UNROLLED GENERATIVE ADVERSARIAL NETWORKS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "list", + "bbox": [ + 113, + 115, + 210, + 149 + ], + "lines": [ + { + "bbox": [ + 111, + 114, + 166, + 127 + ], + "spans": [ + { + "bbox": [ + 111, + 114, + 166, + 127 + ], + "score": 1.0, + "content": "Luke Metz∗", + "type": "text" + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 125, + 169, + 139 + ], + "spans": [ + { + "bbox": [ + 111, + 125, + 169, + 139 + ], + "score": 1.0, + "content": "Google Brain", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 112, + 137, + 211, + 150 + ], + "spans": [ + { + "bbox": [ + 112, + 137, + 211, + 150 + ], + "score": 1.0, + "content": "lmetz@google.com", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 244, + 114, + 293, + 127 + ], + "spans": [ + { + "bbox": [ + 244, + 114, + 293, + 127 + ], + "score": 1.0, + "content": "Ben Poole†", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 243, + 125, + 325, + 139 + ], + "spans": [ + { + "bbox": [ + 243, + 125, + 325, + 139 + ], + "score": 1.0, + "content": "Stanford University", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 244, + 138, + 372, + 149 + ], + "spans": [ + { + "bbox": [ + 244, + 138, + 372, + 149 + ], + "score": 1.0, + "content": "poole@cs.stanford.edu", + "type": "text" + } + ], + "index": 8, + "is_list_start_line": true + } + ], + "index": 4, + "bbox_fs": [ + 111, + 114, + 211, + 150 + ] + }, + { + "type": "list", + "bbox": [ + 244, + 114, + 372, + 149 + ], + "lines": [], + "index": 5, + "bbox_fs": [ + 243, + 114, + 372, + 149 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 406, + 115, + 498, + 149 + ], + "lines": [ + { + "bbox": [ + 406, + 115, + 457, + 127 + ], + "spans": [ + { + "bbox": [ + 406, + 115, + 457, + 127 + ], + "score": 1.0, + "content": "David Pfau", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 406, + 126, + 484, + 139 + ], + "spans": [ + { + "bbox": [ + 406, + 126, + 484, + 139 + ], + "score": 1.0, + "content": "Google DeepMind", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 405, + 138, + 499, + 149 + ], + "spans": [ + { + "bbox": [ + 405, + 138, + 499, + 149 + ], + "score": 1.0, + "content": "pfau@google.com", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6, + "bbox_fs": [ + 405, + 115, + 499, + 149 + ] + }, + { + "type": "text", + "bbox": [ + 112, + 165, + 228, + 199 + ], + "lines": [ + { + "bbox": [ + 111, + 164, + 208, + 177 + ], + "spans": [ + { + "bbox": [ + 111, + 164, + 208, + 177 + ], + "score": 1.0, + "content": "Jascha Sohl-Dickstein", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 112, + 176, + 169, + 188 + ], + "spans": [ + { + "bbox": [ + 112, + 176, + 169, + 188 + ], + "score": 1.0, + "content": "Google Brain", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 113, + 188, + 228, + 200 + ], + "spans": [ + { + "bbox": [ + 113, + 188, + 228, + 200 + ], + "score": 1.0, + "content": "jaschasd@google.com", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11, + "bbox_fs": [ + 111, + 164, + 228, + 200 + ] + }, + { + "type": "title", + "bbox": [ + 278, + 228, + 333, + 240 + ], + "lines": [ + { + "bbox": [ + 276, + 226, + 336, + 242 + ], + "spans": [ + { + "bbox": [ + 276, + 226, + 336, + 242 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 143, + 252, + 468, + 341 + ], + "lines": [ + { + "bbox": [ + 141, + 252, + 469, + 266 + ], + "spans": [ + { + "bbox": [ + 141, + 252, + 469, + 266 + ], + "score": 1.0, + "content": "We introduce a method to stabilize Generative Adversarial Networks (GANs) by", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 142, + 264, + 469, + 276 + ], + "spans": [ + { + "bbox": [ + 142, + 264, + 469, + 276 + ], + "score": 1.0, + "content": "defining the generator objective with respect to an unrolled optimization of the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 274, + 469, + 287 + ], + "spans": [ + { + "bbox": [ + 141, + 274, + 469, + 287 + ], + "score": 1.0, + "content": "discriminator. This allows training to be adjusted between using the optimal dis-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 142, + 286, + 469, + 298 + ], + "spans": [ + { + "bbox": [ + 142, + 286, + 469, + 298 + ], + "score": 1.0, + "content": "criminator in the generator’s objective, which is ideal but infeasible in practice,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 142, + 297, + 470, + 309 + ], + "spans": [ + { + "bbox": [ + 142, + 297, + 470, + 309 + ], + "score": 1.0, + "content": "and using the current value of the discriminator, which is often unstable and leads", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 308, + 470, + 320 + ], + "spans": [ + { + "bbox": [ + 141, + 308, + 470, + 320 + ], + "score": 1.0, + "content": "to poor solutions. We show how this technique solves the common problem of", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 318, + 470, + 332 + ], + "spans": [ + { + "bbox": [ + 141, + 318, + 470, + 332 + ], + "score": 1.0, + "content": "mode collapse, stabilizes training of GANs with complex recurrent generators,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 329, + 452, + 343 + ], + "spans": [ + { + "bbox": [ + 141, + 329, + 452, + 343 + ], + "score": 1.0, + "content": "and increases diversity and coverage of the data distribution by the generator.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 17.5, + "bbox_fs": [ + 141, + 252, + 470, + 343 + ] + }, + { + "type": "title", + "bbox": [ + 109, + 362, + 206, + 375 + ], + "lines": [ + { + "bbox": [ + 105, + 361, + 208, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 208, + 378 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 388, + 505, + 486 + ], + "lines": [ + { + "bbox": [ + 106, + 387, + 506, + 401 + ], + "spans": [ + { + "bbox": [ + 106, + 387, + 506, + 401 + ], + "score": 1.0, + "content": "The use of deep neural networks as generative models for complex data has made great advances", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 398, + 506, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 506, + 412 + ], + "score": 1.0, + "content": "in recent years. This success has been achieved through a surprising diversity of training losses", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 410, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 106, + 410, + 505, + 421 + ], + "score": 1.0, + "content": "and model architectures, including denoising autoencoders (Vincent et al., 2010), variational au-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 420, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 506, + 433 + ], + "score": 1.0, + "content": "toencoders (Kingma & Welling, 2013; Rezende et al., 2014; Gregor et al., 2015; Kulkarni et al.,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 430, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 505, + 444 + ], + "score": 1.0, + "content": "2015; Burda et al., 2015; Kingma et al., 2016), generative stochastic networks (Alain et al., 2015),", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 442, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 505, + 456 + ], + "score": 1.0, + "content": "diffusion probabilistic models (Sohl-Dickstein et al., 2015), autoregressive models (Theis & Bethge,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 452, + 505, + 466 + ], + "spans": [ + { + "bbox": [ + 105, + 452, + 505, + 466 + ], + "score": 1.0, + "content": "2015; van den Oord et al., 2016a;b), real non-volume preserving transformations (Dinh et al., 2014;", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 464, + 505, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 505, + 477 + ], + "score": 1.0, + "content": "2016), Helmholtz machines (Dayan et al., 1995; Bornschein et al., 2015), and Generative Adversar-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 475, + 299, + 487 + ], + "spans": [ + { + "bbox": [ + 106, + 475, + 299, + 487 + ], + "score": 1.0, + "content": "ial Networks (GANs) (Goodfellow et al., 2014).", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 387, + 506, + 487 + ] + }, + { + "type": "title", + "bbox": [ + 109, + 500, + 304, + 511 + ], + "lines": [ + { + "bbox": [ + 106, + 500, + 306, + 513 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 306, + 513 + ], + "score": 1.0, + "content": "1.1 GENERATIVE ADVERSARIAL NETWORKS", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 521, + 505, + 620 + ], + "lines": [ + { + "bbox": [ + 106, + 521, + 505, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 521, + 505, + 534 + ], + "score": 1.0, + "content": "While most deep generative models are trained by maximizing log likelihood or a lower bound on", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 532, + 505, + 544 + ], + "spans": [ + { + "bbox": [ + 106, + 532, + 505, + 544 + ], + "score": 1.0, + "content": "log likelihood, GANs take a radically different approach that does not require inference or explicit", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 542, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 505, + 555 + ], + "score": 1.0, + "content": "calculation of the data likelihood. Instead, two models are used to solve a minimax game: a genera-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 552, + 505, + 568 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 505, + 568 + ], + "score": 1.0, + "content": "tor which samples data, and a discriminator which classifies the data as real or generated. In theory", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 563, + 506, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 506, + 579 + ], + "score": 1.0, + "content": "these models are capable of modeling an arbitrarily complex probability distribution. When using", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 577, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 505, + 588 + ], + "score": 1.0, + "content": "the optimal discriminator for a given class of generators, the original GAN proposed by Goodfellow", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 586, + 505, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 505, + 600 + ], + "score": 1.0, + "content": "et al. minimizes the Jensen-Shannon divergence between the data distribution and the generator,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 598, + 505, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 598, + 505, + 611 + ], + "score": 1.0, + "content": "and extensions generalize this to a wider class of divergences (Nowozin et al., 2016; Sonderby et al.,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 609, + 209, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 609, + 209, + 621 + ], + "score": 1.0, + "content": "2016; Poole et al., 2016).", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 37, + "bbox_fs": [ + 105, + 521, + 506, + 621 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 626, + 505, + 703 + ], + "lines": [ + { + "bbox": [ + 105, + 624, + 505, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 505, + 639 + ], + "score": 1.0, + "content": "The ability to train extremely flexible generating functions, without explicitly computing likeli-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 636, + 506, + 649 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 506, + 649 + ], + "score": 1.0, + "content": "hoods or performing inference, and while targeting more mode-seeking divergences as made GANs", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 647, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 105, + 647, + 506, + 661 + ], + "score": 1.0, + "content": "extremely successful in image generation (Odena et al., 2016; Salimans et al., 2016; Radford et al.,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 658, + 506, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 506, + 672 + ], + "score": 1.0, + "content": "2015), and image super resolution (Ledig et al., 2016). The flexibility of the GAN framework has", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 670, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 106, + 670, + 505, + 681 + ], + "score": 1.0, + "content": "also enabled a number of successful extensions of the technique, for instance for structured predic-", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 681, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 106, + 681, + 505, + 693 + ], + "score": 1.0, + "content": "tion (Reed et al., 2016a;b; Odena et al., 2016), training energy based models (Zhao et al., 2016), and", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 106, + 692, + 412, + 704 + ], + "spans": [ + { + "bbox": [ + 106, + 692, + 412, + 704 + ], + "score": 1.0, + "content": "combining the GAN loss with a mutual information loss (Chen et al., 2016).", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 45, + "bbox_fs": [ + 105, + 624, + 506, + 704 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 182 + ], + "lines": [ + { + "bbox": [ + 106, + 83, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 83, + 505, + 95 + ], + "score": 1.0, + "content": "In practice, however, GANs suffer from many issues, particularly during training. One common", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "failure mode involves the generator collapsing to produce only a single sample or a small family of", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 506, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 506, + 119 + ], + "score": 1.0, + "content": "very similar samples. Another involves the generator and discriminator oscillating during training,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "rather than converging to a fixed point. In addition, if one agent becomes much more powerful than", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 139 + ], + "score": 1.0, + "content": "the other, the learning signal to the other agent becomes useless, and the system does not learn.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 137, + 504, + 149 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 504, + 149 + ], + "score": 1.0, + "content": "To train GANs many tricks must be employed, such as careful selection of architectures (Radford", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 148, + 505, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 505, + 160 + ], + "score": 1.0, + "content": "et al., 2015), minibatch discrimination (Salimans et al., 2016), and noise injection (Salimans et al.,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 158, + 506, + 172 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 506, + 172 + ], + "score": 1.0, + "content": "2016; Sonderby et al., 2016). Even with these tricks the set of hyperparameters for which training is", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 171, + 289, + 183 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 289, + 183 + ], + "score": 1.0, + "content": "successful is generally very small in practice.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 107, + 187, + 505, + 318 + ], + "lines": [ + { + "bbox": [ + 105, + 186, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 506, + 200 + ], + "score": 1.0, + "content": "Once converged, the generative models produced by the GAN training procedure normally do not", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 198, + 505, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 505, + 211 + ], + "score": 1.0, + "content": "cover the whole distribution (Dumoulin et al., 2016; Che et al., 2016), even when targeting a mode-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 209, + 506, + 223 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 506, + 223 + ], + "score": 1.0, + "content": "covering divergence such as KL. Additionally, because it is intractable to compute the GAN training", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 221, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 221, + 505, + 232 + ], + "score": 1.0, + "content": "loss, and because approximate measures of performance such as Parzen window estimates suffer", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 231, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 505, + 243 + ], + "score": 1.0, + "content": "from major flaws (Theis et al., 2016), evaluation of GAN performance is challenging. Currently,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 242, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 505, + 254 + ], + "score": 1.0, + "content": "human judgement of sample quality is one of the leading metrics for evaluating GANs. In practice", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 253, + 505, + 265 + ], + "spans": [ + { + "bbox": [ + 105, + 253, + 505, + 265 + ], + "score": 1.0, + "content": "this metric does not take into account mode dropping if the number of modes is greater than the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 264, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 505, + 276 + ], + "score": 1.0, + "content": "number of samples one is visualizing. In fact, the mode dropping problem generally helps visual", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 275, + 506, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 506, + 288 + ], + "score": 1.0, + "content": "sample quality as the model can choose to focus on only the most common modes. These common", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 286, + 506, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 286, + 506, + 299 + ], + "score": 1.0, + "content": "modes correspond, by definition, to more typical samples. Additionally, the generative model is able", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 297, + 506, + 310 + ], + "spans": [ + { + "bbox": [ + 105, + 297, + 506, + 310 + ], + "score": 1.0, + "content": "to allocate more expressive power to the modes it does cover than it would if it attempted to cover", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 307, + 150, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 150, + 320 + ], + "score": 1.0, + "content": "all modes.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 14.5 + }, + { + "type": "title", + "bbox": [ + 108, + 332, + 322, + 343 + ], + "lines": [ + { + "bbox": [ + 105, + 331, + 323, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 331, + 323, + 345 + ], + "score": 1.0, + "content": "1.2 DIFFERENTIATING THROUGH OPTIMIZATION", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 352, + 505, + 474 + ], + "lines": [ + { + "bbox": [ + 106, + 353, + 504, + 365 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 504, + 365 + ], + "score": 1.0, + "content": "Many optimization schemes, including SGD, RMSProp (Tieleman & Hinton, 2012), and Adam", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 364, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 106, + 364, + 505, + 376 + ], + "score": 1.0, + "content": "(Kingma & Ba, 2014), consist of a sequence of differentiable updates to parameters. Gradients can", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 374, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 505, + 388 + ], + "score": 1.0, + "content": "be backpropagated through unrolled optimization updates in a similar fashion to backpropagation", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 386, + 504, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 504, + 398 + ], + "score": 1.0, + "content": "through a recurrent neural network. The parameters output by the optimizer can thus be included,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 397, + 506, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 506, + 410 + ], + "score": 1.0, + "content": "in a differentiable way, in another objective (Maclaurin et al., 2015). This idea was first suggested", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 407, + 506, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 506, + 421 + ], + "score": 1.0, + "content": "for minimax problems in (Pearlmutter & Siskind, 2008), while (Zhang & Lesser, 2010) provided", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 419, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 419, + 505, + 431 + ], + "score": 1.0, + "content": "a theoretical analysis and experimental results on differentiating through a single step of gradient", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "score": 1.0, + "content": "ascent for simple matrix games. Differentiating through unrolled optimization was first scaled to", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 440, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 440, + 505, + 453 + ], + "score": 1.0, + "content": "deep networks in (Maclaurin et al., 2015), where it was used for hyperparameter optimization. More", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 451, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 505, + 464 + ], + "score": 1.0, + "content": "recently, (Belanger & McCallum, 2015; Han et al., 2016; Andrychowicz et al., 2016) backpropagate", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 462, + 442, + 475 + ], + "spans": [ + { + "bbox": [ + 106, + 462, + 442, + 475 + ], + "score": 1.0, + "content": "through optimization procedures in contexts unrelated to GANs or minimax games.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 106, + 479, + 503, + 502 + ], + "lines": [ + { + "bbox": [ + 105, + 478, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 505, + 493 + ], + "score": 1.0, + "content": "In this work we address the challenges of unstable optimization and mode collapse in GANs by", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 489, + 383, + 505 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 383, + 505 + ], + "score": 1.0, + "content": "unrolling optimization of the discriminator objective during training.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "title", + "bbox": [ + 107, + 517, + 173, + 530 + ], + "lines": [ + { + "bbox": [ + 104, + 515, + 174, + 533 + ], + "spans": [ + { + "bbox": [ + 104, + 515, + 174, + 533 + ], + "score": 1.0, + "content": "2 METHOD", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "title", + "bbox": [ + 107, + 542, + 305, + 553 + ], + "lines": [ + { + "bbox": [ + 105, + 541, + 306, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 306, + 554 + ], + "score": 1.0, + "content": "2.1 GENERATIVE ADVERSARIAL NETWORKS", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 107, + 562, + 503, + 585 + ], + "lines": [ + { + "bbox": [ + 105, + 560, + 504, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 352, + 576 + ], + "score": 1.0, + "content": "The GAN learning problem is to find the optimal parameters", + "type": "text" + }, + { + "bbox": [ + 353, + 563, + 365, + 575 + ], + "score": 0.9, + "content": "\\theta _ { G } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 560, + 464, + 576 + ], + "score": 1.0, + "content": "for a generator function", + "type": "text" + }, + { + "bbox": [ + 465, + 562, + 504, + 574 + ], + "score": 0.92, + "content": "G \\left( z ; \\theta _ { G } \\right)", + "type": "inline_equation" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 573, + 203, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 203, + 587 + ], + "score": 1.0, + "content": "in a minimax objective,", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "interline_equation", + "bbox": [ + 234, + 587, + 378, + 654 + ], + "lines": [ + { + "bbox": [ + 234, + 587, + 378, + 654 + ], + "spans": [ + { + "bbox": [ + 234, + 587, + 378, + 654 + ], + "score": 0.93, + "content": "\\begin{array} { c } { { \\theta _ { G } ^ { * } = \\underset { \\theta _ { G } } { \\mathrm { a r g m i n } } \\underset { \\theta _ { D } } { \\mathrm { m a x } } f \\left( \\theta _ { G } , \\theta _ { D } \\right) } } \\\\ { { \\ \\mathrm { ~ } } } \\\\ { { \\displaystyle \\qquad = \\underset { \\theta _ { G } } { \\mathrm { a r g m i n } } f \\left( \\theta _ { G } , \\theta _ { D } ^ { * } \\left( \\theta _ { G } \\right) \\right) } } \\\\ { { \\theta _ { D } ^ { * } \\left( \\theta _ { G } \\right) = \\underset { \\theta _ { D } } { \\mathrm { a r g m a x } } f \\left( \\theta _ { G } , \\theta _ { D } \\right) , } } \\end{array}", + "type": "interline_equation", + "image_path": "b2008d380b8f084085a49f2a632840f8a4144bbae5800b215f3679c79653344b.jpg" + } + ] + } + ], + "index": 39.5, + "virtual_lines": [ + { + "bbox": [ + 234, + 587, + 378, + 620.5 + ], + "spans": [], + "index": 39 + }, + { + "bbox": [ + 234, + 620.5, + 378, + 654.0 + ], + "spans": [], + "index": 40 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 656, + 248, + 668 + ], + "lines": [ + { + "bbox": [ + 106, + 656, + 248, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 656, + 133, + 669 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 657, + 140, + 668 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 656, + 248, + 669 + ], + "score": 1.0, + "content": "is commonly chosen to be", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 41 + }, + { + "type": "interline_equation", + "bbox": [ + 133, + 671, + 477, + 685 + ], + "lines": [ + { + "bbox": [ + 133, + 671, + 477, + 685 + ], + "spans": [ + { + "bbox": [ + 133, + 671, + 477, + 685 + ], + "score": 0.89, + "content": "f \\left( \\theta _ { G } , \\theta _ { D } \\right) = \\mathbb { E } _ { x \\sim p _ { d a t a } } \\left[ \\log \\left( D \\left( x ; \\theta _ { D } \\right) \\right) \\right] + \\mathbb { E } _ { z \\sim N ( 0 , I ) } \\left[ \\log \\left( 1 - D \\left( G \\left( z ; \\theta _ { G } \\right) ; \\theta _ { D } \\right) \\right) \\right] .", + "type": "interline_equation", + "image_path": "9ee6df4b3e18c286fdee617b2aa953c06b25b1f3424af4b30db2853ba740dd9e.jpg" + } + ] + } + ], + "index": 42, + "virtual_lines": [ + { + "bbox": [ + 133, + 671, + 477, + 685 + ], + "spans": [], + "index": 42 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 687, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 129, + 701 + ], + "score": 1.0, + "content": "Here", + "type": "text" + }, + { + "bbox": [ + 129, + 688, + 161, + 698 + ], + "score": 0.92, + "content": "x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 687, + 246, + 701 + ], + "score": 1.0, + "content": "is the data variable,", + "type": "text" + }, + { + "bbox": [ + 246, + 688, + 276, + 698 + ], + "score": 0.9, + "content": "z \\in { \\mathcal { Z } }", + "type": "inline_equation" + }, + { + "bbox": [ + 277, + 687, + 367, + 701 + ], + "score": 1.0, + "content": "is the latent variable,", + "type": "text" + }, + { + "bbox": [ + 368, + 690, + 390, + 699 + ], + "score": 0.88, + "content": "p _ { d a t a }", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 687, + 506, + 701 + ], + "score": 1.0, + "content": "is the data distribution, the", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 162, + 711 + ], + "score": 1.0, + "content": "discriminator", + "type": "text" + }, + { + "bbox": [ + 163, + 699, + 255, + 710 + ], + "score": 0.89, + "content": "D ( \\cdot ; \\theta _ { D } ) : \\mathcal { X } [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 699, + 446, + 711 + ], + "score": 1.0, + "content": "outputs the estimated probability that a sample", + "type": "text" + }, + { + "bbox": [ + 446, + 701, + 453, + 709 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "comes from", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 192, + 722 + ], + "score": 1.0, + "content": "the data distribution,", + "type": "text" + }, + { + "bbox": [ + 192, + 711, + 205, + 721 + ], + "score": 0.87, + "content": "\\theta _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 709, + 225, + 722 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 225, + 711, + 237, + 721 + ], + "score": 0.87, + "content": "\\theta _ { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "are the discriminator and generator parameters, and the generator", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 142, + 734 + ], + "score": 1.0, + "content": "function", + "type": "text" + }, + { + "bbox": [ + 142, + 720, + 219, + 732 + ], + "score": 0.92, + "content": "G ( \\cdot ; \\theta _ { G } ) : \\mathcal { Z } \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 720, + 505, + 734 + ], + "score": 1.0, + "content": "transforms a sample in the latent space into a sample in the data space.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 44.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 182 + ], + "lines": [ + { + "bbox": [ + 106, + 83, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 83, + 505, + 95 + ], + "score": 1.0, + "content": "In practice, however, GANs suffer from many issues, particularly during training. One common", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "failure mode involves the generator collapsing to produce only a single sample or a small family of", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 506, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 506, + 119 + ], + "score": 1.0, + "content": "very similar samples. Another involves the generator and discriminator oscillating during training,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "rather than converging to a fixed point. In addition, if one agent becomes much more powerful than", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 126, + 505, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 126, + 505, + 139 + ], + "score": 1.0, + "content": "the other, the learning signal to the other agent becomes useless, and the system does not learn.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 137, + 504, + 149 + ], + "spans": [ + { + "bbox": [ + 106, + 137, + 504, + 149 + ], + "score": 1.0, + "content": "To train GANs many tricks must be employed, such as careful selection of architectures (Radford", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 148, + 505, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 148, + 505, + 160 + ], + "score": 1.0, + "content": "et al., 2015), minibatch discrimination (Salimans et al., 2016), and noise injection (Salimans et al.,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 158, + 506, + 172 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 506, + 172 + ], + "score": 1.0, + "content": "2016; Sonderby et al., 2016). Even with these tricks the set of hyperparameters for which training is", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 171, + 289, + 183 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 289, + 183 + ], + "score": 1.0, + "content": "successful is generally very small in practice.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 4, + "bbox_fs": [ + 105, + 83, + 506, + 183 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 187, + 505, + 318 + ], + "lines": [ + { + "bbox": [ + 105, + 186, + 506, + 200 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 506, + 200 + ], + "score": 1.0, + "content": "Once converged, the generative models produced by the GAN training procedure normally do not", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 198, + 505, + 211 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 505, + 211 + ], + "score": 1.0, + "content": "cover the whole distribution (Dumoulin et al., 2016; Che et al., 2016), even when targeting a mode-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 209, + 506, + 223 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 506, + 223 + ], + "score": 1.0, + "content": "covering divergence such as KL. Additionally, because it is intractable to compute the GAN training", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 221, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 105, + 221, + 505, + 232 + ], + "score": 1.0, + "content": "loss, and because approximate measures of performance such as Parzen window estimates suffer", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 231, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 105, + 231, + 505, + 243 + ], + "score": 1.0, + "content": "from major flaws (Theis et al., 2016), evaluation of GAN performance is challenging. Currently,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 242, + 505, + 254 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 505, + 254 + ], + "score": 1.0, + "content": "human judgement of sample quality is one of the leading metrics for evaluating GANs. In practice", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 253, + 505, + 265 + ], + "spans": [ + { + "bbox": [ + 105, + 253, + 505, + 265 + ], + "score": 1.0, + "content": "this metric does not take into account mode dropping if the number of modes is greater than the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 264, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 505, + 276 + ], + "score": 1.0, + "content": "number of samples one is visualizing. In fact, the mode dropping problem generally helps visual", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 275, + 506, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 506, + 288 + ], + "score": 1.0, + "content": "sample quality as the model can choose to focus on only the most common modes. These common", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 286, + 506, + 299 + ], + "spans": [ + { + "bbox": [ + 105, + 286, + 506, + 299 + ], + "score": 1.0, + "content": "modes correspond, by definition, to more typical samples. Additionally, the generative model is able", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 297, + 506, + 310 + ], + "spans": [ + { + "bbox": [ + 105, + 297, + 506, + 310 + ], + "score": 1.0, + "content": "to allocate more expressive power to the modes it does cover than it would if it attempted to cover", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 307, + 150, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 150, + 320 + ], + "score": 1.0, + "content": "all modes.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 186, + 506, + 320 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 332, + 322, + 343 + ], + "lines": [ + { + "bbox": [ + 105, + 331, + 323, + 345 + ], + "spans": [ + { + "bbox": [ + 105, + 331, + 323, + 345 + ], + "score": 1.0, + "content": "1.2 DIFFERENTIATING THROUGH OPTIMIZATION", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 352, + 505, + 474 + ], + "lines": [ + { + "bbox": [ + 106, + 353, + 504, + 365 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 504, + 365 + ], + "score": 1.0, + "content": "Many optimization schemes, including SGD, RMSProp (Tieleman & Hinton, 2012), and Adam", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 364, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 106, + 364, + 505, + 376 + ], + "score": 1.0, + "content": "(Kingma & Ba, 2014), consist of a sequence of differentiable updates to parameters. Gradients can", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 374, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 505, + 388 + ], + "score": 1.0, + "content": "be backpropagated through unrolled optimization updates in a similar fashion to backpropagation", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 386, + 504, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 386, + 504, + 398 + ], + "score": 1.0, + "content": "through a recurrent neural network. The parameters output by the optimizer can thus be included,", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 397, + 506, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 506, + 410 + ], + "score": 1.0, + "content": "in a differentiable way, in another objective (Maclaurin et al., 2015). This idea was first suggested", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 407, + 506, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 506, + 421 + ], + "score": 1.0, + "content": "for minimax problems in (Pearlmutter & Siskind, 2008), while (Zhang & Lesser, 2010) provided", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 419, + 505, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 419, + 505, + 431 + ], + "score": 1.0, + "content": "a theoretical analysis and experimental results on differentiating through a single step of gradient", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "score": 1.0, + "content": "ascent for simple matrix games. Differentiating through unrolled optimization was first scaled to", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 440, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 440, + 505, + 453 + ], + "score": 1.0, + "content": "deep networks in (Maclaurin et al., 2015), where it was used for hyperparameter optimization. More", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 451, + 505, + 464 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 505, + 464 + ], + "score": 1.0, + "content": "recently, (Belanger & McCallum, 2015; Han et al., 2016; Andrychowicz et al., 2016) backpropagate", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 462, + 442, + 475 + ], + "spans": [ + { + "bbox": [ + 106, + 462, + 442, + 475 + ], + "score": 1.0, + "content": "through optimization procedures in contexts unrelated to GANs or minimax games.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 353, + 506, + 475 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 479, + 503, + 502 + ], + "lines": [ + { + "bbox": [ + 105, + 478, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 505, + 493 + ], + "score": 1.0, + "content": "In this work we address the challenges of unstable optimization and mode collapse in GANs by", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 489, + 383, + 505 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 383, + 505 + ], + "score": 1.0, + "content": "unrolling optimization of the discriminator objective during training.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 478, + 505, + 505 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 517, + 173, + 530 + ], + "lines": [ + { + "bbox": [ + 104, + 515, + 174, + 533 + ], + "spans": [ + { + "bbox": [ + 104, + 515, + 174, + 533 + ], + "score": 1.0, + "content": "2 METHOD", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "title", + "bbox": [ + 107, + 542, + 305, + 553 + ], + "lines": [ + { + "bbox": [ + 105, + 541, + 306, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 306, + 554 + ], + "score": 1.0, + "content": "2.1 GENERATIVE ADVERSARIAL NETWORKS", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 107, + 562, + 503, + 585 + ], + "lines": [ + { + "bbox": [ + 105, + 560, + 504, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 352, + 576 + ], + "score": 1.0, + "content": "The GAN learning problem is to find the optimal parameters", + "type": "text" + }, + { + "bbox": [ + 353, + 563, + 365, + 575 + ], + "score": 0.9, + "content": "\\theta _ { G } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 560, + 464, + 576 + ], + "score": 1.0, + "content": "for a generator function", + "type": "text" + }, + { + "bbox": [ + 465, + 562, + 504, + 574 + ], + "score": 0.92, + "content": "G \\left( z ; \\theta _ { G } \\right)", + "type": "inline_equation" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 573, + 203, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 203, + 587 + ], + "score": 1.0, + "content": "in a minimax objective,", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 560, + 504, + 587 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 234, + 587, + 378, + 654 + ], + "lines": [ + { + "bbox": [ + 234, + 587, + 378, + 654 + ], + "spans": [ + { + "bbox": [ + 234, + 587, + 378, + 654 + ], + "score": 0.93, + "content": "\\begin{array} { c } { { \\theta _ { G } ^ { * } = \\underset { \\theta _ { G } } { \\mathrm { a r g m i n } } \\underset { \\theta _ { D } } { \\mathrm { m a x } } f \\left( \\theta _ { G } , \\theta _ { D } \\right) } } \\\\ { { \\ \\mathrm { ~ } } } \\\\ { { \\displaystyle \\qquad = \\underset { \\theta _ { G } } { \\mathrm { a r g m i n } } f \\left( \\theta _ { G } , \\theta _ { D } ^ { * } \\left( \\theta _ { G } \\right) \\right) } } \\\\ { { \\theta _ { D } ^ { * } \\left( \\theta _ { G } \\right) = \\underset { \\theta _ { D } } { \\mathrm { a r g m a x } } f \\left( \\theta _ { G } , \\theta _ { D } \\right) , } } \\end{array}", + "type": "interline_equation", + "image_path": "b2008d380b8f084085a49f2a632840f8a4144bbae5800b215f3679c79653344b.jpg" + } + ] + } + ], + "index": 39.5, + "virtual_lines": [ + { + "bbox": [ + 234, + 587, + 378, + 620.5 + ], + "spans": [], + "index": 39 + }, + { + "bbox": [ + 234, + 620.5, + 378, + 654.0 + ], + "spans": [], + "index": 40 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 656, + 248, + 668 + ], + "lines": [ + { + "bbox": [ + 106, + 656, + 248, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 656, + 133, + 669 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 657, + 140, + 668 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 656, + 248, + 669 + ], + "score": 1.0, + "content": "is commonly chosen to be", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 41, + "bbox_fs": [ + 106, + 656, + 248, + 669 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 133, + 671, + 477, + 685 + ], + "lines": [ + { + "bbox": [ + 133, + 671, + 477, + 685 + ], + "spans": [ + { + "bbox": [ + 133, + 671, + 477, + 685 + ], + "score": 0.89, + "content": "f \\left( \\theta _ { G } , \\theta _ { D } \\right) = \\mathbb { E } _ { x \\sim p _ { d a t a } } \\left[ \\log \\left( D \\left( x ; \\theta _ { D } \\right) \\right) \\right] + \\mathbb { E } _ { z \\sim N ( 0 , I ) } \\left[ \\log \\left( 1 - D \\left( G \\left( z ; \\theta _ { G } \\right) ; \\theta _ { D } \\right) \\right) \\right] .", + "type": "interline_equation", + "image_path": "9ee6df4b3e18c286fdee617b2aa953c06b25b1f3424af4b30db2853ba740dd9e.jpg" + } + ] + } + ], + "index": 42, + "virtual_lines": [ + { + "bbox": [ + 133, + 671, + 477, + 685 + ], + "spans": [], + "index": 42 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 687, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 129, + 701 + ], + "score": 1.0, + "content": "Here", + "type": "text" + }, + { + "bbox": [ + 129, + 688, + 161, + 698 + ], + "score": 0.92, + "content": "x \\in \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 687, + 246, + 701 + ], + "score": 1.0, + "content": "is the data variable,", + "type": "text" + }, + { + "bbox": [ + 246, + 688, + 276, + 698 + ], + "score": 0.9, + "content": "z \\in { \\mathcal { Z } }", + "type": "inline_equation" + }, + { + "bbox": [ + 277, + 687, + 367, + 701 + ], + "score": 1.0, + "content": "is the latent variable,", + "type": "text" + }, + { + "bbox": [ + 368, + 690, + 390, + 699 + ], + "score": 0.88, + "content": "p _ { d a t a }", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 687, + 506, + 701 + ], + "score": 1.0, + "content": "is the data distribution, the", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 162, + 711 + ], + "score": 1.0, + "content": "discriminator", + "type": "text" + }, + { + "bbox": [ + 163, + 699, + 255, + 710 + ], + "score": 0.89, + "content": "D ( \\cdot ; \\theta _ { D } ) : \\mathcal { X } [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 699, + 446, + 711 + ], + "score": 1.0, + "content": "outputs the estimated probability that a sample", + "type": "text" + }, + { + "bbox": [ + 446, + 701, + 453, + 709 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "comes from", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 192, + 722 + ], + "score": 1.0, + "content": "the data distribution,", + "type": "text" + }, + { + "bbox": [ + 192, + 711, + 205, + 721 + ], + "score": 0.87, + "content": "\\theta _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 709, + 225, + 722 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 225, + 711, + 237, + 721 + ], + "score": 0.87, + "content": "\\theta _ { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "are the discriminator and generator parameters, and the generator", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 142, + 734 + ], + "score": 1.0, + "content": "function", + "type": "text" + }, + { + "bbox": [ + 142, + 720, + 219, + 732 + ], + "score": 0.92, + "content": "G ( \\cdot ; \\theta _ { G } ) : \\mathcal { Z } \\mathcal { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 720, + 505, + 734 + ], + "score": 1.0, + "content": "transforms a sample in the latent space into a sample in the data space.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 44.5, + "bbox_fs": [ + 105, + 687, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 335, + 95 + ], + "score": 1.0, + "content": "For the minimax loss in Eq. 4, the optimal discriminator", + "type": "text" + }, + { + "bbox": [ + 335, + 82, + 364, + 95 + ], + "score": 0.92, + "content": "D ^ { \\ast } \\left( x \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "is a known smooth function of the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 328, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 192, + 106 + ], + "score": 1.0, + "content": "generator probability", + "type": "text" + }, + { + "bbox": [ + 193, + 94, + 221, + 106 + ], + "score": 0.92, + "content": "p _ { G } \\left( x \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 93, + 328, + 106 + ], + "score": 1.0, + "content": "(Goodfellow et al., 2014),", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 243, + 109, + 368, + 137 + ], + "lines": [ + { + "bbox": [ + 243, + 109, + 368, + 137 + ], + "spans": [ + { + "bbox": [ + 243, + 109, + 368, + 137 + ], + "score": 0.95, + "content": "D ^ { * } \\left( x \\right) = \\frac { p _ { d a t a } \\left( x \\right) } { p _ { d a t a } \\left( x \\right) + p _ { G } \\left( x \\right) } .", + "type": "interline_equation", + "image_path": "0e2376c104ce7eecf400d8575cb56227acb0e45882302fbde8854a5f2d2b2c6b.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 243, + 109, + 368, + 137 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 141, + 505, + 197 + ], + "lines": [ + { + "bbox": [ + 106, + 141, + 504, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 141, + 372, + 155 + ], + "score": 1.0, + "content": "When the generator loss in Eq. 2 is rewritten directly in terms of", + "type": "text" + }, + { + "bbox": [ + 372, + 142, + 400, + 154 + ], + "score": 0.92, + "content": "p _ { G } \\left( x \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 141, + 491, + 155 + ], + "score": 1.0, + "content": "and Eq. 5 rather than", + "type": "text" + }, + { + "bbox": [ + 491, + 142, + 504, + 153 + ], + "score": 0.87, + "content": "\\theta _ { G }", + "type": "inline_equation" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 152, + 506, + 166 + ], + "spans": [ + { + "bbox": [ + 105, + 152, + 124, + 166 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 153, + 158, + 165 + ], + "score": 0.75, + "content": "\\theta _ { D } ^ { * } \\left( \\theta _ { G } ^ { - } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 152, + 331, + 166 + ], + "score": 1.0, + "content": ", then it is similarly a smooth function of", + "type": "text" + }, + { + "bbox": [ + 331, + 153, + 359, + 165 + ], + "score": 0.91, + "content": "p _ { G } \\left( x \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 152, + 506, + 166 + ], + "score": 1.0, + "content": ". These smoothness guarantees are", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 163, + 506, + 177 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 185, + 177 + ], + "score": 1.0, + "content": "typically lost when", + "type": "text" + }, + { + "bbox": [ + 185, + 164, + 226, + 176 + ], + "score": 0.93, + "content": "D \\left( x ; \\theta _ { D } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 163, + 245, + 177 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 245, + 164, + 284, + 176 + ], + "score": 0.93, + "content": "G \\left( z ; \\theta _ { G } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 163, + 506, + 177 + ], + "score": 1.0, + "content": "are drawn from parametric families. They nonetheless", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 174, + 506, + 188 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 506, + 188 + ], + "score": 1.0, + "content": "suggest that the true generator objective in Eq. 2 will often be well behaved, and is a desirable target", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 185, + 200, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 185, + 200, + 199 + ], + "score": 1.0, + "content": "for direct optimization.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 107, + 202, + 505, + 247 + ], + "lines": [ + { + "bbox": [ + 105, + 203, + 505, + 216 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 351, + 216 + ], + "score": 1.0, + "content": "Explicitly solving for the optimal discriminator parameters", + "type": "text" + }, + { + "bbox": [ + 352, + 203, + 385, + 215 + ], + "score": 0.82, + "content": "\\theta _ { D } ^ { * } \\left( \\theta _ { G } \\right) ", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 203, + 505, + 216 + ], + "score": 1.0, + "content": "for every update step of the", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 213, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 147, + 226 + ], + "score": 1.0, + "content": "generator", + "type": "text" + }, + { + "bbox": [ + 147, + 214, + 156, + 223 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 213, + 505, + 226 + ], + "score": 1.0, + "content": "is computationally infeasible for discriminators based on neural networks. Therefore", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 225, + 505, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 473, + 237 + ], + "score": 1.0, + "content": "this minimax optimization problem is typically solved by alternating gradient descent on", + "type": "text" + }, + { + "bbox": [ + 473, + 225, + 486, + 236 + ], + "score": 0.88, + "content": "\\theta _ { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 225, + 505, + 237 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 236, + 163, + 247 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 146, + 247 + ], + "score": 1.0, + "content": "ascent on", + "type": "text" + }, + { + "bbox": [ + 146, + 236, + 159, + 246 + ], + "score": 0.88, + "content": "\\theta _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 236, + 163, + 247 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 106, + 252, + 505, + 341 + ], + "lines": [ + { + "bbox": [ + 105, + 252, + 506, + 266 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 192, + 266 + ], + "score": 1.0, + "content": "The optimal solution", + "type": "text" + }, + { + "bbox": [ + 192, + 252, + 254, + 264 + ], + "score": 0.92, + "content": "\\theta ^ { * } = \\{ \\theta _ { G } ^ { * } , \\theta _ { D } ^ { * } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 252, + 506, + 266 + ], + "score": 1.0, + "content": "is a fixed point of these iterative learning dynamics. Addition-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 134, + 276 + ], + "score": 1.0, + "content": "ally, if", + "type": "text" + }, + { + "bbox": [ + 135, + 264, + 178, + 276 + ], + "score": 0.93, + "content": "f \\left( { \\theta } _ { G } , { \\theta } _ { D } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 263, + 228, + 276 + ], + "score": 1.0, + "content": "is convex in", + "type": "text" + }, + { + "bbox": [ + 228, + 265, + 241, + 275 + ], + "score": 0.88, + "content": "\\theta _ { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 263, + 303, + 276 + ], + "score": 1.0, + "content": "and concave in", + "type": "text" + }, + { + "bbox": [ + 303, + 264, + 316, + 275 + ], + "score": 0.88, + "content": "\\theta _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 263, + 505, + 276 + ], + "score": 1.0, + "content": ", then alternating gradient descent (ascent) trust", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 274, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 104, + 274, + 505, + 288 + ], + "score": 1.0, + "content": "region updates are guaranteed to converge to the fixed point, under certain additional weak assump-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 285, + 504, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 286, + 303, + 298 + ], + "score": 1.0, + "content": "tions (Juditsky et al., 2011). However in practice", + "type": "text" + }, + { + "bbox": [ + 303, + 285, + 347, + 298 + ], + "score": 0.93, + "content": "f \\left( { \\theta } _ { G } , { \\theta } _ { D } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 286, + 491, + 298 + ], + "score": 1.0, + "content": "is typically very far from convex in", + "type": "text" + }, + { + "bbox": [ + 491, + 286, + 504, + 297 + ], + "score": 0.88, + "content": "\\theta _ { G }", + "type": "inline_equation" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 295, + 506, + 310 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 168, + 310 + ], + "score": 1.0, + "content": "and concave in", + "type": "text" + }, + { + "bbox": [ + 168, + 297, + 181, + 307 + ], + "score": 0.89, + "content": "\\theta _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 295, + 506, + 310 + ], + "score": 1.0, + "content": ", and updates are not constrained in an appropriate way. As a result GAN training", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 307, + 505, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 307, + 505, + 320 + ], + "score": 1.0, + "content": "suffers from mode collapse, undamped oscillations, and other problems detailed in Section 1.1. In", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 318, + 506, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 439, + 331 + ], + "score": 1.0, + "content": "order to address these difficulties, we will introduce a surrogate objective function", + "type": "text" + }, + { + "bbox": [ + 439, + 318, + 489, + 330 + ], + "score": 0.91, + "content": "f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 318, + 506, + 331 + ], + "score": 1.0, + "content": "for", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 329, + 498, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 429, + 343 + ], + "score": 1.0, + "content": "training the generator which more closely resembles the true generator objective", + "type": "text" + }, + { + "bbox": [ + 429, + 330, + 493, + 342 + ], + "score": 0.9, + "content": "f \\left( \\theta _ { G } , \\theta _ { D } ^ { * } \\left( \\theta _ { G } \\right) \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 329, + 498, + 343 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 15.5 + }, + { + "type": "title", + "bbox": [ + 108, + 353, + 214, + 365 + ], + "lines": [ + { + "bbox": [ + 104, + 352, + 216, + 367 + ], + "spans": [ + { + "bbox": [ + 104, + 352, + 216, + 367 + ], + "score": 1.0, + "content": "2.2 UNROLLING GANS", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 374, + 505, + 397 + ], + "lines": [ + { + "bbox": [ + 105, + 373, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 314, + 388 + ], + "score": 1.0, + "content": "A local optimum of the discriminator parameters", + "type": "text" + }, + { + "bbox": [ + 314, + 375, + 327, + 387 + ], + "score": 0.89, + "content": "\\theta _ { D } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 373, + 505, + 388 + ], + "score": 1.0, + "content": "can be expressed as the fixed point of an", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 386, + 239, + 398 + ], + "spans": [ + { + "bbox": [ + 106, + 386, + 239, + 398 + ], + "score": 1.0, + "content": "iterative optimization procedure,", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + }, + { + "type": "interline_equation", + "bbox": [ + 237, + 399, + 373, + 468 + ], + "lines": [ + { + "bbox": [ + 237, + 399, + 373, + 468 + ], + "spans": [ + { + "bbox": [ + 237, + 399, + 373, + 468 + ], + "score": 0.93, + "content": "\\begin{array} { c } { { \\theta _ { D } ^ { 0 } = \\theta _ { D } } } \\\\ { { \\displaystyle } } \\\\ { { \\theta _ { D } ^ { k + 1 } = \\theta _ { D } ^ { k } + \\eta ^ { k } \\displaystyle \\frac { \\mathrm { d } f ( \\theta _ { G } , \\theta _ { D } ^ { k } ) } { \\mathrm { d } \\theta _ { D } ^ { k } } } } \\\\ { { \\displaystyle } } \\\\ { { \\theta _ { D } ^ { * } ( \\theta _ { G } ) = \\displaystyle \\operatorname* { l i m } _ { k \\infty } \\theta _ { D } ^ { k } , } } \\end{array}", + "type": "interline_equation", + "image_path": "f45244ba45a91984f7cab7a13e843b9f8428ca6f8c95d11c332b9e18669248b2.jpg" + } + ] + } + ], + "index": 23.5, + "virtual_lines": [ + { + "bbox": [ + 237, + 399, + 373, + 433.5 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 237, + 433.5, + 373, + 468.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 473, + 504, + 506 + ], + "lines": [ + { + "bbox": [ + 105, + 471, + 506, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 133, + 486 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 472, + 145, + 484 + ], + "score": 0.9, + "content": "\\eta ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 471, + 506, + 486 + ], + "score": 1.0, + "content": "is the learning rate schedule. For clarity, we have expressed Eq. 7 as a full batch steepest", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 483, + 506, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 506, + 496 + ], + "score": 1.0, + "content": "gradient ascent equation. More sophisticated optimizers can be similarly unrolled. In our experi-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 495, + 292, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 495, + 292, + 507 + ], + "score": 1.0, + "content": "ments we unroll Adam (Kingma & Ba, 2014).", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 105, + 511, + 460, + 523 + ], + "lines": [ + { + "bbox": [ + 105, + 510, + 462, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 174, + 526 + ], + "score": 1.0, + "content": "By unrolling for", + "type": "text" + }, + { + "bbox": [ + 174, + 512, + 184, + 522 + ], + "score": 0.83, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 510, + 462, + 526 + ], + "score": 1.0, + "content": "steps, we create a surrogate objective for the update of the generator,", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "interline_equation", + "bbox": [ + 229, + 527, + 381, + 543 + ], + "lines": [ + { + "bbox": [ + 229, + 527, + 381, + 543 + ], + "spans": [ + { + "bbox": [ + 229, + 527, + 381, + 543 + ], + "score": 0.93, + "content": "f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) = f \\left( \\theta _ { G } , \\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) \\right) .", + "type": "interline_equation", + "image_path": "722de6d2f0bca528a8df9e9213bb3bcba721cff99ce6b981b4c7ab2eda98fd02.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 229, + 527, + 381, + 543 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 547, + 504, + 592 + ], + "lines": [ + { + "bbox": [ + 105, + 546, + 504, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 133, + 560 + ], + "score": 1.0, + "content": "When", + "type": "text" + }, + { + "bbox": [ + 133, + 547, + 163, + 558 + ], + "score": 0.89, + "content": "K = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 546, + 466, + 560 + ], + "score": 1.0, + "content": "this objective corresponds exactly to the standard GAN objective, while as", + "type": "text" + }, + { + "bbox": [ + 467, + 547, + 504, + 558 + ], + "score": 0.88, + "content": "K \\infty", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 558, + 506, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 558, + 325, + 572 + ], + "score": 1.0, + "content": "it corresponds to the true generator objective function", + "type": "text" + }, + { + "bbox": [ + 325, + 558, + 387, + 570 + ], + "score": 0.92, + "content": "f \\left( \\theta _ { G } , \\theta _ { D } ^ { * } \\left( G \\right) \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 558, + 506, + 572 + ], + "score": 1.0, + "content": ". By adjusting the number of", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 570, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 169, + 582 + ], + "score": 1.0, + "content": "unrolling steps", + "type": "text" + }, + { + "bbox": [ + 169, + 570, + 180, + 579 + ], + "score": 0.81, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 570, + 505, + 582 + ], + "score": 1.0, + "content": ", we are thus able to interpolate between standard GAN training dynamics with", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 580, + 461, + 594 + ], + "spans": [ + { + "bbox": [ + 105, + 580, + 461, + 594 + ], + "score": 1.0, + "content": "their associated pathologies, and more costly gradient descent on the true generator loss.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31.5 + }, + { + "type": "title", + "bbox": [ + 107, + 605, + 227, + 616 + ], + "lines": [ + { + "bbox": [ + 106, + 605, + 228, + 618 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 228, + 618 + ], + "score": 1.0, + "content": "2.3 PARAMETER UPDATES", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 106, + 625, + 425, + 637 + ], + "lines": [ + { + "bbox": [ + 105, + 623, + 426, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 623, + 426, + 640 + ], + "score": 1.0, + "content": "The generator and discriminator parameter updates using this surrogate loss are", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "interline_equation", + "bbox": [ + 247, + 641, + 363, + 697 + ], + "lines": [ + { + "bbox": [ + 247, + 641, + 363, + 697 + ], + "spans": [ + { + "bbox": [ + 247, + 641, + 363, + 697 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { \\theta _ { G } \\theta _ { G } - \\eta \\frac { \\mathrm { d } f _ { K } ( \\theta _ { G } , \\theta _ { D } ) } { \\mathrm { d } \\theta _ { G } } } \\\\ & { \\theta _ { D } \\theta _ { D } + \\eta \\frac { \\mathrm { d } f ( \\theta _ { G } , \\theta _ { D } ) } { \\mathrm { d } \\theta _ { D } } . } \\end{array}", + "type": "interline_equation", + "image_path": "35895e40c05db7fa3394ccde440178bbdbe1633ad2e976da444caa1f30f35303.jpg" + } + ] + } + ], + "index": 36.5, + "virtual_lines": [ + { + "bbox": [ + 247, + 641, + 363, + 669.0 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 247, + 669.0, + 363, + 697.0 + ], + "spans": [], + "index": 37 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 416, + 712 + ], + "score": 1.0, + "content": "For clarity we use full batch steepest gradient descent (ascent) with stepsize", + "type": "text" + }, + { + "bbox": [ + 417, + 701, + 424, + 710 + ], + "score": 0.79, + "content": "\\eta", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "above, while in ex-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "periments we instead use minibatch Adam for both updates. The gradient in Eq. 10 requires back-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 721, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 505, + 734 + ], + "score": 1.0, + "content": "propagating through the optimization process in Eq. 7. A clear description of differentiation through", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 335, + 95 + ], + "score": 1.0, + "content": "For the minimax loss in Eq. 4, the optimal discriminator", + "type": "text" + }, + { + "bbox": [ + 335, + 82, + 364, + 95 + ], + "score": 0.92, + "content": "D ^ { \\ast } \\left( x \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "is a known smooth function of the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 328, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 192, + 106 + ], + "score": 1.0, + "content": "generator probability", + "type": "text" + }, + { + "bbox": [ + 193, + 94, + 221, + 106 + ], + "score": 0.92, + "content": "p _ { G } \\left( x \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 93, + 328, + 106 + ], + "score": 1.0, + "content": "(Goodfellow et al., 2014),", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 105, + 82, + 505, + 106 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 243, + 109, + 368, + 137 + ], + "lines": [ + { + "bbox": [ + 243, + 109, + 368, + 137 + ], + "spans": [ + { + "bbox": [ + 243, + 109, + 368, + 137 + ], + "score": 0.95, + "content": "D ^ { * } \\left( x \\right) = \\frac { p _ { d a t a } \\left( x \\right) } { p _ { d a t a } \\left( x \\right) + p _ { G } \\left( x \\right) } .", + "type": "interline_equation", + "image_path": "0e2376c104ce7eecf400d8575cb56227acb0e45882302fbde8854a5f2d2b2c6b.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 243, + 109, + 368, + 137 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 141, + 505, + 197 + ], + "lines": [ + { + "bbox": [ + 106, + 141, + 504, + 155 + ], + "spans": [ + { + "bbox": [ + 106, + 141, + 372, + 155 + ], + "score": 1.0, + "content": "When the generator loss in Eq. 2 is rewritten directly in terms of", + "type": "text" + }, + { + "bbox": [ + 372, + 142, + 400, + 154 + ], + "score": 0.92, + "content": "p _ { G } \\left( x \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 141, + 491, + 155 + ], + "score": 1.0, + "content": "and Eq. 5 rather than", + "type": "text" + }, + { + "bbox": [ + 491, + 142, + 504, + 153 + ], + "score": 0.87, + "content": "\\theta _ { G }", + "type": "inline_equation" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 152, + 506, + 166 + ], + "spans": [ + { + "bbox": [ + 105, + 152, + 124, + 166 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 153, + 158, + 165 + ], + "score": 0.75, + "content": "\\theta _ { D } ^ { * } \\left( \\theta _ { G } ^ { - } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 152, + 331, + 166 + ], + "score": 1.0, + "content": ", then it is similarly a smooth function of", + "type": "text" + }, + { + "bbox": [ + 331, + 153, + 359, + 165 + ], + "score": 0.91, + "content": "p _ { G } \\left( x \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 152, + 506, + 166 + ], + "score": 1.0, + "content": ". These smoothness guarantees are", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 163, + 506, + 177 + ], + "spans": [ + { + "bbox": [ + 105, + 163, + 185, + 177 + ], + "score": 1.0, + "content": "typically lost when", + "type": "text" + }, + { + "bbox": [ + 185, + 164, + 226, + 176 + ], + "score": 0.93, + "content": "D \\left( x ; \\theta _ { D } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 163, + 245, + 177 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 245, + 164, + 284, + 176 + ], + "score": 0.93, + "content": "G \\left( z ; \\theta _ { G } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 163, + 506, + 177 + ], + "score": 1.0, + "content": "are drawn from parametric families. They nonetheless", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 174, + 506, + 188 + ], + "spans": [ + { + "bbox": [ + 105, + 174, + 506, + 188 + ], + "score": 1.0, + "content": "suggest that the true generator objective in Eq. 2 will often be well behaved, and is a desirable target", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 185, + 200, + 199 + ], + "spans": [ + { + "bbox": [ + 106, + 185, + 200, + 199 + ], + "score": 1.0, + "content": "for direct optimization.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 141, + 506, + 199 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 202, + 505, + 247 + ], + "lines": [ + { + "bbox": [ + 105, + 203, + 505, + 216 + ], + "spans": [ + { + "bbox": [ + 105, + 203, + 351, + 216 + ], + "score": 1.0, + "content": "Explicitly solving for the optimal discriminator parameters", + "type": "text" + }, + { + "bbox": [ + 352, + 203, + 385, + 215 + ], + "score": 0.82, + "content": "\\theta _ { D } ^ { * } \\left( \\theta _ { G } \\right) ", + "type": "inline_equation" + }, + { + "bbox": [ + 386, + 203, + 505, + 216 + ], + "score": 1.0, + "content": "for every update step of the", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 213, + 505, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 147, + 226 + ], + "score": 1.0, + "content": "generator", + "type": "text" + }, + { + "bbox": [ + 147, + 214, + 156, + 223 + ], + "score": 0.83, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 213, + 505, + 226 + ], + "score": 1.0, + "content": "is computationally infeasible for discriminators based on neural networks. Therefore", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 225, + 505, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 225, + 473, + 237 + ], + "score": 1.0, + "content": "this minimax optimization problem is typically solved by alternating gradient descent on", + "type": "text" + }, + { + "bbox": [ + 473, + 225, + 486, + 236 + ], + "score": 0.88, + "content": "\\theta _ { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 225, + 505, + 237 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 236, + 163, + 247 + ], + "spans": [ + { + "bbox": [ + 105, + 236, + 146, + 247 + ], + "score": 1.0, + "content": "ascent on", + "type": "text" + }, + { + "bbox": [ + 146, + 236, + 159, + 246 + ], + "score": 0.88, + "content": "\\theta _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 236, + 163, + 247 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5, + "bbox_fs": [ + 105, + 203, + 505, + 247 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 252, + 505, + 341 + ], + "lines": [ + { + "bbox": [ + 105, + 252, + 506, + 266 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 192, + 266 + ], + "score": 1.0, + "content": "The optimal solution", + "type": "text" + }, + { + "bbox": [ + 192, + 252, + 254, + 264 + ], + "score": 0.92, + "content": "\\theta ^ { * } = \\{ \\theta _ { G } ^ { * } , \\theta _ { D } ^ { * } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 252, + 506, + 266 + ], + "score": 1.0, + "content": "is a fixed point of these iterative learning dynamics. Addition-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 134, + 276 + ], + "score": 1.0, + "content": "ally, if", + "type": "text" + }, + { + "bbox": [ + 135, + 264, + 178, + 276 + ], + "score": 0.93, + "content": "f \\left( { \\theta } _ { G } , { \\theta } _ { D } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 263, + 228, + 276 + ], + "score": 1.0, + "content": "is convex in", + "type": "text" + }, + { + "bbox": [ + 228, + 265, + 241, + 275 + ], + "score": 0.88, + "content": "\\theta _ { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 263, + 303, + 276 + ], + "score": 1.0, + "content": "and concave in", + "type": "text" + }, + { + "bbox": [ + 303, + 264, + 316, + 275 + ], + "score": 0.88, + "content": "\\theta _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 263, + 505, + 276 + ], + "score": 1.0, + "content": ", then alternating gradient descent (ascent) trust", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 104, + 274, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 104, + 274, + 505, + 288 + ], + "score": 1.0, + "content": "region updates are guaranteed to converge to the fixed point, under certain additional weak assump-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 285, + 504, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 286, + 303, + 298 + ], + "score": 1.0, + "content": "tions (Juditsky et al., 2011). However in practice", + "type": "text" + }, + { + "bbox": [ + 303, + 285, + 347, + 298 + ], + "score": 0.93, + "content": "f \\left( { \\theta } _ { G } , { \\theta } _ { D } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 286, + 491, + 298 + ], + "score": 1.0, + "content": "is typically very far from convex in", + "type": "text" + }, + { + "bbox": [ + 491, + 286, + 504, + 297 + ], + "score": 0.88, + "content": "\\theta _ { G }", + "type": "inline_equation" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 295, + 506, + 310 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 168, + 310 + ], + "score": 1.0, + "content": "and concave in", + "type": "text" + }, + { + "bbox": [ + 168, + 297, + 181, + 307 + ], + "score": 0.89, + "content": "\\theta _ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 295, + 506, + 310 + ], + "score": 1.0, + "content": ", and updates are not constrained in an appropriate way. As a result GAN training", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 307, + 505, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 307, + 505, + 320 + ], + "score": 1.0, + "content": "suffers from mode collapse, undamped oscillations, and other problems detailed in Section 1.1. In", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 318, + 506, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 439, + 331 + ], + "score": 1.0, + "content": "order to address these difficulties, we will introduce a surrogate objective function", + "type": "text" + }, + { + "bbox": [ + 439, + 318, + 489, + 330 + ], + "score": 0.91, + "content": "f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 318, + 506, + 331 + ], + "score": 1.0, + "content": "for", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 329, + 498, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 429, + 343 + ], + "score": 1.0, + "content": "training the generator which more closely resembles the true generator objective", + "type": "text" + }, + { + "bbox": [ + 429, + 330, + 493, + 342 + ], + "score": 0.9, + "content": "f \\left( \\theta _ { G } , \\theta _ { D } ^ { * } \\left( \\theta _ { G } \\right) \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 329, + 498, + 343 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 15.5, + "bbox_fs": [ + 104, + 252, + 506, + 343 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 353, + 214, + 365 + ], + "lines": [ + { + "bbox": [ + 104, + 352, + 216, + 367 + ], + "spans": [ + { + "bbox": [ + 104, + 352, + 216, + 367 + ], + "score": 1.0, + "content": "2.2 UNROLLING GANS", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 106, + 374, + 505, + 397 + ], + "lines": [ + { + "bbox": [ + 105, + 373, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 314, + 388 + ], + "score": 1.0, + "content": "A local optimum of the discriminator parameters", + "type": "text" + }, + { + "bbox": [ + 314, + 375, + 327, + 387 + ], + "score": 0.89, + "content": "\\theta _ { D } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 373, + 505, + 388 + ], + "score": 1.0, + "content": "can be expressed as the fixed point of an", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 386, + 239, + 398 + ], + "spans": [ + { + "bbox": [ + 106, + 386, + 239, + 398 + ], + "score": 1.0, + "content": "iterative optimization procedure,", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 373, + 505, + 398 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 237, + 399, + 373, + 468 + ], + "lines": [ + { + "bbox": [ + 237, + 399, + 373, + 468 + ], + "spans": [ + { + "bbox": [ + 237, + 399, + 373, + 468 + ], + "score": 0.93, + "content": "\\begin{array} { c } { { \\theta _ { D } ^ { 0 } = \\theta _ { D } } } \\\\ { { \\displaystyle } } \\\\ { { \\theta _ { D } ^ { k + 1 } = \\theta _ { D } ^ { k } + \\eta ^ { k } \\displaystyle \\frac { \\mathrm { d } f ( \\theta _ { G } , \\theta _ { D } ^ { k } ) } { \\mathrm { d } \\theta _ { D } ^ { k } } } } \\\\ { { \\displaystyle } } \\\\ { { \\theta _ { D } ^ { * } ( \\theta _ { G } ) = \\displaystyle \\operatorname* { l i m } _ { k \\infty } \\theta _ { D } ^ { k } , } } \\end{array}", + "type": "interline_equation", + "image_path": "f45244ba45a91984f7cab7a13e843b9f8428ca6f8c95d11c332b9e18669248b2.jpg" + } + ] + } + ], + "index": 23.5, + "virtual_lines": [ + { + "bbox": [ + 237, + 399, + 373, + 433.5 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 237, + 433.5, + 373, + 468.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 473, + 504, + 506 + ], + "lines": [ + { + "bbox": [ + 105, + 471, + 506, + 486 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 133, + 486 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 472, + 145, + 484 + ], + "score": 0.9, + "content": "\\eta ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 471, + 506, + 486 + ], + "score": 1.0, + "content": "is the learning rate schedule. For clarity, we have expressed Eq. 7 as a full batch steepest", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 483, + 506, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 506, + 496 + ], + "score": 1.0, + "content": "gradient ascent equation. More sophisticated optimizers can be similarly unrolled. In our experi-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 495, + 292, + 507 + ], + "spans": [ + { + "bbox": [ + 106, + 495, + 292, + 507 + ], + "score": 1.0, + "content": "ments we unroll Adam (Kingma & Ba, 2014).", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 471, + 506, + 507 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 511, + 460, + 523 + ], + "lines": [ + { + "bbox": [ + 105, + 510, + 462, + 526 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 174, + 526 + ], + "score": 1.0, + "content": "By unrolling for", + "type": "text" + }, + { + "bbox": [ + 174, + 512, + 184, + 522 + ], + "score": 0.83, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 510, + 462, + 526 + ], + "score": 1.0, + "content": "steps, we create a surrogate objective for the update of the generator,", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 510, + 462, + 526 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 229, + 527, + 381, + 543 + ], + "lines": [ + { + "bbox": [ + 229, + 527, + 381, + 543 + ], + "spans": [ + { + "bbox": [ + 229, + 527, + 381, + 543 + ], + "score": 0.93, + "content": "f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) = f \\left( \\theta _ { G } , \\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) \\right) .", + "type": "interline_equation", + "image_path": "722de6d2f0bca528a8df9e9213bb3bcba721cff99ce6b981b4c7ab2eda98fd02.jpg" + } + ] + } + ], + "index": 29, + "virtual_lines": [ + { + "bbox": [ + 229, + 527, + 381, + 543 + ], + "spans": [], + "index": 29 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 547, + 504, + 592 + ], + "lines": [ + { + "bbox": [ + 105, + 546, + 504, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 133, + 560 + ], + "score": 1.0, + "content": "When", + "type": "text" + }, + { + "bbox": [ + 133, + 547, + 163, + 558 + ], + "score": 0.89, + "content": "K = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 546, + 466, + 560 + ], + "score": 1.0, + "content": "this objective corresponds exactly to the standard GAN objective, while as", + "type": "text" + }, + { + "bbox": [ + 467, + 547, + 504, + 558 + ], + "score": 0.88, + "content": "K \\infty", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 558, + 506, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 558, + 325, + 572 + ], + "score": 1.0, + "content": "it corresponds to the true generator objective function", + "type": "text" + }, + { + "bbox": [ + 325, + 558, + 387, + 570 + ], + "score": 0.92, + "content": "f \\left( \\theta _ { G } , \\theta _ { D } ^ { * } \\left( G \\right) \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 558, + 506, + 572 + ], + "score": 1.0, + "content": ". By adjusting the number of", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 570, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 169, + 582 + ], + "score": 1.0, + "content": "unrolling steps", + "type": "text" + }, + { + "bbox": [ + 169, + 570, + 180, + 579 + ], + "score": 0.81, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 570, + 505, + 582 + ], + "score": 1.0, + "content": ", we are thus able to interpolate between standard GAN training dynamics with", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 580, + 461, + 594 + ], + "spans": [ + { + "bbox": [ + 105, + 580, + 461, + 594 + ], + "score": 1.0, + "content": "their associated pathologies, and more costly gradient descent on the true generator loss.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 546, + 506, + 594 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 605, + 227, + 616 + ], + "lines": [ + { + "bbox": [ + 106, + 605, + 228, + 618 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 228, + 618 + ], + "score": 1.0, + "content": "2.3 PARAMETER UPDATES", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 106, + 625, + 425, + 637 + ], + "lines": [ + { + "bbox": [ + 105, + 623, + 426, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 623, + 426, + 640 + ], + "score": 1.0, + "content": "The generator and discriminator parameter updates using this surrogate loss are", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35, + "bbox_fs": [ + 105, + 623, + 426, + 640 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 247, + 641, + 363, + 697 + ], + "lines": [ + { + "bbox": [ + 247, + 641, + 363, + 697 + ], + "spans": [ + { + "bbox": [ + 247, + 641, + 363, + 697 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { \\theta _ { G } \\theta _ { G } - \\eta \\frac { \\mathrm { d } f _ { K } ( \\theta _ { G } , \\theta _ { D } ) } { \\mathrm { d } \\theta _ { G } } } \\\\ & { \\theta _ { D } \\theta _ { D } + \\eta \\frac { \\mathrm { d } f ( \\theta _ { G } , \\theta _ { D } ) } { \\mathrm { d } \\theta _ { D } } . } \\end{array}", + "type": "interline_equation", + "image_path": "35895e40c05db7fa3394ccde440178bbdbe1633ad2e976da444caa1f30f35303.jpg" + } + ] + } + ], + "index": 36.5, + "virtual_lines": [ + { + "bbox": [ + 247, + 641, + 363, + 669.0 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 247, + 669.0, + 363, + 697.0 + ], + "spans": [], + "index": 37 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 416, + 712 + ], + "score": 1.0, + "content": "For clarity we use full batch steepest gradient descent (ascent) with stepsize", + "type": "text" + }, + { + "bbox": [ + 417, + 701, + 424, + 710 + ], + "score": 0.79, + "content": "\\eta", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 698, + 505, + 712 + ], + "score": 1.0, + "content": "above, while in ex-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "periments we instead use minibatch Adam for both updates. The gradient in Eq. 10 requires back-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 721, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 505, + 734 + ], + "score": 1.0, + "content": "propagating through the optimization process in Eq. 7. A clear description of differentiation through", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 312, + 506, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 312, + 506, + 325 + ], + "score": 1.0, + "content": "gradient descent is given as Algorithm 2 in (Maclaurin et al., 2015), though in practice the use of", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 323, + 505, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 323, + 505, + 336 + ], + "score": 1.0, + "content": "an automatic differentiation package means this step does not need to be programmed explicitly. A", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 335, + 364, + 347 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 364, + 347 + ], + "score": 1.0, + "content": "pictorial representation of these updates is provided in Figure 1.", + "type": "text", + "cross_page": true + } + ], + "index": 10 + } + ], + "index": 39, + "bbox_fs": [ + 105, + 698, + 505, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 111, + 98, + 503, + 205 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 98, + 503, + 205 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 111, + 98, + 503, + 205 + ], + "spans": [ + { + "bbox": [ + 111, + 98, + 503, + 205 + ], + "score": 0.96, + "type": "image", + "image_path": "b8d41d44f95393092b08110ba25482b02bb1bcb0a2930f383731e4e370791ca3.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 111, + 98, + 503, + 133.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 111, + 133.66666666666666, + 503, + 169.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 111, + 169.33333333333331, + 503, + 204.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 231, + 505, + 287 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 230, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 505, + 244 + ], + "score": 1.0, + "content": "Figure 1: An illustration of the computation graph for an unrolled GAN with 3 unrolling steps.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 241, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 505, + 255 + ], + "score": 1.0, + "content": "The generator update in Equation 10 involves backpropagating the generator gradient (blue arrows)", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 253, + 504, + 266 + ], + "spans": [ + { + "bbox": [ + 105, + 253, + 287, + 266 + ], + "score": 1.0, + "content": "through the unrolled optimization. Each step", + "type": "text" + }, + { + "bbox": [ + 287, + 254, + 294, + 263 + ], + "score": 0.76, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 253, + 493, + 266 + ], + "score": 1.0, + "content": "in the unrolled optimization uses the gradients of", + "type": "text" + }, + { + "bbox": [ + 493, + 254, + 504, + 265 + ], + "score": 0.87, + "content": "f _ { k }", + "type": "inline_equation" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 261, + 507, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 167, + 279 + ], + "score": 1.0, + "content": "with respect to", + "type": "text" + }, + { + "bbox": [ + 168, + 263, + 180, + 276 + ], + "score": 0.9, + "content": "\\theta _ { D } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 261, + 507, + 279 + ], + "score": 1.0, + "content": ", as described in Equation 7 and indicated by the green arrows. The discriminator", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 275, + 429, + 288 + ], + "spans": [ + { + "bbox": [ + 106, + 275, + 429, + 288 + ], + "score": 1.0, + "content": "update in Equation 11 does not depend on the unrolled optimization (red arrow).", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + } + ], + "index": 3.0 + }, + { + "type": "text", + "bbox": [ + 107, + 312, + 505, + 346 + ], + "lines": [ + { + "bbox": [ + 105, + 312, + 506, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 312, + 506, + 325 + ], + "score": 1.0, + "content": "gradient descent is given as Algorithm 2 in (Maclaurin et al., 2015), though in practice the use of", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 323, + 505, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 323, + 505, + 336 + ], + "score": 1.0, + "content": "an automatic differentiation package means this step does not need to be programmed explicitly. A", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 335, + 364, + 347 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 364, + 347 + ], + "score": 1.0, + "content": "pictorial representation of these updates is provided in Figure 1.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 106, + 351, + 505, + 418 + ], + "lines": [ + { + "bbox": [ + 106, + 352, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 505, + 363 + ], + "score": 1.0, + "content": "It is important to distinguish this from an approach suggested in (Goodfellow et al., 2014), that", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 362, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 505, + 375 + ], + "score": 1.0, + "content": "several update steps of the discriminator parameters should be run before each single update step for", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 374, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 374, + 505, + 385 + ], + "score": 1.0, + "content": "the generator. In that approach, the update steps for both models are still gradient descent (ascent)", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 385, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 385, + 505, + 396 + ], + "score": 1.0, + "content": "with respect to fixed values of the other model parameters, rather than the surrogate loss we describe", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 395, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 192, + 408 + ], + "score": 1.0, + "content": "in Eq. 9. Performing", + "type": "text" + }, + { + "bbox": [ + 193, + 396, + 203, + 406 + ], + "score": 0.78, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 395, + 505, + 408 + ], + "score": 1.0, + "content": "steps of discriminator update between each single step of generator update", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 406, + 485, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 306, + 419 + ], + "score": 1.0, + "content": "corresponds to updating the generator parameters", + "type": "text" + }, + { + "bbox": [ + 306, + 406, + 318, + 417 + ], + "score": 0.89, + "content": "\\theta _ { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 406, + 485, + 419 + ], + "score": 1.0, + "content": "using only the first term in Eq. 12 below.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 13.5 + }, + { + "type": "title", + "bbox": [ + 108, + 437, + 267, + 448 + ], + "lines": [ + { + "bbox": [ + 106, + 437, + 268, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 437, + 268, + 450 + ], + "score": 1.0, + "content": "2.4 THE MISSING GRADIENT TERM", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 106, + 460, + 504, + 483 + ], + "lines": [ + { + "bbox": [ + 105, + 459, + 506, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 330, + 474 + ], + "score": 1.0, + "content": "To better understand the behavior of the surrogate loss", + "type": "text" + }, + { + "bbox": [ + 330, + 460, + 381, + 473 + ], + "score": 0.92, + "content": "f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 459, + 506, + 474 + ], + "score": 1.0, + "content": ", we examine its gradient with", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 471, + 265, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 471, + 248, + 484 + ], + "score": 1.0, + "content": "respect to the generator parameters", + "type": "text" + }, + { + "bbox": [ + 249, + 472, + 261, + 482 + ], + "score": 0.89, + "content": "\\theta _ { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 471, + 265, + 484 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "interline_equation", + "bbox": [ + 142, + 493, + 469, + 523 + ], + "lines": [ + { + "bbox": [ + 142, + 493, + 469, + 523 + ], + "spans": [ + { + "bbox": [ + 142, + 493, + 469, + 523 + ], + "score": 0.92, + "content": "\\frac { \\mathrm { d } f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) } { \\mathrm { d } \\theta _ { G } } = \\frac { \\partial f \\left( \\theta _ { G } , \\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) \\right) } { \\partial \\theta _ { G } } + \\frac { \\partial f \\left( \\theta _ { G } , \\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) \\right) } { \\partial \\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) } \\frac { \\mathrm { d } \\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) } { \\mathrm { d } \\theta _ { G } } .", + "type": "interline_equation", + "image_path": "791d5164930dd3fb042c3335cdcf143313e5660a7175699577db372f61561f22.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 142, + 493, + 469, + 503.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 142, + 503.0, + 469, + 513.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 142, + 513.0, + 469, + 523.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 538, + 505, + 594 + ], + "lines": [ + { + "bbox": [ + 106, + 539, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 505, + 551 + ], + "score": 1.0, + "content": "Standard GAN training corresponds exactly to updating the generator parameters using only the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 549, + 506, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 234, + 564 + ], + "score": 1.0, + "content": "first term in this gradient, with", + "type": "text" + }, + { + "bbox": [ + 234, + 549, + 285, + 562 + ], + "score": 0.92, + "content": "\\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) ", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 549, + 506, + 564 + ], + "score": 1.0, + "content": "being the parameters resulting from the discriminator", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 561, + 505, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 446, + 573 + ], + "score": 1.0, + "content": "update step. An optimal generator for any fixed discriminator is a delta function at the", + "type": "text" + }, + { + "bbox": [ + 446, + 563, + 453, + 571 + ], + "score": 0.77, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 561, + 505, + 573 + ], + "score": 1.0, + "content": "to which the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 572, + 505, + 584 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 505, + 584 + ], + "score": 1.0, + "content": "discriminator assigns highest data probability. Therefore, in standard GAN training, each generator", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 583, + 334, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 334, + 595 + ], + "score": 1.0, + "content": "update step is a partial collapse towards a delta function.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 106, + 599, + 505, + 655 + ], + "lines": [ + { + "bbox": [ + 106, + 599, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 505, + 612 + ], + "score": 1.0, + "content": "The second term captures how the discriminator would react to a change in the generator. It reduces", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 610, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 505, + 623 + ], + "score": 1.0, + "content": "the tendency of the generator to engage in mode collapse. For instance, the second term reflects", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 621, + 506, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 506, + 635 + ], + "score": 1.0, + "content": "that as the generator collapses towards a delta function, the discriminator reacts and assigns lower", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 632, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 505, + 646 + ], + "score": 1.0, + "content": "probability to that state, increasing the generator loss. It therefore discourages the generator from", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 643, + 261, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 261, + 657 + ], + "score": 1.0, + "content": "collapsing, and may improve stability.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 106, + 659, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 658, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 120, + 680 + ], + "score": 1.0, + "content": "As", + "type": "text" + }, + { + "bbox": [ + 121, + 662, + 156, + 673 + ], + "score": 0.86, + "content": "K \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 658, + 160, + 680 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 161, + 661, + 175, + 675 + ], + "score": 0.87, + "content": "\\theta _ { D } ^ { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 658, + 289, + 680 + ], + "score": 1.0, + "content": "goes to a local optimum of", + "type": "text" + }, + { + "bbox": [ + 289, + 662, + 296, + 674 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 658, + 328, + 680 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 329, + 660, + 367, + 678 + ], + "score": 0.94, + "content": "\\frac { \\partial f } { \\partial \\theta _ { D } ^ { K } } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 658, + 506, + 680 + ], + "score": 1.0, + "content": ", and therefore the second term in", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 432, + 689 + ], + "score": 1.0, + "content": "Eq. 12 goes to 0 (Danskin, 1967). The gradient of the unrolled surrogate loss", + "type": "text" + }, + { + "bbox": [ + 432, + 676, + 482, + 689 + ], + "score": 0.92, + "content": "f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 676, + 505, + 689 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 687, + 504, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 146, + 700 + ], + "score": 1.0, + "content": "respect to", + "type": "text" + }, + { + "bbox": [ + 147, + 688, + 159, + 699 + ], + "score": 0.89, + "content": "\\theta _ { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 687, + 386, + 700 + ], + "score": 1.0, + "content": "is thus identical to the gradient of the standard GAN loss", + "type": "text" + }, + { + "bbox": [ + 387, + 687, + 430, + 699 + ], + "score": 0.93, + "content": "f \\bar { ( \\theta _ { G } , \\theta _ { D } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 687, + 475, + 700 + ], + "score": 1.0, + "content": "both when", + "type": "text" + }, + { + "bbox": [ + 475, + 688, + 504, + 698 + ], + "score": 0.89, + "content": "K = 0", + "type": "inline_equation" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 148, + 711 + ], + "score": 1.0, + "content": "and when", + "type": "text" + }, + { + "bbox": [ + 149, + 699, + 186, + 709 + ], + "score": 0.9, + "content": "K \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 699, + 252, + 711 + ], + "score": 1.0, + "content": ", where we take", + "type": "text" + }, + { + "bbox": [ + 253, + 700, + 290, + 709 + ], + "score": 0.89, + "content": "K \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "to imply that in the standard GAN the discriminator", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 708, + 504, + 723 + ], + "spans": [ + { + "bbox": [ + 104, + 708, + 453, + 723 + ], + "score": 1.0, + "content": "is also fully optimized between each generator update. Between these two extremes,", + "type": "text" + }, + { + "bbox": [ + 453, + 709, + 504, + 722 + ], + "score": 0.92, + "content": "f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right)", + "type": "inline_equation" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 720, + 504, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 504, + 734 + ], + "score": 1.0, + "content": "captures additional information about the response of the discriminator to changes in the generator.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 35.5 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 8 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 111, + 98, + 503, + 205 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 98, + 503, + 205 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 111, + 98, + 503, + 205 + ], + "spans": [ + { + "bbox": [ + 111, + 98, + 503, + 205 + ], + "score": 0.96, + "type": "image", + "image_path": "b8d41d44f95393092b08110ba25482b02bb1bcb0a2930f383731e4e370791ca3.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 111, + 98, + 503, + 133.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 111, + 133.66666666666666, + 503, + 169.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 111, + 169.33333333333331, + 503, + 204.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 231, + 505, + 287 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 230, + 505, + 244 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 505, + 244 + ], + "score": 1.0, + "content": "Figure 1: An illustration of the computation graph for an unrolled GAN with 3 unrolling steps.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 241, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 241, + 505, + 255 + ], + "score": 1.0, + "content": "The generator update in Equation 10 involves backpropagating the generator gradient (blue arrows)", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 253, + 504, + 266 + ], + "spans": [ + { + "bbox": [ + 105, + 253, + 287, + 266 + ], + "score": 1.0, + "content": "through the unrolled optimization. Each step", + "type": "text" + }, + { + "bbox": [ + 287, + 254, + 294, + 263 + ], + "score": 0.76, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 253, + 493, + 266 + ], + "score": 1.0, + "content": "in the unrolled optimization uses the gradients of", + "type": "text" + }, + { + "bbox": [ + 493, + 254, + 504, + 265 + ], + "score": 0.87, + "content": "f _ { k }", + "type": "inline_equation" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 261, + 507, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 261, + 167, + 279 + ], + "score": 1.0, + "content": "with respect to", + "type": "text" + }, + { + "bbox": [ + 168, + 263, + 180, + 276 + ], + "score": 0.9, + "content": "\\theta _ { D } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 261, + 507, + 279 + ], + "score": 1.0, + "content": ", as described in Equation 7 and indicated by the green arrows. The discriminator", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 275, + 429, + 288 + ], + "spans": [ + { + "bbox": [ + 106, + 275, + 429, + 288 + ], + "score": 1.0, + "content": "update in Equation 11 does not depend on the unrolled optimization (red arrow).", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + } + ], + "index": 3.0 + }, + { + "type": "text", + "bbox": [ + 107, + 312, + 505, + 346 + ], + "lines": [], + "index": 9, + "bbox_fs": [ + 105, + 312, + 506, + 347 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 351, + 505, + 418 + ], + "lines": [ + { + "bbox": [ + 106, + 352, + 505, + 363 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 505, + 363 + ], + "score": 1.0, + "content": "It is important to distinguish this from an approach suggested in (Goodfellow et al., 2014), that", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 362, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 505, + 375 + ], + "score": 1.0, + "content": "several update steps of the discriminator parameters should be run before each single update step for", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 374, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 374, + 505, + 385 + ], + "score": 1.0, + "content": "the generator. In that approach, the update steps for both models are still gradient descent (ascent)", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 385, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 385, + 505, + 396 + ], + "score": 1.0, + "content": "with respect to fixed values of the other model parameters, rather than the surrogate loss we describe", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 395, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 192, + 408 + ], + "score": 1.0, + "content": "in Eq. 9. Performing", + "type": "text" + }, + { + "bbox": [ + 193, + 396, + 203, + 406 + ], + "score": 0.78, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 395, + 505, + 408 + ], + "score": 1.0, + "content": "steps of discriminator update between each single step of generator update", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 406, + 485, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 306, + 419 + ], + "score": 1.0, + "content": "corresponds to updating the generator parameters", + "type": "text" + }, + { + "bbox": [ + 306, + 406, + 318, + 417 + ], + "score": 0.89, + "content": "\\theta _ { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 406, + 485, + 419 + ], + "score": 1.0, + "content": "using only the first term in Eq. 12 below.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 13.5, + "bbox_fs": [ + 105, + 352, + 505, + 419 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 437, + 267, + 448 + ], + "lines": [ + { + "bbox": [ + 106, + 437, + 268, + 450 + ], + "spans": [ + { + "bbox": [ + 106, + 437, + 268, + 450 + ], + "score": 1.0, + "content": "2.4 THE MISSING GRADIENT TERM", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 106, + 460, + 504, + 483 + ], + "lines": [ + { + "bbox": [ + 105, + 459, + 506, + 474 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 330, + 474 + ], + "score": 1.0, + "content": "To better understand the behavior of the surrogate loss", + "type": "text" + }, + { + "bbox": [ + 330, + 460, + 381, + 473 + ], + "score": 0.92, + "content": "f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 459, + 506, + 474 + ], + "score": 1.0, + "content": ", we examine its gradient with", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 471, + 265, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 471, + 248, + 484 + ], + "score": 1.0, + "content": "respect to the generator parameters", + "type": "text" + }, + { + "bbox": [ + 249, + 472, + 261, + 482 + ], + "score": 0.89, + "content": "\\theta _ { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 471, + 265, + 484 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 459, + 506, + 484 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 142, + 493, + 469, + 523 + ], + "lines": [ + { + "bbox": [ + 142, + 493, + 469, + 523 + ], + "spans": [ + { + "bbox": [ + 142, + 493, + 469, + 523 + ], + "score": 0.92, + "content": "\\frac { \\mathrm { d } f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) } { \\mathrm { d } \\theta _ { G } } = \\frac { \\partial f \\left( \\theta _ { G } , \\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) \\right) } { \\partial \\theta _ { G } } + \\frac { \\partial f \\left( \\theta _ { G } , \\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) \\right) } { \\partial \\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) } \\frac { \\mathrm { d } \\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) } { \\mathrm { d } \\theta _ { G } } .", + "type": "interline_equation", + "image_path": "791d5164930dd3fb042c3335cdcf143313e5660a7175699577db372f61561f22.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 142, + 493, + 469, + 503.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 142, + 503.0, + 469, + 513.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 142, + 513.0, + 469, + 523.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 538, + 505, + 594 + ], + "lines": [ + { + "bbox": [ + 106, + 539, + 505, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 505, + 551 + ], + "score": 1.0, + "content": "Standard GAN training corresponds exactly to updating the generator parameters using only the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 549, + 506, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 234, + 564 + ], + "score": 1.0, + "content": "first term in this gradient, with", + "type": "text" + }, + { + "bbox": [ + 234, + 549, + 285, + 562 + ], + "score": 0.92, + "content": "\\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) ", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 549, + 506, + 564 + ], + "score": 1.0, + "content": "being the parameters resulting from the discriminator", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 561, + 505, + 573 + ], + "spans": [ + { + "bbox": [ + 106, + 561, + 446, + 573 + ], + "score": 1.0, + "content": "update step. An optimal generator for any fixed discriminator is a delta function at the", + "type": "text" + }, + { + "bbox": [ + 446, + 563, + 453, + 571 + ], + "score": 0.77, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 561, + 505, + 573 + ], + "score": 1.0, + "content": "to which the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 572, + 505, + 584 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 505, + 584 + ], + "score": 1.0, + "content": "discriminator assigns highest data probability. Therefore, in standard GAN training, each generator", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 583, + 334, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 334, + 595 + ], + "score": 1.0, + "content": "update step is a partial collapse towards a delta function.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 539, + 506, + 595 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 599, + 505, + 655 + ], + "lines": [ + { + "bbox": [ + 106, + 599, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 505, + 612 + ], + "score": 1.0, + "content": "The second term captures how the discriminator would react to a change in the generator. It reduces", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 610, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 505, + 623 + ], + "score": 1.0, + "content": "the tendency of the generator to engage in mode collapse. For instance, the second term reflects", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 621, + 506, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 506, + 635 + ], + "score": 1.0, + "content": "that as the generator collapses towards a delta function, the discriminator reacts and assigns lower", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 632, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 505, + 646 + ], + "score": 1.0, + "content": "probability to that state, increasing the generator loss. It therefore discourages the generator from", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 643, + 261, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 261, + 657 + ], + "score": 1.0, + "content": "collapsing, and may improve stability.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30, + "bbox_fs": [ + 105, + 599, + 506, + 657 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 659, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 658, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 658, + 120, + 680 + ], + "score": 1.0, + "content": "As", + "type": "text" + }, + { + "bbox": [ + 121, + 662, + 156, + 673 + ], + "score": 0.86, + "content": "K \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 658, + 160, + 680 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 161, + 661, + 175, + 675 + ], + "score": 0.87, + "content": "\\theta _ { D } ^ { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 658, + 289, + 680 + ], + "score": 1.0, + "content": "goes to a local optimum of", + "type": "text" + }, + { + "bbox": [ + 289, + 662, + 296, + 674 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 658, + 328, + 680 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 329, + 660, + 367, + 678 + ], + "score": 0.94, + "content": "\\frac { \\partial f } { \\partial \\theta _ { D } ^ { K } } = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 367, + 658, + 506, + 680 + ], + "score": 1.0, + "content": ", and therefore the second term in", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 432, + 689 + ], + "score": 1.0, + "content": "Eq. 12 goes to 0 (Danskin, 1967). The gradient of the unrolled surrogate loss", + "type": "text" + }, + { + "bbox": [ + 432, + 676, + 482, + 689 + ], + "score": 0.92, + "content": "f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 676, + 505, + 689 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 687, + 504, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 146, + 700 + ], + "score": 1.0, + "content": "respect to", + "type": "text" + }, + { + "bbox": [ + 147, + 688, + 159, + 699 + ], + "score": 0.89, + "content": "\\theta _ { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 687, + 386, + 700 + ], + "score": 1.0, + "content": "is thus identical to the gradient of the standard GAN loss", + "type": "text" + }, + { + "bbox": [ + 387, + 687, + 430, + 699 + ], + "score": 0.93, + "content": "f \\bar { ( \\theta _ { G } , \\theta _ { D } ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 687, + 475, + 700 + ], + "score": 1.0, + "content": "both when", + "type": "text" + }, + { + "bbox": [ + 475, + 688, + 504, + 698 + ], + "score": 0.89, + "content": "K = 0", + "type": "inline_equation" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 148, + 711 + ], + "score": 1.0, + "content": "and when", + "type": "text" + }, + { + "bbox": [ + 149, + 699, + 186, + 709 + ], + "score": 0.9, + "content": "K \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 699, + 252, + 711 + ], + "score": 1.0, + "content": ", where we take", + "type": "text" + }, + { + "bbox": [ + 253, + 700, + 290, + 709 + ], + "score": 0.89, + "content": "K \\infty", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "to imply that in the standard GAN the discriminator", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 708, + 504, + 723 + ], + "spans": [ + { + "bbox": [ + 104, + 708, + 453, + 723 + ], + "score": 1.0, + "content": "is also fully optimized between each generator update. Between these two extremes,", + "type": "text" + }, + { + "bbox": [ + 453, + 709, + 504, + 722 + ], + "score": 0.92, + "content": "f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right)", + "type": "inline_equation" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 720, + 504, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 504, + 734 + ], + "score": 1.0, + "content": "captures additional information about the response of the discriminator to changes in the generator.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 35.5, + "bbox_fs": [ + 104, + 658, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 109, + 82, + 313, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 316, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 316, + 96 + ], + "score": 1.0, + "content": "2.5 CONSEQUENCES OF THE SURROGATE LOSS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 104, + 505, + 152 + ], + "lines": [ + { + "bbox": [ + 105, + 103, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 352, + 117 + ], + "score": 1.0, + "content": "GANs can be thought of as a game between the discriminator", + "type": "text" + }, + { + "bbox": [ + 352, + 105, + 367, + 115 + ], + "score": 0.75, + "content": "( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 103, + 439, + 117 + ], + "score": 1.0, + "content": "and the generator", + "type": "text" + }, + { + "bbox": [ + 440, + 105, + 455, + 115 + ], + "score": 0.69, + "content": "( G )", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 103, + 505, + 117 + ], + "score": 1.0, + "content": ". The agents", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "score": 1.0, + "content": "take turns taking actions and updating their parameters until a Nash equilibrium is reached. The", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 125, + 504, + 144 + ], + "spans": [ + { + "bbox": [ + 106, + 125, + 182, + 144 + ], + "score": 1.0, + "content": "optimal action for", + "type": "text" + }, + { + "bbox": [ + 182, + 128, + 192, + 138 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 125, + 334, + 144 + ], + "score": 1.0, + "content": "is to evaluate the probability ratio", + "type": "text" + }, + { + "bbox": [ + 334, + 127, + 393, + 143 + ], + "score": 0.95, + "content": "\\frac { p _ { d a t a } ( x ) } { p _ { G } ( x ) + p _ { d a t a } ( x ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 125, + 497, + 144 + ], + "score": 1.0, + "content": "for the generator’s move", + "type": "text" + }, + { + "bbox": [ + 497, + 131, + 504, + 138 + ], + "score": 0.67, + "content": "x", + "type": "inline_equation" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 141, + 428, + 153 + ], + "spans": [ + { + "bbox": [ + 106, + 141, + 428, + 153 + ], + "score": 1.0, + "content": "(Eq. 5). The optimal generator action is to move its mass to maximize this ratio.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 107, + 158, + 505, + 246 + ], + "lines": [ + { + "bbox": [ + 106, + 157, + 505, + 170 + ], + "spans": [ + { + "bbox": [ + 106, + 157, + 187, + 170 + ], + "score": 1.0, + "content": "The initial move for", + "type": "text" + }, + { + "bbox": [ + 187, + 159, + 196, + 168 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 157, + 505, + 170 + ], + "score": 1.0, + "content": "will be to move as much mass as its parametric family and update step permits", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 169, + 505, + 181 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 427, + 181 + ], + "score": 1.0, + "content": "to the single point that maximizes the ratio of probability densities. The action", + "type": "text" + }, + { + "bbox": [ + 427, + 169, + 437, + 179 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 169, + 505, + 181 + ], + "score": 1.0, + "content": "will then take is", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 180, + 506, + 193 + ], + "spans": [ + { + "bbox": [ + 105, + 180, + 506, + 193 + ], + "score": 1.0, + "content": "quite simple. It will track that point, and to the extent allowed by its own parametric family and", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 191, + 505, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 505, + 203 + ], + "score": 1.0, + "content": "update step assign low data probability to it, and uniform probability everywhere else. This cycle", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 202, + 506, + 215 + ], + "spans": [ + { + "bbox": [ + 105, + 202, + 118, + 215 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 203, + 127, + 212 + ], + "score": 0.81, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 202, + 180, + 215 + ], + "score": 1.0, + "content": "moving and", + "type": "text" + }, + { + "bbox": [ + 180, + 203, + 190, + 212 + ], + "score": 0.79, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 202, + 506, + 215 + ], + "score": 1.0, + "content": "following will repeat forever or converge depending on the rate of change of", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 213, + 505, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 505, + 225 + ], + "score": 1.0, + "content": "the two agents. This is similar to the situation in simple matrix games like rock-paper-scissors and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 224, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 224, + 505, + 236 + ], + "score": 1.0, + "content": "matching pennies, where alternating gradient descent (ascent) with a fixed learning rate is known", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 235, + 356, + 247 + ], + "spans": [ + { + "bbox": [ + 105, + 235, + 356, + 247 + ], + "score": 1.0, + "content": "not to converge (Singh et al., 2000; Bowling & Veloso, 2002).", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 107, + 252, + 505, + 296 + ], + "lines": [ + { + "bbox": [ + 105, + 251, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 251, + 421, + 264 + ], + "score": 1.0, + "content": "In the unrolled case, however, this undesirable behavior no longer occurs. Now", + "type": "text" + }, + { + "bbox": [ + 421, + 252, + 430, + 262 + ], + "score": 0.71, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 251, + 505, + 264 + ], + "score": 1.0, + "content": "’s actions take into", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 263, + 505, + 274 + ], + "spans": [ + { + "bbox": [ + 106, + 263, + 160, + 274 + ], + "score": 1.0, + "content": "account how", + "type": "text" + }, + { + "bbox": [ + 161, + 263, + 171, + 273 + ], + "score": 0.78, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 263, + 284, + 274 + ], + "score": 1.0, + "content": "will respond. In particular,", + "type": "text" + }, + { + "bbox": [ + 285, + 263, + 294, + 273 + ], + "score": 0.77, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 263, + 405, + 274 + ], + "score": 1.0, + "content": "will try to make steps that", + "type": "text" + }, + { + "bbox": [ + 405, + 263, + 415, + 273 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 263, + 505, + 274 + ], + "score": 1.0, + "content": "will have a hard time", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 274, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 274, + 475, + 286 + ], + "score": 1.0, + "content": "responding to. This extra information helps the generator spread its mass to make the next", + "type": "text" + }, + { + "bbox": [ + 475, + 274, + 485, + 284 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 274, + 505, + 286 + ], + "score": 1.0, + "content": "step", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 284, + 288, + 297 + ], + "spans": [ + { + "bbox": [ + 105, + 284, + 288, + 297 + ], + "score": 1.0, + "content": "less effective instead of collapsing to a point.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 107, + 301, + 505, + 379 + ], + "lines": [ + { + "bbox": [ + 106, + 301, + 505, + 314 + ], + "spans": [ + { + "bbox": [ + 106, + 301, + 347, + 314 + ], + "score": 1.0, + "content": "In principle, a surrogate loss function could be used for both", + "type": "text" + }, + { + "bbox": [ + 347, + 302, + 357, + 312 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 301, + 374, + 314 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 375, + 302, + 383, + 312 + ], + "score": 0.75, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 301, + 505, + 314 + ], + "score": 1.0, + "content": ". In the case of 1-step unrolled", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 313, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 505, + 325 + ], + "score": 1.0, + "content": "optimization this is known to lead to convergence for games in which gradient descent (ascent) fails", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 324, + 505, + 336 + ], + "spans": [ + { + "bbox": [ + 106, + 324, + 505, + 336 + ], + "score": 1.0, + "content": "(Zhang & Lesser, 2010). However, the motivation for using the surrogate generator loss in Section", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 334, + 505, + 347 + ], + "spans": [ + { + "bbox": [ + 105, + 334, + 505, + 347 + ], + "score": 1.0, + "content": "2.2, of unrolling the inner of two nested min and max functions, does not apply to using a surrogate", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 345, + 505, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 505, + 358 + ], + "score": 1.0, + "content": "discriminator loss. Additionally, it is more common for the discriminator to overpower the generator", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 356, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 375, + 369 + ], + "score": 1.0, + "content": "than vice-versa when training a GAN. Giving more information to", + "type": "text" + }, + { + "bbox": [ + 375, + 357, + 384, + 366 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 356, + 505, + 369 + ], + "score": 1.0, + "content": "by allowing it to ‘see into the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 368, + 332, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 368, + 332, + 379 + ], + "score": 1.0, + "content": "future’ may thus help the two models be more balanced.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20 + }, + { + "type": "title", + "bbox": [ + 108, + 397, + 200, + 409 + ], + "lines": [ + { + "bbox": [ + 104, + 396, + 202, + 412 + ], + "spans": [ + { + "bbox": [ + 104, + 396, + 202, + 412 + ], + "score": 1.0, + "content": "3 EXPERIMENTS", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 423, + 505, + 534 + ], + "lines": [ + { + "bbox": [ + 105, + 424, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 505, + 436 + ], + "score": 1.0, + "content": "In this section we demonstrate improved mode coverage and stability by applying this technique to", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 434, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 505, + 447 + ], + "score": 1.0, + "content": "five datasets of increasing complexity. Evaluation of generative models is a notoriously hard problem", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 444, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 505, + 459 + ], + "score": 1.0, + "content": "(Theis et al., 2016). As such the de facto standard in GAN literature has become sample quality as", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 457, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 505, + 469 + ], + "score": 1.0, + "content": "evaluated by a human and/or evaluated by a heuristic (Inception score for example, (Salimans et al.,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 466, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 506, + 480 + ], + "score": 1.0, + "content": "2016)). While these evaluation metrics do a reasonable job capturing sample quality, they fail to", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 478, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 505, + 492 + ], + "score": 1.0, + "content": "capture sample diversity. In our first 2 experiments diversity is easily evaluated via visual inspection.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 488, + 506, + 503 + ], + "spans": [ + { + "bbox": [ + 104, + 488, + 506, + 503 + ], + "score": 1.0, + "content": "In our later experiments this is not the case, and we will use a variety of methods to quantify coverage", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 501, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 505, + 513 + ], + "score": 1.0, + "content": "of samples. Our measures are individually strongly suggestive of unrolling reducing mode-collapse", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 512, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 106, + 512, + 505, + 523 + ], + "score": 1.0, + "content": "and improving stability, but none of them alone are conclusive. We believe that taken together", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 520, + 452, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 452, + 537 + ], + "score": 1.0, + "content": "however, they provide extremely compelling evidence for the advantages of unrolling.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 108, + 539, + 505, + 583 + ], + "lines": [ + { + "bbox": [ + 105, + 538, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 506, + 552 + ], + "score": 1.0, + "content": "When doing stochastic optimization, we must choose which minibatches to use in the unrolling", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 550, + 506, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 506, + 562 + ], + "score": 1.0, + "content": "updates in Eq. 7. We experimented with both a fixed minibatch and re-sampled minibatches for", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "score": 1.0, + "content": "each unrolling step, and found it did not significantly impact the result. We use fixed minibatches", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 572, + 244, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 244, + 585 + ], + "score": 1.0, + "content": "for all experiments in this section.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 106, + 589, + 474, + 600 + ], + "lines": [ + { + "bbox": [ + 105, + 587, + 477, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 477, + 603 + ], + "score": 1.0, + "content": "We provide a reference implementation of this technique at github.com/poolio/unrolled gan.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "title", + "bbox": [ + 108, + 616, + 281, + 627 + ], + "lines": [ + { + "bbox": [ + 106, + 615, + 282, + 628 + ], + "spans": [ + { + "bbox": [ + 106, + 615, + 282, + 628 + ], + "score": 1.0, + "content": "3.1 MIXTURE OF GAUSSIANS DATASET", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 107, + 637, + 504, + 704 + ], + "lines": [ + { + "bbox": [ + 106, + 638, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 505, + 650 + ], + "score": 1.0, + "content": "To illustrate the impact of discriminator unrolling, we train a simple GAN architecture on a 2D", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 648, + 506, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 506, + 662 + ], + "score": 1.0, + "content": "mixture of 8 Gaussians arranged in a circle. For a detailed list of architecture and hyperparameters", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 660, + 506, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 506, + 673 + ], + "score": 1.0, + "content": "see Appendix A. Figure 2 shows the dynamics of this model through time. Without unrolling the", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 671, + 506, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 506, + 683 + ], + "score": 1.0, + "content": "generator rotates around the valid modes of the data distribution but is never able to spread out", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 682, + 506, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 263, + 695 + ], + "score": 1.0, + "content": "mass. When adding in unrolling steps", + "type": "text" + }, + { + "bbox": [ + 263, + 682, + 272, + 692 + ], + "score": 0.37, + "content": "\\mathbf { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 682, + 506, + 695 + ], + "score": 1.0, + "content": "quickly learns to spread probability mass and the system", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 693, + 243, + 705 + ], + "spans": [ + { + "bbox": [ + 106, + 693, + 243, + 705 + ], + "score": 1.0, + "content": "converges to the data distribution.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 43.5 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "In Appendix B we perform further experiments on this toy dataset. We explore how unrolling", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 721, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 505, + 733 + ], + "score": 1.0, + "content": "compares to historical averaging, and compares to using the unrolled discriminator to update the", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 47.5 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 109, + 82, + 313, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 316, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 316, + 96 + ], + "score": 1.0, + "content": "2.5 CONSEQUENCES OF THE SURROGATE LOSS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 104, + 505, + 152 + ], + "lines": [ + { + "bbox": [ + 105, + 103, + 505, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 352, + 117 + ], + "score": 1.0, + "content": "GANs can be thought of as a game between the discriminator", + "type": "text" + }, + { + "bbox": [ + 352, + 105, + 367, + 115 + ], + "score": 0.75, + "content": "( D )", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 103, + 439, + 117 + ], + "score": 1.0, + "content": "and the generator", + "type": "text" + }, + { + "bbox": [ + 440, + 105, + 455, + 115 + ], + "score": 0.69, + "content": "( G )", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 103, + 505, + 117 + ], + "score": 1.0, + "content": ". The agents", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 505, + 128 + ], + "score": 1.0, + "content": "take turns taking actions and updating their parameters until a Nash equilibrium is reached. The", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 125, + 504, + 144 + ], + "spans": [ + { + "bbox": [ + 106, + 125, + 182, + 144 + ], + "score": 1.0, + "content": "optimal action for", + "type": "text" + }, + { + "bbox": [ + 182, + 128, + 192, + 138 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 125, + 334, + 144 + ], + "score": 1.0, + "content": "is to evaluate the probability ratio", + "type": "text" + }, + { + "bbox": [ + 334, + 127, + 393, + 143 + ], + "score": 0.95, + "content": "\\frac { p _ { d a t a } ( x ) } { p _ { G } ( x ) + p _ { d a t a } ( x ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 125, + 497, + 144 + ], + "score": 1.0, + "content": "for the generator’s move", + "type": "text" + }, + { + "bbox": [ + 497, + 131, + 504, + 138 + ], + "score": 0.67, + "content": "x", + "type": "inline_equation" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 141, + 428, + 153 + ], + "spans": [ + { + "bbox": [ + 106, + 141, + 428, + 153 + ], + "score": 1.0, + "content": "(Eq. 5). The optimal generator action is to move its mass to maximize this ratio.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2.5, + "bbox_fs": [ + 105, + 103, + 505, + 153 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 158, + 505, + 246 + ], + "lines": [ + { + "bbox": [ + 106, + 157, + 505, + 170 + ], + "spans": [ + { + "bbox": [ + 106, + 157, + 187, + 170 + ], + "score": 1.0, + "content": "The initial move for", + "type": "text" + }, + { + "bbox": [ + 187, + 159, + 196, + 168 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 157, + 505, + 170 + ], + "score": 1.0, + "content": "will be to move as much mass as its parametric family and update step permits", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 169, + 505, + 181 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 427, + 181 + ], + "score": 1.0, + "content": "to the single point that maximizes the ratio of probability densities. The action", + "type": "text" + }, + { + "bbox": [ + 427, + 169, + 437, + 179 + ], + "score": 0.82, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 169, + 505, + 181 + ], + "score": 1.0, + "content": "will then take is", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 180, + 506, + 193 + ], + "spans": [ + { + "bbox": [ + 105, + 180, + 506, + 193 + ], + "score": 1.0, + "content": "quite simple. It will track that point, and to the extent allowed by its own parametric family and", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 191, + 505, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 505, + 203 + ], + "score": 1.0, + "content": "update step assign low data probability to it, and uniform probability everywhere else. This cycle", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 202, + 506, + 215 + ], + "spans": [ + { + "bbox": [ + 105, + 202, + 118, + 215 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 203, + 127, + 212 + ], + "score": 0.81, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 202, + 180, + 215 + ], + "score": 1.0, + "content": "moving and", + "type": "text" + }, + { + "bbox": [ + 180, + 203, + 190, + 212 + ], + "score": 0.79, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 202, + 506, + 215 + ], + "score": 1.0, + "content": "following will repeat forever or converge depending on the rate of change of", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 213, + 505, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 505, + 225 + ], + "score": 1.0, + "content": "the two agents. This is similar to the situation in simple matrix games like rock-paper-scissors and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 224, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 105, + 224, + 505, + 236 + ], + "score": 1.0, + "content": "matching pennies, where alternating gradient descent (ascent) with a fixed learning rate is known", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 235, + 356, + 247 + ], + "spans": [ + { + "bbox": [ + 105, + 235, + 356, + 247 + ], + "score": 1.0, + "content": "not to converge (Singh et al., 2000; Bowling & Veloso, 2002).", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 157, + 506, + 247 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 252, + 505, + 296 + ], + "lines": [ + { + "bbox": [ + 105, + 251, + 505, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 251, + 421, + 264 + ], + "score": 1.0, + "content": "In the unrolled case, however, this undesirable behavior no longer occurs. Now", + "type": "text" + }, + { + "bbox": [ + 421, + 252, + 430, + 262 + ], + "score": 0.71, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 251, + 505, + 264 + ], + "score": 1.0, + "content": "’s actions take into", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 263, + 505, + 274 + ], + "spans": [ + { + "bbox": [ + 106, + 263, + 160, + 274 + ], + "score": 1.0, + "content": "account how", + "type": "text" + }, + { + "bbox": [ + 161, + 263, + 171, + 273 + ], + "score": 0.78, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 263, + 284, + 274 + ], + "score": 1.0, + "content": "will respond. In particular,", + "type": "text" + }, + { + "bbox": [ + 285, + 263, + 294, + 273 + ], + "score": 0.77, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 263, + 405, + 274 + ], + "score": 1.0, + "content": "will try to make steps that", + "type": "text" + }, + { + "bbox": [ + 405, + 263, + 415, + 273 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 263, + 505, + 274 + ], + "score": 1.0, + "content": "will have a hard time", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 274, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 274, + 475, + 286 + ], + "score": 1.0, + "content": "responding to. This extra information helps the generator spread its mass to make the next", + "type": "text" + }, + { + "bbox": [ + 475, + 274, + 485, + 284 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 274, + 505, + 286 + ], + "score": 1.0, + "content": "step", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 284, + 288, + 297 + ], + "spans": [ + { + "bbox": [ + 105, + 284, + 288, + 297 + ], + "score": 1.0, + "content": "less effective instead of collapsing to a point.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 251, + 505, + 297 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 301, + 505, + 379 + ], + "lines": [ + { + "bbox": [ + 106, + 301, + 505, + 314 + ], + "spans": [ + { + "bbox": [ + 106, + 301, + 347, + 314 + ], + "score": 1.0, + "content": "In principle, a surrogate loss function could be used for both", + "type": "text" + }, + { + "bbox": [ + 347, + 302, + 357, + 312 + ], + "score": 0.8, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 301, + 374, + 314 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 375, + 302, + 383, + 312 + ], + "score": 0.75, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 301, + 505, + 314 + ], + "score": 1.0, + "content": ". In the case of 1-step unrolled", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 313, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 505, + 325 + ], + "score": 1.0, + "content": "optimization this is known to lead to convergence for games in which gradient descent (ascent) fails", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 324, + 505, + 336 + ], + "spans": [ + { + "bbox": [ + 106, + 324, + 505, + 336 + ], + "score": 1.0, + "content": "(Zhang & Lesser, 2010). However, the motivation for using the surrogate generator loss in Section", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 334, + 505, + 347 + ], + "spans": [ + { + "bbox": [ + 105, + 334, + 505, + 347 + ], + "score": 1.0, + "content": "2.2, of unrolling the inner of two nested min and max functions, does not apply to using a surrogate", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 345, + 505, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 345, + 505, + 358 + ], + "score": 1.0, + "content": "discriminator loss. Additionally, it is more common for the discriminator to overpower the generator", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 356, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 375, + 369 + ], + "score": 1.0, + "content": "than vice-versa when training a GAN. Giving more information to", + "type": "text" + }, + { + "bbox": [ + 375, + 357, + 384, + 366 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 356, + 505, + 369 + ], + "score": 1.0, + "content": "by allowing it to ‘see into the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 368, + 332, + 379 + ], + "spans": [ + { + "bbox": [ + 105, + 368, + 332, + 379 + ], + "score": 1.0, + "content": "future’ may thus help the two models be more balanced.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 301, + 505, + 379 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 397, + 200, + 409 + ], + "lines": [ + { + "bbox": [ + 104, + 396, + 202, + 412 + ], + "spans": [ + { + "bbox": [ + 104, + 396, + 202, + 412 + ], + "score": 1.0, + "content": "3 EXPERIMENTS", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 423, + 505, + 534 + ], + "lines": [ + { + "bbox": [ + 105, + 424, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 505, + 436 + ], + "score": 1.0, + "content": "In this section we demonstrate improved mode coverage and stability by applying this technique to", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 434, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 505, + 447 + ], + "score": 1.0, + "content": "five datasets of increasing complexity. Evaluation of generative models is a notoriously hard problem", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 444, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 444, + 505, + 459 + ], + "score": 1.0, + "content": "(Theis et al., 2016). As such the de facto standard in GAN literature has become sample quality as", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 457, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 457, + 505, + 469 + ], + "score": 1.0, + "content": "evaluated by a human and/or evaluated by a heuristic (Inception score for example, (Salimans et al.,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 466, + 506, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 506, + 480 + ], + "score": 1.0, + "content": "2016)). While these evaluation metrics do a reasonable job capturing sample quality, they fail to", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 478, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 505, + 492 + ], + "score": 1.0, + "content": "capture sample diversity. In our first 2 experiments diversity is easily evaluated via visual inspection.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 488, + 506, + 503 + ], + "spans": [ + { + "bbox": [ + 104, + 488, + 506, + 503 + ], + "score": 1.0, + "content": "In our later experiments this is not the case, and we will use a variety of methods to quantify coverage", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 501, + 505, + 513 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 505, + 513 + ], + "score": 1.0, + "content": "of samples. Our measures are individually strongly suggestive of unrolling reducing mode-collapse", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 512, + 505, + 523 + ], + "spans": [ + { + "bbox": [ + 106, + 512, + 505, + 523 + ], + "score": 1.0, + "content": "and improving stability, but none of them alone are conclusive. We believe that taken together", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 520, + 452, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 452, + 537 + ], + "score": 1.0, + "content": "however, they provide extremely compelling evidence for the advantages of unrolling.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 29.5, + "bbox_fs": [ + 104, + 424, + 506, + 537 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 539, + 505, + 583 + ], + "lines": [ + { + "bbox": [ + 105, + 538, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 506, + 552 + ], + "score": 1.0, + "content": "When doing stochastic optimization, we must choose which minibatches to use in the unrolling", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 550, + 506, + 562 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 506, + 562 + ], + "score": 1.0, + "content": "updates in Eq. 7. We experimented with both a fixed minibatch and re-sampled minibatches for", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "score": 1.0, + "content": "each unrolling step, and found it did not significantly impact the result. We use fixed minibatches", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 572, + 244, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 244, + 585 + ], + "score": 1.0, + "content": "for all experiments in this section.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36.5, + "bbox_fs": [ + 105, + 538, + 506, + 585 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 589, + 474, + 600 + ], + "lines": [ + { + "bbox": [ + 105, + 587, + 477, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 477, + 603 + ], + "score": 1.0, + "content": "We provide a reference implementation of this technique at github.com/poolio/unrolled gan.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39, + "bbox_fs": [ + 105, + 587, + 477, + 603 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 616, + 281, + 627 + ], + "lines": [ + { + "bbox": [ + 106, + 615, + 282, + 628 + ], + "spans": [ + { + "bbox": [ + 106, + 615, + 282, + 628 + ], + "score": 1.0, + "content": "3.1 MIXTURE OF GAUSSIANS DATASET", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 107, + 637, + 504, + 704 + ], + "lines": [ + { + "bbox": [ + 106, + 638, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 505, + 650 + ], + "score": 1.0, + "content": "To illustrate the impact of discriminator unrolling, we train a simple GAN architecture on a 2D", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 648, + 506, + 662 + ], + "spans": [ + { + "bbox": [ + 105, + 648, + 506, + 662 + ], + "score": 1.0, + "content": "mixture of 8 Gaussians arranged in a circle. For a detailed list of architecture and hyperparameters", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 660, + 506, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 506, + 673 + ], + "score": 1.0, + "content": "see Appendix A. Figure 2 shows the dynamics of this model through time. Without unrolling the", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 671, + 506, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 506, + 683 + ], + "score": 1.0, + "content": "generator rotates around the valid modes of the data distribution but is never able to spread out", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 682, + 506, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 263, + 695 + ], + "score": 1.0, + "content": "mass. When adding in unrolling steps", + "type": "text" + }, + { + "bbox": [ + 263, + 682, + 272, + 692 + ], + "score": 0.37, + "content": "\\mathbf { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 682, + 506, + 695 + ], + "score": 1.0, + "content": "quickly learns to spread probability mass and the system", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 693, + 243, + 705 + ], + "spans": [ + { + "bbox": [ + 106, + 693, + 243, + 705 + ], + "score": 1.0, + "content": "converges to the data distribution.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 43.5, + "bbox_fs": [ + 105, + 638, + 506, + 705 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 723 + ], + "score": 1.0, + "content": "In Appendix B we perform further experiments on this toy dataset. We explore how unrolling", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 721, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 505, + 733 + ], + "score": 1.0, + "content": "compares to historical averaging, and compares to using the unrolled discriminator to update the", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 552, + 506, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 506, + 565 + ], + "score": 1.0, + "content": "generator, but without backpropagating through the generator. In both cases we find that the unrolled", + "type": "text", + "cross_page": true + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 563, + 211, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 211, + 576 + ], + "score": 1.0, + "content": "objective performs better.", + "type": "text", + "cross_page": true + } + ], + "index": 18 + } + ], + "index": 47.5, + "bbox_fs": [ + 105, + 708, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 107, + 80, + 504, + 164 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 80, + 504, + 164 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 80, + 504, + 164 + ], + "spans": [ + { + "bbox": [ + 107, + 80, + 504, + 164 + ], + "score": 0.967, + "type": "image", + "image_path": "a4459b06fa7e9fcdf667d911bf408b5c5ccfe73af1d14e29428d98326008c54a.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 107, + 80, + 504, + 108.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 107, + 108.0, + 504, + 136.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 107, + 136.0, + 504, + 164.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 176, + 505, + 254 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 176, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 506, + 190 + ], + "score": 1.0, + "content": "Figure 2: Unrolling the discriminator stabilizes GAN training on a toy 2D mixture of Gaussians", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 186, + 505, + 202 + ], + "spans": [ + { + "bbox": [ + 104, + 186, + 505, + 202 + ], + "score": 1.0, + "content": "dataset. Columns show a heatmap of the generator distribution after increasing numbers of training", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 199, + 504, + 210 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 504, + 210 + ], + "score": 1.0, + "content": "steps. The final column shows the data distribution. The top row shows training for a GAN with", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 210, + 505, + 222 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 505, + 222 + ], + "score": 1.0, + "content": "10 unrolling steps. Its generator quickly spreads out and converges to the target distribution. The", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 220, + 505, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 220, + 505, + 233 + ], + "score": 1.0, + "content": "bottom row shows standard GAN training. The generator rotates through the modes of the data", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 230, + 505, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 505, + 245 + ], + "score": 1.0, + "content": "distribution. It never converges to a fixed distribution, and only ever assigns significant probability", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 244, + 249, + 255 + ], + "spans": [ + { + "bbox": [ + 106, + 244, + 249, + 255 + ], + "score": 1.0, + "content": "mass to a single data mode at once.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6 + } + ], + "index": 3.5 + }, + { + "type": "image", + "bbox": [ + 106, + 265, + 505, + 475 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 265, + 505, + 475 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 265, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 106, + 265, + 505, + 475 + ], + "score": 0.97, + "type": "image", + "image_path": "ff67f1c079a9b83431ee462581bfb81893ea4eadc9da332a9fff1968becf8c08.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 106, + 265, + 505, + 335.0 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 106, + 335.0, + 505, + 405.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 106, + 405.0, + 505, + 475.0 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 486, + 505, + 531 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 487, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 106, + 487, + 505, + 498 + ], + "score": 1.0, + "content": "Figure 3: Unrolled GAN training increases stability for an RNN generator and convolutional dis-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 498, + 506, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 506, + 510 + ], + "score": 1.0, + "content": "criminator trained on MNIST. The top row was run with 20 unrolling steps. The bottom row is a", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 508, + 506, + 521 + ], + "spans": [ + { + "bbox": [ + 105, + 508, + 506, + 521 + ], + "score": 1.0, + "content": "standard GAN, with 0 unrolling steps. Images are samples from the generator after the indicated", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 519, + 208, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 208, + 532 + ], + "score": 1.0, + "content": "number of training steps.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14.5 + } + ], + "index": 12.75 + }, + { + "type": "text", + "bbox": [ + 107, + 552, + 504, + 575 + ], + "lines": [ + { + "bbox": [ + 105, + 552, + 506, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 506, + 565 + ], + "score": 1.0, + "content": "generator, but without backpropagating through the generator. In both cases we find that the unrolled", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 563, + 211, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 211, + 576 + ], + "score": 1.0, + "content": "objective performs better.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "title", + "bbox": [ + 106, + 590, + 463, + 601 + ], + "lines": [ + { + "bbox": [ + 105, + 589, + 465, + 602 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 465, + 602 + ], + "score": 1.0, + "content": "3.2 PATHOLOGICAL MODEL WITH MISMATCHED GENERATOR AND DISCRIMINATOR", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 610, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 609, + 505, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 505, + 625 + ], + "score": 1.0, + "content": "To evaluate the ability of this approach to improve trainability, we look to a traditionally challenging", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 621, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 505, + 635 + ], + "score": 1.0, + "content": "family of models to train – recurrent neural networks (RNNs). In this experiment we try to generate", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 633, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 455, + 645 + ], + "score": 1.0, + "content": "MNIST samples using an LSTM (Hochreiter & Schmidhuber, 1997). MNIST digits are", + "type": "text" + }, + { + "bbox": [ + 455, + 633, + 482, + 644 + ], + "score": 0.51, + "content": "2 8 \\mathbf { x } 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 482, + 633, + 505, + 645 + ], + "score": 1.0, + "content": "pixel", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 644, + 506, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 506, + 657 + ], + "score": 1.0, + "content": "images. At each timestep of the generator LSTM, it outputs one column of this image, so that", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "score": 1.0, + "content": "after 28 timesteps it has output the entire sample. We use a convolutional neural network as the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "discriminator. See Appendix C for the full model and training details. Unlike in all previously", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "successful GAN models, there is no symmetry between the generator and the discriminator in this", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 687, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 701 + ], + "score": 1.0, + "content": "task, resulting in a more complex power balance. Results can be seen in Figure 3. Once again,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "without unrolling the model quickly collapses, and rotates through a sequence of single modes.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 709, + 504, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 504, + 722 + ], + "score": 1.0, + "content": "Instead of rotating spatially, it cycles through proto-digit like blobs. When running with unrolling", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 721, + 504, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 504, + 733 + ], + "score": 1.0, + "content": "steps the generator disperses and appears to cover the whole data distribution, as in the 2D example.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 25 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 310, + 762 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 107, + 80, + 504, + 164 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 80, + 504, + 164 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 80, + 504, + 164 + ], + "spans": [ + { + "bbox": [ + 107, + 80, + 504, + 164 + ], + "score": 0.967, + "type": "image", + "image_path": "a4459b06fa7e9fcdf667d911bf408b5c5ccfe73af1d14e29428d98326008c54a.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 107, + 80, + 504, + 108.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 107, + 108.0, + 504, + 136.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 107, + 136.0, + 504, + 164.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 176, + 505, + 254 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 176, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 506, + 190 + ], + "score": 1.0, + "content": "Figure 2: Unrolling the discriminator stabilizes GAN training on a toy 2D mixture of Gaussians", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 186, + 505, + 202 + ], + "spans": [ + { + "bbox": [ + 104, + 186, + 505, + 202 + ], + "score": 1.0, + "content": "dataset. Columns show a heatmap of the generator distribution after increasing numbers of training", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 199, + 504, + 210 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 504, + 210 + ], + "score": 1.0, + "content": "steps. The final column shows the data distribution. The top row shows training for a GAN with", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 210, + 505, + 222 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 505, + 222 + ], + "score": 1.0, + "content": "10 unrolling steps. Its generator quickly spreads out and converges to the target distribution. The", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 220, + 505, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 220, + 505, + 233 + ], + "score": 1.0, + "content": "bottom row shows standard GAN training. The generator rotates through the modes of the data", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 230, + 505, + 245 + ], + "spans": [ + { + "bbox": [ + 105, + 230, + 505, + 245 + ], + "score": 1.0, + "content": "distribution. It never converges to a fixed distribution, and only ever assigns significant probability", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 244, + 249, + 255 + ], + "spans": [ + { + "bbox": [ + 106, + 244, + 249, + 255 + ], + "score": 1.0, + "content": "mass to a single data mode at once.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6 + } + ], + "index": 3.5 + }, + { + "type": "image", + "bbox": [ + 106, + 265, + 505, + 475 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 265, + 505, + 475 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 265, + 505, + 475 + ], + "spans": [ + { + "bbox": [ + 106, + 265, + 505, + 475 + ], + "score": 0.97, + "type": "image", + "image_path": "ff67f1c079a9b83431ee462581bfb81893ea4eadc9da332a9fff1968becf8c08.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 106, + 265, + 505, + 335.0 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 106, + 335.0, + 505, + 405.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 106, + 405.0, + 505, + 475.0 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 486, + 505, + 531 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 487, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 106, + 487, + 505, + 498 + ], + "score": 1.0, + "content": "Figure 3: Unrolled GAN training increases stability for an RNN generator and convolutional dis-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 498, + 506, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 506, + 510 + ], + "score": 1.0, + "content": "criminator trained on MNIST. The top row was run with 20 unrolling steps. The bottom row is a", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 508, + 506, + 521 + ], + "spans": [ + { + "bbox": [ + 105, + 508, + 506, + 521 + ], + "score": 1.0, + "content": "standard GAN, with 0 unrolling steps. Images are samples from the generator after the indicated", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 519, + 208, + 532 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 208, + 532 + ], + "score": 1.0, + "content": "number of training steps.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14.5 + } + ], + "index": 12.75 + }, + { + "type": "text", + "bbox": [ + 107, + 552, + 504, + 575 + ], + "lines": [], + "index": 17.5, + "bbox_fs": [ + 105, + 552, + 506, + 576 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 106, + 590, + 463, + 601 + ], + "lines": [ + { + "bbox": [ + 105, + 589, + 465, + 602 + ], + "spans": [ + { + "bbox": [ + 105, + 589, + 465, + 602 + ], + "score": 1.0, + "content": "3.2 PATHOLOGICAL MODEL WITH MISMATCHED GENERATOR AND DISCRIMINATOR", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 610, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 609, + 505, + 625 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 505, + 625 + ], + "score": 1.0, + "content": "To evaluate the ability of this approach to improve trainability, we look to a traditionally challenging", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 621, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 505, + 635 + ], + "score": 1.0, + "content": "family of models to train – recurrent neural networks (RNNs). In this experiment we try to generate", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 633, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 455, + 645 + ], + "score": 1.0, + "content": "MNIST samples using an LSTM (Hochreiter & Schmidhuber, 1997). MNIST digits are", + "type": "text" + }, + { + "bbox": [ + 455, + 633, + 482, + 644 + ], + "score": 0.51, + "content": "2 8 \\mathbf { x } 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 482, + 633, + 505, + 645 + ], + "score": 1.0, + "content": "pixel", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 644, + 506, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 506, + 657 + ], + "score": 1.0, + "content": "images. At each timestep of the generator LSTM, it outputs one column of this image, so that", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 668 + ], + "score": 1.0, + "content": "after 28 timesteps it has output the entire sample. We use a convolutional neural network as the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "discriminator. See Appendix C for the full model and training details. Unlike in all previously", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "successful GAN models, there is no symmetry between the generator and the discriminator in this", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 687, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 701 + ], + "score": 1.0, + "content": "task, resulting in a more complex power balance. Results can be seen in Figure 3. Once again,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 711 + ], + "score": 1.0, + "content": "without unrolling the model quickly collapses, and rotates through a sequence of single modes.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 709, + 504, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 504, + 722 + ], + "score": 1.0, + "content": "Instead of rotating spatially, it cycles through proto-digit like blobs. When running with unrolling", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 721, + 504, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 721, + 504, + 733 + ], + "score": 1.0, + "content": "steps the generator disperses and appears to cover the whole data distribution, as in the 2D example.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 25, + "bbox_fs": [ + 105, + 609, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 59, + 80, + 551, + 139 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 59, + 80, + 551, + 139 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 59, + 80, + 551, + 139 + ], + "spans": [ + { + "bbox": [ + 59, + 80, + 551, + 139 + ], + "score": 0.978, + "html": "
Discriminator SizeUnrolling steps01510
1/4 size of D compared to GModes generated30.6± 20.7365.4 ± 34.75236.4 ± 63.30327.2 ± 74.67
KL(model||data)5.99± 0.425.911 ± 0.144.67 ± 0.434.66 ± 0.46
1/2 size of D compared to GModes generated628.0± 140.9523.6± 55.768732.0± 44.98817.4 ± 37.91
KL(model||data)2.58 ±0.7512.44 ±0.261.66 ± 0.0901.43 ± 0.12
", + "type": "table", + "image_path": "e4cb5f019901d91438706efd6a3956d78e353190e6df31aa58c09fecfa15f59b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 59, + 80, + 551, + 99.66666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 59, + 99.66666666666667, + 551, + 119.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 59, + 119.33333333333334, + 551, + 139.0 + ], + "spans": [], + "index": 2 + } + ] + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 147, + 505, + 203 + ], + "lines": [ + { + "bbox": [ + 106, + 148, + 505, + 160 + ], + "spans": [ + { + "bbox": [ + 106, + 148, + 505, + 160 + ], + "score": 1.0, + "content": "Table 1: Unrolled GANs cover more discrete modes when modeling a dataset with 1,000 data modes,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 158, + 506, + 171 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 341, + 171 + ], + "score": 1.0, + "content": "corresponding to all combinations of three MNIST digits", + "type": "text" + }, + { + "bbox": [ + 342, + 158, + 357, + 169 + ], + "score": 0.87, + "content": "[ 1 0 ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 158, + 506, + 171 + ], + "score": 1.0, + "content": "digit combinations). The number of", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 169, + 505, + 182 + ], + "spans": [ + { + "bbox": [ + 106, + 169, + 505, + 182 + ], + "score": 1.0, + "content": "modes covered is given for different numbers of unrolling steps, and for two different architectures.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 179, + 506, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 179, + 506, + 194 + ], + "score": 1.0, + "content": "The reverse KL divergence between model and data is also given. Standard error is provided for", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 190, + 169, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 190, + 169, + 204 + ], + "score": 1.0, + "content": "both measures.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + }, + { + "type": "title", + "bbox": [ + 108, + 225, + 397, + 235 + ], + "lines": [ + { + "bbox": [ + 105, + 224, + 399, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 224, + 399, + 237 + ], + "score": 1.0, + "content": "3.3 MODE AND MANIFOLD COLLAPSE USING AUGMENTED MNIST", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 107, + 245, + 505, + 290 + ], + "lines": [ + { + "bbox": [ + 105, + 245, + 506, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 506, + 258 + ], + "score": 1.0, + "content": "GANs suffer from two different types of model collapse – collapse to a subset of data modes, and", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 258, + 505, + 268 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 505, + 268 + ], + "score": 1.0, + "content": "collapse to a sub-manifold within the data distribution. In these experiments we isolate both effects", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 268, + 505, + 280 + ], + "spans": [ + { + "bbox": [ + 106, + 268, + 505, + 280 + ], + "score": 1.0, + "content": "using artificially constructed datasets, and demonstrate that unrolling can largely rescue both types", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 278, + 156, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 278, + 156, + 293 + ], + "score": 1.0, + "content": "of collapse.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5 + }, + { + "type": "title", + "bbox": [ + 108, + 306, + 258, + 317 + ], + "lines": [ + { + "bbox": [ + 106, + 306, + 260, + 318 + ], + "spans": [ + { + "bbox": [ + 106, + 306, + 260, + 318 + ], + "score": 1.0, + "content": "3.3.1 DISCRETE MODE COLLAPSE", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 326, + 504, + 381 + ], + "lines": [ + { + "bbox": [ + 105, + 325, + 506, + 339 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 506, + 339 + ], + "score": 1.0, + "content": "To explore the degree to which GANs drop discrete modes in a dataset, we use a technique similar", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 337, + 506, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 337, + 506, + 349 + ], + "score": 1.0, + "content": "to one from (Che et al., 2016). We construct a dataset by stacking three randomly chosen MNIST", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 349, + 505, + 360 + ], + "spans": [ + { + "bbox": [ + 106, + 349, + 505, + 360 + ], + "score": 1.0, + "content": "digits, so as to construct an RGB image with a different MNIST digit in each color channel. This", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 359, + 506, + 372 + ], + "spans": [ + { + "bbox": [ + 105, + 359, + 506, + 372 + ], + "score": 1.0, + "content": "new dataset has 1,000 distinct modes, corresponding to each combination of the ten MNIST classes", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 370, + 193, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 193, + 382 + ], + "score": 1.0, + "content": "in the three channels.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 107, + 387, + 505, + 464 + ], + "lines": [ + { + "bbox": [ + 106, + 388, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 505, + 398 + ], + "score": 1.0, + "content": "We train a GAN on this dataset, and generate samples from the trained model (25,600 samples for all", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 398, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 505, + 410 + ], + "score": 1.0, + "content": "experiments). We then compute the predicted class label of each color channel using a pre-trained", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 409, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 505, + 421 + ], + "score": 1.0, + "content": "MNIST classifier. To evaluate performance, we use two metrics: the number of modes for which the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 420, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 506, + 433 + ], + "score": 1.0, + "content": "generator produced at least one sample, and the KL divergence between the model and the expected", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 430, + 505, + 443 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 505, + 443 + ], + "score": 1.0, + "content": "data distribution. Within this discrete label space, a KL divergence can be estimated tractably be-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 443, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 505, + 453 + ], + "score": 1.0, + "content": "tween the generated samples and the data distribution over classes, where the data distribution is a", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 453, + 279, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 453, + 279, + 465 + ], + "score": 1.0, + "content": "uniform distribution over all 1,000 classes.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 470, + 505, + 514 + ], + "lines": [ + { + "bbox": [ + 106, + 470, + 505, + 483 + ], + "spans": [ + { + "bbox": [ + 106, + 470, + 505, + 483 + ], + "score": 1.0, + "content": "As presented in Table 1, as the number of unrolling steps is increased, both mode coverage and re-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 480, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 505, + 493 + ], + "score": 1.0, + "content": "verse KL divergence improve. Contrary to (Che et al., 2016), we found that reasonably sized models", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 492, + 505, + 504 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 505, + 504 + ], + "score": 1.0, + "content": "(such as the one used in Section 3.4) covered all 1,000 modes even without unrolling. As such we", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 502, + 495, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 495, + 515 + ], + "score": 1.0, + "content": "use smaller convolutional GAN models. Details on the models used are provided in Appendix E.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 107, + 519, + 505, + 575 + ], + "lines": [ + { + "bbox": [ + 106, + 518, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 106, + 518, + 505, + 532 + ], + "score": 1.0, + "content": "We observe an additional interesting effect in this experiment. The benefits of unrolling increase", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 531, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 505, + 543 + ], + "score": 1.0, + "content": "as the discriminator size is reduced. We believe unrolling effectively increases the capacity of the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 540, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 506, + 555 + ], + "score": 1.0, + "content": "discriminator. The unrolled discriminator can better react to any specific way in which the generator", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 551, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 505, + 567 + ], + "score": 1.0, + "content": "is producing non-data-like samples. When the discriminator is weak, the positive impact of unrolling", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 561, + 163, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 163, + 577 + ], + "score": 1.0, + "content": "is thus larger.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32 + }, + { + "type": "title", + "bbox": [ + 107, + 590, + 235, + 601 + ], + "lines": [ + { + "bbox": [ + 106, + 589, + 236, + 603 + ], + "spans": [ + { + "bbox": [ + 106, + 589, + 236, + 603 + ], + "score": 1.0, + "content": "3.3.2 MANIFOLD COLLAPSE", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 107, + 610, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 611, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 505, + 623 + ], + "score": 1.0, + "content": "In addition to discrete modes, we examine the effect of unrolling when modeling continuous mani-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "score": 1.0, + "content": "folds. To get at this quantity, we constructed a dataset consisting of colored MNIST digits. Unlike", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 633, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 505, + 646 + ], + "score": 1.0, + "content": "in the previous experiment, a single MNIST digit was chosen, and then assigned a single monochro-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "score": 1.0, + "content": "matic color. With a perfect generator, one should be able to recover the distribution of colors used to", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 654, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 506, + 668 + ], + "score": 1.0, + "content": "generate the digits. We use colored MNIST digits so that the generator also has to model the digits,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 666, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 506, + 678 + ], + "score": 1.0, + "content": "which makes the task sufficiently complex that the generator is unable to perfectly solve it. The", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "score": 1.0, + "content": "color of each digit is sampled from a 3D normal distribution. Details of this dataset are provided", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "in Appendix F. We will examine the distribution of colors in the samples generated by the trained", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "GAN. As will also be true in the CIFAR10 example in Section 3.4, the lack of diversity in gener-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "ated colors is almost invisible using only visual inspection of the samples. Samples can be found in", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 720, + 159, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 159, + 733 + ], + "score": 1.0, + "content": "Appendix F.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 41 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 303, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 59, + 80, + 551, + 139 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 59, + 80, + 551, + 139 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 59, + 80, + 551, + 139 + ], + "spans": [ + { + "bbox": [ + 59, + 80, + 551, + 139 + ], + "score": 0.978, + "html": "
Discriminator SizeUnrolling steps01510
1/4 size of D compared to GModes generated30.6± 20.7365.4 ± 34.75236.4 ± 63.30327.2 ± 74.67
KL(model||data)5.99± 0.425.911 ± 0.144.67 ± 0.434.66 ± 0.46
1/2 size of D compared to GModes generated628.0± 140.9523.6± 55.768732.0± 44.98817.4 ± 37.91
KL(model||data)2.58 ±0.7512.44 ±0.261.66 ± 0.0901.43 ± 0.12
", + "type": "table", + "image_path": "e4cb5f019901d91438706efd6a3956d78e353190e6df31aa58c09fecfa15f59b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 59, + 80, + 551, + 99.66666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 59, + 99.66666666666667, + 551, + 119.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 59, + 119.33333333333334, + 551, + 139.0 + ], + "spans": [], + "index": 2 + } + ] + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 147, + 505, + 203 + ], + "lines": [ + { + "bbox": [ + 106, + 148, + 505, + 160 + ], + "spans": [ + { + "bbox": [ + 106, + 148, + 505, + 160 + ], + "score": 1.0, + "content": "Table 1: Unrolled GANs cover more discrete modes when modeling a dataset with 1,000 data modes,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 158, + 506, + 171 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 341, + 171 + ], + "score": 1.0, + "content": "corresponding to all combinations of three MNIST digits", + "type": "text" + }, + { + "bbox": [ + 342, + 158, + 357, + 169 + ], + "score": 0.87, + "content": "[ 1 0 ^ { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 158, + 506, + 171 + ], + "score": 1.0, + "content": "digit combinations). The number of", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 169, + 505, + 182 + ], + "spans": [ + { + "bbox": [ + 106, + 169, + 505, + 182 + ], + "score": 1.0, + "content": "modes covered is given for different numbers of unrolling steps, and for two different architectures.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 179, + 506, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 179, + 506, + 194 + ], + "score": 1.0, + "content": "The reverse KL divergence between model and data is also given. Standard error is provided for", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 190, + 169, + 204 + ], + "spans": [ + { + "bbox": [ + 105, + 190, + 169, + 204 + ], + "score": 1.0, + "content": "both measures.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 148, + 506, + 204 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 225, + 397, + 235 + ], + "lines": [ + { + "bbox": [ + 105, + 224, + 399, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 224, + 399, + 237 + ], + "score": 1.0, + "content": "3.3 MODE AND MANIFOLD COLLAPSE USING AUGMENTED MNIST", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 107, + 245, + 505, + 290 + ], + "lines": [ + { + "bbox": [ + 105, + 245, + 506, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 506, + 258 + ], + "score": 1.0, + "content": "GANs suffer from two different types of model collapse – collapse to a subset of data modes, and", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 258, + 505, + 268 + ], + "spans": [ + { + "bbox": [ + 106, + 258, + 505, + 268 + ], + "score": 1.0, + "content": "collapse to a sub-manifold within the data distribution. In these experiments we isolate both effects", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 268, + 505, + 280 + ], + "spans": [ + { + "bbox": [ + 106, + 268, + 505, + 280 + ], + "score": 1.0, + "content": "using artificially constructed datasets, and demonstrate that unrolling can largely rescue both types", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 278, + 156, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 278, + 156, + 293 + ], + "score": 1.0, + "content": "of collapse.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 245, + 506, + 293 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 306, + 258, + 317 + ], + "lines": [ + { + "bbox": [ + 106, + 306, + 260, + 318 + ], + "spans": [ + { + "bbox": [ + 106, + 306, + 260, + 318 + ], + "score": 1.0, + "content": "3.3.1 DISCRETE MODE COLLAPSE", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 326, + 504, + 381 + ], + "lines": [ + { + "bbox": [ + 105, + 325, + 506, + 339 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 506, + 339 + ], + "score": 1.0, + "content": "To explore the degree to which GANs drop discrete modes in a dataset, we use a technique similar", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 337, + 506, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 337, + 506, + 349 + ], + "score": 1.0, + "content": "to one from (Che et al., 2016). We construct a dataset by stacking three randomly chosen MNIST", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 349, + 505, + 360 + ], + "spans": [ + { + "bbox": [ + 106, + 349, + 505, + 360 + ], + "score": 1.0, + "content": "digits, so as to construct an RGB image with a different MNIST digit in each color channel. This", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 359, + 506, + 372 + ], + "spans": [ + { + "bbox": [ + 105, + 359, + 506, + 372 + ], + "score": 1.0, + "content": "new dataset has 1,000 distinct modes, corresponding to each combination of the ten MNIST classes", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 370, + 193, + 382 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 193, + 382 + ], + "score": 1.0, + "content": "in the three channels.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 325, + 506, + 382 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 387, + 505, + 464 + ], + "lines": [ + { + "bbox": [ + 106, + 388, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 106, + 388, + 505, + 398 + ], + "score": 1.0, + "content": "We train a GAN on this dataset, and generate samples from the trained model (25,600 samples for all", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 398, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 505, + 410 + ], + "score": 1.0, + "content": "experiments). We then compute the predicted class label of each color channel using a pre-trained", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 409, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 505, + 421 + ], + "score": 1.0, + "content": "MNIST classifier. To evaluate performance, we use two metrics: the number of modes for which the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 420, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 506, + 433 + ], + "score": 1.0, + "content": "generator produced at least one sample, and the KL divergence between the model and the expected", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 430, + 505, + 443 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 505, + 443 + ], + "score": 1.0, + "content": "data distribution. Within this discrete label space, a KL divergence can be estimated tractably be-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 443, + 505, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 505, + 453 + ], + "score": 1.0, + "content": "tween the generated samples and the data distribution over classes, where the data distribution is a", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 453, + 279, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 453, + 279, + 465 + ], + "score": 1.0, + "content": "uniform distribution over all 1,000 classes.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 388, + 506, + 465 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 470, + 505, + 514 + ], + "lines": [ + { + "bbox": [ + 106, + 470, + 505, + 483 + ], + "spans": [ + { + "bbox": [ + 106, + 470, + 505, + 483 + ], + "score": 1.0, + "content": "As presented in Table 1, as the number of unrolling steps is increased, both mode coverage and re-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 480, + 505, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 505, + 493 + ], + "score": 1.0, + "content": "verse KL divergence improve. Contrary to (Che et al., 2016), we found that reasonably sized models", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 492, + 505, + 504 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 505, + 504 + ], + "score": 1.0, + "content": "(such as the one used in Section 3.4) covered all 1,000 modes even without unrolling. As such we", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 502, + 495, + 515 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 495, + 515 + ], + "score": 1.0, + "content": "use smaller convolutional GAN models. Details on the models used are provided in Appendix E.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5, + "bbox_fs": [ + 105, + 470, + 505, + 515 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 519, + 505, + 575 + ], + "lines": [ + { + "bbox": [ + 106, + 518, + 505, + 532 + ], + "spans": [ + { + "bbox": [ + 106, + 518, + 505, + 532 + ], + "score": 1.0, + "content": "We observe an additional interesting effect in this experiment. The benefits of unrolling increase", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 531, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 505, + 543 + ], + "score": 1.0, + "content": "as the discriminator size is reduced. We believe unrolling effectively increases the capacity of the", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 540, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 506, + 555 + ], + "score": 1.0, + "content": "discriminator. The unrolled discriminator can better react to any specific way in which the generator", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 551, + 505, + 567 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 505, + 567 + ], + "score": 1.0, + "content": "is producing non-data-like samples. When the discriminator is weak, the positive impact of unrolling", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 561, + 163, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 163, + 577 + ], + "score": 1.0, + "content": "is thus larger.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32, + "bbox_fs": [ + 105, + 518, + 506, + 577 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 590, + 235, + 601 + ], + "lines": [ + { + "bbox": [ + 106, + 589, + 236, + 603 + ], + "spans": [ + { + "bbox": [ + 106, + 589, + 236, + 603 + ], + "score": 1.0, + "content": "3.3.2 MANIFOLD COLLAPSE", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 107, + 610, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 611, + 505, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 611, + 505, + 623 + ], + "score": 1.0, + "content": "In addition to discrete modes, we examine the effect of unrolling when modeling continuous mani-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 505, + 635 + ], + "score": 1.0, + "content": "folds. To get at this quantity, we constructed a dataset consisting of colored MNIST digits. Unlike", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 633, + 505, + 646 + ], + "spans": [ + { + "bbox": [ + 105, + 633, + 505, + 646 + ], + "score": 1.0, + "content": "in the previous experiment, a single MNIST digit was chosen, and then assigned a single monochro-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "score": 1.0, + "content": "matic color. With a perfect generator, one should be able to recover the distribution of colors used to", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 654, + 506, + 668 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 506, + 668 + ], + "score": 1.0, + "content": "generate the digits. We use colored MNIST digits so that the generator also has to model the digits,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 666, + 506, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 506, + 678 + ], + "score": 1.0, + "content": "which makes the task sufficiently complex that the generator is unable to perfectly solve it. The", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 505, + 689 + ], + "score": 1.0, + "content": "color of each digit is sampled from a 3D normal distribution. Details of this dataset are provided", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "in Appendix F. We will examine the distribution of colors in the samples generated by the trained", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "GAN. As will also be true in the CIFAR10 example in Section 3.4, the lack of diversity in gener-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "ated colors is almost invisible using only visual inspection of the samples. Samples can be found in", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 720, + 159, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 159, + 733 + ], + "score": 1.0, + "content": "Appendix F.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 41, + "bbox_fs": [ + 105, + 611, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 82, + 80, + 529, + 128 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 82, + 80, + 529, + 128 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 82, + 80, + 529, + 128 + ], + "spans": [ + { + "bbox": [ + 82, + 80, + 529, + 128 + ], + "score": 0.975, + "html": "
Unrolling steps01510
JS divergence with 1/4 layer size0.073 ± 0.00580.142 ± 0.0280.049 ± 0.00210.075 ± 0.012
JS divergence with 1/2 layer size0.095 ± 0.0110.119 ± 0.0100.055 ± 0.00490.074± 0.016
JS divergence with 1/1 layer size0.034 ± 0.00340.050± 0.00260.027 ± 0.00280.025 ± 0.00076
", + "type": "table", + "image_path": "bdf448f52fbc7d7ff1a8589d6a25dd7b7fe2e14cdf8e504055b7f3a333d00a6e.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 82, + 80, + 529, + 96.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 82, + 96.0, + 529, + 112.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 82, + 112.0, + 529, + 128.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 106, + 136, + 505, + 181 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 135, + 505, + 148 + ], + "spans": [ + { + "bbox": [ + 105, + 135, + 505, + 148 + ], + "score": 1.0, + "content": "Table 2: Unrolled GANs better model a continuous distribution. GANs are trained to model ran-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 147, + 505, + 159 + ], + "spans": [ + { + "bbox": [ + 106, + 147, + 505, + 159 + ], + "score": 1.0, + "content": "domly colored MNIST digits, where the color is drawn from a Gaussian distribution. The JS diver-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 158, + 506, + 171 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 506, + 171 + ], + "score": 1.0, + "content": "gence between the data and model distributions over digit colors is then reported, along with standard", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 169, + 492, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 492, + 182 + ], + "score": 1.0, + "content": "error in the JS divergence. More unrolling steps, and larger models, lead to better JS divergence.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "image", + "bbox": [ + 113, + 192, + 488, + 348 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 113, + 192, + 488, + 348 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 192, + 488, + 348 + ], + "spans": [ + { + "bbox": [ + 113, + 192, + 488, + 348 + ], + "score": 0.965, + "type": "image", + "image_path": "3c9a4f15816f9475e20e64fe6d880999dbc71b9039d3a86102ace92afbbde1f2.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 113, + 192, + 488, + 244.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 113, + 244.0, + 488, + 296.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 113, + 296.0, + 488, + 348.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 371, + 505, + 415 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 371, + 505, + 384 + ], + "spans": [ + { + "bbox": [ + 106, + 371, + 505, + 384 + ], + "score": 1.0, + "content": "Figure 4: Visual perception of sample quality and diversity is very similar for models trained with", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 381, + 505, + 395 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 505, + 395 + ], + "score": 1.0, + "content": "different numbers of unrolling steps. Actual sample diversity is higher with more unrolling steps.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 392, + 506, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 506, + 406 + ], + "score": 1.0, + "content": "Each pane shows samples generated after training a model on CIFAR10 with 0, 1, 5, and 10 steps of", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 403, + 149, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 149, + 417 + ], + "score": 1.0, + "content": "unrolling.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11.5 + } + ], + "index": 9.75 + }, + { + "type": "text", + "bbox": [ + 107, + 439, + 505, + 505 + ], + "lines": [ + { + "bbox": [ + 106, + 439, + 505, + 451 + ], + "spans": [ + { + "bbox": [ + 106, + 439, + 505, + 451 + ], + "score": 1.0, + "content": "In order to recover the color the GAN assigned to the digit, we used k-means with 2 clusters, to", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 450, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 505, + 462 + ], + "score": 1.0, + "content": "pick out the foreground color from the background. We then performed this transformation for both", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 461, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 473 + ], + "score": 1.0, + "content": "the training data and the generated images. Next we fit a Gaussian kernel density estimator to both", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 471, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 505, + 484 + ], + "score": 1.0, + "content": "distributions over digit colors. Finally, we computed the JS divergence between the model and data", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 483, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 483, + 505, + 495 + ], + "score": 1.0, + "content": "distributions over colors. Results can be found in Table 2 for several model sizes. Details of the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 493, + 253, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 253, + 506 + ], + "score": 1.0, + "content": "models are provided in Appendix F.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 107, + 511, + 505, + 555 + ], + "lines": [ + { + "bbox": [ + 106, + 511, + 506, + 523 + ], + "spans": [ + { + "bbox": [ + 106, + 511, + 506, + 523 + ], + "score": 1.0, + "content": "In general, the best performing models are unrolled for 5-10 steps, and larger models perform better", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 521, + 506, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 506, + 534 + ], + "score": 1.0, + "content": "than smaller models. Counter-intuitively, taking 1 unrolling step seems to hurt this measure of", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 532, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 505, + 546 + ], + "score": 1.0, + "content": "diversity. We suspect that this is due to it introducing oscillatory dynamics into training. Taking", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 543, + 415, + 557 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 415, + 557 + ], + "score": 1.0, + "content": "more unrolling steps however leads to improved performance with unrolling.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21.5 + }, + { + "type": "title", + "bbox": [ + 108, + 572, + 267, + 583 + ], + "lines": [ + { + "bbox": [ + 106, + 572, + 269, + 585 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 269, + 585 + ], + "score": 1.0, + "content": "3.4 IMAGE MODELING OF CIFAR10", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 106, + 594, + 505, + 693 + ], + "lines": [ + { + "bbox": [ + 106, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "Here we test our technique on a more traditional convolutional GAN architecture and task, similar", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "to those used in (Radford et al., 2015; Salimans et al., 2016). In the previous experiments we tested", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 615, + 506, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 506, + 629 + ], + "score": 1.0, + "content": "models where the standard GAN training algorithm would not converge. In this section we improve", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 627, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 506, + 640 + ], + "score": 1.0, + "content": "a standard model by reducing its tendency to engage in mode collapse. We ran 4 configurations of", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 638, + 506, + 651 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 506, + 651 + ], + "score": 1.0, + "content": "this model, varying the number of unrolling steps to be 0, 1, 5, or 10. Each configuration was run 5", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 648, + 506, + 662 + ], + "spans": [ + { + "bbox": [ + 106, + 648, + 506, + 662 + ], + "score": 1.0, + "content": "times with different random seeds. For full training details see Appendix D. Samples from each of", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 659, + 506, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 506, + 673 + ], + "score": 1.0, + "content": "the 4 configurations can be found in Figure 4. There is no obvious difference in visual quality across", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 670, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 670, + 505, + 684 + ], + "score": 1.0, + "content": "these model configurations. Visual inspection however provides only a poor measure of sample", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 681, + 145, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 145, + 695 + ], + "score": 1.0, + "content": "diversity.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "By training with an unrolled discriminator, we expect to generate more diverse samples which more", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "closely resemble the underlying data distribution. We introduce two techniques to examine sample", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 720, + 399, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 399, + 733 + ], + "score": 1.0, + "content": "diversity: inference via optimization, and pairwise distance distributions.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 752, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 761 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 761 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 82, + 80, + 529, + 128 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 82, + 80, + 529, + 128 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 82, + 80, + 529, + 128 + ], + "spans": [ + { + "bbox": [ + 82, + 80, + 529, + 128 + ], + "score": 0.975, + "html": "
Unrolling steps01510
JS divergence with 1/4 layer size0.073 ± 0.00580.142 ± 0.0280.049 ± 0.00210.075 ± 0.012
JS divergence with 1/2 layer size0.095 ± 0.0110.119 ± 0.0100.055 ± 0.00490.074± 0.016
JS divergence with 1/1 layer size0.034 ± 0.00340.050± 0.00260.027 ± 0.00280.025 ± 0.00076
", + "type": "table", + "image_path": "bdf448f52fbc7d7ff1a8589d6a25dd7b7fe2e14cdf8e504055b7f3a333d00a6e.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 82, + 80, + 529, + 96.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 82, + 96.0, + 529, + 112.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 82, + 112.0, + 529, + 128.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 106, + 136, + 505, + 181 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 135, + 505, + 148 + ], + "spans": [ + { + "bbox": [ + 105, + 135, + 505, + 148 + ], + "score": 1.0, + "content": "Table 2: Unrolled GANs better model a continuous distribution. GANs are trained to model ran-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 147, + 505, + 159 + ], + "spans": [ + { + "bbox": [ + 106, + 147, + 505, + 159 + ], + "score": 1.0, + "content": "domly colored MNIST digits, where the color is drawn from a Gaussian distribution. The JS diver-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 158, + 506, + 171 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 506, + 171 + ], + "score": 1.0, + "content": "gence between the data and model distributions over digit colors is then reported, along with standard", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 169, + 492, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 492, + 182 + ], + "score": 1.0, + "content": "error in the JS divergence. More unrolling steps, and larger models, lead to better JS divergence.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "image", + "bbox": [ + 113, + 192, + 488, + 348 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 113, + 192, + 488, + 348 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 192, + 488, + 348 + ], + "spans": [ + { + "bbox": [ + 113, + 192, + 488, + 348 + ], + "score": 0.965, + "type": "image", + "image_path": "3c9a4f15816f9475e20e64fe6d880999dbc71b9039d3a86102ace92afbbde1f2.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 113, + 192, + 488, + 244.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 113, + 244.0, + 488, + 296.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 113, + 296.0, + 488, + 348.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 371, + 505, + 415 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 371, + 505, + 384 + ], + "spans": [ + { + "bbox": [ + 106, + 371, + 505, + 384 + ], + "score": 1.0, + "content": "Figure 4: Visual perception of sample quality and diversity is very similar for models trained with", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 381, + 505, + 395 + ], + "spans": [ + { + "bbox": [ + 105, + 381, + 505, + 395 + ], + "score": 1.0, + "content": "different numbers of unrolling steps. Actual sample diversity is higher with more unrolling steps.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 392, + 506, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 506, + 406 + ], + "score": 1.0, + "content": "Each pane shows samples generated after training a model on CIFAR10 with 0, 1, 5, and 10 steps of", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 403, + 149, + 417 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 149, + 417 + ], + "score": 1.0, + "content": "unrolling.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11.5 + } + ], + "index": 9.75 + }, + { + "type": "text", + "bbox": [ + 107, + 439, + 505, + 505 + ], + "lines": [ + { + "bbox": [ + 106, + 439, + 505, + 451 + ], + "spans": [ + { + "bbox": [ + 106, + 439, + 505, + 451 + ], + "score": 1.0, + "content": "In order to recover the color the GAN assigned to the digit, we used k-means with 2 clusters, to", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 450, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 450, + 505, + 462 + ], + "score": 1.0, + "content": "pick out the foreground color from the background. We then performed this transformation for both", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 461, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 461, + 505, + 473 + ], + "score": 1.0, + "content": "the training data and the generated images. Next we fit a Gaussian kernel density estimator to both", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 471, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 105, + 471, + 505, + 484 + ], + "score": 1.0, + "content": "distributions over digit colors. Finally, we computed the JS divergence between the model and data", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 483, + 505, + 495 + ], + "spans": [ + { + "bbox": [ + 106, + 483, + 505, + 495 + ], + "score": 1.0, + "content": "distributions over colors. Results can be found in Table 2 for several model sizes. Details of the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 493, + 253, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 493, + 253, + 506 + ], + "score": 1.0, + "content": "models are provided in Appendix F.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 439, + 505, + 506 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 511, + 505, + 555 + ], + "lines": [ + { + "bbox": [ + 106, + 511, + 506, + 523 + ], + "spans": [ + { + "bbox": [ + 106, + 511, + 506, + 523 + ], + "score": 1.0, + "content": "In general, the best performing models are unrolled for 5-10 steps, and larger models perform better", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 521, + 506, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 506, + 534 + ], + "score": 1.0, + "content": "than smaller models. Counter-intuitively, taking 1 unrolling step seems to hurt this measure of", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 532, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 505, + 546 + ], + "score": 1.0, + "content": "diversity. We suspect that this is due to it introducing oscillatory dynamics into training. Taking", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 543, + 415, + 557 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 415, + 557 + ], + "score": 1.0, + "content": "more unrolling steps however leads to improved performance with unrolling.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 511, + 506, + 557 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 572, + 267, + 583 + ], + "lines": [ + { + "bbox": [ + 106, + 572, + 269, + 585 + ], + "spans": [ + { + "bbox": [ + 106, + 572, + 269, + 585 + ], + "score": 1.0, + "content": "3.4 IMAGE MODELING OF CIFAR10", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 106, + 594, + 505, + 693 + ], + "lines": [ + { + "bbox": [ + 106, + 594, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 106, + 594, + 505, + 606 + ], + "score": 1.0, + "content": "Here we test our technique on a more traditional convolutional GAN architecture and task, similar", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "to those used in (Radford et al., 2015; Salimans et al., 2016). In the previous experiments we tested", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 615, + 506, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 506, + 629 + ], + "score": 1.0, + "content": "models where the standard GAN training algorithm would not converge. In this section we improve", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 627, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 506, + 640 + ], + "score": 1.0, + "content": "a standard model by reducing its tendency to engage in mode collapse. We ran 4 configurations of", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 638, + 506, + 651 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 506, + 651 + ], + "score": 1.0, + "content": "this model, varying the number of unrolling steps to be 0, 1, 5, or 10. Each configuration was run 5", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 648, + 506, + 662 + ], + "spans": [ + { + "bbox": [ + 106, + 648, + 506, + 662 + ], + "score": 1.0, + "content": "times with different random seeds. For full training details see Appendix D. Samples from each of", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 659, + 506, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 659, + 506, + 673 + ], + "score": 1.0, + "content": "the 4 configurations can be found in Figure 4. There is no obvious difference in visual quality across", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 670, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 670, + 505, + 684 + ], + "score": 1.0, + "content": "these model configurations. Visual inspection however provides only a poor measure of sample", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 681, + 145, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 145, + 695 + ], + "score": 1.0, + "content": "diversity.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 594, + 506, + 695 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "By training with an unrolled discriminator, we expect to generate more diverse samples which more", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "closely resemble the underlying data distribution. We introduce two techniques to examine sample", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 720, + 399, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 399, + 733 + ], + "score": 1.0, + "content": "diversity: inference via optimization, and pairwise distance distributions.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35, + "bbox_fs": [ + 105, + 698, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 104, + 81, + 508, + 116 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 104, + 81, + 508, + 116 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 104, + 81, + 508, + 116 + ], + "spans": [ + { + "bbox": [ + 104, + 81, + 508, + 116 + ], + "score": 0.962, + "html": "
Unrolling Steps0 steps1 step5 steps10 steps
Average MSE0.0231± 0.00240.0195 ± 0.00210.0200± 0.00230.0181± 0.0018
PercentBestRank0.63%22.97%15.31%61.09 %
", + "type": "table", + "image_path": "b50d528ebc1d7e4f4ff566671309621016069f0eaa4be9ec232cad39b3046c6b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 104, + 81, + 508, + 92.66666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 104, + 92.66666666666667, + 508, + 104.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 104, + 104.33333333333334, + 508, + 116.00000000000001 + ], + "spans": [], + "index": 2 + } + ] + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 124, + 506, + 202 + ], + "lines": [ + { + "bbox": [ + 105, + 124, + 505, + 137 + ], + "spans": [ + { + "bbox": [ + 105, + 124, + 505, + 137 + ], + "score": 1.0, + "content": "Table 3: GANs trained with unrolling are better able to match images in the training set than", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 136, + 506, + 148 + ], + "spans": [ + { + "bbox": [ + 105, + 136, + 506, + 148 + ], + "score": 1.0, + "content": "standard GANs, likely due to mode dropping by the standard GAN. Results show the MSE between", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 146, + 505, + 159 + ], + "spans": [ + { + "bbox": [ + 105, + 146, + 505, + 159 + ], + "score": 1.0, + "content": "training images and the best reconstruction for a model with the given number of unrolling steps.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 157, + 505, + 171 + ], + "spans": [ + { + "bbox": [ + 105, + 157, + 505, + 171 + ], + "score": 1.0, + "content": "The fraction of training images best reconstructed by a given model is given in the final column.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 167, + 506, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 167, + 405, + 182 + ], + "score": 1.0, + "content": "The best reconstructions is found by optimizing the latent representation", + "type": "text" + }, + { + "bbox": [ + 405, + 171, + 412, + 178 + ], + "score": 0.73, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 167, + 506, + 182 + ], + "score": 1.0, + "content": "to produce the closest", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 180, + 506, + 193 + ], + "spans": [ + { + "bbox": [ + 105, + 180, + 197, + 193 + ], + "score": 1.0, + "content": "matching pixel output", + "type": "text" + }, + { + "bbox": [ + 198, + 180, + 236, + 192 + ], + "score": 0.94, + "content": "G \\left( z ; \\theta _ { G } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 180, + 506, + 193 + ], + "score": 1.0, + "content": ". Results are averaged over all 5 runs of each model with different", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 191, + 166, + 202 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 166, + 202 + ], + "score": 1.0, + "content": "random seeds.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6 + }, + { + "type": "title", + "bbox": [ + 108, + 220, + 272, + 231 + ], + "lines": [ + { + "bbox": [ + 106, + 219, + 274, + 232 + ], + "spans": [ + { + "bbox": [ + 106, + 219, + 274, + 232 + ], + "score": 1.0, + "content": "3.4.1 INFERENCE VIA OPTIMIZATION", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 107, + 238, + 505, + 306 + ], + "lines": [ + { + "bbox": [ + 106, + 237, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 505, + 253 + ], + "score": 1.0, + "content": "Since likelihood cannot be tractably computed, over-fitting of GANs is typically tested by taking", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 250, + 505, + 262 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 505, + 262 + ], + "score": 1.0, + "content": "samples and computing the nearest-neighbor images in pixel space from the training data (Goodfel-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 260, + 505, + 274 + ], + "spans": [ + { + "bbox": [ + 105, + 260, + 505, + 274 + ], + "score": 1.0, + "content": "low et al., 2014). We will do the reverse, and measure the ability of the generative model to generate", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 272, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 272, + 505, + 284 + ], + "score": 1.0, + "content": "images that look like specific samples from the training data. If we did this by generating random", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 283, + 505, + 296 + ], + "spans": [ + { + "bbox": [ + 106, + 283, + 505, + 296 + ], + "score": 1.0, + "content": "samples from the model, we would need an exponentially large number of samples. We instead treat", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 293, + 439, + 308 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 221, + 308 + ], + "score": 1.0, + "content": "finding the nearest neighbor", + "type": "text" + }, + { + "bbox": [ + 221, + 295, + 248, + 305 + ], + "score": 0.86, + "content": "x _ { \\mathrm { n e a r e s t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 293, + 318, + 308 + ], + "score": 1.0, + "content": "to a target image", + "type": "text" + }, + { + "bbox": [ + 318, + 295, + 341, + 306 + ], + "score": 0.84, + "content": "x _ { \\mathrm { { t a r g e t } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 293, + 439, + 308 + ], + "score": 1.0, + "content": "as an optimization task,", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 13.5 + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 310, + 387, + 349 + ], + "lines": [ + { + "bbox": [ + 225, + 310, + 387, + 349 + ], + "spans": [ + { + "bbox": [ + 225, + 310, + 387, + 349 + ], + "score": 0.92, + "content": "\\begin{array} { r l } & { z _ { \\mathrm { n e a r e s t } } = \\underset { z } { \\mathrm { a r g m i n } } | | G ( z ; \\theta _ { G } ) - x _ { \\mathrm { t a r g e t } } | \\rvert _ { 2 } ^ { 2 } } \\\\ & { x _ { \\mathrm { n e a r e s t } } = G ( z _ { \\mathrm { n e a r e s t } } ; \\theta _ { G } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "8ebe230db022f5349155b2c7b8f508d347322a1d01e98e07a32852a61e22205d.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 225, + 310, + 387, + 323.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 225, + 323.0, + 387, + 336.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 225, + 336.0, + 387, + 349.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 357, + 505, + 390 + ], + "lines": [ + { + "bbox": [ + 106, + 357, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 106, + 357, + 505, + 370 + ], + "score": 1.0, + "content": "This concept of backpropagating to generate images has been widely used in visualizing features", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 368, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 368, + 505, + 380 + ], + "score": 1.0, + "content": "from discriminative networks (Simonyan et al., 2013; Yosinski et al., 2015; Nguyen et al., 2016) and", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 379, + 421, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 379, + 421, + 392 + ], + "score": 1.0, + "content": "has been applied to explore the visual manifold of GANs in (Zhu et al., 2016).", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 396, + 505, + 451 + ], + "lines": [ + { + "bbox": [ + 105, + 395, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 505, + 410 + ], + "score": 1.0, + "content": "We apply this technique to each of the models trained. We optimize with 3 random starts using", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 406, + 505, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 505, + 419 + ], + "score": 1.0, + "content": "LBFGS, which is the optimizer typically used in similar settings such as style transfer (Johnson", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 416, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 104, + 416, + 506, + 433 + ], + "score": 1.0, + "content": "et al., 2016; Champandard, 2016). Results comparing average mean squared errors between xnearest", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 428, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 123, + 442 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 430, + 147, + 441 + ], + "score": 0.83, + "content": "x _ { \\mathrm { { t a r g e t } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 428, + 505, + 442 + ], + "score": 1.0, + "content": "in pixel space can be found in Table 3. In addition we compute the percent of images for", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 439, + 498, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 498, + 453 + ], + "score": 1.0, + "content": "which a certain configuration achieves the lowest loss when compared to the other configurations.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 456, + 505, + 501 + ], + "lines": [ + { + "bbox": [ + 105, + 456, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 367, + 468 + ], + "score": 1.0, + "content": "In the zero step case, there is poor reconstruction and less than", + "type": "text" + }, + { + "bbox": [ + 367, + 457, + 382, + 467 + ], + "score": 0.83, + "content": "1 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 456, + 505, + 468 + ], + "score": 1.0, + "content": "of the time does it obtain the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 467, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 505, + 480 + ], + "score": 1.0, + "content": "lowest error of the 4 configurations. Taking 1 unrolling step results in a significant improvement in", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 478, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 505, + 491 + ], + "score": 1.0, + "content": "MSE. Taking 10 unrolling steps results in more modest improvement, but continues to reduce the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 491, + 191, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 191, + 500 + ], + "score": 1.0, + "content": "reconstruction MSE.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 107, + 506, + 505, + 613 + ], + "lines": [ + { + "bbox": [ + 105, + 505, + 505, + 520 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 505, + 520 + ], + "score": 1.0, + "content": "To visually see this, we compare the result of the optimization process for 0, 1, 5, and 10 step", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 517, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 106, + 517, + 505, + 531 + ], + "score": 1.0, + "content": "configurations in Figure 5. To select for images where differences in behavior is most apparent,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 528, + 506, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 506, + 542 + ], + "score": 1.0, + "content": "we sort the data by the absolute value of a fractional difference in MSE between the 0 and 10 step", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 540, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 141, + 555 + ], + "score": 1.0, + "content": "models,", + "type": "text" + }, + { + "bbox": [ + 142, + 540, + 212, + 559 + ], + "score": 0.92, + "content": "\\left| \\frac { l _ { 0 s t e p } - l _ { 1 0 s t e p } } { \\frac { 1 } { 2 } ( l _ { 0 s t e p } + l _ { 1 0 s t e p } ) } \\right|", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 543, + 506, + 555 + ], + "score": 1.0, + "content": "This highlights examples where either the 0 or 10 step model cannot", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 557, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 557, + 505, + 569 + ], + "score": 1.0, + "content": "accurately fit the data example but the other can. In Appendix G we show the same comparison for", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 568, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 505, + 582 + ], + "score": 1.0, + "content": "models initialized using different random seeds. Many of the zero step images are fuzzy and ill-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 579, + 505, + 592 + ], + "spans": [ + { + "bbox": [ + 106, + 579, + 505, + 592 + ], + "score": 1.0, + "content": "defined suggesting that these images cannot be generated by the standard GAN generative model,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 590, + 506, + 603 + ], + "spans": [ + { + "bbox": [ + 106, + 590, + 506, + 603 + ], + "score": 1.0, + "content": "and come from a dropped mode. As more unrolling steps are added, the outlines become more clear", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 601, + 495, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 495, + 614 + ], + "score": 1.0, + "content": "and well defined – the model covers more of the distribution and thus can recreate these samples.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 36 + }, + { + "type": "title", + "bbox": [ + 108, + 624, + 233, + 636 + ], + "lines": [ + { + "bbox": [ + 105, + 623, + 234, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 623, + 234, + 637 + ], + "score": 1.0, + "content": "3.4.2 PAIRWISE DISTANCES", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 107, + 643, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "score": 1.0, + "content": "A second complementary approach is to compare statistics of data samples to the corresponding", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "statistics for samples generated by the various models. One particularly simple and relevant statistic", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "is the distribution over pairwise distances between random pairs of samples. In the case of mode", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "collapse, greater probability mass will be concentrated in smaller volumes, and the distribution", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 687, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 701 + ], + "score": 1.0, + "content": "over inter-sample distances should be skewed towards smaller distances. We sample random pairs", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 699, + 504, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 493, + 711 + ], + "score": 1.0, + "content": "of images from each model, as well as from the training data, and compute histograms of the", + "type": "text" + }, + { + "bbox": [ + 494, + 699, + 504, + 710 + ], + "score": 0.85, + "content": "\\ell _ { 2 }", + "type": "inline_equation" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "distances between those sample pairs. As illustrated in Figure 6, the standard GAN, with zero", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 721, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 362, + 733 + ], + "score": 1.0, + "content": "unrolling steps, has its probability mass skewed towards smaller", + "type": "text" + }, + { + "bbox": [ + 362, + 721, + 372, + 732 + ], + "score": 0.87, + "content": "\\ell _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 721, + 505, + 733 + ], + "score": 1.0, + "content": "intersample distances, compared", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 45.5 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 104, + 81, + 508, + 116 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 104, + 81, + 508, + 116 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 104, + 81, + 508, + 116 + ], + "spans": [ + { + "bbox": [ + 104, + 81, + 508, + 116 + ], + "score": 0.962, + "html": "
Unrolling Steps0 steps1 step5 steps10 steps
Average MSE0.0231± 0.00240.0195 ± 0.00210.0200± 0.00230.0181± 0.0018
PercentBestRank0.63%22.97%15.31%61.09 %
", + "type": "table", + "image_path": "b50d528ebc1d7e4f4ff566671309621016069f0eaa4be9ec232cad39b3046c6b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 104, + 81, + 508, + 92.66666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 104, + 92.66666666666667, + 508, + 104.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 104, + 104.33333333333334, + 508, + 116.00000000000001 + ], + "spans": [], + "index": 2 + } + ] + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 124, + 506, + 202 + ], + "lines": [ + { + "bbox": [ + 105, + 124, + 505, + 137 + ], + "spans": [ + { + "bbox": [ + 105, + 124, + 505, + 137 + ], + "score": 1.0, + "content": "Table 3: GANs trained with unrolling are better able to match images in the training set than", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 136, + 506, + 148 + ], + "spans": [ + { + "bbox": [ + 105, + 136, + 506, + 148 + ], + "score": 1.0, + "content": "standard GANs, likely due to mode dropping by the standard GAN. Results show the MSE between", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 146, + 505, + 159 + ], + "spans": [ + { + "bbox": [ + 105, + 146, + 505, + 159 + ], + "score": 1.0, + "content": "training images and the best reconstruction for a model with the given number of unrolling steps.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 157, + 505, + 171 + ], + "spans": [ + { + "bbox": [ + 105, + 157, + 505, + 171 + ], + "score": 1.0, + "content": "The fraction of training images best reconstructed by a given model is given in the final column.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 167, + 506, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 167, + 405, + 182 + ], + "score": 1.0, + "content": "The best reconstructions is found by optimizing the latent representation", + "type": "text" + }, + { + "bbox": [ + 405, + 171, + 412, + 178 + ], + "score": 0.73, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 167, + 506, + 182 + ], + "score": 1.0, + "content": "to produce the closest", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 180, + 506, + 193 + ], + "spans": [ + { + "bbox": [ + 105, + 180, + 197, + 193 + ], + "score": 1.0, + "content": "matching pixel output", + "type": "text" + }, + { + "bbox": [ + 198, + 180, + 236, + 192 + ], + "score": 0.94, + "content": "G \\left( z ; \\theta _ { G } \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 180, + 506, + 193 + ], + "score": 1.0, + "content": ". Results are averaged over all 5 runs of each model with different", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 191, + 166, + 202 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 166, + 202 + ], + "score": 1.0, + "content": "random seeds.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6, + "bbox_fs": [ + 105, + 124, + 506, + 202 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 220, + 272, + 231 + ], + "lines": [ + { + "bbox": [ + 106, + 219, + 274, + 232 + ], + "spans": [ + { + "bbox": [ + 106, + 219, + 274, + 232 + ], + "score": 1.0, + "content": "3.4.1 INFERENCE VIA OPTIMIZATION", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 107, + 238, + 505, + 306 + ], + "lines": [ + { + "bbox": [ + 106, + 237, + 505, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 505, + 253 + ], + "score": 1.0, + "content": "Since likelihood cannot be tractably computed, over-fitting of GANs is typically tested by taking", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 250, + 505, + 262 + ], + "spans": [ + { + "bbox": [ + 105, + 250, + 505, + 262 + ], + "score": 1.0, + "content": "samples and computing the nearest-neighbor images in pixel space from the training data (Goodfel-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 260, + 505, + 274 + ], + "spans": [ + { + "bbox": [ + 105, + 260, + 505, + 274 + ], + "score": 1.0, + "content": "low et al., 2014). We will do the reverse, and measure the ability of the generative model to generate", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 272, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 272, + 505, + 284 + ], + "score": 1.0, + "content": "images that look like specific samples from the training data. If we did this by generating random", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 283, + 505, + 296 + ], + "spans": [ + { + "bbox": [ + 106, + 283, + 505, + 296 + ], + "score": 1.0, + "content": "samples from the model, we would need an exponentially large number of samples. We instead treat", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 293, + 439, + 308 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 221, + 308 + ], + "score": 1.0, + "content": "finding the nearest neighbor", + "type": "text" + }, + { + "bbox": [ + 221, + 295, + 248, + 305 + ], + "score": 0.86, + "content": "x _ { \\mathrm { n e a r e s t } }", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 293, + 318, + 308 + ], + "score": 1.0, + "content": "to a target image", + "type": "text" + }, + { + "bbox": [ + 318, + 295, + 341, + 306 + ], + "score": 0.84, + "content": "x _ { \\mathrm { { t a r g e t } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 342, + 293, + 439, + 308 + ], + "score": 1.0, + "content": "as an optimization task,", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 13.5, + "bbox_fs": [ + 105, + 237, + 505, + 308 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 310, + 387, + 349 + ], + "lines": [ + { + "bbox": [ + 225, + 310, + 387, + 349 + ], + "spans": [ + { + "bbox": [ + 225, + 310, + 387, + 349 + ], + "score": 0.92, + "content": "\\begin{array} { r l } & { z _ { \\mathrm { n e a r e s t } } = \\underset { z } { \\mathrm { a r g m i n } } | | G ( z ; \\theta _ { G } ) - x _ { \\mathrm { t a r g e t } } | \\rvert _ { 2 } ^ { 2 } } \\\\ & { x _ { \\mathrm { n e a r e s t } } = G ( z _ { \\mathrm { n e a r e s t } } ; \\theta _ { G } ) . } \\end{array}", + "type": "interline_equation", + "image_path": "8ebe230db022f5349155b2c7b8f508d347322a1d01e98e07a32852a61e22205d.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 225, + 310, + 387, + 323.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 225, + 323.0, + 387, + 336.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 225, + 336.0, + 387, + 349.0 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 357, + 505, + 390 + ], + "lines": [ + { + "bbox": [ + 106, + 357, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 106, + 357, + 505, + 370 + ], + "score": 1.0, + "content": "This concept of backpropagating to generate images has been widely used in visualizing features", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 368, + 505, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 368, + 505, + 380 + ], + "score": 1.0, + "content": "from discriminative networks (Simonyan et al., 2013; Yosinski et al., 2015; Nguyen et al., 2016) and", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 379, + 421, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 379, + 421, + 392 + ], + "score": 1.0, + "content": "has been applied to explore the visual manifold of GANs in (Zhu et al., 2016).", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21, + "bbox_fs": [ + 106, + 357, + 505, + 392 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 396, + 505, + 451 + ], + "lines": [ + { + "bbox": [ + 105, + 395, + 505, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 505, + 410 + ], + "score": 1.0, + "content": "We apply this technique to each of the models trained. We optimize with 3 random starts using", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 406, + 505, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 505, + 419 + ], + "score": 1.0, + "content": "LBFGS, which is the optimizer typically used in similar settings such as style transfer (Johnson", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 416, + 506, + 433 + ], + "spans": [ + { + "bbox": [ + 104, + 416, + 506, + 433 + ], + "score": 1.0, + "content": "et al., 2016; Champandard, 2016). Results comparing average mean squared errors between xnearest", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 428, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 428, + 123, + 442 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 430, + 147, + 441 + ], + "score": 0.83, + "content": "x _ { \\mathrm { { t a r g e t } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 147, + 428, + 505, + 442 + ], + "score": 1.0, + "content": "in pixel space can be found in Table 3. In addition we compute the percent of images for", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 439, + 498, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 498, + 453 + ], + "score": 1.0, + "content": "which a certain configuration achieves the lowest loss when compared to the other configurations.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25, + "bbox_fs": [ + 104, + 395, + 506, + 453 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 456, + 505, + 501 + ], + "lines": [ + { + "bbox": [ + 105, + 456, + 505, + 468 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 367, + 468 + ], + "score": 1.0, + "content": "In the zero step case, there is poor reconstruction and less than", + "type": "text" + }, + { + "bbox": [ + 367, + 457, + 382, + 467 + ], + "score": 0.83, + "content": "1 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 456, + 505, + 468 + ], + "score": 1.0, + "content": "of the time does it obtain the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 467, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 505, + 480 + ], + "score": 1.0, + "content": "lowest error of the 4 configurations. Taking 1 unrolling step results in a significant improvement in", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 478, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 505, + 491 + ], + "score": 1.0, + "content": "MSE. Taking 10 unrolling steps results in more modest improvement, but continues to reduce the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 491, + 191, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 191, + 500 + ], + "score": 1.0, + "content": "reconstruction MSE.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 456, + 505, + 500 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 506, + 505, + 613 + ], + "lines": [ + { + "bbox": [ + 105, + 505, + 505, + 520 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 505, + 520 + ], + "score": 1.0, + "content": "To visually see this, we compare the result of the optimization process for 0, 1, 5, and 10 step", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 517, + 505, + 531 + ], + "spans": [ + { + "bbox": [ + 106, + 517, + 505, + 531 + ], + "score": 1.0, + "content": "configurations in Figure 5. To select for images where differences in behavior is most apparent,", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 528, + 506, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 506, + 542 + ], + "score": 1.0, + "content": "we sort the data by the absolute value of a fractional difference in MSE between the 0 and 10 step", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 540, + 506, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 543, + 141, + 555 + ], + "score": 1.0, + "content": "models,", + "type": "text" + }, + { + "bbox": [ + 142, + 540, + 212, + 559 + ], + "score": 0.92, + "content": "\\left| \\frac { l _ { 0 s t e p } - l _ { 1 0 s t e p } } { \\frac { 1 } { 2 } ( l _ { 0 s t e p } + l _ { 1 0 s t e p } ) } \\right|", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 543, + 506, + 555 + ], + "score": 1.0, + "content": "This highlights examples where either the 0 or 10 step model cannot", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 557, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 557, + 505, + 569 + ], + "score": 1.0, + "content": "accurately fit the data example but the other can. In Appendix G we show the same comparison for", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 568, + 505, + 582 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 505, + 582 + ], + "score": 1.0, + "content": "models initialized using different random seeds. Many of the zero step images are fuzzy and ill-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 579, + 505, + 592 + ], + "spans": [ + { + "bbox": [ + 106, + 579, + 505, + 592 + ], + "score": 1.0, + "content": "defined suggesting that these images cannot be generated by the standard GAN generative model,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 590, + 506, + 603 + ], + "spans": [ + { + "bbox": [ + 106, + 590, + 506, + 603 + ], + "score": 1.0, + "content": "and come from a dropped mode. As more unrolling steps are added, the outlines become more clear", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 601, + 495, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 601, + 495, + 614 + ], + "score": 1.0, + "content": "and well defined – the model covers more of the distribution and thus can recreate these samples.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 36, + "bbox_fs": [ + 105, + 505, + 506, + 614 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 624, + 233, + 636 + ], + "lines": [ + { + "bbox": [ + 105, + 623, + 234, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 623, + 234, + 637 + ], + "score": 1.0, + "content": "3.4.2 PAIRWISE DISTANCES", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 107, + 643, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 644, + 505, + 657 + ], + "score": 1.0, + "content": "A second complementary approach is to compare statistics of data samples to the corresponding", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 505, + 667 + ], + "score": 1.0, + "content": "statistics for samples generated by the various models. One particularly simple and relevant statistic", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 678 + ], + "score": 1.0, + "content": "is the distribution over pairwise distances between random pairs of samples. In the case of mode", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "score": 1.0, + "content": "collapse, greater probability mass will be concentrated in smaller volumes, and the distribution", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 687, + 505, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 701 + ], + "score": 1.0, + "content": "over inter-sample distances should be skewed towards smaller distances. We sample random pairs", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 699, + 504, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 493, + 711 + ], + "score": 1.0, + "content": "of images from each model, as well as from the training data, and compute histograms of the", + "type": "text" + }, + { + "bbox": [ + 494, + 699, + 504, + 710 + ], + "score": 0.85, + "content": "\\ell _ { 2 }", + "type": "inline_equation" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "distances between those sample pairs. As illustrated in Figure 6, the standard GAN, with zero", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 721, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 362, + 733 + ], + "score": 1.0, + "content": "unrolling steps, has its probability mass skewed towards smaller", + "type": "text" + }, + { + "bbox": [ + 362, + 721, + 372, + 732 + ], + "score": 0.87, + "content": "\\ell _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 721, + 505, + 733 + ], + "score": 1.0, + "content": "intersample distances, compared", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 389, + 505, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 505, + 402 + ], + "score": 1.0, + "content": "to real data. As the number of unrolling steps is increased, the histograms over intersample distances", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 401, + 506, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 401, + 506, + 414 + ], + "score": 1.0, + "content": "increasingly come to resemble that for the data distribution. This is further evidence in support of", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 413, + 344, + 424 + ], + "spans": [ + { + "bbox": [ + 106, + 413, + 344, + 424 + ], + "score": 1.0, + "content": "unrolling decreasing the mode collapse behavior of GANs.", + "type": "text", + "cross_page": true + } + ], + "index": 12 + } + ], + "index": 45.5, + "bbox_fs": [ + 105, + 644, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 81, + 510, + 279 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 81, + 510, + 279 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 81, + 510, + 279 + ], + "spans": [ + { + "bbox": [ + 108, + 81, + 510, + 279 + ], + "score": 0.971, + "type": "image", + "image_path": "3f34af532391d651dea1658c3e0b24559b5a1f84c026a63aecddf941edff8580.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 108, + 81, + 510, + 147.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 147.0, + 510, + 213.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 213.0, + 510, + 279.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 291, + 505, + 369 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 291, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 106, + 291, + 505, + 303 + ], + "score": 1.0, + "content": "Figure 5: Training set images are more accurately reconstructed using GANs trained with unrolling", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 302, + 506, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 302, + 506, + 315 + ], + "score": 1.0, + "content": "than by a standard (0 step) GAN, likely due to mode dropping by the standard GAN. Raw data is", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 312, + 505, + 327 + ], + "spans": [ + { + "bbox": [ + 104, + 312, + 505, + 327 + ], + "score": 1.0, + "content": "on the left, and the optimized images to reach this target follow for 0, 1, 5, and 10 unrolling steps.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 324, + 505, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 505, + 336 + ], + "score": 1.0, + "content": "The reconstruction MSE is listed below each sample. A random 1280 images where selected from", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 335, + 505, + 347 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 505, + 347 + ], + "score": 1.0, + "content": "the training set, and corresponding best reconstructions for each model were found via optimiza-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 347, + 505, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 347, + 505, + 358 + ], + "score": 1.0, + "content": "tion. Shown here are the eight images with the largest absolute fractional difference between GANs", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 356, + 258, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 258, + 370 + ], + "score": 1.0, + "content": "trained with 0 and 10 unrolling steps.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 107, + 389, + 504, + 423 + ], + "lines": [ + { + "bbox": [ + 105, + 389, + 505, + 402 + ], + "spans": [ + { + "bbox": [ + 105, + 389, + 505, + 402 + ], + "score": 1.0, + "content": "to real data. As the number of unrolling steps is increased, the histograms over intersample distances", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 401, + 506, + 414 + ], + "spans": [ + { + "bbox": [ + 106, + 401, + 506, + 414 + ], + "score": 1.0, + "content": "increasingly come to resemble that for the data distribution. This is further evidence in support of", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 413, + 344, + 424 + ], + "spans": [ + { + "bbox": [ + 106, + 413, + 344, + 424 + ], + "score": 1.0, + "content": "unrolling decreasing the mode collapse behavior of GANs.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11 + }, + { + "type": "title", + "bbox": [ + 108, + 441, + 190, + 453 + ], + "lines": [ + { + "bbox": [ + 105, + 439, + 192, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 192, + 456 + ], + "score": 1.0, + "content": "4 DISCUSSION", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 466, + 505, + 510 + ], + "lines": [ + { + "bbox": [ + 105, + 465, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 505, + 479 + ], + "score": 1.0, + "content": "In this work we developed a method to stabilize GAN training and reduce mode collapse by defining", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 478, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 505, + 489 + ], + "score": 1.0, + "content": "the generator objective with respect to unrolled optimization of the discriminator. We then demon-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 488, + 505, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 505, + 501 + ], + "score": 1.0, + "content": "strated the application of this method to several tasks, where it either rescued unstable training, or", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 500, + 404, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 404, + 511 + ], + "score": 1.0, + "content": "reduced the tendency of the model to drop regions of the data distribution.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 107, + 516, + 505, + 594 + ], + "lines": [ + { + "bbox": [ + 105, + 515, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 505, + 529 + ], + "score": 1.0, + "content": "The main drawback to this method is computational cost of each training step, which increases", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 527, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 505, + 540 + ], + "score": 1.0, + "content": "linearly with the number of unrolling steps. There is a tradeoff between better approximating the true", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 538, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 506, + 552 + ], + "score": 1.0, + "content": "generator loss and the computation required to make this estimate. Depending on the architecture,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 549, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 506, + 563 + ], + "score": 1.0, + "content": "one unrolling step can be enough. In other more unstable models, such as the RNN case, more are", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 560, + 505, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 505, + 573 + ], + "score": 1.0, + "content": "needed to stabilize training. We have some initial positive results suggesting it may be sufficient", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 572, + 505, + 584 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 505, + 584 + ], + "score": 1.0, + "content": "to further perturb the training gradient in the same direction that a single unrolling step perturbs it.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 582, + 417, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 417, + 595 + ], + "score": 1.0, + "content": "While this is more computationally efficient, further investigation is required.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 599, + 505, + 676 + ], + "lines": [ + { + "bbox": [ + 106, + 599, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 505, + 612 + ], + "score": 1.0, + "content": "The method presented here bridges some of the gap between theoretical and practical results for", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 609, + 506, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 506, + 623 + ], + "score": 1.0, + "content": "training of GANs. We believe developing better update rules for the generator and discriminator is", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 621, + 505, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 505, + 633 + ], + "score": 1.0, + "content": "an important line of work for GAN training. In this work we have only considered a small fraction", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 632, + 504, + 644 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 420, + 644 + ], + "score": 1.0, + "content": "of the design space. For instance, the approach could be extended to unroll", + "type": "text" + }, + { + "bbox": [ + 420, + 632, + 429, + 642 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 632, + 494, + 644 + ], + "score": 1.0, + "content": "when updating", + "type": "text" + }, + { + "bbox": [ + 494, + 632, + 504, + 642 + ], + "score": 0.76, + "content": "D", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 643, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 505, + 655 + ], + "score": 1.0, + "content": "as well – letting the discriminator react to how the generator would move. It is also possible to", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 654, + 506, + 666 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 186, + 666 + ], + "score": 1.0, + "content": "unroll sequences of", + "type": "text" + }, + { + "bbox": [ + 186, + 654, + 195, + 664 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 654, + 213, + 666 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 213, + 654, + 223, + 664 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 654, + 438, + 666 + ], + "score": 1.0, + "content": "updates. This would make updates that are recursive:", + "type": "text" + }, + { + "bbox": [ + 438, + 654, + 447, + 664 + ], + "score": 0.77, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 654, + 506, + 666 + ], + "score": 1.0, + "content": "could react to", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 665, + 343, + 677 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 221, + 677 + ], + "score": 1.0, + "content": "maximize performance as if", + "type": "text" + }, + { + "bbox": [ + 221, + 666, + 230, + 675 + ], + "score": 0.85, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 665, + 248, + 677 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 248, + 666, + 257, + 675 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 665, + 343, + 677 + ], + "score": 1.0, + "content": "had already updated.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28 + }, + { + "type": "title", + "bbox": [ + 108, + 690, + 200, + 700 + ], + "lines": [ + { + "bbox": [ + 107, + 691, + 200, + 702 + ], + "spans": [ + { + "bbox": [ + 107, + 691, + 200, + 702 + ], + "score": 1.0, + "content": "ACKNOWLEDGMENTS", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "We would like to thank Laurent Dinh, David Dohan, Vincent Dumoulin, Liam Fedus, Ishaan Gul-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "score": 1.0, + "content": "rajani, Julian Ibarz, Eric Jang, Matthew Johnson, Marc Lanctot, Augustus Odena, Gabriel Pereyra,", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 301, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 81, + 510, + 279 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 81, + 510, + 279 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 81, + 510, + 279 + ], + "spans": [ + { + "bbox": [ + 108, + 81, + 510, + 279 + ], + "score": 0.971, + "type": "image", + "image_path": "3f34af532391d651dea1658c3e0b24559b5a1f84c026a63aecddf941edff8580.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 108, + 81, + 510, + 147.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 147.0, + 510, + 213.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 213.0, + 510, + 279.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 291, + 505, + 369 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 291, + 505, + 303 + ], + "spans": [ + { + "bbox": [ + 106, + 291, + 505, + 303 + ], + "score": 1.0, + "content": "Figure 5: Training set images are more accurately reconstructed using GANs trained with unrolling", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 302, + 506, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 302, + 506, + 315 + ], + "score": 1.0, + "content": "than by a standard (0 step) GAN, likely due to mode dropping by the standard GAN. Raw data is", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 312, + 505, + 327 + ], + "spans": [ + { + "bbox": [ + 104, + 312, + 505, + 327 + ], + "score": 1.0, + "content": "on the left, and the optimized images to reach this target follow for 0, 1, 5, and 10 unrolling steps.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 324, + 505, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 505, + 336 + ], + "score": 1.0, + "content": "The reconstruction MSE is listed below each sample. A random 1280 images where selected from", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 335, + 505, + 347 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 505, + 347 + ], + "score": 1.0, + "content": "the training set, and corresponding best reconstructions for each model were found via optimiza-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 347, + 505, + 358 + ], + "spans": [ + { + "bbox": [ + 106, + 347, + 505, + 358 + ], + "score": 1.0, + "content": "tion. Shown here are the eight images with the largest absolute fractional difference between GANs", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 356, + 258, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 258, + 370 + ], + "score": 1.0, + "content": "trained with 0 and 10 unrolling steps.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 107, + 389, + 504, + 423 + ], + "lines": [], + "index": 11, + "bbox_fs": [ + 105, + 389, + 506, + 424 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 108, + 441, + 190, + 453 + ], + "lines": [ + { + "bbox": [ + 105, + 439, + 192, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 439, + 192, + 456 + ], + "score": 1.0, + "content": "4 DISCUSSION", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 466, + 505, + 510 + ], + "lines": [ + { + "bbox": [ + 105, + 465, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 505, + 479 + ], + "score": 1.0, + "content": "In this work we developed a method to stabilize GAN training and reduce mode collapse by defining", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 478, + 505, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 505, + 489 + ], + "score": 1.0, + "content": "the generator objective with respect to unrolled optimization of the discriminator. We then demon-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 488, + 505, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 505, + 501 + ], + "score": 1.0, + "content": "strated the application of this method to several tasks, where it either rescued unstable training, or", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 500, + 404, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 404, + 511 + ], + "score": 1.0, + "content": "reduced the tendency of the model to drop regions of the data distribution.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 465, + 505, + 511 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 516, + 505, + 594 + ], + "lines": [ + { + "bbox": [ + 105, + 515, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 505, + 529 + ], + "score": 1.0, + "content": "The main drawback to this method is computational cost of each training step, which increases", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 527, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 527, + 505, + 540 + ], + "score": 1.0, + "content": "linearly with the number of unrolling steps. There is a tradeoff between better approximating the true", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 538, + 506, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 538, + 506, + 552 + ], + "score": 1.0, + "content": "generator loss and the computation required to make this estimate. Depending on the architecture,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 549, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 549, + 506, + 563 + ], + "score": 1.0, + "content": "one unrolling step can be enough. In other more unstable models, such as the RNN case, more are", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 560, + 505, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 505, + 573 + ], + "score": 1.0, + "content": "needed to stabilize training. We have some initial positive results suggesting it may be sufficient", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 572, + 505, + 584 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 505, + 584 + ], + "score": 1.0, + "content": "to further perturb the training gradient in the same direction that a single unrolling step perturbs it.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 582, + 417, + 595 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 417, + 595 + ], + "score": 1.0, + "content": "While this is more computationally efficient, further investigation is required.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 515, + 506, + 595 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 599, + 505, + 676 + ], + "lines": [ + { + "bbox": [ + 106, + 599, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 505, + 612 + ], + "score": 1.0, + "content": "The method presented here bridges some of the gap between theoretical and practical results for", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 609, + 506, + 623 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 506, + 623 + ], + "score": 1.0, + "content": "training of GANs. We believe developing better update rules for the generator and discriminator is", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 621, + 505, + 633 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 505, + 633 + ], + "score": 1.0, + "content": "an important line of work for GAN training. In this work we have only considered a small fraction", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 632, + 504, + 644 + ], + "spans": [ + { + "bbox": [ + 105, + 632, + 420, + 644 + ], + "score": 1.0, + "content": "of the design space. For instance, the approach could be extended to unroll", + "type": "text" + }, + { + "bbox": [ + 420, + 632, + 429, + 642 + ], + "score": 0.8, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 632, + 494, + 644 + ], + "score": 1.0, + "content": "when updating", + "type": "text" + }, + { + "bbox": [ + 494, + 632, + 504, + 642 + ], + "score": 0.76, + "content": "D", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 643, + 505, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 643, + 505, + 655 + ], + "score": 1.0, + "content": "as well – letting the discriminator react to how the generator would move. It is also possible to", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 654, + 506, + 666 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 186, + 666 + ], + "score": 1.0, + "content": "unroll sequences of", + "type": "text" + }, + { + "bbox": [ + 186, + 654, + 195, + 664 + ], + "score": 0.82, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 654, + 213, + 666 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 213, + 654, + 223, + 664 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 654, + 438, + 666 + ], + "score": 1.0, + "content": "updates. This would make updates that are recursive:", + "type": "text" + }, + { + "bbox": [ + 438, + 654, + 447, + 664 + ], + "score": 0.77, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 654, + 506, + 666 + ], + "score": 1.0, + "content": "could react to", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 665, + 343, + 677 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 221, + 677 + ], + "score": 1.0, + "content": "maximize performance as if", + "type": "text" + }, + { + "bbox": [ + 221, + 666, + 230, + 675 + ], + "score": 0.85, + "content": "G", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 665, + 248, + 677 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 248, + 666, + 257, + 675 + ], + "score": 0.83, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 665, + 343, + 677 + ], + "score": 1.0, + "content": "had already updated.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 599, + 506, + 677 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 690, + 200, + 700 + ], + "lines": [ + { + "bbox": [ + 107, + 691, + 200, + 702 + ], + "spans": [ + { + "bbox": [ + 107, + 691, + 200, + 702 + ], + "score": 1.0, + "content": "ACKNOWLEDGMENTS", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 503, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "We would like to thank Laurent Dinh, David Dohan, Vincent Dumoulin, Liam Fedus, Ishaan Gul-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "score": 1.0, + "content": "rajani, Julian Ibarz, Eric Jang, Matthew Johnson, Marc Lanctot, Augustus Odena, Gabriel Pereyra,", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 709, + 505, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 147, + 82, + 462, + 343 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 147, + 82, + 462, + 343 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 147, + 82, + 462, + 343 + ], + "spans": [ + { + "bbox": [ + 147, + 82, + 462, + 343 + ], + "score": 0.972, + "type": "image", + "image_path": "34368461772c640ab500044ba570ada5248c204871939191a4efbed64c4bf4ee.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 147, + 82, + 462, + 169.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 147, + 169.0, + 462, + 256.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 147, + 256.0, + 462, + 343.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 357, + 505, + 413 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 358, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 106, + 358, + 505, + 370 + ], + "score": 1.0, + "content": "Figure 6: As the number of unrolling steps in GAN training is increased, the distribution of pairwise", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 367, + 506, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 506, + 381 + ], + "score": 1.0, + "content": "distances between model samples more closely resembles the same distribution for the data. Here", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 379, + 506, + 392 + ], + "spans": [ + { + "bbox": [ + 105, + 379, + 506, + 392 + ], + "score": 1.0, + "content": "we plot histograms of pairwise distances between randomly selected samples. The red line gives", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "score": 1.0, + "content": "pairwise distances in the data, while each of the five blue lines in each plot represents a model", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 401, + 475, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 475, + 414 + ], + "score": 1.0, + "content": "trained with a different random seed. The vertical lines are the medians of each distribution.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + } + ], + "index": 3.0 + }, + { + "type": "text", + "bbox": [ + 106, + 431, + 504, + 454 + ], + "lines": [ + { + "bbox": [ + 106, + 431, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 106, + 431, + 505, + 444 + ], + "score": 1.0, + "content": "Colin Raffel, Sam Schoenholz, Ayush Sekhari, Jon Shlens, and Dale Schuurmans for insightful", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 442, + 343, + 455 + ], + "spans": [ + { + "bbox": [ + 106, + 442, + 343, + 455 + ], + "score": 1.0, + "content": "conversation, as well as the rest of the Google Brain Team.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 147, + 82, + 462, + 343 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 147, + 82, + 462, + 343 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 147, + 82, + 462, + 343 + ], + "spans": [ + { + "bbox": [ + 147, + 82, + 462, + 343 + ], + "score": 0.972, + "type": "image", + "image_path": "34368461772c640ab500044ba570ada5248c204871939191a4efbed64c4bf4ee.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 147, + 82, + 462, + 169.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 147, + 169.0, + 462, + 256.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 147, + 256.0, + 462, + 343.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 357, + 505, + 413 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 358, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 106, + 358, + 505, + 370 + ], + "score": 1.0, + "content": "Figure 6: As the number of unrolling steps in GAN training is increased, the distribution of pairwise", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 367, + 506, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 367, + 506, + 381 + ], + "score": 1.0, + "content": "distances between model samples more closely resembles the same distribution for the data. Here", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 379, + 506, + 392 + ], + "spans": [ + { + "bbox": [ + 105, + 379, + 506, + 392 + ], + "score": 1.0, + "content": "we plot histograms of pairwise distances between randomly selected samples. The red line gives", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "spans": [ + { + "bbox": [ + 105, + 390, + 506, + 403 + ], + "score": 1.0, + "content": "pairwise distances in the data, while each of the five blue lines in each plot represents a model", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 401, + 475, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 475, + 414 + ], + "score": 1.0, + "content": "trained with a different random seed. The vertical lines are the medians of each distribution.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + } + ], + "index": 3.0 + }, + { + "type": "text", + "bbox": [ + 106, + 431, + 504, + 454 + ], + "lines": [ + { + "bbox": [ + 106, + 431, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 106, + 431, + 505, + 444 + ], + "score": 1.0, + "content": "Colin Raffel, Sam Schoenholz, Ayush Sekhari, Jon Shlens, and Dale Schuurmans for insightful", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 442, + 343, + 455 + ], + "spans": [ + { + "bbox": [ + 106, + 442, + 343, + 455 + ], + "score": 1.0, + "content": "conversation, as well as the rest of the Google Brain Team.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 106, + 431, + 505, + 455 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 82, + 176, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 176, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 176, + 95 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 100, + 504, + 133 + ], + "lines": [ + { + "bbox": [ + 105, + 97, + 506, + 114 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 506, + 114 + ], + "score": 1.0, + "content": "Guillaume Alain, Yoshua Bengio, Li Yao, Jason Yosinski, Eric Thibodeau-Laufer, Saizheng Zhang,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 111, + 505, + 123 + ], + "spans": [ + { + "bbox": [ + 115, + 111, + 505, + 123 + ], + "score": 1.0, + "content": "and Pascal Vincent. Gsns : Generative stochastic networks. arXiv preprint arXiv:1503.05571,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 121, + 142, + 133 + ], + "spans": [ + { + "bbox": [ + 115, + 121, + 142, + 133 + ], + "score": 1.0, + "content": "2015.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 106, + 140, + 504, + 174 + ], + "lines": [ + { + "bbox": [ + 106, + 140, + 505, + 152 + ], + "spans": [ + { + "bbox": [ + 106, + 140, + 505, + 152 + ], + "score": 1.0, + "content": "Marcin Andrychowicz, Misha Denil, Sergio Gomez, Matthew W Hoffman, David Pfau, Tom Schaul,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 151, + 505, + 164 + ], + "spans": [ + { + "bbox": [ + 116, + 151, + 505, + 164 + ], + "score": 1.0, + "content": "and Nando de Freitas. Learning to learn by gradient descent by gradient descent. arXiv preprint", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 162, + 219, + 174 + ], + "spans": [ + { + "bbox": [ + 115, + 162, + 219, + 174 + ], + "score": 1.0, + "content": "arXiv:1606.04474, 2016.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 105, + 180, + 504, + 203 + ], + "lines": [ + { + "bbox": [ + 105, + 179, + 505, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 179, + 505, + 194 + ], + "score": 1.0, + "content": "David Belanger and Andrew McCallum. Structured prediction energy networks. arXiv preprint", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 116, + 192, + 219, + 202 + ], + "spans": [ + { + "bbox": [ + 116, + 192, + 219, + 202 + ], + "score": 1.0, + "content": "arXiv:1511.06350, 2015.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 104, + 210, + 505, + 233 + ], + "lines": [ + { + "bbox": [ + 106, + 210, + 505, + 223 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 505, + 223 + ], + "score": 1.0, + "content": "Jorg Bornschein, Samira Shabanian, Asja Fischer, and Yoshua Bengio. Bidirectional helmholtz", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 222, + 322, + 232 + ], + "spans": [ + { + "bbox": [ + 116, + 222, + 322, + 232 + ], + "score": 1.0, + "content": "machines. arXiv preprint arXiv:1506.03877, 2015.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 105, + 239, + 505, + 262 + ], + "lines": [ + { + "bbox": [ + 105, + 237, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 506, + 253 + ], + "score": 1.0, + "content": "Michael Bowling and Manuela Veloso. Multiagent learning using a variable learning rate. Artificial", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 117, + 250, + 263, + 263 + ], + "spans": [ + { + "bbox": [ + 117, + 250, + 263, + 263 + ], + "score": 1.0, + "content": "Intelligence, 136(2):215–250, 2002.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 105, + 268, + 505, + 291 + ], + "lines": [ + { + "bbox": [ + 106, + 268, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 268, + 505, + 282 + ], + "score": 1.0, + "content": "Yuri Burda, Roger B. Grosse, and Ruslan Salakhutdinov. Importance weighted autoencoders. arXiv", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 280, + 254, + 292 + ], + "spans": [ + { + "bbox": [ + 115, + 280, + 254, + 292 + ], + "score": 1.0, + "content": "preprint arXiv:1509.00519, 2015.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 105, + 298, + 504, + 321 + ], + "lines": [ + { + "bbox": [ + 106, + 298, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 505, + 311 + ], + "score": 1.0, + "content": "Alex J. Champandard. Semantic style transfer and turning two-bit doodles into fine artworks. arXiv", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 114, + 309, + 254, + 321 + ], + "spans": [ + { + "bbox": [ + 114, + 309, + 254, + 321 + ], + "score": 1.0, + "content": "preprint arXiv:1603.01768, 2016.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 106, + 327, + 504, + 351 + ], + "lines": [ + { + "bbox": [ + 106, + 326, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 106, + 326, + 505, + 340 + ], + "score": 1.0, + "content": "Tong Che, Yanran Li, Athul Paul Jacob, Yoshua Bengio, and Wenjie Li. Mode regularized generative", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 338, + 370, + 351 + ], + "spans": [ + { + "bbox": [ + 115, + 338, + 370, + 351 + ], + "score": 1.0, + "content": "adversarial networks. arXiv preprint arXiv: 1612.02136, 2016.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 107, + 357, + 503, + 390 + ], + "lines": [ + { + "bbox": [ + 106, + 357, + 504, + 369 + ], + "spans": [ + { + "bbox": [ + 106, + 357, + 504, + 369 + ], + "score": 1.0, + "content": "Xi Chen, Yan Duan, Rein Houthooft, John Schulman, Ilya Sutskever, and Pieter Abbeel. Info-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 368, + 505, + 381 + ], + "spans": [ + { + "bbox": [ + 115, + 368, + 505, + 381 + ], + "score": 1.0, + "content": "gan: Interpretable representation learning by information maximizing generative adversarial nets.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 116, + 379, + 279, + 392 + ], + "spans": [ + { + "bbox": [ + 116, + 379, + 279, + 392 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1606.03657, 2016.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 105, + 397, + 504, + 420 + ], + "lines": [ + { + "bbox": [ + 105, + 397, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 505, + 411 + ], + "score": 1.0, + "content": "John M Danskin. The theory of max-min and its application to weapons allocation problems, vol-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 409, + 321, + 420 + ], + "spans": [ + { + "bbox": [ + 116, + 409, + 321, + 420 + ], + "score": 1.0, + "content": "ume 5. Springer Science & Business Media, 1967.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 107, + 426, + 503, + 450 + ], + "lines": [ + { + "bbox": [ + 106, + 426, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 505, + 439 + ], + "score": 1.0, + "content": "Peter Dayan, Geoffrey E Hinton, Radford M Neal, and Richard S Zemel. The helmholtz machine.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 438, + 285, + 450 + ], + "spans": [ + { + "bbox": [ + 115, + 438, + 285, + 450 + ], + "score": 1.0, + "content": "Neural computation, 7(5):889–904, 1995.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + }, + { + "type": "text", + "bbox": [ + 106, + 456, + 503, + 479 + ], + "lines": [ + { + "bbox": [ + 105, + 456, + 505, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 505, + 470 + ], + "score": 1.0, + "content": "Laurent Dinh, David Krueger, and Yoshua Bengio. NICE: non-linear independent components esti-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 468, + 308, + 479 + ], + "spans": [ + { + "bbox": [ + 115, + 468, + 308, + 479 + ], + "score": 1.0, + "content": "mation. arXiv preprint arXiv:1410.8516, 2014.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 106, + 485, + 504, + 509 + ], + "lines": [ + { + "bbox": [ + 105, + 484, + 505, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 505, + 499 + ], + "score": 1.0, + "content": "Laurent Dinh, Jascha Sohl-Dickstein, and Samy Bengio. Density estimation using real NVP. arXiv", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 497, + 254, + 508 + ], + "spans": [ + { + "bbox": [ + 115, + 497, + 254, + 508 + ], + "score": 1.0, + "content": "preprint arXiv:1605.08803, 2016.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 106, + 514, + 505, + 548 + ], + "lines": [ + { + "bbox": [ + 106, + 513, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 106, + 513, + 505, + 528 + ], + "score": 1.0, + "content": "Vincent Dumoulin, Ishmael Belghazi, Ben Poole, Alex Lamb, Martin Arjovsky, Olivier Mastropi-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 526, + 505, + 538 + ], + "spans": [ + { + "bbox": [ + 115, + 526, + 505, + 538 + ], + "score": 1.0, + "content": "etro, and Aaron Courville. Adversarially learned inference. arXiv preprint arXiv:1606.00704,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 535, + 143, + 550 + ], + "spans": [ + { + "bbox": [ + 115, + 535, + 143, + 550 + ], + "score": 1.0, + "content": "2016.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 107, + 555, + 503, + 590 + ], + "lines": [ + { + "bbox": [ + 105, + 555, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 505, + 569 + ], + "score": 1.0, + "content": "Xavier Glorot and Yoshua Bengio. Understanding the difficulty of training deep feedforward neural", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 567, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 116, + 567, + 505, + 579 + ], + "score": 1.0, + "content": "networks. In JMLR W&CP: Proceedings of the Thirteenth International Conference on Artificial", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 116, + 577, + 434, + 590 + ], + "spans": [ + { + "bbox": [ + 116, + 577, + 434, + 590 + ], + "score": 1.0, + "content": "Intelligence and Statistics (AISTATS 2010), volume 9, pp. 249–256, May 2010.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 106, + 595, + 505, + 651 + ], + "lines": [ + { + "bbox": [ + 105, + 595, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 505, + 608 + ], + "score": 1.0, + "content": "Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 605, + 506, + 621 + ], + "spans": [ + { + "bbox": [ + 115, + 605, + 506, + 621 + ], + "score": 1.0, + "content": "Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Z. Ghahramani, M. Welling,", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 115, + 617, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 115, + 617, + 505, + 631 + ], + "score": 1.0, + "content": "C. Cortes, N. D. Lawrence, and K. Q. Weinberger (eds.), Advances in Neural Information Pro-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 627, + 505, + 643 + ], + "spans": [ + { + "bbox": [ + 115, + 627, + 505, + 643 + ], + "score": 1.0, + "content": "cessing Systems 27, pp. 2672–2680. Curran Associates, Inc., 2014. URL http://papers.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 639, + 422, + 653 + ], + "spans": [ + { + "bbox": [ + 115, + 639, + 422, + 653 + ], + "score": 1.0, + "content": "nips.cc/paper/5423-generative-adversarial-nets.pdf.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38 + }, + { + "type": "text", + "bbox": [ + 107, + 658, + 505, + 703 + ], + "lines": [ + { + "bbox": [ + 106, + 658, + 505, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 658, + 505, + 670 + ], + "score": 1.0, + "content": "Karol Gregor, Ivo Danihelka, Alex Graves, and Daan Wierstra. DRAW: A recurrent neural network", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 669, + 504, + 681 + ], + "spans": [ + { + "bbox": [ + 116, + 669, + 504, + 681 + ], + "score": 1.0, + "content": "for image generation. In Proceedings of The 32nd International Conference on Machine Learn-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 680, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 115, + 680, + 505, + 693 + ], + "score": 1.0, + "content": "ing, pp. 1462–1471, 2015. URL http://www.jmlr.org/proceedings/papers/v37/", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 116, + 693, + 199, + 703 + ], + "spans": [ + { + "bbox": [ + 116, + 693, + 199, + 703 + ], + "score": 1.0, + "content": "gregor15.html.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 42.5 + }, + { + "type": "text", + "bbox": [ + 108, + 709, + 502, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 708, + 504, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 708, + 504, + 723 + ], + "score": 1.0, + "content": "Tian Han, Yang Lu, Song-Chun Zhu, and Ying Nian Wu. Alternating back-propagation for generator", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 115, + 720, + 399, + 733 + ], + "spans": [ + { + "bbox": [ + 115, + 720, + 399, + 733 + ], + "score": 1.0, + "content": "network, 2016. URL https://arxiv.org/abs/1606.08571.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 82, + 176, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 176, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 176, + 95 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 100, + 504, + 133 + ], + "lines": [ + { + "bbox": [ + 105, + 97, + 506, + 114 + ], + "spans": [ + { + "bbox": [ + 105, + 97, + 506, + 114 + ], + "score": 1.0, + "content": "Guillaume Alain, Yoshua Bengio, Li Yao, Jason Yosinski, Eric Thibodeau-Laufer, Saizheng Zhang,", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 115, + 111, + 505, + 123 + ], + "spans": [ + { + "bbox": [ + 115, + 111, + 505, + 123 + ], + "score": 1.0, + "content": "and Pascal Vincent. Gsns : Generative stochastic networks. arXiv preprint arXiv:1503.05571,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 115, + 121, + 142, + 133 + ], + "spans": [ + { + "bbox": [ + 115, + 121, + 142, + 133 + ], + "score": 1.0, + "content": "2015.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 97, + 506, + 133 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 140, + 504, + 174 + ], + "lines": [ + { + "bbox": [ + 106, + 140, + 505, + 152 + ], + "spans": [ + { + "bbox": [ + 106, + 140, + 505, + 152 + ], + "score": 1.0, + "content": "Marcin Andrychowicz, Misha Denil, Sergio Gomez, Matthew W Hoffman, David Pfau, Tom Schaul,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 151, + 505, + 164 + ], + "spans": [ + { + "bbox": [ + 116, + 151, + 505, + 164 + ], + "score": 1.0, + "content": "and Nando de Freitas. Learning to learn by gradient descent by gradient descent. arXiv preprint", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 162, + 219, + 174 + ], + "spans": [ + { + "bbox": [ + 115, + 162, + 219, + 174 + ], + "score": 1.0, + "content": "arXiv:1606.04474, 2016.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5, + "bbox_fs": [ + 106, + 140, + 505, + 174 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 180, + 504, + 203 + ], + "lines": [ + { + "bbox": [ + 105, + 179, + 505, + 194 + ], + "spans": [ + { + "bbox": [ + 105, + 179, + 505, + 194 + ], + "score": 1.0, + "content": "David Belanger and Andrew McCallum. Structured prediction energy networks. arXiv preprint", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 116, + 192, + 219, + 202 + ], + "spans": [ + { + "bbox": [ + 116, + 192, + 219, + 202 + ], + "score": 1.0, + "content": "arXiv:1511.06350, 2015.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 179, + 505, + 202 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 210, + 505, + 233 + ], + "lines": [ + { + "bbox": [ + 106, + 210, + 505, + 223 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 505, + 223 + ], + "score": 1.0, + "content": "Jorg Bornschein, Samira Shabanian, Asja Fischer, and Yoshua Bengio. Bidirectional helmholtz", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 116, + 222, + 322, + 232 + ], + "spans": [ + { + "bbox": [ + 116, + 222, + 322, + 232 + ], + "score": 1.0, + "content": "machines. arXiv preprint arXiv:1506.03877, 2015.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5, + "bbox_fs": [ + 106, + 210, + 505, + 232 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 239, + 505, + 262 + ], + "lines": [ + { + "bbox": [ + 105, + 237, + 506, + 253 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 506, + 253 + ], + "score": 1.0, + "content": "Michael Bowling and Manuela Veloso. Multiagent learning using a variable learning rate. Artificial", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 117, + 250, + 263, + 263 + ], + "spans": [ + { + "bbox": [ + 117, + 250, + 263, + 263 + ], + "score": 1.0, + "content": "Intelligence, 136(2):215–250, 2002.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 237, + 506, + 263 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 268, + 505, + 291 + ], + "lines": [ + { + "bbox": [ + 106, + 268, + 505, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 268, + 505, + 282 + ], + "score": 1.0, + "content": "Yuri Burda, Roger B. Grosse, and Ruslan Salakhutdinov. Importance weighted autoencoders. arXiv", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 280, + 254, + 292 + ], + "spans": [ + { + "bbox": [ + 115, + 280, + 254, + 292 + ], + "score": 1.0, + "content": "preprint arXiv:1509.00519, 2015.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5, + "bbox_fs": [ + 106, + 268, + 505, + 292 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 298, + 504, + 321 + ], + "lines": [ + { + "bbox": [ + 106, + 298, + 505, + 311 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 505, + 311 + ], + "score": 1.0, + "content": "Alex J. Champandard. Semantic style transfer and turning two-bit doodles into fine artworks. arXiv", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 114, + 309, + 254, + 321 + ], + "spans": [ + { + "bbox": [ + 114, + 309, + 254, + 321 + ], + "score": 1.0, + "content": "preprint arXiv:1603.01768, 2016.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 106, + 298, + 505, + 321 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 327, + 504, + 351 + ], + "lines": [ + { + "bbox": [ + 106, + 326, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 106, + 326, + 505, + 340 + ], + "score": 1.0, + "content": "Tong Che, Yanran Li, Athul Paul Jacob, Yoshua Bengio, and Wenjie Li. Mode regularized generative", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 338, + 370, + 351 + ], + "spans": [ + { + "bbox": [ + 115, + 338, + 370, + 351 + ], + "score": 1.0, + "content": "adversarial networks. arXiv preprint arXiv: 1612.02136, 2016.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 106, + 326, + 505, + 351 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 357, + 503, + 390 + ], + "lines": [ + { + "bbox": [ + 106, + 357, + 504, + 369 + ], + "spans": [ + { + "bbox": [ + 106, + 357, + 504, + 369 + ], + "score": 1.0, + "content": "Xi Chen, Yan Duan, Rein Houthooft, John Schulman, Ilya Sutskever, and Pieter Abbeel. Info-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 368, + 505, + 381 + ], + "spans": [ + { + "bbox": [ + 115, + 368, + 505, + 381 + ], + "score": 1.0, + "content": "gan: Interpretable representation learning by information maximizing generative adversarial nets.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 116, + 379, + 279, + 392 + ], + "spans": [ + { + "bbox": [ + 116, + 379, + 279, + 392 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1606.03657, 2016.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 106, + 357, + 505, + 392 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 397, + 504, + 420 + ], + "lines": [ + { + "bbox": [ + 105, + 397, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 505, + 411 + ], + "score": 1.0, + "content": "John M Danskin. The theory of max-min and its application to weapons allocation problems, vol-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 116, + 409, + 321, + 420 + ], + "spans": [ + { + "bbox": [ + 116, + 409, + 321, + 420 + ], + "score": 1.0, + "content": "ume 5. Springer Science & Business Media, 1967.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5, + "bbox_fs": [ + 105, + 397, + 505, + 420 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 426, + 503, + 450 + ], + "lines": [ + { + "bbox": [ + 106, + 426, + 505, + 439 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 505, + 439 + ], + "score": 1.0, + "content": "Peter Dayan, Geoffrey E Hinton, Radford M Neal, and Richard S Zemel. The helmholtz machine.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 115, + 438, + 285, + 450 + ], + "spans": [ + { + "bbox": [ + 115, + 438, + 285, + 450 + ], + "score": 1.0, + "content": "Neural computation, 7(5):889–904, 1995.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5, + "bbox_fs": [ + 106, + 426, + 505, + 450 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 456, + 503, + 479 + ], + "lines": [ + { + "bbox": [ + 105, + 456, + 505, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 505, + 470 + ], + "score": 1.0, + "content": "Laurent Dinh, David Krueger, and Yoshua Bengio. NICE: non-linear independent components esti-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 115, + 468, + 308, + 479 + ], + "spans": [ + { + "bbox": [ + 115, + 468, + 308, + 479 + ], + "score": 1.0, + "content": "mation. arXiv preprint arXiv:1410.8516, 2014.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 456, + 505, + 479 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 485, + 504, + 509 + ], + "lines": [ + { + "bbox": [ + 105, + 484, + 505, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 505, + 499 + ], + "score": 1.0, + "content": "Laurent Dinh, Jascha Sohl-Dickstein, and Samy Bengio. Density estimation using real NVP. arXiv", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 497, + 254, + 508 + ], + "spans": [ + { + "bbox": [ + 115, + 497, + 254, + 508 + ], + "score": 1.0, + "content": "preprint arXiv:1605.08803, 2016.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 484, + 505, + 508 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 514, + 505, + 548 + ], + "lines": [ + { + "bbox": [ + 106, + 513, + 505, + 528 + ], + "spans": [ + { + "bbox": [ + 106, + 513, + 505, + 528 + ], + "score": 1.0, + "content": "Vincent Dumoulin, Ishmael Belghazi, Ben Poole, Alex Lamb, Martin Arjovsky, Olivier Mastropi-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 526, + 505, + 538 + ], + "spans": [ + { + "bbox": [ + 115, + 526, + 505, + 538 + ], + "score": 1.0, + "content": "etro, and Aaron Courville. Adversarially learned inference. arXiv preprint arXiv:1606.00704,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 115, + 535, + 143, + 550 + ], + "spans": [ + { + "bbox": [ + 115, + 535, + 143, + 550 + ], + "score": 1.0, + "content": "2016.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 106, + 513, + 505, + 550 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 555, + 503, + 590 + ], + "lines": [ + { + "bbox": [ + 105, + 555, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 505, + 569 + ], + "score": 1.0, + "content": "Xavier Glorot and Yoshua Bengio. Understanding the difficulty of training deep feedforward neural", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 567, + 505, + 579 + ], + "spans": [ + { + "bbox": [ + 116, + 567, + 505, + 579 + ], + "score": 1.0, + "content": "networks. In JMLR W&CP: Proceedings of the Thirteenth International Conference on Artificial", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 116, + 577, + 434, + 590 + ], + "spans": [ + { + "bbox": [ + 116, + 577, + 434, + 590 + ], + "score": 1.0, + "content": "Intelligence and Statistics (AISTATS 2010), volume 9, pp. 249–256, May 2010.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 555, + 505, + 590 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 595, + 505, + 651 + ], + "lines": [ + { + "bbox": [ + 105, + 595, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 505, + 608 + ], + "score": 1.0, + "content": "Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 605, + 506, + 621 + ], + "spans": [ + { + "bbox": [ + 115, + 605, + 506, + 621 + ], + "score": 1.0, + "content": "Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Z. Ghahramani, M. Welling,", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 115, + 617, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 115, + 617, + 505, + 631 + ], + "score": 1.0, + "content": "C. Cortes, N. D. Lawrence, and K. Q. Weinberger (eds.), Advances in Neural Information Pro-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 627, + 505, + 643 + ], + "spans": [ + { + "bbox": [ + 115, + 627, + 505, + 643 + ], + "score": 1.0, + "content": "cessing Systems 27, pp. 2672–2680. Curran Associates, Inc., 2014. URL http://papers.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 115, + 639, + 422, + 653 + ], + "spans": [ + { + "bbox": [ + 115, + 639, + 422, + 653 + ], + "score": 1.0, + "content": "nips.cc/paper/5423-generative-adversarial-nets.pdf.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38, + "bbox_fs": [ + 105, + 595, + 506, + 653 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 658, + 505, + 703 + ], + "lines": [ + { + "bbox": [ + 106, + 658, + 505, + 670 + ], + "spans": [ + { + "bbox": [ + 106, + 658, + 505, + 670 + ], + "score": 1.0, + "content": "Karol Gregor, Ivo Danihelka, Alex Graves, and Daan Wierstra. DRAW: A recurrent neural network", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 116, + 669, + 504, + 681 + ], + "spans": [ + { + "bbox": [ + 116, + 669, + 504, + 681 + ], + "score": 1.0, + "content": "for image generation. In Proceedings of The 32nd International Conference on Machine Learn-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 115, + 680, + 505, + 693 + ], + "spans": [ + { + "bbox": [ + 115, + 680, + 505, + 693 + ], + "score": 1.0, + "content": "ing, pp. 1462–1471, 2015. URL http://www.jmlr.org/proceedings/papers/v37/", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 116, + 693, + 199, + 703 + ], + "spans": [ + { + "bbox": [ + 116, + 693, + 199, + 703 + ], + "score": 1.0, + "content": "gregor15.html.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 42.5, + "bbox_fs": [ + 106, + 658, + 505, + 703 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 709, + 502, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 708, + 504, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 708, + 504, + 723 + ], + "score": 1.0, + "content": "Tian Han, Yang Lu, Song-Chun Zhu, and Ying Nian Wu. Alternating back-propagation for generator", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 115, + 720, + 399, + 733 + ], + "spans": [ + { + "bbox": [ + 115, + 720, + 399, + 733 + ], + "score": 1.0, + "content": "network, 2016. URL https://arxiv.org/abs/1606.08571.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 45.5, + "bbox_fs": [ + 106, + 708, + 504, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 502, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Sepp Hochreiter and Jurgen Schmidhuber. Long short-term memory. ¨ Neural Comput., 9(8):1735–", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 117, + 93, + 504, + 106 + ], + "spans": [ + { + "bbox": [ + 117, + 93, + 504, + 106 + ], + "score": 1.0, + "content": "1780, November 1997. ISSN 0899-7667. doi: 10.1162/neco.1997.9.8.1735. URL http://dx.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 117, + 105, + 325, + 117 + ], + "spans": [ + { + "bbox": [ + 117, + 105, + 325, + 117 + ], + "score": 1.0, + "content": "doi.org/10.1162/neco.1997.9.8.1735.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 107, + 124, + 505, + 169 + ], + "lines": [ + { + "bbox": [ + 105, + 124, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 124, + 505, + 138 + ], + "score": 1.0, + "content": "Sergey Ioffe and Christian Szegedy. Batch normalization: Accelerating deep network training by", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 135, + 505, + 148 + ], + "spans": [ + { + "bbox": [ + 115, + 135, + 505, + 148 + ], + "score": 1.0, + "content": "reducing internal covariate shift. In Proceedings of the 32nd International Conference on Machine", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 147, + 504, + 159 + ], + "spans": [ + { + "bbox": [ + 116, + 147, + 504, + 159 + ], + "score": 1.0, + "content": "Learning, ICML 2015, Lille, France, 6-11 July 2015, pp. 448–456, 2015. URL http://jmlr.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 158, + 355, + 170 + ], + "spans": [ + { + "bbox": [ + 116, + 158, + 355, + 170 + ], + "score": 1.0, + "content": "org/proceedings/papers/v37/ioffe15.html.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 106, + 177, + 504, + 200 + ], + "lines": [ + { + "bbox": [ + 105, + 177, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 177, + 505, + 190 + ], + "score": 1.0, + "content": "Justin Johnson, Alexandre Alahi, and Fei-Fei Li. Perceptual losses for real-time style transfer and", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 189, + 349, + 201 + ], + "spans": [ + { + "bbox": [ + 115, + 189, + 349, + 201 + ], + "score": 1.0, + "content": "super-resolution. arXiv preprint arXiv:1603.08155, 2016.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 107, + 208, + 506, + 242 + ], + "lines": [ + { + "bbox": [ + 106, + 208, + 505, + 220 + ], + "spans": [ + { + "bbox": [ + 106, + 208, + 505, + 220 + ], + "score": 1.0, + "content": "Anatoli Juditsky, Arkadi Nemirovski, et al. First order methods for nonsmooth convex large-scale", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 219, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 115, + 219, + 505, + 232 + ], + "score": 1.0, + "content": "optimization, i: general purpose methods. Optimization for Machine Learning, pp. 121–148,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 229, + 143, + 242 + ], + "spans": [ + { + "bbox": [ + 115, + 229, + 143, + 242 + ], + "score": 1.0, + "content": "2011.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 250, + 504, + 273 + ], + "lines": [ + { + "bbox": [ + 105, + 249, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 249, + 506, + 264 + ], + "score": 1.0, + "content": "Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 262, + 214, + 273 + ], + "spans": [ + { + "bbox": [ + 116, + 262, + 214, + 273 + ], + "score": 1.0, + "content": "arXiv:1412.6980, 2014.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 105, + 281, + 503, + 304 + ], + "lines": [ + { + "bbox": [ + 105, + 280, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 505, + 295 + ], + "score": 1.0, + "content": "Diederik P Kingma and Max Welling. Auto-encoding variational bayes, 2013. URL https:", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 117, + 293, + 270, + 304 + ], + "spans": [ + { + "bbox": [ + 117, + 293, + 270, + 304 + ], + "score": 1.0, + "content": "//arxiv.org/abs/1312.6114.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 106, + 312, + 504, + 335 + ], + "lines": [ + { + "bbox": [ + 105, + 311, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 506, + 326 + ], + "score": 1.0, + "content": "Diederik P. Kingma, Tim Salimans, and Max Welling. Improving variational inference with inverse", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 324, + 223, + 336 + ], + "spans": [ + { + "bbox": [ + 115, + 324, + 223, + 336 + ], + "score": 1.0, + "content": "autoregressive flow. 2016.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 107, + 343, + 503, + 367 + ], + "lines": [ + { + "bbox": [ + 106, + 343, + 505, + 356 + ], + "spans": [ + { + "bbox": [ + 106, + 343, + 505, + 356 + ], + "score": 1.0, + "content": "Tejas D. Kulkarni, Will Whitney, Pushmeet Kohli, and Joshua B. Tenenbaum. Deep convolutional", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 116, + 354, + 384, + 367 + ], + "spans": [ + { + "bbox": [ + 116, + 354, + 384, + 367 + ], + "score": 1.0, + "content": "inverse graphics network. arXiv preprint arXiv:1503.03167, 2015.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 107, + 374, + 505, + 408 + ], + "lines": [ + { + "bbox": [ + 105, + 374, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 505, + 388 + ], + "score": 1.0, + "content": "Christian Ledig, Lucas Theis, Ferenc Huszar, Jose Caballero, Andrew Aitken, Alykhan Tejani, Jo-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 384, + 506, + 400 + ], + "spans": [ + { + "bbox": [ + 115, + 384, + 506, + 400 + ], + "score": 1.0, + "content": "hannes Totz, Zehan Wang, and Wenzhe Shi. Photo-realistic single image super-resolution using a", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 397, + 489, + 409 + ], + "spans": [ + { + "bbox": [ + 115, + 397, + 489, + 409 + ], + "score": 1.0, + "content": "generative adversarial network, 2016. URL https://arxiv.org/abs/1609.04802.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 416, + 504, + 440 + ], + "lines": [ + { + "bbox": [ + 106, + 416, + 504, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 504, + 430 + ], + "score": 1.0, + "content": "Dougal Maclaurin, David Duvenaud, and Ryan P. Adams. Gradient-based hyperparameter optimiza-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 116, + 427, + 271, + 440 + ], + "spans": [ + { + "bbox": [ + 116, + 427, + 271, + 440 + ], + "score": 1.0, + "content": "tion through reversible learning, 2015.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 106, + 447, + 504, + 482 + ], + "lines": [ + { + "bbox": [ + 105, + 447, + 506, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 506, + 461 + ], + "score": 1.0, + "content": "Anh Nguyen, Alexey Dosovitskiy, Jason Yosinski, Thomas Brox, and Jeff Clune. Synthesizing", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 116, + 459, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 116, + 459, + 505, + 471 + ], + "score": 1.0, + "content": "the preferred inputs for neurons in neural networks via deep generator networks. arXiv preprint", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 470, + 219, + 481 + ], + "spans": [ + { + "bbox": [ + 116, + 470, + 219, + 481 + ], + "score": 1.0, + "content": "arXiv:1605.09304, 2016.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 104, + 489, + 504, + 513 + ], + "lines": [ + { + "bbox": [ + 105, + 489, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 505, + 503 + ], + "score": 1.0, + "content": "Sebastian Nowozin, Botond Cseke, and Ryota Tomioka. f-gan: Training generative neural samplers", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 501, + 452, + 513 + ], + "spans": [ + { + "bbox": [ + 115, + 501, + 452, + 513 + ], + "score": 1.0, + "content": "using variational divergence minimization. arXiv preprint arXiv:1606.00709, 2016.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 106, + 521, + 503, + 544 + ], + "lines": [ + { + "bbox": [ + 106, + 521, + 504, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 521, + 504, + 534 + ], + "score": 1.0, + "content": "Augustus Odena, Christopher Olah, and Jonathon Shlens. Conditional image synthesis with auxil-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 532, + 358, + 544 + ], + "spans": [ + { + "bbox": [ + 115, + 532, + 358, + 544 + ], + "score": 1.0, + "content": "iary classifier gans. arXiv preprint arXiv:1610.09585, 2016.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + }, + { + "type": "text", + "bbox": [ + 106, + 551, + 505, + 597 + ], + "lines": [ + { + "bbox": [ + 106, + 552, + 505, + 564 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 505, + 564 + ], + "score": 1.0, + "content": "Barak A. Pearlmutter and Jeffrey Mark Siskind. Reverse-mode ad in a functional framework:", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 564, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 115, + 564, + 505, + 576 + ], + "score": 1.0, + "content": "Lambda the ultimate backpropagator. ACM Trans. Program. Lang. Syst., 30(2):7:1–7:36, March", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 573, + 504, + 586 + ], + "spans": [ + { + "bbox": [ + 115, + 573, + 504, + 586 + ], + "score": 1.0, + "content": "2008. ISSN 0164-0925. doi: 10.1145/1330017.1330018. URL http://doi.acm.org/10.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 116, + 586, + 241, + 596 + ], + "spans": [ + { + "bbox": [ + 116, + 586, + 241, + 596 + ], + "score": 1.0, + "content": "1145/1330017.1330018.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 106, + 605, + 504, + 628 + ], + "lines": [ + { + "bbox": [ + 106, + 604, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 106, + 604, + 505, + 618 + ], + "score": 1.0, + "content": "Ben Poole, Alexander A Alemi, Jascha Sohl-Dickstein, and Anelia Angelova. Improved generator", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 617, + 359, + 628 + ], + "spans": [ + { + "bbox": [ + 115, + 617, + 359, + 628 + ], + "score": 1.0, + "content": "objectives for gans. arXiv preprint arXiv:1612.02780, 2016.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 105, + 636, + 504, + 659 + ], + "lines": [ + { + "bbox": [ + 105, + 635, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 635, + 506, + 650 + ], + "score": 1.0, + "content": "Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learning with deep", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 647, + 468, + 660 + ], + "spans": [ + { + "bbox": [ + 115, + 647, + 468, + 660 + ], + "score": 1.0, + "content": "convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434, 2015.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 106, + 667, + 504, + 690 + ], + "lines": [ + { + "bbox": [ + 105, + 666, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 681 + ], + "score": 1.0, + "content": "Scott Reed, Zeynep Akata, Santosh Mohan, Samuel Tenka, Bernt Schiele, and Honglak Lee. Learn-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 678, + 299, + 691 + ], + "spans": [ + { + "bbox": [ + 115, + 678, + 299, + 691 + ], + "score": 1.0, + "content": "ing what and where to draw. In NIPS, 2016a.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5 + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 504, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 504, + 712 + ], + "score": 1.0, + "content": "Scott Reed, Zeynep Akata, Xinchen Yan, Lajanugen Logeswaran, Bernt Schiele, and Honglak Lee.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 116, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "Generative adversarial text-to-image synthesis. In Proceedings of The 33rd International Confer-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 116, + 721, + 258, + 733 + ], + "spans": [ + { + "bbox": [ + 116, + 721, + 258, + 733 + ], + "score": 1.0, + "content": "ence on Machine Learning, 2016b.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 108, + 82, + 502, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "Sepp Hochreiter and Jurgen Schmidhuber. Long short-term memory. ¨ Neural Comput., 9(8):1735–", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 117, + 93, + 504, + 106 + ], + "spans": [ + { + "bbox": [ + 117, + 93, + 504, + 106 + ], + "score": 1.0, + "content": "1780, November 1997. ISSN 0899-7667. doi: 10.1162/neco.1997.9.8.1735. URL http://dx.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 117, + 105, + 325, + 117 + ], + "spans": [ + { + "bbox": [ + 117, + 105, + 325, + 117 + ], + "score": 1.0, + "content": "doi.org/10.1162/neco.1997.9.8.1735.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 81, + 505, + 117 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 124, + 505, + 169 + ], + "lines": [ + { + "bbox": [ + 105, + 124, + 505, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 124, + 505, + 138 + ], + "score": 1.0, + "content": "Sergey Ioffe and Christian Szegedy. Batch normalization: Accelerating deep network training by", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 135, + 505, + 148 + ], + "spans": [ + { + "bbox": [ + 115, + 135, + 505, + 148 + ], + "score": 1.0, + "content": "reducing internal covariate shift. In Proceedings of the 32nd International Conference on Machine", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 116, + 147, + 504, + 159 + ], + "spans": [ + { + "bbox": [ + 116, + 147, + 504, + 159 + ], + "score": 1.0, + "content": "Learning, ICML 2015, Lille, France, 6-11 July 2015, pp. 448–456, 2015. URL http://jmlr.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 116, + 158, + 355, + 170 + ], + "spans": [ + { + "bbox": [ + 116, + 158, + 355, + 170 + ], + "score": 1.0, + "content": "org/proceedings/papers/v37/ioffe15.html.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 124, + 505, + 170 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 177, + 504, + 200 + ], + "lines": [ + { + "bbox": [ + 105, + 177, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 105, + 177, + 505, + 190 + ], + "score": 1.0, + "content": "Justin Johnson, Alexandre Alahi, and Fei-Fei Li. Perceptual losses for real-time style transfer and", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 189, + 349, + 201 + ], + "spans": [ + { + "bbox": [ + 115, + 189, + 349, + 201 + ], + "score": 1.0, + "content": "super-resolution. arXiv preprint arXiv:1603.08155, 2016.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 177, + 505, + 201 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 208, + 506, + 242 + ], + "lines": [ + { + "bbox": [ + 106, + 208, + 505, + 220 + ], + "spans": [ + { + "bbox": [ + 106, + 208, + 505, + 220 + ], + "score": 1.0, + "content": "Anatoli Juditsky, Arkadi Nemirovski, et al. First order methods for nonsmooth convex large-scale", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 115, + 219, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 115, + 219, + 505, + 232 + ], + "score": 1.0, + "content": "optimization, i: general purpose methods. Optimization for Machine Learning, pp. 121–148,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 115, + 229, + 143, + 242 + ], + "spans": [ + { + "bbox": [ + 115, + 229, + 143, + 242 + ], + "score": 1.0, + "content": "2011.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 106, + 208, + 505, + 242 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 250, + 504, + 273 + ], + "lines": [ + { + "bbox": [ + 105, + 249, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 249, + 506, + 264 + ], + "score": 1.0, + "content": "Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 262, + 214, + 273 + ], + "spans": [ + { + "bbox": [ + 116, + 262, + 214, + 273 + ], + "score": 1.0, + "content": "arXiv:1412.6980, 2014.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 249, + 506, + 273 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 281, + 503, + 304 + ], + "lines": [ + { + "bbox": [ + 105, + 280, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 505, + 295 + ], + "score": 1.0, + "content": "Diederik P Kingma and Max Welling. Auto-encoding variational bayes, 2013. URL https:", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 117, + 293, + 270, + 304 + ], + "spans": [ + { + "bbox": [ + 117, + 293, + 270, + 304 + ], + "score": 1.0, + "content": "//arxiv.org/abs/1312.6114.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 280, + 505, + 304 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 312, + 504, + 335 + ], + "lines": [ + { + "bbox": [ + 105, + 311, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 506, + 326 + ], + "score": 1.0, + "content": "Diederik P. Kingma, Tim Salimans, and Max Welling. Improving variational inference with inverse", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 324, + 223, + 336 + ], + "spans": [ + { + "bbox": [ + 115, + 324, + 223, + 336 + ], + "score": 1.0, + "content": "autoregressive flow. 2016.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 311, + 506, + 336 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 343, + 503, + 367 + ], + "lines": [ + { + "bbox": [ + 106, + 343, + 505, + 356 + ], + "spans": [ + { + "bbox": [ + 106, + 343, + 505, + 356 + ], + "score": 1.0, + "content": "Tejas D. Kulkarni, Will Whitney, Pushmeet Kohli, and Joshua B. Tenenbaum. Deep convolutional", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 116, + 354, + 384, + 367 + ], + "spans": [ + { + "bbox": [ + 116, + 354, + 384, + 367 + ], + "score": 1.0, + "content": "inverse graphics network. arXiv preprint arXiv:1503.03167, 2015.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 106, + 343, + 505, + 367 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 374, + 505, + 408 + ], + "lines": [ + { + "bbox": [ + 105, + 374, + 505, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 505, + 388 + ], + "score": 1.0, + "content": "Christian Ledig, Lucas Theis, Ferenc Huszar, Jose Caballero, Andrew Aitken, Alykhan Tejani, Jo-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 384, + 506, + 400 + ], + "spans": [ + { + "bbox": [ + 115, + 384, + 506, + 400 + ], + "score": 1.0, + "content": "hannes Totz, Zehan Wang, and Wenzhe Shi. Photo-realistic single image super-resolution using a", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 115, + 397, + 489, + 409 + ], + "spans": [ + { + "bbox": [ + 115, + 397, + 489, + 409 + ], + "score": 1.0, + "content": "generative adversarial network, 2016. URL https://arxiv.org/abs/1609.04802.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 374, + 506, + 409 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 416, + 504, + 440 + ], + "lines": [ + { + "bbox": [ + 106, + 416, + 504, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 416, + 504, + 430 + ], + "score": 1.0, + "content": "Dougal Maclaurin, David Duvenaud, and Ryan P. Adams. Gradient-based hyperparameter optimiza-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 116, + 427, + 271, + 440 + ], + "spans": [ + { + "bbox": [ + 116, + 427, + 271, + 440 + ], + "score": 1.0, + "content": "tion through reversible learning, 2015.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 106, + 416, + 504, + 440 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 447, + 504, + 482 + ], + "lines": [ + { + "bbox": [ + 105, + 447, + 506, + 461 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 506, + 461 + ], + "score": 1.0, + "content": "Anh Nguyen, Alexey Dosovitskiy, Jason Yosinski, Thomas Brox, and Jeff Clune. Synthesizing", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 116, + 459, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 116, + 459, + 505, + 471 + ], + "score": 1.0, + "content": "the preferred inputs for neurons in neural networks via deep generator networks. arXiv preprint", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 470, + 219, + 481 + ], + "spans": [ + { + "bbox": [ + 116, + 470, + 219, + 481 + ], + "score": 1.0, + "content": "arXiv:1605.09304, 2016.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26, + "bbox_fs": [ + 105, + 447, + 506, + 481 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 489, + 504, + 513 + ], + "lines": [ + { + "bbox": [ + 105, + 489, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 105, + 489, + 505, + 503 + ], + "score": 1.0, + "content": "Sebastian Nowozin, Botond Cseke, and Ryota Tomioka. f-gan: Training generative neural samplers", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 115, + 501, + 452, + 513 + ], + "spans": [ + { + "bbox": [ + 115, + 501, + 452, + 513 + ], + "score": 1.0, + "content": "using variational divergence minimization. arXiv preprint arXiv:1606.00709, 2016.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 489, + 505, + 513 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 521, + 503, + 544 + ], + "lines": [ + { + "bbox": [ + 106, + 521, + 504, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 521, + 504, + 534 + ], + "score": 1.0, + "content": "Augustus Odena, Christopher Olah, and Jonathon Shlens. Conditional image synthesis with auxil-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 115, + 532, + 358, + 544 + ], + "spans": [ + { + "bbox": [ + 115, + 532, + 358, + 544 + ], + "score": 1.0, + "content": "iary classifier gans. arXiv preprint arXiv:1610.09585, 2016.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5, + "bbox_fs": [ + 106, + 521, + 504, + 544 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 551, + 505, + 597 + ], + "lines": [ + { + "bbox": [ + 106, + 552, + 505, + 564 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 505, + 564 + ], + "score": 1.0, + "content": "Barak A. Pearlmutter and Jeffrey Mark Siskind. Reverse-mode ad in a functional framework:", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 564, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 115, + 564, + 505, + 576 + ], + "score": 1.0, + "content": "Lambda the ultimate backpropagator. ACM Trans. Program. Lang. Syst., 30(2):7:1–7:36, March", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 115, + 573, + 504, + 586 + ], + "spans": [ + { + "bbox": [ + 115, + 573, + 504, + 586 + ], + "score": 1.0, + "content": "2008. ISSN 0164-0925. doi: 10.1145/1330017.1330018. URL http://doi.acm.org/10.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 116, + 586, + 241, + 596 + ], + "spans": [ + { + "bbox": [ + 116, + 586, + 241, + 596 + ], + "score": 1.0, + "content": "1145/1330017.1330018.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33.5, + "bbox_fs": [ + 106, + 552, + 505, + 596 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 605, + 504, + 628 + ], + "lines": [ + { + "bbox": [ + 106, + 604, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 106, + 604, + 505, + 618 + ], + "score": 1.0, + "content": "Ben Poole, Alexander A Alemi, Jascha Sohl-Dickstein, and Anelia Angelova. Improved generator", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 115, + 617, + 359, + 628 + ], + "spans": [ + { + "bbox": [ + 115, + 617, + 359, + 628 + ], + "score": 1.0, + "content": "objectives for gans. arXiv preprint arXiv:1612.02780, 2016.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5, + "bbox_fs": [ + 106, + 604, + 505, + 628 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 636, + 504, + 659 + ], + "lines": [ + { + "bbox": [ + 105, + 635, + 506, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 635, + 506, + 650 + ], + "score": 1.0, + "content": "Alec Radford, Luke Metz, and Soumith Chintala. Unsupervised representation learning with deep", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 115, + 647, + 468, + 660 + ], + "spans": [ + { + "bbox": [ + 115, + 647, + 468, + 660 + ], + "score": 1.0, + "content": "convolutional generative adversarial networks. arXiv preprint arXiv:1511.06434, 2015.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 635, + 506, + 660 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 667, + 504, + 690 + ], + "lines": [ + { + "bbox": [ + 105, + 666, + 505, + 681 + ], + "spans": [ + { + "bbox": [ + 105, + 666, + 505, + 681 + ], + "score": 1.0, + "content": "Scott Reed, Zeynep Akata, Santosh Mohan, Samuel Tenka, Bernt Schiele, and Honglak Lee. Learn-", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 115, + 678, + 299, + 691 + ], + "spans": [ + { + "bbox": [ + 115, + 678, + 299, + 691 + ], + "score": 1.0, + "content": "ing what and where to draw. In NIPS, 2016a.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 666, + 505, + 691 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 504, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 504, + 712 + ], + "score": 1.0, + "content": "Scott Reed, Zeynep Akata, Xinchen Yan, Lajanugen Logeswaran, Bernt Schiele, and Honglak Lee.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 116, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "Generative adversarial text-to-image synthesis. In Proceedings of The 33rd International Confer-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 116, + 721, + 258, + 733 + ], + "spans": [ + { + "bbox": [ + 116, + 721, + 258, + 733 + ], + "score": 1.0, + "content": "ence on Machine Learning, 2016b.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 698, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 503, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Danilo Jimenez Rezende, Shakir Mohamed, and Daan Wierstra. Stochastic backpropagation and", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 116, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "variational inference in deep latent gaussian models. In International Conference on Machine", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 104, + 221, + 116 + ], + "spans": [ + { + "bbox": [ + 116, + 104, + 221, + 116 + ], + "score": 1.0, + "content": "Learning. Citeseer, 2014.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 123, + 503, + 146 + ], + "lines": [ + { + "bbox": [ + 106, + 122, + 505, + 136 + ], + "spans": [ + { + "bbox": [ + 106, + 122, + 505, + 136 + ], + "score": 1.0, + "content": "Tim Salimans, Ian J. Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 134, + 436, + 147 + ], + "spans": [ + { + "bbox": [ + 115, + 134, + 436, + 147 + ], + "score": 1.0, + "content": "Improved techniques for training gans. arXiv preprint arXiv:1606.03498, 2016.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 105, + 153, + 504, + 176 + ], + "lines": [ + { + "bbox": [ + 107, + 154, + 504, + 165 + ], + "spans": [ + { + "bbox": [ + 107, + 154, + 504, + 165 + ], + "score": 1.0, + "content": "Karen Simonyan, Andrea Vedaldi, and Andrew Zisserman. Deep inside convolutional networks: Vi-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 164, + 504, + 176 + ], + "spans": [ + { + "bbox": [ + 115, + 164, + 504, + 176 + ], + "score": 1.0, + "content": "sualising image classification models and saliency maps. arXiv preprint arXiv:1312.6034, 2013.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 108, + 182, + 504, + 217 + ], + "lines": [ + { + "bbox": [ + 106, + 183, + 505, + 196 + ], + "spans": [ + { + "bbox": [ + 106, + 183, + 505, + 196 + ], + "score": 1.0, + "content": "Satinder Singh, Michael Kearns, and Yishay Mansour. Nash convergence of gradient dynamics", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 195, + 505, + 206 + ], + "spans": [ + { + "bbox": [ + 115, + 195, + 505, + 206 + ], + "score": 1.0, + "content": "in general-sum games. In Proceedings of the Sixteenth conference on Uncertainty in artificial", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 116, + 205, + 390, + 217 + ], + "spans": [ + { + "bbox": [ + 116, + 205, + 390, + 217 + ], + "score": 1.0, + "content": "intelligence, pp. 541–548. Morgan Kaufmann Publishers Inc., 2000.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 106, + 223, + 506, + 268 + ], + "lines": [ + { + "bbox": [ + 104, + 223, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 104, + 223, + 506, + 237 + ], + "score": 1.0, + "content": "Jascha Sohl-Dickstein, Eric A. Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsuper-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 235, + 506, + 247 + ], + "spans": [ + { + "bbox": [ + 116, + 235, + 506, + 247 + ], + "score": 1.0, + "content": "vised learning using nonequilibrium thermodynamics. In Proceedings of The 32nd International", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 245, + 505, + 259 + ], + "spans": [ + { + "bbox": [ + 115, + 245, + 505, + 259 + ], + "score": 1.0, + "content": "Conference on Machine Learning, pp. 2256–2265, 2015. URL http://arxiv.org/abs/", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 257, + 181, + 268 + ], + "spans": [ + { + "bbox": [ + 116, + 257, + 181, + 268 + ], + "score": 1.0, + "content": "1503.03585.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 106, + 275, + 505, + 309 + ], + "lines": [ + { + "bbox": [ + 106, + 275, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 106, + 275, + 505, + 288 + ], + "score": 1.0, + "content": "Casper Kaae Sonderby, Jose Caballero, Lucas Theis, Wenzhe Shi, and Ferenc Huszar. Amortised", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 286, + 504, + 299 + ], + "spans": [ + { + "bbox": [ + 115, + 286, + 504, + 299 + ], + "score": 1.0, + "content": "map inference for image super-resolution, 2016. URL https://arxiv.org/abs/1610.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 297, + 164, + 310 + ], + "spans": [ + { + "bbox": [ + 116, + 297, + 164, + 310 + ], + "score": 1.0, + "content": "04490v1.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 106, + 316, + 505, + 349 + ], + "lines": [ + { + "bbox": [ + 105, + 316, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 316, + 505, + 329 + ], + "score": 1.0, + "content": "L. Theis and M. Bethge. Generative image modeling using spatial lstms. In Advances in Neu-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 327, + 504, + 340 + ], + "spans": [ + { + "bbox": [ + 116, + 327, + 504, + 340 + ], + "score": 1.0, + "content": "ral Information Processing Systems 28, Dec 2015. URL http://arxiv.org/abs/1506.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 117, + 338, + 158, + 351 + ], + "spans": [ + { + "bbox": [ + 117, + 339, + 153, + 349 + ], + "score": 0.3, + "content": "0 3 4 7 8 /", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 338, + 158, + 351 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 357, + 504, + 391 + ], + "lines": [ + { + "bbox": [ + 105, + 357, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 357, + 505, + 370 + ], + "score": 1.0, + "content": "L. Theis, A. van den Oord, and M. Bethge. A note on the evaluation of generative models. In In-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 114, + 367, + 504, + 381 + ], + "spans": [ + { + "bbox": [ + 114, + 367, + 504, + 381 + ], + "score": 1.0, + "content": "ternational Conference on Learning Representations, Apr 2016. URL http://arxiv.org/", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 380, + 206, + 391 + ], + "spans": [ + { + "bbox": [ + 116, + 380, + 206, + 391 + ], + "score": 1.0, + "content": "abs/1511.01844.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 104, + 397, + 504, + 421 + ], + "lines": [ + { + "bbox": [ + 105, + 397, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 505, + 411 + ], + "score": 1.0, + "content": "T. Tieleman and G. Hinton. Lecture 6.5—RmsProp: Divide the gradient by a running average of its", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 409, + 438, + 422 + ], + "spans": [ + { + "bbox": [ + 115, + 409, + 438, + 422 + ], + "score": 1.0, + "content": "recent magnitude. COURSERA: Neural Networks for Machine Learning, 2012.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 106, + 428, + 504, + 461 + ], + "lines": [ + { + "bbox": [ + 106, + 428, + 504, + 440 + ], + "spans": [ + { + "bbox": [ + 106, + 428, + 504, + 440 + ], + "score": 1.0, + "content": "Aaron van den Oord, Nal Kalchbrenner, and Koray Kavukcuoglu. Pixel recurrent neural networks.¨", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 439, + 504, + 451 + ], + "spans": [ + { + "bbox": [ + 115, + 439, + 504, + 451 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1601.06759, abs/1601.06759, 2016a. URL http://arxiv.org/abs/", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 450, + 182, + 461 + ], + "spans": [ + { + "bbox": [ + 116, + 450, + 182, + 461 + ], + "score": 1.0, + "content": "1601.06759.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 106, + 468, + 504, + 502 + ], + "lines": [ + { + "bbox": [ + 106, + 469, + 504, + 481 + ], + "spans": [ + { + "bbox": [ + 106, + 469, + 504, + 481 + ], + "score": 1.0, + "content": "Aaron van den Oord, Nal Kalchbrenner, Oriol Vinyals, Lasse Espeholt, Alex Graves, and Ko-¨", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 116, + 480, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 116, + 480, + 505, + 492 + ], + "score": 1.0, + "content": "ray Kavukcuoglu. Conditional image generation with pixelcnn decoders. arXiv preprint", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 116, + 491, + 224, + 503 + ], + "spans": [ + { + "bbox": [ + 116, + 491, + 224, + 503 + ], + "score": 1.0, + "content": "arXiv:1606.05328, 2016b.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 106, + 509, + 505, + 555 + ], + "lines": [ + { + "bbox": [ + 106, + 510, + 504, + 522 + ], + "spans": [ + { + "bbox": [ + 106, + 510, + 504, + 522 + ], + "score": 1.0, + "content": "Pascal Vincent, Hugo Larochelle, Isabelle Lajoie, Yoshua Bengio, and Pierre-Antoine Manzagol.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 116, + 521, + 506, + 534 + ], + "spans": [ + { + "bbox": [ + 116, + 521, + 506, + 534 + ], + "score": 1.0, + "content": "Stacked denoising autoencoders: Learning useful representations in a deep network with a local", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 532, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 115, + 532, + 505, + 543 + ], + "score": 1.0, + "content": "denoising criterion. J. Mach. Learn. Res., 11:3371–3408, December 2010. ISSN 1532-4435.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 543, + 437, + 555 + ], + "spans": [ + { + "bbox": [ + 116, + 543, + 437, + 555 + ], + "score": 1.0, + "content": "URL http://dl.acm.org/citation.cfm?id=1756006.1953039.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 106, + 561, + 502, + 585 + ], + "lines": [ + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "score": 1.0, + "content": "Jason Yosinski, Jeff Clune, Anh Nguyen, Thomas Fuchs, and Hod Lipson. Understanding neural", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 572, + 429, + 586 + ], + "spans": [ + { + "bbox": [ + 115, + 572, + 429, + 586 + ], + "score": 1.0, + "content": "networks through deep visualization. arXiv preprint arXiv:1506.06579, 2015.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5 + }, + { + "type": "text", + "bbox": [ + 106, + 591, + 504, + 615 + ], + "lines": [ + { + "bbox": [ + 105, + 590, + 505, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 505, + 605 + ], + "score": 1.0, + "content": "Chongjie Zhang and Victor R Lesser. Multi-agent learning with policy prediction. In Proceedings", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 602, + 401, + 615 + ], + "spans": [ + { + "bbox": [ + 116, + 602, + 401, + 615 + ], + "score": 1.0, + "content": "of the Twenty-Fourth AAAI Conference on Artificial Intelligence, 2010.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 106, + 621, + 503, + 645 + ], + "lines": [ + { + "bbox": [ + 106, + 621, + 504, + 634 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 504, + 634 + ], + "score": 1.0, + "content": "Junbo Zhao, Michael Mathieu, and Yann LeCun. Energy-based generative adversarial network.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 632, + 278, + 645 + ], + "spans": [ + { + "bbox": [ + 116, + 632, + 278, + 645 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1609.03126, 2016.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5 + }, + { + "type": "text", + "bbox": [ + 106, + 651, + 505, + 685 + ], + "lines": [ + { + "bbox": [ + 106, + 651, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 106, + 651, + 505, + 663 + ], + "score": 1.0, + "content": "Jun-Yan Zhu, Philipp Krahenb ¨ uhl, Eli Shechtman, and Alexei A. Efros. Generative visual manipula- ¨", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 662, + 506, + 676 + ], + "spans": [ + { + "bbox": [ + 115, + 662, + 506, + 676 + ], + "score": 1.0, + "content": "tion on the natural image manifold. In Proceedings of European Conference on Computer Vision", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 673, + 178, + 686 + ], + "spans": [ + { + "bbox": [ + 116, + 673, + 178, + 686 + ], + "score": 1.0, + "content": "(ECCV), 2016.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 763 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 763 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 503, + 116 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Danilo Jimenez Rezende, Shakir Mohamed, and Daan Wierstra. Stochastic backpropagation and", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 116, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 116, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "variational inference in deep latent gaussian models. In International Conference on Machine", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 104, + 221, + 116 + ], + "spans": [ + { + "bbox": [ + 116, + 104, + 221, + 116 + ], + "score": 1.0, + "content": "Learning. Citeseer, 2014.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 105, + 82, + 505, + 116 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 123, + 503, + 146 + ], + "lines": [ + { + "bbox": [ + 106, + 122, + 505, + 136 + ], + "spans": [ + { + "bbox": [ + 106, + 122, + 505, + 136 + ], + "score": 1.0, + "content": "Tim Salimans, Ian J. Goodfellow, Wojciech Zaremba, Vicki Cheung, Alec Radford, and Xi Chen.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 134, + 436, + 147 + ], + "spans": [ + { + "bbox": [ + 115, + 134, + 436, + 147 + ], + "score": 1.0, + "content": "Improved techniques for training gans. arXiv preprint arXiv:1606.03498, 2016.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 106, + 122, + 505, + 147 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 153, + 504, + 176 + ], + "lines": [ + { + "bbox": [ + 107, + 154, + 504, + 165 + ], + "spans": [ + { + "bbox": [ + 107, + 154, + 504, + 165 + ], + "score": 1.0, + "content": "Karen Simonyan, Andrea Vedaldi, and Andrew Zisserman. Deep inside convolutional networks: Vi-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 164, + 504, + 176 + ], + "spans": [ + { + "bbox": [ + 115, + 164, + 504, + 176 + ], + "score": 1.0, + "content": "sualising image classification models and saliency maps. arXiv preprint arXiv:1312.6034, 2013.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 107, + 154, + 504, + 176 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 182, + 504, + 217 + ], + "lines": [ + { + "bbox": [ + 106, + 183, + 505, + 196 + ], + "spans": [ + { + "bbox": [ + 106, + 183, + 505, + 196 + ], + "score": 1.0, + "content": "Satinder Singh, Michael Kearns, and Yishay Mansour. Nash convergence of gradient dynamics", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 115, + 195, + 505, + 206 + ], + "spans": [ + { + "bbox": [ + 115, + 195, + 505, + 206 + ], + "score": 1.0, + "content": "in general-sum games. In Proceedings of the Sixteenth conference on Uncertainty in artificial", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 116, + 205, + 390, + 217 + ], + "spans": [ + { + "bbox": [ + 116, + 205, + 390, + 217 + ], + "score": 1.0, + "content": "intelligence, pp. 541–548. Morgan Kaufmann Publishers Inc., 2000.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8, + "bbox_fs": [ + 106, + 183, + 505, + 217 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 223, + 506, + 268 + ], + "lines": [ + { + "bbox": [ + 104, + 223, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 104, + 223, + 506, + 237 + ], + "score": 1.0, + "content": "Jascha Sohl-Dickstein, Eric A. Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsuper-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 116, + 235, + 506, + 247 + ], + "spans": [ + { + "bbox": [ + 116, + 235, + 506, + 247 + ], + "score": 1.0, + "content": "vised learning using nonequilibrium thermodynamics. In Proceedings of The 32nd International", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 245, + 505, + 259 + ], + "spans": [ + { + "bbox": [ + 115, + 245, + 505, + 259 + ], + "score": 1.0, + "content": "Conference on Machine Learning, pp. 2256–2265, 2015. URL http://arxiv.org/abs/", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 116, + 257, + 181, + 268 + ], + "spans": [ + { + "bbox": [ + 116, + 257, + 181, + 268 + ], + "score": 1.0, + "content": "1503.03585.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11.5, + "bbox_fs": [ + 104, + 223, + 506, + 268 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 275, + 505, + 309 + ], + "lines": [ + { + "bbox": [ + 106, + 275, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 106, + 275, + 505, + 288 + ], + "score": 1.0, + "content": "Casper Kaae Sonderby, Jose Caballero, Lucas Theis, Wenzhe Shi, and Ferenc Huszar. Amortised", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 286, + 504, + 299 + ], + "spans": [ + { + "bbox": [ + 115, + 286, + 504, + 299 + ], + "score": 1.0, + "content": "map inference for image super-resolution, 2016. URL https://arxiv.org/abs/1610.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 116, + 297, + 164, + 310 + ], + "spans": [ + { + "bbox": [ + 116, + 297, + 164, + 310 + ], + "score": 1.0, + "content": "04490v1.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15, + "bbox_fs": [ + 106, + 275, + 505, + 310 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 316, + 505, + 349 + ], + "lines": [ + { + "bbox": [ + 105, + 316, + 505, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 316, + 505, + 329 + ], + "score": 1.0, + "content": "L. Theis and M. Bethge. Generative image modeling using spatial lstms. In Advances in Neu-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 116, + 327, + 504, + 340 + ], + "spans": [ + { + "bbox": [ + 116, + 327, + 504, + 340 + ], + "score": 1.0, + "content": "ral Information Processing Systems 28, Dec 2015. URL http://arxiv.org/abs/1506.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 117, + 338, + 158, + 351 + ], + "spans": [ + { + "bbox": [ + 117, + 339, + 153, + 349 + ], + "score": 0.3, + "content": "0 3 4 7 8 /", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 338, + 158, + 351 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 316, + 505, + 351 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 357, + 504, + 391 + ], + "lines": [ + { + "bbox": [ + 105, + 357, + 505, + 370 + ], + "spans": [ + { + "bbox": [ + 105, + 357, + 505, + 370 + ], + "score": 1.0, + "content": "L. Theis, A. van den Oord, and M. Bethge. A note on the evaluation of generative models. In In-", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 114, + 367, + 504, + 381 + ], + "spans": [ + { + "bbox": [ + 114, + 367, + 504, + 381 + ], + "score": 1.0, + "content": "ternational Conference on Learning Representations, Apr 2016. URL http://arxiv.org/", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 380, + 206, + 391 + ], + "spans": [ + { + "bbox": [ + 116, + 380, + 206, + 391 + ], + "score": 1.0, + "content": "abs/1511.01844.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 357, + 505, + 391 + ] + }, + { + "type": "text", + "bbox": [ + 104, + 397, + 504, + 421 + ], + "lines": [ + { + "bbox": [ + 105, + 397, + 505, + 411 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 505, + 411 + ], + "score": 1.0, + "content": "T. Tieleman and G. Hinton. Lecture 6.5—RmsProp: Divide the gradient by a running average of its", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 409, + 438, + 422 + ], + "spans": [ + { + "bbox": [ + 115, + 409, + 438, + 422 + ], + "score": 1.0, + "content": "recent magnitude. COURSERA: Neural Networks for Machine Learning, 2012.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 397, + 505, + 422 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 428, + 504, + 461 + ], + "lines": [ + { + "bbox": [ + 106, + 428, + 504, + 440 + ], + "spans": [ + { + "bbox": [ + 106, + 428, + 504, + 440 + ], + "score": 1.0, + "content": "Aaron van den Oord, Nal Kalchbrenner, and Koray Kavukcuoglu. Pixel recurrent neural networks.¨", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 439, + 504, + 451 + ], + "spans": [ + { + "bbox": [ + 115, + 439, + 504, + 451 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1601.06759, abs/1601.06759, 2016a. URL http://arxiv.org/abs/", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 116, + 450, + 182, + 461 + ], + "spans": [ + { + "bbox": [ + 116, + 450, + 182, + 461 + ], + "score": 1.0, + "content": "1601.06759.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26, + "bbox_fs": [ + 106, + 428, + 504, + 461 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 468, + 504, + 502 + ], + "lines": [ + { + "bbox": [ + 106, + 469, + 504, + 481 + ], + "spans": [ + { + "bbox": [ + 106, + 469, + 504, + 481 + ], + "score": 1.0, + "content": "Aaron van den Oord, Nal Kalchbrenner, Oriol Vinyals, Lasse Espeholt, Alex Graves, and Ko-¨", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 116, + 480, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 116, + 480, + 505, + 492 + ], + "score": 1.0, + "content": "ray Kavukcuoglu. Conditional image generation with pixelcnn decoders. arXiv preprint", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 116, + 491, + 224, + 503 + ], + "spans": [ + { + "bbox": [ + 116, + 491, + 224, + 503 + ], + "score": 1.0, + "content": "arXiv:1606.05328, 2016b.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29, + "bbox_fs": [ + 106, + 469, + 505, + 503 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 509, + 505, + 555 + ], + "lines": [ + { + "bbox": [ + 106, + 510, + 504, + 522 + ], + "spans": [ + { + "bbox": [ + 106, + 510, + 504, + 522 + ], + "score": 1.0, + "content": "Pascal Vincent, Hugo Larochelle, Isabelle Lajoie, Yoshua Bengio, and Pierre-Antoine Manzagol.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 116, + 521, + 506, + 534 + ], + "spans": [ + { + "bbox": [ + 116, + 521, + 506, + 534 + ], + "score": 1.0, + "content": "Stacked denoising autoencoders: Learning useful representations in a deep network with a local", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 115, + 532, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 115, + 532, + 505, + 543 + ], + "score": 1.0, + "content": "denoising criterion. J. Mach. Learn. Res., 11:3371–3408, December 2010. ISSN 1532-4435.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 116, + 543, + 437, + 555 + ], + "spans": [ + { + "bbox": [ + 116, + 543, + 437, + 555 + ], + "score": 1.0, + "content": "URL http://dl.acm.org/citation.cfm?id=1756006.1953039.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32.5, + "bbox_fs": [ + 106, + 510, + 506, + 555 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 561, + 502, + 585 + ], + "lines": [ + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 505, + 574 + ], + "score": 1.0, + "content": "Jason Yosinski, Jeff Clune, Anh Nguyen, Thomas Fuchs, and Hod Lipson. Understanding neural", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 115, + 572, + 429, + 586 + ], + "spans": [ + { + "bbox": [ + 115, + 572, + 429, + 586 + ], + "score": 1.0, + "content": "networks through deep visualization. arXiv preprint arXiv:1506.06579, 2015.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 561, + 505, + 586 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 591, + 504, + 615 + ], + "lines": [ + { + "bbox": [ + 105, + 590, + 505, + 605 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 505, + 605 + ], + "score": 1.0, + "content": "Chongjie Zhang and Victor R Lesser. Multi-agent learning with policy prediction. In Proceedings", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 116, + 602, + 401, + 615 + ], + "spans": [ + { + "bbox": [ + 116, + 602, + 401, + 615 + ], + "score": 1.0, + "content": "of the Twenty-Fourth AAAI Conference on Artificial Intelligence, 2010.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 590, + 505, + 615 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 621, + 503, + 645 + ], + "lines": [ + { + "bbox": [ + 106, + 621, + 504, + 634 + ], + "spans": [ + { + "bbox": [ + 106, + 621, + 504, + 634 + ], + "score": 1.0, + "content": "Junbo Zhao, Michael Mathieu, and Yann LeCun. Energy-based generative adversarial network.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 116, + 632, + 278, + 645 + ], + "spans": [ + { + "bbox": [ + 116, + 632, + 278, + 645 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1609.03126, 2016.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39.5, + "bbox_fs": [ + 106, + 621, + 504, + 645 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 651, + 505, + 685 + ], + "lines": [ + { + "bbox": [ + 106, + 651, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 106, + 651, + 505, + 663 + ], + "score": 1.0, + "content": "Jun-Yan Zhu, Philipp Krahenb ¨ uhl, Eli Shechtman, and Alexei A. Efros. Generative visual manipula- ¨", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 115, + 662, + 506, + 676 + ], + "spans": [ + { + "bbox": [ + 115, + 662, + 506, + 676 + ], + "score": 1.0, + "content": "tion on the natural image manifold. In Proceedings of European Conference on Computer Vision", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 116, + 673, + 178, + 686 + ], + "spans": [ + { + "bbox": [ + 116, + 673, + 178, + 686 + ], + "score": 1.0, + "content": "(ECCV), 2016.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42, + "bbox_fs": [ + 106, + 651, + 506, + 686 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 79, + 194, + 101 + ], + "lines": [ + { + "bbox": [ + 104, + 77, + 196, + 103 + ], + "spans": [ + { + "bbox": [ + 104, + 77, + 196, + 103 + ], + "score": 1.0, + "content": "Appendix", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 107, + 118, + 306, + 132 + ], + "lines": [ + { + "bbox": [ + 105, + 117, + 307, + 134 + ], + "spans": [ + { + "bbox": [ + 105, + 117, + 307, + 134 + ], + "score": 1.0, + "content": "A 2D GAUSSIAN TRAINING DETAILS", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 143, + 486, + 155 + ], + "lines": [ + { + "bbox": [ + 106, + 142, + 487, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 142, + 487, + 156 + ], + "score": 1.0, + "content": "Network architecture and experimental details for the experiment in Section 3.1 are as follows:", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 106, + 160, + 504, + 182 + ], + "lines": [ + { + "bbox": [ + 105, + 158, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 506, + 173 + ], + "score": 1.0, + "content": "The dataset is sampled from a mixture of 8 Gaussians of standard deviation 0.02. The means are", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 172, + 277, + 183 + ], + "spans": [ + { + "bbox": [ + 106, + 172, + 277, + 183 + ], + "score": 1.0, + "content": "equally spaced around a circle of radius 2.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 107, + 188, + 505, + 221 + ], + "lines": [ + { + "bbox": [ + 106, + 188, + 505, + 200 + ], + "spans": [ + { + "bbox": [ + 106, + 188, + 505, + 200 + ], + "score": 1.0, + "content": "The generator network consists of a fully connected network with 2 hidden layers of size 128 with", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 199, + 505, + 212 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 505, + 212 + ], + "score": 1.0, + "content": "relu activations followed by a linear projection to 2 dimensions. All weights are initialized to be", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 210, + 231, + 223 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 231, + 223 + ], + "score": 1.0, + "content": "orthogonal with scaling of 0.8.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 107, + 227, + 505, + 261 + ], + "lines": [ + { + "bbox": [ + 105, + 226, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 505, + 240 + ], + "score": 1.0, + "content": "The discriminator network first scales its input down by a factor of 4 (to roughly scale to (-1,1)),", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 237, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 505, + 250 + ], + "score": 1.0, + "content": "followed by 1 layer fully connected network with relu activations to a linear layer to of size 1 to act", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 248, + 156, + 262 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 156, + 262 + ], + "score": 1.0, + "content": "as the logit.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 107, + 265, + 505, + 310 + ], + "lines": [ + { + "bbox": [ + 105, + 265, + 506, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 265, + 210, + 279 + ], + "score": 1.0, + "content": "The generator minimizes", + "type": "text" + }, + { + "bbox": [ + 210, + 266, + 370, + 278 + ], + "score": 0.92, + "content": "\\mathcal { L } _ { G } = \\log ( D ( x ) ) + \\log ( 1 - D ( G ( z ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 265, + 506, + 279 + ], + "score": 1.0, + "content": "and the discriminator minimizes", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 276, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 106, + 277, + 277, + 289 + ], + "score": 0.89, + "content": "\\mathcal { L } _ { D } \\stackrel { = } { = } - \\log ( D ( x ) ) - \\log ( 1 - D ( \\stackrel { . } { G } ( z ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 277, + 276, + 306, + 289 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 306, + 279, + 313, + 287 + ], + "score": 0.58, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 276, + 485, + 289 + ], + "score": 1.0, + "content": "is sampled from the data distribution and", + "type": "text" + }, + { + "bbox": [ + 486, + 278, + 505, + 288 + ], + "score": 0.82, + "content": "z \\sim", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 107, + 286, + 506, + 301 + ], + "spans": [ + { + "bbox": [ + 107, + 288, + 151, + 299 + ], + "score": 0.9, + "content": "\\mathcal { N } ( 0 , I _ { 2 5 6 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 286, + 506, + 301 + ], + "score": 1.0, + "content": ". Both networks are optimized using Adam (Kingma & Ba, 2014) with a learning rate of", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 299, + 177, + 310 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 143, + 310 + ], + "score": 1.0, + "content": "1e-4 and", + "type": "text" + }, + { + "bbox": [ + 144, + 299, + 173, + 310 + ], + "score": 0.9, + "content": "\\beta _ { 1 } { = } 0 . 5", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 299, + 177, + 310 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 107, + 316, + 504, + 338 + ], + "lines": [ + { + "bbox": [ + 105, + 314, + 506, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 506, + 329 + ], + "score": 1.0, + "content": "The network is trained by alternating updates of the generator and the discriminator. One step", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 324, + 247, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 176, + 340 + ], + "score": 1.0, + "content": "consists of either", + "type": "text" + }, + { + "bbox": [ + 176, + 327, + 185, + 337 + ], + "score": 0.54, + "content": "\\mathbf { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 324, + 197, + 340 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 197, + 327, + 206, + 337 + ], + "score": 0.6, + "content": "\\mathbf { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 324, + 247, + 340 + ], + "score": 1.0, + "content": "updating.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "title", + "bbox": [ + 108, + 353, + 364, + 366 + ], + "lines": [ + { + "bbox": [ + 105, + 353, + 365, + 367 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 365, + 367 + ], + "score": 1.0, + "content": "B MORE MIXTURE OF GAUSSIAN EXPERIMENTS", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "title", + "bbox": [ + 106, + 378, + 353, + 390 + ], + "lines": [ + { + "bbox": [ + 106, + 378, + 354, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 354, + 390 + ], + "score": 1.0, + "content": "B.1 EFFECTS OF TIME DELAY / HISTORICAL AVERAGING", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 398, + 505, + 432 + ], + "lines": [ + { + "bbox": [ + 105, + 397, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 505, + 412 + ], + "score": 1.0, + "content": "Another comparison we looked at was with regard to historical averaging based approaches. Re-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 408, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 505, + 423 + ], + "score": 1.0, + "content": "cently similarly inspired approaches have been used in (Salimans et al., 2016) to stabilize training.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 421, + 410, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 410, + 433 + ], + "score": 1.0, + "content": "For our study, we looked at taking an ensemble of discriminators over time.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 437, + 444, + 450 + ], + "lines": [ + { + "bbox": [ + 105, + 435, + 447, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 447, + 452 + ], + "score": 1.0, + "content": "First, we looked at taking an ensemble of the last N steps, as shown in Figure App.1.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "image", + "bbox": [ + 107, + 461, + 503, + 571 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 461, + 503, + 571 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 461, + 503, + 571 + ], + "spans": [ + { + "bbox": [ + 107, + 461, + 503, + 571 + ], + "score": 0.967, + "type": "image", + "image_path": "ca00dc71a894afed5412e7d9c9476916ab9ab1f2648dedea7d529b310373bd67.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 107, + 461, + 503, + 497.6666666666667 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 107, + 497.6666666666667, + 503, + 534.3333333333334 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 107, + 534.3333333333334, + 503, + 571.0 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 582, + 505, + 627 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "Figure App.1: Historical averaging does not visibly increase stability on the mixture of Gaussians", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 595, + 505, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 505, + 605 + ], + "score": 1.0, + "content": "task. Each row corresponds to an ensemble of discriminators which consists of the indicated number", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "score": 1.0, + "content": "of immediately preceding discriminators. The columns correspond to different numbers of training", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 615, + 132, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 132, + 630 + ], + "score": 1.0, + "content": "steps.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5 + } + ], + "index": 25.75 + }, + { + "type": "text", + "bbox": [ + 107, + 638, + 505, + 682 + ], + "lines": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "To further explore this idea, we ran experiments with an ensemble of 5 discriminators, but with", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 649, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 106, + 649, + 506, + 661 + ], + "score": 1.0, + "content": "different periods between replacing discriminators in the ensemble. For example, if I sample at a", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 660, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 505, + 672 + ], + "score": 1.0, + "content": "rate of 100, it would take 500 steps to replace all 5 discriminators. Results can be seen in Figure", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 669, + 138, + 685 + ], + "spans": [ + { + "bbox": [ + 104, + 669, + 138, + 685 + ], + "score": 1.0, + "content": "App.2.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "We observe that given longer and longer time delays, the model becomes less and less stable. We", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 699, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 506, + 711 + ], + "score": 1.0, + "content": "hypothesize that this is due to the initial shape of the discriminator loss surface. When training, the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "discriminator’s estimates of probability densities are only accurate on regions where it was trained.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "score": 1.0, + "content": "When fixing this discriminator, we are removing the feedback between the generator exploitation", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35.5 + } + ], + "page_idx": 14, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 301, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 79, + 194, + 101 + ], + "lines": [ + { + "bbox": [ + 104, + 77, + 196, + 103 + ], + "spans": [ + { + "bbox": [ + 104, + 77, + 196, + 103 + ], + "score": 1.0, + "content": "Appendix", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 107, + 118, + 306, + 132 + ], + "lines": [ + { + "bbox": [ + 105, + 117, + 307, + 134 + ], + "spans": [ + { + "bbox": [ + 105, + 117, + 307, + 134 + ], + "score": 1.0, + "content": "A 2D GAUSSIAN TRAINING DETAILS", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 143, + 486, + 155 + ], + "lines": [ + { + "bbox": [ + 106, + 142, + 487, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 142, + 487, + 156 + ], + "score": 1.0, + "content": "Network architecture and experimental details for the experiment in Section 3.1 are as follows:", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2, + "bbox_fs": [ + 106, + 142, + 487, + 156 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 160, + 504, + 182 + ], + "lines": [ + { + "bbox": [ + 105, + 158, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 158, + 506, + 173 + ], + "score": 1.0, + "content": "The dataset is sampled from a mixture of 8 Gaussians of standard deviation 0.02. The means are", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 172, + 277, + 183 + ], + "spans": [ + { + "bbox": [ + 106, + 172, + 277, + 183 + ], + "score": 1.0, + "content": "equally spaced around a circle of radius 2.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 105, + 158, + 506, + 183 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 188, + 505, + 221 + ], + "lines": [ + { + "bbox": [ + 106, + 188, + 505, + 200 + ], + "spans": [ + { + "bbox": [ + 106, + 188, + 505, + 200 + ], + "score": 1.0, + "content": "The generator network consists of a fully connected network with 2 hidden layers of size 128 with", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 199, + 505, + 212 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 505, + 212 + ], + "score": 1.0, + "content": "relu activations followed by a linear projection to 2 dimensions. All weights are initialized to be", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 210, + 231, + 223 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 231, + 223 + ], + "score": 1.0, + "content": "orthogonal with scaling of 0.8.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6, + "bbox_fs": [ + 106, + 188, + 505, + 223 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 227, + 505, + 261 + ], + "lines": [ + { + "bbox": [ + 105, + 226, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 505, + 240 + ], + "score": 1.0, + "content": "The discriminator network first scales its input down by a factor of 4 (to roughly scale to (-1,1)),", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 237, + 505, + 250 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 505, + 250 + ], + "score": 1.0, + "content": "followed by 1 layer fully connected network with relu activations to a linear layer to of size 1 to act", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 248, + 156, + 262 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 156, + 262 + ], + "score": 1.0, + "content": "as the logit.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 226, + 505, + 262 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 265, + 505, + 310 + ], + "lines": [ + { + "bbox": [ + 105, + 265, + 506, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 265, + 210, + 279 + ], + "score": 1.0, + "content": "The generator minimizes", + "type": "text" + }, + { + "bbox": [ + 210, + 266, + 370, + 278 + ], + "score": 0.92, + "content": "\\mathcal { L } _ { G } = \\log ( D ( x ) ) + \\log ( 1 - D ( G ( z ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 265, + 506, + 279 + ], + "score": 1.0, + "content": "and the discriminator minimizes", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 276, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 106, + 277, + 277, + 289 + ], + "score": 0.89, + "content": "\\mathcal { L } _ { D } \\stackrel { = } { = } - \\log ( D ( x ) ) - \\log ( 1 - D ( \\stackrel { . } { G } ( z ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 277, + 276, + 306, + 289 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 306, + 279, + 313, + 287 + ], + "score": 0.58, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 314, + 276, + 485, + 289 + ], + "score": 1.0, + "content": "is sampled from the data distribution and", + "type": "text" + }, + { + "bbox": [ + 486, + 278, + 505, + 288 + ], + "score": 0.82, + "content": "z \\sim", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 107, + 286, + 506, + 301 + ], + "spans": [ + { + "bbox": [ + 107, + 288, + 151, + 299 + ], + "score": 0.9, + "content": "\\mathcal { N } ( 0 , I _ { 2 5 6 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 286, + 506, + 301 + ], + "score": 1.0, + "content": ". Both networks are optimized using Adam (Kingma & Ba, 2014) with a learning rate of", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 299, + 177, + 310 + ], + "spans": [ + { + "bbox": [ + 106, + 299, + 143, + 310 + ], + "score": 1.0, + "content": "1e-4 and", + "type": "text" + }, + { + "bbox": [ + 144, + 299, + 173, + 310 + ], + "score": 0.9, + "content": "\\beta _ { 1 } { = } 0 . 5", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 299, + 177, + 310 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 265, + 506, + 310 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 316, + 504, + 338 + ], + "lines": [ + { + "bbox": [ + 105, + 314, + 506, + 329 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 506, + 329 + ], + "score": 1.0, + "content": "The network is trained by alternating updates of the generator and the discriminator. One step", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 324, + 247, + 340 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 176, + 340 + ], + "score": 1.0, + "content": "consists of either", + "type": "text" + }, + { + "bbox": [ + 176, + 327, + 185, + 337 + ], + "score": 0.54, + "content": "\\mathbf { G }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 324, + 197, + 340 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 197, + 327, + 206, + 337 + ], + "score": 0.6, + "content": "\\mathbf { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 206, + 324, + 247, + 340 + ], + "score": 1.0, + "content": "updating.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 314, + 506, + 340 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 353, + 364, + 366 + ], + "lines": [ + { + "bbox": [ + 105, + 353, + 365, + 367 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 365, + 367 + ], + "score": 1.0, + "content": "B MORE MIXTURE OF GAUSSIAN EXPERIMENTS", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "title", + "bbox": [ + 106, + 378, + 353, + 390 + ], + "lines": [ + { + "bbox": [ + 106, + 378, + 354, + 390 + ], + "spans": [ + { + "bbox": [ + 106, + 378, + 354, + 390 + ], + "score": 1.0, + "content": "B.1 EFFECTS OF TIME DELAY / HISTORICAL AVERAGING", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 398, + 505, + 432 + ], + "lines": [ + { + "bbox": [ + 105, + 397, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 397, + 505, + 412 + ], + "score": 1.0, + "content": "Another comparison we looked at was with regard to historical averaging based approaches. Re-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 408, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 408, + 505, + 423 + ], + "score": 1.0, + "content": "cently similarly inspired approaches have been used in (Salimans et al., 2016) to stabilize training.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 421, + 410, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 421, + 410, + 433 + ], + "score": 1.0, + "content": "For our study, we looked at taking an ensemble of discriminators over time.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 397, + 505, + 433 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 437, + 444, + 450 + ], + "lines": [ + { + "bbox": [ + 105, + 435, + 447, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 447, + 452 + ], + "score": 1.0, + "content": "First, we looked at taking an ensemble of the last N steps, as shown in Figure App.1.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22, + "bbox_fs": [ + 105, + 435, + 447, + 452 + ] + }, + { + "type": "image", + "bbox": [ + 107, + 461, + 503, + 571 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 461, + 503, + 571 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 461, + 503, + 571 + ], + "spans": [ + { + "bbox": [ + 107, + 461, + 503, + 571 + ], + "score": 0.967, + "type": "image", + "image_path": "ca00dc71a894afed5412e7d9c9476916ab9ab1f2648dedea7d529b310373bd67.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 107, + 461, + 503, + 497.6666666666667 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 107, + 497.6666666666667, + 503, + 534.3333333333334 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 107, + 534.3333333333334, + 503, + 571.0 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 582, + 505, + 627 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 583, + 505, + 595 + ], + "spans": [ + { + "bbox": [ + 106, + 583, + 505, + 595 + ], + "score": 1.0, + "content": "Figure App.1: Historical averaging does not visibly increase stability on the mixture of Gaussians", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 595, + 505, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 505, + 605 + ], + "score": 1.0, + "content": "task. Each row corresponds to an ensemble of discriminators which consists of the indicated number", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 605, + 505, + 618 + ], + "score": 1.0, + "content": "of immediately preceding discriminators. The columns correspond to different numbers of training", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 615, + 132, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 132, + 630 + ], + "score": 1.0, + "content": "steps.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5 + } + ], + "index": 25.75 + }, + { + "type": "text", + "bbox": [ + 107, + 638, + 505, + 682 + ], + "lines": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "To further explore this idea, we ran experiments with an ensemble of 5 discriminators, but with", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 649, + 506, + 661 + ], + "spans": [ + { + "bbox": [ + 106, + 649, + 506, + 661 + ], + "score": 1.0, + "content": "different periods between replacing discriminators in the ensemble. For example, if I sample at a", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 660, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 505, + 672 + ], + "score": 1.0, + "content": "rate of 100, it would take 500 steps to replace all 5 discriminators. Results can be seen in Figure", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 104, + 669, + 138, + 685 + ], + "spans": [ + { + "bbox": [ + 104, + 669, + 138, + 685 + ], + "score": 1.0, + "content": "App.2.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31.5, + "bbox_fs": [ + 104, + 637, + 506, + 685 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 687, + 506, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 506, + 700 + ], + "score": 1.0, + "content": "We observe that given longer and longer time delays, the model becomes less and less stable. We", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 699, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 506, + 711 + ], + "score": 1.0, + "content": "hypothesize that this is due to the initial shape of the discriminator loss surface. When training, the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "discriminator’s estimates of probability densities are only accurate on regions where it was trained.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "score": 1.0, + "content": "When fixing this discriminator, we are removing the feedback between the generator exploitation", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 271, + 505, + 283 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 505, + 283 + ], + "score": 1.0, + "content": "and the discriminators ability to move. As a result, the generator is able to exploit these fixed areas", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 282, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 106, + 282, + 505, + 294 + ], + "score": 1.0, + "content": "of poor performance for older discriminators in the ensemble. New discriminators (over)compensate", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 291, + 261, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 261, + 307 + ], + "score": 1.0, + "content": "for this, leading the system to diverge.", + "type": "text", + "cross_page": true + } + ], + "index": 9 + } + ], + "index": 35.5, + "bbox_fs": [ + 105, + 687, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 107, + 81, + 503, + 192 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 81, + 503, + 192 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 81, + 503, + 192 + ], + "spans": [ + { + "bbox": [ + 107, + 81, + 503, + 192 + ], + "score": 0.946, + "type": "image", + "image_path": "04adbd77c06eb06d8623be8ab50fe44136a6a0398a13d8fdd2a583c617dbcff0.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 107, + 81, + 503, + 118.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 107, + 118.0, + 503, + 155.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 107, + 155.0, + 503, + 192.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 204, + 505, + 249 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 205, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 106, + 205, + 505, + 217 + ], + "score": 1.0, + "content": "Figure App.2: Introducing longer time delays between the discriminator ensemble results in insta-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 215, + 505, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 452, + 227 + ], + "score": 1.0, + "content": "bility and probability distributions that are not in the window being visualized. The", + "type": "text" + }, + { + "bbox": [ + 452, + 217, + 460, + 226 + ], + "score": 0.37, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 215, + 505, + 227 + ], + "score": 1.0, + "content": "axis is the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 227, + 505, + 239 + ], + "spans": [ + { + "bbox": [ + 106, + 227, + 505, + 239 + ], + "score": 1.0, + "content": "number of weight updates and the y axis is how many steps to skip between discriminator updates", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 107, + 239, + 303, + 249 + ], + "spans": [ + { + "bbox": [ + 107, + 239, + 303, + 249 + ], + "score": 1.0, + "content": "when selecting the ensemble of 5 discriminators.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 107, + 271, + 504, + 304 + ], + "lines": [ + { + "bbox": [ + 106, + 271, + 505, + 283 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 505, + 283 + ], + "score": 1.0, + "content": "and the discriminators ability to move. As a result, the generator is able to exploit these fixed areas", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 282, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 106, + 282, + 505, + 294 + ], + "score": 1.0, + "content": "of poor performance for older discriminators in the ensemble. New discriminators (over)compensate", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 291, + 261, + 307 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 261, + 307 + ], + "score": 1.0, + "content": "for this, leading the system to diverge.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8 + }, + { + "type": "title", + "bbox": [ + 108, + 319, + 288, + 330 + ], + "lines": [ + { + "bbox": [ + 105, + 318, + 290, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 290, + 331 + ], + "score": 1.0, + "content": "B.2 EFFECTS OF THE SECOND GRADIENT", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 107, + 340, + 505, + 417 + ], + "lines": [ + { + "bbox": [ + 105, + 340, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 505, + 354 + ], + "score": 1.0, + "content": "A second factor we analyzed is the effect of backpropagating the learning signal through the un-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 351, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 505, + 364 + ], + "score": 1.0, + "content": "rolling in Equation 12. We can turn on or off this backpropagation through the unrolling by in-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 362, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 505, + 374 + ], + "score": 1.0, + "content": "troducing stop gradient calls into our computation graph between each unrolling step. With the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 374, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 505, + 385 + ], + "score": 1.0, + "content": "stop gradient in place, the update signal corresponds only to the first term in Equation 12. We", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 384, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 505, + 396 + ], + "score": 1.0, + "content": "looked at 3 configurations: without stop gradients; vanilla unrolled GAN, with stop gradients; and", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 395, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 307, + 408 + ], + "score": 1.0, + "content": "with stop gradients but taking the average over the", + "type": "text" + }, + { + "bbox": [ + 307, + 396, + 314, + 405 + ], + "score": 0.8, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 395, + 505, + 408 + ], + "score": 1.0, + "content": "unrolling steps instead of taking the final value.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 406, + 250, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 250, + 419 + ], + "score": 1.0, + "content": "Results can be see in Figure App.3.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 107, + 422, + 505, + 478 + ], + "lines": [ + { + "bbox": [ + 107, + 423, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 107, + 423, + 505, + 435 + ], + "score": 1.0, + "content": "We initially observed no difference between unrolling with and without the second gradient, as both", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 432, + 506, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 506, + 448 + ], + "score": 1.0, + "content": "required 3 unrolling steps to become stable. When the discriminator is unrolled to convergence,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 445, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 106, + 445, + 505, + 457 + ], + "score": 1.0, + "content": "the second gradient term becomes zero. Due to the simplicity of the problem, we suspect that the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 455, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 505, + 469 + ], + "score": 1.0, + "content": "discriminator nearly converged for every generator step, and the second gradient term was thus", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 466, + 149, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 149, + 479 + ], + "score": 1.0, + "content": "irrelevant.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 107, + 483, + 504, + 527 + ], + "lines": [ + { + "bbox": [ + 105, + 483, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 505, + 496 + ], + "score": 1.0, + "content": "To test this, we modified the dynamics to perform five generator steps for each discriminator update.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 494, + 506, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 494, + 506, + 507 + ], + "score": 1.0, + "content": "Results are shown in Figure App.4. With the discriminator now kept out of equilibrium, successful", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 506, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 505, + 518 + ], + "score": 1.0, + "content": "training can be achieved with half as many unrolling steps when using both terms in the gradient", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 516, + 266, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 266, + 529 + ], + "score": 1.0, + "content": "than when only including the first term.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5 + }, + { + "type": "title", + "bbox": [ + 107, + 545, + 302, + 558 + ], + "lines": [ + { + "bbox": [ + 105, + 543, + 303, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 303, + 560 + ], + "score": 1.0, + "content": "C RNN MNIST TRAINING DETAILS", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 108, + 571, + 396, + 582 + ], + "lines": [ + { + "bbox": [ + 106, + 570, + 398, + 584 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 398, + 584 + ], + "score": 1.0, + "content": "The network architecture for the experiment in Section 3.2 is as follows:", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 108, + 588, + 263, + 599 + ], + "lines": [ + { + "bbox": [ + 106, + 586, + 264, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 586, + 264, + 601 + ], + "score": 1.0, + "content": "The MNIST dataset is scaled to [-1, 1).", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 604, + 505, + 671 + ], + "lines": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "The generator first scales the 256D noise vector through a 256 unit fully connected layer with relu", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 615, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 505, + 628 + ], + "score": 1.0, + "content": "activation. This is then fed into the initial state of a 256D LSTM(Hochreiter & Schmidhuber, 1997)", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "score": 1.0, + "content": "that runs 28 steps corresponding to the number of columns in MNIST. The resulting sequence of ac-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "tivations is projected through a fully connected layer with 28 outputs with a tanh activation function.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 649, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 106, + 649, + 505, + 660 + ], + "score": 1.0, + "content": "All weights are initialized via the ”Xavier” initialization (Glorot & Bengio, 2010). The forget bias", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 660, + 235, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 235, + 671 + ], + "score": 1.0, + "content": "on the LSTM is initialized to 1.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 108, + 677, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 397, + 689 + ], + "score": 1.0, + "content": "The discriminator network feeds the input into a Convolution(16, stride", + "type": "text" + }, + { + "bbox": [ + 397, + 678, + 409, + 687 + ], + "score": 0.71, + "content": "^ { = 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 677, + 505, + 689 + ], + "score": 1.0, + "content": ") followed by a Convo-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 687, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 171, + 699 + ], + "score": 1.0, + "content": "lution(32, stride", + "type": "text" + }, + { + "bbox": [ + 171, + 688, + 183, + 698 + ], + "score": 0.48, + "content": "^ { = 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 687, + 330, + 699 + ], + "score": 1.0, + "content": ") followed by Convolution(32, stride", + "type": "text" + }, + { + "bbox": [ + 330, + 688, + 342, + 698 + ], + "score": 0.68, + "content": "{ \\boldsymbol { \\mathbf { \\mathit { \\varepsilon } } } } = 2 { \\boldsymbol { \\mathbf { \\mathit { \\varepsilon } } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 687, + 505, + 699 + ], + "score": 1.0, + "content": "). All convolutions have stride 2. As in", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "(Radford et al., 2015) leaky rectifiers are used with a 0.3 leak. Batch normalization is applied after", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "each layer (Ioffe & Szegedy, 2015). The resulting 4D tensor is then flattened and a linear projection", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 720, + 231, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 231, + 733 + ], + "score": 1.0, + "content": "is performed to a single scalar.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38 + } + ], + "page_idx": 15, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 301, + 752, + 311, + 760 + ], + "lines": [] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 107, + 81, + 503, + 192 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 81, + 503, + 192 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 81, + 503, + 192 + ], + "spans": [ + { + "bbox": [ + 107, + 81, + 503, + 192 + ], + "score": 0.946, + "type": "image", + "image_path": "04adbd77c06eb06d8623be8ab50fe44136a6a0398a13d8fdd2a583c617dbcff0.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 107, + 81, + 503, + 118.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 107, + 118.0, + 503, + 155.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 107, + 155.0, + 503, + 192.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 204, + 505, + 249 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 205, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 106, + 205, + 505, + 217 + ], + "score": 1.0, + "content": "Figure App.2: Introducing longer time delays between the discriminator ensemble results in insta-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 215, + 505, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 215, + 452, + 227 + ], + "score": 1.0, + "content": "bility and probability distributions that are not in the window being visualized. The", + "type": "text" + }, + { + "bbox": [ + 452, + 217, + 460, + 226 + ], + "score": 0.37, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 460, + 215, + 505, + 227 + ], + "score": 1.0, + "content": "axis is the", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 227, + 505, + 239 + ], + "spans": [ + { + "bbox": [ + 106, + 227, + 505, + 239 + ], + "score": 1.0, + "content": "number of weight updates and the y axis is how many steps to skip between discriminator updates", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 107, + 239, + 303, + 249 + ], + "spans": [ + { + "bbox": [ + 107, + 239, + 303, + 249 + ], + "score": 1.0, + "content": "when selecting the ensemble of 5 discriminators.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 107, + 271, + 504, + 304 + ], + "lines": [], + "index": 8, + "bbox_fs": [ + 105, + 271, + 505, + 307 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 108, + 319, + 288, + 330 + ], + "lines": [ + { + "bbox": [ + 105, + 318, + 290, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 290, + 331 + ], + "score": 1.0, + "content": "B.2 EFFECTS OF THE SECOND GRADIENT", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 107, + 340, + 505, + 417 + ], + "lines": [ + { + "bbox": [ + 105, + 340, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 505, + 354 + ], + "score": 1.0, + "content": "A second factor we analyzed is the effect of backpropagating the learning signal through the un-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 351, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 351, + 505, + 364 + ], + "score": 1.0, + "content": "rolling in Equation 12. We can turn on or off this backpropagation through the unrolling by in-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 362, + 505, + 374 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 505, + 374 + ], + "score": 1.0, + "content": "troducing stop gradient calls into our computation graph between each unrolling step. With the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 374, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 105, + 374, + 505, + 385 + ], + "score": 1.0, + "content": "stop gradient in place, the update signal corresponds only to the first term in Equation 12. We", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 384, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 505, + 396 + ], + "score": 1.0, + "content": "looked at 3 configurations: without stop gradients; vanilla unrolled GAN, with stop gradients; and", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 395, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 307, + 408 + ], + "score": 1.0, + "content": "with stop gradients but taking the average over the", + "type": "text" + }, + { + "bbox": [ + 307, + 396, + 314, + 405 + ], + "score": 0.8, + "content": "k", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 395, + 505, + 408 + ], + "score": 1.0, + "content": "unrolling steps instead of taking the final value.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 406, + 250, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 406, + 250, + 419 + ], + "score": 1.0, + "content": "Results can be see in Figure App.3.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 340, + 505, + 419 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 422, + 505, + 478 + ], + "lines": [ + { + "bbox": [ + 107, + 423, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 107, + 423, + 505, + 435 + ], + "score": 1.0, + "content": "We initially observed no difference between unrolling with and without the second gradient, as both", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 432, + 506, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 506, + 448 + ], + "score": 1.0, + "content": "required 3 unrolling steps to become stable. When the discriminator is unrolled to convergence,", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 445, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 106, + 445, + 505, + 457 + ], + "score": 1.0, + "content": "the second gradient term becomes zero. Due to the simplicity of the problem, we suspect that the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 455, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 505, + 469 + ], + "score": 1.0, + "content": "discriminator nearly converged for every generator step, and the second gradient term was thus", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 466, + 149, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 466, + 149, + 479 + ], + "score": 1.0, + "content": "irrelevant.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 423, + 506, + 479 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 483, + 504, + 527 + ], + "lines": [ + { + "bbox": [ + 105, + 483, + 505, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 483, + 505, + 496 + ], + "score": 1.0, + "content": "To test this, we modified the dynamics to perform five generator steps for each discriminator update.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 494, + 506, + 507 + ], + "spans": [ + { + "bbox": [ + 105, + 494, + 506, + 507 + ], + "score": 1.0, + "content": "Results are shown in Figure App.4. With the discriminator now kept out of equilibrium, successful", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 506, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 106, + 506, + 505, + 518 + ], + "score": 1.0, + "content": "training can be achieved with half as many unrolling steps when using both terms in the gradient", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 516, + 266, + 529 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 266, + 529 + ], + "score": 1.0, + "content": "than when only including the first term.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 483, + 506, + 529 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 545, + 302, + 558 + ], + "lines": [ + { + "bbox": [ + 105, + 543, + 303, + 560 + ], + "spans": [ + { + "bbox": [ + 105, + 543, + 303, + 560 + ], + "score": 1.0, + "content": "C RNN MNIST TRAINING DETAILS", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 108, + 571, + 396, + 582 + ], + "lines": [ + { + "bbox": [ + 106, + 570, + 398, + 584 + ], + "spans": [ + { + "bbox": [ + 106, + 570, + 398, + 584 + ], + "score": 1.0, + "content": "The network architecture for the experiment in Section 3.2 is as follows:", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28, + "bbox_fs": [ + 106, + 570, + 398, + 584 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 588, + 263, + 599 + ], + "lines": [ + { + "bbox": [ + 106, + 586, + 264, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 586, + 264, + 601 + ], + "score": 1.0, + "content": "The MNIST dataset is scaled to [-1, 1).", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29, + "bbox_fs": [ + 106, + 586, + 264, + 601 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 604, + 505, + 671 + ], + "lines": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "The generator first scales the 256D noise vector through a 256 unit fully connected layer with relu", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 615, + 505, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 505, + 628 + ], + "score": 1.0, + "content": "activation. This is then fed into the initial state of a 256D LSTM(Hochreiter & Schmidhuber, 1997)", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "score": 1.0, + "content": "that runs 28 steps corresponding to the number of columns in MNIST. The resulting sequence of ac-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 505, + 650 + ], + "score": 1.0, + "content": "tivations is projected through a fully connected layer with 28 outputs with a tanh activation function.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 649, + 505, + 660 + ], + "spans": [ + { + "bbox": [ + 106, + 649, + 505, + 660 + ], + "score": 1.0, + "content": "All weights are initialized via the ”Xavier” initialization (Glorot & Bengio, 2010). The forget bias", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 660, + 235, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 660, + 235, + 671 + ], + "score": 1.0, + "content": "on the LSTM is initialized to 1.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 605, + 505, + 671 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 677, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 505, + 689 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 397, + 689 + ], + "score": 1.0, + "content": "The discriminator network feeds the input into a Convolution(16, stride", + "type": "text" + }, + { + "bbox": [ + 397, + 678, + 409, + 687 + ], + "score": 0.71, + "content": "^ { = 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 677, + 505, + 689 + ], + "score": 1.0, + "content": ") followed by a Convo-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 687, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 171, + 699 + ], + "score": 1.0, + "content": "lution(32, stride", + "type": "text" + }, + { + "bbox": [ + 171, + 688, + 183, + 698 + ], + "score": 0.48, + "content": "^ { = 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 687, + 330, + 699 + ], + "score": 1.0, + "content": ") followed by Convolution(32, stride", + "type": "text" + }, + { + "bbox": [ + 330, + 688, + 342, + 698 + ], + "score": 0.68, + "content": "{ \\boldsymbol { \\mathbf { \\mathit { \\varepsilon } } } } = 2 { \\boldsymbol { \\mathbf { \\mathit { \\varepsilon } } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 687, + 505, + 699 + ], + "score": 1.0, + "content": "). All convolutions have stride 2. As in", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 698, + 506, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 506, + 711 + ], + "score": 1.0, + "content": "(Radford et al., 2015) leaky rectifiers are used with a 0.3 leak. Batch normalization is applied after", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 723 + ], + "score": 1.0, + "content": "each layer (Ioffe & Szegedy, 2015). The resulting 4D tensor is then flattened and a linear projection", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 720, + 231, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 231, + 733 + ], + "score": 1.0, + "content": "is performed to a single scalar.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38, + "bbox_fs": [ + 105, + 677, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 104, + 77, + 505, + 424 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 104, + 77, + 505, + 424 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 104, + 77, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 104, + 77, + 505, + 424 + ], + "score": 0.56, + "type": "image", + "image_path": "e082e1fa15fe3c8c65693a9ca8c06d3ede7c2ddcf2d0287c257abe4e8c2ff9b6.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 104, + 77, + 505, + 192.66666666666669 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 104, + 192.66666666666669, + 505, + 308.33333333333337 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 104, + 308.33333333333337, + 505, + 424.00000000000006 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 435, + 505, + 480 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 435, + 505, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 505, + 448 + ], + "score": 1.0, + "content": "Figure App.3: If the discriminator remains nearly at its optimum during learning, then performance", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 446, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 505, + 459 + ], + "score": 1.0, + "content": "is nearly identical with and without the second gradient term in Equation 12. As shown in Figure", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 458, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 458, + 505, + 471 + ], + "score": 1.0, + "content": "App.4, when the discriminator lags behind the generator, backpropagating through unrolling aids", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 470, + 161, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 161, + 480 + ], + "score": 1.0, + "content": "convergence.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 107, + 505, + 505, + 550 + ], + "lines": [ + { + "bbox": [ + 105, + 505, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 247, + 518 + ], + "score": 1.0, + "content": "The generator network minimises", + "type": "text" + }, + { + "bbox": [ + 248, + 505, + 337, + 518 + ], + "score": 0.92, + "content": "\\mathcal { L } _ { G } = \\log ( D ( G ( z ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 505, + 475, + 518 + ], + "score": 1.0, + "content": "and the discriminator minimizes", + "type": "text" + }, + { + "bbox": [ + 476, + 506, + 505, + 517 + ], + "score": 0.87, + "content": "\\mathcal { L } _ { D } =", + "type": "inline_equation" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 516, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 106, + 516, + 237, + 529 + ], + "score": 0.91, + "content": "\\log ( D ( x ) ) + \\log ( 1 - D ( G ( z ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 516, + 505, + 529 + ], + "score": 1.0, + "content": ". Both networks are trained with Adam(Kingma & Ba, 2014) with", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 528, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 210, + 540 + ], + "score": 1.0, + "content": "learning rates of 1e-4 and", + "type": "text" + }, + { + "bbox": [ + 211, + 528, + 240, + 539 + ], + "score": 0.89, + "content": "\\beta _ { 1 } { = } 0 . 5", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 528, + 505, + 540 + ], + "score": 1.0, + "content": ". The network is trained alternating updating the generator and the", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 538, + 397, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 538, + 397, + 551 + ], + "score": 1.0, + "content": "discriminator for 150k steps. One step consists of just 1 network update.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8.5 + }, + { + "type": "title", + "bbox": [ + 107, + 571, + 327, + 585 + ], + "lines": [ + { + "bbox": [ + 105, + 570, + 328, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 328, + 587 + ], + "score": 1.0, + "content": "D CIFAR10/MNIST TRAINING DETAILS", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 108, + 600, + 504, + 623 + ], + "lines": [ + { + "bbox": [ + 106, + 600, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 505, + 613 + ], + "score": 1.0, + "content": "The network architectures for the discriminator, generator, and encoder as as follows. All convolu-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 611, + 469, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 611, + 469, + 623 + ], + "score": 1.0, + "content": "tions have a kernel size of 3x3 with batch normalization and leaky ReLU’s with a 0.3 leak.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 108, + 628, + 253, + 640 + ], + "lines": [ + { + "bbox": [ + 106, + 628, + 254, + 640 + ], + "spans": [ + { + "bbox": [ + 106, + 628, + 254, + 640 + ], + "score": 1.0, + "content": "The generator network is defined as:", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "table", + "bbox": [ + 107, + 645, + 332, + 736 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 107, + 645, + 332, + 736 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 645, + 332, + 736 + ], + "spans": [ + { + "bbox": [ + 107, + 645, + 332, + 736 + ], + "score": 0.972, + "html": "
number outputsstride
Input: z ~ N(0,I256)Fully connectedReshape to image 4,4,512ConvolutionConvolutionConvolutionConvolution4*4*512256128641or32221
", + "type": "table", + "image_path": "6be1d7b077edb834c2bd6ee68d568dbb574dbb6746300a22e384363f25abe428.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 107, + 645, + 332, + 658.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 107, + 658.0, + 332, + 671.0 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 107, + 671.0, + 332, + 684.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 107, + 684.0, + 332, + 697.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 107, + 697.0, + 332, + 710.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 107, + 710.0, + 332, + 723.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 107, + 723.0, + 332, + 736.0 + ], + "spans": [], + "index": 21 + } + ] + } + ], + "index": 18 + } + ], + "page_idx": 16, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "17", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 104, + 77, + 505, + 424 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 104, + 77, + 505, + 424 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 104, + 77, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 104, + 77, + 505, + 424 + ], + "score": 0.56, + "type": "image", + "image_path": "e082e1fa15fe3c8c65693a9ca8c06d3ede7c2ddcf2d0287c257abe4e8c2ff9b6.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 104, + 77, + 505, + 192.66666666666669 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 104, + 192.66666666666669, + 505, + 308.33333333333337 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 104, + 308.33333333333337, + 505, + 424.00000000000006 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 435, + 505, + 480 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 435, + 505, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 505, + 448 + ], + "score": 1.0, + "content": "Figure App.3: If the discriminator remains nearly at its optimum during learning, then performance", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 446, + 505, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 505, + 459 + ], + "score": 1.0, + "content": "is nearly identical with and without the second gradient term in Equation 12. As shown in Figure", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 458, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 458, + 505, + 471 + ], + "score": 1.0, + "content": "App.4, when the discriminator lags behind the generator, backpropagating through unrolling aids", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 470, + 161, + 480 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 161, + 480 + ], + "score": 1.0, + "content": "convergence.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 4.5 + } + ], + "index": 2.75 + }, + { + "type": "text", + "bbox": [ + 107, + 505, + 505, + 550 + ], + "lines": [ + { + "bbox": [ + 105, + 505, + 505, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 505, + 247, + 518 + ], + "score": 1.0, + "content": "The generator network minimises", + "type": "text" + }, + { + "bbox": [ + 248, + 505, + 337, + 518 + ], + "score": 0.92, + "content": "\\mathcal { L } _ { G } = \\log ( D ( G ( z ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 505, + 475, + 518 + ], + "score": 1.0, + "content": "and the discriminator minimizes", + "type": "text" + }, + { + "bbox": [ + 476, + 506, + 505, + 517 + ], + "score": 0.87, + "content": "\\mathcal { L } _ { D } =", + "type": "inline_equation" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 516, + 505, + 529 + ], + "spans": [ + { + "bbox": [ + 106, + 516, + 237, + 529 + ], + "score": 0.91, + "content": "\\log ( D ( x ) ) + \\log ( 1 - D ( G ( z ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 516, + 505, + 529 + ], + "score": 1.0, + "content": ". Both networks are trained with Adam(Kingma & Ba, 2014) with", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 528, + 505, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 528, + 210, + 540 + ], + "score": 1.0, + "content": "learning rates of 1e-4 and", + "type": "text" + }, + { + "bbox": [ + 211, + 528, + 240, + 539 + ], + "score": 0.89, + "content": "\\beta _ { 1 } { = } 0 . 5", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 528, + 505, + 540 + ], + "score": 1.0, + "content": ". The network is trained alternating updating the generator and the", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 538, + 397, + 551 + ], + "spans": [ + { + "bbox": [ + 106, + 538, + 397, + 551 + ], + "score": 1.0, + "content": "discriminator for 150k steps. One step consists of just 1 network update.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8.5, + "bbox_fs": [ + 105, + 505, + 505, + 551 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 571, + 327, + 585 + ], + "lines": [ + { + "bbox": [ + 105, + 570, + 328, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 328, + 587 + ], + "score": 1.0, + "content": "D CIFAR10/MNIST TRAINING DETAILS", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 108, + 600, + 504, + 623 + ], + "lines": [ + { + "bbox": [ + 106, + 600, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 505, + 613 + ], + "score": 1.0, + "content": "The network architectures for the discriminator, generator, and encoder as as follows. All convolu-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 611, + 469, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 611, + 469, + 623 + ], + "score": 1.0, + "content": "tions have a kernel size of 3x3 with batch normalization and leaky ReLU’s with a 0.3 leak.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 106, + 600, + 505, + 623 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 628, + 253, + 640 + ], + "lines": [ + { + "bbox": [ + 106, + 628, + 254, + 640 + ], + "spans": [ + { + "bbox": [ + 106, + 628, + 254, + 640 + ], + "score": 1.0, + "content": "The generator network is defined as:", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14, + "bbox_fs": [ + 106, + 628, + 254, + 640 + ] + }, + { + "type": "table", + "bbox": [ + 107, + 645, + 332, + 736 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 107, + 645, + 332, + 736 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 645, + 332, + 736 + ], + "spans": [ + { + "bbox": [ + 107, + 645, + 332, + 736 + ], + "score": 0.972, + "html": "
number outputsstride
Input: z ~ N(0,I256)Fully connectedReshape to image 4,4,512ConvolutionConvolutionConvolutionConvolution4*4*512256128641or32221
", + "type": "table", + "image_path": "6be1d7b077edb834c2bd6ee68d568dbb574dbb6746300a22e384363f25abe428.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 107, + 645, + 332, + 658.0 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 107, + 658.0, + 332, + 671.0 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 107, + 671.0, + 332, + 684.0 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 107, + 684.0, + 332, + 697.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 107, + 697.0, + 332, + 710.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 107, + 710.0, + 332, + 723.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 107, + 723.0, + 332, + 736.0 + ], + "spans": [], + "index": 21 + } + ] + } + ], + "index": 18 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 106, + 93, + 503, + 251 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 93, + 503, + 251 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 93, + 503, + 251 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 503, + 251 + ], + "score": 0.736, + "type": "image", + "image_path": "ddba8731c308e072b5532fd2f2dfbcf72fd2fc3c5d92dcb601c2c03608beee85.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 106, + 93, + 503, + 145.66666666666666 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 145.66666666666666, + 503, + 198.33333333333331 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 198.33333333333331, + 503, + 250.99999999999997 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 181, + 253, + 429, + 265 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 182, + 252, + 429, + 267 + ], + "spans": [ + { + "bbox": [ + 182, + 252, + 429, + 267 + ], + "score": 1.0, + "content": "Unrolled GAN with 5 G Steps per D without second gradient", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + } + ], + "index": 3.0 + }, + { + "type": "table", + "bbox": [ + 107, + 604, + 325, + 683 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 231, + 79, + 379, + 91 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 231, + 78, + 380, + 92 + ], + "spans": [ + { + "bbox": [ + 231, + 78, + 380, + 92 + ], + "score": 1.0, + "content": "Unrolled GAN with 5 G Steps per D", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 107, + 604, + 325, + 683 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 604, + 325, + 683 + ], + "spans": [ + { + "bbox": [ + 107, + 604, + 325, + 683 + ], + "score": 0.976, + "html": "
number outputsstride
Input: x~ Pdata or G Transposed Convolution Transposed Convolution Transposed Convolution64 128 256222
Flatten
Fully Connected1
", + "type": "table", + "image_path": "21b4b254a8e79c21939f50e66bbaf743c06f0c33a79d176b714a375db90c6857.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 107, + 604, + 325, + 619.8 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 107, + 619.8, + 325, + 635.5999999999999 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 107, + 635.5999999999999, + 325, + 651.3999999999999 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 107, + 651.3999999999999, + 325, + 667.1999999999998 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 107, + 667.1999999999998, + 325, + 682.9999999999998 + ], + "spans": [], + "index": 17 + } + ] + } + ], + "index": 7.5 + }, + { + "type": "image", + "bbox": [ + 106, + 268, + 503, + 425 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 268, + 503, + 425 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 268, + 503, + 425 + ], + "spans": [ + { + "bbox": [ + 106, + 268, + 503, + 425 + ], + "score": 0.775, + "type": "image", + "image_path": "c5b3fa4bd5bb790af426480883819ba983aa6618ec11502817b7cc7fa39e22d5.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 106, + 268, + 503, + 320.3333333333333 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 106, + 320.3333333333333, + 503, + 372.66666666666663 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 106, + 372.66666666666663, + 503, + 424.99999999999994 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 437, + 505, + 483 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 438, + 505, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 505, + 451 + ], + "score": 1.0, + "content": "Figure App.4: Backpropagating through the unrolling process aids convergence when the dis-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 448, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 506, + 462 + ], + "score": 1.0, + "content": "criminator does not fully converge between generator updates. When taking 5 generator steps per", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 459, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 505, + 473 + ], + "score": 1.0, + "content": "discriminator step unrolling greatly increases stability, requiring only 5 unrolling steps to converge.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 470, + 436, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 470, + 436, + 484 + ], + "score": 1.0, + "content": "Without the second gradient it requires 10 unrolling steps. Also see Figure App.3.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5 + } + ], + "index": 7.75 + }, + { + "type": "text", + "bbox": [ + 107, + 589, + 269, + 600 + ], + "lines": [ + { + "bbox": [ + 106, + 587, + 270, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 270, + 601 + ], + "score": 1.0, + "content": "The discriminator network is defined as:", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 686, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 247, + 700 + ], + "score": 1.0, + "content": "The generator network minimises", + "type": "text" + }, + { + "bbox": [ + 248, + 687, + 337, + 700 + ], + "score": 0.92, + "content": "\\mathcal { L } _ { G } = \\log ( D ( G ( z ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 686, + 475, + 700 + ], + "score": 1.0, + "content": "and the discriminator minimizes", + "type": "text" + }, + { + "bbox": [ + 476, + 688, + 505, + 699 + ], + "score": 0.86, + "content": "\\mathcal { L } _ { D } =", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 234, + 711 + ], + "score": 0.9, + "content": "\\log ( D ( x ) ) + \\log ( 1 - D ( G ( z ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 698, + 506, + 712 + ], + "score": 1.0, + "content": ". The networks are trained with Adam with a generator learning rate", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "of 1e-4, and a discriminator learning rate of 2e-4. The network is trained alternating updating the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 720, + 469, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 248, + 733 + ], + "score": 1.0, + "content": "generator and the discriminator for", + "type": "text" + }, + { + "bbox": [ + 249, + 721, + 270, + 731 + ], + "score": 0.63, + "content": "1 0 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 720, + 469, + 733 + ], + "score": 1.0, + "content": "steps. One step consists of just 1 network update.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5 + } + ], + "page_idx": 17, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "18", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 106, + 93, + 503, + 251 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 93, + 503, + 251 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 93, + 503, + 251 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 503, + 251 + ], + "score": 0.736, + "type": "image", + "image_path": "ddba8731c308e072b5532fd2f2dfbcf72fd2fc3c5d92dcb601c2c03608beee85.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 106, + 93, + 503, + 145.66666666666666 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 145.66666666666666, + 503, + 198.33333333333331 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 198.33333333333331, + 503, + 250.99999999999997 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 181, + 253, + 429, + 265 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 182, + 252, + 429, + 267 + ], + "spans": [ + { + "bbox": [ + 182, + 252, + 429, + 267 + ], + "score": 1.0, + "content": "Unrolled GAN with 5 G Steps per D without second gradient", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + } + ], + "index": 3.0 + }, + { + "type": "table", + "bbox": [ + 107, + 604, + 325, + 683 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 231, + 79, + 379, + 91 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 231, + 78, + 380, + 92 + ], + "spans": [ + { + "bbox": [ + 231, + 78, + 380, + 92 + ], + "score": 1.0, + "content": "Unrolled GAN with 5 G Steps per D", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 107, + 604, + 325, + 683 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 604, + 325, + 683 + ], + "spans": [ + { + "bbox": [ + 107, + 604, + 325, + 683 + ], + "score": 0.976, + "html": "
number outputsstride
Input: x~ Pdata or G Transposed Convolution Transposed Convolution Transposed Convolution64 128 256222
Flatten
Fully Connected1
", + "type": "table", + "image_path": "21b4b254a8e79c21939f50e66bbaf743c06f0c33a79d176b714a375db90c6857.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 107, + 604, + 325, + 619.8 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 107, + 619.8, + 325, + 635.5999999999999 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 107, + 635.5999999999999, + 325, + 651.3999999999999 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 107, + 651.3999999999999, + 325, + 667.1999999999998 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 107, + 667.1999999999998, + 325, + 682.9999999999998 + ], + "spans": [], + "index": 17 + } + ] + } + ], + "index": 7.5 + }, + { + "type": "image", + "bbox": [ + 106, + 268, + 503, + 425 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 268, + 503, + 425 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 268, + 503, + 425 + ], + "spans": [ + { + "bbox": [ + 106, + 268, + 503, + 425 + ], + "score": 0.775, + "type": "image", + "image_path": "c5b3fa4bd5bb790af426480883819ba983aa6618ec11502817b7cc7fa39e22d5.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 106, + 268, + 503, + 320.3333333333333 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 106, + 320.3333333333333, + 503, + 372.66666666666663 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 106, + 372.66666666666663, + 503, + 424.99999999999994 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 437, + 505, + 483 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 438, + 505, + 451 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 505, + 451 + ], + "score": 1.0, + "content": "Figure App.4: Backpropagating through the unrolling process aids convergence when the dis-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 448, + 506, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 506, + 462 + ], + "score": 1.0, + "content": "criminator does not fully converge between generator updates. When taking 5 generator steps per", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 459, + 505, + 473 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 505, + 473 + ], + "score": 1.0, + "content": "discriminator step unrolling greatly increases stability, requiring only 5 unrolling steps to converge.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 470, + 436, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 470, + 436, + 484 + ], + "score": 1.0, + "content": "Without the second gradient it requires 10 unrolling steps. Also see Figure App.3.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5 + } + ], + "index": 7.75 + }, + { + "type": "text", + "bbox": [ + 107, + 589, + 269, + 600 + ], + "lines": [ + { + "bbox": [ + 106, + 587, + 270, + 601 + ], + "spans": [ + { + "bbox": [ + 106, + 587, + 270, + 601 + ], + "score": 1.0, + "content": "The discriminator network is defined as:", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12, + "bbox_fs": [ + 106, + 587, + 270, + 601 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 686, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 686, + 247, + 700 + ], + "score": 1.0, + "content": "The generator network minimises", + "type": "text" + }, + { + "bbox": [ + 248, + 687, + 337, + 700 + ], + "score": 0.92, + "content": "\\mathcal { L } _ { G } = \\log ( D ( G ( z ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 686, + 475, + 700 + ], + "score": 1.0, + "content": "and the discriminator minimizes", + "type": "text" + }, + { + "bbox": [ + 476, + 688, + 505, + 699 + ], + "score": 0.86, + "content": "\\mathcal { L } _ { D } =", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 234, + 711 + ], + "score": 0.9, + "content": "\\log ( D ( x ) ) + \\log ( 1 - D ( G ( z ) ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 698, + 506, + 712 + ], + "score": 1.0, + "content": ". The networks are trained with Adam with a generator learning rate", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "of 1e-4, and a discriminator learning rate of 2e-4. The network is trained alternating updating the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 720, + 469, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 248, + 733 + ], + "score": 1.0, + "content": "generator and the discriminator for", + "type": "text" + }, + { + "bbox": [ + 249, + 721, + 270, + 731 + ], + "score": 0.63, + "content": "1 0 0 \\mathrm { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 720, + 469, + 733 + ], + "score": 1.0, + "content": "steps. One step consists of just 1 network update.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 686, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 233, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 79, + 234, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 234, + 96 + ], + "score": 1.0, + "content": "E 1000 CLASS MNIST", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table", + "bbox": [ + 106, + 108, + 326, + 198 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 108, + 326, + 198 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 108, + 326, + 198 + ], + "spans": [ + { + "bbox": [ + 106, + 108, + 326, + 198 + ], + "score": 0.971, + "html": "
number outputsstride
Input: z ~ N(0,I256)Fully connectedReshape to image 4,4,64ConvolutionConvolutionConvolutionConvolution4 *4*643216832221
", + "type": "table", + "image_path": "6d161339c120af4e827f0777703e338fce4dd94b3a55a7a2c9761991b51c1f31.jpg" + } + ] + } + ], + "index": 3.5, + "virtual_lines": [ + { + "bbox": [ + 106, + 108, + 326, + 123.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 123.0, + 326, + 138.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 138.0, + 326, + 153.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 106, + 153.0, + 326, + 168.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 106, + 168.0, + 326, + 183.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 106, + 183.0, + 326, + 198.0 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 105, + 202, + 506, + 225 + ], + "lines": [ + { + "bbox": [ + 105, + 200, + 506, + 216 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 320, + 216 + ], + "score": 1.0, + "content": "The discriminator network is parametrized by a size", + "type": "text" + }, + { + "bbox": [ + 321, + 203, + 330, + 213 + ], + "score": 0.3, + "content": "\\mathrm { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 200, + 506, + 216 + ], + "score": 1.0, + "content": "and is defined as follows. In our tests, we", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 214, + 197, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 197, + 225 + ], + "score": 1.0, + "content": "used X of 1/4 and 1/2.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "table", + "bbox": [ + 107, + 228, + 325, + 308 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 107, + 228, + 325, + 308 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 107, + 228, + 325, + 308 + ], + "spans": [ + { + "bbox": [ + 107, + 228, + 325, + 308 + ], + "score": 0.978, + "html": "
number outputsstride
Input: x ~ Pdata or G Transposed Convolution Transposed Convolution8*X 16*X222
Transposed Convolution Flatten32*X
Fully Connected1
", + "type": "table", + "image_path": "d861a766290ac6cab3787aa8042ed8fa20fa08a94a861ed7ebddde0d0efa7d41.jpg" + } + ] + } + ], + "index": 11.5, + "virtual_lines": [ + { + "bbox": [ + 107, + 228, + 325, + 241.33333333333334 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 107, + 241.33333333333334, + 325, + 254.66666666666669 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 107, + 254.66666666666669, + 325, + 268.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 107, + 268.0, + 325, + 281.3333333333333 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 107, + 281.3333333333333, + 325, + 294.66666666666663 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 107, + 294.66666666666663, + 325, + 307.99999999999994 + ], + "spans": [], + "index": 14 + } + ] + } + ], + "index": 11.5 + }, + { + "type": "title", + "bbox": [ + 107, + 326, + 269, + 340 + ], + "lines": [ + { + "bbox": [ + 105, + 326, + 271, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 326, + 271, + 342 + ], + "score": 1.0, + "content": "F COLORED MNIST DATASET", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "title", + "bbox": [ + 107, + 356, + 171, + 367 + ], + "lines": [ + { + "bbox": [ + 105, + 355, + 172, + 368 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 172, + 368 + ], + "score": 1.0, + "content": "F.1 DATASET", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 379, + 505, + 456 + ], + "lines": [ + { + "bbox": [ + 105, + 378, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 105, + 378, + 325, + 393 + ], + "score": 1.0, + "content": "To generate this dataset we first took the mnist digit,", + "type": "text" + }, + { + "bbox": [ + 325, + 380, + 331, + 389 + ], + "score": 0.67, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 378, + 505, + 393 + ], + "score": 1.0, + "content": ", scaled between 0 and 1. For each image", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 390, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 390, + 185, + 403 + ], + "score": 1.0, + "content": "we sample a color,", + "type": "text" + }, + { + "bbox": [ + 186, + 390, + 195, + 401 + ], + "score": 0.78, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 390, + 326, + 403 + ], + "score": 1.0, + "content": ", normally distributed with mean", + "type": "text" + }, + { + "bbox": [ + 326, + 391, + 338, + 401 + ], + "score": 0.58, + "content": "\\scriptstyle = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 390, + 368, + 403 + ], + "score": 1.0, + "content": "and std", + "type": "text" + }, + { + "bbox": [ + 369, + 390, + 388, + 401 + ], + "score": 0.28, + "content": "\\scriptstyle \\mathtt { = 0 . 5 }", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 390, + 505, + 403 + ], + "score": 1.0, + "content": ". To generate a colored digit", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 401, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 197, + 414 + ], + "score": 1.0, + "content": "between (-1, 1) we do", + "type": "text" + }, + { + "bbox": [ + 197, + 401, + 263, + 413 + ], + "score": 0.93, + "content": "I * C + \\mathsf { \\bar { ( } } I - 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 401, + 505, + 414 + ], + "score": 1.0, + "content": ". Finally, we add a small amount of pixel independent noise", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 412, + 504, + 424 + ], + "spans": [ + { + "bbox": [ + 106, + 412, + 280, + 424 + ], + "score": 1.0, + "content": "sampled from a normal distribution with std", + "type": "text" + }, + { + "bbox": [ + 281, + 412, + 299, + 423 + ], + "score": 0.49, + "content": "= 0 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 412, + 504, + 424 + ], + "score": 1.0, + "content": ", and the resulting values are cliped between (-1, 1).", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 422, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 105, + 422, + 505, + 435 + ], + "score": 1.0, + "content": "When visualized, this generates images and samples that can be seen in figure App.5. Once again", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 433, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 505, + 447 + ], + "score": 1.0, + "content": "it is very hard to visually see differences in sample diversity when comparing the 128 and the 512", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 446, + 162, + 456 + ], + "spans": [ + { + "bbox": [ + 106, + 446, + 162, + 456 + ], + "score": 1.0, + "content": "sized models.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20 + }, + { + "type": "image", + "bbox": [ + 124, + 470, + 487, + 591 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 124, + 470, + 487, + 591 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 124, + 470, + 487, + 591 + ], + "spans": [ + { + "bbox": [ + 124, + 470, + 487, + 591 + ], + "score": 0.975, + "type": "image", + "image_path": "44d318f018c78dbdaec0c5211309ab172baeb975108ebe77f750982a4193cef6.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 124, + 470, + 487, + 510.3333333333333 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 124, + 510.3333333333333, + 487, + 550.6666666666666 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 124, + 550.6666666666666, + 487, + 591.0 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 602, + 505, + 636 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 602, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 505, + 614 + ], + "score": 1.0, + "content": "Figure App.5: Right: samples from the data distribution. Middle: Samples from 1/4 size model with", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 612, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 104, + 612, + 505, + 627 + ], + "score": 1.0, + "content": "0 look ahead steps (worst diversity). Left: Samples from 1/1 size model with 10 look ahead steps", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 624, + 174, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 174, + 637 + ], + "score": 1.0, + "content": "(most diversity).", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28 + } + ], + "index": 26.5 + }, + { + "type": "title", + "bbox": [ + 107, + 669, + 169, + 681 + ], + "lines": [ + { + "bbox": [ + 105, + 668, + 171, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 171, + 683 + ], + "score": 1.0, + "content": "F.2 MODELS", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 108, + 692, + 505, + 715 + ], + "lines": [ + { + "bbox": [ + 106, + 692, + 506, + 705 + ], + "spans": [ + { + "bbox": [ + 106, + 692, + 365, + 705 + ], + "score": 1.0, + "content": "The models used in this section are parametrized by a variable", + "type": "text" + }, + { + "bbox": [ + 365, + 693, + 375, + 703 + ], + "score": 0.34, + "content": "\\mathrm { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 692, + 506, + 705 + ], + "score": 1.0, + "content": "to control capacity. A value of", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 702, + 496, + 716 + ], + "spans": [ + { + "bbox": [ + 106, + 704, + 126, + 714 + ], + "score": 0.76, + "content": "{ \\bf X } { = } 1", + "type": "inline_equation" + }, + { + "bbox": [ + 126, + 702, + 496, + 716 + ], + "score": 1.0, + "content": "is same architecture used in the cifar10 experiments. We used 1/4, 1/2 and 1 as these values.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 108, + 720, + 253, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 720, + 254, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 254, + 733 + ], + "score": 1.0, + "content": "The generator network is defined as:", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + } + ], + "page_idx": 18, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "19", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 81, + 233, + 94 + ], + "lines": [ + { + "bbox": [ + 105, + 79, + 234, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 79, + 234, + 96 + ], + "score": 1.0, + "content": "E 1000 CLASS MNIST", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table", + "bbox": [ + 106, + 108, + 326, + 198 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 108, + 326, + 198 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 108, + 326, + 198 + ], + "spans": [ + { + "bbox": [ + 106, + 108, + 326, + 198 + ], + "score": 0.971, + "html": "
number outputsstride
Input: z ~ N(0,I256)Fully connectedReshape to image 4,4,64ConvolutionConvolutionConvolutionConvolution4 *4*643216832221
", + "type": "table", + "image_path": "6d161339c120af4e827f0777703e338fce4dd94b3a55a7a2c9761991b51c1f31.jpg" + } + ] + } + ], + "index": 3.5, + "virtual_lines": [ + { + "bbox": [ + 106, + 108, + 326, + 123.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 123.0, + 326, + 138.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 138.0, + 326, + 153.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 106, + 153.0, + 326, + 168.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 106, + 168.0, + 326, + 183.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 106, + 183.0, + 326, + 198.0 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 105, + 202, + 506, + 225 + ], + "lines": [ + { + "bbox": [ + 105, + 200, + 506, + 216 + ], + "spans": [ + { + "bbox": [ + 105, + 200, + 320, + 216 + ], + "score": 1.0, + "content": "The discriminator network is parametrized by a size", + "type": "text" + }, + { + "bbox": [ + 321, + 203, + 330, + 213 + ], + "score": 0.3, + "content": "\\mathrm { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 200, + 506, + 216 + ], + "score": 1.0, + "content": "and is defined as follows. In our tests, we", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 214, + 197, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 197, + 225 + ], + "score": 1.0, + "content": "used X of 1/4 and 1/2.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 200, + 506, + 225 + ] + }, + { + "type": "table", + "bbox": [ + 107, + 228, + 325, + 308 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 107, + 228, + 325, + 308 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 107, + 228, + 325, + 308 + ], + "spans": [ + { + "bbox": [ + 107, + 228, + 325, + 308 + ], + "score": 0.978, + "html": "
number outputsstride
Input: x ~ Pdata or G Transposed Convolution Transposed Convolution8*X 16*X222
Transposed Convolution Flatten32*X
Fully Connected1
", + "type": "table", + "image_path": "d861a766290ac6cab3787aa8042ed8fa20fa08a94a861ed7ebddde0d0efa7d41.jpg" + } + ] + } + ], + "index": 11.5, + "virtual_lines": [ + { + "bbox": [ + 107, + 228, + 325, + 241.33333333333334 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 107, + 241.33333333333334, + 325, + 254.66666666666669 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 107, + 254.66666666666669, + 325, + 268.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 107, + 268.0, + 325, + 281.3333333333333 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 107, + 281.3333333333333, + 325, + 294.66666666666663 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 107, + 294.66666666666663, + 325, + 307.99999999999994 + ], + "spans": [], + "index": 14 + } + ] + } + ], + "index": 11.5 + }, + { + "type": "title", + "bbox": [ + 107, + 326, + 269, + 340 + ], + "lines": [ + { + "bbox": [ + 105, + 326, + 271, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 326, + 271, + 342 + ], + "score": 1.0, + "content": "F COLORED MNIST DATASET", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "title", + "bbox": [ + 107, + 356, + 171, + 367 + ], + "lines": [ + { + "bbox": [ + 105, + 355, + 172, + 368 + ], + "spans": [ + { + "bbox": [ + 105, + 355, + 172, + 368 + ], + "score": 1.0, + "content": "F.1 DATASET", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 106, + 379, + 505, + 456 + ], + "lines": [ + { + "bbox": [ + 105, + 378, + 505, + 393 + ], + "spans": [ + { + "bbox": [ + 105, + 378, + 325, + 393 + ], + "score": 1.0, + "content": "To generate this dataset we first took the mnist digit,", + "type": "text" + }, + { + "bbox": [ + 325, + 380, + 331, + 389 + ], + "score": 0.67, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 378, + 505, + 393 + ], + "score": 1.0, + "content": ", scaled between 0 and 1. For each image", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 390, + 505, + 403 + ], + "spans": [ + { + "bbox": [ + 106, + 390, + 185, + 403 + ], + "score": 1.0, + "content": "we sample a color,", + "type": "text" + }, + { + "bbox": [ + 186, + 390, + 195, + 401 + ], + "score": 0.78, + "content": "C", + "type": "inline_equation" + }, + { + "bbox": [ + 195, + 390, + 326, + 403 + ], + "score": 1.0, + "content": ", normally distributed with mean", + "type": "text" + }, + { + "bbox": [ + 326, + 391, + 338, + 401 + ], + "score": 0.58, + "content": "\\scriptstyle = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 390, + 368, + 403 + ], + "score": 1.0, + "content": "and std", + "type": "text" + }, + { + "bbox": [ + 369, + 390, + 388, + 401 + ], + "score": 0.28, + "content": "\\scriptstyle \\mathtt { = 0 . 5 }", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 390, + 505, + 403 + ], + "score": 1.0, + "content": ". To generate a colored digit", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 401, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 197, + 414 + ], + "score": 1.0, + "content": "between (-1, 1) we do", + "type": "text" + }, + { + "bbox": [ + 197, + 401, + 263, + 413 + ], + "score": 0.93, + "content": "I * C + \\mathsf { \\bar { ( } } I - 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 401, + 505, + 414 + ], + "score": 1.0, + "content": ". Finally, we add a small amount of pixel independent noise", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 412, + 504, + 424 + ], + "spans": [ + { + "bbox": [ + 106, + 412, + 280, + 424 + ], + "score": 1.0, + "content": "sampled from a normal distribution with std", + "type": "text" + }, + { + "bbox": [ + 281, + 412, + 299, + 423 + ], + "score": 0.49, + "content": "= 0 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 412, + 504, + 424 + ], + "score": 1.0, + "content": ", and the resulting values are cliped between (-1, 1).", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 422, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 105, + 422, + 505, + 435 + ], + "score": 1.0, + "content": "When visualized, this generates images and samples that can be seen in figure App.5. Once again", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 433, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 505, + 447 + ], + "score": 1.0, + "content": "it is very hard to visually see differences in sample diversity when comparing the 128 and the 512", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 446, + 162, + 456 + ], + "spans": [ + { + "bbox": [ + 106, + 446, + 162, + 456 + ], + "score": 1.0, + "content": "sized models.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 378, + 505, + 456 + ] + }, + { + "type": "image", + "bbox": [ + 124, + 470, + 487, + 591 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 124, + 470, + 487, + 591 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 124, + 470, + 487, + 591 + ], + "spans": [ + { + "bbox": [ + 124, + 470, + 487, + 591 + ], + "score": 0.975, + "type": "image", + "image_path": "44d318f018c78dbdaec0c5211309ab172baeb975108ebe77f750982a4193cef6.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 124, + 470, + 487, + 510.3333333333333 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 124, + 510.3333333333333, + 487, + 550.6666666666666 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 124, + 550.6666666666666, + 487, + 591.0 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 602, + 505, + 636 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 602, + 505, + 614 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 505, + 614 + ], + "score": 1.0, + "content": "Figure App.5: Right: samples from the data distribution. Middle: Samples from 1/4 size model with", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 104, + 612, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 104, + 612, + 505, + 627 + ], + "score": 1.0, + "content": "0 look ahead steps (worst diversity). Left: Samples from 1/1 size model with 10 look ahead steps", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 624, + 174, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 624, + 174, + 637 + ], + "score": 1.0, + "content": "(most diversity).", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28 + } + ], + "index": 26.5 + }, + { + "type": "title", + "bbox": [ + 107, + 669, + 169, + 681 + ], + "lines": [ + { + "bbox": [ + 105, + 668, + 171, + 683 + ], + "spans": [ + { + "bbox": [ + 105, + 668, + 171, + 683 + ], + "score": 1.0, + "content": "F.2 MODELS", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 108, + 692, + 505, + 715 + ], + "lines": [ + { + "bbox": [ + 106, + 692, + 506, + 705 + ], + "spans": [ + { + "bbox": [ + 106, + 692, + 365, + 705 + ], + "score": 1.0, + "content": "The models used in this section are parametrized by a variable", + "type": "text" + }, + { + "bbox": [ + 365, + 693, + 375, + 703 + ], + "score": 0.34, + "content": "\\mathrm { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 692, + 506, + 705 + ], + "score": 1.0, + "content": "to control capacity. A value of", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 702, + 496, + 716 + ], + "spans": [ + { + "bbox": [ + 106, + 704, + 126, + 714 + ], + "score": 0.76, + "content": "{ \\bf X } { = } 1", + "type": "inline_equation" + }, + { + "bbox": [ + 126, + 702, + 496, + 716 + ], + "score": 1.0, + "content": "is same architecture used in the cifar10 experiments. We used 1/4, 1/2 and 1 as these values.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 106, + 692, + 506, + 716 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 720, + 253, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 720, + 254, + 733 + ], + "spans": [ + { + "bbox": [ + 106, + 720, + 254, + 733 + ], + "score": 1.0, + "content": "The generator network is defined as:", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33, + "bbox_fs": [ + 106, + 720, + 254, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 107, + 80, + 344, + 171 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 107, + 80, + 344, + 171 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 80, + 344, + 171 + ], + "spans": [ + { + "bbox": [ + 107, + 80, + 344, + 171 + ], + "score": 0.977, + "html": "
number outputsstride
Input: z~ N(0,I256)Fully connectedReshape to image 4,4,512*XConvolutionConvolutionConvolutionConvolution4*4*512*X256*X128*X64*X32221
", + "type": "table", + "image_path": "de1a5ae153f5aca675e1d81f0274ce2bb52042972834a41845eb58fa8a6854f0.jpg" + } + ] + } + ], + "index": 2.5, + "virtual_lines": [ + { + "bbox": [ + 107, + 80, + 344, + 95.16666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 107, + 95.16666666666667, + 344, + 110.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 107, + 110.33333333333334, + 344, + 125.50000000000001 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 107, + 125.50000000000001, + 344, + 140.66666666666669 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 107, + 140.66666666666669, + 344, + 155.83333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 107, + 155.83333333333334, + 344, + 171.0 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 108, + 175, + 269, + 186 + ], + "lines": [ + { + "bbox": [ + 106, + 173, + 270, + 187 + ], + "spans": [ + { + "bbox": [ + 106, + 173, + 270, + 187 + ], + "score": 1.0, + "content": "The discriminator network is defined as:", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "table", + "bbox": [ + 107, + 190, + 326, + 270 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 107, + 190, + 326, + 270 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 107, + 190, + 326, + 270 + ], + "spans": [ + { + "bbox": [ + 107, + 190, + 326, + 270 + ], + "score": 0.964, + "html": "
number outputsstride
Input: x ~ Pdata or G Transposed Convolution Transposed Convolution Transposed Convolution Flatten Fully Connected64*X 128*X 256*X222
", + "type": "table", + "image_path": "8185cc38813b0aac909bf7ec1d0d173e742b2ff60e4a2f788d6a7b905fcd642a.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 107, + 190, + 326, + 206.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 107, + 206.0, + 326, + 222.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 107, + 222.0, + 326, + 238.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 107, + 238.0, + 326, + 254.0 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 107, + 254.0, + 326, + 270.0 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 9 + }, + { + "type": "title", + "bbox": [ + 107, + 282, + 335, + 295 + ], + "lines": [ + { + "bbox": [ + 105, + 280, + 336, + 297 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 336, + 297 + ], + "score": 1.0, + "content": "G OPTIMIZATION BASED VISUALIZATIONS", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 307, + 505, + 351 + ], + "lines": [ + { + "bbox": [ + 105, + 307, + 506, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 506, + 320 + ], + "score": 1.0, + "content": "More examples of model based optimization. We performed 5 runs with different seeds of each of", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 318, + 505, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 318, + 505, + 330 + ], + "score": 1.0, + "content": "of the unrolling steps configuration. Bellow are comparisons for each run index. Ideally this would", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 329, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 505, + 342 + ], + "score": 1.0, + "content": "be a many to many comparison, but for space efficiency we grouped the runs by the index in which", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 339, + 167, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 167, + 353 + ], + "score": 1.0, + "content": "they were run.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14.5 + } + ], + "page_idx": 19, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 107, + 80, + 344, + 171 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 107, + 80, + 344, + 171 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 107, + 80, + 344, + 171 + ], + "spans": [ + { + "bbox": [ + 107, + 80, + 344, + 171 + ], + "score": 0.977, + "html": "
number outputsstride
Input: z~ N(0,I256)Fully connectedReshape to image 4,4,512*XConvolutionConvolutionConvolutionConvolution4*4*512*X256*X128*X64*X32221
", + "type": "table", + "image_path": "de1a5ae153f5aca675e1d81f0274ce2bb52042972834a41845eb58fa8a6854f0.jpg" + } + ] + } + ], + "index": 2.5, + "virtual_lines": [ + { + "bbox": [ + 107, + 80, + 344, + 95.16666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 107, + 95.16666666666667, + 344, + 110.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 107, + 110.33333333333334, + 344, + 125.50000000000001 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 107, + 125.50000000000001, + 344, + 140.66666666666669 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 107, + 140.66666666666669, + 344, + 155.83333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 107, + 155.83333333333334, + 344, + 171.0 + ], + "spans": [], + "index": 5 + } + ] + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 108, + 175, + 269, + 186 + ], + "lines": [ + { + "bbox": [ + 106, + 173, + 270, + 187 + ], + "spans": [ + { + "bbox": [ + 106, + 173, + 270, + 187 + ], + "score": 1.0, + "content": "The discriminator network is defined as:", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6, + "bbox_fs": [ + 106, + 173, + 270, + 187 + ] + }, + { + "type": "table", + "bbox": [ + 107, + 190, + 326, + 270 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 107, + 190, + 326, + 270 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 107, + 190, + 326, + 270 + ], + "spans": [ + { + "bbox": [ + 107, + 190, + 326, + 270 + ], + "score": 0.964, + "html": "
number outputsstride
Input: x ~ Pdata or G Transposed Convolution Transposed Convolution Transposed Convolution Flatten Fully Connected64*X 128*X 256*X222
", + "type": "table", + "image_path": "8185cc38813b0aac909bf7ec1d0d173e742b2ff60e4a2f788d6a7b905fcd642a.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 107, + 190, + 326, + 206.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 107, + 206.0, + 326, + 222.0 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 107, + 222.0, + 326, + 238.0 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 107, + 238.0, + 326, + 254.0 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 107, + 254.0, + 326, + 270.0 + ], + "spans": [], + "index": 11 + } + ] + } + ], + "index": 9 + }, + { + "type": "title", + "bbox": [ + 107, + 282, + 335, + 295 + ], + "lines": [ + { + "bbox": [ + 105, + 280, + 336, + 297 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 336, + 297 + ], + "score": 1.0, + "content": "G OPTIMIZATION BASED VISUALIZATIONS", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 106, + 307, + 505, + 351 + ], + "lines": [ + { + "bbox": [ + 105, + 307, + 506, + 320 + ], + "spans": [ + { + "bbox": [ + 105, + 307, + 506, + 320 + ], + "score": 1.0, + "content": "More examples of model based optimization. We performed 5 runs with different seeds of each of", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 318, + 505, + 330 + ], + "spans": [ + { + "bbox": [ + 106, + 318, + 505, + 330 + ], + "score": 1.0, + "content": "of the unrolling steps configuration. Bellow are comparisons for each run index. Ideally this would", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 329, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 505, + 342 + ], + "score": 1.0, + "content": "be a many to many comparison, but for space efficiency we grouped the runs by the index in which", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 339, + 167, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 167, + 353 + ], + "score": 1.0, + "content": "they were run.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14.5, + "bbox_fs": [ + 105, + 307, + 506, + 353 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 116, + 91, + 503, + 700 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 91, + 503, + 700 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 91, + 503, + 700 + ], + "spans": [ + { + "bbox": [ + 116, + 91, + 503, + 700 + ], + "score": 0.953, + "type": "image", + "image_path": "8d38e597124cc936be0c8530d6fb7e8b1ecc41f5e50351124daf12a2d78e11fa.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 91, + 503, + 294.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 294.0, + 503, + 497.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 497.0, + 503, + 700.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 181, + 708, + 430, + 720 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 180, + 707, + 430, + 722 + ], + "spans": [ + { + "bbox": [ + 180, + 707, + 430, + 722 + ], + "score": 1.0, + "content": "Figure App.6: Samples from 1/5 with different random seeds.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ], + "page_idx": 20, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 116, + 91, + 503, + 700 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 91, + 503, + 700 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 91, + 503, + 700 + ], + "spans": [ + { + "bbox": [ + 116, + 91, + 503, + 700 + ], + "score": 0.953, + "type": "image", + "image_path": "8d38e597124cc936be0c8530d6fb7e8b1ecc41f5e50351124daf12a2d78e11fa.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 91, + 503, + 294.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 294.0, + 503, + 497.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 497.0, + 503, + 700.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 181, + 708, + 430, + 720 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 180, + 707, + 430, + 722 + ], + "spans": [ + { + "bbox": [ + 180, + 707, + 430, + 722 + ], + "score": 1.0, + "content": "Figure App.6: Samples from 1/5 with different random seeds.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 116, + 90, + 502, + 698 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 90, + 502, + 698 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 90, + 502, + 698 + ], + "spans": [ + { + "bbox": [ + 116, + 90, + 502, + 698 + ], + "score": 0.872, + "type": "image", + "image_path": "be94e2bb107541e788983b4ad2bdc31bebdff50fbbb8529c915238f8cca809e8.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 90, + 502, + 292.66666666666663 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 292.66666666666663, + 502, + 495.33333333333326 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 495.33333333333326, + 502, + 697.9999999999999 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 181, + 708, + 430, + 720 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 180, + 707, + 430, + 722 + ], + "spans": [ + { + "bbox": [ + 180, + 707, + 430, + 722 + ], + "score": 1.0, + "content": "Figure App.7: Samples from 2/5 with different random seeds.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ], + "page_idx": 21, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 116, + 90, + 502, + 698 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 90, + 502, + 698 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 90, + 502, + 698 + ], + "spans": [ + { + "bbox": [ + 116, + 90, + 502, + 698 + ], + "score": 0.872, + "type": "image", + "image_path": "be94e2bb107541e788983b4ad2bdc31bebdff50fbbb8529c915238f8cca809e8.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 90, + 502, + 292.66666666666663 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 292.66666666666663, + 502, + 495.33333333333326 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 495.33333333333326, + 502, + 697.9999999999999 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 181, + 708, + 430, + 720 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 180, + 707, + 430, + 722 + ], + "spans": [ + { + "bbox": [ + 180, + 707, + 430, + 722 + ], + "score": 1.0, + "content": "Figure App.7: Samples from 2/5 with different random seeds.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 116, + 89, + 502, + 697 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 89, + 502, + 697 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 89, + 502, + 697 + ], + "spans": [ + { + "bbox": [ + 116, + 89, + 502, + 697 + ], + "score": 0.945, + "type": "image", + "image_path": "e17957e3a6c6fffe5b489567eb4830a1720a2ae85eadaf27b1b6017a82527e8b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 89, + 502, + 291.66666666666663 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 291.66666666666663, + 502, + 494.33333333333326 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 494.33333333333326, + 502, + 696.9999999999999 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 181, + 708, + 430, + 720 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 180, + 707, + 430, + 722 + ], + "spans": [ + { + "bbox": [ + 180, + 707, + 430, + 722 + ], + "score": 1.0, + "content": "Figure App.8: Samples from 3/5 with different random seeds.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ], + "page_idx": 22, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 764 + ], + "score": 1.0, + "content": "23", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 116, + 89, + 502, + 697 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 89, + 502, + 697 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 89, + 502, + 697 + ], + "spans": [ + { + "bbox": [ + 116, + 89, + 502, + 697 + ], + "score": 0.945, + "type": "image", + "image_path": "e17957e3a6c6fffe5b489567eb4830a1720a2ae85eadaf27b1b6017a82527e8b.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 89, + 502, + 291.66666666666663 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 291.66666666666663, + 502, + 494.33333333333326 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 494.33333333333326, + 502, + 696.9999999999999 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 181, + 708, + 430, + 720 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 180, + 707, + 430, + 722 + ], + "spans": [ + { + "bbox": [ + 180, + 707, + 430, + 722 + ], + "score": 1.0, + "content": "Figure App.8: Samples from 3/5 with different random seeds.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 116, + 93, + 502, + 697 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 93, + 502, + 697 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 93, + 502, + 697 + ], + "spans": [ + { + "bbox": [ + 116, + 93, + 502, + 697 + ], + "score": 0.953, + "type": "image", + "image_path": "fbca68637cf3798a1a739daff1548046fba7251170be090a9d1d669d0aa5603e.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 93, + 502, + 294.33333333333337 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 294.33333333333337, + 502, + 495.66666666666674 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 495.66666666666674, + 502, + 697.0000000000001 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 181, + 708, + 430, + 720 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 180, + 707, + 430, + 721 + ], + "spans": [ + { + "bbox": [ + 180, + 707, + 430, + 721 + ], + "score": 1.0, + "content": "Figure App.9: Samples from 4/5 with different random seeds.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ], + "page_idx": 23, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 116, + 93, + 502, + 697 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 93, + 502, + 697 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 93, + 502, + 697 + ], + "spans": [ + { + "bbox": [ + 116, + 93, + 502, + 697 + ], + "score": 0.953, + "type": "image", + "image_path": "fbca68637cf3798a1a739daff1548046fba7251170be090a9d1d669d0aa5603e.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 93, + 502, + 294.33333333333337 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 294.33333333333337, + 502, + 495.66666666666674 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 495.66666666666674, + 502, + 697.0000000000001 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 181, + 708, + 430, + 720 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 180, + 707, + 430, + 721 + ], + "spans": [ + { + "bbox": [ + 180, + 707, + 430, + 721 + ], + "score": 1.0, + "content": "Figure App.9: Samples from 4/5 with different random seeds.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 116, + 89, + 502, + 698 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 89, + 502, + 698 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 89, + 502, + 698 + ], + "spans": [ + { + "bbox": [ + 116, + 89, + 502, + 698 + ], + "score": 0.948, + "type": "image", + "image_path": "c67273dbe4f3a1a6aee81a0d0854850226af640cdd34215ec9e0663f82c11a7d.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 89, + 502, + 292.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 292.0, + 502, + 495.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 495.0, + 502, + 698.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 178, + 708, + 432, + 720 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 177, + 707, + 433, + 722 + ], + "spans": [ + { + "bbox": [ + 177, + 707, + 433, + 722 + ], + "score": 1.0, + "content": "Figure App.10: Samples from 5/5 with different random seeds.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ], + "page_idx": 24, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2017", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 298, + 749, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 116, + 89, + 502, + 698 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 116, + 89, + 502, + 698 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 116, + 89, + 502, + 698 + ], + "spans": [ + { + "bbox": [ + 116, + 89, + 502, + 698 + ], + "score": 0.948, + "type": "image", + "image_path": "c67273dbe4f3a1a6aee81a0d0854850226af640cdd34215ec9e0663f82c11a7d.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 116, + 89, + 502, + 292.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 116, + 292.0, + 502, + 495.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 116, + 495.0, + 502, + 698.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 178, + 708, + 432, + 720 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 177, + 707, + 433, + 722 + ], + "spans": [ + { + "bbox": [ + 177, + 707, + 433, + 722 + ], + "score": 1.0, + "content": "Figure App.10: Samples from 5/5 with different random seeds.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/train/BydrOIcle/BydrOIcle_model.json b/parse/train/BydrOIcle/BydrOIcle_model.json new file mode 100644 index 0000000000000000000000000000000000000000..b97be290c771136d59322705b711d0ec9dedac1e --- /dev/null +++ b/parse/train/BydrOIcle/BydrOIcle_model.json @@ -0,0 +1,32838 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 398, + 702, + 1302, + 702, + 1302, + 948, + 398, + 948 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1078, + 1404, + 1078, + 1404, + 1352, + 298, + 1352 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1448, + 1403, + 1448, + 1403, + 1724, + 298, + 1724 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 299, + 1739, + 1404, + 1739, + 1404, + 1954, + 299, + 1954 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 315, + 321, + 586, + 321, + 586, + 414, + 315, + 414 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 313, + 460, + 635, + 460, + 635, + 553, + 313, + 553 + ], + "score": 0.951 + }, + { + "category_id": 2, + "poly": [ + 328, + 1974, + 1215, + 1974, + 1215, + 2034, + 328, + 2034 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 680, + 319, + 1035, + 319, + 1035, + 414, + 680, + 414 + ], + "score": 0.94 + }, + { + "category_id": 1, + "poly": [ + 1130, + 320, + 1385, + 320, + 1385, + 415, + 1130, + 415 + ], + "score": 0.93 + }, + { + "category_id": 0, + "poly": [ + 296, + 220, + 1374, + 220, + 1374, + 269, + 296, + 269 + ], + "score": 0.928 + }, + { + "category_id": 0, + "poly": [ + 303, + 1008, + 573, + 1008, + 573, + 1043, + 303, + 1043 + ], + "score": 0.894 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 815, + 75, + 815, + 105, + 299, + 105 + ], + "score": 0.868 + }, + { + "category_id": 0, + "poly": [ + 773, + 634, + 927, + 634, + 927, + 667, + 773, + 667 + ], + "score": 0.853 + }, + { + "category_id": 0, + "poly": [ + 303, + 1391, + 847, + 1391, + 847, + 1421, + 303, + 1421 + ], + "score": 0.788 + }, + { + "category_id": 2, + "poly": [ + 842, + 2089, + 857, + 2089, + 857, + 2112, + 842, + 2112 + ], + "score": 0.686 + }, + { + "category_id": 1, + "poly": [ + 303, + 1391, + 847, + 1391, + 847, + 1421, + 303, + 1421 + ], + "score": 0.108 + }, + { + "category_id": 15, + "poly": [ + 328.0, + 1970.0, + 1213.0, + 1970.0, + 1213.0, + 2008.0, + 328.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 330.0, + 2000.0, + 877.0, + 2000.0, + 877.0, + 2039.0, + 330.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 218.0, + 1380.0, + 218.0, + 1380.0, + 276.0, + 295.0, + 276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1004.0, + 579.0, + 1004.0, + 579.0, + 1051.0, + 294.0, + 1051.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 630.0, + 934.0, + 630.0, + 934.0, + 673.0, + 768.0, + 673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1390.0, + 852.0, + 1390.0, + 852.0, + 1425.0, + 296.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 2088.0, + 860.0, + 2088.0, + 860.0, + 2118.0, + 841.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 700.0, + 1304.0, + 700.0, + 1304.0, + 741.0, + 394.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 735.0, + 1305.0, + 735.0, + 1305.0, + 769.0, + 395.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 763.0, + 1305.0, + 763.0, + 1305.0, + 799.0, + 394.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 796.0, + 1305.0, + 796.0, + 1305.0, + 829.0, + 395.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 826.0, + 1306.0, + 826.0, + 1306.0, + 860.0, + 395.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 857.0, + 1306.0, + 857.0, + 1306.0, + 891.0, + 394.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 886.0, + 1306.0, + 886.0, + 1306.0, + 923.0, + 394.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 916.0, + 1257.0, + 916.0, + 1257.0, + 953.0, + 394.0, + 953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1076.0, + 1406.0, + 1076.0, + 1406.0, + 1114.0, + 295.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1106.0, + 1406.0, + 1106.0, + 1406.0, + 1145.0, + 293.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1140.0, + 1405.0, + 1140.0, + 1405.0, + 1172.0, + 296.0, + 1172.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1168.0, + 1406.0, + 1168.0, + 1406.0, + 1205.0, + 294.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1197.0, + 1405.0, + 1197.0, + 1405.0, + 1236.0, + 293.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1228.0, + 1404.0, + 1228.0, + 1404.0, + 1267.0, + 293.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1258.0, + 1405.0, + 1258.0, + 1405.0, + 1296.0, + 292.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1290.0, + 1405.0, + 1290.0, + 1405.0, + 1327.0, + 294.0, + 1327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1320.0, + 833.0, + 1320.0, + 833.0, + 1355.0, + 295.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1448.0, + 1405.0, + 1448.0, + 1405.0, + 1484.0, + 296.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1480.0, + 1404.0, + 1480.0, + 1404.0, + 1513.0, + 296.0, + 1513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1508.0, + 1404.0, + 1508.0, + 1404.0, + 1544.0, + 293.0, + 1544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1535.0, + 1405.0, + 1535.0, + 1405.0, + 1580.0, + 292.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1565.0, + 1407.0, + 1565.0, + 1407.0, + 1610.0, + 292.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1603.0, + 1403.0, + 1603.0, + 1403.0, + 1636.0, + 296.0, + 1636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1629.0, + 1405.0, + 1629.0, + 1405.0, + 1668.0, + 292.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1662.0, + 1405.0, + 1662.0, + 1405.0, + 1698.0, + 294.0, + 1698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1693.0, + 582.0, + 1693.0, + 582.0, + 1726.0, + 296.0, + 1726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1736.0, + 1405.0, + 1736.0, + 1405.0, + 1775.0, + 293.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1768.0, + 1406.0, + 1768.0, + 1406.0, + 1805.0, + 293.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1799.0, + 1406.0, + 1799.0, + 1406.0, + 1837.0, + 294.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1829.0, + 1406.0, + 1829.0, + 1406.0, + 1867.0, + 294.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1862.0, + 1403.0, + 1862.0, + 1403.0, + 1893.0, + 295.0, + 1893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1892.0, + 1405.0, + 1892.0, + 1405.0, + 1926.0, + 295.0, + 1926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1923.0, + 1146.0, + 1923.0, + 1146.0, + 1957.0, + 295.0, + 1957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 318.0, + 462.0, + 318.0, + 462.0, + 353.0, + 309.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 348.0, + 472.0, + 348.0, + 472.0, + 387.0, + 311.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 381.0, + 588.0, + 381.0, + 588.0, + 419.0, + 312.0, + 419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 311.0, + 458.0, + 578.0, + 458.0, + 578.0, + 493.0, + 311.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 491.0, + 471.0, + 491.0, + 471.0, + 524.0, + 313.0, + 524.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 316.0, + 524.0, + 636.0, + 524.0, + 636.0, + 556.0, + 316.0, + 556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 317.0, + 816.0, + 317.0, + 816.0, + 353.0, + 678.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 348.0, + 905.0, + 348.0, + 905.0, + 387.0, + 677.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 385.0, + 1036.0, + 385.0, + 1036.0, + 414.0, + 678.0, + 414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 321.0, + 1271.0, + 321.0, + 1271.0, + 353.0, + 1129.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 351.0, + 1346.0, + 351.0, + 1346.0, + 387.0, + 1129.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1127.0, + 384.0, + 1388.0, + 384.0, + 1388.0, + 416.0, + 1127.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1390.0, + 852.0, + 1390.0, + 852.0, + 1425.0, + 296.0, + 1425.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 521, + 1403, + 521, + 1403, + 885, + 298, + 885 + ], + "score": 0.985 + }, + { + "category_id": 1, + "poly": [ + 298, + 980, + 1404, + 980, + 1404, + 1317, + 298, + 1317 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 229, + 1404, + 229, + 1404, + 506, + 298, + 506 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1911, + 1403, + 1911, + 1403, + 2035, + 298, + 2035 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 295, + 1332, + 1398, + 1332, + 1398, + 1395, + 295, + 1395 + ], + "score": 0.952 + }, + { + "category_id": 1, + "poly": [ + 298, + 1562, + 1399, + 1562, + 1399, + 1627, + 298, + 1627 + ], + "score": 0.943 + }, + { + "category_id": 1, + "poly": [ + 300, + 1824, + 690, + 1824, + 690, + 1856, + 300, + 1856 + ], + "score": 0.928 + }, + { + "category_id": 0, + "poly": [ + 298, + 1506, + 849, + 1506, + 849, + 1538, + 298, + 1538 + ], + "score": 0.923 + }, + { + "category_id": 0, + "poly": [ + 300, + 923, + 895, + 923, + 895, + 955, + 300, + 955 + ], + "score": 0.91 + }, + { + "category_id": 0, + "poly": [ + 299, + 1437, + 481, + 1437, + 481, + 1474, + 299, + 1474 + ], + "score": 0.906 + }, + { + "category_id": 8, + "poly": [ + 708, + 1635, + 1049, + 1635, + 1049, + 1691, + 708, + 1691 + ], + "score": 0.893 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 814, + 76, + 814, + 104, + 300, + 104 + ], + "score": 0.893 + }, + { + "category_id": 8, + "poly": [ + 748, + 1698, + 1048, + 1698, + 1048, + 1751, + 748, + 1751 + ], + "score": 0.891 + }, + { + "category_id": 8, + "poly": [ + 651, + 1759, + 1010, + 1759, + 1010, + 1813, + 651, + 1813 + ], + "score": 0.886 + }, + { + "category_id": 8, + "poly": [ + 369, + 1864, + 1325, + 1864, + 1325, + 1905, + 369, + 1905 + ], + "score": 0.884 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1701, + 1400, + 1701, + 1400, + 1730, + 1366, + 1730 + ], + "score": 0.877 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1870, + 1400, + 1870, + 1400, + 1899, + 1366, + 1899 + ], + "score": 0.871 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1762, + 1400, + 1762, + 1400, + 1791, + 1366, + 1791 + ], + "score": 0.868 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1639, + 1400, + 1639, + 1400, + 1669, + 1366, + 1669 + ], + "score": 0.866 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.726 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.229 + }, + { + "category_id": 14, + "poly": [ + 650, + 1631, + 1052, + 1631, + 1052, + 1817, + 650, + 1817 + ], + "score": 0.93, + "latex": "\\begin{array} { c } { { \\theta _ { G } ^ { * } = \\underset { \\theta _ { G } } { \\mathrm { a r g m i n } } \\underset { \\theta _ { D } } { \\mathrm { m a x } } f \\left( \\theta _ { G } , \\theta _ { D } \\right) } } \\\\ { { \\ \\mathrm { ~ } } } \\\\ { { \\displaystyle \\qquad = \\underset { \\theta _ { G } } { \\mathrm { a r g m i n } } f \\left( \\theta _ { G } , \\theta _ { D } ^ { * } \\left( \\theta _ { G } \\right) \\right) } } \\\\ { { \\theta _ { D } ^ { * } \\left( \\theta _ { G } \\right) = \\underset { \\theta _ { D } } { \\mathrm { a r g m a x } } f \\left( \\theta _ { G } , \\theta _ { D } \\right) , } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1292, + 1563, + 1400, + 1563, + 1400, + 1597, + 1292, + 1597 + ], + "score": 0.92, + "latex": "G \\left( z ; \\theta _ { G } \\right)" + }, + { + "category_id": 13, + "poly": [ + 397, + 2002, + 611, + 2002, + 611, + 2036, + 397, + 2036 + ], + "score": 0.92, + "latex": "G ( \\cdot ; \\theta _ { G } ) : \\mathcal { Z } \\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 360, + 1912, + 448, + 1912, + 448, + 1940, + 360, + 1940 + ], + "score": 0.92, + "latex": "x \\in \\mathcal { X }" + }, + { + "category_id": 13, + "poly": [ + 686, + 1913, + 769, + 1913, + 769, + 1940, + 686, + 1940 + ], + "score": 0.9, + "latex": "z \\in { \\mathcal { Z } }" + }, + { + "category_id": 13, + "poly": [ + 981, + 1565, + 1016, + 1565, + 1016, + 1598, + 981, + 1598 + ], + "score": 0.9, + "latex": "\\theta _ { G } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 453, + 1942, + 709, + 1942, + 709, + 1974, + 453, + 1974 + ], + "score": 0.89, + "latex": "D ( \\cdot ; \\theta _ { D } ) : \\mathcal { X } [ 0 , 1 ]" + }, + { + "category_id": 14, + "poly": [ + 372, + 1865, + 1327, + 1865, + 1327, + 1905, + 372, + 1905 + ], + "score": 0.89, + "latex": "f \\left( \\theta _ { G } , \\theta _ { D } \\right) = \\mathbb { E } _ { x \\sim p _ { d a t a } } \\left[ \\log \\left( D \\left( x ; \\theta _ { D } \\right) \\right) \\right] + \\mathbb { E } _ { z \\sim N ( 0 , I ) } \\left[ \\log \\left( 1 - D \\left( G \\left( z ; \\theta _ { G } \\right) ; \\theta _ { D } \\right) \\right) \\right] ." + }, + { + "category_id": 13, + "poly": [ + 1023, + 1918, + 1086, + 1918, + 1086, + 1944, + 1023, + 1944 + ], + "score": 0.88, + "latex": "p _ { d a t a }" + }, + { + "category_id": 13, + "poly": [ + 626, + 1975, + 660, + 1975, + 660, + 2003, + 626, + 2003 + ], + "score": 0.87, + "latex": "\\theta _ { G }" + }, + { + "category_id": 13, + "poly": [ + 536, + 1975, + 572, + 1975, + 572, + 2003, + 536, + 2003 + ], + "score": 0.87, + "latex": "\\theta _ { D }" + }, + { + "category_id": 13, + "poly": [ + 372, + 1826, + 391, + 1826, + 391, + 1857, + 372, + 1857 + ], + "score": 0.85, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 1241, + 1948, + 1261, + 1948, + 1261, + 1970, + 1241, + 1970 + ], + "score": 0.76, + "latex": "x" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1504.0, + 851.0, + 1504.0, + 851.0, + 1541.0, + 294.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 920.0, + 899.0, + 920.0, + 899.0, + 961.0, + 294.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1433.0, + 486.0, + 1433.0, + 486.0, + 1481.0, + 291.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 817.0, + 73.0, + 817.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2120.0, + 838.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 519.0, + 1407.0, + 519.0, + 1407.0, + 558.0, + 293.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 551.0, + 1403.0, + 551.0, + 1403.0, + 587.0, + 293.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 581.0, + 1407.0, + 581.0, + 1407.0, + 620.0, + 292.0, + 620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 614.0, + 1405.0, + 614.0, + 1405.0, + 647.0, + 294.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 643.0, + 1403.0, + 643.0, + 1403.0, + 676.0, + 294.0, + 676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 673.0, + 1405.0, + 673.0, + 1405.0, + 708.0, + 293.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 703.0, + 1405.0, + 703.0, + 1405.0, + 738.0, + 293.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 736.0, + 1405.0, + 736.0, + 1405.0, + 769.0, + 294.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 765.0, + 1406.0, + 765.0, + 1406.0, + 800.0, + 292.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 795.0, + 1407.0, + 795.0, + 1407.0, + 832.0, + 293.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 825.0, + 1407.0, + 825.0, + 1407.0, + 863.0, + 292.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 854.0, + 418.0, + 854.0, + 418.0, + 889.0, + 294.0, + 889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 983.0, + 1402.0, + 983.0, + 1402.0, + 1014.0, + 296.0, + 1014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1012.0, + 1405.0, + 1012.0, + 1405.0, + 1047.0, + 296.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1040.0, + 1404.0, + 1040.0, + 1404.0, + 1078.0, + 293.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1073.0, + 1402.0, + 1073.0, + 1402.0, + 1108.0, + 294.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1104.0, + 1406.0, + 1104.0, + 1406.0, + 1139.0, + 294.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1132.0, + 1408.0, + 1132.0, + 1408.0, + 1171.0, + 292.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1165.0, + 1405.0, + 1165.0, + 1405.0, + 1199.0, + 294.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1194.0, + 1405.0, + 1194.0, + 1405.0, + 1229.0, + 294.0, + 1229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1224.0, + 1404.0, + 1224.0, + 1404.0, + 1259.0, + 296.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1255.0, + 1404.0, + 1255.0, + 1404.0, + 1291.0, + 293.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1285.0, + 1228.0, + 1285.0, + 1228.0, + 1321.0, + 295.0, + 1321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 231.0, + 1405.0, + 231.0, + 1405.0, + 264.0, + 296.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 263.0, + 1404.0, + 263.0, + 1404.0, + 296.0, + 296.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 288.0, + 1406.0, + 288.0, + 1406.0, + 332.0, + 292.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 320.0, + 1405.0, + 320.0, + 1405.0, + 357.0, + 294.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 352.0, + 1405.0, + 352.0, + 1405.0, + 388.0, + 294.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 382.0, + 1402.0, + 382.0, + 1402.0, + 415.0, + 296.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 412.0, + 1403.0, + 412.0, + 1403.0, + 447.0, + 293.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 441.0, + 1406.0, + 441.0, + 1406.0, + 480.0, + 293.0, + 480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 476.0, + 803.0, + 476.0, + 803.0, + 509.0, + 296.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1909.0, + 359.0, + 1909.0, + 359.0, + 1949.0, + 292.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1909.0, + 685.0, + 1909.0, + 685.0, + 1949.0, + 449.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 770.0, + 1909.0, + 1022.0, + 1909.0, + 1022.0, + 1949.0, + 770.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1909.0, + 1407.0, + 1909.0, + 1407.0, + 1949.0, + 1087.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1943.0, + 452.0, + 1943.0, + 452.0, + 1975.0, + 296.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 1943.0, + 1240.0, + 1943.0, + 1240.0, + 1975.0, + 710.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1262.0, + 1943.0, + 1405.0, + 1943.0, + 1405.0, + 1975.0, + 1262.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1972.0, + 535.0, + 1972.0, + 535.0, + 2007.0, + 294.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 1972.0, + 625.0, + 1972.0, + 625.0, + 2007.0, + 573.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 1972.0, + 1404.0, + 1972.0, + 1404.0, + 2007.0, + 661.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2000.0, + 396.0, + 2000.0, + 396.0, + 2040.0, + 293.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 612.0, + 2000.0, + 1403.0, + 2000.0, + 1403.0, + 2040.0, + 612.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1328.0, + 1403.0, + 1328.0, + 1403.0, + 1370.0, + 293.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1360.0, + 1065.0, + 1360.0, + 1065.0, + 1403.0, + 294.0, + 1403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1558.0, + 980.0, + 1558.0, + 980.0, + 1601.0, + 293.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1017.0, + 1558.0, + 1291.0, + 1558.0, + 1291.0, + 1601.0, + 1017.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1594.0, + 565.0, + 1594.0, + 565.0, + 1631.0, + 294.0, + 1631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1824.0, + 371.0, + 1824.0, + 371.0, + 1860.0, + 297.0, + 1860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1824.0, + 691.0, + 1824.0, + 691.0, + 1860.0, + 392.0, + 1860.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 700, + 1405, + 700, + 1405, + 949, + 297, + 949 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 393, + 1404, + 393, + 1404, + 549, + 298, + 549 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 562, + 1404, + 562, + 1404, + 688, + 298, + 688 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 1520, + 1402, + 1520, + 1402, + 1646, + 298, + 1646 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1942, + 1401, + 1942, + 1401, + 2036, + 298, + 2036 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 299, + 1314, + 1402, + 1314, + 1402, + 1407, + 299, + 1407 + ], + "score": 0.97 + }, + { + "category_id": 8, + "poly": [ + 679, + 310, + 1019, + 310, + 1019, + 384, + 679, + 384 + ], + "score": 0.957 + }, + { + "category_id": 1, + "poly": [ + 293, + 228, + 1402, + 228, + 1402, + 294, + 293, + 294 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 297, + 1041, + 1403, + 1041, + 1403, + 1104, + 297, + 1104 + ], + "score": 0.947 + }, + { + "category_id": 8, + "poly": [ + 688, + 1860, + 999, + 1860, + 999, + 1929, + 688, + 1929 + ], + "score": 0.943 + }, + { + "category_id": 8, + "poly": [ + 640, + 1468, + 1057, + 1468, + 1057, + 1511, + 640, + 1511 + ], + "score": 0.941 + }, + { + "category_id": 8, + "poly": [ + 696, + 1162, + 1035, + 1162, + 1035, + 1239, + 696, + 1239 + ], + "score": 0.937 + }, + { + "category_id": 8, + "poly": [ + 690, + 1784, + 1008, + 1784, + 1008, + 1854, + 690, + 1854 + ], + "score": 0.936 + }, + { + "category_id": 1, + "poly": [ + 293, + 1421, + 1280, + 1421, + 1280, + 1455, + 293, + 1455 + ], + "score": 0.929 + }, + { + "category_id": 1, + "poly": [ + 296, + 1738, + 1181, + 1738, + 1181, + 1772, + 296, + 1772 + ], + "score": 0.927 + }, + { + "category_id": 8, + "poly": [ + 662, + 1245, + 897, + 1245, + 897, + 1297, + 662, + 1297 + ], + "score": 0.919 + }, + { + "category_id": 0, + "poly": [ + 299, + 1681, + 632, + 1681, + 632, + 1713, + 299, + 1713 + ], + "score": 0.917 + }, + { + "category_id": 2, + "poly": [ + 299, + 74, + 815, + 74, + 815, + 105, + 299, + 105 + ], + "score": 0.914 + }, + { + "category_id": 0, + "poly": [ + 300, + 983, + 597, + 983, + 597, + 1015, + 300, + 1015 + ], + "score": 0.913 + }, + { + "category_id": 8, + "poly": [ + 719, + 1117, + 827, + 1117, + 827, + 1154, + 719, + 1154 + ], + "score": 0.904 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1880, + 1399, + 1880, + 1399, + 1911, + 1352, + 1911 + ], + "score": 0.896 + }, + { + "category_id": 9, + "poly": [ + 1366, + 329, + 1400, + 329, + 1400, + 359, + 1366, + 359 + ], + "score": 0.89 + }, + { + "category_id": 9, + "poly": [ + 1352, + 1806, + 1400, + 1806, + 1400, + 1836, + 1352, + 1836 + ], + "score": 0.889 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1474, + 1400, + 1474, + 1400, + 1504, + 1366, + 1504 + ], + "score": 0.887 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1251, + 1400, + 1251, + 1400, + 1280, + 1366, + 1280 + ], + "score": 0.88 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1122, + 1400, + 1122, + 1400, + 1151, + 1366, + 1151 + ], + "score": 0.876 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1187, + 1400, + 1187, + 1400, + 1217, + 1366, + 1217 + ], + "score": 0.862 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.664 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.493 + }, + { + "category_id": 14, + "poly": [ + 676, + 305, + 1024, + 305, + 1024, + 383, + 676, + 383 + ], + "score": 0.95, + "latex": "D ^ { * } \\left( x \\right) = \\frac { p _ { d a t a } \\left( x \\right) } { p _ { d a t a } \\left( x \\right) + p _ { G } \\left( x \\right) } ." + }, + { + "category_id": 14, + "poly": [ + 660, + 1111, + 1038, + 1111, + 1038, + 1301, + 660, + 1301 + ], + "score": 0.93, + "latex": "\\begin{array} { c } { { \\theta _ { D } ^ { 0 } = \\theta _ { D } } } \\\\ { { \\displaystyle } } \\\\ { { \\theta _ { D } ^ { k + 1 } = \\theta _ { D } ^ { k } + \\eta ^ { k } \\displaystyle \\frac { \\mathrm { d } f ( \\theta _ { G } , \\theta _ { D } ^ { k } ) } { \\mathrm { d } \\theta _ { D } ^ { k } } } } \\\\ { { \\displaystyle } } \\\\ { { \\theta _ { D } ^ { * } ( \\theta _ { G } ) = \\displaystyle \\operatorname* { l i m } _ { k \\infty } \\theta _ { D } ^ { k } , } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 516, + 456, + 630, + 456, + 630, + 490, + 516, + 490 + ], + "score": 0.93, + "latex": "D \\left( x ; \\theta _ { D } \\right)" + }, + { + "category_id": 13, + "poly": [ + 844, + 794, + 965, + 794, + 965, + 828, + 844, + 828 + ], + "score": 0.93, + "latex": "f \\left( { \\theta } _ { G } , { \\theta } _ { D } \\right)" + }, + { + "category_id": 13, + "poly": [ + 682, + 456, + 790, + 456, + 790, + 490, + 682, + 490 + ], + "score": 0.93, + "latex": "G \\left( z ; \\theta _ { G } \\right)" + }, + { + "category_id": 13, + "poly": [ + 375, + 734, + 495, + 734, + 495, + 767, + 375, + 767 + ], + "score": 0.93, + "latex": "f \\left( { \\theta } _ { G } , { \\theta } _ { D } \\right)" + }, + { + "category_id": 14, + "poly": [ + 687, + 1782, + 1009, + 1782, + 1009, + 1937, + 687, + 1937 + ], + "score": 0.93, + "latex": "\\begin{array} { r l } & { \\theta _ { G } \\theta _ { G } - \\eta \\frac { \\mathrm { d } f _ { K } ( \\theta _ { G } , \\theta _ { D } ) } { \\mathrm { d } \\theta _ { G } } } \\\\ & { \\theta _ { D } \\theta _ { D } + \\eta \\frac { \\mathrm { d } f ( \\theta _ { G } , \\theta _ { D } ) } { \\mathrm { d } \\theta _ { D } } . } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 638, + 1466, + 1061, + 1466, + 1061, + 1511, + 638, + 1511 + ], + "score": 0.93, + "latex": "f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) = f \\left( \\theta _ { G } , \\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) \\right) ." + }, + { + "category_id": 13, + "poly": [ + 932, + 230, + 1013, + 230, + 1013, + 264, + 932, + 264 + ], + "score": 0.92, + "latex": "D ^ { \\ast } \\left( x \\right)" + }, + { + "category_id": 13, + "poly": [ + 1035, + 395, + 1112, + 395, + 1112, + 429, + 1035, + 429 + ], + "score": 0.92, + "latex": "p _ { G } \\left( x \\right)" + }, + { + "category_id": 13, + "poly": [ + 905, + 1552, + 1075, + 1552, + 1075, + 1586, + 905, + 1586 + ], + "score": 0.92, + "latex": "f \\left( \\theta _ { G } , \\theta _ { D } ^ { * } \\left( G \\right) \\right)" + }, + { + "category_id": 13, + "poly": [ + 536, + 702, + 706, + 702, + 706, + 736, + 536, + 736 + ], + "score": 0.92, + "latex": "\\theta ^ { * } = \\{ \\theta _ { G } ^ { * } , \\theta _ { D } ^ { * } \\}" + }, + { + "category_id": 13, + "poly": [ + 537, + 262, + 614, + 262, + 614, + 295, + 537, + 295 + ], + "score": 0.92, + "latex": "p _ { G } \\left( x \\right)" + }, + { + "category_id": 13, + "poly": [ + 922, + 426, + 999, + 426, + 999, + 459, + 922, + 459 + ], + "score": 0.91, + "latex": "p _ { G } \\left( x \\right)" + }, + { + "category_id": 13, + "poly": [ + 1221, + 885, + 1360, + 885, + 1360, + 917, + 1221, + 917 + ], + "score": 0.91, + "latex": "f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right)" + }, + { + "category_id": 13, + "poly": [ + 1194, + 917, + 1370, + 917, + 1370, + 950, + 1194, + 950 + ], + "score": 0.9, + "latex": "f \\left( \\theta _ { G } , \\theta _ { D } ^ { * } \\left( \\theta _ { G } \\right) \\right)" + }, + { + "category_id": 13, + "poly": [ + 372, + 1312, + 404, + 1312, + 404, + 1347, + 372, + 1347 + ], + "score": 0.9, + "latex": "\\eta ^ { k }" + }, + { + "category_id": 13, + "poly": [ + 874, + 1043, + 910, + 1043, + 910, + 1076, + 874, + 1076 + ], + "score": 0.89, + "latex": "\\theta _ { D } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 371, + 1522, + 454, + 1522, + 454, + 1550, + 371, + 1550 + ], + "score": 0.89, + "latex": "K = 0" + }, + { + "category_id": 13, + "poly": [ + 468, + 826, + 503, + 826, + 503, + 855, + 468, + 855 + ], + "score": 0.89, + "latex": "\\theta _ { D }" + }, + { + "category_id": 13, + "poly": [ + 1298, + 1522, + 1401, + 1522, + 1401, + 1550, + 1298, + 1550 + ], + "score": 0.88, + "latex": "K \\infty" + }, + { + "category_id": 13, + "poly": [ + 1316, + 626, + 1351, + 626, + 1351, + 656, + 1316, + 656 + ], + "score": 0.88, + "latex": "\\theta _ { G }" + }, + { + "category_id": 13, + "poly": [ + 408, + 657, + 443, + 657, + 443, + 686, + 408, + 686 + ], + "score": 0.88, + "latex": "\\theta _ { D }" + }, + { + "category_id": 13, + "poly": [ + 843, + 734, + 879, + 734, + 879, + 764, + 843, + 764 + ], + "score": 0.88, + "latex": "\\theta _ { D }" + }, + { + "category_id": 13, + "poly": [ + 636, + 737, + 670, + 737, + 670, + 765, + 636, + 765 + ], + "score": 0.88, + "latex": "\\theta _ { G }" + }, + { + "category_id": 13, + "poly": [ + 1366, + 796, + 1401, + 796, + 1401, + 825, + 1366, + 825 + ], + "score": 0.88, + "latex": "\\theta _ { G }" + }, + { + "category_id": 13, + "poly": [ + 1366, + 396, + 1401, + 396, + 1401, + 427, + 1366, + 427 + ], + "score": 0.87, + "latex": "\\theta _ { G }" + }, + { + "category_id": 13, + "poly": [ + 485, + 1424, + 513, + 1424, + 513, + 1450, + 485, + 1450 + ], + "score": 0.83, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 411, + 596, + 436, + 596, + 436, + 622, + 411, + 622 + ], + "score": 0.83, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 978, + 564, + 1072, + 564, + 1072, + 598, + 978, + 598 + ], + "score": 0.82, + "latex": "\\theta _ { D } ^ { * } \\left( \\theta _ { G } \\right) " + }, + { + "category_id": 13, + "poly": [ + 472, + 1584, + 500, + 1584, + 500, + 1610, + 472, + 1610 + ], + "score": 0.81, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 1159, + 1948, + 1178, + 1948, + 1178, + 1974, + 1159, + 1974 + ], + "score": 0.79, + "latex": "\\eta" + }, + { + "category_id": 13, + "poly": [ + 347, + 425, + 440, + 425, + 440, + 459, + 347, + 459 + ], + "score": 0.75, + "latex": "\\theta _ { D } ^ { * } \\left( \\theta _ { G } ^ { - } \\right)" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1681.0, + 634.0, + 1681.0, + 634.0, + 1717.0, + 295.0, + 1717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 980.0, + 601.0, + 980.0, + 601.0, + 1020.0, + 291.0, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2118.0, + 838.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 700.0, + 535.0, + 700.0, + 535.0, + 741.0, + 292.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 700.0, + 1406.0, + 700.0, + 1406.0, + 741.0, + 707.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 731.0, + 374.0, + 731.0, + 374.0, + 769.0, + 294.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 731.0, + 635.0, + 731.0, + 635.0, + 769.0, + 496.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 671.0, + 731.0, + 842.0, + 731.0, + 842.0, + 769.0, + 671.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 880.0, + 731.0, + 1405.0, + 731.0, + 1405.0, + 769.0, + 880.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 763.0, + 1405.0, + 763.0, + 1405.0, + 800.0, + 291.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 795.0, + 843.0, + 795.0, + 843.0, + 829.0, + 295.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 966.0, + 795.0, + 1365.0, + 795.0, + 1365.0, + 829.0, + 966.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 821.0, + 467.0, + 821.0, + 467.0, + 862.0, + 292.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 821.0, + 1406.0, + 821.0, + 1406.0, + 862.0, + 504.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 855.0, + 1405.0, + 855.0, + 1405.0, + 889.0, + 295.0, + 889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 884.0, + 1220.0, + 884.0, + 1220.0, + 920.0, + 292.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1361.0, + 884.0, + 1406.0, + 884.0, + 1406.0, + 920.0, + 1361.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 915.0, + 1193.0, + 915.0, + 1193.0, + 953.0, + 294.0, + 953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1371.0, + 915.0, + 1385.0, + 915.0, + 1385.0, + 953.0, + 1371.0, + 953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 394.0, + 1034.0, + 394.0, + 1034.0, + 431.0, + 296.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.0, + 394.0, + 1365.0, + 394.0, + 1365.0, + 431.0, + 1113.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 424.0, + 346.0, + 424.0, + 346.0, + 462.0, + 294.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 441.0, + 424.0, + 921.0, + 424.0, + 921.0, + 462.0, + 441.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 424.0, + 1406.0, + 424.0, + 1406.0, + 462.0, + 1000.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 454.0, + 515.0, + 454.0, + 515.0, + 492.0, + 292.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 631.0, + 454.0, + 681.0, + 454.0, + 681.0, + 492.0, + 631.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 791.0, + 454.0, + 1407.0, + 454.0, + 1407.0, + 492.0, + 791.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 486.0, + 1406.0, + 486.0, + 1406.0, + 523.0, + 293.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 516.0, + 558.0, + 516.0, + 558.0, + 554.0, + 295.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 564.0, + 977.0, + 564.0, + 977.0, + 600.0, + 294.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 564.0, + 1405.0, + 564.0, + 1405.0, + 600.0, + 1073.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 594.0, + 410.0, + 594.0, + 410.0, + 629.0, + 292.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 594.0, + 1405.0, + 594.0, + 1405.0, + 629.0, + 437.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 625.0, + 1315.0, + 625.0, + 1315.0, + 661.0, + 293.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1352.0, + 625.0, + 1405.0, + 625.0, + 1405.0, + 661.0, + 1352.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 657.0, + 407.0, + 657.0, + 407.0, + 687.0, + 293.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 444.0, + 657.0, + 453.0, + 657.0, + 453.0, + 687.0, + 444.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1518.0, + 370.0, + 1518.0, + 370.0, + 1557.0, + 293.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 1518.0, + 1297.0, + 1518.0, + 1297.0, + 1557.0, + 455.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1518.0, + 1406.0, + 1518.0, + 1406.0, + 1557.0, + 1402.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1552.0, + 904.0, + 1552.0, + 904.0, + 1589.0, + 293.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 1552.0, + 1407.0, + 1552.0, + 1407.0, + 1589.0, + 1076.0, + 1589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1584.0, + 471.0, + 1584.0, + 471.0, + 1617.0, + 296.0, + 1617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 501.0, + 1584.0, + 1404.0, + 1584.0, + 1404.0, + 1617.0, + 501.0, + 1617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1612.0, + 1283.0, + 1612.0, + 1283.0, + 1650.0, + 292.0, + 1650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1939.0, + 1158.0, + 1939.0, + 1158.0, + 1979.0, + 292.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1179.0, + 1939.0, + 1405.0, + 1939.0, + 1405.0, + 1979.0, + 1179.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1972.0, + 1405.0, + 1972.0, + 1405.0, + 2008.0, + 292.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 2004.0, + 1405.0, + 2004.0, + 1405.0, + 2039.0, + 292.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1311.0, + 371.0, + 1311.0, + 371.0, + 1351.0, + 294.0, + 1351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1311.0, + 1406.0, + 1311.0, + 1406.0, + 1351.0, + 405.0, + 1351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1344.0, + 1406.0, + 1344.0, + 1406.0, + 1380.0, + 294.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1376.0, + 813.0, + 1376.0, + 813.0, + 1409.0, + 295.0, + 1409.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 229.0, + 931.0, + 229.0, + 931.0, + 265.0, + 295.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1014.0, + 229.0, + 1405.0, + 229.0, + 1405.0, + 265.0, + 1014.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 261.0, + 536.0, + 261.0, + 536.0, + 296.0, + 293.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 615.0, + 261.0, + 912.0, + 261.0, + 912.0, + 296.0, + 615.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1038.0, + 873.0, + 1038.0, + 873.0, + 1079.0, + 292.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 911.0, + 1038.0, + 1405.0, + 1038.0, + 1405.0, + 1079.0, + 911.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1073.0, + 665.0, + 1073.0, + 665.0, + 1108.0, + 295.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1418.0, + 484.0, + 1418.0, + 484.0, + 1462.0, + 293.0, + 1462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 1418.0, + 1285.0, + 1418.0, + 1285.0, + 1462.0, + 514.0, + 1462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1733.0, + 1185.0, + 1733.0, + 1185.0, + 1780.0, + 292.0, + 1780.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 976, + 1405, + 976, + 1405, + 1163, + 297, + 1163 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 1497, + 1403, + 1497, + 1403, + 1652, + 297, + 1652 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 296, + 1833, + 1404, + 1833, + 1404, + 2035, + 296, + 2035 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 1666, + 1404, + 1666, + 1404, + 1822, + 297, + 1822 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 868, + 1405, + 868, + 1405, + 962, + 298, + 962 + ], + "score": 0.973 + }, + { + "category_id": 3, + "poly": [ + 310, + 274, + 1399, + 274, + 1399, + 572, + 310, + 572 + ], + "score": 0.96 + }, + { + "category_id": 1, + "poly": [ + 296, + 1278, + 1402, + 1278, + 1402, + 1343, + 296, + 1343 + ], + "score": 0.955 + }, + { + "category_id": 8, + "poly": [ + 395, + 1370, + 1306, + 1370, + 1306, + 1453, + 395, + 1453 + ], + "score": 0.953 + }, + { + "category_id": 0, + "poly": [ + 300, + 1215, + 742, + 1215, + 742, + 1247, + 300, + 1247 + ], + "score": 0.905 + }, + { + "category_id": 4, + "poly": [ + 295, + 642, + 1405, + 642, + 1405, + 798, + 295, + 798 + ], + "score": 0.904 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 815, + 76, + 815, + 104, + 299, + 104 + ], + "score": 0.895 + }, + { + "category_id": 9, + "poly": [ + 1353, + 1398, + 1399, + 1398, + 1399, + 1428, + 1353, + 1428 + ], + "score": 0.882 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2111, + 841, + 2111 + ], + "score": 0.773 + }, + { + "category_id": 1, + "poly": [ + 295, + 642, + 1405, + 642, + 1405, + 798, + 295, + 798 + ], + "score": 0.093 + }, + { + "category_id": 13, + "poly": [ + 914, + 1834, + 1020, + 1834, + 1020, + 1884, + 914, + 1884 + ], + "score": 0.94, + "latex": "\\frac { \\partial f } { \\partial \\theta _ { D } ^ { K } } = 0" + }, + { + "category_id": 13, + "poly": [ + 1075, + 1911, + 1196, + 1911, + 1196, + 1944, + 1075, + 1944 + ], + "score": 0.93, + "latex": "f \\bar { ( \\theta _ { G } , \\theta _ { D } ) }" + }, + { + "category_id": 13, + "poly": [ + 919, + 1280, + 1059, + 1280, + 1059, + 1314, + 919, + 1314 + ], + "score": 0.92, + "latex": "f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right)" + }, + { + "category_id": 13, + "poly": [ + 1260, + 1972, + 1400, + 1972, + 1400, + 2006, + 1260, + 2006 + ], + "score": 0.92, + "latex": "f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right)" + }, + { + "category_id": 13, + "poly": [ + 1202, + 1880, + 1341, + 1880, + 1341, + 1914, + 1202, + 1914 + ], + "score": 0.92, + "latex": "f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right)" + }, + { + "category_id": 14, + "poly": [ + 396, + 1370, + 1304, + 1370, + 1304, + 1454, + 396, + 1454 + ], + "score": 0.92, + "latex": "\\frac { \\mathrm { d } f _ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) } { \\mathrm { d } \\theta _ { G } } = \\frac { \\partial f \\left( \\theta _ { G } , \\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) \\right) } { \\partial \\theta _ { G } } + \\frac { \\partial f \\left( \\theta _ { G } , \\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) \\right) } { \\partial \\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) } \\frac { \\mathrm { d } \\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) } { \\mathrm { d } \\theta _ { G } } ." + }, + { + "category_id": 13, + "poly": [ + 651, + 1527, + 792, + 1527, + 792, + 1563, + 651, + 1563 + ], + "score": 0.92, + "latex": "\\theta _ { D } ^ { K } \\left( \\theta _ { G } , \\theta _ { D } \\right) " + }, + { + "category_id": 13, + "poly": [ + 467, + 733, + 502, + 733, + 502, + 768, + 467, + 768 + ], + "score": 0.9, + "latex": "\\theta _ { D } ^ { k }" + }, + { + "category_id": 13, + "poly": [ + 414, + 1944, + 518, + 1944, + 518, + 1971, + 414, + 1971 + ], + "score": 0.9, + "latex": "K \\infty" + }, + { + "category_id": 13, + "poly": [ + 692, + 1312, + 726, + 1312, + 726, + 1341, + 692, + 1341 + ], + "score": 0.89, + "latex": "\\theta _ { G }" + }, + { + "category_id": 13, + "poly": [ + 409, + 1912, + 444, + 1912, + 444, + 1942, + 409, + 1942 + ], + "score": 0.89, + "latex": "\\theta _ { G }" + }, + { + "category_id": 13, + "poly": [ + 851, + 1130, + 885, + 1130, + 885, + 1161, + 851, + 1161 + ], + "score": 0.89, + "latex": "\\theta _ { G }" + }, + { + "category_id": 13, + "poly": [ + 703, + 1945, + 807, + 1945, + 807, + 1970, + 703, + 1970 + ], + "score": 0.89, + "latex": "K \\infty" + }, + { + "category_id": 13, + "poly": [ + 1322, + 1913, + 1401, + 1913, + 1401, + 1940, + 1322, + 1940 + ], + "score": 0.89, + "latex": "K = 0" + }, + { + "category_id": 13, + "poly": [ + 448, + 1838, + 487, + 1838, + 487, + 1875, + 448, + 1875 + ], + "score": 0.87, + "latex": "\\theta _ { D } ^ { K }" + }, + { + "category_id": 13, + "poly": [ + 1371, + 706, + 1400, + 706, + 1400, + 737, + 1371, + 737 + ], + "score": 0.87, + "latex": "f _ { k }" + }, + { + "category_id": 13, + "poly": [ + 337, + 1840, + 435, + 1840, + 435, + 1870, + 337, + 1870 + ], + "score": 0.86, + "latex": "K \\infty" + }, + { + "category_id": 13, + "poly": [ + 804, + 1841, + 823, + 1841, + 823, + 1873, + 804, + 1873 + ], + "score": 0.84, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 537, + 1100, + 566, + 1100, + 566, + 1128, + 537, + 1128 + ], + "score": 0.78, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 1240, + 1564, + 1260, + 1564, + 1260, + 1587, + 1240, + 1587 + ], + "score": 0.77, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 799, + 706, + 818, + 706, + 818, + 732, + 799, + 732 + ], + "score": 0.76, + "latex": "k" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 279.0, + 1399.0, + 279.0, + 1399.0, + 315.0, + 1183.0, + 315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 728.0, + 340.0, + 737.0, + 340.0, + 737.0, + 351.0, + 728.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1201.0, + 327.0, + 1226.0, + 327.0, + 1226.0, + 347.0, + 1201.0, + 347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1231.0, + 310.0, + 1393.0, + 310.0, + 1393.0, + 363.0, + 1231.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 357.0, + 749.0, + 357.0, + 749.0, + 414.0, + 692.0, + 414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 357.0, + 1045.0, + 357.0, + 1045.0, + 413.0, + 985.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1229.0, + 346.0, + 1395.0, + 346.0, + 1395.0, + 407.0, + 1229.0, + 407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 394.0, + 561.0, + 394.0, + 561.0, + 449.0, + 449.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 423.0, + 591.0, + 423.0, + 591.0, + 444.0, + 571.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 618.0, + 398.0, + 687.0, + 398.0, + 687.0, + 438.0, + 618.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 423.0, + 712.0, + 423.0, + 712.0, + 439.0, + 695.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 733.0, + 396.0, + 744.0, + 396.0, + 744.0, + 406.0, + 733.0, + 406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 764.0, + 394.0, + 877.0, + 394.0, + 877.0, + 449.0, + 764.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 422.0, + 910.0, + 422.0, + 910.0, + 445.0, + 883.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 398.0, + 990.0, + 398.0, + 990.0, + 437.0, + 921.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1007.0, + 430.0, + 1018.0, + 430.0, + 1018.0, + 440.0, + 1007.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 394.0, + 1170.0, + 394.0, + 1170.0, + 450.0, + 1060.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 430.0, + 1231.0, + 430.0, + 1231.0, + 460.0, + 1200.0, + 460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1235.0, + 391.0, + 1351.0, + 391.0, + 1351.0, + 464.0, + 1235.0, + 464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 502.0, + 470.0, + 510.0, + 470.0, + 510.0, + 481.0, + 502.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 454.0, + 856.0, + 454.0, + 856.0, + 498.0, + 804.0, + 498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1086.0, + 452.0, + 1144.0, + 452.0, + 1144.0, + 500.0, + 1086.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1235.0, + 450.0, + 1359.0, + 450.0, + 1359.0, + 494.0, + 1235.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 488.0, + 506.0, + 539.0, + 506.0, + 539.0, + 559.0, + 488.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 805.0, + 508.0, + 852.0, + 508.0, + 852.0, + 557.0, + 805.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 508.0, + 1138.0, + 508.0, + 1138.0, + 557.0, + 1092.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 361.5, + 380.0, + 361.5, + 380.0, + 372.5, + 344.0, + 372.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.75, + 407.0, + 356.75, + 407.0, + 356.75, + 439.0, + 346.75, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 519.75, + 465.0, + 522.75, + 465.0, + 522.75, + 485.0, + 519.75, + 485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1215.0, + 746.0, + 1215.0, + 746.0, + 1251.0, + 295.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 641.0, + 1405.0, + 641.0, + 1405.0, + 680.0, + 293.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 672.0, + 1405.0, + 672.0, + 1405.0, + 711.0, + 294.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 703.0, + 798.0, + 703.0, + 798.0, + 741.0, + 294.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 703.0, + 1370.0, + 703.0, + 1370.0, + 741.0, + 819.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 727.0, + 466.0, + 727.0, + 466.0, + 776.0, + 292.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 727.0, + 1410.0, + 727.0, + 1410.0, + 776.0, + 503.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 764.0, + 1193.0, + 764.0, + 1193.0, + 801.0, + 295.0, + 801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 862.0, + 2086.0, + 862.0, + 2118.0, + 839.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 979.0, + 1405.0, + 979.0, + 1405.0, + 1011.0, + 295.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1008.0, + 1405.0, + 1008.0, + 1405.0, + 1043.0, + 292.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1040.0, + 1403.0, + 1040.0, + 1403.0, + 1072.0, + 296.0, + 1072.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1070.0, + 1403.0, + 1070.0, + 1403.0, + 1102.0, + 296.0, + 1102.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1099.0, + 536.0, + 1099.0, + 536.0, + 1135.0, + 294.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 1099.0, + 1405.0, + 1099.0, + 1405.0, + 1135.0, + 567.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1130.0, + 850.0, + 1130.0, + 850.0, + 1166.0, + 295.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 1130.0, + 1349.0, + 1130.0, + 1349.0, + 1166.0, + 886.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1499.0, + 1404.0, + 1499.0, + 1404.0, + 1532.0, + 296.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1525.0, + 650.0, + 1525.0, + 650.0, + 1567.0, + 293.0, + 1567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 793.0, + 1525.0, + 1408.0, + 1525.0, + 1408.0, + 1567.0, + 793.0, + 1567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1559.0, + 1239.0, + 1559.0, + 1239.0, + 1594.0, + 295.0, + 1594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 1559.0, + 1404.0, + 1559.0, + 1404.0, + 1594.0, + 1261.0, + 1594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1590.0, + 1404.0, + 1590.0, + 1404.0, + 1624.0, + 295.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1620.0, + 928.0, + 1620.0, + 928.0, + 1655.0, + 293.0, + 1655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1830.0, + 336.0, + 1830.0, + 336.0, + 1891.0, + 293.0, + 1891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1830.0, + 447.0, + 1830.0, + 447.0, + 1891.0, + 436.0, + 1891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 488.0, + 1830.0, + 803.0, + 1830.0, + 803.0, + 1891.0, + 488.0, + 1891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 824.0, + 1830.0, + 913.0, + 1830.0, + 913.0, + 1891.0, + 824.0, + 1891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 1830.0, + 1406.0, + 1830.0, + 1406.0, + 1891.0, + 1021.0, + 1891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1879.0, + 1201.0, + 1879.0, + 1201.0, + 1915.0, + 293.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1342.0, + 1879.0, + 1405.0, + 1879.0, + 1405.0, + 1915.0, + 1342.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1911.0, + 408.0, + 1911.0, + 408.0, + 1945.0, + 293.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 445.0, + 1911.0, + 1074.0, + 1911.0, + 1074.0, + 1945.0, + 445.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 1911.0, + 1321.0, + 1911.0, + 1321.0, + 1945.0, + 1197.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1911.0, + 1407.0, + 1911.0, + 1407.0, + 1945.0, + 1402.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1942.0, + 413.0, + 1942.0, + 413.0, + 1976.0, + 295.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 519.0, + 1942.0, + 702.0, + 1942.0, + 702.0, + 1976.0, + 519.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 808.0, + 1942.0, + 1405.0, + 1942.0, + 1405.0, + 1976.0, + 808.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1969.0, + 1259.0, + 1969.0, + 1259.0, + 2010.0, + 291.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 2001.0, + 1401.0, + 2001.0, + 1401.0, + 2039.0, + 295.0, + 2039.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1665.0, + 1405.0, + 1665.0, + 1405.0, + 1701.0, + 295.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1695.0, + 1405.0, + 1695.0, + 1405.0, + 1733.0, + 294.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1725.0, + 1406.0, + 1725.0, + 1406.0, + 1765.0, + 292.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1756.0, + 1405.0, + 1756.0, + 1405.0, + 1796.0, + 292.0, + 1796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1788.0, + 726.0, + 1788.0, + 726.0, + 1827.0, + 294.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 868.0, + 1407.0, + 868.0, + 1407.0, + 904.0, + 293.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 898.0, + 1405.0, + 898.0, + 1405.0, + 934.0, + 292.0, + 934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 931.0, + 1012.0, + 931.0, + 1012.0, + 965.0, + 294.0, + 965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1277.0, + 918.0, + 1277.0, + 918.0, + 1317.0, + 294.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 1277.0, + 1406.0, + 1277.0, + 1406.0, + 1317.0, + 1060.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1310.0, + 691.0, + 1310.0, + 691.0, + 1347.0, + 295.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 727.0, + 1310.0, + 737.0, + 1310.0, + 737.0, + 1347.0, + 727.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 641.0, + 1405.0, + 641.0, + 1405.0, + 680.0, + 293.0, + 680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 672.0, + 1405.0, + 672.0, + 1405.0, + 711.0, + 294.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 703.0, + 798.0, + 703.0, + 798.0, + 741.0, + 294.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 703.0, + 1370.0, + 703.0, + 1370.0, + 741.0, + 819.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 727.0, + 466.0, + 727.0, + 466.0, + 776.0, + 292.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 727.0, + 1410.0, + 727.0, + 1410.0, + 776.0, + 503.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 764.0, + 1193.0, + 764.0, + 1193.0, + 801.0, + 295.0, + 801.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1177, + 1404, + 1177, + 1404, + 1484, + 298, + 1484 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 439, + 1404, + 439, + 1404, + 685, + 298, + 685 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 838, + 1404, + 838, + 1404, + 1053, + 298, + 1053 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 299, + 1772, + 1402, + 1772, + 1402, + 1957, + 299, + 1957 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 700, + 1403, + 700, + 1403, + 823, + 298, + 823 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 299, + 289, + 1403, + 289, + 1403, + 424, + 299, + 424 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 300, + 1499, + 1403, + 1499, + 1403, + 1622, + 300, + 1622 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 298, + 1972, + 1400, + 1972, + 1400, + 2034, + 298, + 2034 + ], + "score": 0.942 + }, + { + "category_id": 1, + "poly": [ + 296, + 1637, + 1319, + 1637, + 1319, + 1669, + 296, + 1669 + ], + "score": 0.915 + }, + { + "category_id": 0, + "poly": [ + 301, + 1104, + 557, + 1104, + 557, + 1138, + 301, + 1138 + ], + "score": 0.894 + }, + { + "category_id": 2, + "poly": [ + 300, + 76, + 815, + 76, + 815, + 104, + 300, + 104 + ], + "score": 0.883 + }, + { + "category_id": 0, + "poly": [ + 302, + 1712, + 782, + 1712, + 782, + 1743, + 302, + 1743 + ], + "score": 0.877 + }, + { + "category_id": 0, + "poly": [ + 303, + 230, + 872, + 230, + 872, + 261, + 303, + 261 + ], + "score": 0.835 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 858, + 2088, + 858, + 2112, + 841, + 2112 + ], + "score": 0.696 + }, + { + "category_id": 2, + "poly": [ + 841, + 2088, + 859, + 2088, + 859, + 2112, + 841, + 2112 + ], + "score": 0.093 + }, + { + "category_id": 13, + "poly": [ + 929, + 353, + 1093, + 353, + 1093, + 398, + 929, + 398 + ], + "score": 0.95, + "latex": "\\frac { p _ { d a t a } ( x ) } { p _ { G } ( x ) + p _ { d a t a } ( x ) }" + }, + { + "category_id": 13, + "poly": [ + 1043, + 993, + 1067, + 993, + 1067, + 1019, + 1043, + 1019 + ], + "score": 0.82, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1188, + 472, + 1214, + 472, + 1214, + 498, + 1188, + 498 + ], + "score": 0.82, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 508, + 358, + 534, + 358, + 534, + 385, + 508, + 385 + ], + "score": 0.82, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 329, + 564, + 354, + 564, + 354, + 590, + 329, + 590 + ], + "score": 0.81, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 521, + 442, + 546, + 442, + 546, + 468, + 521, + 468 + ], + "score": 0.8, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1126, + 733, + 1153, + 733, + 1153, + 759, + 1126, + 759 + ], + "score": 0.8, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 965, + 840, + 992, + 840, + 992, + 867, + 965, + 867 + ], + "score": 0.8, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1321, + 763, + 1348, + 763, + 1348, + 789, + 1321, + 789 + ], + "score": 0.8, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 501, + 564, + 528, + 564, + 528, + 590, + 501, + 590 + ], + "score": 0.79, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 448, + 733, + 475, + 733, + 475, + 759, + 448, + 759 + ], + "score": 0.78, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 792, + 733, + 818, + 733, + 818, + 760, + 792, + 760 + ], + "score": 0.77, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1042, + 839, + 1066, + 839, + 1066, + 867, + 1042, + 867 + ], + "score": 0.75, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 979, + 292, + 1022, + 292, + 1022, + 321, + 979, + 321 + ], + "score": 0.75, + "latex": "( D )" + }, + { + "category_id": 13, + "poly": [ + 1172, + 702, + 1196, + 702, + 1196, + 728, + 1172, + 728 + ], + "score": 0.71, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1223, + 292, + 1265, + 292, + 1265, + 322, + 1223, + 322 + ], + "score": 0.69, + "latex": "( G )" + }, + { + "category_id": 13, + "poly": [ + 1382, + 364, + 1401, + 364, + 1401, + 385, + 1382, + 385 + ], + "score": 0.67, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 732, + 1896, + 758, + 1896, + 758, + 1923, + 732, + 1923 + ], + "score": 0.37, + "latex": "\\mathbf { G }" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1100.0, + 562.0, + 1100.0, + 562.0, + 1145.0, + 291.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1711.0, + 786.0, + 1711.0, + 786.0, + 1747.0, + 295.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 228.0, + 879.0, + 228.0, + 879.0, + 267.0, + 296.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2085.0, + 861.0, + 2085.0, + 861.0, + 2120.0, + 839.0, + 2120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2085.0, + 862.0, + 2085.0, + 862.0, + 2119.0, + 838.0, + 2119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1178.0, + 1405.0, + 1178.0, + 1405.0, + 1213.0, + 293.0, + 1213.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1206.0, + 1404.0, + 1206.0, + 1404.0, + 1243.0, + 293.0, + 1243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1235.0, + 1405.0, + 1235.0, + 1405.0, + 1275.0, + 293.0, + 1275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1271.0, + 1404.0, + 1271.0, + 1404.0, + 1303.0, + 294.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1297.0, + 1406.0, + 1297.0, + 1406.0, + 1336.0, + 292.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1329.0, + 1405.0, + 1329.0, + 1405.0, + 1367.0, + 292.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1356.0, + 1406.0, + 1356.0, + 1406.0, + 1399.0, + 291.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1392.0, + 1405.0, + 1392.0, + 1405.0, + 1427.0, + 294.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1423.0, + 1404.0, + 1423.0, + 1404.0, + 1455.0, + 296.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1447.0, + 1257.0, + 1447.0, + 1257.0, + 1492.0, + 292.0, + 1492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 438.0, + 520.0, + 438.0, + 520.0, + 474.0, + 295.0, + 474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 438.0, + 1404.0, + 438.0, + 1404.0, + 474.0, + 547.0, + 474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 471.0, + 1187.0, + 471.0, + 1187.0, + 503.0, + 293.0, + 503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 471.0, + 1405.0, + 471.0, + 1405.0, + 503.0, + 1215.0, + 503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 501.0, + 1406.0, + 501.0, + 1406.0, + 537.0, + 293.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 532.0, + 1405.0, + 532.0, + 1405.0, + 566.0, + 294.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 562.0, + 328.0, + 562.0, + 328.0, + 599.0, + 293.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 562.0, + 500.0, + 562.0, + 500.0, + 599.0, + 355.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 529.0, + 562.0, + 1408.0, + 562.0, + 1408.0, + 599.0, + 529.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 592.0, + 1404.0, + 592.0, + 1404.0, + 626.0, + 294.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 624.0, + 1405.0, + 624.0, + 1405.0, + 658.0, + 293.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 655.0, + 990.0, + 655.0, + 990.0, + 688.0, + 294.0, + 688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 838.0, + 964.0, + 838.0, + 964.0, + 873.0, + 296.0, + 873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 993.0, + 838.0, + 1041.0, + 838.0, + 1041.0, + 873.0, + 993.0, + 873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 838.0, + 1405.0, + 838.0, + 1405.0, + 873.0, + 1067.0, + 873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 870.0, + 1404.0, + 870.0, + 1404.0, + 904.0, + 293.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 900.0, + 1404.0, + 900.0, + 1404.0, + 934.0, + 296.0, + 934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 928.0, + 1404.0, + 928.0, + 1404.0, + 966.0, + 293.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 960.0, + 1404.0, + 960.0, + 1404.0, + 995.0, + 295.0, + 995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 990.0, + 1042.0, + 990.0, + 1042.0, + 1027.0, + 293.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 990.0, + 1404.0, + 990.0, + 1404.0, + 1027.0, + 1068.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1023.0, + 924.0, + 1023.0, + 924.0, + 1054.0, + 294.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1774.0, + 1404.0, + 1774.0, + 1404.0, + 1806.0, + 297.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1802.0, + 1406.0, + 1802.0, + 1406.0, + 1839.0, + 294.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1834.0, + 1406.0, + 1834.0, + 1406.0, + 1870.0, + 294.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1865.0, + 1406.0, + 1865.0, + 1406.0, + 1899.0, + 293.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1895.0, + 731.0, + 1895.0, + 731.0, + 1931.0, + 293.0, + 1931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 759.0, + 1895.0, + 1406.0, + 1895.0, + 1406.0, + 1931.0, + 759.0, + 1931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1927.0, + 676.0, + 1927.0, + 676.0, + 1959.0, + 295.0, + 1959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 699.0, + 1171.0, + 699.0, + 1171.0, + 734.0, + 293.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 699.0, + 1405.0, + 699.0, + 1405.0, + 734.0, + 1197.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 732.0, + 447.0, + 732.0, + 447.0, + 762.0, + 296.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 732.0, + 791.0, + 732.0, + 791.0, + 762.0, + 476.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 732.0, + 1125.0, + 732.0, + 1125.0, + 762.0, + 819.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 732.0, + 1403.0, + 732.0, + 1403.0, + 762.0, + 1154.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 763.0, + 1320.0, + 763.0, + 1320.0, + 796.0, + 293.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1349.0, + 763.0, + 1405.0, + 763.0, + 1405.0, + 796.0, + 1349.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 791.0, + 800.0, + 791.0, + 800.0, + 825.0, + 293.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 288.0, + 978.0, + 288.0, + 978.0, + 325.0, + 294.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 288.0, + 1222.0, + 288.0, + 1222.0, + 325.0, + 1023.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1266.0, + 288.0, + 1404.0, + 288.0, + 1404.0, + 325.0, + 1266.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 319.0, + 1405.0, + 319.0, + 1405.0, + 356.0, + 293.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 348.0, + 507.0, + 348.0, + 507.0, + 400.0, + 297.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 348.0, + 928.0, + 348.0, + 928.0, + 400.0, + 535.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 348.0, + 1381.0, + 348.0, + 1381.0, + 400.0, + 1094.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 348.0, + 1405.0, + 348.0, + 1405.0, + 400.0, + 1402.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 393.0, + 1191.0, + 393.0, + 1191.0, + 427.0, + 297.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1497.0, + 1407.0, + 1497.0, + 1407.0, + 1535.0, + 294.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1529.0, + 1407.0, + 1529.0, + 1407.0, + 1563.0, + 294.0, + 1563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1561.0, + 1404.0, + 1561.0, + 1404.0, + 1597.0, + 294.0, + 1597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1589.0, + 680.0, + 1589.0, + 680.0, + 1626.0, + 294.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1968.0, + 1405.0, + 1968.0, + 1405.0, + 2011.0, + 292.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2003.0, + 1404.0, + 2003.0, + 1404.0, + 2037.0, + 293.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1632.0, + 1326.0, + 1632.0, + 1326.0, + 1677.0, + 294.0, + 1677.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1697, + 1404, + 1697, + 1404, + 2034, + 298, + 2034 + ], + "score": 0.982 + }, + { + "category_id": 3, + "poly": [ + 297, + 737, + 1404, + 737, + 1404, + 1321, + 297, + 1321 + ], + "score": 0.97 + }, + { + "category_id": 3, + "poly": [ + 298, + 223, + 1402, + 223, + 1402, + 457, + 298, + 457 + ], + "score": 0.967 + }, + { + "category_id": 4, + "poly": [ + 296, + 491, + 1405, + 491, + 1405, + 706, + 296, + 706 + ], + "score": 0.961 + }, + { + "category_id": 4, + "poly": [ + 296, + 1351, + 1405, + 1351, + 1405, + 1476, + 296, + 1476 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 299, + 1536, + 1402, + 1536, + 1402, + 1598, + 299, + 1598 + ], + "score": 0.948 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.888 + }, + { + "category_id": 0, + "poly": [ + 296, + 1641, + 1287, + 1641, + 1287, + 1670, + 296, + 1670 + ], + "score": 0.832 + }, + { + "category_id": 2, + "poly": [ + 840, + 2089, + 859, + 2089, + 859, + 2112, + 840, + 2112 + ], + "score": 0.777 + }, + { + "category_id": 13, + "poly": [ + 1266, + 1760, + 1340, + 1760, + 1340, + 1789, + 1266, + 1789 + ], + "score": 0.51, + "latex": "2 8 \\mathbf { x } 2 8" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 730.0, + 830.0, + 730.0, + 830.0, + 1030.0, + 757.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 734.0, + 1103.0, + 734.0, + 1103.0, + 1038.0, + 1043.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 735.0, + 375.0, + 735.0, + 375.0, + 1297.0, + 325.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 735.0, + 446.0, + 735.0, + 446.0, + 1026.0, + 393.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.0, + 736.0, + 481.0, + 736.0, + 481.0, + 1298.0, + 429.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 466.0, + 736.0, + 510.0, + 736.0, + 510.0, + 1294.0, + 466.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 496.0, + 736.0, + 547.0, + 736.0, + 547.0, + 1030.0, + 496.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 736.0, + 615.0, + 736.0, + 615.0, + 1021.0, + 571.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 603.0, + 736.0, + 651.0, + 736.0, + 651.0, + 1024.0, + 603.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 736.0, + 717.0, + 736.0, + 717.0, + 1021.0, + 673.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 806.0, + 736.0, + 858.0, + 736.0, + 858.0, + 1023.0, + 806.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 736.0, + 893.0, + 736.0, + 893.0, + 1022.0, + 845.0, + 1022.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 876.0, + 736.0, + 927.0, + 736.0, + 927.0, + 1024.0, + 876.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 913.0, + 736.0, + 965.0, + 736.0, + 965.0, + 1024.0, + 913.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 736.0, + 999.0, + 736.0, + 999.0, + 1024.0, + 947.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1013.0, + 736.0, + 1064.0, + 736.0, + 1064.0, + 1026.0, + 1013.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1152.0, + 736.0, + 1204.0, + 736.0, + 1204.0, + 1037.0, + 1152.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 737.0, + 579.0, + 737.0, + 579.0, + 1030.0, + 531.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 638.0, + 737.0, + 682.0, + 737.0, + 682.0, + 1020.0, + 638.0, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 737.0, + 1132.0, + 737.0, + 1132.0, + 1025.0, + 1087.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 737.0, + 1167.0, + 737.0, + 1167.0, + 1028.0, + 1123.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1328.0, + 737.0, + 1372.0, + 737.0, + 1372.0, + 1293.0, + 1328.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1362.0, + 737.0, + 1406.0, + 737.0, + 1406.0, + 1296.0, + 1362.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 738.0, + 339.0, + 738.0, + 339.0, + 1024.0, + 295.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 738.0, + 784.0, + 738.0, + 784.0, + 1021.0, + 740.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 738.0, + 1026.0, + 738.0, + 1026.0, + 1021.0, + 981.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1191.0, + 738.0, + 1235.0, + 738.0, + 1235.0, + 1021.0, + 1191.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 739.0, + 751.0, + 739.0, + 751.0, + 1019.0, + 711.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1226.0, + 739.0, + 1270.0, + 739.0, + 1270.0, + 1021.0, + 1226.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 366.0, + 740.0, + 407.0, + 740.0, + 407.0, + 1019.0, + 366.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 740.0, + 1301.0, + 740.0, + 1301.0, + 1020.0, + 1260.0, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1296.0, + 740.0, + 1336.0, + 740.0, + 1336.0, + 1021.0, + 1296.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 994.0, + 874.0, + 994.0, + 874.0, + 1183.0, + 746.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1003.0, + 441.0, + 1003.0, + 441.0, + 1293.0, + 404.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 1005.0, + 666.0, + 1005.0, + 666.0, + 1270.0, + 597.0, + 1270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1009.0, + 628.0, + 1009.0, + 628.0, + 1299.0, + 568.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 1009.0, + 898.0, + 1009.0, + 898.0, + 1297.0, + 845.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 1009.0, + 1164.0, + 1009.0, + 1164.0, + 1201.0, + 1123.0, + 1201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1250.0, + 1009.0, + 1303.0, + 1009.0, + 1303.0, + 1295.0, + 1250.0, + 1295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 1010.0, + 576.0, + 1010.0, + 576.0, + 1293.0, + 535.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 1011.0, + 409.0, + 1011.0, + 409.0, + 1293.0, + 369.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 507.0, + 1011.0, + 544.0, + 1011.0, + 544.0, + 1291.0, + 507.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 633.0, + 1011.0, + 693.0, + 1011.0, + 693.0, + 1096.0, + 633.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.0, + 1011.0, + 759.0, + 1011.0, + 759.0, + 1141.0, + 705.0, + 1141.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 1011.0, + 1096.0, + 1011.0, + 1096.0, + 1152.0, + 1054.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 1011.0, + 1130.0, + 1011.0, + 1130.0, + 1152.0, + 1089.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 1013.0, + 339.0, + 1013.0, + 339.0, + 1292.0, + 302.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 1013.0, + 724.0, + 1013.0, + 724.0, + 1130.0, + 672.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 1013.0, + 795.0, + 1013.0, + 795.0, + 1094.0, + 741.0, + 1094.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 1013.0, + 1023.0, + 1013.0, + 1023.0, + 1100.0, + 990.0, + 1100.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1019.0, + 1013.0, + 1059.0, + 1013.0, + 1059.0, + 1259.0, + 1019.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.0, + 1014.0, + 1234.0, + 1014.0, + 1234.0, + 1117.0, + 1196.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1298.0, + 1014.0, + 1335.0, + 1014.0, + 1335.0, + 1204.0, + 1298.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 1015.0, + 990.0, + 1015.0, + 990.0, + 1116.0, + 956.0, + 1116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 1019.0, + 1263.0, + 1019.0, + 1263.0, + 1098.0, + 1233.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 647.0, + 1081.0, + 682.0, + 1081.0, + 682.0, + 1123.0, + 647.0, + 1123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 750.0, + 1087.0, + 781.0, + 1087.0, + 781.0, + 1122.0, + 750.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 969.0, + 1107.0, + 979.0, + 1107.0, + 979.0, + 1121.0, + 969.0, + 1121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 1107.0, + 1220.0, + 1107.0, + 1220.0, + 1121.0, + 1210.0, + 1121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 1116.0, + 785.0, + 1116.0, + 785.0, + 1195.0, + 648.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 1113.0, + 852.0, + 1113.0, + 852.0, + 1164.0, + 813.0, + 1164.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 969.0, + 1127.0, + 979.0, + 1127.0, + 979.0, + 1137.0, + 969.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 1127.0, + 1011.0, + 1127.0, + 1011.0, + 1137.0, + 999.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 1128.0, + 1220.0, + 1128.0, + 1220.0, + 1137.0, + 1210.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 1146.0, + 721.0, + 1146.0, + 721.0, + 1263.0, + 648.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 778.0, + 1142.0, + 852.0, + 1142.0, + 852.0, + 1263.0, + 778.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 965.0, + 1161.0, + 980.0, + 1161.0, + 980.0, + 1181.0, + 965.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1068.0, + 1162.0, + 1084.0, + 1162.0, + 1084.0, + 1179.0, + 1068.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1099.0, + 1150.0, + 1117.0, + 1150.0, + 1117.0, + 1182.0, + 1099.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 1164.0, + 1220.0, + 1164.0, + 1220.0, + 1177.0, + 1210.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 646.0, + 1180.0, + 683.0, + 1180.0, + 683.0, + 1255.0, + 646.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 1171.0, + 785.0, + 1171.0, + 785.0, + 1262.0, + 707.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 810.0, + 1177.0, + 854.0, + 1177.0, + 854.0, + 1262.0, + 810.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 1177.0, + 983.0, + 1177.0, + 983.0, + 1252.0, + 961.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 1228.0, + 1015.0, + 1228.0, + 1015.0, + 1245.0, + 999.0, + 1245.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1063.0, + 1177.0, + 1088.0, + 1177.0, + 1088.0, + 1252.0, + 1063.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1106.0, + 1179.0, + 1112.0, + 1179.0, + 1112.0, + 1185.0, + 1106.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 1186.0, + 1160.0, + 1186.0, + 1160.0, + 1257.0, + 1131.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1203.0, + 1195.0, + 1224.0, + 1195.0, + 1224.0, + 1250.0, + 1203.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1239.0, + 1216.0, + 1257.0, + 1216.0, + 1257.0, + 1247.0, + 1239.0, + 1247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1291.0, + 1182.0, + 1334.0, + 1182.0, + 1334.0, + 1291.0, + 1291.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 604.0, + 1239.0, + 861.0, + 1239.0, + 861.0, + 1298.0, + 604.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 1253.0, + 1233.0, + 1253.0, + 1233.0, + 1287.0, + 1192.0, + 1287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 1281.0, + 488.0, + 1281.0, + 488.0, + 1329.0, + 388.0, + 1329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 660.0, + 1281.0, + 764.0, + 1281.0, + 764.0, + 1332.0, + 660.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 1284.0, + 1037.0, + 1284.0, + 1037.0, + 1326.0, + 936.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 1281.0, + 1323.0, + 1281.0, + 1323.0, + 1329.0, + 1210.0, + 1329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 1153.5, + 774.0, + 1153.5, + 774.0, + 1188.5, + 757.0, + 1188.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1217.0, + 1251.5, + 1279.0, + 1251.5, + 1279.0, + 1287.5, + 1217.0, + 1287.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 431.0, + 402.0, + 431.0, + 402.0, + 457.0, + 346.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 429.0, + 567.0, + 429.0, + 567.0, + 459.0, + 498.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 654.0, + 430.0, + 728.0, + 430.0, + 728.0, + 456.0, + 654.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 430.0, + 886.0, + 430.0, + 886.0, + 456.0, + 812.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 969.0, + 430.0, + 1046.0, + 430.0, + 1046.0, + 457.0, + 969.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 430.0, + 1206.0, + 430.0, + 1206.0, + 456.0, + 1131.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1296.0, + 431.0, + 1357.0, + 431.0, + 1357.0, + 457.0, + 1296.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 489.0, + 1406.0, + 489.0, + 1406.0, + 528.0, + 293.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 518.0, + 1405.0, + 518.0, + 1405.0, + 562.0, + 291.0, + 562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 555.0, + 1402.0, + 555.0, + 1402.0, + 586.0, + 295.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 584.0, + 1405.0, + 584.0, + 1405.0, + 618.0, + 296.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 613.0, + 1403.0, + 613.0, + 1403.0, + 648.0, + 294.0, + 648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 640.0, + 1405.0, + 640.0, + 1405.0, + 682.0, + 293.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 678.0, + 693.0, + 678.0, + 693.0, + 709.0, + 296.0, + 709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1353.0, + 1403.0, + 1353.0, + 1403.0, + 1385.0, + 295.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1384.0, + 1407.0, + 1384.0, + 1407.0, + 1417.0, + 295.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1412.0, + 1406.0, + 1412.0, + 1406.0, + 1449.0, + 294.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1443.0, + 580.0, + 1443.0, + 580.0, + 1480.0, + 294.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1637.0, + 1293.0, + 1637.0, + 1293.0, + 1674.0, + 294.0, + 1674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2087.0, + 862.0, + 2087.0, + 862.0, + 2118.0, + 840.0, + 2118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1693.0, + 1405.0, + 1693.0, + 1405.0, + 1738.0, + 293.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1726.0, + 1405.0, + 1726.0, + 1405.0, + 1765.0, + 293.0, + 1765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1759.0, + 1265.0, + 1759.0, + 1265.0, + 1793.0, + 294.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1341.0, + 1759.0, + 1405.0, + 1759.0, + 1405.0, + 1793.0, + 1341.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1790.0, + 1406.0, + 1790.0, + 1406.0, + 1825.0, + 294.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1821.0, + 1405.0, + 1821.0, + 1405.0, + 1856.0, + 294.0, + 1856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1851.0, + 1404.0, + 1851.0, + 1404.0, + 1885.0, + 294.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1882.0, + 1405.0, + 1882.0, + 1405.0, + 1916.0, + 294.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1909.0, + 1405.0, + 1909.0, + 1405.0, + 1949.0, + 292.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1942.0, + 1404.0, + 1942.0, + 1404.0, + 1977.0, + 296.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1972.0, + 1402.0, + 1972.0, + 1402.0, + 2007.0, + 294.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2003.0, + 1402.0, + 2003.0, + 1402.0, + 2037.0, + 294.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1534.0, + 1406.0, + 1534.0, + 1406.0, + 1571.0, + 293.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1566.0, + 587.0, + 1566.0, + 587.0, + 1600.0, + 294.0, + 1600.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1697, + 1404, + 1697, + 1404, + 2034, + 298, + 2034 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1076, + 1403, + 1076, + 1403, + 1290, + 298, + 1290 + ], + "score": 0.981 + }, + { + "category_id": 5, + "poly": [ + 165, + 224, + 1533, + 224, + 1533, + 388, + 165, + 388 + ], + "score": 0.978, + "html": "
Discriminator SizeUnrolling steps01510
1/4 size of D compared to GModes generated30.6± 20.7365.4 ± 34.75236.4 ± 63.30327.2 ± 74.67
KL(model||data)5.99± 0.425.911 ± 0.144.67 ± 0.434.66 ± 0.46
1/2 size of D compared to GModes generated628.0± 140.9523.6± 55.768732.0± 44.98817.4 ± 37.91
KL(model||data)2.58 ±0.7512.44 ±0.261.66 ± 0.0901.43 ± 0.12
" + }, + { + "category_id": 1, + "poly": [ + 299, + 908, + 1402, + 908, + 1402, + 1060, + 299, + 1060 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 299, + 683, + 1403, + 683, + 1403, + 808, + 299, + 808 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 299, + 1306, + 1403, + 1306, + 1403, + 1428, + 299, + 1428 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 299, + 1443, + 1403, + 1443, + 1403, + 1599, + 299, + 1599 + ], + "score": 0.975 + }, + { + "category_id": 0, + "poly": [ + 302, + 851, + 718, + 851, + 718, + 882, + 302, + 882 + ], + "score": 0.894 + }, + { + "category_id": 0, + "poly": [ + 299, + 1641, + 654, + 1641, + 654, + 1672, + 299, + 1672 + ], + "score": 0.892 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 814, + 75, + 814, + 105, + 299, + 105 + ], + "score": 0.884 + }, + { + "category_id": 0, + "poly": [ + 300, + 625, + 1104, + 625, + 1104, + 655, + 300, + 655 + ], + "score": 0.836 + }, + { + "category_id": 2, + "poly": [ + 842, + 2088, + 858, + 2088, + 858, + 2111, + 842, + 2111 + ], + "score": 0.647 + }, + { + "category_id": 1, + "poly": [ + 297, + 410, + 1405, + 410, + 1405, + 564, + 297, + 564 + ], + "score": 0.539 + }, + { + "category_id": 6, + "poly": [ + 297, + 410, + 1405, + 410, + 1405, + 564, + 297, + 564 + ], + "score": 0.351 + }, + { + "category_id": 2, + "poly": [ + 842, + 2088, + 859, + 2088, + 859, + 2111, + 842, + 2111 + ], + "score": 0.245 + }, + { + "category_id": 13, + "poly": [ + 950, + 440, + 994, + 440, + 994, + 471, + 950, + 471 + ], + "score": 0.87, + "latex": "[ 1 0 ^ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 764, + 352, + 910, + 352, + 910, + 382, + 764, + 382 + ], + "score": 0.55, + "latex": "2 . 5 8 \\pm 0 . 7 5 1" + }, + { + "category_id": 13, + "poly": [ + 1351, + 261, + 1514, + 261, + 1514, + 285, + 1351, + 285 + ], + "score": 0.46, + "latex": "{ \\bf 3 2 7 . 2 \\pm 7 4 . 6 7 }" + }, + { + "category_id": 13, + "poly": [ + 1156, + 260, + 1320, + 260, + 1320, + 285, + 1156, + 285 + ], + "score": 0.43, + "latex": "2 3 6 . 4 \\pm 6 3 . 3 0" + }, + { + "category_id": 13, + "poly": [ + 1163, + 352, + 1315, + 352, + 1315, + 382, + 1163, + 382 + ], + "score": 0.41, + "latex": "1 . 6 6 \\pm 0 . 0 9 0" + }, + { + "category_id": 13, + "poly": [ + 1362, + 351, + 1502, + 351, + 1502, + 382, + 1362, + 382 + ], + "score": 0.37, + "latex": "\\mathbf { 1 . 4 3 \\pm 0 . 1 2 }" + }, + { + "category_id": 13, + "poly": [ + 974, + 352, + 1105, + 352, + 1105, + 382, + 974, + 382 + ], + "score": 0.35, + "latex": "2 . 4 4 \\pm 0 . 2 6" + }, + { + "category_id": 13, + "poly": [ + 950, + 320, + 1126, + 320, + 1126, + 349, + 950, + 349 + ], + "score": 0.3, + "latex": "5 2 3 . 6 \\pm 5 5 . 7 6 8" + }, + { + "category_id": 13, + "poly": [ + 965, + 259, + 1113, + 259, + 1113, + 286, + 965, + 286 + ], + "score": 0.27, + "latex": "6 5 . 4 \\pm 3 4 . 7 5" + }, + { + "category_id": 13, + "poly": [ + 1158, + 320, + 1320, + 320, + 1320, + 349, + 1158, + 349 + ], + "score": 0.27, + "latex": "7 3 2 . 0 \\pm 4 4 . 9 8" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 851.0, + 723.0, + 851.0, + 723.0, + 884.0, + 296.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1638.0, + 657.0, + 1638.0, + 657.0, + 1676.0, + 295.0, + 1676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 623.0, + 1110.0, + 623.0, + 1110.0, + 660.0, + 293.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2119.0, + 840.0, + 2119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 412.0, + 1403.0, + 412.0, + 1403.0, + 445.0, + 296.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 441.0, + 949.0, + 441.0, + 949.0, + 475.0, + 294.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 441.0, + 1406.0, + 441.0, + 1406.0, + 475.0, + 995.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 472.0, + 1403.0, + 472.0, + 1403.0, + 506.0, + 295.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 499.0, + 1406.0, + 499.0, + 1406.0, + 539.0, + 294.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 530.0, + 471.0, + 530.0, + 471.0, + 567.0, + 294.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2087.0, + 861.0, + 2087.0, + 861.0, + 2119.0, + 840.0, + 2119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1698.0, + 1404.0, + 1698.0, + 1404.0, + 1733.0, + 294.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1729.0, + 1404.0, + 1729.0, + 1404.0, + 1764.0, + 294.0, + 1764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1760.0, + 1404.0, + 1760.0, + 1404.0, + 1795.0, + 294.0, + 1795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1790.0, + 1405.0, + 1790.0, + 1405.0, + 1825.0, + 294.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1818.0, + 1406.0, + 1818.0, + 1406.0, + 1857.0, + 292.0, + 1857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1851.0, + 1406.0, + 1851.0, + 1406.0, + 1885.0, + 294.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1879.0, + 1405.0, + 1879.0, + 1405.0, + 1915.0, + 293.0, + 1915.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1911.0, + 1404.0, + 1911.0, + 1404.0, + 1946.0, + 293.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1939.0, + 1406.0, + 1939.0, + 1406.0, + 1980.0, + 292.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1970.0, + 1405.0, + 1970.0, + 1405.0, + 2010.0, + 292.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2002.0, + 444.0, + 2002.0, + 444.0, + 2038.0, + 293.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1078.0, + 1404.0, + 1078.0, + 1404.0, + 1108.0, + 297.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1108.0, + 1405.0, + 1108.0, + 1405.0, + 1141.0, + 293.0, + 1141.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1138.0, + 1405.0, + 1138.0, + 1405.0, + 1172.0, + 294.0, + 1172.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1168.0, + 1406.0, + 1168.0, + 1406.0, + 1203.0, + 292.0, + 1203.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1196.0, + 1404.0, + 1196.0, + 1404.0, + 1232.0, + 293.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1231.0, + 1404.0, + 1231.0, + 1404.0, + 1261.0, + 296.0, + 1261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1260.0, + 775.0, + 1260.0, + 775.0, + 1292.0, + 296.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 905.0, + 1406.0, + 905.0, + 1406.0, + 943.0, + 294.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 937.0, + 1406.0, + 937.0, + 1406.0, + 972.0, + 293.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 970.0, + 1404.0, + 970.0, + 1404.0, + 1002.0, + 297.0, + 1002.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 998.0, + 1406.0, + 998.0, + 1406.0, + 1035.0, + 294.0, + 1035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1030.0, + 538.0, + 1030.0, + 538.0, + 1063.0, + 294.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 681.0, + 1407.0, + 681.0, + 1407.0, + 719.0, + 293.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 717.0, + 1404.0, + 717.0, + 1404.0, + 747.0, + 297.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 747.0, + 1403.0, + 747.0, + 1403.0, + 779.0, + 295.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 773.0, + 434.0, + 773.0, + 434.0, + 814.0, + 292.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1306.0, + 1404.0, + 1306.0, + 1404.0, + 1342.0, + 295.0, + 1342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1336.0, + 1405.0, + 1336.0, + 1405.0, + 1370.0, + 294.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1369.0, + 1404.0, + 1369.0, + 1404.0, + 1402.0, + 295.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1397.0, + 1376.0, + 1397.0, + 1376.0, + 1433.0, + 294.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1441.0, + 1405.0, + 1441.0, + 1405.0, + 1480.0, + 295.0, + 1480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1476.0, + 1404.0, + 1476.0, + 1404.0, + 1509.0, + 295.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1502.0, + 1407.0, + 1502.0, + 1407.0, + 1542.0, + 293.0, + 1542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1533.0, + 1405.0, + 1533.0, + 1405.0, + 1576.0, + 292.0, + 1576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1561.0, + 453.0, + 1561.0, + 453.0, + 1605.0, + 293.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 412.0, + 1403.0, + 412.0, + 1403.0, + 445.0, + 296.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 441.0, + 949.0, + 441.0, + 949.0, + 475.0, + 294.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 441.0, + 1406.0, + 441.0, + 1406.0, + 475.0, + 995.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 472.0, + 1403.0, + 472.0, + 1403.0, + 506.0, + 295.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 499.0, + 1406.0, + 499.0, + 1406.0, + 539.0, + 294.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 530.0, + 471.0, + 530.0, + 471.0, + 567.0, + 294.0, + 567.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1650, + 1404, + 1650, + 1404, + 1926, + 297, + 1926 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1220, + 1403, + 1220, + 1403, + 1405, + 298, + 1405 + ], + "score": 0.977 + }, + { + "category_id": 5, + "poly": [ + 229, + 224, + 1470, + 224, + 1470, + 356, + 229, + 356 + ], + "score": 0.975, + "html": "
Unrolling steps01510
JS divergence with 1/4 layer size0.073 ± 0.00580.142 ± 0.0280.049 ± 0.00210.075 ± 0.012
JS divergence with 1/2 layer size0.095 ± 0.0110.119 ± 0.0100.055 ± 0.00490.074± 0.016
JS divergence with 1/1 layer size0.034 ± 0.00340.050± 0.00260.027 ± 0.00280.025 ± 0.00076
" + }, + { + "category_id": 1, + "poly": [ + 299, + 1420, + 1404, + 1420, + 1404, + 1543, + 299, + 1543 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 300, + 1942, + 1401, + 1942, + 1401, + 2034, + 300, + 2034 + ], + "score": 0.971 + }, + { + "category_id": 3, + "poly": [ + 315, + 535, + 1357, + 535, + 1357, + 968, + 315, + 968 + ], + "score": 0.965 + }, + { + "category_id": 4, + "poly": [ + 297, + 1032, + 1405, + 1032, + 1405, + 1155, + 297, + 1155 + ], + "score": 0.939 + }, + { + "category_id": 0, + "poly": [ + 302, + 1589, + 744, + 1589, + 744, + 1621, + 302, + 1621 + ], + "score": 0.888 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 814, + 75, + 814, + 105, + 298, + 105 + ], + "score": 0.888 + }, + { + "category_id": 2, + "poly": [ + 841, + 2089, + 858, + 2089, + 858, + 2111, + 841, + 2111 + ], + "score": 0.793 + }, + { + "category_id": 6, + "poly": [ + 297, + 379, + 1405, + 379, + 1405, + 504, + 297, + 504 + ], + "score": 0.629 + }, + { + "category_id": 1, + "poly": [ + 297, + 379, + 1405, + 379, + 1405, + 504, + 297, + 504 + ], + "score": 0.496 + }, + { + "category_id": 13, + "poly": [ + 642, + 257, + 818, + 257, + 818, + 286, + 642, + 286 + ], + "score": 0.34, + "latex": "\\overline { { 0 . 0 7 3 \\pm 0 . 0 0 5 8 } }" + }, + { + "category_id": 13, + "poly": [ + 1055, + 258, + 1230, + 258, + 1230, + 286, + 1055, + 286 + ], + "score": 0.31, + "latex": "\\mathbf { 0 . 0 4 9 } \\pm \\mathbf { 0 . 0 0 2 1 }" + }, + { + "category_id": 13, + "poly": [ + 1263, + 318, + 1455, + 318, + 1455, + 351, + 1263, + 351 + ], + "score": 0.3, + "latex": "\\mathbf { 0 . 0 2 5 \\pm 0 . 0 0 0 7 6 }" + }, + { + "category_id": 13, + "poly": [ + 642, + 319, + 819, + 319, + 819, + 350, + 642, + 350 + ], + "score": 0.28, + "latex": "\\overline { { 0 . 0 3 4 \\pm 0 . 0 0 3 4 } }" + }, + { + "category_id": 13, + "poly": [ + 1277, + 258, + 1439, + 258, + 1439, + 286, + 1277, + 286 + ], + "score": 0.25, + "latex": "\\overline { { 0 . 0 7 5 \\pm 0 . 0 1 2 } }" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 536.0, + 485.0, + 536.0, + 485.0, + 572.0, + 404.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 665.0, + 535.0, + 748.0, + 535.0, + 748.0, + 575.0, + 665.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 533.0, + 1009.0, + 533.0, + 1009.0, + 573.0, + 926.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 534.0, + 1277.0, + 534.0, + 1277.0, + 573.0, + 1181.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1314.0, + 573.0, + 1348.0, + 573.0, + 1348.0, + 608.0, + 1314.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1117.0, + 656.0, + 1141.0, + 656.0, + 1141.0, + 675.0, + 1117.0, + 675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1117.0, + 685.0, + 1139.0, + 685.0, + 1139.0, + 704.0, + 1117.0, + 704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1118.0, + 703.0, + 1141.0, + 703.0, + 1141.0, + 729.0, + 1118.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.0, + 719.0, + 1143.0, + 719.0, + 1143.0, + 765.0, + 1113.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 779.0, + 985.0, + 779.0, + 985.0, + 806.0, + 953.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1032.0, + 1405.0, + 1032.0, + 1405.0, + 1068.0, + 295.0, + 1068.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1060.0, + 1405.0, + 1060.0, + 1405.0, + 1098.0, + 292.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1091.0, + 1407.0, + 1091.0, + 1407.0, + 1130.0, + 292.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1120.0, + 415.0, + 1120.0, + 415.0, + 1161.0, + 293.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1590.0, + 748.0, + 1590.0, + 748.0, + 1625.0, + 295.0, + 1625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2086.0, + 860.0, + 2086.0, + 860.0, + 2116.0, + 839.0, + 2116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 377.0, + 1403.0, + 377.0, + 1403.0, + 412.0, + 294.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 410.0, + 1405.0, + 410.0, + 1405.0, + 443.0, + 296.0, + 443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 440.0, + 1406.0, + 440.0, + 1406.0, + 475.0, + 292.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 471.0, + 1369.0, + 471.0, + 1369.0, + 506.0, + 294.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1651.0, + 1404.0, + 1651.0, + 1404.0, + 1684.0, + 296.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1681.0, + 1404.0, + 1681.0, + 1404.0, + 1714.0, + 295.0, + 1714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1710.0, + 1406.0, + 1710.0, + 1406.0, + 1749.0, + 294.0, + 1749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1742.0, + 1406.0, + 1742.0, + 1406.0, + 1779.0, + 294.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1774.0, + 1407.0, + 1774.0, + 1407.0, + 1810.0, + 295.0, + 1810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1802.0, + 1407.0, + 1802.0, + 1407.0, + 1839.0, + 295.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1832.0, + 1406.0, + 1832.0, + 1406.0, + 1870.0, + 294.0, + 1870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1862.0, + 1404.0, + 1862.0, + 1404.0, + 1902.0, + 292.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1892.0, + 404.0, + 1892.0, + 404.0, + 1931.0, + 294.0, + 1931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1222.0, + 1405.0, + 1222.0, + 1405.0, + 1254.0, + 296.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1250.0, + 1405.0, + 1250.0, + 1405.0, + 1286.0, + 292.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1281.0, + 1405.0, + 1281.0, + 1405.0, + 1316.0, + 294.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1311.0, + 1405.0, + 1311.0, + 1405.0, + 1347.0, + 294.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1343.0, + 1404.0, + 1343.0, + 1404.0, + 1375.0, + 296.0, + 1375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1372.0, + 704.0, + 1372.0, + 704.0, + 1407.0, + 293.0, + 1407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1421.0, + 1406.0, + 1421.0, + 1406.0, + 1454.0, + 295.0, + 1454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1449.0, + 1408.0, + 1449.0, + 1408.0, + 1486.0, + 293.0, + 1486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1480.0, + 1405.0, + 1480.0, + 1405.0, + 1519.0, + 293.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1511.0, + 1155.0, + 1511.0, + 1155.0, + 1549.0, + 293.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1941.0, + 1406.0, + 1941.0, + 1406.0, + 1978.0, + 295.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1971.0, + 1403.0, + 1971.0, + 1403.0, + 2008.0, + 295.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2000.0, + 1111.0, + 2000.0, + 1111.0, + 2038.0, + 294.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 377.0, + 1403.0, + 377.0, + 1403.0, + 412.0, + 294.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 410.0, + 1405.0, + 410.0, + 1405.0, + 443.0, + 296.0, + 443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 440.0, + 1406.0, + 440.0, + 1406.0, + 475.0, + 292.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 471.0, + 1369.0, + 471.0, + 1369.0, + 506.0, + 294.0, + 506.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1407, + 1405, + 1407, + 1405, + 1703, + 298, + 1703 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1788, + 1403, + 1788, + 1403, + 2035, + 298, + 2035 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 663, + 1403, + 663, + 1403, + 850, + 298, + 850 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 1100, + 1403, + 1100, + 1403, + 1254, + 298, + 1254 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1268, + 1404, + 1268, + 1404, + 1392, + 298, + 1392 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 299, + 993, + 1404, + 993, + 1404, + 1086, + 299, + 1086 + ], + "score": 0.971 + }, + { + "category_id": 5, + "poly": [ + 289, + 225, + 1413, + 225, + 1413, + 324, + 289, + 324 + ], + "score": 0.962, + "html": "
Unrolling Steps0 steps1 step5 steps10 steps
Average MSE0.0231± 0.00240.0195 ± 0.00210.0200± 0.00230.0181± 0.0018
PercentBestRank0.63%22.97%15.31%61.09 %
" + }, + { + "category_id": 0, + "poly": [ + 300, + 1735, + 649, + 1735, + 649, + 1768, + 300, + 1768 + ], + "score": 0.909 + }, + { + "category_id": 0, + "poly": [ + 302, + 612, + 758, + 612, + 758, + 644, + 302, + 644 + ], + "score": 0.908 + }, + { + "category_id": 9, + "poly": [ + 1353, + 931, + 1399, + 931, + 1399, + 960, + 1353, + 960 + ], + "score": 0.907 + }, + { + "category_id": 2, + "poly": [ + 298, + 75, + 815, + 75, + 815, + 105, + 298, + 105 + ], + "score": 0.894 + }, + { + "category_id": 9, + "poly": [ + 1352, + 873, + 1399, + 873, + 1399, + 904, + 1352, + 904 + ], + "score": 0.892 + }, + { + "category_id": 1, + "poly": [ + 295, + 346, + 1406, + 346, + 1406, + 562, + 295, + 562 + ], + "score": 0.799 + }, + { + "category_id": 2, + "poly": [ + 840, + 2088, + 859, + 2088, + 859, + 2111, + 840, + 2111 + ], + "score": 0.795 + }, + { + "category_id": 8, + "poly": [ + 629, + 867, + 1074, + 867, + 1074, + 920, + 629, + 920 + ], + "score": 0.696 + }, + { + "category_id": 8, + "poly": [ + 626, + 928, + 911, + 928, + 911, + 965, + 626, + 965 + ], + "score": 0.666 + }, + { + "category_id": 8, + "poly": [ + 625, + 866, + 1074, + 866, + 1074, + 966, + 625, + 966 + ], + "score": 0.296 + }, + { + "category_id": 6, + "poly": [ + 295, + 346, + 1406, + 346, + 1406, + 562, + 295, + 562 + ], + "score": 0.144 + }, + { + "category_id": 13, + "poly": [ + 550, + 500, + 658, + 500, + 658, + 534, + 550, + 534 + ], + "score": 0.94, + "latex": "G \\left( z ; \\theta _ { G } \\right)" + }, + { + "category_id": 13, + "poly": [ + 397, + 1500, + 590, + 1500, + 590, + 1555, + 397, + 1555 + ], + "score": 0.92, + "latex": "\\left| \\frac { l _ { 0 s t e p } - l _ { 1 0 s t e p } } { \\frac { 1 } { 2 } ( l _ { 0 s t e p } + l _ { 1 0 s t e p } ) } \\right|" + }, + { + "category_id": 14, + "poly": [ + 625, + 863, + 1077, + 863, + 1077, + 970, + 625, + 970 + ], + "score": 0.92, + "latex": "\\begin{array} { r l } & { z _ { \\mathrm { n e a r e s t } } = \\underset { z } { \\mathrm { a r g m i n } } | | G ( z ; \\theta _ { G } ) - x _ { \\mathrm { t a r g e t } } | \\rvert _ { 2 } ^ { 2 } } \\\\ & { x _ { \\mathrm { n e a r e s t } } = G ( z _ { \\mathrm { n e a r e s t } } ; \\theta _ { G } ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1008, + 2004, + 1034, + 2004, + 1034, + 2034, + 1008, + 2034 + ], + "score": 0.87, + "latex": "\\ell _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 615, + 822, + 689, + 822, + 689, + 848, + 615, + 848 + ], + "score": 0.86, + "latex": "x _ { \\mathrm { n e a r e s t } }" + }, + { + "category_id": 13, + "poly": [ + 1373, + 1943, + 1401, + 1943, + 1401, + 1974, + 1373, + 1974 + ], + "score": 0.85, + "latex": "\\ell _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 885, + 822, + 949, + 822, + 949, + 851, + 885, + 851 + ], + "score": 0.84, + "latex": "x _ { \\mathrm { { t a r g e t } } }" + }, + { + "category_id": 13, + "poly": [ + 1022, + 1270, + 1063, + 1270, + 1063, + 1298, + 1022, + 1298 + ], + "score": 0.83, + "latex": "1 \\%" + }, + { + "category_id": 13, + "poly": [ + 344, + 1196, + 409, + 1196, + 409, + 1226, + 344, + 1226 + ], + "score": 0.83, + "latex": "x _ { \\mathrm { { t a r g e t } } }" + }, + { + "category_id": 13, + "poly": [ + 1126, + 475, + 1146, + 475, + 1146, + 497, + 1126, + 497 + ], + "score": 0.73, + "latex": "z" + }, + { + "category_id": 13, + "poly": [ + 598, + 288, + 675, + 288, + 675, + 318, + 598, + 318 + ], + "score": 0.58, + "latex": "0 . 6 3 \\%" + }, + { + "category_id": 13, + "poly": [ + 1035, + 288, + 1123, + 288, + 1123, + 318, + 1035, + 318 + ], + "score": 0.54, + "latex": "1 5 . 3 1 \\%" + }, + { + "category_id": 13, + "poly": [ + 1252, + 288, + 1344, + 288, + 1344, + 318, + 1252, + 318 + ], + "score": 0.48, + "latex": "6 1 . 0 9 \\%" + }, + { + "category_id": 13, + "poly": [ + 812, + 288, + 903, + 288, + 903, + 318, + 812, + 318 + ], + "score": 0.46, + "latex": "2 2 . 9 7 \\%" + }, + { + "category_id": 13, + "poly": [ + 542, + 257, + 733, + 257, + 733, + 286, + 542, + 286 + ], + "score": 0.44, + "latex": "0 . 0 2 3 1 \\pm 0 . 0 0 2 4" + }, + { + "category_id": 13, + "poly": [ + 984, + 257, + 1173, + 257, + 1173, + 286, + 984, + 286 + ], + "score": 0.37, + "latex": "0 . 0 2 0 0 \\pm 0 . 0 0 2 3" + }, + { + "category_id": 13, + "poly": [ + 1204, + 256, + 1395, + 256, + 1395, + 286, + 1204, + 286 + ], + "score": 0.34, + "latex": "\\mathbf { 0 . 0 1 8 1 } \\pm \\mathbf { 0 . 0 0 1 8 }" + }, + { + "category_id": 13, + "poly": [ + 763, + 257, + 952, + 257, + 952, + 286, + 763, + 286 + ], + "score": 0.28, + "latex": "0 . 0 1 9 5 \\pm 0 . 0 0 2 1" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1733.0, + 651.0, + 1733.0, + 651.0, + 1771.0, + 294.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 611.0, + 762.0, + 611.0, + 762.0, + 646.0, + 295.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 815.0, + 73.0, + 815.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2087.0, + 861.0, + 2087.0, + 861.0, + 2117.0, + 839.0, + 2117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 347.0, + 1404.0, + 347.0, + 1404.0, + 381.0, + 294.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 379.0, + 1406.0, + 379.0, + 1406.0, + 413.0, + 294.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 407.0, + 1403.0, + 407.0, + 1403.0, + 444.0, + 293.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 437.0, + 1404.0, + 437.0, + 1404.0, + 475.0, + 294.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 466.0, + 1125.0, + 466.0, + 1125.0, + 506.0, + 293.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 466.0, + 1407.0, + 466.0, + 1407.0, + 506.0, + 1147.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 500.0, + 549.0, + 500.0, + 549.0, + 538.0, + 293.0, + 538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 500.0, + 1407.0, + 500.0, + 1407.0, + 538.0, + 659.0, + 538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 532.0, + 462.0, + 532.0, + 462.0, + 563.0, + 294.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1405.0, + 1405.0, + 1405.0, + 1405.0, + 1446.0, + 293.0, + 1446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1437.0, + 1405.0, + 1437.0, + 1405.0, + 1475.0, + 295.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1467.0, + 1406.0, + 1467.0, + 1406.0, + 1507.0, + 292.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1510.0, + 393.0, + 1510.0, + 393.0, + 1544.0, + 295.0, + 1544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 609.0, + 1509.0, + 1406.0, + 1509.0, + 1406.0, + 1544.0, + 609.0, + 1544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1549.0, + 1405.0, + 1549.0, + 1405.0, + 1583.0, + 293.0, + 1583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1579.0, + 1405.0, + 1579.0, + 1405.0, + 1617.0, + 293.0, + 1617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1611.0, + 1403.0, + 1611.0, + 1403.0, + 1645.0, + 295.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1640.0, + 1406.0, + 1640.0, + 1406.0, + 1675.0, + 295.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1670.0, + 1377.0, + 1670.0, + 1377.0, + 1706.0, + 293.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1789.0, + 1405.0, + 1789.0, + 1405.0, + 1826.0, + 293.0, + 1826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1821.0, + 1405.0, + 1821.0, + 1405.0, + 1855.0, + 294.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1851.0, + 1404.0, + 1851.0, + 1404.0, + 1885.0, + 293.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1883.0, + 1405.0, + 1883.0, + 1405.0, + 1916.0, + 296.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1909.0, + 1405.0, + 1909.0, + 1405.0, + 1948.0, + 292.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1943.0, + 1372.0, + 1943.0, + 1372.0, + 1977.0, + 294.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1973.0, + 1404.0, + 1973.0, + 1404.0, + 2007.0, + 294.0, + 2007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2003.0, + 1007.0, + 2003.0, + 1007.0, + 2037.0, + 296.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 2003.0, + 1405.0, + 2003.0, + 1405.0, + 2037.0, + 1035.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 660.0, + 1404.0, + 660.0, + 1404.0, + 703.0, + 295.0, + 703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 696.0, + 1404.0, + 696.0, + 1404.0, + 730.0, + 294.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 724.0, + 1405.0, + 724.0, + 1405.0, + 762.0, + 293.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 758.0, + 1403.0, + 758.0, + 1403.0, + 790.0, + 296.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 787.0, + 1405.0, + 787.0, + 1405.0, + 823.0, + 296.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 814.0, + 614.0, + 814.0, + 614.0, + 856.0, + 293.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 814.0, + 884.0, + 814.0, + 884.0, + 856.0, + 690.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 950.0, + 814.0, + 1222.0, + 814.0, + 1222.0, + 856.0, + 950.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1098.0, + 1405.0, + 1098.0, + 1405.0, + 1139.0, + 294.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1129.0, + 1404.0, + 1129.0, + 1404.0, + 1166.0, + 293.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1157.0, + 1407.0, + 1157.0, + 1407.0, + 1204.0, + 291.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1190.0, + 343.0, + 1190.0, + 343.0, + 1230.0, + 292.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 1190.0, + 1404.0, + 1190.0, + 1404.0, + 1230.0, + 410.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1222.0, + 1385.0, + 1222.0, + 1385.0, + 1259.0, + 294.0, + 1259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1269.0, + 1021.0, + 1269.0, + 1021.0, + 1302.0, + 294.0, + 1302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1064.0, + 1269.0, + 1404.0, + 1269.0, + 1404.0, + 1302.0, + 1064.0, + 1302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1298.0, + 1405.0, + 1298.0, + 1405.0, + 1336.0, + 292.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1330.0, + 1405.0, + 1330.0, + 1405.0, + 1366.0, + 293.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1364.0, + 532.0, + 1364.0, + 532.0, + 1391.0, + 293.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 992.0, + 1404.0, + 992.0, + 1404.0, + 1030.0, + 295.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1024.0, + 1404.0, + 1024.0, + 1404.0, + 1058.0, + 295.0, + 1058.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1055.0, + 1170.0, + 1055.0, + 1170.0, + 1089.0, + 295.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 347.0, + 1404.0, + 347.0, + 1404.0, + 381.0, + 294.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 379.0, + 1406.0, + 379.0, + 1406.0, + 413.0, + 294.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 407.0, + 1403.0, + 407.0, + 1403.0, + 444.0, + 293.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 437.0, + 1404.0, + 437.0, + 1404.0, + 475.0, + 294.0, + 475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 466.0, + 1125.0, + 466.0, + 1125.0, + 506.0, + 293.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 466.0, + 1407.0, + 466.0, + 1407.0, + 506.0, + 1147.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 500.0, + 549.0, + 500.0, + 549.0, + 538.0, + 293.0, + 538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 500.0, + 1407.0, + 500.0, + 1407.0, + 538.0, + 659.0, + 538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 532.0, + 462.0, + 532.0, + 462.0, + 563.0, + 294.0, + 563.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1665, + 1403, + 1665, + 1403, + 1880, + 298, + 1880 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1435, + 1403, + 1435, + 1403, + 1650, + 298, + 1650 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 299, + 1297, + 1403, + 1297, + 1403, + 1418, + 299, + 1418 + ], + "score": 0.977 + }, + { + "category_id": 3, + "poly": [ + 301, + 227, + 1418, + 227, + 1418, + 775, + 301, + 775 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 299, + 1083, + 1402, + 1083, + 1402, + 1176, + 299, + 1176 + ], + "score": 0.96 + }, + { + "category_id": 4, + "poly": [ + 296, + 809, + 1404, + 809, + 1404, + 1025, + 296, + 1025 + ], + "score": 0.958 + }, + { + "category_id": 1, + "poly": [ + 299, + 1971, + 1399, + 1971, + 1399, + 2034, + 299, + 2034 + ], + "score": 0.942 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.883 + }, + { + "category_id": 0, + "poly": [ + 300, + 1226, + 529, + 1226, + 529, + 1260, + 300, + 1260 + ], + "score": 0.882 + }, + { + "category_id": 0, + "poly": [ + 301, + 1919, + 557, + 1919, + 557, + 1947, + 301, + 1947 + ], + "score": 0.873 + }, + { + "category_id": 2, + "poly": [ + 837, + 2088, + 865, + 2088, + 865, + 2112, + 837, + 2112 + ], + "score": 0.845 + }, + { + "category_id": 4, + "poly": [ + 297, + 808, + 1404, + 808, + 1404, + 1024, + 297, + 1024 + ], + "score": 0.172 + }, + { + "category_id": 13, + "poly": [ + 615, + 1850, + 639, + 1850, + 639, + 1876, + 615, + 1876 + ], + "score": 0.85, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 690, + 1850, + 716, + 1850, + 716, + 1876, + 690, + 1876 + ], + "score": 0.83, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 519, + 1819, + 544, + 1819, + 544, + 1846, + 519, + 1846 + ], + "score": 0.82, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 594, + 1819, + 621, + 1819, + 621, + 1845, + 594, + 1845 + ], + "score": 0.81, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1168, + 1757, + 1193, + 1757, + 1193, + 1785, + 1168, + 1785 + ], + "score": 0.8, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1219, + 1818, + 1244, + 1818, + 1244, + 1846, + 1219, + 1846 + ], + "score": 0.77, + "latex": "G" + }, + { + "category_id": 13, + "poly": [ + 1374, + 1758, + 1401, + 1758, + 1401, + 1784, + 1374, + 1784 + ], + "score": 0.76, + "latex": "D" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 229.0, + 390.0, + 229.0, + 390.0, + 259.0, + 326.0, + 259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 222.0, + 841.0, + 222.0, + 841.0, + 264.0, + 417.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 226.0, + 970.0, + 226.0, + 970.0, + 261.0, + 904.0, + 261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 225.0, + 1416.0, + 225.0, + 1416.0, + 264.0, + 999.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 257.0, + 836.0, + 257.0, + 836.0, + 353.0, + 395.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 355.0, + 502.0, + 355.0, + 502.0, + 390.0, + 426.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 355.0, + 616.0, + 355.0, + 616.0, + 388.0, + 527.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 640.0, + 355.0, + 716.0, + 355.0, + 716.0, + 390.0, + 640.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 357.0, + 829.0, + 357.0, + 829.0, + 390.0, + 742.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 355.0, + 1086.0, + 355.0, + 1086.0, + 390.0, + 999.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 355.0, + 1194.0, + 355.0, + 1194.0, + 390.0, + 1105.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 355.0, + 1301.0, + 355.0, + 1301.0, + 390.0, + 1213.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1320.0, + 357.0, + 1408.0, + 357.0, + 1408.0, + 390.0, + 1320.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 480.0, + 510.0, + 480.0, + 510.0, + 522.0, + 421.0, + 522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 483.0, + 829.0, + 483.0, + 829.0, + 520.0, + 526.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 485.0, + 1086.0, + 485.0, + 1086.0, + 519.0, + 999.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1104.0, + 485.0, + 1193.0, + 485.0, + 1193.0, + 519.0, + 1104.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1212.0, + 485.0, + 1299.0, + 485.0, + 1299.0, + 519.0, + 1212.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1327.0, + 485.0, + 1401.0, + 485.0, + 1401.0, + 520.0, + 1327.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 612.0, + 508.0, + 612.0, + 508.0, + 645.0, + 421.0, + 645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 612.0, + 615.0, + 612.0, + 615.0, + 647.0, + 527.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.0, + 612.0, + 723.0, + 612.0, + 723.0, + 647.0, + 634.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 612.0, + 829.0, + 612.0, + 829.0, + 647.0, + 742.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 612.0, + 1086.0, + 612.0, + 1086.0, + 645.0, + 999.0, + 645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.0, + 612.0, + 1189.0, + 612.0, + 1189.0, + 647.0, + 1113.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1213.0, + 612.0, + 1302.0, + 612.0, + 1302.0, + 645.0, + 1213.0, + 645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1321.0, + 615.0, + 1406.0, + 615.0, + 1406.0, + 644.0, + 1321.0, + 644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 660.0, + 951.0, + 660.0, + 951.0, + 726.0, + 889.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 739.0, + 508.0, + 739.0, + 508.0, + 777.0, + 417.0, + 777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 739.0, + 617.0, + 739.0, + 617.0, + 777.0, + 526.0, + 777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.0, + 740.0, + 723.0, + 740.0, + 723.0, + 774.0, + 634.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 741.0, + 829.0, + 741.0, + 829.0, + 776.0, + 742.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 741.0, + 1088.0, + 741.0, + 1088.0, + 774.0, + 999.0, + 774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1104.0, + 740.0, + 1408.0, + 740.0, + 1408.0, + 777.0, + 1104.0, + 777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 810.0, + 1404.0, + 810.0, + 1404.0, + 844.0, + 295.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 840.0, + 1406.0, + 840.0, + 1406.0, + 875.0, + 295.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 866.0, + 1405.0, + 866.0, + 1405.0, + 908.0, + 291.0, + 908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 900.0, + 1405.0, + 900.0, + 1405.0, + 936.0, + 294.0, + 936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 931.0, + 1404.0, + 931.0, + 1404.0, + 966.0, + 294.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 964.0, + 1404.0, + 964.0, + 1404.0, + 995.0, + 295.0, + 995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 990.0, + 717.0, + 990.0, + 717.0, + 1029.0, + 293.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1221.0, + 535.0, + 1221.0, + 535.0, + 1268.0, + 292.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 1921.0, + 558.0, + 1921.0, + 558.0, + 1950.0, + 299.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 809.0, + 1405.0, + 809.0, + 1405.0, + 847.0, + 294.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 841.0, + 1406.0, + 841.0, + 1406.0, + 875.0, + 295.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 867.0, + 1405.0, + 867.0, + 1405.0, + 909.0, + 291.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 900.0, + 1405.0, + 900.0, + 1405.0, + 936.0, + 294.0, + 936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 931.0, + 1404.0, + 931.0, + 1404.0, + 967.0, + 292.0, + 967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 962.0, + 1404.0, + 962.0, + 1404.0, + 996.0, + 295.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 991.0, + 717.0, + 991.0, + 717.0, + 1030.0, + 292.0, + 1030.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1665.0, + 1405.0, + 1665.0, + 1405.0, + 1700.0, + 296.0, + 1700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1694.0, + 1408.0, + 1694.0, + 1408.0, + 1733.0, + 292.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1727.0, + 1405.0, + 1727.0, + 1405.0, + 1761.0, + 294.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1757.0, + 1167.0, + 1757.0, + 1167.0, + 1791.0, + 294.0, + 1791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 1757.0, + 1373.0, + 1757.0, + 1373.0, + 1791.0, + 1194.0, + 1791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1787.0, + 1405.0, + 1787.0, + 1405.0, + 1822.0, + 292.0, + 1822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1818.0, + 518.0, + 1818.0, + 518.0, + 1852.0, + 294.0, + 1852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 1818.0, + 593.0, + 1818.0, + 593.0, + 1852.0, + 545.0, + 1852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 622.0, + 1818.0, + 1218.0, + 1818.0, + 1218.0, + 1852.0, + 622.0, + 1852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 1818.0, + 1407.0, + 1818.0, + 1407.0, + 1852.0, + 1245.0, + 1852.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1849.0, + 614.0, + 1849.0, + 614.0, + 1883.0, + 294.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 640.0, + 1849.0, + 689.0, + 1849.0, + 689.0, + 1883.0, + 640.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1849.0, + 955.0, + 1849.0, + 955.0, + 1883.0, + 717.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1433.0, + 1405.0, + 1433.0, + 1405.0, + 1470.0, + 294.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1465.0, + 1404.0, + 1465.0, + 1404.0, + 1502.0, + 293.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1496.0, + 1407.0, + 1496.0, + 1407.0, + 1534.0, + 292.0, + 1534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1525.0, + 1407.0, + 1525.0, + 1407.0, + 1565.0, + 292.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1556.0, + 1405.0, + 1556.0, + 1405.0, + 1593.0, + 293.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1589.0, + 1404.0, + 1589.0, + 1404.0, + 1624.0, + 294.0, + 1624.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1618.0, + 1159.0, + 1618.0, + 1159.0, + 1654.0, + 294.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1294.0, + 1404.0, + 1294.0, + 1404.0, + 1333.0, + 292.0, + 1333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1329.0, + 1403.0, + 1329.0, + 1403.0, + 1361.0, + 294.0, + 1361.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1357.0, + 1405.0, + 1357.0, + 1405.0, + 1394.0, + 294.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1390.0, + 1124.0, + 1390.0, + 1124.0, + 1422.0, + 295.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1082.0, + 1404.0, + 1082.0, + 1404.0, + 1119.0, + 294.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1116.0, + 1406.0, + 1116.0, + 1406.0, + 1150.0, + 295.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1148.0, + 956.0, + 1148.0, + 956.0, + 1179.0, + 295.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1971.0, + 1404.0, + 1971.0, + 1404.0, + 2006.0, + 296.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2000.0, + 1405.0, + 2000.0, + 1405.0, + 2041.0, + 293.0, + 2041.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 411, + 230, + 1285, + 230, + 1285, + 953, + 411, + 953 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 297, + 1199, + 1400, + 1199, + 1400, + 1263, + 297, + 1263 + ], + "score": 0.934 + }, + { + "category_id": 4, + "poly": [ + 295, + 994, + 1405, + 994, + 1405, + 1148, + 295, + 1148 + ], + "score": 0.927 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.892 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 863, + 2088, + 863, + 2112, + 836, + 2112 + ], + "score": 0.845 + }, + { + "category_id": 15, + "poly": [ + 690.0, + 229.0, + 1067.0, + 229.0, + 1067.0, + 263.0, + 690.0, + 263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 247.0, + 475.0, + 247.0, + 475.0, + 279.0, + 435.0, + 279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 266.0, + 477.0, + 266.0, + 477.0, + 324.0, + 434.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1168.0, + 276.0, + 1245.0, + 276.0, + 1245.0, + 315.0, + 1168.0, + 315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 316.0, + 476.0, + 316.0, + 476.0, + 347.0, + 435.0, + 347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 308.0, + 1268.0, + 308.0, + 1268.0, + 356.0, + 1171.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 338.0, + 476.0, + 338.0, + 476.0, + 369.0, + 434.0, + 369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 362.0, + 476.0, + 362.0, + 476.0, + 392.0, + 435.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 386.0, + 477.0, + 386.0, + 477.0, + 416.0, + 435.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 414.0, + 476.0, + 414.0, + 476.0, + 443.0, + 435.0, + 443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 401.0, + 631.0, + 401.0, + 631.0, + 432.0, + 590.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 402.0, + 809.0, + 402.0, + 809.0, + 432.0, + 769.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 401.0, + 988.0, + 401.0, + 988.0, + 432.0, + 947.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1126.0, + 401.0, + 1166.0, + 401.0, + 1166.0, + 432.0, + 1126.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 435.0, + 477.0, + 435.0, + 477.0, + 567.0, + 410.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 440.0, + 1268.0, + 440.0, + 1268.0, + 521.0, + 1170.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 507.0, + 473.0, + 507.0, + 473.0, + 534.0, + 437.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 530.0, + 473.0, + 530.0, + 473.0, + 557.0, + 437.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 549.0, + 488.0, + 549.0, + 488.0, + 693.0, + 397.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 591.0, + 568.0, + 631.0, + 568.0, + 631.0, + 599.0, + 591.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 569.0, + 808.0, + 569.0, + 808.0, + 597.0, + 767.0, + 597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 569.0, + 988.0, + 569.0, + 988.0, + 599.0, + 947.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1126.0, + 568.0, + 1166.0, + 568.0, + 1166.0, + 599.0, + 1126.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 604.0, + 474.0, + 604.0, + 474.0, + 631.0, + 435.0, + 631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 603.0, + 1246.0, + 603.0, + 1246.0, + 650.0, + 1170.0, + 650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 626.0, + 474.0, + 626.0, + 474.0, + 653.0, + 436.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 671.0, + 476.0, + 671.0, + 476.0, + 702.0, + 435.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 638.0, + 1269.0, + 638.0, + 1269.0, + 690.0, + 1170.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 694.0, + 476.0, + 694.0, + 476.0, + 725.0, + 435.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 717.0, + 477.0, + 717.0, + 477.0, + 748.0, + 435.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 744.0, + 473.0, + 744.0, + 473.0, + 775.0, + 434.0, + 775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 591.0, + 733.0, + 631.0, + 733.0, + 631.0, + 764.0, + 591.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 770.0, + 733.0, + 809.0, + 733.0, + 809.0, + 764.0, + 770.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 949.0, + 736.0, + 986.0, + 736.0, + 986.0, + 763.0, + 949.0, + 763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1127.0, + 733.0, + 1167.0, + 733.0, + 1167.0, + 764.0, + 1127.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 766.0, + 478.0, + 766.0, + 478.0, + 822.0, + 433.0, + 822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 772.0, + 1230.0, + 772.0, + 1230.0, + 814.0, + 1154.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 814.0, + 477.0, + 814.0, + 477.0, + 845.0, + 434.0, + 845.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1155.0, + 807.0, + 1268.0, + 807.0, + 1268.0, + 853.0, + 1155.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 837.0, + 476.0, + 837.0, + 476.0, + 867.0, + 434.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 858.0, + 477.0, + 858.0, + 477.0, + 893.0, + 434.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 884.0, + 476.0, + 884.0, + 476.0, + 913.0, + 435.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 899.0, + 631.0, + 899.0, + 631.0, + 930.0, + 590.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 899.0, + 810.0, + 899.0, + 810.0, + 930.0, + 769.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 899.0, + 988.0, + 899.0, + 988.0, + 930.0, + 947.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1127.0, + 899.0, + 1166.0, + 899.0, + 1166.0, + 930.0, + 1127.0, + 930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 918.0, + 927.0, + 918.0, + 927.0, + 958.0, + 829.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 996.0, + 1405.0, + 996.0, + 1405.0, + 1029.0, + 295.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1022.0, + 1406.0, + 1022.0, + 1406.0, + 1060.0, + 293.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1055.0, + 1406.0, + 1055.0, + 1406.0, + 1091.0, + 294.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1086.0, + 1406.0, + 1086.0, + 1406.0, + 1121.0, + 294.0, + 1121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1115.0, + 1320.0, + 1115.0, + 1320.0, + 1150.0, + 294.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1199.0, + 1404.0, + 1199.0, + 1404.0, + 1235.0, + 296.0, + 1235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1228.0, + 955.0, + 1228.0, + 955.0, + 1264.0, + 295.0, + 1264.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 299, + 75, + 815, + 75, + 815, + 105, + 299, + 105 + ], + "score": 0.888 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2113, + 836, + 2113 + ], + "score": 0.831 + }, + { + "category_id": 0, + "poly": [ + 299, + 228, + 489, + 228, + 489, + 261, + 299, + 261 + ], + "score": 0.737 + }, + { + "category_id": 1, + "poly": [ + 298, + 278, + 1402, + 278, + 1402, + 370, + 298, + 370 + ], + "score": 0.718 + }, + { + "category_id": 1, + "poly": [ + 292, + 747, + 1403, + 747, + 1403, + 810, + 292, + 810 + ], + "score": 0.707 + }, + { + "category_id": 1, + "poly": [ + 294, + 829, + 1402, + 829, + 1402, + 893, + 294, + 893 + ], + "score": 0.69 + }, + { + "category_id": 1, + "poly": [ + 301, + 1972, + 1396, + 1972, + 1396, + 2034, + 301, + 2034 + ], + "score": 0.678 + }, + { + "category_id": 1, + "poly": [ + 297, + 1348, + 1401, + 1348, + 1401, + 1414, + 297, + 1414 + ], + "score": 0.678 + }, + { + "category_id": 1, + "poly": [ + 294, + 665, + 1404, + 665, + 1404, + 730, + 294, + 730 + ], + "score": 0.671 + }, + { + "category_id": 1, + "poly": [ + 291, + 584, + 1403, + 584, + 1403, + 648, + 291, + 648 + ], + "score": 0.643 + }, + { + "category_id": 1, + "poly": [ + 292, + 502, + 1402, + 502, + 1402, + 566, + 292, + 566 + ], + "score": 0.64 + }, + { + "category_id": 1, + "poly": [ + 292, + 1104, + 1401, + 1104, + 1401, + 1169, + 292, + 1169 + ], + "score": 0.637 + }, + { + "category_id": 1, + "poly": [ + 296, + 1430, + 1403, + 1430, + 1403, + 1524, + 296, + 1524 + ], + "score": 0.637 + }, + { + "category_id": 1, + "poly": [ + 297, + 1268, + 1399, + 1268, + 1399, + 1332, + 297, + 1332 + ], + "score": 0.635 + }, + { + "category_id": 1, + "poly": [ + 298, + 1542, + 1399, + 1542, + 1399, + 1639, + 298, + 1639 + ], + "score": 0.632 + }, + { + "category_id": 1, + "poly": [ + 299, + 1185, + 1399, + 1185, + 1399, + 1250, + 299, + 1250 + ], + "score": 0.609 + }, + { + "category_id": 1, + "poly": [ + 296, + 389, + 1400, + 389, + 1400, + 484, + 296, + 484 + ], + "score": 0.609 + }, + { + "category_id": 1, + "poly": [ + 298, + 992, + 1398, + 992, + 1398, + 1086, + 298, + 1086 + ], + "score": 0.594 + }, + { + "category_id": 1, + "poly": [ + 296, + 909, + 1401, + 909, + 1401, + 975, + 296, + 975 + ], + "score": 0.593 + }, + { + "category_id": 1, + "poly": [ + 297, + 1654, + 1403, + 1654, + 1403, + 1810, + 297, + 1810 + ], + "score": 0.565 + }, + { + "category_id": 1, + "poly": [ + 298, + 1828, + 1404, + 1828, + 1404, + 1953, + 298, + 1953 + ], + "score": 0.516 + }, + { + "category_id": 1, + "poly": [ + 299, + 228, + 489, + 228, + 489, + 261, + 299, + 261 + ], + "score": 0.131 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 228.0, + 490.0, + 228.0, + 490.0, + 265.0, + 297.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 271.0, + 1407.0, + 271.0, + 1407.0, + 318.0, + 292.0, + 318.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 310.0, + 1404.0, + 310.0, + 1404.0, + 343.0, + 322.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 337.0, + 397.0, + 337.0, + 397.0, + 371.0, + 322.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 745.0, + 1405.0, + 745.0, + 1405.0, + 784.0, + 296.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 779.0, + 707.0, + 779.0, + 707.0, + 812.0, + 320.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 829.0, + 1405.0, + 829.0, + 1405.0, + 865.0, + 296.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 861.0, + 707.0, + 861.0, + 707.0, + 894.0, + 319.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1968.0, + 1401.0, + 1968.0, + 1401.0, + 2009.0, + 295.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 2002.0, + 1111.0, + 2002.0, + 1111.0, + 2037.0, + 322.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1347.0, + 1405.0, + 1347.0, + 1405.0, + 1387.0, + 293.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1382.0, + 707.0, + 1382.0, + 707.0, + 1413.0, + 320.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 661.0, + 1406.0, + 661.0, + 1406.0, + 704.0, + 292.0, + 704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 696.0, + 732.0, + 696.0, + 732.0, + 732.0, + 325.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 584.0, + 1405.0, + 584.0, + 1405.0, + 620.0, + 295.0, + 620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 617.0, + 895.0, + 617.0, + 895.0, + 647.0, + 323.0, + 647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 498.0, + 1405.0, + 498.0, + 1405.0, + 540.0, + 292.0, + 540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 534.0, + 611.0, + 534.0, + 611.0, + 563.0, + 323.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1103.0, + 1404.0, + 1103.0, + 1404.0, + 1143.0, + 294.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1137.0, + 892.0, + 1137.0, + 892.0, + 1167.0, + 324.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1427.0, + 1403.0, + 1427.0, + 1403.0, + 1469.0, + 295.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1462.0, + 1405.0, + 1462.0, + 1405.0, + 1496.0, + 321.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1487.0, + 399.0, + 1487.0, + 399.0, + 1528.0, + 320.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1267.0, + 1403.0, + 1267.0, + 1403.0, + 1306.0, + 293.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1300.0, + 857.0, + 1300.0, + 857.0, + 1332.0, + 322.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1542.0, + 1405.0, + 1542.0, + 1405.0, + 1581.0, + 293.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1576.0, + 1405.0, + 1576.0, + 1405.0, + 1610.0, + 323.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1605.0, + 1206.0, + 1605.0, + 1206.0, + 1640.0, + 324.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1185.0, + 1403.0, + 1185.0, + 1403.0, + 1221.0, + 295.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1218.0, + 794.0, + 1218.0, + 794.0, + 1250.0, + 321.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 390.0, + 1404.0, + 390.0, + 1404.0, + 424.0, + 295.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 421.0, + 1405.0, + 421.0, + 1405.0, + 457.0, + 323.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 451.0, + 611.0, + 451.0, + 611.0, + 485.0, + 321.0, + 485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 993.0, + 1402.0, + 993.0, + 1402.0, + 1027.0, + 295.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1024.0, + 1405.0, + 1024.0, + 1405.0, + 1060.0, + 322.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1055.0, + 777.0, + 1055.0, + 777.0, + 1089.0, + 324.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 908.0, + 1403.0, + 908.0, + 1403.0, + 947.0, + 295.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 941.0, + 1029.0, + 941.0, + 1029.0, + 976.0, + 321.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1655.0, + 1404.0, + 1655.0, + 1404.0, + 1691.0, + 294.0, + 1691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1681.0, + 1407.0, + 1681.0, + 1407.0, + 1727.0, + 320.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1715.0, + 1405.0, + 1715.0, + 1405.0, + 1753.0, + 322.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1744.0, + 1403.0, + 1744.0, + 1403.0, + 1788.0, + 320.0, + 1788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1777.0, + 1174.0, + 1777.0, + 1174.0, + 1815.0, + 322.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1830.0, + 1404.0, + 1830.0, + 1404.0, + 1862.0, + 296.0, + 1862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1860.0, + 1402.0, + 1860.0, + 1402.0, + 1893.0, + 323.0, + 1893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1889.0, + 1404.0, + 1889.0, + 1404.0, + 1926.0, + 322.0, + 1926.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1925.0, + 554.0, + 1925.0, + 554.0, + 1954.0, + 323.0, + 1954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 228.0, + 490.0, + 228.0, + 490.0, + 265.0, + 297.0, + 265.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 300, + 230, + 1397, + 230, + 1397, + 324, + 300, + 324 + ], + "score": 0.885 + }, + { + "category_id": 1, + "poly": [ + 295, + 1244, + 1402, + 1244, + 1402, + 1340, + 295, + 1340 + ], + "score": 0.874 + }, + { + "category_id": 1, + "poly": [ + 298, + 346, + 1404, + 346, + 1404, + 471, + 298, + 471 + ], + "score": 0.872 + }, + { + "category_id": 1, + "poly": [ + 299, + 954, + 1399, + 954, + 1399, + 1021, + 299, + 1021 + ], + "score": 0.872 + }, + { + "category_id": 1, + "poly": [ + 297, + 868, + 1401, + 868, + 1401, + 933, + 297, + 933 + ], + "score": 0.862 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 815, + 75, + 815, + 105, + 299, + 105 + ], + "score": 0.861 + }, + { + "category_id": 1, + "poly": [ + 293, + 783, + 1399, + 783, + 1399, + 847, + 293, + 847 + ], + "score": 0.859 + }, + { + "category_id": 1, + "poly": [ + 297, + 1158, + 1402, + 1158, + 1402, + 1223, + 297, + 1223 + ], + "score": 0.847 + }, + { + "category_id": 1, + "poly": [ + 295, + 695, + 1402, + 695, + 1402, + 761, + 295, + 761 + ], + "score": 0.847 + }, + { + "category_id": 1, + "poly": [ + 299, + 1942, + 1401, + 1942, + 1401, + 2035, + 299, + 2035 + ], + "score": 0.843 + }, + { + "category_id": 1, + "poly": [ + 297, + 1681, + 1402, + 1681, + 1402, + 1746, + 297, + 1746 + ], + "score": 0.832 + }, + { + "category_id": 1, + "poly": [ + 295, + 1854, + 1400, + 1854, + 1400, + 1919, + 295, + 1919 + ], + "score": 0.83 + }, + { + "category_id": 1, + "poly": [ + 292, + 1768, + 1400, + 1768, + 1400, + 1833, + 292, + 1833 + ], + "score": 0.823 + }, + { + "category_id": 1, + "poly": [ + 295, + 493, + 1401, + 493, + 1401, + 558, + 295, + 558 + ], + "score": 0.821 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 864, + 2088, + 864, + 2113, + 836, + 2113 + ], + "score": 0.818 + }, + { + "category_id": 1, + "poly": [ + 291, + 1361, + 1402, + 1361, + 1402, + 1426, + 291, + 1426 + ], + "score": 0.803 + }, + { + "category_id": 1, + "poly": [ + 298, + 1041, + 1405, + 1041, + 1405, + 1136, + 298, + 1136 + ], + "score": 0.799 + }, + { + "category_id": 1, + "poly": [ + 298, + 578, + 1406, + 578, + 1406, + 673, + 298, + 673 + ], + "score": 0.781 + }, + { + "category_id": 1, + "poly": [ + 296, + 1448, + 1398, + 1448, + 1398, + 1512, + 296, + 1512 + ], + "score": 0.774 + }, + { + "category_id": 1, + "poly": [ + 296, + 1533, + 1404, + 1533, + 1404, + 1660, + 296, + 1660 + ], + "score": 0.768 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 227.0, + 1403.0, + 227.0, + 1403.0, + 269.0, + 294.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 260.0, + 1401.0, + 260.0, + 1401.0, + 295.0, + 325.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 292.0, + 903.0, + 292.0, + 903.0, + 326.0, + 325.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1242.0, + 1406.0, + 1242.0, + 1406.0, + 1283.0, + 294.0, + 1283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1276.0, + 1403.0, + 1276.0, + 1403.0, + 1310.0, + 323.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1308.0, + 611.0, + 1308.0, + 611.0, + 1337.0, + 323.0, + 1337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 345.0, + 1404.0, + 345.0, + 1404.0, + 384.0, + 293.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 376.0, + 1405.0, + 376.0, + 1405.0, + 413.0, + 320.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 410.0, + 1401.0, + 410.0, + 1401.0, + 442.0, + 323.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 439.0, + 987.0, + 439.0, + 987.0, + 473.0, + 323.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 955.0, + 1403.0, + 955.0, + 1403.0, + 991.0, + 296.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 986.0, + 1069.0, + 986.0, + 1069.0, + 1021.0, + 323.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 866.0, + 1406.0, + 866.0, + 1406.0, + 907.0, + 292.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 901.0, + 622.0, + 901.0, + 622.0, + 934.0, + 322.0, + 934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 779.0, + 1403.0, + 779.0, + 1403.0, + 821.0, + 292.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 814.0, + 752.0, + 814.0, + 752.0, + 846.0, + 326.0, + 846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1156.0, + 1401.0, + 1156.0, + 1401.0, + 1196.0, + 295.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1188.0, + 755.0, + 1188.0, + 755.0, + 1224.0, + 324.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 692.0, + 1406.0, + 692.0, + 1406.0, + 735.0, + 293.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 728.0, + 596.0, + 728.0, + 596.0, + 759.0, + 323.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1939.0, + 1402.0, + 1939.0, + 1402.0, + 1979.0, + 293.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1972.0, + 1405.0, + 1972.0, + 1405.0, + 2009.0, + 324.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 2003.0, + 718.0, + 2003.0, + 718.0, + 2037.0, + 324.0, + 2037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1679.0, + 1403.0, + 1679.0, + 1403.0, + 1719.0, + 295.0, + 1719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1714.0, + 999.0, + 1714.0, + 999.0, + 1747.0, + 322.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1850.0, + 1404.0, + 1850.0, + 1404.0, + 1893.0, + 294.0, + 1893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1885.0, + 831.0, + 1885.0, + 831.0, + 1920.0, + 322.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1764.0, + 1406.0, + 1764.0, + 1406.0, + 1808.0, + 294.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1798.0, + 1300.0, + 1798.0, + 1300.0, + 1834.0, + 320.0, + 1834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 493.0, + 1405.0, + 493.0, + 1405.0, + 529.0, + 294.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 526.0, + 971.0, + 526.0, + 971.0, + 559.0, + 322.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1360.0, + 1404.0, + 1360.0, + 1404.0, + 1399.0, + 294.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1394.0, + 1258.0, + 1394.0, + 1258.0, + 1427.0, + 322.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1040.0, + 1403.0, + 1040.0, + 1403.0, + 1079.0, + 293.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1069.0, + 1407.0, + 1069.0, + 1407.0, + 1112.0, + 321.0, + 1112.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1103.0, + 1359.0, + 1103.0, + 1359.0, + 1138.0, + 321.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 579.0, + 1404.0, + 579.0, + 1404.0, + 613.0, + 296.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 611.0, + 1403.0, + 611.0, + 1403.0, + 645.0, + 322.0, + 645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 638.0, + 399.0, + 638.0, + 399.0, + 674.0, + 320.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1448.0, + 1402.0, + 1448.0, + 1402.0, + 1484.0, + 297.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1479.0, + 997.0, + 1479.0, + 997.0, + 1513.0, + 321.0, + 1513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1535.0, + 1404.0, + 1535.0, + 1404.0, + 1568.0, + 295.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1567.0, + 1404.0, + 1567.0, + 1404.0, + 1600.0, + 322.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1594.0, + 1402.0, + 1594.0, + 1402.0, + 1630.0, + 322.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1628.0, + 671.0, + 1628.0, + 671.0, + 1658.0, + 324.0, + 1658.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 300, + 75, + 815, + 75, + 815, + 105, + 300, + 105 + ], + "score": 0.886 + }, + { + "category_id": 1, + "poly": [ + 299, + 229, + 1399, + 229, + 1399, + 323, + 299, + 323 + ], + "score": 0.866 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2112, + 836, + 2112 + ], + "score": 0.843 + }, + { + "category_id": 1, + "poly": [ + 291, + 1105, + 1401, + 1105, + 1401, + 1171, + 291, + 1171 + ], + "score": 0.842 + }, + { + "category_id": 1, + "poly": [ + 296, + 1302, + 1402, + 1302, + 1402, + 1397, + 296, + 1397 + ], + "score": 0.839 + }, + { + "category_id": 1, + "poly": [ + 295, + 1559, + 1397, + 1559, + 1397, + 1626, + 295, + 1626 + ], + "score": 0.835 + }, + { + "category_id": 1, + "poly": [ + 297, + 1189, + 1400, + 1189, + 1400, + 1283, + 297, + 1283 + ], + "score": 0.833 + }, + { + "category_id": 1, + "poly": [ + 298, + 993, + 1401, + 993, + 1401, + 1087, + 298, + 1087 + ], + "score": 0.829 + }, + { + "category_id": 1, + "poly": [ + 297, + 621, + 1406, + 621, + 1406, + 746, + 297, + 746 + ], + "score": 0.817 + }, + { + "category_id": 1, + "poly": [ + 297, + 765, + 1405, + 765, + 1405, + 859, + 297, + 859 + ], + "score": 0.816 + }, + { + "category_id": 1, + "poly": [ + 296, + 1725, + 1398, + 1725, + 1398, + 1792, + 296, + 1792 + ], + "score": 0.814 + }, + { + "category_id": 1, + "poly": [ + 302, + 508, + 1402, + 508, + 1402, + 603, + 302, + 603 + ], + "score": 0.813 + }, + { + "category_id": 1, + "poly": [ + 295, + 1642, + 1400, + 1642, + 1400, + 1709, + 295, + 1709 + ], + "score": 0.795 + }, + { + "category_id": 1, + "poly": [ + 296, + 879, + 1404, + 879, + 1404, + 972, + 296, + 972 + ], + "score": 0.784 + }, + { + "category_id": 1, + "poly": [ + 294, + 425, + 1400, + 425, + 1400, + 490, + 294, + 490 + ], + "score": 0.783 + }, + { + "category_id": 1, + "poly": [ + 296, + 1416, + 1404, + 1416, + 1404, + 1543, + 296, + 1543 + ], + "score": 0.778 + }, + { + "category_id": 1, + "poly": [ + 296, + 1809, + 1405, + 1809, + 1405, + 1904, + 296, + 1904 + ], + "score": 0.773 + }, + { + "category_id": 1, + "poly": [ + 295, + 342, + 1399, + 342, + 1399, + 407, + 295, + 407 + ], + "score": 0.771 + }, + { + "category_id": 13, + "poly": [ + 326, + 943, + 427, + 943, + 427, + 972, + 326, + 972 + ], + "score": 0.3, + "latex": "0 3 4 7 8 /" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2122.0, + 832.0, + 2122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 229.0, + 1404.0, + 229.0, + 1404.0, + 264.0, + 294.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 259.0, + 1404.0, + 259.0, + 1404.0, + 297.0, + 323.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 291.0, + 614.0, + 291.0, + 614.0, + 324.0, + 323.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1104.0, + 1403.0, + 1104.0, + 1403.0, + 1143.0, + 294.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1137.0, + 1217.0, + 1137.0, + 1217.0, + 1173.0, + 322.0, + 1173.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1303.0, + 1400.0, + 1303.0, + 1400.0, + 1337.0, + 295.0, + 1337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1335.0, + 1404.0, + 1335.0, + 1404.0, + 1369.0, + 323.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1364.0, + 624.0, + 1364.0, + 624.0, + 1398.0, + 323.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1560.0, + 1403.0, + 1560.0, + 1403.0, + 1596.0, + 294.0, + 1596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1591.0, + 1194.0, + 1591.0, + 1194.0, + 1628.0, + 321.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1190.0, + 1402.0, + 1190.0, + 1402.0, + 1224.0, + 296.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1221.0, + 1402.0, + 1221.0, + 1402.0, + 1255.0, + 322.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1252.0, + 506.0, + 1252.0, + 506.0, + 1282.0, + 323.0, + 1282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 994.0, + 1403.0, + 994.0, + 1403.0, + 1028.0, + 294.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 1022.0, + 1402.0, + 1022.0, + 1402.0, + 1061.0, + 319.0, + 1061.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1057.0, + 573.0, + 1057.0, + 573.0, + 1087.0, + 324.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 620.0, + 1406.0, + 620.0, + 1406.0, + 659.0, + 291.0, + 659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 654.0, + 1406.0, + 654.0, + 1406.0, + 687.0, + 324.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 682.0, + 1404.0, + 682.0, + 1404.0, + 720.0, + 321.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 716.0, + 505.0, + 716.0, + 505.0, + 746.0, + 324.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 766.0, + 1405.0, + 766.0, + 1405.0, + 800.0, + 296.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 797.0, + 1402.0, + 797.0, + 1402.0, + 831.0, + 321.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 826.0, + 456.0, + 826.0, + 456.0, + 862.0, + 323.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1726.0, + 1402.0, + 1726.0, + 1402.0, + 1762.0, + 295.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1757.0, + 774.0, + 1757.0, + 774.0, + 1792.0, + 324.0, + 1792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 510.0, + 1403.0, + 510.0, + 1403.0, + 545.0, + 296.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 542.0, + 1404.0, + 542.0, + 1404.0, + 574.0, + 321.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 570.0, + 1084.0, + 570.0, + 1084.0, + 604.0, + 323.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1639.0, + 1405.0, + 1639.0, + 1405.0, + 1682.0, + 294.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1674.0, + 1116.0, + 1674.0, + 1116.0, + 1709.0, + 323.0, + 1709.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 878.0, + 1404.0, + 878.0, + 1404.0, + 916.0, + 294.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 911.0, + 1402.0, + 911.0, + 1402.0, + 945.0, + 323.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 428.0, + 940.0, + 441.0, + 940.0, + 441.0, + 975.0, + 428.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 428.0, + 1401.0, + 428.0, + 1401.0, + 460.0, + 298.0, + 460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 458.0, + 1400.0, + 458.0, + 1400.0, + 491.0, + 322.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1418.0, + 1402.0, + 1418.0, + 1402.0, + 1451.0, + 295.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1448.0, + 1406.0, + 1448.0, + 1406.0, + 1484.0, + 323.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 1478.0, + 1405.0, + 1478.0, + 1405.0, + 1510.0, + 321.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1510.0, + 1216.0, + 1510.0, + 1216.0, + 1543.0, + 324.0, + 1543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1810.0, + 1403.0, + 1810.0, + 1403.0, + 1844.0, + 296.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1839.0, + 1406.0, + 1839.0, + 1406.0, + 1878.0, + 320.0, + 1878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1871.0, + 497.0, + 1871.0, + 497.0, + 1906.0, + 323.0, + 1906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 341.0, + 1403.0, + 341.0, + 1403.0, + 380.0, + 295.0, + 380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 373.0, + 1213.0, + 373.0, + 1213.0, + 409.0, + 321.0, + 409.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 299, + 738, + 1403, + 738, + 1403, + 863, + 299, + 863 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 298, + 1773, + 1404, + 1773, + 1404, + 1896, + 298, + 1896 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 299, + 1911, + 1403, + 1911, + 1403, + 2035, + 299, + 2035 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 299, + 523, + 1404, + 523, + 1404, + 616, + 299, + 616 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 298, + 631, + 1404, + 631, + 1404, + 725, + 298, + 725 + ], + "score": 0.967 + }, + { + "category_id": 3, + "poly": [ + 298, + 1282, + 1399, + 1282, + 1399, + 1587, + 298, + 1587 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 298, + 878, + 1401, + 878, + 1401, + 941, + 298, + 941 + ], + "score": 0.947 + }, + { + "category_id": 1, + "poly": [ + 296, + 445, + 1400, + 445, + 1400, + 508, + 296, + 508 + ], + "score": 0.932 + }, + { + "category_id": 4, + "poly": [ + 296, + 1619, + 1405, + 1619, + 1405, + 1744, + 296, + 1744 + ], + "score": 0.923 + }, + { + "category_id": 0, + "poly": [ + 300, + 220, + 539, + 220, + 539, + 282, + 300, + 282 + ], + "score": 0.917 + }, + { + "category_id": 1, + "poly": [ + 298, + 1216, + 1236, + 1216, + 1236, + 1250, + 298, + 1250 + ], + "score": 0.916 + }, + { + "category_id": 0, + "poly": [ + 298, + 329, + 851, + 329, + 851, + 368, + 298, + 368 + ], + "score": 0.911 + }, + { + "category_id": 1, + "poly": [ + 296, + 398, + 1352, + 398, + 1352, + 432, + 296, + 432 + ], + "score": 0.911 + }, + { + "category_id": 0, + "poly": [ + 301, + 983, + 1012, + 983, + 1012, + 1019, + 301, + 1019 + ], + "score": 0.901 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 815, + 75, + 815, + 105, + 299, + 105 + ], + "score": 0.892 + }, + { + "category_id": 1, + "poly": [ + 299, + 1108, + 1403, + 1108, + 1403, + 1202, + 299, + 1202 + ], + "score": 0.861 + }, + { + "category_id": 2, + "poly": [ + 837, + 2088, + 864, + 2088, + 864, + 2112, + 837, + 2112 + ], + "score": 0.853 + }, + { + "category_id": 0, + "poly": [ + 297, + 1052, + 982, + 1052, + 982, + 1084, + 297, + 1084 + ], + "score": 0.648 + }, + { + "category_id": 1, + "poly": [ + 297, + 1052, + 982, + 1052, + 982, + 1084, + 297, + 1084 + ], + "score": 0.279 + }, + { + "category_id": 13, + "poly": [ + 585, + 740, + 1030, + 740, + 1030, + 773, + 585, + 773 + ], + "score": 0.92, + "latex": "\\mathcal { L } _ { G } = \\log ( D ( x ) ) + \\log ( 1 - D ( G ( z ) ) )" + }, + { + "category_id": 13, + "poly": [ + 298, + 801, + 421, + 801, + 421, + 833, + 298, + 833 + ], + "score": 0.9, + "latex": "\\mathcal { N } ( 0 , I _ { 2 5 6 } )" + }, + { + "category_id": 13, + "poly": [ + 400, + 833, + 482, + 833, + 482, + 863, + 400, + 863 + ], + "score": 0.9, + "latex": "\\beta _ { 1 } { = } 0 . 5" + }, + { + "category_id": 13, + "poly": [ + 296, + 770, + 770, + 770, + 770, + 804, + 296, + 804 + ], + "score": 0.89, + "latex": "\\mathcal { L } _ { D } \\stackrel { = } { = } - \\log ( D ( x ) ) - \\log ( 1 - D ( \\stackrel { . } { G } ( z ) ) )" + }, + { + "category_id": 13, + "poly": [ + 1350, + 774, + 1403, + 774, + 1403, + 800, + 1350, + 800 + ], + "score": 0.82, + "latex": "z \\sim" + }, + { + "category_id": 13, + "poly": [ + 549, + 910, + 573, + 910, + 573, + 937, + 549, + 937 + ], + "score": 0.6, + "latex": "\\mathbf { D }" + }, + { + "category_id": 13, + "poly": [ + 852, + 776, + 872, + 776, + 872, + 798, + 852, + 798 + ], + "score": 0.58, + "latex": "\\mathbf { X }" + }, + { + "category_id": 13, + "poly": [ + 491, + 910, + 516, + 910, + 516, + 937, + 491, + 937 + ], + "score": 0.54, + "latex": "\\mathbf { G }" + }, + { + "category_id": 15, + "poly": [ + 1161.0, + 1297.0, + 1171.0, + 1297.0, + 1171.0, + 1305.0, + 1161.0, + 1305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1305.0, + 342.0, + 1305.0, + 342.0, + 1327.0, + 323.0, + 1327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1321.0, + 331.0, + 1321.0, + 331.0, + 1523.0, + 294.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 946.0, + 1326.0, + 956.0, + 1326.0, + 956.0, + 1336.0, + 946.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1336.0, + 1331.0, + 1346.0, + 1331.0, + 1346.0, + 1340.0, + 1336.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 1366.0, + 866.0, + 1366.0, + 866.0, + 1375.0, + 856.0, + 1375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1370.0, + 342.0, + 1370.0, + 342.0, + 1394.0, + 323.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 976.0, + 1371.0, + 990.0, + 1371.0, + 990.0, + 1380.0, + 976.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1361.0, + 1378.0, + 1371.0, + 1378.0, + 1371.0, + 1388.0, + 1361.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1171.0, + 1389.0, + 1183.0, + 1389.0, + 1183.0, + 1399.0, + 1171.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 586.0, + 1395.0, + 596.0, + 1395.0, + 596.0, + 1404.0, + 586.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1039.0, + 1394.0, + 1051.0, + 1394.0, + 1051.0, + 1405.0, + 1039.0, + 1405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1428.0, + 691.0, + 1428.0, + 691.0, + 1440.0, + 681.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1436.0, + 352.0, + 1436.0, + 352.0, + 1460.0, + 322.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1445.0, + 406.0, + 1445.0, + 406.0, + 1456.0, + 393.0, + 1456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1126.0, + 1443.0, + 1136.0, + 1443.0, + 1136.0, + 1455.0, + 1126.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 753.0, + 1451.0, + 766.0, + 1451.0, + 766.0, + 1461.0, + 753.0, + 1461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 859.0, + 1460.0, + 871.0, + 1460.0, + 871.0, + 1471.0, + 859.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 1457.0, + 1067.0, + 1457.0, + 1067.0, + 1469.0, + 1056.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1234.0, + 1457.0, + 1245.0, + 1457.0, + 1245.0, + 1469.0, + 1234.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 952.0, + 1493.0, + 965.0, + 1493.0, + 965.0, + 1504.0, + 952.0, + 1504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1334.0, + 1498.0, + 1344.0, + 1498.0, + 1344.0, + 1508.0, + 1334.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 701.0, + 1513.0, + 711.0, + 1513.0, + 711.0, + 1522.0, + 701.0, + 1522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1261.0, + 1509.0, + 1271.0, + 1509.0, + 1271.0, + 1521.0, + 1261.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 1518.0, + 763.0, + 1518.0, + 763.0, + 1529.0, + 751.0, + 1529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1161.0, + 1523.0, + 1174.0, + 1523.0, + 1174.0, + 1534.0, + 1161.0, + 1534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1545.0, + 412.0, + 1545.0, + 412.0, + 1561.0, + 397.0, + 1561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 1540.0, + 523.0, + 1540.0, + 523.0, + 1565.0, + 476.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 566.0, + 1540.0, + 622.0, + 1540.0, + 622.0, + 1565.0, + 566.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 1540.0, + 717.0, + 1540.0, + 717.0, + 1565.0, + 661.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 755.0, + 1540.0, + 811.0, + 1540.0, + 811.0, + 1565.0, + 755.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1540.0, + 906.0, + 1540.0, + 906.0, + 1565.0, + 848.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.0, + 1540.0, + 1000.0, + 1540.0, + 1000.0, + 1565.0, + 944.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 1540.0, + 1095.0, + 1540.0, + 1095.0, + 1565.0, + 1037.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 1540.0, + 1189.0, + 1540.0, + 1189.0, + 1565.0, + 1132.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1226.0, + 1540.0, + 1284.0, + 1540.0, + 1284.0, + 1565.0, + 1226.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1321.0, + 1540.0, + 1378.0, + 1540.0, + 1378.0, + 1565.0, + 1321.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 1555.0, + 971.0, + 1555.0, + 971.0, + 1588.0, + 830.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1499.5, + 348.0, + 1499.5, + 348.0, + 1523.5, + 323.0, + 1523.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1621.0, + 1403.0, + 1621.0, + 1403.0, + 1653.0, + 295.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1653.0, + 1403.0, + 1653.0, + 1403.0, + 1683.0, + 295.0, + 1683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1681.0, + 1405.0, + 1681.0, + 1405.0, + 1718.0, + 294.0, + 1718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1711.0, + 369.0, + 1711.0, + 369.0, + 1752.0, + 293.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 215.0, + 547.0, + 215.0, + 547.0, + 287.0, + 290.0, + 287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 326.0, + 853.0, + 326.0, + 853.0, + 374.0, + 294.0, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 982.0, + 1015.0, + 982.0, + 1015.0, + 1022.0, + 294.0, + 1022.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 833.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2124.0, + 833.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1051.0, + 984.0, + 1051.0, + 984.0, + 1086.0, + 295.0, + 1086.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 737.0, + 584.0, + 737.0, + 584.0, + 777.0, + 293.0, + 777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 737.0, + 1407.0, + 737.0, + 1407.0, + 777.0, + 1031.0, + 777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 771.0, + 769.0, + 851.0, + 769.0, + 851.0, + 805.0, + 771.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 769.0, + 1349.0, + 769.0, + 1349.0, + 805.0, + 873.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 769.0, + 1407.0, + 769.0, + 1407.0, + 805.0, + 1404.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 797.0, + 297.0, + 797.0, + 297.0, + 837.0, + 292.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 422.0, + 797.0, + 1408.0, + 797.0, + 1408.0, + 837.0, + 422.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 832.0, + 399.0, + 832.0, + 399.0, + 863.0, + 295.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 483.0, + 832.0, + 492.0, + 832.0, + 492.0, + 863.0, + 483.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1772.0, + 1405.0, + 1772.0, + 1405.0, + 1808.0, + 293.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1804.0, + 1406.0, + 1804.0, + 1406.0, + 1837.0, + 296.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1836.0, + 1404.0, + 1836.0, + 1404.0, + 1868.0, + 293.0, + 1868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1860.0, + 384.0, + 1860.0, + 384.0, + 1903.0, + 289.0, + 1903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1910.0, + 1407.0, + 1910.0, + 1407.0, + 1946.0, + 295.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1942.0, + 1407.0, + 1942.0, + 1407.0, + 1976.0, + 293.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1971.0, + 1404.0, + 1971.0, + 1404.0, + 2009.0, + 293.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2001.0, + 1405.0, + 2001.0, + 1405.0, + 2040.0, + 294.0, + 2040.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 524.0, + 1403.0, + 524.0, + 1403.0, + 557.0, + 295.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 555.0, + 1404.0, + 555.0, + 1404.0, + 589.0, + 295.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 585.0, + 643.0, + 585.0, + 643.0, + 620.0, + 295.0, + 620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 628.0, + 1404.0, + 628.0, + 1404.0, + 668.0, + 293.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 661.0, + 1405.0, + 661.0, + 1405.0, + 695.0, + 294.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 691.0, + 434.0, + 691.0, + 434.0, + 730.0, + 293.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 873.0, + 1406.0, + 873.0, + 1406.0, + 916.0, + 294.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 902.0, + 490.0, + 902.0, + 490.0, + 947.0, + 292.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 902.0, + 548.0, + 902.0, + 548.0, + 947.0, + 517.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 902.0, + 687.0, + 902.0, + 687.0, + 947.0, + 574.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 441.0, + 1406.0, + 441.0, + 1406.0, + 483.0, + 294.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 478.0, + 770.0, + 478.0, + 770.0, + 511.0, + 296.0, + 511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1211.0, + 1242.0, + 1211.0, + 1242.0, + 1256.0, + 292.0, + 1256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 397.0, + 1354.0, + 397.0, + 1354.0, + 436.0, + 295.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1105.0, + 1404.0, + 1105.0, + 1404.0, + 1145.0, + 293.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1136.0, + 1404.0, + 1136.0, + 1404.0, + 1177.0, + 293.0, + 1177.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1171.0, + 1141.0, + 1171.0, + 1141.0, + 1205.0, + 294.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1051.0, + 984.0, + 1051.0, + 984.0, + 1086.0, + 295.0, + 1086.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 14, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 945, + 1403, + 945, + 1403, + 1160, + 298, + 1160 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 299, + 1680, + 1404, + 1680, + 1404, + 1865, + 299, + 1865 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 299, + 1344, + 1402, + 1344, + 1402, + 1466, + 299, + 1466 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 299, + 1174, + 1403, + 1174, + 1403, + 1328, + 299, + 1328 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 300, + 1881, + 1402, + 1881, + 1402, + 2034, + 300, + 2034 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 299, + 753, + 1402, + 753, + 1402, + 846, + 299, + 846 + ], + "score": 0.961 + }, + { + "category_id": 3, + "poly": [ + 299, + 227, + 1399, + 227, + 1399, + 535, + 299, + 535 + ], + "score": 0.946 + }, + { + "category_id": 4, + "poly": [ + 297, + 569, + 1404, + 569, + 1404, + 692, + 297, + 692 + ], + "score": 0.93 + }, + { + "category_id": 0, + "poly": [ + 299, + 1514, + 840, + 1514, + 840, + 1551, + 299, + 1551 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 301, + 1587, + 1100, + 1587, + 1100, + 1619, + 301, + 1619 + ], + "score": 0.917 + }, + { + "category_id": 1, + "poly": [ + 300, + 1634, + 733, + 1634, + 733, + 1665, + 300, + 1665 + ], + "score": 0.913 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 815, + 76, + 815, + 104, + 299, + 104 + ], + "score": 0.893 + }, + { + "category_id": 0, + "poly": [ + 301, + 887, + 801, + 887, + 801, + 917, + 301, + 917 + ], + "score": 0.86 + }, + { + "category_id": 2, + "poly": [ + 837, + 2089, + 865, + 2089, + 865, + 2112, + 837, + 2112 + ], + "score": 0.84 + }, + { + "category_id": 13, + "poly": [ + 855, + 1100, + 874, + 1100, + 874, + 1126, + 855, + 1126 + ], + "score": 0.8, + "latex": "k" + }, + { + "category_id": 13, + "poly": [ + 1105, + 1884, + 1138, + 1884, + 1138, + 1910, + 1105, + 1910 + ], + "score": 0.71, + "latex": "^ { = 2 }" + }, + { + "category_id": 13, + "poly": [ + 918, + 1913, + 952, + 1913, + 952, + 1940, + 918, + 1940 + ], + "score": 0.68, + "latex": "{ \\boldsymbol { \\mathbf { \\mathit { \\varepsilon } } } } = 2 { \\boldsymbol { \\mathbf { \\mathit { \\varepsilon } } } }" + }, + { + "category_id": 13, + "poly": [ + 477, + 1913, + 509, + 1913, + 509, + 1940, + 477, + 1940 + ], + "score": 0.48, + "latex": "^ { = 2 }" + }, + { + "category_id": 13, + "poly": [ + 1258, + 605, + 1278, + 605, + 1278, + 629, + 1258, + 629 + ], + "score": 0.37, + "latex": "\\mathbf { X }" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 230.0, + 333.0, + 230.0, + 333.0, + 496.0, + 297.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1061.0, + 242.0, + 1073.0, + 242.0, + 1073.0, + 255.0, + 1061.0, + 255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 255.0, + 339.0, + 255.0, + 339.0, + 275.0, + 325.0, + 275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 794.0, + 256.0, + 807.0, + 256.0, + 807.0, + 266.0, + 794.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 852.0, + 279.0, + 862.0, + 279.0, + 862.0, + 288.0, + 852.0, + 288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 280.0, + 964.0, + 280.0, + 964.0, + 289.0, + 953.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 278.0, + 1261.0, + 278.0, + 1261.0, + 288.0, + 1252.0, + 288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1349.0, + 279.0, + 1360.0, + 279.0, + 1360.0, + 288.0, + 1349.0, + 288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 319.0, + 352.0, + 319.0, + 352.0, + 343.0, + 323.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 316.0, + 797.0, + 316.0, + 797.0, + 326.0, + 784.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 932.0, + 333.0, + 944.0, + 333.0, + 944.0, + 341.0, + 932.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.0, + 343.0, + 695.0, + 343.0, + 695.0, + 354.0, + 684.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 344.0, + 1061.0, + 344.0, + 1061.0, + 355.0, + 1050.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1150.0, + 343.0, + 1165.0, + 343.0, + 1165.0, + 358.0, + 1150.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 385.0, + 359.0, + 385.0, + 359.0, + 408.0, + 323.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 472.0, + 395.0, + 481.0, + 395.0, + 481.0, + 405.0, + 472.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 599.0, + 384.0, + 610.0, + 384.0, + 610.0, + 394.0, + 599.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 877.0, + 380.0, + 889.0, + 380.0, + 889.0, + 391.0, + 877.0, + 391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 405.0, + 767.0, + 405.0, + 767.0, + 415.0, + 757.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 966.0, + 409.0, + 978.0, + 409.0, + 978.0, + 421.0, + 966.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 450.0, + 368.0, + 450.0, + 368.0, + 473.0, + 324.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 491.0, + 414.0, + 491.0, + 414.0, + 512.0, + 395.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 488.0, + 524.0, + 488.0, + 524.0, + 514.0, + 475.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 488.0, + 623.0, + 488.0, + 623.0, + 514.0, + 564.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 488.0, + 716.0, + 488.0, + 716.0, + 514.0, + 659.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 753.0, + 488.0, + 811.0, + 488.0, + 811.0, + 514.0, + 753.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 827.0, + 487.0, + 1000.0, + 487.0, + 1000.0, + 539.0, + 827.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 488.0, + 1095.0, + 488.0, + 1095.0, + 514.0, + 1038.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 488.0, + 1190.0, + 488.0, + 1190.0, + 514.0, + 1131.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1227.0, + 488.0, + 1285.0, + 488.0, + 1285.0, + 514.0, + 1227.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1320.0, + 488.0, + 1378.0, + 488.0, + 1378.0, + 514.0, + 1320.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 570.0, + 1404.0, + 570.0, + 1404.0, + 603.0, + 295.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 599.0, + 1257.0, + 599.0, + 1257.0, + 633.0, + 294.0, + 633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1279.0, + 599.0, + 1404.0, + 599.0, + 1404.0, + 633.0, + 1279.0, + 633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 632.0, + 1404.0, + 632.0, + 1404.0, + 664.0, + 295.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 299.0, + 664.0, + 844.0, + 664.0, + 844.0, + 693.0, + 299.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1511.0, + 843.0, + 1511.0, + 843.0, + 1558.0, + 294.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 886.0, + 807.0, + 886.0, + 807.0, + 921.0, + 294.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2125.0, + 831.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 945.0, + 1403.0, + 945.0, + 1403.0, + 984.0, + 293.0, + 984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 975.0, + 1404.0, + 975.0, + 1404.0, + 1013.0, + 293.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1007.0, + 1405.0, + 1007.0, + 1405.0, + 1041.0, + 294.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1039.0, + 1405.0, + 1039.0, + 1405.0, + 1071.0, + 293.0, + 1071.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1067.0, + 1405.0, + 1067.0, + 1405.0, + 1102.0, + 293.0, + 1102.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1099.0, + 854.0, + 1099.0, + 854.0, + 1134.0, + 294.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1099.0, + 1405.0, + 1099.0, + 1405.0, + 1134.0, + 875.0, + 1134.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1128.0, + 697.0, + 1128.0, + 697.0, + 1166.0, + 293.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1682.0, + 1403.0, + 1682.0, + 1403.0, + 1714.0, + 297.0, + 1714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1711.0, + 1405.0, + 1711.0, + 1405.0, + 1747.0, + 294.0, + 1747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1741.0, + 1404.0, + 1741.0, + 1404.0, + 1778.0, + 293.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1772.0, + 1403.0, + 1772.0, + 1403.0, + 1808.0, + 294.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1804.0, + 1404.0, + 1804.0, + 1404.0, + 1836.0, + 295.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1835.0, + 654.0, + 1835.0, + 654.0, + 1865.0, + 294.0, + 1865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1344.0, + 1403.0, + 1344.0, + 1403.0, + 1380.0, + 294.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1374.0, + 1406.0, + 1374.0, + 1406.0, + 1410.0, + 294.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1407.0, + 1403.0, + 1407.0, + 1403.0, + 1440.0, + 295.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1435.0, + 740.0, + 1435.0, + 740.0, + 1470.0, + 294.0, + 1470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1176.0, + 1404.0, + 1176.0, + 1404.0, + 1209.0, + 298.0, + 1209.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1202.0, + 1407.0, + 1202.0, + 1407.0, + 1246.0, + 293.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1238.0, + 1404.0, + 1238.0, + 1404.0, + 1271.0, + 295.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1264.0, + 1405.0, + 1264.0, + 1405.0, + 1304.0, + 293.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1296.0, + 416.0, + 1296.0, + 416.0, + 1332.0, + 292.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1881.0, + 1104.0, + 1881.0, + 1104.0, + 1914.0, + 296.0, + 1914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1139.0, + 1881.0, + 1403.0, + 1881.0, + 1403.0, + 1914.0, + 1139.0, + 1914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1911.0, + 476.0, + 1911.0, + 476.0, + 1944.0, + 296.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 1911.0, + 917.0, + 1911.0, + 917.0, + 1944.0, + 510.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 1911.0, + 1404.0, + 1911.0, + 1404.0, + 1944.0, + 953.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1940.0, + 1406.0, + 1940.0, + 1406.0, + 1977.0, + 296.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1971.0, + 1406.0, + 1971.0, + 1406.0, + 2009.0, + 294.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2002.0, + 642.0, + 2002.0, + 642.0, + 2038.0, + 293.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 754.0, + 1404.0, + 754.0, + 1404.0, + 787.0, + 295.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 784.0, + 1404.0, + 784.0, + 1404.0, + 818.0, + 295.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 810.0, + 727.0, + 810.0, + 727.0, + 853.0, + 294.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1584.0, + 1106.0, + 1584.0, + 1106.0, + 1623.0, + 296.0, + 1623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1629.0, + 736.0, + 1629.0, + 736.0, + 1671.0, + 295.0, + 1671.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 15, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1403, + 1404, + 1403, + 1404, + 1530, + 298, + 1530 + ], + "score": 0.974 + }, + { + "category_id": 5, + "poly": [ + 298, + 1792, + 923, + 1792, + 923, + 2046, + 298, + 2046 + ], + "score": 0.972, + "html": "
number outputsstride
Input: z ~ N(0,I256)Fully connectedReshape to image 4,4,512ConvolutionConvolutionConvolutionConvolution4*4*512256128641or32221
" + }, + { + "category_id": 4, + "poly": [ + 295, + 1210, + 1404, + 1210, + 1404, + 1334, + 295, + 1334 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 300, + 1667, + 1401, + 1667, + 1401, + 1731, + 300, + 1731 + ], + "score": 0.949 + }, + { + "category_id": 0, + "poly": [ + 299, + 1588, + 909, + 1588, + 909, + 1626, + 299, + 1626 + ], + "score": 0.925 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.893 + }, + { + "category_id": 2, + "poly": [ + 835, + 2087, + 865, + 2087, + 865, + 2113, + 835, + 2113 + ], + "score": 0.823 + }, + { + "category_id": 1, + "poly": [ + 300, + 1746, + 705, + 1746, + 705, + 1778, + 300, + 1778 + ], + "score": 0.794 + }, + { + "category_id": 3, + "poly": [ + 289, + 215, + 1404, + 215, + 1404, + 1179, + 289, + 1179 + ], + "score": 0.56 + }, + { + "category_id": 6, + "poly": [ + 300, + 1746, + 705, + 1746, + 705, + 1778, + 300, + 1778 + ], + "score": 0.141 + }, + { + "category_id": 13, + "poly": [ + 689, + 1405, + 938, + 1405, + 938, + 1439, + 689, + 1439 + ], + "score": 0.92, + "latex": "\\mathcal { L } _ { G } = \\log ( D ( G ( z ) ) )" + }, + { + "category_id": 13, + "poly": [ + 297, + 1436, + 660, + 1436, + 660, + 1470, + 297, + 1470 + ], + "score": 0.91, + "latex": "\\log ( D ( x ) ) + \\log ( 1 - D ( G ( z ) ) )" + }, + { + "category_id": 13, + "poly": [ + 389, + 1824, + 563, + 1824, + 563, + 1858, + 389, + 1858 + ], + "score": 0.89, + "latex": "z \\sim \\mathcal { N } ( 0 , I _ { 2 5 6 } )" + }, + { + "category_id": 13, + "poly": [ + 587, + 1469, + 668, + 1469, + 668, + 1498, + 587, + 1498 + ], + "score": 0.89, + "latex": "\\beta _ { 1 } { = } 0 . 5" + }, + { + "category_id": 13, + "poly": [ + 1323, + 1406, + 1403, + 1406, + 1403, + 1437, + 1323, + 1437 + ], + "score": 0.87, + "latex": "\\mathcal { L } _ { D } =" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1210.0, + 1405.0, + 1210.0, + 1405.0, + 1246.0, + 294.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1240.0, + 1404.0, + 1240.0, + 1404.0, + 1276.0, + 292.0, + 1276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1273.0, + 1404.0, + 1273.0, + 1404.0, + 1309.0, + 294.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1306.0, + 448.0, + 1306.0, + 448.0, + 1336.0, + 294.0, + 1336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1585.0, + 913.0, + 1585.0, + 913.0, + 1633.0, + 292.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2125.0, + 832.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 217.0, + 939.0, + 217.0, + 939.0, + 255.0, + 763.0, + 255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 279.0, + 343.0, + 279.0, + 343.0, + 298.0, + 328.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 345.0, + 342.0, + 345.0, + 342.0, + 363.0, + 326.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 357.0, + 336.0, + 357.0, + 336.0, + 439.0, + 297.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 409.0, + 344.0, + 409.0, + 344.0, + 431.0, + 325.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 428.0, + 333.0, + 428.0, + 333.0, + 468.0, + 304.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 470.0, + 331.0, + 470.0, + 331.0, + 526.0, + 304.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 477.0, + 342.0, + 477.0, + 342.0, + 496.0, + 328.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 541.0, + 357.0, + 541.0, + 357.0, + 565.0, + 325.0, + 565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 606.0, + 357.0, + 606.0, + 357.0, + 632.0, + 325.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 647.0, + 416.0, + 647.0, + 416.0, + 669.0, + 397.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 474.0, + 643.0, + 528.0, + 643.0, + 528.0, + 670.0, + 474.0, + 670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 642.0, + 629.0, + 642.0, + 629.0, + 671.0, + 564.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 645.0, + 721.0, + 645.0, + 721.0, + 670.0, + 659.0, + 670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 752.0, + 645.0, + 815.0, + 645.0, + 815.0, + 670.0, + 752.0, + 670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 818.0, + 642.0, + 1003.0, + 642.0, + 1003.0, + 696.0, + 818.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1035.0, + 645.0, + 1097.0, + 645.0, + 1097.0, + 670.0, + 1035.0, + 670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 645.0, + 1192.0, + 645.0, + 1192.0, + 670.0, + 1129.0, + 670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.0, + 645.0, + 1286.0, + 645.0, + 1286.0, + 670.0, + 1224.0, + 670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 645.0, + 1381.0, + 645.0, + 1381.0, + 670.0, + 1317.0, + 670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 625.0, + 694.0, + 1076.0, + 694.0, + 1076.0, + 733.0, + 625.0, + 733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 751.0, + 610.0, + 751.0, + 610.0, + 766.0, + 595.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1163.0, + 753.0, + 1173.0, + 753.0, + 1173.0, + 762.0, + 1163.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1325.0, + 757.0, + 1335.0, + 757.0, + 1335.0, + 767.0, + 1325.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 765.0, + 343.0, + 765.0, + 343.0, + 781.0, + 328.0, + 781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 831.0, + 342.0, + 831.0, + 342.0, + 848.0, + 328.0, + 848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 839.0, + 336.0, + 839.0, + 336.0, + 922.0, + 297.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 893.0, + 344.0, + 893.0, + 344.0, + 916.0, + 325.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 909.0, + 334.0, + 909.0, + 334.0, + 960.0, + 302.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 947.0, + 333.0, + 947.0, + 333.0, + 1013.0, + 303.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 958.0, + 344.0, + 958.0, + 344.0, + 982.0, + 325.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1024.0, + 357.0, + 1024.0, + 357.0, + 1050.0, + 325.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1090.0, + 355.0, + 1090.0, + 355.0, + 1114.0, + 325.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 1133.0, + 415.0, + 1133.0, + 415.0, + 1150.0, + 398.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1132.0, + 524.0, + 1132.0, + 524.0, + 1150.0, + 478.0, + 1150.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 566.0, + 1128.0, + 626.0, + 1128.0, + 626.0, + 1154.0, + 566.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 1128.0, + 720.0, + 1128.0, + 720.0, + 1154.0, + 658.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 750.0, + 1125.0, + 815.0, + 1125.0, + 815.0, + 1155.0, + 750.0, + 1155.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 1128.0, + 1002.0, + 1128.0, + 1002.0, + 1181.0, + 819.0, + 1181.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 1128.0, + 1097.0, + 1128.0, + 1097.0, + 1154.0, + 1036.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 1128.0, + 1191.0, + 1128.0, + 1191.0, + 1154.0, + 1130.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1225.0, + 1129.0, + 1283.0, + 1129.0, + 1283.0, + 1151.0, + 1225.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1319.0, + 1128.0, + 1379.0, + 1128.0, + 1379.0, + 1154.0, + 1319.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1745.0, + 708.0, + 1745.0, + 708.0, + 1778.0, + 295.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1404.0, + 688.0, + 1404.0, + 688.0, + 1440.0, + 293.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 1404.0, + 1322.0, + 1404.0, + 1322.0, + 1440.0, + 939.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1435.0, + 296.0, + 1435.0, + 296.0, + 1471.0, + 293.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 1435.0, + 1405.0, + 1435.0, + 1405.0, + 1471.0, + 661.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1468.0, + 586.0, + 1468.0, + 586.0, + 1501.0, + 294.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 669.0, + 1468.0, + 1404.0, + 1468.0, + 1404.0, + 1501.0, + 669.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1496.0, + 1105.0, + 1496.0, + 1105.0, + 1532.0, + 295.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1667.0, + 1403.0, + 1667.0, + 1403.0, + 1703.0, + 296.0, + 1703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1699.0, + 1303.0, + 1699.0, + 1303.0, + 1731.0, + 296.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1745.0, + 708.0, + 1745.0, + 708.0, + 1778.0, + 295.0, + 1778.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 16, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 298, + 1678, + 905, + 1678, + 905, + 1898, + 298, + 1898 + ], + "score": 0.976, + "html": "
number outputsstride
Input: x~ Pdata or G Transposed Convolution Transposed Convolution Transposed Convolution64 128 256222
Flatten
Fully Connected1
" + }, + { + "category_id": 1, + "poly": [ + 297, + 1910, + 1405, + 1910, + 1405, + 2036, + 297, + 2036 + ], + "score": 0.949 + }, + { + "category_id": 4, + "poly": [ + 295, + 1216, + 1405, + 1216, + 1405, + 1342, + 295, + 1342 + ], + "score": 0.946 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 815, + 75, + 815, + 104, + 299, + 104 + ], + "score": 0.896 + }, + { + "category_id": 2, + "poly": [ + 836, + 2087, + 865, + 2087, + 865, + 2113, + 836, + 2113 + ], + "score": 0.865 + }, + { + "category_id": 4, + "poly": [ + 504, + 705, + 1194, + 705, + 1194, + 738, + 504, + 738 + ], + "score": 0.801 + }, + { + "category_id": 1, + "poly": [ + 299, + 1637, + 748, + 1637, + 748, + 1668, + 299, + 1668 + ], + "score": 0.783 + }, + { + "category_id": 3, + "poly": [ + 297, + 746, + 1399, + 746, + 1399, + 1182, + 297, + 1182 + ], + "score": 0.775 + }, + { + "category_id": 3, + "poly": [ + 295, + 261, + 1399, + 261, + 1399, + 698, + 295, + 698 + ], + "score": 0.736 + }, + { + "category_id": 6, + "poly": [ + 643, + 221, + 1055, + 221, + 1055, + 254, + 643, + 254 + ], + "score": 0.267 + }, + { + "category_id": 6, + "poly": [ + 299, + 1637, + 748, + 1637, + 748, + 1668, + 299, + 1668 + ], + "score": 0.17 + }, + { + "category_id": 13, + "poly": [ + 689, + 1910, + 938, + 1910, + 938, + 1945, + 689, + 1945 + ], + "score": 0.92, + "latex": "\\mathcal { L } _ { G } = \\log ( D ( G ( z ) ) )" + }, + { + "category_id": 13, + "poly": [ + 297, + 1942, + 651, + 1942, + 651, + 1976, + 297, + 1976 + ], + "score": 0.9, + "latex": "\\log ( D ( x ) ) + \\log ( 1 - D ( G ( z ) ) )" + }, + { + "category_id": 13, + "poly": [ + 1323, + 1912, + 1403, + 1912, + 1403, + 1943, + 1323, + 1943 + ], + "score": 0.86, + "latex": "\\mathcal { L } _ { D } =" + }, + { + "category_id": 13, + "poly": [ + 388, + 1717, + 503, + 1717, + 503, + 1742, + 388, + 1742 + ], + "score": 0.74, + "latex": "x \\sim p _ { d a t a }" + }, + { + "category_id": 13, + "poly": [ + 692, + 2004, + 750, + 2004, + 750, + 2032, + 692, + 2032 + ], + "score": 0.63, + "latex": "1 0 0 \\mathrm { k }" + }, + { + "category_id": 13, + "poly": [ + 538, + 1713, + 563, + 1713, + 563, + 1739, + 538, + 1739 + ], + "score": 0.48, + "latex": "\\overline { G }" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1217.0, + 1405.0, + 1217.0, + 1405.0, + 1253.0, + 294.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1247.0, + 1406.0, + 1247.0, + 1406.0, + 1284.0, + 294.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1275.0, + 1405.0, + 1275.0, + 1405.0, + 1316.0, + 293.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1308.0, + 1212.0, + 1308.0, + 1212.0, + 1345.0, + 295.0, + 1345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 72.0, + 817.0, + 72.0, + 817.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2123.0, + 832.0, + 2123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 506.0, + 701.0, + 1194.0, + 701.0, + 1194.0, + 742.0, + 506.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 763.0, + 510.0, + 763.0, + 510.0, + 773.0, + 500.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 565.0, + 762.0, + 578.0, + 762.0, + 578.0, + 772.0, + 565.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 767.0, + 345.0, + 767.0, + 345.0, + 789.0, + 326.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 675.0, + 784.0, + 690.0, + 784.0, + 690.0, + 799.0, + 675.0, + 799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1247.0, + 787.0, + 1259.0, + 787.0, + 1259.0, + 798.0, + 1247.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 833.0, + 345.0, + 833.0, + 345.0, + 856.0, + 326.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 836.0, + 605.0, + 836.0, + 605.0, + 847.0, + 595.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 830.0, + 882.0, + 830.0, + 882.0, + 839.0, + 872.0, + 839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 845.0, + 337.0, + 845.0, + 337.0, + 1026.0, + 297.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 400.0, + 844.0, + 410.0, + 844.0, + 410.0, + 854.0, + 400.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1141.0, + 848.0, + 1151.0, + 848.0, + 1151.0, + 858.0, + 1141.0, + 858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 854.0, + 485.0, + 854.0, + 485.0, + 866.0, + 475.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 852.0, + 691.0, + 852.0, + 691.0, + 864.0, + 680.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 888.0, + 1066.0, + 888.0, + 1066.0, + 900.0, + 1056.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 894.0, + 989.0, + 894.0, + 989.0, + 905.0, + 978.0, + 905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 900.0, + 342.0, + 900.0, + 342.0, + 920.0, + 327.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 900.0, + 705.0, + 900.0, + 705.0, + 910.0, + 695.0, + 910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 861.0, + 900.0, + 871.0, + 900.0, + 871.0, + 909.0, + 861.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1143.0, + 903.0, + 1154.0, + 903.0, + 1154.0, + 912.0, + 1143.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 912.0, + 405.0, + 912.0, + 405.0, + 923.0, + 392.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 500.0, + 911.0, + 511.0, + 911.0, + 511.0, + 921.0, + 500.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 965.0, + 342.0, + 965.0, + 342.0, + 987.0, + 327.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 979.0, + 961.0, + 991.0, + 961.0, + 991.0, + 970.0, + 979.0, + 970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 961.0, + 1065.0, + 961.0, + 1065.0, + 973.0, + 1053.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1138.0, + 967.0, + 1150.0, + 967.0, + 1150.0, + 977.0, + 1138.0, + 977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 665.0, + 974.0, + 675.0, + 974.0, + 675.0, + 983.0, + 665.0, + 983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 600.0, + 990.0, + 611.0, + 990.0, + 611.0, + 999.0, + 600.0, + 999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 988.0, + 861.0, + 988.0, + 861.0, + 999.0, + 851.0, + 999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 326.0, + 1030.0, + 356.0, + 1030.0, + 356.0, + 1055.0, + 326.0, + 1055.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1095.0, + 356.0, + 1095.0, + 356.0, + 1121.0, + 325.0, + 1121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 790.0, + 1094.0, + 800.0, + 1094.0, + 800.0, + 1103.0, + 790.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1135.0, + 416.0, + 1135.0, + 416.0, + 1157.0, + 397.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 1133.0, + 526.0, + 1133.0, + 526.0, + 1160.0, + 475.0, + 1160.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 1132.0, + 626.0, + 1132.0, + 626.0, + 1161.0, + 563.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 656.0, + 1132.0, + 722.0, + 1132.0, + 722.0, + 1161.0, + 656.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 750.0, + 1132.0, + 815.0, + 1132.0, + 815.0, + 1161.0, + 750.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 1132.0, + 911.0, + 1132.0, + 911.0, + 1161.0, + 845.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 1132.0, + 1005.0, + 1132.0, + 1005.0, + 1161.0, + 939.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 1132.0, + 1098.0, + 1132.0, + 1098.0, + 1161.0, + 1033.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1127.0, + 1132.0, + 1192.0, + 1132.0, + 1192.0, + 1161.0, + 1127.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1221.0, + 1132.0, + 1286.0, + 1132.0, + 1286.0, + 1161.0, + 1221.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1316.0, + 1132.0, + 1380.0, + 1132.0, + 1380.0, + 1161.0, + 1316.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 821.0, + 1144.0, + 981.0, + 1144.0, + 981.0, + 1185.0, + 821.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 283.0, + 345.0, + 283.0, + 345.0, + 307.0, + 325.0, + 307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 287.0, + 882.0, + 287.0, + 882.0, + 297.0, + 872.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 977.0, + 301.0, + 987.0, + 301.0, + 987.0, + 312.0, + 977.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 339.0, + 759.0, + 339.0, + 759.0, + 348.0, + 749.0, + 348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 340.0, + 990.0, + 340.0, + 990.0, + 350.0, + 980.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 349.0, + 344.0, + 349.0, + 344.0, + 371.0, + 325.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 348.0, + 689.0, + 348.0, + 689.0, + 357.0, + 679.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 361.0, + 339.0, + 361.0, + 339.0, + 544.0, + 295.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 415.0, + 342.0, + 415.0, + 342.0, + 437.0, + 327.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 857.0, + 430.0, + 870.0, + 430.0, + 870.0, + 442.0, + 857.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 430.0, + 990.0, + 430.0, + 990.0, + 441.0, + 980.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 482.0, + 342.0, + 482.0, + 342.0, + 503.0, + 327.0, + 503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 544.0, + 357.0, + 544.0, + 357.0, + 574.0, + 324.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 610.0, + 358.0, + 610.0, + 358.0, + 638.0, + 323.0, + 638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 652.0, + 417.0, + 652.0, + 417.0, + 673.0, + 397.0, + 673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 650.0, + 527.0, + 650.0, + 527.0, + 676.0, + 475.0, + 676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 564.0, + 650.0, + 626.0, + 650.0, + 626.0, + 676.0, + 564.0, + 676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 656.0, + 649.0, + 722.0, + 649.0, + 722.0, + 678.0, + 656.0, + 678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 750.0, + 647.0, + 816.0, + 647.0, + 816.0, + 678.0, + 750.0, + 678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 844.0, + 647.0, + 911.0, + 647.0, + 911.0, + 677.0, + 844.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 647.0, + 1004.0, + 647.0, + 1004.0, + 677.0, + 939.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 647.0, + 1099.0, + 647.0, + 1099.0, + 677.0, + 1034.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1127.0, + 647.0, + 1193.0, + 647.0, + 1193.0, + 677.0, + 1127.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1221.0, + 647.0, + 1287.0, + 647.0, + 1287.0, + 677.0, + 1221.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1316.0, + 647.0, + 1380.0, + 647.0, + 1380.0, + 677.0, + 1316.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 661.0, + 981.0, + 661.0, + 981.0, + 702.0, + 820.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 643.0, + 217.0, + 1056.0, + 217.0, + 1056.0, + 258.0, + 643.0, + 258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1633.0, + 750.0, + 1633.0, + 750.0, + 1670.0, + 295.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1908.0, + 688.0, + 1908.0, + 688.0, + 1947.0, + 292.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 1908.0, + 1322.0, + 1908.0, + 1322.0, + 1947.0, + 939.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 1941.0, + 1406.0, + 1941.0, + 1406.0, + 1978.0, + 652.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1971.0, + 1405.0, + 1971.0, + 1405.0, + 2008.0, + 294.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 2002.0, + 691.0, + 2002.0, + 691.0, + 2038.0, + 292.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 2002.0, + 1304.0, + 2002.0, + 1304.0, + 2038.0, + 751.0, + 2038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1633.0, + 750.0, + 1633.0, + 750.0, + 1670.0, + 295.0, + 1670.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 17, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1053, + 1405, + 1053, + 1405, + 1268, + 297, + 1268 + ], + "score": 0.978 + }, + { + "category_id": 5, + "poly": [ + 298, + 635, + 905, + 635, + 905, + 857, + 298, + 857 + ], + "score": 0.978, + "html": "
number outputsstride
Input: x ~ Pdata or G Transposed Convolution Transposed Convolution8*X 16*X222
Transposed Convolution Flatten32*X
Fully Connected1
" + }, + { + "category_id": 3, + "poly": [ + 345, + 1308, + 1353, + 1308, + 1353, + 1643, + 345, + 1643 + ], + "score": 0.975 + }, + { + "category_id": 5, + "poly": [ + 297, + 301, + 908, + 301, + 908, + 552, + 297, + 552 + ], + "score": 0.971, + "html": "
number outputsstride
Input: z ~ N(0,I256)Fully connectedReshape to image 4,4,64ConvolutionConvolutionConvolutionConvolution4 *4*643216832221
" + }, + { + "category_id": 1, + "poly": [ + 301, + 1924, + 1404, + 1924, + 1404, + 1988, + 301, + 1988 + ], + "score": 0.944 + }, + { + "category_id": 4, + "poly": [ + 297, + 1674, + 1404, + 1674, + 1404, + 1768, + 297, + 1768 + ], + "score": 0.915 + }, + { + "category_id": 0, + "poly": [ + 299, + 908, + 749, + 908, + 749, + 946, + 299, + 946 + ], + "score": 0.913 + }, + { + "category_id": 1, + "poly": [ + 292, + 562, + 1406, + 562, + 1406, + 625, + 292, + 625 + ], + "score": 0.898 + }, + { + "category_id": 1, + "poly": [ + 300, + 2002, + 705, + 2002, + 705, + 2035, + 300, + 2035 + ], + "score": 0.897 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 815, + 75, + 815, + 105, + 299, + 105 + ], + "score": 0.894 + }, + { + "category_id": 0, + "poly": [ + 299, + 1860, + 471, + 1860, + 471, + 1892, + 299, + 1892 + ], + "score": 0.892 + }, + { + "category_id": 0, + "poly": [ + 298, + 989, + 476, + 989, + 476, + 1022, + 298, + 1022 + ], + "score": 0.883 + }, + { + "category_id": 2, + "poly": [ + 836, + 2088, + 865, + 2088, + 865, + 2112, + 836, + 2112 + ], + "score": 0.851 + }, + { + "category_id": 0, + "poly": [ + 298, + 225, + 649, + 225, + 649, + 263, + 298, + 263 + ], + "score": 0.588 + }, + { + "category_id": 6, + "poly": [ + 298, + 225, + 649, + 225, + 649, + 263, + 298, + 263 + ], + "score": 0.253 + }, + { + "category_id": 13, + "poly": [ + 549, + 1115, + 733, + 1115, + 733, + 1149, + 549, + 1149 + ], + "score": 0.93, + "latex": "I * C + \\mathsf { \\bar { ( } } I - 1 )" + }, + { + "category_id": 13, + "poly": [ + 389, + 332, + 563, + 332, + 563, + 367, + 389, + 367 + ], + "score": 0.88, + "latex": "z \\sim \\mathcal { N } ( 0 , I _ { 2 5 6 } )" + }, + { + "category_id": 13, + "poly": [ + 389, + 673, + 504, + 673, + 504, + 699, + 389, + 699 + ], + "score": 0.8, + "latex": "x \\sim p _ { d a t a }" + }, + { + "category_id": 13, + "poly": [ + 517, + 1086, + 542, + 1086, + 542, + 1114, + 517, + 1114 + ], + "score": 0.78, + "latex": "C" + }, + { + "category_id": 13, + "poly": [ + 297, + 1956, + 350, + 1956, + 350, + 1984, + 297, + 1984 + ], + "score": 0.76, + "latex": "{ \\bf X } { = } 1" + }, + { + "category_id": 13, + "poly": [ + 904, + 1056, + 922, + 1056, + 922, + 1083, + 904, + 1083 + ], + "score": 0.67, + "latex": "I" + }, + { + "category_id": 13, + "poly": [ + 907, + 1087, + 941, + 1087, + 941, + 1114, + 907, + 1114 + ], + "score": 0.58, + "latex": "\\scriptstyle = 0" + }, + { + "category_id": 13, + "poly": [ + 538, + 670, + 563, + 670, + 563, + 696, + 538, + 696 + ], + "score": 0.52, + "latex": "\\overline { G }" + }, + { + "category_id": 13, + "poly": [ + 781, + 1147, + 833, + 1147, + 833, + 1175, + 781, + 1175 + ], + "score": 0.49, + "latex": "= 0 . 2" + }, + { + "category_id": 13, + "poly": [ + 1016, + 1927, + 1042, + 1927, + 1042, + 1953, + 1016, + 1953 + ], + "score": 0.34, + "latex": "\\mathrm { X }" + }, + { + "category_id": 13, + "poly": [ + 892, + 564, + 918, + 564, + 918, + 592, + 892, + 592 + ], + "score": 0.3, + "latex": "\\mathrm { X }" + }, + { + "category_id": 13, + "poly": [ + 1025, + 1086, + 1079, + 1086, + 1079, + 1115, + 1025, + 1115 + ], + "score": 0.28, + "latex": "\\scriptstyle \\mathtt { = 0 . 5 }" + }, + { + "category_id": 15, + "poly": [ + 600.0, + 1258.0, + 1402.0, + 1258.0, + 1402.0, + 1692.0, + 600.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 717.0, + 1301.0, + 792.0, + 1301.0, + 792.0, + 1625.0, + 717.0, + 1625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 582.0, + 1306.0, + 649.0, + 1306.0, + 649.0, + 1649.0, + 582.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 630.0, + 1308.0, + 689.0, + 1308.0, + 689.0, + 1647.0, + 630.0, + 1647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 677.0, + 1308.0, + 736.0, + 1308.0, + 736.0, + 1647.0, + 677.0, + 1647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 818.0, + 1314.0, + 846.0, + 1314.0, + 846.0, + 1355.0, + 818.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 1318.0, + 387.0, + 1318.0, + 387.0, + 1355.0, + 356.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1318.0, + 424.0, + 1318.0, + 424.0, + 1355.0, + 393.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 1319.0, + 551.0, + 1319.0, + 551.0, + 1349.0, + 522.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1230.0, + 1319.0, + 1263.0, + 1319.0, + 1263.0, + 1354.0, + 1230.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 1320.0, + 506.0, + 1320.0, + 506.0, + 1346.0, + 482.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1320.0, + 1320.0, + 1342.0, + 1320.0, + 1342.0, + 1349.0, + 1320.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 1323.0, + 1053.0, + 1323.0, + 1053.0, + 1347.0, + 1028.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 1325.0, + 543.0, + 1325.0, + 543.0, + 1628.0, + 452.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1325.0, + 587.0, + 1325.0, + 587.0, + 1620.0, + 503.0, + 1620.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 1326.0, + 1094.0, + 1326.0, + 1094.0, + 1345.0, + 1075.0, + 1345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1334.0, + 612.0, + 1334.0, + 612.0, + 1617.0, + 538.0, + 1617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 1336.0, + 417.0, + 1336.0, + 417.0, + 1652.0, + 339.0, + 1652.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1341.0, + 439.0, + 1341.0, + 439.0, + 1614.0, + 383.0, + 1614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 422.0, + 1341.0, + 481.0, + 1341.0, + 481.0, + 1618.0, + 422.0, + 1618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 571.0, + 1351.0, + 579.0, + 1351.0, + 579.0, + 1360.0, + 571.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 1372.0, + 1196.0, + 1372.0, + 1196.0, + 1385.0, + 1188.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1319.0, + 1358.0, + 1341.0, + 1358.0, + 1341.0, + 1389.0, + 1319.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 1384.0, + 858.0, + 1384.0, + 858.0, + 1392.0, + 847.0, + 1392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.0, + 1404.0, + 1151.0, + 1404.0, + 1151.0, + 1419.0, + 1142.0, + 1419.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 802.0, + 1421.0, + 822.0, + 1421.0, + 822.0, + 1449.0, + 802.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1187.0, + 1434.0, + 1215.0, + 1434.0, + 1215.0, + 1492.0, + 1187.0, + 1492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1285.0, + 1582.0, + 1297.0, + 1582.0, + 1297.0, + 1591.0, + 1285.0, + 1591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 1606.0, + 428.0, + 1606.0, + 428.0, + 1635.0, + 399.0, + 1635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 468.0, + 1599.0, + 514.0, + 1599.0, + 514.0, + 1636.0, + 468.0, + 1636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 1598.0, + 554.0, + 1598.0, + 554.0, + 1637.0, + 518.0, + 1637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 1598.0, + 601.0, + 1598.0, + 601.0, + 1639.0, + 556.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 1597.0, + 762.0, + 1597.0, + 762.0, + 1637.0, + 714.0, + 1637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 1618.0, + 796.0, + 1618.0, + 796.0, + 1628.0, + 786.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 820.0, + 1605.0, + 841.0, + 1605.0, + 841.0, + 1632.0, + 820.0, + 1632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 1601.0, + 884.0, + 1601.0, + 884.0, + 1634.0, + 856.0, + 1634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1113.0, + 1600.0, + 1140.0, + 1600.0, + 1140.0, + 1633.0, + 1113.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1159.0, + 1613.0, + 1173.0, + 1613.0, + 1173.0, + 1626.0, + 1159.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 1606.0, + 1215.0, + 1606.0, + 1215.0, + 1627.0, + 1195.0, + 1627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1278.0, + 1613.0, + 1296.0, + 1613.0, + 1296.0, + 1625.0, + 1278.0, + 1625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.25, + 1424.5, + 1129.25, + 1424.5, + 1129.25, + 1444.5, + 1089.25, + 1444.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 1594.0, + 462.0, + 1594.0, + 462.0, + 1638.5, + 443.0, + 1638.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1673.0, + 1405.0, + 1673.0, + 1405.0, + 1708.0, + 293.0, + 1708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1702.0, + 1405.0, + 1702.0, + 1405.0, + 1742.0, + 291.0, + 1742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1736.0, + 486.0, + 1736.0, + 486.0, + 1771.0, + 296.0, + 1771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 907.0, + 755.0, + 907.0, + 755.0, + 951.0, + 292.0, + 951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1856.0, + 475.0, + 1856.0, + 475.0, + 1898.0, + 292.0, + 1898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 988.0, + 480.0, + 988.0, + 480.0, + 1024.0, + 294.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2085.0, + 870.0, + 2085.0, + 870.0, + 2124.0, + 832.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 222.0, + 651.0, + 222.0, + 651.0, + 267.0, + 293.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 222.0, + 651.0, + 222.0, + 651.0, + 267.0, + 293.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1050.0, + 903.0, + 1050.0, + 903.0, + 1092.0, + 294.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 923.0, + 1050.0, + 1405.0, + 1050.0, + 1405.0, + 1092.0, + 923.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1085.0, + 516.0, + 1085.0, + 516.0, + 1120.0, + 295.0, + 1120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 1085.0, + 906.0, + 1085.0, + 906.0, + 1120.0, + 543.0, + 1120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 1085.0, + 1024.0, + 1085.0, + 1024.0, + 1120.0, + 942.0, + 1120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1080.0, + 1085.0, + 1405.0, + 1085.0, + 1405.0, + 1120.0, + 1080.0, + 1120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1114.0, + 548.0, + 1114.0, + 548.0, + 1151.0, + 294.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 734.0, + 1114.0, + 1405.0, + 1114.0, + 1405.0, + 1151.0, + 734.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1146.0, + 780.0, + 1146.0, + 780.0, + 1179.0, + 295.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 834.0, + 1146.0, + 1402.0, + 1146.0, + 1402.0, + 1179.0, + 834.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1174.0, + 1403.0, + 1174.0, + 1403.0, + 1211.0, + 294.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1204.0, + 1405.0, + 1204.0, + 1405.0, + 1242.0, + 292.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1239.0, + 452.0, + 1239.0, + 452.0, + 1268.0, + 295.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1924.0, + 1015.0, + 1924.0, + 1015.0, + 1960.0, + 297.0, + 1960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1924.0, + 1406.0, + 1924.0, + 1406.0, + 1960.0, + 1043.0, + 1960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1952.0, + 1379.0, + 1952.0, + 1379.0, + 1991.0, + 351.0, + 1991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 558.0, + 891.0, + 558.0, + 891.0, + 600.0, + 293.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 558.0, + 1407.0, + 558.0, + 1407.0, + 600.0, + 919.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 595.0, + 549.0, + 595.0, + 549.0, + 625.0, + 294.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 2001.0, + 707.0, + 2001.0, + 707.0, + 2037.0, + 296.0, + 2037.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 18, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 298, + 224, + 956, + 224, + 956, + 475, + 298, + 475 + ], + "score": 0.977, + "html": "
number outputsstride
Input: z~ N(0,I256)Fully connectedReshape to image 4,4,512*XConvolutionConvolutionConvolutionConvolution4*4*512*X256*X128*X64*X32221
" + }, + { + "category_id": 1, + "poly": [ + 297, + 854, + 1405, + 854, + 1405, + 977, + 297, + 977 + ], + "score": 0.973 + }, + { + "category_id": 5, + "poly": [ + 298, + 529, + 907, + 529, + 907, + 750, + 298, + 750 + ], + "score": 0.964, + "html": "
number outputsstride
Input: x ~ Pdata or G Transposed Convolution Transposed Convolution Transposed Convolution Flatten Fully Connected64*X 128*X 256*X222
" + }, + { + "category_id": 0, + "poly": [ + 298, + 784, + 931, + 784, + 931, + 820, + 298, + 820 + ], + "score": 0.919 + }, + { + "category_id": 2, + "poly": [ + 299, + 75, + 814, + 75, + 814, + 105, + 299, + 105 + ], + "score": 0.886 + }, + { + "category_id": 1, + "poly": [ + 302, + 487, + 748, + 487, + 748, + 518, + 302, + 518 + ], + "score": 0.863 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 866, + 2088, + 866, + 2113, + 835, + 2113 + ], + "score": 0.849 + }, + { + "category_id": 13, + "poly": [ + 389, + 256, + 563, + 256, + 563, + 290, + 389, + 290 + ], + "score": 0.84, + "latex": "z \\sim \\mathcal { N } ( 0 , I _ { 2 5 6 } )" + }, + { + "category_id": 13, + "poly": [ + 388, + 567, + 503, + 567, + 503, + 593, + 388, + 593 + ], + "score": 0.73, + "latex": "x \\sim p _ { d a t a }" + }, + { + "category_id": 13, + "poly": [ + 538, + 562, + 562, + 562, + 562, + 589, + 538, + 589 + ], + "score": 0.43, + "latex": "\\overline { G }" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 780.0, + 935.0, + 780.0, + 935.0, + 825.0, + 294.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2083.0, + 872.0, + 2083.0, + 872.0, + 2124.0, + 829.0, + 2124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 853.0, + 1408.0, + 853.0, + 1408.0, + 889.0, + 294.0, + 889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 884.0, + 1403.0, + 884.0, + 1403.0, + 917.0, + 295.0, + 917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 916.0, + 1405.0, + 916.0, + 1405.0, + 952.0, + 292.0, + 952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 944.0, + 464.0, + 944.0, + 464.0, + 981.0, + 293.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 483.0, + 750.0, + 483.0, + 750.0, + 520.0, + 297.0, + 520.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 19, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 324, + 255, + 1398, + 255, + 1398, + 1945, + 324, + 1945 + ], + "score": 0.953 + }, + { + "category_id": 4, + "poly": [ + 503, + 1967, + 1195, + 1967, + 1195, + 2002, + 503, + 2002 + ], + "score": 0.925 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.884 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 863, + 2088, + 863, + 2113, + 834, + 2113 + ], + "score": 0.826 + }, + { + "category_id": 15, + "poly": [ + 344.0, + 255.0, + 412.0, + 255.0, + 412.0, + 287.0, + 344.0, + 287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 253.0, + 522.0, + 253.0, + 522.0, + 289.0, + 437.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 253.0, + 625.0, + 253.0, + 625.0, + 288.0, + 537.0, + 288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 253.0, + 841.0, + 253.0, + 841.0, + 287.0, + 641.0, + 287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 253.0, + 970.0, + 253.0, + 970.0, + 290.0, + 899.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 250.0, + 1397.0, + 250.0, + 1397.0, + 293.0, + 990.0, + 293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 285.0, + 833.0, + 285.0, + 833.0, + 375.0, + 327.0, + 375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 441.0, + 377.0, + 518.0, + 377.0, + 518.0, + 414.0, + 441.0, + 414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 380.0, + 624.0, + 380.0, + 624.0, + 410.0, + 538.0, + 410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 646.0, + 377.0, + 723.0, + 377.0, + 723.0, + 414.0, + 646.0, + 414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 380.0, + 828.0, + 380.0, + 828.0, + 412.0, + 745.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 378.0, + 1386.0, + 378.0, + 1386.0, + 414.0, + 986.0, + 414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 503.0, + 522.0, + 503.0, + 522.0, + 541.0, + 435.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 503.0, + 626.0, + 503.0, + 626.0, + 541.0, + 538.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 503.0, + 729.0, + 503.0, + 729.0, + 541.0, + 641.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 507.0, + 828.0, + 507.0, + 828.0, + 539.0, + 745.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 505.0, + 1389.0, + 505.0, + 1389.0, + 541.0, + 991.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 630.0, + 522.0, + 630.0, + 522.0, + 667.0, + 435.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 634.0, + 624.0, + 634.0, + 624.0, + 666.0, + 540.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 643.0, + 634.0, + 725.0, + 634.0, + 725.0, + 666.0, + 643.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 634.0, + 828.0, + 634.0, + 828.0, + 666.0, + 745.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 632.0, + 1385.0, + 632.0, + 1385.0, + 667.0, + 989.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 690.0, + 1382.0, + 690.0, + 1382.0, + 764.0, + 1277.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 757.0, + 522.0, + 757.0, + 522.0, + 794.0, + 435.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 758.0, + 623.0, + 758.0, + 623.0, + 790.0, + 536.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 757.0, + 729.0, + 757.0, + 729.0, + 794.0, + 641.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 757.0, + 830.0, + 757.0, + 830.0, + 794.0, + 744.0, + 794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 993.0, + 761.0, + 1077.0, + 761.0, + 1077.0, + 792.0, + 993.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 756.0, + 1380.0, + 756.0, + 1380.0, + 793.0, + 1103.0, + 793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 868.0, + 776.0, + 1402.0, + 776.0, + 1402.0, + 902.0, + 868.0, + 902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 887.0, + 520.0, + 887.0, + 520.0, + 919.0, + 435.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 887.0, + 624.0, + 887.0, + 624.0, + 917.0, + 538.0, + 917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 639.0, + 887.0, + 725.0, + 887.0, + 725.0, + 919.0, + 639.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 887.0, + 830.0, + 887.0, + 830.0, + 919.0, + 745.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 887.0, + 1076.0, + 887.0, + 1076.0, + 919.0, + 991.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 887.0, + 1386.0, + 887.0, + 1386.0, + 917.0, + 1092.0, + 917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 864.0, + 909.0, + 1406.0, + 909.0, + 1406.0, + 1023.0, + 864.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 1010.0, + 522.0, + 1010.0, + 522.0, + 1048.0, + 434.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1014.0, + 623.0, + 1014.0, + 623.0, + 1046.0, + 538.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 639.0, + 1010.0, + 727.0, + 1010.0, + 727.0, + 1048.0, + 639.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 753.0, + 1014.0, + 822.0, + 1014.0, + 822.0, + 1046.0, + 753.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 1012.0, + 1077.0, + 1012.0, + 1077.0, + 1044.0, + 988.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 1009.0, + 1389.0, + 1009.0, + 1389.0, + 1050.0, + 1092.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 364.0, + 1068.0, + 410.0, + 1068.0, + 410.0, + 1102.0, + 364.0, + 1102.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1072.0, + 368.0, + 1072.0, + 368.0, + 1109.0, + 347.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1096.0, + 511.0, + 1096.0, + 511.0, + 1111.0, + 489.0, + 1111.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 1137.0, + 520.0, + 1137.0, + 520.0, + 1175.0, + 437.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1139.0, + 742.0, + 1139.0, + 742.0, + 1171.0, + 547.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1141.0, + 830.0, + 1141.0, + 830.0, + 1171.0, + 745.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 1139.0, + 1387.0, + 1139.0, + 1387.0, + 1175.0, + 989.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1201.0, + 384.0, + 1201.0, + 384.0, + 1237.0, + 341.0, + 1237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 630.0, + 1186.0, + 822.0, + 1186.0, + 822.0, + 1255.0, + 630.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 879.0, + 1171.0, + 1398.0, + 1171.0, + 1398.0, + 1273.0, + 879.0, + 1273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 1264.0, + 832.0, + 1264.0, + 832.0, + 1301.0, + 435.0, + 1301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1004.0, + 1266.0, + 1066.0, + 1266.0, + 1066.0, + 1300.0, + 1004.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 1268.0, + 1197.0, + 1268.0, + 1197.0, + 1300.0, + 1096.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 1266.0, + 1386.0, + 1266.0, + 1386.0, + 1300.0, + 1199.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 1324.0, + 802.0, + 1324.0, + 802.0, + 1370.0, + 736.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 1395.0, + 830.0, + 1395.0, + 830.0, + 1424.0, + 435.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 997.0, + 1391.0, + 1074.0, + 1391.0, + 1074.0, + 1428.0, + 997.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1101.0, + 1395.0, + 1177.0, + 1395.0, + 1177.0, + 1424.0, + 1101.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 1395.0, + 1386.0, + 1395.0, + 1386.0, + 1426.0, + 1199.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 1520.0, + 830.0, + 1520.0, + 830.0, + 1551.0, + 435.0, + 1551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 1518.0, + 1077.0, + 1518.0, + 1077.0, + 1555.0, + 988.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 1520.0, + 1298.0, + 1520.0, + 1298.0, + 1555.0, + 1094.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 1518.0, + 1387.0, + 1518.0, + 1387.0, + 1555.0, + 1301.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1544.0, + 838.0, + 1544.0, + 838.0, + 1651.0, + 329.0, + 1651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1646.0, + 832.0, + 1646.0, + 832.0, + 1682.0, + 430.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 1646.0, + 1387.0, + 1646.0, + 1387.0, + 1682.0, + 989.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 1775.0, + 536.0, + 1775.0, + 536.0, + 1807.0, + 435.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 1775.0, + 740.0, + 1775.0, + 740.0, + 1807.0, + 540.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1775.0, + 830.0, + 1775.0, + 830.0, + 1807.0, + 745.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 1771.0, + 1079.0, + 1771.0, + 1079.0, + 1809.0, + 988.0, + 1809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 1771.0, + 1182.0, + 1771.0, + 1182.0, + 1809.0, + 1094.0, + 1809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 1771.0, + 1285.0, + 1771.0, + 1285.0, + 1809.0, + 1199.0, + 1809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 1775.0, + 1386.0, + 1775.0, + 1386.0, + 1807.0, + 1301.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 1902.0, + 520.0, + 1902.0, + 520.0, + 1933.0, + 435.0, + 1933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1900.0, + 623.0, + 1900.0, + 623.0, + 1932.0, + 538.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 1902.0, + 830.0, + 1902.0, + 830.0, + 1933.0, + 641.0, + 1933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 1902.0, + 1072.0, + 1902.0, + 1072.0, + 1933.0, + 999.0, + 1933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 1902.0, + 1180.0, + 1902.0, + 1180.0, + 1933.0, + 1096.0, + 1933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 1902.0, + 1387.0, + 1902.0, + 1387.0, + 1933.0, + 1197.0, + 1933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 833.5, + 364.0, + 833.5, + 364.0, + 883.5, + 356.0, + 883.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 1193.5, + 631.0, + 1193.5, + 631.0, + 1246.5, + 522.0, + 1246.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1325.0, + 385.0, + 1325.0, + 385.0, + 1358.0, + 358.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.0, + 1325.0, + 1364.0, + 1325.0, + 1364.0, + 1364.0, + 1299.0, + 1364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 501.0, + 1965.0, + 1196.0, + 1965.0, + 1196.0, + 2006.0, + 501.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2085.0, + 868.0, + 2085.0, + 868.0, + 2124.0, + 830.0, + 2124.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 20, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 4, + "poly": [ + 503, + 1967, + 1195, + 1967, + 1195, + 2002, + 503, + 2002 + ], + "score": 0.923 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 105, + 299, + 105 + ], + "score": 0.885 + }, + { + "category_id": 3, + "poly": [ + 324, + 250, + 1397, + 250, + 1397, + 1940, + 324, + 1940 + ], + "score": 0.872 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 865, + 2088, + 865, + 2113, + 834, + 2113 + ], + "score": 0.843 + }, + { + "category_id": 3, + "poly": [ + 324, + 248, + 1396, + 248, + 1396, + 1934, + 324, + 1934 + ], + "score": 0.693 + }, + { + "category_id": 15, + "poly": [ + 501.0, + 1965.0, + 1196.0, + 1965.0, + 1196.0, + 2006.0, + 501.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 71.0, + 816.0, + 71.0, + 816.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 256.0, + 835.0, + 256.0, + 835.0, + 286.0, + 345.0, + 286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 252.0, + 1394.0, + 252.0, + 1394.0, + 290.0, + 895.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 377.0, + 833.0, + 377.0, + 833.0, + 413.0, + 432.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 377.0, + 1390.0, + 377.0, + 1390.0, + 413.0, + 985.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 428.0, + 502.0, + 835.0, + 502.0, + 835.0, + 541.0, + 428.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 500.0, + 1392.0, + 500.0, + 1392.0, + 541.0, + 985.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 631.0, + 824.0, + 631.0, + 824.0, + 666.0, + 430.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 631.0, + 1388.0, + 631.0, + 1388.0, + 666.0, + 989.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 428.0, + 687.0, + 523.0, + 687.0, + 523.0, + 739.0, + 428.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 698.0, + 606.0, + 698.0, + 606.0, + 728.0, + 567.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 689.0, + 818.0, + 689.0, + 818.0, + 741.0, + 736.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 677.0, + 974.0, + 677.0, + 974.0, + 739.0, + 892.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 759.0, + 520.0, + 759.0, + 520.0, + 791.0, + 435.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 759.0, + 624.0, + 759.0, + 624.0, + 791.0, + 540.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 755.0, + 727.0, + 755.0, + 727.0, + 792.0, + 642.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 752.0, + 759.0, + 824.0, + 759.0, + 824.0, + 791.0, + 752.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 753.0, + 1383.0, + 753.0, + 1383.0, + 795.0, + 986.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 780.0, + 1062.0, + 780.0, + 1062.0, + 899.0, + 323.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1293.0, + 813.0, + 1385.0, + 813.0, + 1385.0, + 866.0, + 1293.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 884.0, + 831.0, + 884.0, + 831.0, + 920.0, + 430.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 886.0, + 1077.0, + 886.0, + 1077.0, + 918.0, + 992.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 882.0, + 1183.0, + 882.0, + 1183.0, + 920.0, + 1095.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 886.0, + 1284.0, + 886.0, + 1284.0, + 918.0, + 1198.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 882.0, + 1388.0, + 882.0, + 1388.0, + 920.0, + 1300.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 414.0, + 922.0, + 534.0, + 922.0, + 534.0, + 988.0, + 414.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 923.0, + 444.0, + 923.0, + 444.0, + 991.0, + 336.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1009.0, + 835.0, + 1009.0, + 835.0, + 1050.0, + 430.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 1011.0, + 1390.0, + 1011.0, + 1390.0, + 1047.0, + 991.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 1060.0, + 1374.0, + 1060.0, + 1374.0, + 1132.0, + 1005.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 1138.0, + 523.0, + 1138.0, + 523.0, + 1175.0, + 433.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 1138.0, + 829.0, + 1138.0, + 829.0, + 1173.0, + 536.0, + 1173.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 1138.0, + 1390.0, + 1138.0, + 1390.0, + 1173.0, + 996.0, + 1173.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 1265.0, + 625.0, + 1265.0, + 625.0, + 1298.0, + 433.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 1267.0, + 727.0, + 1267.0, + 727.0, + 1298.0, + 642.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 752.0, + 1267.0, + 822.0, + 1267.0, + 822.0, + 1298.0, + 752.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1263.0, + 1383.0, + 1263.0, + 1383.0, + 1304.0, + 992.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 1393.0, + 727.0, + 1393.0, + 727.0, + 1425.0, + 435.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1393.0, + 829.0, + 1393.0, + 829.0, + 1425.0, + 745.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 1390.0, + 1390.0, + 1390.0, + 1390.0, + 1431.0, + 989.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 1520.0, + 521.0, + 1520.0, + 521.0, + 1552.0, + 437.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 1520.0, + 630.0, + 1520.0, + 630.0, + 1552.0, + 540.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 1518.0, + 826.0, + 1518.0, + 826.0, + 1554.0, + 641.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 1516.0, + 1390.0, + 1516.0, + 1390.0, + 1557.0, + 989.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 1647.0, + 521.0, + 1647.0, + 521.0, + 1679.0, + 435.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 1647.0, + 727.0, + 1647.0, + 727.0, + 1679.0, + 536.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1647.0, + 829.0, + 1647.0, + 829.0, + 1677.0, + 745.0, + 1677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 1645.0, + 1388.0, + 1645.0, + 1388.0, + 1680.0, + 991.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 1774.0, + 534.0, + 1774.0, + 534.0, + 1805.0, + 435.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1770.0, + 831.0, + 1770.0, + 831.0, + 1806.0, + 539.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1772.0, + 1386.0, + 1772.0, + 1386.0, + 1807.0, + 992.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 318.0, + 1798.0, + 848.0, + 1798.0, + 848.0, + 1917.0, + 318.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 1811.0, + 1077.0, + 1811.0, + 1077.0, + 1899.0, + 881.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1150.0, + 1845.0, + 1165.0, + 1845.0, + 1165.0, + 1859.0, + 1150.0, + 1859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1900.0, + 521.0, + 1900.0, + 521.0, + 1930.0, + 430.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1900.0, + 626.0, + 1900.0, + 626.0, + 1930.0, + 538.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 1900.0, + 729.0, + 1900.0, + 729.0, + 1930.0, + 641.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1900.0, + 831.0, + 1900.0, + 831.0, + 1932.0, + 745.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 1900.0, + 1077.0, + 1900.0, + 1077.0, + 1932.0, + 985.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1101.0, + 1900.0, + 1176.0, + 1900.0, + 1176.0, + 1932.0, + 1101.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 1900.0, + 1284.0, + 1900.0, + 1284.0, + 1932.0, + 1198.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 1900.0, + 1387.0, + 1900.0, + 1387.0, + 1932.0, + 1302.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2125.0, + 829.0, + 2125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 252.0, + 838.0, + 252.0, + 838.0, + 289.0, + 342.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 253.0, + 1395.0, + 253.0, + 1395.0, + 290.0, + 899.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.0, + 372.0, + 831.0, + 372.0, + 831.0, + 414.0, + 429.0, + 414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 377.0, + 1387.0, + 377.0, + 1387.0, + 412.0, + 992.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 518.0, + 417.0, + 518.0, + 417.0, + 531.0, + 391.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 507.0, + 833.0, + 507.0, + 833.0, + 542.0, + 433.0, + 542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 503.0, + 1391.0, + 503.0, + 1391.0, + 544.0, + 988.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 979.0, + 533.0, + 1398.0, + 533.0, + 1398.0, + 641.0, + 979.0, + 641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.0, + 629.0, + 822.0, + 629.0, + 822.0, + 665.0, + 431.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 627.0, + 1386.0, + 627.0, + 1386.0, + 671.0, + 989.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 414.0, + 667.0, + 840.0, + 667.0, + 840.0, + 755.0, + 414.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 886.0, + 676.0, + 981.0, + 676.0, + 981.0, + 745.0, + 886.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.0, + 754.0, + 826.0, + 754.0, + 826.0, + 796.0, + 431.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 756.0, + 1384.0, + 756.0, + 1384.0, + 797.0, + 990.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 780.0, + 849.0, + 780.0, + 849.0, + 899.0, + 322.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.0, + 882.0, + 829.0, + 882.0, + 829.0, + 920.0, + 431.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 883.0, + 1080.0, + 883.0, + 1080.0, + 920.0, + 992.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 883.0, + 1296.0, + 883.0, + 1296.0, + 920.0, + 1096.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 883.0, + 1387.0, + 883.0, + 1387.0, + 920.0, + 1301.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 922.0, + 626.0, + 922.0, + 626.0, + 994.0, + 323.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 1013.0, + 833.0, + 1013.0, + 833.0, + 1048.0, + 433.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1011.0, + 1389.0, + 1011.0, + 1389.0, + 1046.0, + 992.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 974.0, + 1039.0, + 1397.0, + 1039.0, + 1397.0, + 1151.0, + 974.0, + 1151.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 1138.0, + 521.0, + 1138.0, + 521.0, + 1175.0, + 433.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 1138.0, + 637.0, + 1138.0, + 637.0, + 1170.0, + 535.0, + 1170.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 638.0, + 1138.0, + 739.0, + 1138.0, + 739.0, + 1170.0, + 638.0, + 1170.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1139.0, + 829.0, + 1139.0, + 829.0, + 1171.0, + 745.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1141.0, + 1386.0, + 1141.0, + 1386.0, + 1171.0, + 1001.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 1225.0, + 389.0, + 1225.0, + 389.0, + 1253.0, + 344.0, + 1253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 1264.0, + 627.0, + 1264.0, + 627.0, + 1301.0, + 433.0, + 1301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 1268.0, + 726.0, + 1268.0, + 726.0, + 1299.0, + 642.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 728.0, + 1264.0, + 825.0, + 1264.0, + 825.0, + 1301.0, + 728.0, + 1301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1266.0, + 1384.0, + 1266.0, + 1384.0, + 1301.0, + 992.0, + 1301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 1392.0, + 831.0, + 1392.0, + 831.0, + 1428.0, + 435.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 1391.0, + 1389.0, + 1391.0, + 1389.0, + 1431.0, + 990.0, + 1431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 1521.0, + 519.0, + 1521.0, + 519.0, + 1552.0, + 437.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1521.0, + 624.0, + 1521.0, + 624.0, + 1552.0, + 538.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 1521.0, + 748.0, + 1521.0, + 748.0, + 1552.0, + 642.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 752.0, + 1521.0, + 823.0, + 1521.0, + 823.0, + 1552.0, + 752.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1517.0, + 1387.0, + 1517.0, + 1387.0, + 1554.0, + 992.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 1647.0, + 521.0, + 1647.0, + 521.0, + 1679.0, + 437.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 1647.0, + 624.0, + 1647.0, + 624.0, + 1679.0, + 540.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 1647.0, + 726.0, + 1647.0, + 726.0, + 1679.0, + 642.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1647.0, + 829.0, + 1647.0, + 829.0, + 1679.0, + 745.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 1647.0, + 1387.0, + 1647.0, + 1387.0, + 1679.0, + 994.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 1684.0, + 545.0, + 1684.0, + 545.0, + 1766.0, + 338.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 1776.0, + 831.0, + 1776.0, + 831.0, + 1805.0, + 435.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1770.0, + 1389.0, + 1770.0, + 1389.0, + 1809.0, + 992.0, + 1809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 1794.0, + 855.0, + 1794.0, + 855.0, + 1921.0, + 314.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1837.0, + 963.0, + 1837.0, + 963.0, + 1885.0, + 895.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 1896.0, + 522.0, + 1896.0, + 522.0, + 1932.0, + 434.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1896.0, + 624.0, + 1896.0, + 624.0, + 1932.0, + 539.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 1900.0, + 726.0, + 1900.0, + 726.0, + 1932.0, + 642.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1900.0, + 831.0, + 1900.0, + 831.0, + 1932.0, + 745.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 1900.0, + 1078.0, + 1900.0, + 1078.0, + 1932.0, + 994.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1104.0, + 1904.0, + 1175.0, + 1904.0, + 1175.0, + 1932.0, + 1104.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 1902.0, + 1283.0, + 1902.0, + 1283.0, + 1932.0, + 1199.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1301.0, + 1902.0, + 1386.0, + 1902.0, + 1386.0, + 1934.0, + 1301.0, + 1934.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 21, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 323, + 248, + 1397, + 248, + 1397, + 1938, + 323, + 1938 + ], + "score": 0.945 + }, + { + "category_id": 4, + "poly": [ + 503, + 1967, + 1195, + 1967, + 1195, + 2002, + 503, + 2002 + ], + "score": 0.925 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 105, + 299, + 105 + ], + "score": 0.886 + }, + { + "category_id": 2, + "poly": [ + 835, + 2088, + 865, + 2088, + 865, + 2113, + 835, + 2113 + ], + "score": 0.862 + }, + { + "category_id": 15, + "poly": [ + 341.0, + 252.0, + 838.0, + 252.0, + 838.0, + 289.0, + 341.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 897.0, + 252.0, + 1394.0, + 252.0, + 1394.0, + 288.0, + 897.0, + 288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 377.0, + 832.0, + 377.0, + 832.0, + 412.0, + 449.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 379.0, + 1076.0, + 379.0, + 1076.0, + 411.0, + 988.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 377.0, + 1388.0, + 377.0, + 1388.0, + 412.0, + 1095.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 502.0, + 827.0, + 502.0, + 827.0, + 541.0, + 423.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 506.0, + 1386.0, + 506.0, + 1386.0, + 541.0, + 990.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 539.0, + 1381.0, + 539.0, + 1381.0, + 633.0, + 985.0, + 633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 428.0, + 626.0, + 829.0, + 626.0, + 829.0, + 668.0, + 428.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 632.0, + 1198.0, + 632.0, + 1198.0, + 664.0, + 994.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 632.0, + 1286.0, + 632.0, + 1286.0, + 664.0, + 1199.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 636.0, + 1372.0, + 636.0, + 1372.0, + 660.0, + 1317.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 664.0, + 1392.0, + 664.0, + 1392.0, + 767.0, + 981.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 759.0, + 831.0, + 759.0, + 831.0, + 795.0, + 427.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 756.0, + 1390.0, + 756.0, + 1390.0, + 797.0, + 988.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 786.0, + 1395.0, + 786.0, + 1395.0, + 895.0, + 908.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 884.0, + 741.0, + 884.0, + 741.0, + 920.0, + 436.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 888.0, + 829.0, + 888.0, + 829.0, + 920.0, + 744.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 886.0, + 1386.0, + 886.0, + 1386.0, + 918.0, + 988.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 919.0, + 1390.0, + 919.0, + 1390.0, + 1021.0, + 935.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 1009.0, + 825.0, + 1009.0, + 825.0, + 1050.0, + 434.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 1011.0, + 1390.0, + 1011.0, + 1390.0, + 1046.0, + 987.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 1045.0, + 834.0, + 1045.0, + 834.0, + 1140.0, + 372.0, + 1140.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.0, + 1133.0, + 833.0, + 1133.0, + 833.0, + 1176.0, + 429.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 1138.0, + 1078.0, + 1138.0, + 1078.0, + 1175.0, + 990.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 1140.0, + 1179.0, + 1140.0, + 1179.0, + 1171.0, + 1097.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 1140.0, + 1284.0, + 1140.0, + 1284.0, + 1171.0, + 1199.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 1140.0, + 1386.0, + 1140.0, + 1386.0, + 1171.0, + 1302.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 1196.0, + 1067.0, + 1196.0, + 1067.0, + 1244.0, + 1001.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1227.0, + 1201.0, + 1289.0, + 1201.0, + 1289.0, + 1242.0, + 1227.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1293.0, + 1194.0, + 1368.0, + 1194.0, + 1368.0, + 1248.0, + 1293.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.0, + 1265.0, + 832.0, + 1265.0, + 832.0, + 1300.0, + 429.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 1261.0, + 1390.0, + 1261.0, + 1390.0, + 1304.0, + 994.0, + 1304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1298.0, + 605.0, + 1298.0, + 605.0, + 1404.0, + 323.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 1317.0, + 825.0, + 1317.0, + 825.0, + 1389.0, + 686.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.0, + 1391.0, + 823.0, + 1391.0, + 823.0, + 1427.0, + 431.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 1393.0, + 1076.0, + 1393.0, + 1076.0, + 1425.0, + 988.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 1391.0, + 1386.0, + 1391.0, + 1386.0, + 1427.0, + 1095.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1258.0, + 1432.0, + 1381.0, + 1432.0, + 1381.0, + 1503.0, + 1258.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.0, + 1516.0, + 832.0, + 1516.0, + 832.0, + 1557.0, + 431.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 1516.0, + 1390.0, + 1516.0, + 1390.0, + 1555.0, + 986.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 1547.0, + 432.0, + 1547.0, + 432.0, + 1648.0, + 327.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.0, + 1645.0, + 625.0, + 1645.0, + 625.0, + 1682.0, + 429.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 629.0, + 1658.0, + 651.0, + 1658.0, + 651.0, + 1673.0, + 629.0, + 1673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 1645.0, + 827.0, + 1645.0, + 827.0, + 1680.0, + 653.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 1645.0, + 1181.0, + 1645.0, + 1181.0, + 1682.0, + 998.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 1647.0, + 1284.0, + 1647.0, + 1284.0, + 1678.0, + 1199.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 1647.0, + 1386.0, + 1647.0, + 1386.0, + 1678.0, + 1302.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1770.0, + 832.0, + 1770.0, + 832.0, + 1811.0, + 436.0, + 1811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 1770.0, + 1392.0, + 1770.0, + 1392.0, + 1811.0, + 988.0, + 1811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 1831.0, + 961.0, + 1831.0, + 961.0, + 1884.0, + 904.0, + 1884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 1900.0, + 521.0, + 1900.0, + 521.0, + 1932.0, + 427.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 1900.0, + 623.0, + 1900.0, + 623.0, + 1932.0, + 541.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 647.0, + 1898.0, + 831.0, + 1898.0, + 831.0, + 1936.0, + 647.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 1898.0, + 1078.0, + 1898.0, + 1078.0, + 1936.0, + 990.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1100.0, + 1898.0, + 1177.0, + 1898.0, + 1177.0, + 1936.0, + 1100.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 1900.0, + 1282.0, + 1900.0, + 1282.0, + 1932.0, + 1199.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 1900.0, + 1385.0, + 1900.0, + 1385.0, + 1932.0, + 1302.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 501.0, + 1965.0, + 1196.0, + 1965.0, + 1196.0, + 2006.0, + 501.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 71.0, + 816.0, + 71.0, + 816.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 2085.0, + 869.0, + 2085.0, + 869.0, + 2123.0, + 830.0, + 2123.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 22, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 324, + 259, + 1397, + 259, + 1397, + 1938, + 324, + 1938 + ], + "score": 0.953 + }, + { + "category_id": 4, + "poly": [ + 503, + 1968, + 1195, + 1968, + 1195, + 2002, + 503, + 2002 + ], + "score": 0.925 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 104, + 299, + 104 + ], + "score": 0.885 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 865, + 2088, + 865, + 2113, + 834, + 2113 + ], + "score": 0.863 + }, + { + "category_id": 15, + "poly": [ + 342.0, + 257.0, + 410.0, + 257.0, + 410.0, + 289.0, + 342.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 253.0, + 520.0, + 253.0, + 520.0, + 289.0, + 436.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 251.0, + 839.0, + 251.0, + 839.0, + 291.0, + 540.0, + 291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 259.0, + 963.0, + 259.0, + 963.0, + 285.0, + 901.0, + 285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 253.0, + 1077.0, + 253.0, + 1077.0, + 289.0, + 994.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 257.0, + 1179.0, + 257.0, + 1179.0, + 287.0, + 1095.0, + 287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1193.0, + 251.0, + 1394.0, + 251.0, + 1394.0, + 291.0, + 1193.0, + 291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 378.0, + 521.0, + 378.0, + 521.0, + 415.0, + 433.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 543.0, + 378.0, + 620.0, + 378.0, + 620.0, + 415.0, + 543.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 379.0, + 727.0, + 379.0, + 727.0, + 411.0, + 642.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 379.0, + 828.0, + 379.0, + 828.0, + 411.0, + 747.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 378.0, + 1388.0, + 378.0, + 1388.0, + 413.0, + 987.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 504.0, + 625.0, + 504.0, + 625.0, + 541.0, + 433.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 504.0, + 833.0, + 504.0, + 833.0, + 541.0, + 641.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 504.0, + 1390.0, + 504.0, + 1390.0, + 539.0, + 987.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 633.0, + 520.0, + 633.0, + 520.0, + 665.0, + 437.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 632.0, + 829.0, + 632.0, + 829.0, + 665.0, + 545.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 630.0, + 1390.0, + 630.0, + 1390.0, + 670.0, + 996.0, + 670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 659.0, + 648.0, + 659.0, + 648.0, + 767.0, + 433.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 760.0, + 695.0, + 796.0, + 695.0, + 796.0, + 733.0, + 760.0, + 733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 755.0, + 520.0, + 755.0, + 520.0, + 791.0, + 436.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 759.0, + 624.0, + 759.0, + 624.0, + 791.0, + 540.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 759.0, + 725.0, + 759.0, + 725.0, + 791.0, + 642.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 758.0, + 824.0, + 758.0, + 824.0, + 795.0, + 749.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 753.0, + 1079.0, + 753.0, + 1079.0, + 795.0, + 986.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 753.0, + 1282.0, + 753.0, + 1282.0, + 795.0, + 1092.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1304.0, + 753.0, + 1380.0, + 753.0, + 1380.0, + 795.0, + 1304.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 887.0, + 520.0, + 887.0, + 520.0, + 919.0, + 437.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 887.0, + 624.0, + 887.0, + 624.0, + 919.0, + 540.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 887.0, + 725.0, + 887.0, + 725.0, + 919.0, + 642.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 887.0, + 829.0, + 887.0, + 829.0, + 919.0, + 745.0, + 919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 885.0, + 1388.0, + 885.0, + 1388.0, + 921.0, + 989.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 932.0, + 917.0, + 1398.0, + 917.0, + 1398.0, + 1013.0, + 932.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 1007.0, + 524.0, + 1007.0, + 524.0, + 1049.0, + 433.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1013.0, + 637.0, + 1013.0, + 637.0, + 1045.0, + 538.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 1013.0, + 740.0, + 1013.0, + 740.0, + 1045.0, + 641.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 1010.0, + 831.0, + 1010.0, + 831.0, + 1047.0, + 743.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 1011.0, + 1388.0, + 1011.0, + 1388.0, + 1047.0, + 985.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 1137.0, + 833.0, + 1137.0, + 833.0, + 1178.0, + 432.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1139.0, + 1388.0, + 1139.0, + 1388.0, + 1174.0, + 992.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 443.0, + 1267.0, + 514.0, + 1267.0, + 514.0, + 1299.0, + 443.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 1267.0, + 727.0, + 1267.0, + 727.0, + 1297.0, + 540.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 1267.0, + 829.0, + 1267.0, + 829.0, + 1299.0, + 747.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1263.0, + 1388.0, + 1263.0, + 1388.0, + 1299.0, + 992.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 1294.0, + 833.0, + 1294.0, + 833.0, + 1403.0, + 336.0, + 1403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 1393.0, + 521.0, + 1393.0, + 521.0, + 1425.0, + 432.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 538.0, + 1395.0, + 727.0, + 1395.0, + 727.0, + 1425.0, + 538.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 738.0, + 1393.0, + 831.0, + 1393.0, + 831.0, + 1425.0, + 738.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1390.0, + 1390.0, + 1390.0, + 1390.0, + 1428.0, + 992.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 1519.0, + 831.0, + 1519.0, + 831.0, + 1554.0, + 435.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1517.0, + 1387.0, + 1517.0, + 1387.0, + 1558.0, + 992.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 1645.0, + 833.0, + 1645.0, + 833.0, + 1680.0, + 432.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 1645.0, + 1284.0, + 1645.0, + 1284.0, + 1680.0, + 994.0, + 1680.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 1647.0, + 1388.0, + 1647.0, + 1388.0, + 1679.0, + 1302.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 1706.0, + 573.0, + 1706.0, + 573.0, + 1727.0, + 540.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 1712.0, + 659.0, + 1712.0, + 659.0, + 1729.0, + 641.0, + 1729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 1775.0, + 829.0, + 1775.0, + 829.0, + 1806.0, + 435.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1775.0, + 1077.0, + 1775.0, + 1077.0, + 1805.0, + 992.0, + 1805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1082.0, + 1775.0, + 1387.0, + 1775.0, + 1387.0, + 1806.0, + 1082.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1334.0, + 1852.0, + 1377.0, + 1852.0, + 1377.0, + 1885.0, + 1334.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1901.0, + 520.0, + 1901.0, + 520.0, + 1932.0, + 430.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 540.0, + 1901.0, + 622.0, + 1901.0, + 622.0, + 1932.0, + 540.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 1901.0, + 727.0, + 1901.0, + 727.0, + 1932.0, + 642.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1901.0, + 829.0, + 1901.0, + 829.0, + 1932.0, + 745.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 1899.0, + 1077.0, + 1899.0, + 1077.0, + 1936.0, + 987.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 1901.0, + 1178.0, + 1901.0, + 1178.0, + 1932.0, + 1095.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 1901.0, + 1282.0, + 1901.0, + 1282.0, + 1932.0, + 1198.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 1901.0, + 1387.0, + 1901.0, + 1387.0, + 1932.0, + 1302.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 1689.5, + 442.0, + 1689.5, + 442.0, + 1740.0, + 372.0, + 1740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 501.0, + 1965.0, + 1195.0, + 1965.0, + 1195.0, + 2005.0, + 501.0, + 2005.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 73.0, + 816.0, + 73.0, + 816.0, + 108.0, + 296.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2084.0, + 871.0, + 2084.0, + 871.0, + 2125.0, + 829.0, + 2125.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 23, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 323, + 248, + 1397, + 248, + 1397, + 1939, + 323, + 1939 + ], + "score": 0.948 + }, + { + "category_id": 4, + "poly": [ + 496, + 1967, + 1202, + 1967, + 1202, + 2002, + 496, + 2002 + ], + "score": 0.925 + }, + { + "category_id": 2, + "poly": [ + 299, + 76, + 814, + 76, + 814, + 105, + 299, + 105 + ], + "score": 0.885 + }, + { + "category_id": 2, + "poly": [ + 834, + 2088, + 865, + 2088, + 865, + 2113, + 834, + 2113 + ], + "score": 0.864 + }, + { + "category_id": 15, + "poly": [ + 345.0, + 256.0, + 836.0, + 256.0, + 836.0, + 286.0, + 345.0, + 286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 252.0, + 1394.0, + 252.0, + 1394.0, + 290.0, + 899.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.0, + 377.0, + 831.0, + 377.0, + 831.0, + 413.0, + 431.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 379.0, + 1076.0, + 379.0, + 1076.0, + 411.0, + 992.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 379.0, + 1183.0, + 379.0, + 1183.0, + 411.0, + 1097.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 379.0, + 1284.0, + 379.0, + 1284.0, + 411.0, + 1199.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 379.0, + 1386.0, + 379.0, + 1386.0, + 411.0, + 1302.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 502.0, + 832.0, + 502.0, + 832.0, + 541.0, + 434.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 506.0, + 1076.0, + 506.0, + 1076.0, + 538.0, + 992.0, + 538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1102.0, + 506.0, + 1386.0, + 506.0, + 1386.0, + 538.0, + 1102.0, + 538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 631.0, + 517.0, + 631.0, + 517.0, + 668.0, + 440.0, + 668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 633.0, + 625.0, + 633.0, + 625.0, + 664.0, + 541.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 633.0, + 726.0, + 633.0, + 726.0, + 664.0, + 642.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 633.0, + 829.0, + 633.0, + 829.0, + 664.0, + 744.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 633.0, + 1078.0, + 633.0, + 1078.0, + 664.0, + 992.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 631.0, + 1388.0, + 631.0, + 1388.0, + 666.0, + 1095.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 442.0, + 756.0, + 834.0, + 756.0, + 834.0, + 797.0, + 442.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 760.0, + 1381.0, + 760.0, + 1381.0, + 791.0, + 992.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 438.0, + 806.0, + 840.0, + 806.0, + 840.0, + 885.0, + 438.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 806.0, + 987.0, + 806.0, + 987.0, + 875.0, + 889.0, + 875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.0, + 886.0, + 522.0, + 886.0, + 522.0, + 916.0, + 429.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 886.0, + 829.0, + 886.0, + 829.0, + 916.0, + 539.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 883.0, + 1078.0, + 883.0, + 1078.0, + 920.0, + 990.0, + 920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 886.0, + 1183.0, + 886.0, + 1183.0, + 918.0, + 1097.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 886.0, + 1284.0, + 886.0, + 1284.0, + 918.0, + 1199.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1302.0, + 886.0, + 1385.0, + 886.0, + 1385.0, + 918.0, + 1302.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 914.0, + 842.0, + 914.0, + 842.0, + 1023.0, + 324.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 1011.0, + 829.0, + 1011.0, + 829.0, + 1047.0, + 425.0, + 1047.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1010.0, + 1392.0, + 1010.0, + 1392.0, + 1051.0, + 992.0, + 1051.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 1045.0, + 985.0, + 1045.0, + 985.0, + 1148.0, + 882.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 1140.0, + 521.0, + 1140.0, + 521.0, + 1172.0, + 434.0, + 1172.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1140.0, + 625.0, + 1140.0, + 625.0, + 1172.0, + 539.0, + 1172.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 1140.0, + 728.0, + 1140.0, + 728.0, + 1172.0, + 642.0, + 1172.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 1140.0, + 829.0, + 1140.0, + 829.0, + 1172.0, + 744.0, + 1172.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1140.0, + 1080.0, + 1140.0, + 1080.0, + 1172.0, + 992.0, + 1172.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 1138.0, + 1386.0, + 1138.0, + 1386.0, + 1174.0, + 1093.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1267.0, + 521.0, + 1267.0, + 521.0, + 1299.0, + 436.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1267.0, + 625.0, + 1267.0, + 625.0, + 1299.0, + 539.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 1267.0, + 726.0, + 1267.0, + 726.0, + 1299.0, + 642.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 1265.0, + 831.0, + 1265.0, + 831.0, + 1302.0, + 743.0, + 1302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 1265.0, + 1388.0, + 1265.0, + 1388.0, + 1301.0, + 988.0, + 1301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 729.0, + 1329.0, + 831.0, + 1329.0, + 831.0, + 1359.0, + 729.0, + 1359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1394.0, + 521.0, + 1394.0, + 521.0, + 1426.0, + 436.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 1394.0, + 623.0, + 1394.0, + 623.0, + 1426.0, + 541.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 1394.0, + 726.0, + 1394.0, + 726.0, + 1426.0, + 642.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 1394.0, + 829.0, + 1394.0, + 829.0, + 1426.0, + 746.0, + 1426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 1394.0, + 1076.0, + 1394.0, + 1076.0, + 1424.0, + 988.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 1392.0, + 1388.0, + 1392.0, + 1388.0, + 1427.0, + 1093.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 1517.0, + 517.0, + 1517.0, + 517.0, + 1554.0, + 440.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 1517.0, + 625.0, + 1517.0, + 625.0, + 1554.0, + 537.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 640.0, + 1519.0, + 728.0, + 1519.0, + 728.0, + 1551.0, + 640.0, + 1551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 1513.0, + 831.0, + 1513.0, + 831.0, + 1555.0, + 742.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 1517.0, + 1383.0, + 1517.0, + 1383.0, + 1558.0, + 988.0, + 1558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 1648.0, + 522.0, + 1648.0, + 522.0, + 1679.0, + 434.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1648.0, + 627.0, + 1648.0, + 627.0, + 1679.0, + 539.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 1646.0, + 832.0, + 1646.0, + 832.0, + 1681.0, + 645.0, + 1681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1646.0, + 1388.0, + 1646.0, + 1388.0, + 1681.0, + 992.0, + 1681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1773.0, + 726.0, + 1773.0, + 726.0, + 1808.0, + 436.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 728.0, + 1774.0, + 823.0, + 1774.0, + 823.0, + 1806.0, + 728.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 1774.0, + 1386.0, + 1774.0, + 1386.0, + 1806.0, + 999.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 888.0, + 1829.0, + 974.0, + 1829.0, + 974.0, + 1899.0, + 888.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 1851.0, + 1025.0, + 1851.0, + 1025.0, + 1866.0, + 1009.0, + 1866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 1901.0, + 521.0, + 1901.0, + 521.0, + 1933.0, + 434.0, + 1933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 1901.0, + 623.0, + 1901.0, + 623.0, + 1933.0, + 541.0, + 1933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 644.0, + 1901.0, + 726.0, + 1901.0, + 726.0, + 1933.0, + 644.0, + 1933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 1901.0, + 831.0, + 1901.0, + 831.0, + 1933.0, + 744.0, + 1933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 1893.0, + 1081.0, + 1893.0, + 1081.0, + 1936.0, + 986.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 1899.0, + 1183.0, + 1899.0, + 1183.0, + 1931.0, + 1095.0, + 1931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1201.0, + 1901.0, + 1287.0, + 1901.0, + 1287.0, + 1931.0, + 1201.0, + 1931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 1901.0, + 1388.0, + 1901.0, + 1388.0, + 1931.0, + 1300.0, + 1931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 1965.0, + 1204.0, + 1965.0, + 1204.0, + 2006.0, + 494.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 71.0, + 816.0, + 71.0, + 816.0, + 108.0, + 295.0, + 108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 2083.0, + 871.0, + 2083.0, + 871.0, + 2124.0, + 829.0, + 2124.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 24, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/CU0APx9LMaL/images/0a233ba78d8aee5f9c455f2ba91df694b349684b59f498ebb491da70235c132a.jpg b/parse/train/CU0APx9LMaL/images/0a233ba78d8aee5f9c455f2ba91df694b349684b59f498ebb491da70235c132a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2e071206da019f1bd9a343ede1e9700133553449 --- /dev/null +++ b/parse/train/CU0APx9LMaL/images/0a233ba78d8aee5f9c455f2ba91df694b349684b59f498ebb491da70235c132a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1fcc290cb4fa8ebbfdeba7ba002097a493f6c551617a8e4758799e39c5dabfdb +size 23834 diff --git a/parse/train/CU0APx9LMaL/images/2aa7682ff6f6d64290a750a2d26b39eb19920f1daafef8c9f300ce841533edbc.jpg b/parse/train/CU0APx9LMaL/images/2aa7682ff6f6d64290a750a2d26b39eb19920f1daafef8c9f300ce841533edbc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0c03f0e64026c2bc5421b6fbd24c1661686dd815 --- /dev/null +++ b/parse/train/CU0APx9LMaL/images/2aa7682ff6f6d64290a750a2d26b39eb19920f1daafef8c9f300ce841533edbc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4c28349ae4bdcf1ee61290deb9a27a860a630ea63bd4d98dcba8e46bc065dad4 +size 110683 diff --git a/parse/train/CU0APx9LMaL/images/2d0af86ca8edae773833df1fed2d52c7d3e28cdb3474189a864fb45257097cfe.jpg b/parse/train/CU0APx9LMaL/images/2d0af86ca8edae773833df1fed2d52c7d3e28cdb3474189a864fb45257097cfe.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2959dc66d8787103f9a390284fe807ccf5236c65 --- /dev/null +++ b/parse/train/CU0APx9LMaL/images/2d0af86ca8edae773833df1fed2d52c7d3e28cdb3474189a864fb45257097cfe.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e882d212d58b1cbb52f8e5a722bf58104f2dbb75e8b20bc52fad99f2b800184a +size 31184 diff --git a/parse/train/CU0APx9LMaL/images/35a23870e5ce55539da32c082d7b165e304948c4426de4dd851989434dd2fb1b.jpg b/parse/train/CU0APx9LMaL/images/35a23870e5ce55539da32c082d7b165e304948c4426de4dd851989434dd2fb1b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2f8adcb592da3f9da937d24843fcf96cbed4da62 --- /dev/null +++ b/parse/train/CU0APx9LMaL/images/35a23870e5ce55539da32c082d7b165e304948c4426de4dd851989434dd2fb1b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:732163595dde5918bf8165a651807701c070d87924e84b66c60ec49d155c99b4 +size 89754 diff --git a/parse/train/CU0APx9LMaL/images/497bc85272436c739e1a2bbaeb0089c0470480b07bbfc555b59bdd88e62e8650.jpg b/parse/train/CU0APx9LMaL/images/497bc85272436c739e1a2bbaeb0089c0470480b07bbfc555b59bdd88e62e8650.jpg new file mode 100644 index 0000000000000000000000000000000000000000..87cdb90af9aaa7b11b10e7cbb3c75d023a5285da --- /dev/null +++ b/parse/train/CU0APx9LMaL/images/497bc85272436c739e1a2bbaeb0089c0470480b07bbfc555b59bdd88e62e8650.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5cbf8e98b1e15ac33fda1be7d9854c1f88ef946314e41852adb6e0d5c3f50710 +size 14501 diff --git a/parse/train/CU0APx9LMaL/images/5088061f5a759f979d88de131bf572316d35e3563432652d93718bd7dc1a3b95.jpg b/parse/train/CU0APx9LMaL/images/5088061f5a759f979d88de131bf572316d35e3563432652d93718bd7dc1a3b95.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2a08d7c8d4af8a6f0676861ca54df34a6e9a992f --- /dev/null +++ b/parse/train/CU0APx9LMaL/images/5088061f5a759f979d88de131bf572316d35e3563432652d93718bd7dc1a3b95.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:905eef70e29e243e3e29154566416fbd61ba7d6157002ba0456659761c26a774 +size 25167 diff --git a/parse/train/CU0APx9LMaL/images/77ea062c368189785e7300ffa778f2a24a06c1b7ad9649bb85fb2f173a8c6f6c.jpg b/parse/train/CU0APx9LMaL/images/77ea062c368189785e7300ffa778f2a24a06c1b7ad9649bb85fb2f173a8c6f6c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d604bdeb395c32ed3f21d67ef5803d35d06e677f --- /dev/null +++ b/parse/train/CU0APx9LMaL/images/77ea062c368189785e7300ffa778f2a24a06c1b7ad9649bb85fb2f173a8c6f6c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89ee46e3211e200278c934ccdd8fd49141829c8353e24bacd9fee44fefdbdc42 +size 51735 diff --git a/parse/train/CU0APx9LMaL/images/8c90c5857af224c30d93df8385c104f07c53878301e83d77c1faa5adac0b2f6f.jpg b/parse/train/CU0APx9LMaL/images/8c90c5857af224c30d93df8385c104f07c53878301e83d77c1faa5adac0b2f6f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b50506d00e237cf717729ef20f78d11f44f48822 --- /dev/null +++ b/parse/train/CU0APx9LMaL/images/8c90c5857af224c30d93df8385c104f07c53878301e83d77c1faa5adac0b2f6f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cdf2d7fb36ebbbc3669a3771173d63937d5109d0c40481f324a770da2c15532a +size 49310 diff --git a/parse/train/CU0APx9LMaL/images/c0aa9cae86b38f7e55589b7ef2336de40eafa2e09cefd877788706683dd174ce.jpg b/parse/train/CU0APx9LMaL/images/c0aa9cae86b38f7e55589b7ef2336de40eafa2e09cefd877788706683dd174ce.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5d1859585ac6afa25178da3104ac17697262f705 --- /dev/null +++ b/parse/train/CU0APx9LMaL/images/c0aa9cae86b38f7e55589b7ef2336de40eafa2e09cefd877788706683dd174ce.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:099b5fb951699a5157d456e4c6e1c80a5d9937353d1a1ce22247d31e5953c741 +size 80714 diff --git a/parse/train/CU0APx9LMaL/images/c0ccbaca6895863c9d9bf36bcfa08a7e0052cf30b1cbfb715abb577138a20acd.jpg b/parse/train/CU0APx9LMaL/images/c0ccbaca6895863c9d9bf36bcfa08a7e0052cf30b1cbfb715abb577138a20acd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4ea216d5c4201e695a2510dadb9456f1d0fb2efe --- /dev/null +++ b/parse/train/CU0APx9LMaL/images/c0ccbaca6895863c9d9bf36bcfa08a7e0052cf30b1cbfb715abb577138a20acd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:80b0ce8768fca591e58913542e6a623bda9e9860615711e09fce490442ef1767 +size 60973 diff --git a/parse/train/CU0APx9LMaL/images/d186067c75eeae1e888505f877c7e2769abdcbca921b47fe3e35c848d17cfbda.jpg b/parse/train/CU0APx9LMaL/images/d186067c75eeae1e888505f877c7e2769abdcbca921b47fe3e35c848d17cfbda.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7f83987cc04d2d4d837a1543dcf4bccbb84216ce --- /dev/null +++ b/parse/train/CU0APx9LMaL/images/d186067c75eeae1e888505f877c7e2769abdcbca921b47fe3e35c848d17cfbda.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fa295ce23a2aaa7bf1ddbd77ba6bbca3055ff6791055ab9503529bcd601539c9 +size 54092 diff --git a/parse/train/CU0APx9LMaL/images/d689f534f876402c93481a57d3d7f0f2779c26e73ff384cf57206a8d4bfda0da.jpg b/parse/train/CU0APx9LMaL/images/d689f534f876402c93481a57d3d7f0f2779c26e73ff384cf57206a8d4bfda0da.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9cc1e1dc56ea13b3c399765dcc20f72a48edea56 --- /dev/null +++ b/parse/train/CU0APx9LMaL/images/d689f534f876402c93481a57d3d7f0f2779c26e73ff384cf57206a8d4bfda0da.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee0b6bbec7b320b2a23659ec6cf46fdf62f4d457241fda50c374542f37ddd103 +size 40411 diff --git a/parse/train/CU0APx9LMaL/images/fc463be9d566d406967f817c7edad10477d1d7a40f7bac6c5148f38ff17efec1.jpg b/parse/train/CU0APx9LMaL/images/fc463be9d566d406967f817c7edad10477d1d7a40f7bac6c5148f38ff17efec1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bbbf525cad7d3323e48b48add73792c2d3c2b41e --- /dev/null +++ b/parse/train/CU0APx9LMaL/images/fc463be9d566d406967f817c7edad10477d1d7a40f7bac6c5148f38ff17efec1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:acb9cfe7443c9ac4ab039ead81cdcb3e05534d10b78b2cf5719d0fdc56569ef7 +size 20303 diff --git a/parse/train/D51irFX8UOG/images/7920639e68061430b573897ebe7641d2ca4ac915f32c7d043f64dbefee5aebd8.jpg b/parse/train/D51irFX8UOG/images/7920639e68061430b573897ebe7641d2ca4ac915f32c7d043f64dbefee5aebd8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..99f6ea6179af0b9a0db603a21790ea914e7b9926 --- /dev/null +++ b/parse/train/D51irFX8UOG/images/7920639e68061430b573897ebe7641d2ca4ac915f32c7d043f64dbefee5aebd8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b178fe01bff3260509f5489e3d859096c7cb3cacc907c4ff7593cd42a5bc53a8 +size 127229 diff --git a/parse/train/D51irFX8UOG/images/9cf7b1cc6c901c6ac0f059d5ab800e637e03f5099173829a4f075b8914f8d589.jpg b/parse/train/D51irFX8UOG/images/9cf7b1cc6c901c6ac0f059d5ab800e637e03f5099173829a4f075b8914f8d589.jpg new file mode 100644 index 0000000000000000000000000000000000000000..219d107acd6354dd1288b57a6fd826508a5ce668 --- /dev/null +++ b/parse/train/D51irFX8UOG/images/9cf7b1cc6c901c6ac0f059d5ab800e637e03f5099173829a4f075b8914f8d589.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2a59e793fc2a786e6023ac86f3a90af338c352763d04802ed672d754daef6023 +size 6973 diff --git a/parse/train/D51irFX8UOG/images/b35b9d95dfe62a7b91d54288971b68ebfcd93d44cf96f1b479beba0ba81a1227.jpg b/parse/train/D51irFX8UOG/images/b35b9d95dfe62a7b91d54288971b68ebfcd93d44cf96f1b479beba0ba81a1227.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3f8169860d912cd4fe421216a843334d8c79fc74 --- /dev/null +++ b/parse/train/D51irFX8UOG/images/b35b9d95dfe62a7b91d54288971b68ebfcd93d44cf96f1b479beba0ba81a1227.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b89a253e0c9a604feddc7cc4f4e267313796c978feeca832bfce8f591f32a21b +size 4234 diff --git a/parse/train/D51irFX8UOG/images/cd02b6c85d6f351e0911d4510ef2aab7b9033b557ebb1339a4b7aa758b6ae07d.jpg b/parse/train/D51irFX8UOG/images/cd02b6c85d6f351e0911d4510ef2aab7b9033b557ebb1339a4b7aa758b6ae07d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..397a136cb35cc04adbd7b0a626f88ffb152a43b1 --- /dev/null +++ b/parse/train/D51irFX8UOG/images/cd02b6c85d6f351e0911d4510ef2aab7b9033b557ebb1339a4b7aa758b6ae07d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:838853d543d1cfb25e7ffbd24cc96a2cc10d92e4c79bbdb11cad29ef6c0cda22 +size 88504 diff --git a/parse/train/H1xSNiRcF7/H1xSNiRcF7_middle.json b/parse/train/H1xSNiRcF7/H1xSNiRcF7_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..3ffdef05fbd94e43a012d8694e38e0b4c7345b85 --- /dev/null +++ b/parse/train/H1xSNiRcF7/H1xSNiRcF7_middle.json @@ -0,0 +1,33380 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 90, + 389, + 127 + ], + "lines": [ + { + "bbox": [ + 106, + 88, + 323, + 108 + ], + "spans": [ + { + "bbox": [ + 106, + 88, + 323, + 108 + ], + "score": 1.0, + "content": "SMOOTHING THE GEOMETRY", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 109, + 390, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 109, + 390, + 128 + ], + "score": 1.0, + "content": "OF PROBABILISTIC BOX EMBEDDINGS", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 112, + 145, + 455, + 179 + ], + "lines": [ + { + "bbox": [ + 111, + 146, + 456, + 158 + ], + "spans": [ + { + "bbox": [ + 111, + 146, + 140, + 158 + ], + "score": 1.0, + "content": "Xiang", + "type": "text" + }, + { + "bbox": [ + 140, + 146, + 155, + 157 + ], + "score": 0.58, + "content": "\\mathbf { L i } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 146, + 456, + 158 + ], + "score": 1.0, + "content": ", Luke Vilnis∗, Dongxu Zhang, Michael Boratko & Andrew McCallum", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 111, + 156, + 304, + 170 + ], + "spans": [ + { + "bbox": [ + 111, + 156, + 304, + 170 + ], + "score": 1.0, + "content": "College of Information and Computer Sciences", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 112, + 169, + 264, + 180 + ], + "spans": [ + { + "bbox": [ + 112, + 169, + 264, + 180 + ], + "score": 1.0, + "content": "University of Massachusetts Amherst", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3 + }, + { + "type": "title", + "bbox": [ + 278, + 208, + 333, + 220 + ], + "lines": [ + { + "bbox": [ + 276, + 208, + 335, + 221 + ], + "spans": [ + { + "bbox": [ + 276, + 208, + 335, + 221 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 143, + 234, + 468, + 464 + ], + "lines": [ + { + "bbox": [ + 142, + 235, + 469, + 246 + ], + "spans": [ + { + "bbox": [ + 142, + 235, + 469, + 246 + ], + "score": 1.0, + "content": "There is growing interest in geometrically-inspired embeddings for learning hier-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 142, + 245, + 469, + 257 + ], + "spans": [ + { + "bbox": [ + 142, + 245, + 469, + 257 + ], + "score": 1.0, + "content": "archies, partial orders, and lattice structures, with natural applications to transitive", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 256, + 469, + 268 + ], + "spans": [ + { + "bbox": [ + 141, + 256, + 469, + 268 + ], + "score": 1.0, + "content": "relational data such as entailment graphs. Recent work has extended these ideas", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 267, + 470, + 280 + ], + "spans": [ + { + "bbox": [ + 141, + 267, + 470, + 280 + ], + "score": 1.0, + "content": "beyond deterministic hierarchies to probabilistically calibrated models, which en-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 277, + 469, + 291 + ], + "spans": [ + { + "bbox": [ + 141, + 277, + 469, + 291 + ], + "score": 1.0, + "content": "able learning from uncertain supervision and inferring soft-inclusions among con-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 288, + 470, + 302 + ], + "spans": [ + { + "bbox": [ + 141, + 288, + 470, + 302 + ], + "score": 1.0, + "content": "cepts, while maintaining the geometric inductive bias of hierarchical embedding", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 300, + 470, + 311 + ], + "spans": [ + { + "bbox": [ + 141, + 300, + 470, + 311 + ], + "score": 1.0, + "content": "models. We build on the Box Lattice model of Vilnis et al. (2018), which showed", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 311, + 470, + 323 + ], + "spans": [ + { + "bbox": [ + 141, + 311, + 470, + 323 + ], + "score": 1.0, + "content": "promising results in modeling soft-inclusions through an overlapping hierarchy of", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 322, + 469, + 334 + ], + "spans": [ + { + "bbox": [ + 141, + 322, + 469, + 334 + ], + "score": 1.0, + "content": "sets, parameterized as high-dimensional hyperrectangles (boxes). However, the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 332, + 469, + 346 + ], + "spans": [ + { + "bbox": [ + 141, + 332, + 469, + 346 + ], + "score": 1.0, + "content": "hard edges of the boxes present difficulties for standard gradient based optimiza-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 344, + 469, + 356 + ], + "spans": [ + { + "bbox": [ + 141, + 344, + 469, + 356 + ], + "score": 1.0, + "content": "tion; that work employed a special surrogate function for the disjoint case, but", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 354, + 470, + 367 + ], + "spans": [ + { + "bbox": [ + 141, + 354, + 470, + 367 + ], + "score": 1.0, + "content": "we find this method to be fragile. In this work, we present a novel hierarchical", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 142, + 366, + 469, + 378 + ], + "spans": [ + { + "bbox": [ + 142, + 366, + 469, + 378 + ], + "score": 1.0, + "content": "embedding model, inspired by a relaxation of box embeddings into parameter-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 377, + 469, + 389 + ], + "spans": [ + { + "bbox": [ + 141, + 377, + 469, + 389 + ], + "score": 1.0, + "content": "ized density functions using Gaussian convolutions over the boxes. Our approach", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 388, + 469, + 399 + ], + "spans": [ + { + "bbox": [ + 141, + 388, + 469, + 399 + ], + "score": 1.0, + "content": "provides an alternative surrogate to the original lattice measure that improves the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 399, + 469, + 411 + ], + "spans": [ + { + "bbox": [ + 141, + 399, + 469, + 411 + ], + "score": 1.0, + "content": "robustness of optimization in the disjoint case, while also preserving the desir-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 142, + 410, + 470, + 422 + ], + "spans": [ + { + "bbox": [ + 142, + 410, + 470, + 422 + ], + "score": 1.0, + "content": "able properties with respect to the original lattice. We demonstrate increased or", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 142, + 421, + 469, + 433 + ], + "spans": [ + { + "bbox": [ + 142, + 421, + 469, + 433 + ], + "score": 1.0, + "content": "matching performance on WordNet hypernymy prediction, Flickr caption entail-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 432, + 469, + 443 + ], + "spans": [ + { + "bbox": [ + 141, + 432, + 469, + 443 + ], + "score": 1.0, + "content": "ment and a MovieLens-based market basket dataset. We show especially marked", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 142, + 443, + 469, + 455 + ], + "spans": [ + { + "bbox": [ + 142, + 443, + 469, + 455 + ], + "score": 1.0, + "content": "improvements in the case of sparse data, where many conditional probabilities", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 453, + 367, + 466 + ], + "spans": [ + { + "bbox": [ + 141, + 453, + 367, + 466 + ], + "score": 1.0, + "content": "should be low, and thus boxes should be nearly disjoint.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 16 + }, + { + "type": "title", + "bbox": [ + 108, + 487, + 206, + 500 + ], + "lines": [ + { + "bbox": [ + 105, + 486, + 208, + 503 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 208, + 503 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 512, + 505, + 578 + ], + "lines": [ + { + "bbox": [ + 105, + 511, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 505, + 525 + ], + "score": 1.0, + "content": "Embedding methods have long been a key technique in machine learning, providing a natural way", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 523, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 505, + 537 + ], + "score": 1.0, + "content": "to convert semantic problems into geometric problems. Early examples include the vector space", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 535, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 535, + 505, + 546 + ], + "score": 1.0, + "content": "(Salton et al., 1975) and latent semantic indexing (Deerwester et al., 1990) models for information", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 545, + 505, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 505, + 558 + ], + "score": 1.0, + "content": "retrieval. Embeddings experienced a renaissance after the publication of Word2Vec (Mikolov et al.,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 556, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 505, + 569 + ], + "score": 1.0, + "content": "2013), a neural word embedding method (Bengio et al., 2003; Mnih & Hinton, 2009) that could run", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 568, + 175, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 175, + 579 + ], + "score": 1.0, + "content": "at massive scale.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 30.5 + }, + { + "type": "text", + "bbox": [ + 107, + 584, + 505, + 672 + ], + "lines": [ + { + "bbox": [ + 105, + 583, + 505, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 505, + 597 + ], + "score": 1.0, + "content": "Recent years have seen an interest in structured or geometric representations. Instead of represent-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 596, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 505, + 608 + ], + "score": 1.0, + "content": "ing e.g. images, words, sentences, or knowledge base concepts with points, these methods instead", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 607, + 505, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 607, + 505, + 619 + ], + "score": 1.0, + "content": "associate them with more complex geometric structures. These objects can be density functions, as", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 616, + 506, + 631 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 506, + 631 + ], + "score": 1.0, + "content": "in Gaussian embeddings (Vilnis & McCallum, 2015; Athiwaratkun & Wilson, 2017; 2018), convex", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 628, + 505, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 505, + 642 + ], + "score": 1.0, + "content": "cones, as in order embeddings (Vendrov et al., 2016; Lai & Hockenmaier, 2017), or axis-aligned hy-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 639, + 506, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 506, + 652 + ], + "score": 1.0, + "content": "perrectangles, as in box embeddings (Vilnis et al., 2018; Subramanian & Chakrabarti, 2018). These", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 651, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 505, + 663 + ], + "score": 1.0, + "content": "geometric objects more naturally express ideas of asymmetry, entailment, ordering, and transitive", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 662, + 494, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 494, + 673 + ], + "score": 1.0, + "content": "relations than simple points in a vector space, and provide a strong inductive bias for these tasks.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 108, + 678, + 504, + 711 + ], + "lines": [ + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "score": 1.0, + "content": "In this work, we focus on the probabilistic Box Lattice model of Vilnis et al. (2018), because of its", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 690, + 505, + 702 + ], + "spans": [ + { + "bbox": [ + 106, + 690, + 505, + 702 + ], + "score": 1.0, + "content": "strong empirical performance in modeling transitive relations, probabilistic interpretation (edges in", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 104, + 699, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 104, + 699, + 505, + 713 + ], + "score": 1.0, + "content": "a relational DAG are replaced with conditional probabilities), and ability to model complex joint", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 43 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 125, + 721, + 198, + 731 + ], + "lines": [ + { + "bbox": [ + 123, + 719, + 200, + 734 + ], + "spans": [ + { + "bbox": [ + 123, + 719, + 200, + 734 + ], + "score": 1.0, + "content": "* Equal contribution.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 90, + 389, + 127 + ], + "lines": [ + { + "bbox": [ + 106, + 88, + 323, + 108 + ], + "spans": [ + { + "bbox": [ + 106, + 88, + 323, + 108 + ], + "score": 1.0, + "content": "SMOOTHING THE GEOMETRY", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 109, + 390, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 109, + 390, + 128 + ], + "score": 1.0, + "content": "OF PROBABILISTIC BOX EMBEDDINGS", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "list", + "bbox": [ + 112, + 145, + 455, + 179 + ], + "lines": [ + { + "bbox": [ + 111, + 146, + 456, + 158 + ], + "spans": [ + { + "bbox": [ + 111, + 146, + 140, + 158 + ], + "score": 1.0, + "content": "Xiang", + "type": "text" + }, + { + "bbox": [ + 140, + 146, + 155, + 157 + ], + "score": 0.58, + "content": "\\mathbf { L i } ^ { * }", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 146, + 456, + 158 + ], + "score": 1.0, + "content": ", Luke Vilnis∗, Dongxu Zhang, Michael Boratko & Andrew McCallum", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 111, + 156, + 304, + 170 + ], + "spans": [ + { + "bbox": [ + 111, + 156, + 304, + 170 + ], + "score": 1.0, + "content": "College of Information and Computer Sciences", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 112, + 169, + 264, + 180 + ], + "spans": [ + { + "bbox": [ + 112, + 169, + 264, + 180 + ], + "score": 1.0, + "content": "University of Massachusetts Amherst", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + } + ], + "index": 3, + "bbox_fs": [ + 111, + 146, + 456, + 180 + ] + }, + { + "type": "title", + "bbox": [ + 278, + 208, + 333, + 220 + ], + "lines": [ + { + "bbox": [ + 276, + 208, + 335, + 221 + ], + "spans": [ + { + "bbox": [ + 276, + 208, + 335, + 221 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 143, + 234, + 468, + 464 + ], + "lines": [ + { + "bbox": [ + 142, + 235, + 469, + 246 + ], + "spans": [ + { + "bbox": [ + 142, + 235, + 469, + 246 + ], + "score": 1.0, + "content": "There is growing interest in geometrically-inspired embeddings for learning hier-", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 142, + 245, + 469, + 257 + ], + "spans": [ + { + "bbox": [ + 142, + 245, + 469, + 257 + ], + "score": 1.0, + "content": "archies, partial orders, and lattice structures, with natural applications to transitive", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 256, + 469, + 268 + ], + "spans": [ + { + "bbox": [ + 141, + 256, + 469, + 268 + ], + "score": 1.0, + "content": "relational data such as entailment graphs. Recent work has extended these ideas", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 267, + 470, + 280 + ], + "spans": [ + { + "bbox": [ + 141, + 267, + 470, + 280 + ], + "score": 1.0, + "content": "beyond deterministic hierarchies to probabilistically calibrated models, which en-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 277, + 469, + 291 + ], + "spans": [ + { + "bbox": [ + 141, + 277, + 469, + 291 + ], + "score": 1.0, + "content": "able learning from uncertain supervision and inferring soft-inclusions among con-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 288, + 470, + 302 + ], + "spans": [ + { + "bbox": [ + 141, + 288, + 470, + 302 + ], + "score": 1.0, + "content": "cepts, while maintaining the geometric inductive bias of hierarchical embedding", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 300, + 470, + 311 + ], + "spans": [ + { + "bbox": [ + 141, + 300, + 470, + 311 + ], + "score": 1.0, + "content": "models. We build on the Box Lattice model of Vilnis et al. (2018), which showed", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 311, + 470, + 323 + ], + "spans": [ + { + "bbox": [ + 141, + 311, + 470, + 323 + ], + "score": 1.0, + "content": "promising results in modeling soft-inclusions through an overlapping hierarchy of", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 322, + 469, + 334 + ], + "spans": [ + { + "bbox": [ + 141, + 322, + 469, + 334 + ], + "score": 1.0, + "content": "sets, parameterized as high-dimensional hyperrectangles (boxes). However, the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 332, + 469, + 346 + ], + "spans": [ + { + "bbox": [ + 141, + 332, + 469, + 346 + ], + "score": 1.0, + "content": "hard edges of the boxes present difficulties for standard gradient based optimiza-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 141, + 344, + 469, + 356 + ], + "spans": [ + { + "bbox": [ + 141, + 344, + 469, + 356 + ], + "score": 1.0, + "content": "tion; that work employed a special surrogate function for the disjoint case, but", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 141, + 354, + 470, + 367 + ], + "spans": [ + { + "bbox": [ + 141, + 354, + 470, + 367 + ], + "score": 1.0, + "content": "we find this method to be fragile. In this work, we present a novel hierarchical", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 142, + 366, + 469, + 378 + ], + "spans": [ + { + "bbox": [ + 142, + 366, + 469, + 378 + ], + "score": 1.0, + "content": "embedding model, inspired by a relaxation of box embeddings into parameter-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 141, + 377, + 469, + 389 + ], + "spans": [ + { + "bbox": [ + 141, + 377, + 469, + 389 + ], + "score": 1.0, + "content": "ized density functions using Gaussian convolutions over the boxes. Our approach", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 141, + 388, + 469, + 399 + ], + "spans": [ + { + "bbox": [ + 141, + 388, + 469, + 399 + ], + "score": 1.0, + "content": "provides an alternative surrogate to the original lattice measure that improves the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 141, + 399, + 469, + 411 + ], + "spans": [ + { + "bbox": [ + 141, + 399, + 469, + 411 + ], + "score": 1.0, + "content": "robustness of optimization in the disjoint case, while also preserving the desir-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 142, + 410, + 470, + 422 + ], + "spans": [ + { + "bbox": [ + 142, + 410, + 470, + 422 + ], + "score": 1.0, + "content": "able properties with respect to the original lattice. We demonstrate increased or", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 142, + 421, + 469, + 433 + ], + "spans": [ + { + "bbox": [ + 142, + 421, + 469, + 433 + ], + "score": 1.0, + "content": "matching performance on WordNet hypernymy prediction, Flickr caption entail-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 432, + 469, + 443 + ], + "spans": [ + { + "bbox": [ + 141, + 432, + 469, + 443 + ], + "score": 1.0, + "content": "ment and a MovieLens-based market basket dataset. We show especially marked", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 142, + 443, + 469, + 455 + ], + "spans": [ + { + "bbox": [ + 142, + 443, + 469, + 455 + ], + "score": 1.0, + "content": "improvements in the case of sparse data, where many conditional probabilities", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 453, + 367, + 466 + ], + "spans": [ + { + "bbox": [ + 141, + 453, + 367, + 466 + ], + "score": 1.0, + "content": "should be low, and thus boxes should be nearly disjoint.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 16, + "bbox_fs": [ + 141, + 235, + 470, + 466 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 487, + 206, + 500 + ], + "lines": [ + { + "bbox": [ + 105, + 486, + 208, + 503 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 208, + 503 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "text", + "bbox": [ + 107, + 512, + 505, + 578 + ], + "lines": [ + { + "bbox": [ + 105, + 511, + 505, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 511, + 505, + 525 + ], + "score": 1.0, + "content": "Embedding methods have long been a key technique in machine learning, providing a natural way", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 523, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 505, + 537 + ], + "score": 1.0, + "content": "to convert semantic problems into geometric problems. Early examples include the vector space", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 535, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 106, + 535, + 505, + 546 + ], + "score": 1.0, + "content": "(Salton et al., 1975) and latent semantic indexing (Deerwester et al., 1990) models for information", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 545, + 505, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 505, + 558 + ], + "score": 1.0, + "content": "retrieval. Embeddings experienced a renaissance after the publication of Word2Vec (Mikolov et al.,", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 556, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 556, + 505, + 569 + ], + "score": 1.0, + "content": "2013), a neural word embedding method (Bengio et al., 2003; Mnih & Hinton, 2009) that could run", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 568, + 175, + 579 + ], + "spans": [ + { + "bbox": [ + 105, + 568, + 175, + 579 + ], + "score": 1.0, + "content": "at massive scale.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 511, + 505, + 579 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 584, + 505, + 672 + ], + "lines": [ + { + "bbox": [ + 105, + 583, + 505, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 505, + 597 + ], + "score": 1.0, + "content": "Recent years have seen an interest in structured or geometric representations. Instead of represent-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 596, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 505, + 608 + ], + "score": 1.0, + "content": "ing e.g. images, words, sentences, or knowledge base concepts with points, these methods instead", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 607, + 505, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 607, + 505, + 619 + ], + "score": 1.0, + "content": "associate them with more complex geometric structures. These objects can be density functions, as", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 616, + 506, + 631 + ], + "spans": [ + { + "bbox": [ + 105, + 616, + 506, + 631 + ], + "score": 1.0, + "content": "in Gaussian embeddings (Vilnis & McCallum, 2015; Athiwaratkun & Wilson, 2017; 2018), convex", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 628, + 505, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 505, + 642 + ], + "score": 1.0, + "content": "cones, as in order embeddings (Vendrov et al., 2016; Lai & Hockenmaier, 2017), or axis-aligned hy-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 639, + 506, + 652 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 506, + 652 + ], + "score": 1.0, + "content": "perrectangles, as in box embeddings (Vilnis et al., 2018; Subramanian & Chakrabarti, 2018). These", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 651, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 651, + 505, + 663 + ], + "score": 1.0, + "content": "geometric objects more naturally express ideas of asymmetry, entailment, ordering, and transitive", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 662, + 494, + 673 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 494, + 673 + ], + "score": 1.0, + "content": "relations than simple points in a vector space, and provide a strong inductive bias for these tasks.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 583, + 506, + 673 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 678, + 504, + 711 + ], + "lines": [ + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "score": 1.0, + "content": "In this work, we focus on the probabilistic Box Lattice model of Vilnis et al. (2018), because of its", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 690, + 505, + 702 + ], + "spans": [ + { + "bbox": [ + 106, + 690, + 505, + 702 + ], + "score": 1.0, + "content": "strong empirical performance in modeling transitive relations, probabilistic interpretation (edges in", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 104, + 699, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 104, + 699, + 505, + 713 + ], + "score": 1.0, + "content": "a relational DAG are replaced with conditional probabilities), and ability to model complex joint", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 83, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 505, + 95 + ], + "score": 1.0, + "content": "probability distributions including negative correlations. Box embeddings (BE) are a generalization", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "of order embeddings (OE) (Vendrov et al., 2016) and probabilistic order embeddings (POE) (Lai", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "score": 1.0, + "content": "& Hockenmaier, 2017) that replace the vector lattice ordering (notions of overlapping and enclos-", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 506, + 129 + ], + "score": 1.0, + "content": "ing convex cones) in OE and POE with a more general notion of overlapping boxes (products of", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 125, + 150, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 125, + 150, + 140 + ], + "score": 1.0, + "content": "intervals).", + "type": "text", + "cross_page": true + } + ], + "index": 4 + } + ], + "index": 43, + "bbox_fs": [ + 104, + 678, + 505, + 713 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 137 + ], + "lines": [ + { + "bbox": [ + 105, + 83, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 505, + 95 + ], + "score": 1.0, + "content": "probability distributions including negative correlations. Box embeddings (BE) are a generalization", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 94, + 505, + 106 + ], + "score": 1.0, + "content": "of order embeddings (OE) (Vendrov et al., 2016) and probabilistic order embeddings (POE) (Lai", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 105, + 104, + 506, + 117 + ], + "score": 1.0, + "content": "& Hockenmaier, 2017) that replace the vector lattice ordering (notions of overlapping and enclos-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 114, + 506, + 129 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 506, + 129 + ], + "score": 1.0, + "content": "ing convex cones) in OE and POE with a more general notion of overlapping boxes (products of", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 125, + 150, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 125, + 150, + 140 + ], + "score": 1.0, + "content": "intervals).", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 107, + 143, + 505, + 264 + ], + "lines": [ + { + "bbox": [ + 106, + 142, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 142, + 505, + 156 + ], + "score": 1.0, + "content": "While intuitively appealing, the “hard edges” of boxes and their ability to become easily disjoint,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 154, + 506, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 154, + 506, + 167 + ], + "score": 1.0, + "content": "present difficulties for gradient-based optimization: when two boxes are disjoint in the model, but", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 165, + 506, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 506, + 178 + ], + "score": 1.0, + "content": "have overlap in the ground truth, no gradient can flow to the model to correct the problem. This is of", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 176, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 506, + 189 + ], + "score": 1.0, + "content": "special concern for (pseudo-)sparse data, where many boxes should have nearly zero overlap, while", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 187, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 187, + 505, + 199 + ], + "score": 1.0, + "content": "others should have very high overlap. This is especially pronounced in the case of e.g. market basket", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 198, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 505, + 210 + ], + "score": 1.0, + "content": "models for recommendation, where most items should not be recommended, and entailment tasks,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 209, + 505, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 505, + 222 + ], + "score": 1.0, + "content": "most of which are currently artificially resampled into a 1:1 ratio of positive to negative examples.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 219, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 106, + 219, + 505, + 232 + ], + "score": 1.0, + "content": "To address the disjoint case, Vilnis et al. (2018) introduce an ad-hoc surrogate function. In contrast,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 231, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 505, + 243 + ], + "score": 1.0, + "content": "we look at this problem as inspiration for a new model, based on the intuition of relaxing the hard", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 242, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 505, + 255 + ], + "score": 1.0, + "content": "edges of the boxes into smoothed density functions, using a Gaussian convolution with the original", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 252, + 136, + 266 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 136, + 266 + ], + "score": 1.0, + "content": "boxes.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 108, + 270, + 504, + 303 + ], + "lines": [ + { + "bbox": [ + 106, + 269, + 506, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 269, + 506, + 282 + ], + "score": 1.0, + "content": "We demonstrate the superiority of our approach to modeling transitive relations on WordNet, Flickr", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 280, + 505, + 293 + ], + "spans": [ + { + "bbox": [ + 106, + 280, + 505, + 293 + ], + "score": 1.0, + "content": "caption entailment, and a MovieLens-based market basket dataset. We match or beat existing state", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 291, + 453, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 453, + 305 + ], + "score": 1.0, + "content": "of the art results, while showing substantial improvements in the pseudosparse regime.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17 + }, + { + "type": "title", + "bbox": [ + 108, + 318, + 211, + 331 + ], + "lines": [ + { + "bbox": [ + 104, + 317, + 213, + 334 + ], + "spans": [ + { + "bbox": [ + 104, + 317, + 213, + 334 + ], + "score": 1.0, + "content": "2 RELATED WORK", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 344, + 505, + 454 + ], + "lines": [ + { + "bbox": [ + 106, + 343, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 343, + 505, + 357 + ], + "score": 1.0, + "content": "As mentioned in the introduction, there is much related work on structured or geometric embeddings.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 353, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 505, + 367 + ], + "score": 1.0, + "content": "Most relevant to this work are the order embeddings of Vendrov et al. (2016), which embed a non-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 366, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 505, + 378 + ], + "score": 1.0, + "content": "probabilistic DAG or lattice in a vector space with order given by inclusion of embeddings’ forward", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 376, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 505, + 389 + ], + "score": 1.0, + "content": "cones, the probabilistic extension of that model due to Lai & Hockenmaier (2017), and the box", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 387, + 506, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 506, + 400 + ], + "score": 1.0, + "content": "lattice or box embedding model of Vilnis et al. (2018), which we extend. Concurrently to Vilnis", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 399, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 505, + 412 + ], + "score": 1.0, + "content": "et al. (2018), another hyperrectangle-based generalization of order embeddings was proposed by", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 409, + 506, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 506, + 422 + ], + "score": 1.0, + "content": "Subramanian & Chakrabarti (2018), also called box embeddings. The difference between the two", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 420, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 505, + 433 + ], + "score": 1.0, + "content": "models lies in the interpretation: the former is a probabilistic model that assigns edges conditional", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 432, + 506, + 443 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 506, + 443 + ], + "score": 1.0, + "content": "probabilities according to degrees of overlap, while the latter is a deterministic model in the style of", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 443, + 479, + 455 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 479, + 455 + ], + "score": 1.0, + "content": "order embeddings — an edge is considered present only if one box entirely encloses another.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 24.5 + }, + { + "type": "text", + "bbox": [ + 107, + 459, + 504, + 536 + ], + "lines": [ + { + "bbox": [ + 105, + 459, + 506, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 506, + 472 + ], + "score": 1.0, + "content": "Methods based on embedding points in hyperbolic space (Nickel & Kiela, 2017; Ganea et al., 2018)", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 470, + 505, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 505, + 482 + ], + "score": 1.0, + "content": "have also recently been proposed for learning hierarchical embeddings. These models, similar", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 482, + 505, + 494 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 505, + 494 + ], + "score": 1.0, + "content": "to order embeddings and the box embeddings of Subramanian & Chakrabarti (2018), are non-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 492, + 506, + 505 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 506, + 505 + ], + "score": 1.0, + "content": "probabilistic and optimize an energy function. Additionally, while the negative curvature of hy-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 504, + 506, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 506, + 516 + ], + "score": 1.0, + "content": "perbolic space is attractively biased towards learning tree structures (since distances between points", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 515, + 506, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 506, + 527 + ], + "score": 1.0, + "content": "increase the farther they are from the origin), this constant curvature makes the models not as suit-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 525, + 256, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 256, + 538 + ], + "score": 1.0, + "content": "able for learning non-treelike DAGs.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 108, + 542, + 504, + 586 + ], + "lines": [ + { + "bbox": [ + 106, + 542, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 106, + 542, + 505, + 555 + ], + "score": 1.0, + "content": "Our approach to smoothing the energy landscape of the model using Gaussian convolution is com-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 553, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 106, + 553, + 505, + 565 + ], + "score": 1.0, + "content": "mon in mollified optimization and continuation methods, and is increasingly making its way into", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 564, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 505, + 576 + ], + "score": 1.0, + "content": "machine learning models such as Mollifying Networks (Gulcehre et al., 2016b), diffusion-trained", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 576, + 434, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 576, + 434, + 587 + ], + "score": 1.0, + "content": "networks (Mobahi, 2016), and noisy activation functions (Gulcehre et al., 2016a).", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 107, + 592, + 505, + 647 + ], + "lines": [ + { + "bbox": [ + 106, + 591, + 505, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 591, + 505, + 605 + ], + "score": 1.0, + "content": "Our focus on embedding orderings and transitive relations is a subset of knowledge graph embed-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 602, + 505, + 615 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 505, + 615 + ], + "score": 1.0, + "content": "ding. While this field is very large, the main difference of our probabilistic approach is that we seek", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 613, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 505, + 627 + ], + "score": 1.0, + "content": "to learn an embedding model which maps concepts to subsets of event space, giving our model an", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 625, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 625, + 505, + 637 + ], + "score": 1.0, + "content": "inductive bias especially suited for transitive relations as well as fuzzy concepts of inclusion and", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 636, + 153, + 647 + ], + "spans": [ + { + "bbox": [ + 106, + 636, + 153, + 647 + ], + "score": 1.0, + "content": "entailment.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 43 + }, + { + "type": "title", + "bbox": [ + 108, + 662, + 200, + 676 + ], + "lines": [ + { + "bbox": [ + 104, + 661, + 201, + 678 + ], + "spans": [ + { + "bbox": [ + 104, + 661, + 201, + 678 + ], + "score": 1.0, + "content": "3 BACKGROUND", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 46 + }, + { + "type": "text", + "bbox": [ + 108, + 687, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "We begin with a brief overview of two methods for representing ontologies as geometric objects.", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "First, we review some definitions from order theory, a useful formalism for describing ontologies,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "then we introduce the vector and box lattices. Figure 1 shows a simple two-dimensional example of", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 720, + 195, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 195, + 733 + ], + "score": 1.0, + "content": "these representations.", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 48.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 763 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 137 + ], + "lines": [], + "index": 2, + "bbox_fs": [ + 105, + 83, + 506, + 140 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 143, + 505, + 264 + ], + "lines": [ + { + "bbox": [ + 106, + 142, + 505, + 156 + ], + "spans": [ + { + "bbox": [ + 106, + 142, + 505, + 156 + ], + "score": 1.0, + "content": "While intuitively appealing, the “hard edges” of boxes and their ability to become easily disjoint,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 154, + 506, + 167 + ], + "spans": [ + { + "bbox": [ + 105, + 154, + 506, + 167 + ], + "score": 1.0, + "content": "present difficulties for gradient-based optimization: when two boxes are disjoint in the model, but", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 165, + 506, + 178 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 506, + 178 + ], + "score": 1.0, + "content": "have overlap in the ground truth, no gradient can flow to the model to correct the problem. This is of", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 176, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 105, + 176, + 506, + 189 + ], + "score": 1.0, + "content": "special concern for (pseudo-)sparse data, where many boxes should have nearly zero overlap, while", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 187, + 505, + 199 + ], + "spans": [ + { + "bbox": [ + 105, + 187, + 505, + 199 + ], + "score": 1.0, + "content": "others should have very high overlap. This is especially pronounced in the case of e.g. market basket", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 198, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 105, + 198, + 505, + 210 + ], + "score": 1.0, + "content": "models for recommendation, where most items should not be recommended, and entailment tasks,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 209, + 505, + 222 + ], + "spans": [ + { + "bbox": [ + 105, + 209, + 505, + 222 + ], + "score": 1.0, + "content": "most of which are currently artificially resampled into a 1:1 ratio of positive to negative examples.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 219, + 505, + 232 + ], + "spans": [ + { + "bbox": [ + 106, + 219, + 505, + 232 + ], + "score": 1.0, + "content": "To address the disjoint case, Vilnis et al. (2018) introduce an ad-hoc surrogate function. In contrast,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 231, + 505, + 243 + ], + "spans": [ + { + "bbox": [ + 106, + 231, + 505, + 243 + ], + "score": 1.0, + "content": "we look at this problem as inspiration for a new model, based on the intuition of relaxing the hard", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 242, + 505, + 255 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 505, + 255 + ], + "score": 1.0, + "content": "edges of the boxes into smoothed density functions, using a Gaussian convolution with the original", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 252, + 136, + 266 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 136, + 266 + ], + "score": 1.0, + "content": "boxes.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 142, + 506, + 266 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 270, + 504, + 303 + ], + "lines": [ + { + "bbox": [ + 106, + 269, + 506, + 282 + ], + "spans": [ + { + "bbox": [ + 106, + 269, + 506, + 282 + ], + "score": 1.0, + "content": "We demonstrate the superiority of our approach to modeling transitive relations on WordNet, Flickr", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 280, + 505, + 293 + ], + "spans": [ + { + "bbox": [ + 106, + 280, + 505, + 293 + ], + "score": 1.0, + "content": "caption entailment, and a MovieLens-based market basket dataset. We match or beat existing state", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 291, + 453, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 453, + 305 + ], + "score": 1.0, + "content": "of the art results, while showing substantial improvements in the pseudosparse regime.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 269, + 506, + 305 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 318, + 211, + 331 + ], + "lines": [ + { + "bbox": [ + 104, + 317, + 213, + 334 + ], + "spans": [ + { + "bbox": [ + 104, + 317, + 213, + 334 + ], + "score": 1.0, + "content": "2 RELATED WORK", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 107, + 344, + 505, + 454 + ], + "lines": [ + { + "bbox": [ + 106, + 343, + 505, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 343, + 505, + 357 + ], + "score": 1.0, + "content": "As mentioned in the introduction, there is much related work on structured or geometric embeddings.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 353, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 505, + 367 + ], + "score": 1.0, + "content": "Most relevant to this work are the order embeddings of Vendrov et al. (2016), which embed a non-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 366, + 505, + 378 + ], + "spans": [ + { + "bbox": [ + 105, + 366, + 505, + 378 + ], + "score": 1.0, + "content": "probabilistic DAG or lattice in a vector space with order given by inclusion of embeddings’ forward", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 376, + 505, + 389 + ], + "spans": [ + { + "bbox": [ + 105, + 376, + 505, + 389 + ], + "score": 1.0, + "content": "cones, the probabilistic extension of that model due to Lai & Hockenmaier (2017), and the box", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 387, + 506, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 387, + 506, + 400 + ], + "score": 1.0, + "content": "lattice or box embedding model of Vilnis et al. (2018), which we extend. Concurrently to Vilnis", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 399, + 505, + 412 + ], + "spans": [ + { + "bbox": [ + 105, + 399, + 505, + 412 + ], + "score": 1.0, + "content": "et al. (2018), another hyperrectangle-based generalization of order embeddings was proposed by", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 409, + 506, + 422 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 506, + 422 + ], + "score": 1.0, + "content": "Subramanian & Chakrabarti (2018), also called box embeddings. The difference between the two", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 420, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 505, + 433 + ], + "score": 1.0, + "content": "models lies in the interpretation: the former is a probabilistic model that assigns edges conditional", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 432, + 506, + 443 + ], + "spans": [ + { + "bbox": [ + 105, + 432, + 506, + 443 + ], + "score": 1.0, + "content": "probabilities according to degrees of overlap, while the latter is a deterministic model in the style of", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 443, + 479, + 455 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 479, + 455 + ], + "score": 1.0, + "content": "order embeddings — an edge is considered present only if one box entirely encloses another.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 343, + 506, + 455 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 459, + 504, + 536 + ], + "lines": [ + { + "bbox": [ + 105, + 459, + 506, + 472 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 506, + 472 + ], + "score": 1.0, + "content": "Methods based on embedding points in hyperbolic space (Nickel & Kiela, 2017; Ganea et al., 2018)", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 470, + 505, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 505, + 482 + ], + "score": 1.0, + "content": "have also recently been proposed for learning hierarchical embeddings. These models, similar", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 482, + 505, + 494 + ], + "spans": [ + { + "bbox": [ + 105, + 482, + 505, + 494 + ], + "score": 1.0, + "content": "to order embeddings and the box embeddings of Subramanian & Chakrabarti (2018), are non-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 492, + 506, + 505 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 506, + 505 + ], + "score": 1.0, + "content": "probabilistic and optimize an energy function. Additionally, while the negative curvature of hy-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 504, + 506, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 506, + 516 + ], + "score": 1.0, + "content": "perbolic space is attractively biased towards learning tree structures (since distances between points", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 515, + 506, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 515, + 506, + 527 + ], + "score": 1.0, + "content": "increase the farther they are from the origin), this constant curvature makes the models not as suit-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 525, + 256, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 525, + 256, + 538 + ], + "score": 1.0, + "content": "able for learning non-treelike DAGs.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 459, + 506, + 538 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 542, + 504, + 586 + ], + "lines": [ + { + "bbox": [ + 106, + 542, + 505, + 555 + ], + "spans": [ + { + "bbox": [ + 106, + 542, + 505, + 555 + ], + "score": 1.0, + "content": "Our approach to smoothing the energy landscape of the model using Gaussian convolution is com-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 553, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 106, + 553, + 505, + 565 + ], + "score": 1.0, + "content": "mon in mollified optimization and continuation methods, and is increasingly making its way into", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 564, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 505, + 576 + ], + "score": 1.0, + "content": "machine learning models such as Mollifying Networks (Gulcehre et al., 2016b), diffusion-trained", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 576, + 434, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 576, + 434, + 587 + ], + "score": 1.0, + "content": "networks (Mobahi, 2016), and noisy activation functions (Gulcehre et al., 2016a).", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 542, + 505, + 587 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 592, + 505, + 647 + ], + "lines": [ + { + "bbox": [ + 106, + 591, + 505, + 605 + ], + "spans": [ + { + "bbox": [ + 106, + 591, + 505, + 605 + ], + "score": 1.0, + "content": "Our focus on embedding orderings and transitive relations is a subset of knowledge graph embed-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 602, + 505, + 615 + ], + "spans": [ + { + "bbox": [ + 105, + 602, + 505, + 615 + ], + "score": 1.0, + "content": "ding. While this field is very large, the main difference of our probabilistic approach is that we seek", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 613, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 505, + 627 + ], + "score": 1.0, + "content": "to learn an embedding model which maps concepts to subsets of event space, giving our model an", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 625, + 505, + 637 + ], + "spans": [ + { + "bbox": [ + 106, + 625, + 505, + 637 + ], + "score": 1.0, + "content": "inductive bias especially suited for transitive relations as well as fuzzy concepts of inclusion and", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 636, + 153, + 647 + ], + "spans": [ + { + "bbox": [ + 106, + 636, + 153, + 647 + ], + "score": 1.0, + "content": "entailment.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 43, + "bbox_fs": [ + 105, + 591, + 505, + 647 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 662, + 200, + 676 + ], + "lines": [ + { + "bbox": [ + 104, + 661, + 201, + 678 + ], + "spans": [ + { + "bbox": [ + 104, + 661, + 201, + 678 + ], + "score": 1.0, + "content": "3 BACKGROUND", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 46 + }, + { + "type": "text", + "bbox": [ + 108, + 687, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "We begin with a brief overview of two methods for representing ontologies as geometric objects.", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 506, + 712 + ], + "score": 1.0, + "content": "First, we review some definitions from order theory, a useful formalism for describing ontologies,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 506, + 722 + ], + "score": 1.0, + "content": "then we introduce the vector and box lattices. Figure 1 shows a simple two-dimensional example of", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 720, + 195, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 195, + 733 + ], + "score": 1.0, + "content": "these representations.", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 48.5, + "bbox_fs": [ + 105, + 687, + 506, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 129, + 85, + 493, + 209 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 129, + 85, + 493, + 209 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 129, + 85, + 493, + 209 + ], + "spans": [ + { + "bbox": [ + 129, + 85, + 493, + 209 + ], + "score": 0.954, + "type": "image", + "image_path": "e33c5ed448bf97a33daa06440935051df615504b85b184de3319c90014042b76.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 129, + 85, + 493, + 126.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 129, + 126.33333333333334, + 493, + 167.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 129, + 167.66666666666669, + 493, + 209.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 218, + 505, + 252 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 217, + 505, + 231 + ], + "spans": [ + { + "bbox": [ + 105, + 217, + 505, + 231 + ], + "score": 1.0, + "content": "Figure 1: Comparison between the Order Embedding (vector lattice) and Box Embedding represen-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 229, + 504, + 241 + ], + "spans": [ + { + "bbox": [ + 105, + 229, + 504, + 241 + ], + "score": 1.0, + "content": "tations for a simple ontology. Regions represent concepts and overlaps represent their entailment.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 240, + 316, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 240, + 316, + 253 + ], + "score": 1.0, + "content": "Shading represents density in the probabilistic case.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 108, + 270, + 273, + 281 + ], + "lines": [ + { + "bbox": [ + 106, + 270, + 274, + 283 + ], + "spans": [ + { + "bbox": [ + 106, + 270, + 274, + 283 + ], + "score": 1.0, + "content": "3.1 PARTIAL ORDERS AND LATTICES", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 105, + 290, + 504, + 314 + ], + "lines": [ + { + "bbox": [ + 105, + 290, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 305, + 304 + ], + "score": 1.0, + "content": "A non-strict partially ordered set (poset) is a pair", + "type": "text" + }, + { + "bbox": [ + 305, + 291, + 326, + 302 + ], + "score": 0.83, + "content": "P , \\preceq", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 290, + 356, + 304 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 356, + 291, + 365, + 301 + ], + "score": 0.84, + "content": "P", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 290, + 415, + 304 + ], + "score": 1.0, + "content": "is a set, and", + "type": "text" + }, + { + "bbox": [ + 415, + 292, + 425, + 302 + ], + "score": 0.8, + "content": "\\preceq", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 290, + 505, + 304 + ], + "score": 1.0, + "content": "is a binary relation.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 301, + 183, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 301, + 135, + 315 + ], + "score": 1.0, + "content": "For all", + "type": "text" + }, + { + "bbox": [ + 136, + 302, + 179, + 314 + ], + "score": 0.93, + "content": "a , b , c \\in P", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 301, + 183, + 315 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 142, + 321, + 311, + 362 + ], + "lines": [ + { + "bbox": [ + 141, + 321, + 219, + 334 + ], + "spans": [ + { + "bbox": [ + 141, + 321, + 193, + 334 + ], + "score": 1.0, + "content": "Reflexivity:", + "type": "text" + }, + { + "bbox": [ + 194, + 322, + 219, + 333 + ], + "score": 0.88, + "content": "a \\preceq a", + "type": "inline_equation" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 335, + 309, + 348 + ], + "spans": [ + { + "bbox": [ + 141, + 335, + 209, + 348 + ], + "score": 1.0, + "content": "Antisymmetry:", + "type": "text" + }, + { + "bbox": [ + 210, + 336, + 252, + 347 + ], + "score": 0.89, + "content": "a \\preceq b \\preceq a", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 335, + 285, + 348 + ], + "score": 1.0, + "content": "implies", + "type": "text" + }, + { + "bbox": [ + 286, + 336, + 309, + 345 + ], + "score": 0.85, + "content": "a = b", + "type": "inline_equation" + } + ], + "index": 10 + }, + { + "bbox": [ + 142, + 349, + 297, + 363 + ], + "spans": [ + { + "bbox": [ + 142, + 349, + 197, + 363 + ], + "score": 1.0, + "content": "Transitivity:", + "type": "text" + }, + { + "bbox": [ + 198, + 350, + 240, + 361 + ], + "score": 0.87, + "content": "a \\preceq b \\preceq c", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 349, + 272, + 363 + ], + "score": 1.0, + "content": "implies", + "type": "text" + }, + { + "bbox": [ + 272, + 351, + 297, + 361 + ], + "score": 0.72, + "content": "a \\preceq c", + "type": "inline_equation" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 108, + 369, + 504, + 403 + ], + "lines": [ + { + "bbox": [ + 106, + 369, + 505, + 381 + ], + "spans": [ + { + "bbox": [ + 106, + 369, + 505, + 381 + ], + "score": 1.0, + "content": "This generalizes the standard concept of a totally ordered set to allow some elements to be incom-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 380, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 505, + 394 + ], + "score": 1.0, + "content": "parable. Posets provide a good formalism for the kind of acyclic directed graph data found in many", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 107, + 392, + 275, + 403 + ], + "spans": [ + { + "bbox": [ + 107, + 392, + 275, + 403 + ], + "score": 1.0, + "content": "knowledge bases with transitive relations.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 408, + 504, + 442 + ], + "lines": [ + { + "bbox": [ + 105, + 407, + 506, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 506, + 421 + ], + "score": 1.0, + "content": "A lattice is a poset where any subset of elements has a single unique least upper bound, and greatest", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 420, + 504, + 432 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 286, + 432 + ], + "score": 1.0, + "content": "lower bound. In a bounded lattice, the set", + "type": "text" + }, + { + "bbox": [ + 287, + 420, + 296, + 429 + ], + "score": 0.79, + "content": "P", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 420, + 439, + 432 + ], + "score": 1.0, + "content": "contains two additional elements,", + "type": "text" + }, + { + "bbox": [ + 439, + 420, + 471, + 431 + ], + "score": 0.72, + "content": "{ \\mathsf { T } } \\left( t o p \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 420, + 495, + 432 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 495, + 420, + 504, + 430 + ], + "score": 0.56, + "content": "\\perp", + "type": "inline_equation" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 430, + 460, + 443 + ], + "spans": [ + { + "bbox": [ + 106, + 430, + 460, + 443 + ], + "score": 1.0, + "content": "(bottom), which denote the least upper bound and greatest lower bound of the entire set.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 108, + 447, + 505, + 480 + ], + "lines": [ + { + "bbox": [ + 105, + 446, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 300, + 460 + ], + "score": 1.0, + "content": "A lattice is equipped with two binary operations,", + "type": "text" + }, + { + "bbox": [ + 300, + 448, + 309, + 458 + ], + "score": 0.35, + "content": "\\vee", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 446, + 352, + 460 + ], + "score": 1.0, + "content": "(join), and", + "type": "text" + }, + { + "bbox": [ + 353, + 448, + 361, + 457 + ], + "score": 0.39, + "content": "\\wedge", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 446, + 392, + 460 + ], + "score": 1.0, + "content": "(meet).", + "type": "text" + }, + { + "bbox": [ + 392, + 447, + 412, + 458 + ], + "score": 0.62, + "content": "a \\lor b", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 446, + 505, + 460 + ], + "score": 1.0, + "content": "denotes the least upper", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 458, + 505, + 470 + ], + "spans": [ + { + "bbox": [ + 106, + 458, + 146, + 470 + ], + "score": 1.0, + "content": "bound of", + "type": "text" + }, + { + "bbox": [ + 146, + 459, + 184, + 469 + ], + "score": 0.92, + "content": "a , b \\in P", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 458, + 206, + 470 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 206, + 459, + 230, + 469 + ], + "score": 0.87, + "content": "a \\wedge b", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 458, + 505, + 470 + ], + "score": 1.0, + "content": "denotes their greatest lower bound. A bounded lattice must satisfy", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 469, + 174, + 482 + ], + "spans": [ + { + "bbox": [ + 106, + 469, + 174, + 482 + ], + "score": 1.0, + "content": "these properties:", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 141, + 489, + 419, + 558 + ], + "lines": [ + { + "bbox": [ + 142, + 489, + 280, + 500 + ], + "spans": [ + { + "bbox": [ + 142, + 489, + 203, + 500 + ], + "score": 1.0, + "content": "Idempotency:", + "type": "text" + }, + { + "bbox": [ + 203, + 489, + 280, + 499 + ], + "score": 0.64, + "content": "a \\wedge a = a \\vee a = a", + "type": "inline_equation" + } + ], + "index": 21 + }, + { + "bbox": [ + 143, + 503, + 343, + 514 + ], + "spans": [ + { + "bbox": [ + 143, + 503, + 213, + 514 + ], + "score": 1.0, + "content": "Commutativity:", + "type": "text" + }, + { + "bbox": [ + 213, + 503, + 269, + 514 + ], + "score": 0.9, + "content": "a \\wedge b = b \\wedge a", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 503, + 286, + 514 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 287, + 503, + 343, + 514 + ], + "score": 0.85, + "content": "a \\vee b = b \\vee a", + "type": "inline_equation" + } + ], + "index": 22 + }, + { + "bbox": [ + 142, + 516, + 417, + 529 + ], + "spans": [ + { + "bbox": [ + 142, + 516, + 202, + 529 + ], + "score": 1.0, + "content": "Associativity:", + "type": "text" + }, + { + "bbox": [ + 203, + 517, + 297, + 529 + ], + "score": 0.88, + "content": "a \\wedge b \\wedge c = a \\wedge ( b \\wedge c )", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 516, + 315, + 529 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 315, + 516, + 417, + 529 + ], + "score": 0.85, + "content": "( a \\lor b \\lor c ) = a \\lor ( b \\lor c )", + "type": "inline_equation" + } + ], + "index": 23 + }, + { + "bbox": [ + 141, + 531, + 344, + 543 + ], + "spans": [ + { + "bbox": [ + 141, + 531, + 196, + 543 + ], + "score": 1.0, + "content": "Absorption:", + "type": "text" + }, + { + "bbox": [ + 196, + 531, + 261, + 543 + ], + "score": 0.93, + "content": "a \\vee ( a \\wedge b ) = a", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 531, + 279, + 543 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 279, + 531, + 344, + 543 + ], + "score": 0.89, + "content": "a \\wedge ( a \\vee b ) = a", + "type": "inline_equation" + } + ], + "index": 24 + }, + { + "bbox": [ + 141, + 545, + 235, + 557 + ], + "spans": [ + { + "bbox": [ + 141, + 545, + 186, + 557 + ], + "score": 1.0, + "content": "Bounded:", + "type": "text" + }, + { + "bbox": [ + 187, + 545, + 235, + 557 + ], + "score": 0.71, + "content": "\\perp \\preceq a \\preceq \\top", + "type": "inline_equation" + } + ], + "index": 25 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 108, + 565, + 504, + 610 + ], + "lines": [ + { + "bbox": [ + 105, + 565, + 505, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 258, + 578 + ], + "score": 1.0, + "content": "Note that the extended real numbers,", + "type": "text" + }, + { + "bbox": [ + 258, + 565, + 320, + 577 + ], + "score": 0.93, + "content": "\\mathbb { R } \\cup \\{ - \\infty , \\infty \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 565, + 505, + 578 + ], + "score": 1.0, + "content": ", form a bounded lattice (and in fact, a totally", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 577, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 347, + 588 + ], + "score": 1.0, + "content": "ordered set) under the min and max operations as the meet", + "type": "text" + }, + { + "bbox": [ + 347, + 577, + 361, + 587 + ], + "score": 0.65, + "content": "( \\wedge )", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 577, + 398, + 588 + ], + "score": 1.0, + "content": "and join", + "type": "text" + }, + { + "bbox": [ + 399, + 577, + 412, + 587 + ], + "score": 0.38, + "content": "( \\vee )", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 577, + 505, + 588 + ], + "score": 1.0, + "content": "operations. So do sets", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 587, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 251, + 599 + ], + "score": 1.0, + "content": "partially ordered by inclusion, with", + "type": "text" + }, + { + "bbox": [ + 251, + 588, + 259, + 597 + ], + "score": 0.63, + "content": "\\cap", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 587, + 277, + 599 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 278, + 588, + 286, + 597 + ], + "score": 0.36, + "content": "\\cup", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 587, + 298, + 599 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 298, + 588, + 307, + 597 + ], + "score": 0.78, + "content": "\\wedge", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 587, + 325, + 599 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 325, + 588, + 333, + 597 + ], + "score": 0.59, + "content": "\\vee", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 587, + 505, + 599 + ], + "score": 1.0, + "content": ". Thinking of these special cases gives the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 598, + 284, + 611 + ], + "spans": [ + { + "bbox": [ + 106, + 598, + 284, + 611 + ], + "score": 1.0, + "content": "intuition for the fourth property, absorption.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 106, + 614, + 505, + 648 + ], + "lines": [ + { + "bbox": [ + 105, + 614, + 504, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 123, + 628 + ], + "score": 1.0, + "content": "The", + "type": "text" + }, + { + "bbox": [ + 124, + 616, + 132, + 625 + ], + "score": 0.72, + "content": "\\wedge", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 614, + 149, + 628 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 149, + 616, + 157, + 625 + ], + "score": 0.26, + "content": "\\vee", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 614, + 420, + 628 + ], + "score": 1.0, + "content": "operations can be swapped, along with reversing the poset relation", + "type": "text" + }, + { + "bbox": [ + 420, + 616, + 429, + 626 + ], + "score": 0.77, + "content": "\\preceq", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 614, + 504, + 628 + ], + "score": 1.0, + "content": ", to give a valid lat-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 625, + 505, + 638 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 505, + 638 + ], + "score": 1.0, + "content": "tice, called the dual lattice. In the real numbers this just corresponds to a sign change. A semilattice", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 637, + 254, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 637, + 254, + 648 + ], + "score": 1.0, + "content": "has only a meet or join, but not both.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 106, + 653, + 504, + 676 + ], + "lines": [ + { + "bbox": [ + 106, + 654, + 505, + 665 + ], + "spans": [ + { + "bbox": [ + 106, + 654, + 394, + 665 + ], + "score": 1.0, + "content": "Note. In the rest of the paper, when the context is clear, we will also use", + "type": "text" + }, + { + "bbox": [ + 395, + 655, + 403, + 664 + ], + "score": 0.73, + "content": "\\wedge", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 654, + 421, + 665 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 421, + 655, + 429, + 664 + ], + "score": 0.3, + "content": "\\vee", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 654, + 505, + 665 + ], + "score": 1.0, + "content": "to denote min and", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 664, + 393, + 676 + ], + "spans": [ + { + "bbox": [ + 105, + 664, + 393, + 676 + ], + "score": 1.0, + "content": "max of real numbers, in order to clarify the intuition behind our model.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "title", + "bbox": [ + 107, + 689, + 208, + 700 + ], + "lines": [ + { + "bbox": [ + 105, + 688, + 209, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 209, + 701 + ], + "score": 1.0, + "content": "3.2 VECTOR LATTICE", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "A vector lattice, also known as a Riesz space (Zaanen, 1997), or Hilbert lattice when the accompa-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 720, + 469, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 469, + 733 + ], + "score": 1.0, + "content": "nying vector space has an inner product, is a vector space endowed with a lattice structure.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 301, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 129, + 85, + 493, + 209 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 129, + 85, + 493, + 209 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 129, + 85, + 493, + 209 + ], + "spans": [ + { + "bbox": [ + 129, + 85, + 493, + 209 + ], + "score": 0.954, + "type": "image", + "image_path": "e33c5ed448bf97a33daa06440935051df615504b85b184de3319c90014042b76.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 129, + 85, + 493, + 126.33333333333334 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 129, + 126.33333333333334, + 493, + 167.66666666666669 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 129, + 167.66666666666669, + 493, + 209.00000000000003 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 218, + 505, + 252 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 217, + 505, + 231 + ], + "spans": [ + { + "bbox": [ + 105, + 217, + 505, + 231 + ], + "score": 1.0, + "content": "Figure 1: Comparison between the Order Embedding (vector lattice) and Box Embedding represen-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 229, + 504, + 241 + ], + "spans": [ + { + "bbox": [ + 105, + 229, + 504, + 241 + ], + "score": 1.0, + "content": "tations for a simple ontology. Regions represent concepts and overlaps represent their entailment.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 240, + 316, + 253 + ], + "spans": [ + { + "bbox": [ + 106, + 240, + 316, + 253 + ], + "score": 1.0, + "content": "Shading represents density in the probabilistic case.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 108, + 270, + 273, + 281 + ], + "lines": [ + { + "bbox": [ + 106, + 270, + 274, + 283 + ], + "spans": [ + { + "bbox": [ + 106, + 270, + 274, + 283 + ], + "score": 1.0, + "content": "3.1 PARTIAL ORDERS AND LATTICES", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 105, + 290, + 504, + 314 + ], + "lines": [ + { + "bbox": [ + 105, + 290, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 305, + 304 + ], + "score": 1.0, + "content": "A non-strict partially ordered set (poset) is a pair", + "type": "text" + }, + { + "bbox": [ + 305, + 291, + 326, + 302 + ], + "score": 0.83, + "content": "P , \\preceq", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 290, + 356, + 304 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 356, + 291, + 365, + 301 + ], + "score": 0.84, + "content": "P", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 290, + 415, + 304 + ], + "score": 1.0, + "content": "is a set, and", + "type": "text" + }, + { + "bbox": [ + 415, + 292, + 425, + 302 + ], + "score": 0.8, + "content": "\\preceq", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 290, + 505, + 304 + ], + "score": 1.0, + "content": "is a binary relation.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 301, + 183, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 301, + 135, + 315 + ], + "score": 1.0, + "content": "For all", + "type": "text" + }, + { + "bbox": [ + 136, + 302, + 179, + 314 + ], + "score": 0.93, + "content": "a , b , c \\in P", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 301, + 183, + 315 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 105, + 290, + 505, + 315 + ] + }, + { + "type": "text", + "bbox": [ + 142, + 321, + 311, + 362 + ], + "lines": [ + { + "bbox": [ + 141, + 321, + 219, + 334 + ], + "spans": [ + { + "bbox": [ + 141, + 321, + 193, + 334 + ], + "score": 1.0, + "content": "Reflexivity:", + "type": "text" + }, + { + "bbox": [ + 194, + 322, + 219, + 333 + ], + "score": 0.88, + "content": "a \\preceq a", + "type": "inline_equation" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 335, + 309, + 348 + ], + "spans": [ + { + "bbox": [ + 141, + 335, + 209, + 348 + ], + "score": 1.0, + "content": "Antisymmetry:", + "type": "text" + }, + { + "bbox": [ + 210, + 336, + 252, + 347 + ], + "score": 0.89, + "content": "a \\preceq b \\preceq a", + "type": "inline_equation" + }, + { + "bbox": [ + 253, + 335, + 285, + 348 + ], + "score": 1.0, + "content": "implies", + "type": "text" + }, + { + "bbox": [ + 286, + 336, + 309, + 345 + ], + "score": 0.85, + "content": "a = b", + "type": "inline_equation" + } + ], + "index": 10 + }, + { + "bbox": [ + 142, + 349, + 297, + 363 + ], + "spans": [ + { + "bbox": [ + 142, + 349, + 197, + 363 + ], + "score": 1.0, + "content": "Transitivity:", + "type": "text" + }, + { + "bbox": [ + 198, + 350, + 240, + 361 + ], + "score": 0.87, + "content": "a \\preceq b \\preceq c", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 349, + 272, + 363 + ], + "score": 1.0, + "content": "implies", + "type": "text" + }, + { + "bbox": [ + 272, + 351, + 297, + 361 + ], + "score": 0.72, + "content": "a \\preceq c", + "type": "inline_equation" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 141, + 321, + 309, + 363 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 369, + 504, + 403 + ], + "lines": [ + { + "bbox": [ + 106, + 369, + 505, + 381 + ], + "spans": [ + { + "bbox": [ + 106, + 369, + 505, + 381 + ], + "score": 1.0, + "content": "This generalizes the standard concept of a totally ordered set to allow some elements to be incom-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 380, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 505, + 394 + ], + "score": 1.0, + "content": "parable. Posets provide a good formalism for the kind of acyclic directed graph data found in many", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 107, + 392, + 275, + 403 + ], + "spans": [ + { + "bbox": [ + 107, + 392, + 275, + 403 + ], + "score": 1.0, + "content": "knowledge bases with transitive relations.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13, + "bbox_fs": [ + 106, + 369, + 505, + 403 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 408, + 504, + 442 + ], + "lines": [ + { + "bbox": [ + 105, + 407, + 506, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 407, + 506, + 421 + ], + "score": 1.0, + "content": "A lattice is a poset where any subset of elements has a single unique least upper bound, and greatest", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 420, + 504, + 432 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 286, + 432 + ], + "score": 1.0, + "content": "lower bound. In a bounded lattice, the set", + "type": "text" + }, + { + "bbox": [ + 287, + 420, + 296, + 429 + ], + "score": 0.79, + "content": "P", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 420, + 439, + 432 + ], + "score": 1.0, + "content": "contains two additional elements,", + "type": "text" + }, + { + "bbox": [ + 439, + 420, + 471, + 431 + ], + "score": 0.72, + "content": "{ \\mathsf { T } } \\left( t o p \\right)", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 420, + 495, + 432 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 495, + 420, + 504, + 430 + ], + "score": 0.56, + "content": "\\perp", + "type": "inline_equation" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 430, + 460, + 443 + ], + "spans": [ + { + "bbox": [ + 106, + 430, + 460, + 443 + ], + "score": 1.0, + "content": "(bottom), which denote the least upper bound and greatest lower bound of the entire set.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 407, + 506, + 443 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 447, + 505, + 480 + ], + "lines": [ + { + "bbox": [ + 105, + 446, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 300, + 460 + ], + "score": 1.0, + "content": "A lattice is equipped with two binary operations,", + "type": "text" + }, + { + "bbox": [ + 300, + 448, + 309, + 458 + ], + "score": 0.35, + "content": "\\vee", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 446, + 352, + 460 + ], + "score": 1.0, + "content": "(join), and", + "type": "text" + }, + { + "bbox": [ + 353, + 448, + 361, + 457 + ], + "score": 0.39, + "content": "\\wedge", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 446, + 392, + 460 + ], + "score": 1.0, + "content": "(meet).", + "type": "text" + }, + { + "bbox": [ + 392, + 447, + 412, + 458 + ], + "score": 0.62, + "content": "a \\lor b", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 446, + 505, + 460 + ], + "score": 1.0, + "content": "denotes the least upper", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 458, + 505, + 470 + ], + "spans": [ + { + "bbox": [ + 106, + 458, + 146, + 470 + ], + "score": 1.0, + "content": "bound of", + "type": "text" + }, + { + "bbox": [ + 146, + 459, + 184, + 469 + ], + "score": 0.92, + "content": "a , b \\in P", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 458, + 206, + 470 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 206, + 459, + 230, + 469 + ], + "score": 0.87, + "content": "a \\wedge b", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 458, + 505, + 470 + ], + "score": 1.0, + "content": "denotes their greatest lower bound. A bounded lattice must satisfy", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 469, + 174, + 482 + ], + "spans": [ + { + "bbox": [ + 106, + 469, + 174, + 482 + ], + "score": 1.0, + "content": "these properties:", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 446, + 505, + 482 + ] + }, + { + "type": "list", + "bbox": [ + 141, + 489, + 419, + 558 + ], + "lines": [ + { + "bbox": [ + 142, + 489, + 280, + 500 + ], + "spans": [ + { + "bbox": [ + 142, + 489, + 203, + 500 + ], + "score": 1.0, + "content": "Idempotency:", + "type": "text" + }, + { + "bbox": [ + 203, + 489, + 280, + 499 + ], + "score": 0.64, + "content": "a \\wedge a = a \\vee a = a", + "type": "inline_equation" + } + ], + "index": 21, + "is_list_start_line": true + }, + { + "bbox": [ + 143, + 503, + 343, + 514 + ], + "spans": [ + { + "bbox": [ + 143, + 503, + 213, + 514 + ], + "score": 1.0, + "content": "Commutativity:", + "type": "text" + }, + { + "bbox": [ + 213, + 503, + 269, + 514 + ], + "score": 0.9, + "content": "a \\wedge b = b \\wedge a", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 503, + 286, + 514 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 287, + 503, + 343, + 514 + ], + "score": 0.85, + "content": "a \\vee b = b \\vee a", + "type": "inline_equation" + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 142, + 516, + 417, + 529 + ], + "spans": [ + { + "bbox": [ + 142, + 516, + 202, + 529 + ], + "score": 1.0, + "content": "Associativity:", + "type": "text" + }, + { + "bbox": [ + 203, + 517, + 297, + 529 + ], + "score": 0.88, + "content": "a \\wedge b \\wedge c = a \\wedge ( b \\wedge c )", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 516, + 315, + 529 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 315, + 516, + 417, + 529 + ], + "score": 0.85, + "content": "( a \\lor b \\lor c ) = a \\lor ( b \\lor c )", + "type": "inline_equation" + } + ], + "index": 23, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 531, + 344, + 543 + ], + "spans": [ + { + "bbox": [ + 141, + 531, + 196, + 543 + ], + "score": 1.0, + "content": "Absorption:", + "type": "text" + }, + { + "bbox": [ + 196, + 531, + 261, + 543 + ], + "score": 0.93, + "content": "a \\vee ( a \\wedge b ) = a", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 531, + 279, + 543 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 279, + 531, + 344, + 543 + ], + "score": 0.89, + "content": "a \\wedge ( a \\vee b ) = a", + "type": "inline_equation" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 545, + 235, + 557 + ], + "spans": [ + { + "bbox": [ + 141, + 545, + 186, + 557 + ], + "score": 1.0, + "content": "Bounded:", + "type": "text" + }, + { + "bbox": [ + 187, + 545, + 235, + 557 + ], + "score": 0.71, + "content": "\\perp \\preceq a \\preceq \\top", + "type": "inline_equation" + } + ], + "index": 25, + "is_list_start_line": true + } + ], + "index": 23, + "bbox_fs": [ + 141, + 489, + 417, + 557 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 565, + 504, + 610 + ], + "lines": [ + { + "bbox": [ + 105, + 565, + 505, + 578 + ], + "spans": [ + { + "bbox": [ + 105, + 565, + 258, + 578 + ], + "score": 1.0, + "content": "Note that the extended real numbers,", + "type": "text" + }, + { + "bbox": [ + 258, + 565, + 320, + 577 + ], + "score": 0.93, + "content": "\\mathbb { R } \\cup \\{ - \\infty , \\infty \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 565, + 505, + 578 + ], + "score": 1.0, + "content": ", form a bounded lattice (and in fact, a totally", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 577, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 106, + 577, + 347, + 588 + ], + "score": 1.0, + "content": "ordered set) under the min and max operations as the meet", + "type": "text" + }, + { + "bbox": [ + 347, + 577, + 361, + 587 + ], + "score": 0.65, + "content": "( \\wedge )", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 577, + 398, + 588 + ], + "score": 1.0, + "content": "and join", + "type": "text" + }, + { + "bbox": [ + 399, + 577, + 412, + 587 + ], + "score": 0.38, + "content": "( \\vee )", + "type": "inline_equation" + }, + { + "bbox": [ + 413, + 577, + 505, + 588 + ], + "score": 1.0, + "content": "operations. So do sets", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 587, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 251, + 599 + ], + "score": 1.0, + "content": "partially ordered by inclusion, with", + "type": "text" + }, + { + "bbox": [ + 251, + 588, + 259, + 597 + ], + "score": 0.63, + "content": "\\cap", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 587, + 277, + 599 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 278, + 588, + 286, + 597 + ], + "score": 0.36, + "content": "\\cup", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 587, + 298, + 599 + ], + "score": 1.0, + "content": "as", + "type": "text" + }, + { + "bbox": [ + 298, + 588, + 307, + 597 + ], + "score": 0.78, + "content": "\\wedge", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 587, + 325, + 599 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 325, + 588, + 333, + 597 + ], + "score": 0.59, + "content": "\\vee", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 587, + 505, + 599 + ], + "score": 1.0, + "content": ". Thinking of these special cases gives the", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 598, + 284, + 611 + ], + "spans": [ + { + "bbox": [ + 106, + 598, + 284, + 611 + ], + "score": 1.0, + "content": "intuition for the fourth property, absorption.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5, + "bbox_fs": [ + 105, + 565, + 505, + 611 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 614, + 505, + 648 + ], + "lines": [ + { + "bbox": [ + 105, + 614, + 504, + 628 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 123, + 628 + ], + "score": 1.0, + "content": "The", + "type": "text" + }, + { + "bbox": [ + 124, + 616, + 132, + 625 + ], + "score": 0.72, + "content": "\\wedge", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 614, + 149, + 628 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 149, + 616, + 157, + 625 + ], + "score": 0.26, + "content": "\\vee", + "type": "inline_equation" + }, + { + "bbox": [ + 158, + 614, + 420, + 628 + ], + "score": 1.0, + "content": "operations can be swapped, along with reversing the poset relation", + "type": "text" + }, + { + "bbox": [ + 420, + 616, + 429, + 626 + ], + "score": 0.77, + "content": "\\preceq", + "type": "inline_equation" + }, + { + "bbox": [ + 430, + 614, + 504, + 628 + ], + "score": 1.0, + "content": ", to give a valid lat-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 625, + 505, + 638 + ], + "spans": [ + { + "bbox": [ + 105, + 625, + 505, + 638 + ], + "score": 1.0, + "content": "tice, called the dual lattice. In the real numbers this just corresponds to a sign change. A semilattice", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 637, + 254, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 637, + 254, + 648 + ], + "score": 1.0, + "content": "has only a meet or join, but not both.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 105, + 614, + 505, + 648 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 653, + 504, + 676 + ], + "lines": [ + { + "bbox": [ + 106, + 654, + 505, + 665 + ], + "spans": [ + { + "bbox": [ + 106, + 654, + 394, + 665 + ], + "score": 1.0, + "content": "Note. In the rest of the paper, when the context is clear, we will also use", + "type": "text" + }, + { + "bbox": [ + 395, + 655, + 403, + 664 + ], + "score": 0.73, + "content": "\\wedge", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 654, + 421, + 665 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 421, + 655, + 429, + 664 + ], + "score": 0.3, + "content": "\\vee", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 654, + 505, + 665 + ], + "score": 1.0, + "content": "to denote min and", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 664, + 393, + 676 + ], + "spans": [ + { + "bbox": [ + 105, + 664, + 393, + 676 + ], + "score": 1.0, + "content": "max of real numbers, in order to clarify the intuition behind our model.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 654, + 505, + 676 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 689, + 208, + 700 + ], + "lines": [ + { + "bbox": [ + 105, + 688, + 209, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 209, + 701 + ], + "score": 1.0, + "content": "3.2 VECTOR LATTICE", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "A vector lattice, also known as a Riesz space (Zaanen, 1997), or Hilbert lattice when the accompa-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 720, + 469, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 469, + 733 + ], + "score": 1.0, + "content": "nying vector space has an inner product, is a vector space endowed with a lattice structure.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5, + "bbox_fs": [ + 105, + 709, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 337, + 96 + ], + "score": 1.0, + "content": "A standard choice of partial order for the vector lattice", + "type": "text" + }, + { + "bbox": [ + 337, + 83, + 351, + 92 + ], + "score": 0.88, + "content": "\\mathbb { R } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "is to use the product order from the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 92, + 342, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 92, + 299, + 106 + ], + "score": 1.0, + "content": "underlying real numbers, which specifies for all", + "type": "text" + }, + { + "bbox": [ + 299, + 94, + 342, + 105 + ], + "score": 0.9, + "content": "\\mathbf { x } , \\mathbf { y } \\in \\mathbb { R } ^ { n }", + "type": "inline_equation" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 232, + 109, + 378, + 123 + ], + "lines": [ + { + "bbox": [ + 232, + 109, + 378, + 123 + ], + "spans": [ + { + "bbox": [ + 232, + 109, + 378, + 123 + ], + "score": 0.92, + "content": "\\mathbf { x } \\preceq \\mathbf { y } \\iff \\forall i \\in \\{ 1 . . n \\} , \\ x _ { i } \\leq y _ { i }", + "type": "interline_equation", + "image_path": "39447f8d1918348420bed1c934f9decaf1395429fb960ee0d6b944d409ab3817.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 232, + 109, + 378, + 123 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 126, + 505, + 182 + ], + "lines": [ + { + "bbox": [ + 106, + 127, + 504, + 139 + ], + "spans": [ + { + "bbox": [ + 106, + 127, + 504, + 139 + ], + "score": 1.0, + "content": "Under this order, meet and join operations are pointwise min and max, which gives a lattice struc-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 137, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 505, + 151 + ], + "score": 1.0, + "content": "ture. In this formalism, the Order Embeddings of Vendrov et al. (2016) embed partial orders as", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 149, + 505, + 161 + ], + "spans": [ + { + "bbox": [ + 105, + 149, + 505, + 161 + ], + "score": 1.0, + "content": "vectors using the reverse product order, corresponding to the dual lattice, and restrict the vectors to", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 159, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 293, + 173 + ], + "score": 1.0, + "content": "be positive. The vector of all zeroes represents", + "type": "text" + }, + { + "bbox": [ + 294, + 160, + 302, + 170 + ], + "score": 0.74, + "content": "\\top", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 159, + 506, + 173 + ], + "score": 1.0, + "content": ", and embedded objects become “more specific” as", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 170, + 258, + 184 + ], + "spans": [ + { + "bbox": [ + 106, + 170, + 258, + 184 + ], + "score": 1.0, + "content": "they get farther away from the origin.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 187, + 505, + 221 + ], + "lines": [ + { + "bbox": [ + 105, + 186, + 505, + 202 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 505, + 202 + ], + "score": 1.0, + "content": "Figure 1b demonstrates a toy, two-dimensional example of the Order Embedding vector lattice rep-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 199, + 505, + 211 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 505, + 211 + ], + "score": 1.0, + "content": "resentation of a simple ontology. Shading represents the probability measure assigned to this lattice", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 210, + 348, + 221 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 348, + 221 + ], + "score": 1.0, + "content": "in the probabilistic extension of Lai & Hockenmaier (2017).", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "title", + "bbox": [ + 107, + 234, + 191, + 245 + ], + "lines": [ + { + "bbox": [ + 105, + 232, + 192, + 247 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 192, + 247 + ], + "score": 1.0, + "content": "3.3 BOX LATTICE", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 107, + 254, + 505, + 288 + ], + "lines": [ + { + "bbox": [ + 106, + 255, + 505, + 267 + ], + "spans": [ + { + "bbox": [ + 106, + 255, + 505, + 267 + ], + "score": 1.0, + "content": "Vilnis et al. (2018) introduced a box lattice, wherein each concept in a knowledge graph is associated", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 266, + 505, + 278 + ], + "spans": [ + { + "bbox": [ + 106, + 266, + 505, + 278 + ], + "score": 1.0, + "content": "with two vectors, the minimum and maximum coordinates of an axis-aligned hyperrectangle, or box", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 277, + 196, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 277, + 196, + 289 + ], + "score": 1.0, + "content": "(product of intervals).", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 107, + 293, + 505, + 338 + ], + "lines": [ + { + "bbox": [ + 106, + 294, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 106, + 294, + 505, + 306 + ], + "score": 1.0, + "content": "Using the notion of set inclusion between boxes, there is a natural partial order and lattice structure.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 304, + 506, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 181, + 318 + ], + "score": 1.0, + "content": "To represent a box", + "type": "text" + }, + { + "bbox": [ + 181, + 307, + 189, + 315 + ], + "score": 0.51, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 304, + 240, + 318 + ], + "score": 1.0, + "content": ", let the pairs", + "type": "text" + }, + { + "bbox": [ + 241, + 305, + 291, + 317 + ], + "score": 0.93, + "content": "( x _ { m , i } , x _ { M , i } )", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 304, + 506, + 318 + ], + "score": 1.0, + "content": "be the maximum and minimum of the interval at each", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 315, + 504, + 328 + ], + "spans": [ + { + "bbox": [ + 105, + 315, + 151, + 328 + ], + "score": 1.0, + "content": "coordinate", + "type": "text" + }, + { + "bbox": [ + 151, + 317, + 156, + 326 + ], + "score": 0.64, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 315, + 496, + 328 + ], + "score": 1.0, + "content": ". Then the box lattice structure (least upper bounds and greatest lower bounds), with", + "type": "text" + }, + { + "bbox": [ + 496, + 316, + 504, + 326 + ], + "score": 0.52, + "content": "\\vee", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 327, + 394, + 339 + ], + "spans": [ + { + "bbox": [ + 106, + 327, + 123, + 339 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 327, + 132, + 337 + ], + "score": 0.8, + "content": "\\wedge", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 327, + 394, + 339 + ], + "score": 1.0, + "content": "denoting max and min when applied to the scalar coordinates, is", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16.5 + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 358, + 385, + 411 + ], + "lines": [ + { + "bbox": [ + 225, + 358, + 385, + 411 + ], + "spans": [ + { + "bbox": [ + 225, + 358, + 385, + 411 + ], + "score": 0.87, + "content": "\\begin{array} { l } { { \\displaystyle { \\bf x } \\wedge { \\bf y } = \\prod _ { i } [ x _ { m , i } \\vee y _ { m , i } , x _ { M , i } \\wedge y _ { M , i } ] } } \\\\ { { \\displaystyle { \\bf x } \\vee { \\bf y } = \\prod _ { i } [ x _ { m , i } \\wedge y _ { m , i } , x _ { M , i } \\vee y _ { M , i } ] } } \\end{array}", + "type": "interline_equation", + "image_path": "dfd1b2c9110ec9bc78da993096d360ceef0c1b07d83878a0a797e9687962d9c5.jpg" + } + ] + } + ], + "index": 20.5, + "virtual_lines": [ + { + "bbox": [ + 225, + 358, + 385, + 371.25 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 225, + 371.25, + 385, + 384.5 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 225, + 384.5, + 385, + 397.75 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 225, + 397.75, + 385, + 411.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 415, + 505, + 449 + ], + "lines": [ + { + "bbox": [ + 105, + 414, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 131, + 428 + ], + "score": 1.0, + "content": "Here,", + "type": "text" + }, + { + "bbox": [ + 132, + 415, + 143, + 428 + ], + "score": 0.83, + "content": "\\prod", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 414, + 505, + 428 + ], + "score": 1.0, + "content": "denotes a set (cartesian) product — the lattice meet is the largest box contained entirely", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 426, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 155, + 438 + ], + "score": 1.0, + "content": "within both", + "type": "text" + }, + { + "bbox": [ + 155, + 428, + 163, + 437 + ], + "score": 0.69, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 426, + 181, + 438 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 182, + 428, + 189, + 438 + ], + "score": 0.68, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 426, + 505, + 438 + ], + "score": 1.0, + "content": ", or bottom (the empty set) where no intersection exists, and the lattice join is", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 437, + 276, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 238, + 450 + ], + "score": 1.0, + "content": "the smallest box containing both", + "type": "text" + }, + { + "bbox": [ + 239, + 439, + 246, + 447 + ], + "score": 0.69, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 437, + 263, + 450 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 264, + 439, + 271, + 449 + ], + "score": 0.57, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 437, + 276, + 450 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 106, + 453, + 505, + 532 + ], + "lines": [ + { + "bbox": [ + 106, + 454, + 506, + 466 + ], + "spans": [ + { + "bbox": [ + 106, + 454, + 506, + 466 + ], + "score": 1.0, + "content": "To associate a measure, marginal probabilities of (collections of) events are given by the volume of", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 465, + 506, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 506, + 478 + ], + "score": 1.0, + "content": "boxes, their complements, and intersections under a suitable probability measure. Under the uniform", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 475, + 504, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 177, + 489 + ], + "score": 1.0, + "content": "measure, if event", + "type": "text" + }, + { + "bbox": [ + 178, + 478, + 186, + 486 + ], + "score": 0.65, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 475, + 377, + 489 + ], + "score": 1.0, + "content": "has an associated box with interval boundaries", + "type": "text" + }, + { + "bbox": [ + 378, + 476, + 417, + 488 + ], + "score": 0.94, + "content": "( x _ { m } , x _ { M } )", + "type": "inline_equation" + }, + { + "bbox": [ + 418, + 475, + 483, + 489 + ], + "score": 1.0, + "content": ", the probability", + "type": "text" + }, + { + "bbox": [ + 484, + 476, + 504, + 488 + ], + "score": 0.9, + "content": "p ( \\mathbf { x } )", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 487, + 506, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 154, + 500 + ], + "score": 1.0, + "content": "is given by", + "type": "text" + }, + { + "bbox": [ + 154, + 487, + 227, + 500 + ], + "score": 0.91, + "content": "\\textstyle \\prod _ { i } ^ { n } ( x _ { M , i } - x _ { m , i } )", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 487, + 506, + 500 + ], + "score": 1.0, + "content": ". Use of the uniform measure requires the boxes to be constrained to", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 497, + 506, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 216, + 511 + ], + "score": 1.0, + "content": "the unit hypercube, so that", + "type": "text" + }, + { + "bbox": [ + 217, + 498, + 256, + 510 + ], + "score": 0.85, + "content": "p ( \\mathbf { x } ) \\leq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 497, + 261, + 511 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 261, + 498, + 283, + 510 + ], + "score": 0.83, + "content": "p ( \\bot )", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 497, + 387, + 511 + ], + "score": 1.0, + "content": "is taken to be zero, since", + "type": "text" + }, + { + "bbox": [ + 388, + 498, + 397, + 508 + ], + "score": 0.66, + "content": "\\perp", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 497, + 506, + 511 + ], + "score": 1.0, + "content": "is an empty set. As boxes", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 509, + 506, + 521 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 506, + 521 + ], + "score": 1.0, + "content": "are simply special cases of sets, it is intuitive that this is a valid probability measure, but it can also", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 519, + 491, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 491, + 533 + ], + "score": 1.0, + "content": "be shown to be compatible with the meet semilattice structure in a precise sense (Leader, 1971).", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 105, + 536, + 504, + 560 + ], + "lines": [ + { + "bbox": [ + 104, + 536, + 505, + 550 + ], + "spans": [ + { + "bbox": [ + 104, + 536, + 505, + 550 + ], + "score": 1.0, + "content": "Figure 1c demonstrates a toy, two-dimensional example of the Box Embedding lattice representation", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 546, + 193, + 561 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 193, + 561 + ], + "score": 1.0, + "content": "of a simple ontology.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5 + }, + { + "type": "title", + "bbox": [ + 107, + 575, + 173, + 588 + ], + "lines": [ + { + "bbox": [ + 105, + 574, + 174, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 174, + 589 + ], + "score": 1.0, + "content": "4 METHOD", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "title", + "bbox": [ + 107, + 599, + 340, + 611 + ], + "lines": [ + { + "bbox": [ + 105, + 599, + 340, + 612 + ], + "spans": [ + { + "bbox": [ + 105, + 599, + 340, + 612 + ], + "score": 1.0, + "content": "4.1 MOTIVATION: OPTIMIZATION AND SPARSE DATA", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 106, + 619, + 505, + 676 + ], + "lines": [ + { + "bbox": [ + 106, + 620, + 505, + 632 + ], + "spans": [ + { + "bbox": [ + 106, + 620, + 505, + 632 + ], + "score": 1.0, + "content": "When using gradient-based optimization to learn box embeddings, an immediate problem identified", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 632, + 505, + 644 + ], + "spans": [ + { + "bbox": [ + 106, + 632, + 505, + 644 + ], + "score": 1.0, + "content": "in the original work is that when two concepts are incorrectly given as disjoint by the model, no", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 642, + 506, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 506, + 655 + ], + "score": 1.0, + "content": "gradient signal can flow since the meet (intersection) is exactly zero, with zero derivative. To see", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 653, + 505, + 665 + ], + "spans": [ + { + "bbox": [ + 106, + 653, + 505, + 665 + ], + "score": 1.0, + "content": "this, note that for a pair of 1-dimensional boxes (intervals), the volume of the meet under the uniform", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 664, + 252, + 677 + ], + "spans": [ + { + "bbox": [ + 105, + 664, + 142, + 677 + ], + "score": 1.0, + "content": "measure", + "type": "text" + }, + { + "bbox": [ + 142, + 667, + 149, + 676 + ], + "score": 0.79, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 664, + 252, + 677 + ], + "score": 1.0, + "content": "as given in Section 3.3 is", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 39 + }, + { + "type": "interline_equation", + "bbox": [ + 205, + 696, + 405, + 711 + ], + "lines": [ + { + "bbox": [ + 205, + 696, + 405, + 711 + ], + "spans": [ + { + "bbox": [ + 205, + 696, + 405, + 711 + ], + "score": 0.85, + "content": "p ( \\mathbf { x } \\wedge \\mathbf { y } ) = m _ { h } ( \\operatorname* { m i n } ( x _ { M } , y _ { M } ) - \\operatorname* { m a x } ( x _ { m } , y _ { m } ) )", + "type": "interline_equation", + "image_path": "8b81c5c769e5c8e08ee0accfb5907e109e1db0fbc9508b4f7340f4a95fe91297.jpg" + } + ] + } + ], + "index": 42, + "virtual_lines": [ + { + "bbox": [ + 205, + 696, + 405, + 711 + ], + "spans": [], + "index": 42 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 720, + 394, + 733 + ], + "lines": [ + { + "bbox": [ + 106, + 718, + 395, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 718, + 133, + 734 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 722, + 148, + 732 + ], + "score": 0.86, + "content": "m _ { h }", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 718, + 273, + 734 + ], + "score": 1.0, + "content": "is the standard hinge function,", + "type": "text" + }, + { + "bbox": [ + 273, + 720, + 391, + 732 + ], + "score": 0.91, + "content": "m _ { h } ( x ) = 0 \\lor x = \\operatorname* { m a x } ( 0 , x )", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 718, + 395, + 734 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 43 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 310, + 762 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 8 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 337, + 96 + ], + "score": 1.0, + "content": "A standard choice of partial order for the vector lattice", + "type": "text" + }, + { + "bbox": [ + 337, + 83, + 351, + 92 + ], + "score": 0.88, + "content": "\\mathbb { R } ^ { n }", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "is to use the product order from the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 92, + 342, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 92, + 299, + 106 + ], + "score": 1.0, + "content": "underlying real numbers, which specifies for all", + "type": "text" + }, + { + "bbox": [ + 299, + 94, + 342, + 105 + ], + "score": 0.9, + "content": "\\mathbf { x } , \\mathbf { y } \\in \\mathbb { R } ^ { n }", + "type": "inline_equation" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 105, + 81, + 505, + 106 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 232, + 109, + 378, + 123 + ], + "lines": [ + { + "bbox": [ + 232, + 109, + 378, + 123 + ], + "spans": [ + { + "bbox": [ + 232, + 109, + 378, + 123 + ], + "score": 0.92, + "content": "\\mathbf { x } \\preceq \\mathbf { y } \\iff \\forall i \\in \\{ 1 . . n \\} , \\ x _ { i } \\leq y _ { i }", + "type": "interline_equation", + "image_path": "39447f8d1918348420bed1c934f9decaf1395429fb960ee0d6b944d409ab3817.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 232, + 109, + 378, + 123 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 126, + 505, + 182 + ], + "lines": [ + { + "bbox": [ + 106, + 127, + 504, + 139 + ], + "spans": [ + { + "bbox": [ + 106, + 127, + 504, + 139 + ], + "score": 1.0, + "content": "Under this order, meet and join operations are pointwise min and max, which gives a lattice struc-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 137, + 505, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 505, + 151 + ], + "score": 1.0, + "content": "ture. In this formalism, the Order Embeddings of Vendrov et al. (2016) embed partial orders as", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 149, + 505, + 161 + ], + "spans": [ + { + "bbox": [ + 105, + 149, + 505, + 161 + ], + "score": 1.0, + "content": "vectors using the reverse product order, corresponding to the dual lattice, and restrict the vectors to", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 159, + 506, + 173 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 293, + 173 + ], + "score": 1.0, + "content": "be positive. The vector of all zeroes represents", + "type": "text" + }, + { + "bbox": [ + 294, + 160, + 302, + 170 + ], + "score": 0.74, + "content": "\\top", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 159, + 506, + 173 + ], + "score": 1.0, + "content": ", and embedded objects become “more specific” as", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 170, + 258, + 184 + ], + "spans": [ + { + "bbox": [ + 106, + 170, + 258, + 184 + ], + "score": 1.0, + "content": "they get farther away from the origin.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 127, + 506, + 184 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 187, + 505, + 221 + ], + "lines": [ + { + "bbox": [ + 105, + 186, + 505, + 202 + ], + "spans": [ + { + "bbox": [ + 105, + 186, + 505, + 202 + ], + "score": 1.0, + "content": "Figure 1b demonstrates a toy, two-dimensional example of the Order Embedding vector lattice rep-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 199, + 505, + 211 + ], + "spans": [ + { + "bbox": [ + 106, + 199, + 505, + 211 + ], + "score": 1.0, + "content": "resentation of a simple ontology. Shading represents the probability measure assigned to this lattice", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 210, + 348, + 221 + ], + "spans": [ + { + "bbox": [ + 106, + 210, + 348, + 221 + ], + "score": 1.0, + "content": "in the probabilistic extension of Lai & Hockenmaier (2017).", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 186, + 505, + 221 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 234, + 191, + 245 + ], + "lines": [ + { + "bbox": [ + 105, + 232, + 192, + 247 + ], + "spans": [ + { + "bbox": [ + 105, + 232, + 192, + 247 + ], + "score": 1.0, + "content": "3.3 BOX LATTICE", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 107, + 254, + 505, + 288 + ], + "lines": [ + { + "bbox": [ + 106, + 255, + 505, + 267 + ], + "spans": [ + { + "bbox": [ + 106, + 255, + 505, + 267 + ], + "score": 1.0, + "content": "Vilnis et al. (2018) introduced a box lattice, wherein each concept in a knowledge graph is associated", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 266, + 505, + 278 + ], + "spans": [ + { + "bbox": [ + 106, + 266, + 505, + 278 + ], + "score": 1.0, + "content": "with two vectors, the minimum and maximum coordinates of an axis-aligned hyperrectangle, or box", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 277, + 196, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 277, + 196, + 289 + ], + "score": 1.0, + "content": "(product of intervals).", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13, + "bbox_fs": [ + 105, + 255, + 505, + 289 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 293, + 505, + 338 + ], + "lines": [ + { + "bbox": [ + 106, + 294, + 505, + 306 + ], + "spans": [ + { + "bbox": [ + 106, + 294, + 505, + 306 + ], + "score": 1.0, + "content": "Using the notion of set inclusion between boxes, there is a natural partial order and lattice structure.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 304, + 506, + 318 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 181, + 318 + ], + "score": 1.0, + "content": "To represent a box", + "type": "text" + }, + { + "bbox": [ + 181, + 307, + 189, + 315 + ], + "score": 0.51, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 304, + 240, + 318 + ], + "score": 1.0, + "content": ", let the pairs", + "type": "text" + }, + { + "bbox": [ + 241, + 305, + 291, + 317 + ], + "score": 0.93, + "content": "( x _ { m , i } , x _ { M , i } )", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 304, + 506, + 318 + ], + "score": 1.0, + "content": "be the maximum and minimum of the interval at each", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 315, + 504, + 328 + ], + "spans": [ + { + "bbox": [ + 105, + 315, + 151, + 328 + ], + "score": 1.0, + "content": "coordinate", + "type": "text" + }, + { + "bbox": [ + 151, + 317, + 156, + 326 + ], + "score": 0.64, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 315, + 496, + 328 + ], + "score": 1.0, + "content": ". Then the box lattice structure (least upper bounds and greatest lower bounds), with", + "type": "text" + }, + { + "bbox": [ + 496, + 316, + 504, + 326 + ], + "score": 0.52, + "content": "\\vee", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 327, + 394, + 339 + ], + "spans": [ + { + "bbox": [ + 106, + 327, + 123, + 339 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 327, + 132, + 337 + ], + "score": 0.8, + "content": "\\wedge", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 327, + 394, + 339 + ], + "score": 1.0, + "content": "denoting max and min when applied to the scalar coordinates, is", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 294, + 506, + 339 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 358, + 385, + 411 + ], + "lines": [ + { + "bbox": [ + 225, + 358, + 385, + 411 + ], + "spans": [ + { + "bbox": [ + 225, + 358, + 385, + 411 + ], + "score": 0.87, + "content": "\\begin{array} { l } { { \\displaystyle { \\bf x } \\wedge { \\bf y } = \\prod _ { i } [ x _ { m , i } \\vee y _ { m , i } , x _ { M , i } \\wedge y _ { M , i } ] } } \\\\ { { \\displaystyle { \\bf x } \\vee { \\bf y } = \\prod _ { i } [ x _ { m , i } \\wedge y _ { m , i } , x _ { M , i } \\vee y _ { M , i } ] } } \\end{array}", + "type": "interline_equation", + "image_path": "dfd1b2c9110ec9bc78da993096d360ceef0c1b07d83878a0a797e9687962d9c5.jpg" + } + ] + } + ], + "index": 20.5, + "virtual_lines": [ + { + "bbox": [ + 225, + 358, + 385, + 371.25 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 225, + 371.25, + 385, + 384.5 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 225, + 384.5, + 385, + 397.75 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 225, + 397.75, + 385, + 411.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 415, + 505, + 449 + ], + "lines": [ + { + "bbox": [ + 105, + 414, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 131, + 428 + ], + "score": 1.0, + "content": "Here,", + "type": "text" + }, + { + "bbox": [ + 132, + 415, + 143, + 428 + ], + "score": 0.83, + "content": "\\prod", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 414, + 505, + 428 + ], + "score": 1.0, + "content": "denotes a set (cartesian) product — the lattice meet is the largest box contained entirely", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 426, + 505, + 438 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 155, + 438 + ], + "score": 1.0, + "content": "within both", + "type": "text" + }, + { + "bbox": [ + 155, + 428, + 163, + 437 + ], + "score": 0.69, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 426, + 181, + 438 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 182, + 428, + 189, + 438 + ], + "score": 0.68, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 426, + 505, + 438 + ], + "score": 1.0, + "content": ", or bottom (the empty set) where no intersection exists, and the lattice join is", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 437, + 276, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 238, + 450 + ], + "score": 1.0, + "content": "the smallest box containing both", + "type": "text" + }, + { + "bbox": [ + 239, + 439, + 246, + 447 + ], + "score": 0.69, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 246, + 437, + 263, + 450 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 264, + 439, + 271, + 449 + ], + "score": 0.57, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 437, + 276, + 450 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 414, + 505, + 450 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 453, + 505, + 532 + ], + "lines": [ + { + "bbox": [ + 106, + 454, + 506, + 466 + ], + "spans": [ + { + "bbox": [ + 106, + 454, + 506, + 466 + ], + "score": 1.0, + "content": "To associate a measure, marginal probabilities of (collections of) events are given by the volume of", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 465, + 506, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 506, + 478 + ], + "score": 1.0, + "content": "boxes, their complements, and intersections under a suitable probability measure. Under the uniform", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 475, + 504, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 177, + 489 + ], + "score": 1.0, + "content": "measure, if event", + "type": "text" + }, + { + "bbox": [ + 178, + 478, + 186, + 486 + ], + "score": 0.65, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 475, + 377, + 489 + ], + "score": 1.0, + "content": "has an associated box with interval boundaries", + "type": "text" + }, + { + "bbox": [ + 378, + 476, + 417, + 488 + ], + "score": 0.94, + "content": "( x _ { m } , x _ { M } )", + "type": "inline_equation" + }, + { + "bbox": [ + 418, + 475, + 483, + 489 + ], + "score": 1.0, + "content": ", the probability", + "type": "text" + }, + { + "bbox": [ + 484, + 476, + 504, + 488 + ], + "score": 0.9, + "content": "p ( \\mathbf { x } )", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 487, + 506, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 154, + 500 + ], + "score": 1.0, + "content": "is given by", + "type": "text" + }, + { + "bbox": [ + 154, + 487, + 227, + 500 + ], + "score": 0.91, + "content": "\\textstyle \\prod _ { i } ^ { n } ( x _ { M , i } - x _ { m , i } )", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 487, + 506, + 500 + ], + "score": 1.0, + "content": ". Use of the uniform measure requires the boxes to be constrained to", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 497, + 506, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 216, + 511 + ], + "score": 1.0, + "content": "the unit hypercube, so that", + "type": "text" + }, + { + "bbox": [ + 217, + 498, + 256, + 510 + ], + "score": 0.85, + "content": "p ( \\mathbf { x } ) \\leq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 497, + 261, + 511 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 261, + 498, + 283, + 510 + ], + "score": 0.83, + "content": "p ( \\bot )", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 497, + 387, + 511 + ], + "score": 1.0, + "content": "is taken to be zero, since", + "type": "text" + }, + { + "bbox": [ + 388, + 498, + 397, + 508 + ], + "score": 0.66, + "content": "\\perp", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 497, + 506, + 511 + ], + "score": 1.0, + "content": "is an empty set. As boxes", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 509, + 506, + 521 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 506, + 521 + ], + "score": 1.0, + "content": "are simply special cases of sets, it is intuitive that this is a valid probability measure, but it can also", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 519, + 491, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 491, + 533 + ], + "score": 1.0, + "content": "be shown to be compatible with the meet semilattice structure in a precise sense (Leader, 1971).", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 454, + 506, + 533 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 536, + 504, + 560 + ], + "lines": [ + { + "bbox": [ + 104, + 536, + 505, + 550 + ], + "spans": [ + { + "bbox": [ + 104, + 536, + 505, + 550 + ], + "score": 1.0, + "content": "Figure 1c demonstrates a toy, two-dimensional example of the Box Embedding lattice representation", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 546, + 193, + 561 + ], + "spans": [ + { + "bbox": [ + 105, + 546, + 193, + 561 + ], + "score": 1.0, + "content": "of a simple ontology.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33.5, + "bbox_fs": [ + 104, + 536, + 505, + 561 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 575, + 173, + 588 + ], + "lines": [ + { + "bbox": [ + 105, + 574, + 174, + 589 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 174, + 589 + ], + "score": 1.0, + "content": "4 METHOD", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 35 + }, + { + "type": "title", + "bbox": [ + 107, + 599, + 340, + 611 + ], + "lines": [ + { + "bbox": [ + 105, + 599, + 340, + 612 + ], + "spans": [ + { + "bbox": [ + 105, + 599, + 340, + 612 + ], + "score": 1.0, + "content": "4.1 MOTIVATION: OPTIMIZATION AND SPARSE DATA", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 106, + 619, + 505, + 676 + ], + "lines": [ + { + "bbox": [ + 106, + 620, + 505, + 632 + ], + "spans": [ + { + "bbox": [ + 106, + 620, + 505, + 632 + ], + "score": 1.0, + "content": "When using gradient-based optimization to learn box embeddings, an immediate problem identified", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 632, + 505, + 644 + ], + "spans": [ + { + "bbox": [ + 106, + 632, + 505, + 644 + ], + "score": 1.0, + "content": "in the original work is that when two concepts are incorrectly given as disjoint by the model, no", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 642, + 506, + 655 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 506, + 655 + ], + "score": 1.0, + "content": "gradient signal can flow since the meet (intersection) is exactly zero, with zero derivative. To see", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 653, + 505, + 665 + ], + "spans": [ + { + "bbox": [ + 106, + 653, + 505, + 665 + ], + "score": 1.0, + "content": "this, note that for a pair of 1-dimensional boxes (intervals), the volume of the meet under the uniform", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 664, + 252, + 677 + ], + "spans": [ + { + "bbox": [ + 105, + 664, + 142, + 677 + ], + "score": 1.0, + "content": "measure", + "type": "text" + }, + { + "bbox": [ + 142, + 667, + 149, + 676 + ], + "score": 0.79, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 664, + 252, + 677 + ], + "score": 1.0, + "content": "as given in Section 3.3 is", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 39, + "bbox_fs": [ + 105, + 620, + 506, + 677 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 205, + 696, + 405, + 711 + ], + "lines": [ + { + "bbox": [ + 205, + 696, + 405, + 711 + ], + "spans": [ + { + "bbox": [ + 205, + 696, + 405, + 711 + ], + "score": 0.85, + "content": "p ( \\mathbf { x } \\wedge \\mathbf { y } ) = m _ { h } ( \\operatorname* { m i n } ( x _ { M } , y _ { M } ) - \\operatorname* { m a x } ( x _ { m } , y _ { m } ) )", + "type": "interline_equation", + "image_path": "8b81c5c769e5c8e08ee0accfb5907e109e1db0fbc9508b4f7340f4a95fe91297.jpg" + } + ] + } + ], + "index": 42, + "virtual_lines": [ + { + "bbox": [ + 205, + 696, + 405, + 711 + ], + "spans": [], + "index": 42 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 720, + 394, + 733 + ], + "lines": [ + { + "bbox": [ + 106, + 718, + 395, + 734 + ], + "spans": [ + { + "bbox": [ + 106, + 718, + 133, + 734 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 722, + 148, + 732 + ], + "score": 0.86, + "content": "m _ { h }", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 718, + 273, + 734 + ], + "score": 1.0, + "content": "is the standard hinge function,", + "type": "text" + }, + { + "bbox": [ + 273, + 720, + 391, + 732 + ], + "score": 0.91, + "content": "m _ { h } ( x ) = 0 \\lor x = \\operatorname* { m a x } ( 0 , x )", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 718, + 395, + 734 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 43, + "bbox_fs": [ + 106, + 718, + 395, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 149 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 504, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 504, + 95 + ], + "score": 1.0, + "content": "The hinge function has a large flat plateau at 0 when intervals are disjoint. This issue is especially", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "problematic when the lattice to be embedded is (pseudo-)sparse, that is, most boxes should have", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 505, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 505, + 119 + ], + "score": 1.0, + "content": "very little or no intersection, since if training accidentally makes two boxes disjoint there is no way", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "to recover with the naive measure. The authors propose a surrogate function to optimize in this case,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 124, + 506, + 141 + ], + "spans": [ + { + "bbox": [ + 104, + 124, + 506, + 141 + ], + "score": 1.0, + "content": "but we will use a more principled framework to develop alternate measures that avoid this pathology,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 320, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 320, + 150 + ], + "score": 1.0, + "content": "improving both optimization and final model quality.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 107, + 162, + 225, + 173 + ], + "lines": [ + { + "bbox": [ + 106, + 162, + 227, + 174 + ], + "spans": [ + { + "bbox": [ + 106, + 162, + 227, + 174 + ], + "score": 1.0, + "content": "4.2 RELAXED GEOMETRY", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "image", + "bbox": [ + 112, + 183, + 498, + 284 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 112, + 183, + 498, + 284 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 183, + 498, + 284 + ], + "spans": [ + { + "bbox": [ + 112, + 183, + 498, + 284 + ], + "score": 0.964, + "type": "image", + "image_path": "52e273973b88b2eeb82ffad0414a0376e7873c9a60e503fb0bddeea225d759b0.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 112, + 183, + 498, + 216.66666666666666 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 112, + 216.66666666666666, + 498, + 250.33333333333331 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 112, + 250.33333333333331, + 498, + 284.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 293, + 505, + 326 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 292, + 506, + 306 + ], + "spans": [ + { + "bbox": [ + 106, + 292, + 506, + 306 + ], + "score": 1.0, + "content": "Figure 2: One-dimensional example demonstrating two disjoint indicators of intervals before and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 303, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 505, + 316 + ], + "score": 1.0, + "content": "after the application of a smoothing kernel. The area under the purple product curve is proportional", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 314, + 205, + 328 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 205, + 328 + ], + "score": 1.0, + "content": "to the degree of overlap.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 107, + 340, + 505, + 407 + ], + "lines": [ + { + "bbox": [ + 106, + 341, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 341, + 505, + 352 + ], + "score": 1.0, + "content": "The intuition behind our approach is that the “hard edges” of the standard box embeddings lead to", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 352, + 506, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 506, + 364 + ], + "score": 1.0, + "content": "unwanted gradient sparsity, and we seek a relaxation of this assumption that maintains the desirable", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 362, + 506, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 506, + 376 + ], + "score": 1.0, + "content": "properties of the base lattice model while enabling better optimization and preserving a geometric", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 375, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 375, + 505, + 385 + ], + "score": 1.0, + "content": "intuition. For ease of exposition, we will refer to 1-dimensional intervals in this section, but the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 384, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 505, + 396 + ], + "score": 1.0, + "content": "results carry through from the representation of boxes as products of intervals and their volumes", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 396, + 265, + 407 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 265, + 407 + ], + "score": 1.0, + "content": "under the associated product measures.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 107, + 412, + 505, + 446 + ], + "lines": [ + { + "bbox": [ + 106, + 412, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 106, + 412, + 505, + 424 + ], + "score": 1.0, + "content": "The first observation is that, considering boxes as indicator functions of intervals, we can rewrite the", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 423, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 235, + 437 + ], + "score": 1.0, + "content": "measure of the joint probability", + "type": "text" + }, + { + "bbox": [ + 236, + 424, + 273, + 435 + ], + "score": 0.92, + "content": "p ( \\mathbf { x } \\wedge \\mathbf { y } )", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 423, + 347, + 437 + ], + "score": 1.0, + "content": "between intervals", + "type": "text" + }, + { + "bbox": [ + 348, + 423, + 388, + 435 + ], + "score": 0.93, + "content": "\\mathbf { x } = [ a , b ]", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 423, + 406, + 437 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 407, + 423, + 447, + 435 + ], + "score": 0.93, + "content": "\\mathbf { y } = [ c , d ]", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 423, + 505, + 437 + ], + "score": 1.0, + "content": "as an integral", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 433, + 244, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 433, + 244, + 447 + ], + "score": 1.0, + "content": "of the product of those indicators:", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "interline_equation", + "bbox": [ + 233, + 450, + 378, + 477 + ], + "lines": [ + { + "bbox": [ + 233, + 450, + 378, + 477 + ], + "spans": [ + { + "bbox": [ + 233, + 450, + 378, + 477 + ], + "score": 0.95, + "content": "p ( \\mathbf { x } \\wedge \\mathbf { y } ) = \\int _ { \\mathbb { R } } \\mathbb { 1 } _ { [ a , b ] } ( x ) \\mathbb { 1 } _ { [ c , d ] } ( x ) d x", + "type": "interline_equation", + "image_path": "65de705b3e3c43047eb214d038bfb961d3325075a55b7b8985e5db77c003751f.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 233, + 450, + 378, + 477 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 481, + 496, + 493 + ], + "lines": [ + { + "bbox": [ + 105, + 479, + 498, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 498, + 496 + ], + "score": 1.0, + "content": "since the product has support (and is equal to 1) only in the areas where the two intervals overlap.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 106, + 497, + 505, + 565 + ], + "lines": [ + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "score": 1.0, + "content": "A solution suggests itself in replacing these indicator functions with functions of infinite support. We", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 510, + 505, + 521 + ], + "spans": [ + { + "bbox": [ + 106, + 510, + 505, + 521 + ], + "score": 1.0, + "content": "elect for kernel smoothing, specifically convolution with a normalized Gaussian kernel, equivalent to", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 520, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 106, + 520, + 505, + 533 + ], + "score": 1.0, + "content": "an application of the diffusion equation to the original functional form of the embeddings (indicator", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 530, + 505, + 544 + ], + "spans": [ + { + "bbox": [ + 105, + 530, + 505, + 544 + ], + "score": 1.0, + "content": "functions) and a common approach to mollified optimization and energy smoothing (Neelakantan", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 541, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 506, + 555 + ], + "score": 1.0, + "content": "et al., 2015; Gulcehre et al., 2016b; Mobahi, 2016). This approach is demonstrated in one dimension", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 552, + 156, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 156, + 566 + ], + "score": 1.0, + "content": "in Figure 2.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 107, + 569, + 408, + 582 + ], + "lines": [ + { + "bbox": [ + 106, + 568, + 408, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 568, + 182, + 583 + ], + "score": 1.0, + "content": "Specifically, given", + "type": "text" + }, + { + "bbox": [ + 182, + 570, + 222, + 582 + ], + "score": 0.93, + "content": "\\mathbf { x } = [ a , b ]", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 568, + 408, + 583 + ], + "score": 1.0, + "content": ", we associate the smoothed indicator function", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "interline_equation", + "bbox": [ + 125, + 587, + 486, + 615 + ], + "lines": [ + { + "bbox": [ + 125, + 587, + 486, + 615 + ], + "spans": [ + { + "bbox": [ + 125, + 587, + 486, + 615 + ], + "score": 0.93, + "content": "f ( x ; a , b , \\sigma ^ { 2 } ) = \\mathbb { 1 } _ { [ a , b ] } ( x ) * \\phi ( x ; \\sigma ^ { 2 } ) = \\int _ { \\mathbb { R } } \\mathbb { 1 } _ { [ a , b ] } ( z ) \\phi ( x - z ; \\sigma ^ { 2 } ) d z = \\int _ { a } ^ { b } \\phi ( x - z ; \\sigma ^ { 2 } ) d z", + "type": "interline_equation", + "image_path": "f2a08231891c3a463390427732186839d80db3482d50ce1fc2ab5c2c47ad3a89.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 125, + 587, + 486, + 596.3333333333334 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 125, + 596.3333333333334, + 486, + 605.6666666666667 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 125, + 605.6666666666667, + 486, + 615.0000000000001 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 626, + 504, + 649 + ], + "lines": [ + { + "bbox": [ + 105, + 624, + 504, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 312, + 639 + ], + "score": 1.0, + "content": "We then wish to evaluate, for two lattice elements", + "type": "text" + }, + { + "bbox": [ + 312, + 628, + 320, + 636 + ], + "score": 0.75, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 624, + 338, + 639 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 338, + 628, + 346, + 637 + ], + "score": 0.45, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 624, + 497, + 639 + ], + "score": 1.0, + "content": "with associated smoothed indicators", + "type": "text" + }, + { + "bbox": [ + 497, + 627, + 504, + 638 + ], + "score": 0.83, + "content": "f", + "type": "inline_equation" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 635, + 136, + 651 + ], + "spans": [ + { + "bbox": [ + 104, + 635, + 123, + 651 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 639, + 129, + 649 + ], + "score": 0.79, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 130, + 635, + 136, + 651 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5 + }, + { + "type": "interline_equation", + "bbox": [ + 212, + 653, + 398, + 680 + ], + "lines": [ + { + "bbox": [ + 212, + 653, + 398, + 680 + ], + "spans": [ + { + "bbox": [ + 212, + 653, + 398, + 680 + ], + "score": 0.94, + "content": "p _ { \\phi } ( \\mathbf { x } \\wedge \\mathbf { y } ) = \\int _ { \\mathbb { R } } f ( x ; a , b , \\sigma _ { 1 } ^ { 2 } ) g ( x ; c , d , \\sigma _ { 2 } ^ { 2 } ) d x", + "type": "interline_equation", + "image_path": "0e7fe20b5e8fdaef9ca655bab7cbce18de5ca7a755561b768cfc86f9df7fcec6.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 212, + 653, + 398, + 680 + ], + "spans": [], + "index": 36 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 684, + 281, + 695 + ], + "lines": [ + { + "bbox": [ + 107, + 683, + 282, + 696 + ], + "spans": [ + { + "bbox": [ + 107, + 683, + 282, + 696 + ], + "score": 1.0, + "content": "This integral admits a closed form solution.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 106, + 708, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 185, + 723 + ], + "score": 1.0, + "content": "Proposition 1. Let", + "type": "text" + }, + { + "bbox": [ + 186, + 709, + 268, + 722 + ], + "score": 0.91, + "content": "\\begin{array} { r } { m _ { \\Phi } ( x ) = \\int \\Phi ( x ) d x } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 708, + 506, + 723 + ], + "score": 1.0, + "content": "be an antiderivative of the standard normal CDF. Then the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 719, + 243, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 243, + 734 + ], + "score": 1.0, + "content": "solution to equation 2 is given by,", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 505, + 149 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 504, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 504, + 95 + ], + "score": 1.0, + "content": "The hinge function has a large flat plateau at 0 when intervals are disjoint. This issue is especially", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 506, + 106 + ], + "score": 1.0, + "content": "problematic when the lattice to be embedded is (pseudo-)sparse, that is, most boxes should have", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 103, + 505, + 119 + ], + "spans": [ + { + "bbox": [ + 105, + 103, + 505, + 119 + ], + "score": 1.0, + "content": "very little or no intersection, since if training accidentally makes two boxes disjoint there is no way", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 105, + 115, + 505, + 128 + ], + "score": 1.0, + "content": "to recover with the naive measure. The authors propose a surrogate function to optimize in this case,", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 124, + 506, + 141 + ], + "spans": [ + { + "bbox": [ + 104, + 124, + 506, + 141 + ], + "score": 1.0, + "content": "but we will use a more principled framework to develop alternate measures that avoid this pathology,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 137, + 320, + 150 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 320, + 150 + ], + "score": 1.0, + "content": "improving both optimization and final model quality.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5, + "bbox_fs": [ + 104, + 82, + 506, + 150 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 162, + 225, + 173 + ], + "lines": [ + { + "bbox": [ + 106, + 162, + 227, + 174 + ], + "spans": [ + { + "bbox": [ + 106, + 162, + 227, + 174 + ], + "score": 1.0, + "content": "4.2 RELAXED GEOMETRY", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "image", + "bbox": [ + 112, + 183, + 498, + 284 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 112, + 183, + 498, + 284 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 112, + 183, + 498, + 284 + ], + "spans": [ + { + "bbox": [ + 112, + 183, + 498, + 284 + ], + "score": 0.964, + "type": "image", + "image_path": "52e273973b88b2eeb82ffad0414a0376e7873c9a60e503fb0bddeea225d759b0.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 112, + 183, + 498, + 216.66666666666666 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 112, + 216.66666666666666, + 498, + 250.33333333333331 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 112, + 250.33333333333331, + 498, + 284.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 293, + 505, + 326 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 292, + 506, + 306 + ], + "spans": [ + { + "bbox": [ + 106, + 292, + 506, + 306 + ], + "score": 1.0, + "content": "Figure 2: One-dimensional example demonstrating two disjoint indicators of intervals before and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 303, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 105, + 303, + 505, + 316 + ], + "score": 1.0, + "content": "after the application of a smoothing kernel. The area under the purple product curve is proportional", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 314, + 205, + 328 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 205, + 328 + ], + "score": 1.0, + "content": "to the degree of overlap.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 107, + 340, + 505, + 407 + ], + "lines": [ + { + "bbox": [ + 106, + 341, + 505, + 352 + ], + "spans": [ + { + "bbox": [ + 106, + 341, + 505, + 352 + ], + "score": 1.0, + "content": "The intuition behind our approach is that the “hard edges” of the standard box embeddings lead to", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 352, + 506, + 364 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 506, + 364 + ], + "score": 1.0, + "content": "unwanted gradient sparsity, and we seek a relaxation of this assumption that maintains the desirable", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 362, + 506, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 506, + 376 + ], + "score": 1.0, + "content": "properties of the base lattice model while enabling better optimization and preserving a geometric", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 375, + 505, + 385 + ], + "spans": [ + { + "bbox": [ + 106, + 375, + 505, + 385 + ], + "score": 1.0, + "content": "intuition. For ease of exposition, we will refer to 1-dimensional intervals in this section, but the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 384, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 505, + 396 + ], + "score": 1.0, + "content": "results carry through from the representation of boxes as products of intervals and their volumes", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 396, + 265, + 407 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 265, + 407 + ], + "score": 1.0, + "content": "under the associated product measures.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 15.5, + "bbox_fs": [ + 105, + 341, + 506, + 407 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 412, + 505, + 446 + ], + "lines": [ + { + "bbox": [ + 106, + 412, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 106, + 412, + 505, + 424 + ], + "score": 1.0, + "content": "The first observation is that, considering boxes as indicator functions of intervals, we can rewrite the", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 423, + 505, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 235, + 437 + ], + "score": 1.0, + "content": "measure of the joint probability", + "type": "text" + }, + { + "bbox": [ + 236, + 424, + 273, + 435 + ], + "score": 0.92, + "content": "p ( \\mathbf { x } \\wedge \\mathbf { y } )", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 423, + 347, + 437 + ], + "score": 1.0, + "content": "between intervals", + "type": "text" + }, + { + "bbox": [ + 348, + 423, + 388, + 435 + ], + "score": 0.93, + "content": "\\mathbf { x } = [ a , b ]", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 423, + 406, + 437 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 407, + 423, + 447, + 435 + ], + "score": 0.93, + "content": "\\mathbf { y } = [ c , d ]", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 423, + 505, + 437 + ], + "score": 1.0, + "content": "as an integral", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 433, + 244, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 433, + 244, + 447 + ], + "score": 1.0, + "content": "of the product of those indicators:", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 412, + 505, + 447 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 233, + 450, + 378, + 477 + ], + "lines": [ + { + "bbox": [ + 233, + 450, + 378, + 477 + ], + "spans": [ + { + "bbox": [ + 233, + 450, + 378, + 477 + ], + "score": 0.95, + "content": "p ( \\mathbf { x } \\wedge \\mathbf { y } ) = \\int _ { \\mathbb { R } } \\mathbb { 1 } _ { [ a , b ] } ( x ) \\mathbb { 1 } _ { [ c , d ] } ( x ) d x", + "type": "interline_equation", + "image_path": "65de705b3e3c43047eb214d038bfb961d3325075a55b7b8985e5db77c003751f.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 233, + 450, + 378, + 477 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 481, + 496, + 493 + ], + "lines": [ + { + "bbox": [ + 105, + 479, + 498, + 496 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 498, + 496 + ], + "score": 1.0, + "content": "since the product has support (and is equal to 1) only in the areas where the two intervals overlap.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 479, + 498, + 496 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 497, + 505, + 565 + ], + "lines": [ + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "score": 1.0, + "content": "A solution suggests itself in replacing these indicator functions with functions of infinite support. We", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 510, + 505, + 521 + ], + "spans": [ + { + "bbox": [ + 106, + 510, + 505, + 521 + ], + "score": 1.0, + "content": "elect for kernel smoothing, specifically convolution with a normalized Gaussian kernel, equivalent to", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 520, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 106, + 520, + 505, + 533 + ], + "score": 1.0, + "content": "an application of the diffusion equation to the original functional form of the embeddings (indicator", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 530, + 505, + 544 + ], + "spans": [ + { + "bbox": [ + 105, + 530, + 505, + 544 + ], + "score": 1.0, + "content": "functions) and a common approach to mollified optimization and energy smoothing (Neelakantan", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 541, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 506, + 555 + ], + "score": 1.0, + "content": "et al., 2015; Gulcehre et al., 2016b; Mobahi, 2016). This approach is demonstrated in one dimension", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 552, + 156, + 566 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 156, + 566 + ], + "score": 1.0, + "content": "in Figure 2.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 498, + 506, + 566 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 569, + 408, + 582 + ], + "lines": [ + { + "bbox": [ + 106, + 568, + 408, + 583 + ], + "spans": [ + { + "bbox": [ + 106, + 568, + 182, + 583 + ], + "score": 1.0, + "content": "Specifically, given", + "type": "text" + }, + { + "bbox": [ + 182, + 570, + 222, + 582 + ], + "score": 0.93, + "content": "\\mathbf { x } = [ a , b ]", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 568, + 408, + 583 + ], + "score": 1.0, + "content": ", we associate the smoothed indicator function", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30, + "bbox_fs": [ + 106, + 568, + 408, + 583 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 125, + 587, + 486, + 615 + ], + "lines": [ + { + "bbox": [ + 125, + 587, + 486, + 615 + ], + "spans": [ + { + "bbox": [ + 125, + 587, + 486, + 615 + ], + "score": 0.93, + "content": "f ( x ; a , b , \\sigma ^ { 2 } ) = \\mathbb { 1 } _ { [ a , b ] } ( x ) * \\phi ( x ; \\sigma ^ { 2 } ) = \\int _ { \\mathbb { R } } \\mathbb { 1 } _ { [ a , b ] } ( z ) \\phi ( x - z ; \\sigma ^ { 2 } ) d z = \\int _ { a } ^ { b } \\phi ( x - z ; \\sigma ^ { 2 } ) d z", + "type": "interline_equation", + "image_path": "f2a08231891c3a463390427732186839d80db3482d50ce1fc2ab5c2c47ad3a89.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 125, + 587, + 486, + 596.3333333333334 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 125, + 596.3333333333334, + 486, + 605.6666666666667 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 125, + 605.6666666666667, + 486, + 615.0000000000001 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 626, + 504, + 649 + ], + "lines": [ + { + "bbox": [ + 105, + 624, + 504, + 639 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 312, + 639 + ], + "score": 1.0, + "content": "We then wish to evaluate, for two lattice elements", + "type": "text" + }, + { + "bbox": [ + 312, + 628, + 320, + 636 + ], + "score": 0.75, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 624, + 338, + 639 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 338, + 628, + 346, + 637 + ], + "score": 0.45, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 347, + 624, + 497, + 639 + ], + "score": 1.0, + "content": "with associated smoothed indicators", + "type": "text" + }, + { + "bbox": [ + 497, + 627, + 504, + 638 + ], + "score": 0.83, + "content": "f", + "type": "inline_equation" + } + ], + "index": 34 + }, + { + "bbox": [ + 104, + 635, + 136, + 651 + ], + "spans": [ + { + "bbox": [ + 104, + 635, + 123, + 651 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 639, + 129, + 649 + ], + "score": 0.79, + "content": "g", + "type": "inline_equation" + }, + { + "bbox": [ + 130, + 635, + 136, + 651 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5, + "bbox_fs": [ + 104, + 624, + 504, + 651 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 212, + 653, + 398, + 680 + ], + "lines": [ + { + "bbox": [ + 212, + 653, + 398, + 680 + ], + "spans": [ + { + "bbox": [ + 212, + 653, + 398, + 680 + ], + "score": 0.94, + "content": "p _ { \\phi } ( \\mathbf { x } \\wedge \\mathbf { y } ) = \\int _ { \\mathbb { R } } f ( x ; a , b , \\sigma _ { 1 } ^ { 2 } ) g ( x ; c , d , \\sigma _ { 2 } ^ { 2 } ) d x", + "type": "interline_equation", + "image_path": "0e7fe20b5e8fdaef9ca655bab7cbce18de5ca7a755561b768cfc86f9df7fcec6.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 212, + 653, + 398, + 680 + ], + "spans": [], + "index": 36 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 684, + 281, + 695 + ], + "lines": [ + { + "bbox": [ + 107, + 683, + 282, + 696 + ], + "spans": [ + { + "bbox": [ + 107, + 683, + 282, + 696 + ], + "score": 1.0, + "content": "This integral admits a closed form solution.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37, + "bbox_fs": [ + 107, + 683, + 282, + 696 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 708, + 504, + 732 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 506, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 185, + 723 + ], + "score": 1.0, + "content": "Proposition 1. Let", + "type": "text" + }, + { + "bbox": [ + 186, + 709, + 268, + 722 + ], + "score": 0.91, + "content": "\\begin{array} { r } { m _ { \\Phi } ( x ) = \\int \\Phi ( x ) d x } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 708, + 506, + 723 + ], + "score": 1.0, + "content": "be an antiderivative of the standard normal CDF. Then the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 719, + 243, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 243, + 734 + ], + "score": 1.0, + "content": "solution to equation 2 is given by,", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 105, + 708, + 506, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "interline_equation", + "bbox": [ + 155, + 96, + 455, + 131 + ], + "lines": [ + { + "bbox": [ + 155, + 96, + 455, + 131 + ], + "spans": [ + { + "bbox": [ + 155, + 96, + 455, + 131 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { p _ { \\phi } ( \\mathbf { x } \\wedge \\mathbf { y } ) = \\sigma \\left( m _ { \\Phi } ( \\frac { b - c } { \\sigma } ) + m _ { \\Phi } ( \\frac { a - d } { \\sigma } ) - m _ { \\Phi } ( \\frac { b - d } { \\sigma } ) - m _ { \\Phi } ( \\frac { a - c } { \\sigma } ) \\right) } \\\\ & { \\qquad \\approx \\left( \\rho \\operatorname { s o f t } ( \\frac { b - c } { \\rho } ) + \\rho \\operatorname { s o f t } ( \\frac { a - d } { \\rho } ) \\right) - \\left( \\rho \\operatorname { s o f t } ( \\frac { b - d } { \\rho } ) + \\rho \\operatorname { s o f t } ( \\frac { a - c } { \\rho } ) \\right) } \\end{array}", + "type": "interline_equation", + "image_path": "dfced9ab50f8fb9bf2009489d2a8081d61924d11e4adaa5c2093964d30174ff3.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 155, + 96, + 455, + 107.66666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 155, + 107.66666666666667, + 455, + 119.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 155, + 119.33333333333334, + 455, + 131.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 141, + 506, + 168 + ], + "lines": [ + { + "bbox": [ + 105, + 140, + 506, + 159 + ], + "spans": [ + { + "bbox": [ + 105, + 140, + 133, + 159 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 141, + 196, + 155 + ], + "score": 0.9, + "content": "\\sigma = \\sqrt { \\sigma _ { 1 } ^ { 2 } + \\sigma _ { 2 } ^ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 140, + 200, + 159 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 201, + 142, + 311, + 155 + ], + "score": 0.87, + "content": "\\operatorname { s o f t } ( x ) = \\log ( 1 + \\exp ( x ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 140, + 506, + 159 + ], + "score": 1.0, + "content": "is the softplus function, the antiderivative of the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 152, + 237, + 169 + ], + "spans": [ + { + "bbox": [ + 105, + 152, + 237, + 169 + ], + "score": 1.0, + "content": "logistic sigmoid, and ρ = σ1.702 .", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 106, + 177, + 504, + 200 + ], + "lines": [ + { + "bbox": [ + 106, + 177, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 106, + 177, + 505, + 190 + ], + "score": 1.0, + "content": "Proof. The first line is proved in Appendix A, the second approximation follows from the approxi-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 188, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 106, + 188, + 147, + 201 + ], + "score": 1.0, + "content": "mation of", + "type": "text" + }, + { + "bbox": [ + 148, + 189, + 156, + 198 + ], + "score": 0.84, + "content": "\\Phi", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 188, + 367, + 201 + ], + "score": 1.0, + "content": "by a logistic sigmoid given in Bowling et al. (2009).", + "type": "text" + }, + { + "bbox": [ + 495, + 190, + 505, + 199 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 106, + 212, + 434, + 224 + ], + "lines": [ + { + "bbox": [ + 105, + 211, + 434, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 211, + 277, + 225 + ], + "score": 1.0, + "content": "Note that, in the zero-temperature limit, as", + "type": "text" + }, + { + "bbox": [ + 278, + 214, + 285, + 224 + ], + "score": 0.81, + "content": "\\rho", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 211, + 434, + 225 + ], + "score": 1.0, + "content": "goes to zero, we recover the formula", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "interline_equation", + "bbox": [ + 146, + 226, + 465, + 279 + ], + "lines": [ + { + "bbox": [ + 146, + 226, + 465, + 279 + ], + "spans": [ + { + "bbox": [ + 146, + 226, + 465, + 279 + ], + "score": 0.92, + "content": "{ \\begin{array} { l } { p _ { \\phi } ( \\mathbf { x } \\wedge \\mathbf { y } ) = \\operatorname* { l i m } _ { \\rho \\to 0 } \\left( \\rho { \\mathrm { s o f t } } ( { \\frac { b - c } { \\rho } } ) + \\rho { \\mathrm { s o f t } } ( { \\frac { a - d } { \\rho } } ) \\right) - \\left( \\rho { \\mathrm { s o f t } } ( { \\frac { b - d } { \\rho } } ) + \\rho { \\mathrm { s o f t } } ( { \\frac { a - c } { \\rho } } ) \\right) } \\\\ { \\qquad = \\left( m _ { h } ( b - c ) + m _ { h } ( a - d ) \\right) - \\left( m _ { h } ( b - d ) + m _ { h } ( a - c ) \\right) } \\\\ { \\qquad = m _ { h } ( b \\wedge d - a \\vee c ) } \\end{array} }", + "type": "interline_equation", + "image_path": "03b772feac1e3a638d4c6b0c9b9b017d4732d377a5de9395086d7fcff5ac59f8.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 146, + 226, + 465, + 243.66666666666666 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 146, + 243.66666666666666, + 465, + 261.3333333333333 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 146, + 261.3333333333333, + 465, + 279.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 281, + 503, + 326 + ], + "lines": [ + { + "bbox": [ + 106, + 281, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 106, + 281, + 262, + 294 + ], + "score": 1.0, + "content": "with equality in the last line because", + "type": "text" + }, + { + "bbox": [ + 262, + 281, + 285, + 294 + ], + "score": 0.92, + "content": "( a , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 281, + 306, + 294 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 306, + 281, + 328, + 294 + ], + "score": 0.92, + "content": "( c , d )", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 281, + 505, + 294 + ], + "score": 1.0, + "content": "are intervals. This last line is exactly our", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 293, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 505, + 304 + ], + "score": 1.0, + "content": "original equation equation 1, which is expected from convolution with a zero-bandwidth kernel (a", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 304, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 304, + 505, + 316 + ], + "score": 1.0, + "content": "Dirac delta function, the identity element under convolution). This is true for both the exact formula", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 314, + 304, + 328 + ], + "spans": [ + { + "bbox": [ + 106, + 315, + 131, + 328 + ], + "score": 1.0, + "content": "using", + "type": "text" + }, + { + "bbox": [ + 131, + 314, + 172, + 327 + ], + "score": 0.93, + "content": "\\textstyle \\int \\Phi ( x ) d x", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 315, + 304, + 328 + ], + "score": 1.0, + "content": ", and the softplus approximation.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 106, + 331, + 505, + 365 + ], + "lines": [ + { + "bbox": [ + 106, + 330, + 506, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 330, + 198, + 344 + ], + "score": 1.0, + "content": "Unfortunately, for any", + "type": "text" + }, + { + "bbox": [ + 198, + 332, + 223, + 343 + ], + "score": 0.9, + "content": "\\rho > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 330, + 506, + 344 + ], + "score": 1.0, + "content": ", multiplication of Gaussian-smoothed indicators does not give a valid", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 342, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 352, + 356 + ], + "score": 1.0, + "content": "meet operation on a function lattice, for the simple reason that", + "type": "text" + }, + { + "bbox": [ + 352, + 342, + 382, + 354 + ], + "score": 0.92, + "content": "f ^ { 2 } \\neq f", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 342, + 506, + 356 + ], + "score": 1.0, + "content": ", except in the case of indicator", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 353, + 367, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 367, + 366 + ], + "score": 1.0, + "content": "functions, violating the idempotency requirement of Section 3.1.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 107, + 370, + 504, + 393 + ], + "lines": [ + { + "bbox": [ + 105, + 369, + 506, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 425, + 384 + ], + "score": 1.0, + "content": "More importantly, for practical considerations, if we are to treat the outputs of", + "type": "text" + }, + { + "bbox": [ + 426, + 372, + 437, + 383 + ], + "score": 0.87, + "content": "p _ { \\phi }", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 369, + 506, + 384 + ], + "score": 1.0, + "content": "as probabilities,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 381, + 184, + 393 + ], + "spans": [ + { + "bbox": [ + 106, + 381, + 184, + 393 + ], + "score": 1.0, + "content": "the consequence is", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 397, + 386, + 424 + ], + "lines": [ + { + "bbox": [ + 225, + 397, + 386, + 424 + ], + "spans": [ + { + "bbox": [ + 225, + 397, + 386, + 424 + ], + "score": 0.94, + "content": "p _ { \\phi } ( \\mathbf { x } | \\mathbf { x } ) = \\frac { p _ { \\phi } ( \\mathbf { x } , \\mathbf { x } ) } { p _ { \\phi } ( \\mathbf { x } ) } = \\frac { p _ { \\phi } ( \\mathbf { x } \\wedge \\mathbf { x } ) } { p _ { \\phi } ( \\mathbf { x } ) } \\neq 1", + "type": "interline_equation", + "image_path": "876a5a3f61060e6111fc2326df5ec8f61afae5c9aa38ad91752cad5d0e64273d.jpg" + } + ] + } + ], + "index": 20.5, + "virtual_lines": [ + { + "bbox": [ + 225, + 397, + 386, + 410.5 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 225, + 410.5, + 386, + 424.0 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 427, + 505, + 461 + ], + "lines": [ + { + "bbox": [ + 106, + 428, + 504, + 440 + ], + "spans": [ + { + "bbox": [ + 106, + 428, + 504, + 440 + ], + "score": 1.0, + "content": "which complicates our applications that train on conditional probabilities. However, by a modifica-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 438, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 280, + 452 + ], + "score": 1.0, + "content": "tion of equation 3, we can obtain a function", + "type": "text" + }, + { + "bbox": [ + 280, + 441, + 287, + 450 + ], + "score": 0.81, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 438, + 326, + 452 + ], + "score": 1.0, + "content": "such that", + "type": "text" + }, + { + "bbox": [ + 326, + 438, + 393, + 451 + ], + "score": 0.94, + "content": "p ( \\mathbf { x } \\wedge \\mathbf { \\bar { x } } ) = p ( \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 438, + 505, + 452 + ], + "score": 1.0, + "content": ", while retaining the smooth", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 450, + 296, + 461 + ], + "spans": [ + { + "bbox": [ + 106, + 450, + 296, + 461 + ], + "score": 1.0, + "content": "optimization properties of the Gaussian model.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 108, + 466, + 432, + 478 + ], + "lines": [ + { + "bbox": [ + 106, + 465, + 432, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 241, + 479 + ], + "score": 1.0, + "content": "Recall that for the hinge function", + "type": "text" + }, + { + "bbox": [ + 241, + 469, + 256, + 478 + ], + "score": 0.87, + "content": "m _ { h }", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 465, + 329, + 479 + ], + "score": 1.0, + "content": "and two intervals", + "type": "text" + }, + { + "bbox": [ + 329, + 466, + 351, + 479 + ], + "score": 0.92, + "content": "( a , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 465, + 370, + 479 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 370, + 466, + 392, + 479 + ], + "score": 0.93, + "content": "( c , d )", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 465, + 432, + 479 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "interline_equation", + "bbox": [ + 147, + 483, + 465, + 498 + ], + "lines": [ + { + "bbox": [ + 147, + 483, + 465, + 498 + ], + "spans": [ + { + "bbox": [ + 147, + 483, + 465, + 498 + ], + "score": 0.89, + "content": "\\bigl ( m _ { h } ( b - c ) + m _ { h } ( a - d ) \\bigr ) - \\bigl ( m _ { h } ( b - d ) + m _ { h } ( a - c ) \\bigr ) = m _ { h } ( b \\wedge d - a \\vee c )", + "type": "interline_equation", + "image_path": "ded1db3dd855abd51ac121bf2cea0a3cbbb86c25e931235f4837f6642e38cec3.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 147, + 483, + 465, + 498 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 501, + 503, + 524 + ], + "lines": [ + { + "bbox": [ + 106, + 501, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 106, + 501, + 505, + 514 + ], + "score": 1.0, + "content": "where the left hand side is the zero-temperature limit of the Gaussian model from equation 3. This", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 512, + 388, + 524 + ], + "spans": [ + { + "bbox": [ + 106, + 512, + 252, + 524 + ], + "score": 1.0, + "content": "identity is true of the hinge function", + "type": "text" + }, + { + "bbox": [ + 252, + 514, + 267, + 523 + ], + "score": 0.87, + "content": "m _ { h }", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 512, + 388, + 524 + ], + "score": 1.0, + "content": ", but not the softplus function.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 106, + 528, + 505, + 573 + ], + "lines": [ + { + "bbox": [ + 106, + 529, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 529, + 505, + 541 + ], + "score": 1.0, + "content": "However, an equation with a similar functional form as equation 6 (on both the left- and right-hand", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 540, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 505, + 552 + ], + "score": 1.0, + "content": "sides) is true not only of the hinge function from the unsmoothed model, but also true of the softplus.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 550, + 505, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 177, + 564 + ], + "score": 1.0, + "content": "For two intervals", + "type": "text" + }, + { + "bbox": [ + 177, + 551, + 221, + 563 + ], + "score": 0.93, + "content": "\\mathbf { x } = ( a , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 550, + 234, + 564 + ], + "score": 1.0, + "content": "an", + "type": "text" + }, + { + "bbox": [ + 234, + 551, + 278, + 563 + ], + "score": 0.93, + "content": "\\mathbf { y } = ( c , d )", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 550, + 505, + 564 + ], + "score": 1.0, + "content": ", by the commutativity of min and max with monotonic", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 561, + 185, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 185, + 574 + ], + "score": 1.0, + "content": "functions, we have", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30.5 + }, + { + "type": "interline_equation", + "bbox": [ + 142, + 577, + 470, + 592 + ], + "lines": [ + { + "bbox": [ + 142, + 577, + 470, + 592 + ], + "spans": [ + { + "bbox": [ + 142, + 577, + 470, + 592 + ], + "score": 0.89, + "content": "{ \\bigl ( } \\operatorname { s o f t } ( b - c ) \\lor \\operatorname { s o f t } ( a - d ) { \\bigr ) } \\land { \\bigl ( } \\operatorname { s o f t } ( b - d ) \\lor \\operatorname { s o f t } ( a - c ) { \\bigr ) } = \\operatorname { s o f t } ( b \\land d - a \\lor c )", + "type": "interline_equation", + "image_path": "a6ac4e6db7c1097f3f21faf20c6218ee82ade51dc605203fb45b069ff7c2d83d.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 142, + 577, + 470, + 592 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 595, + 503, + 630 + ], + "lines": [ + { + "bbox": [ + 105, + 595, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 505, + 608 + ], + "score": 1.0, + "content": "In the zero-temperature limit, all terms in equations 3 and 7 are equivalent. However, outside of this,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 606, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 224, + 620 + ], + "score": 1.0, + "content": "equation 7 is idempotent for", + "type": "text" + }, + { + "bbox": [ + 225, + 606, + 328, + 618 + ], + "score": 0.92, + "content": "\\mathbf { x } = \\mathbf { y } = ( { a } , { \\bar { b } } ) = ( { c } , { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 328, + 606, + 505, + 620 + ], + "score": 1.0, + "content": "(when considered as a measure of overlap,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 618, + 350, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 618, + 350, + 630 + ], + "score": 1.0, + "content": "made precise in the next paragraph), while equation 3 is not.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 106, + 633, + 503, + 657 + ], + "lines": [ + { + "bbox": [ + 106, + 633, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 633, + 273, + 648 + ], + "score": 1.0, + "content": "This inspires us to define the probabilities", + "type": "text" + }, + { + "bbox": [ + 273, + 634, + 293, + 646 + ], + "score": 0.92, + "content": "p ( \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 633, + 310, + 648 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 311, + 634, + 341, + 646 + ], + "score": 0.92, + "content": "p ( \\mathbf { x } , \\mathbf { y } )", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 633, + 505, + 648 + ], + "score": 1.0, + "content": "using a normalized version of equation 7", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 645, + 418, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 645, + 418, + 658 + ], + "score": 1.0, + "content": "in place of equation 3. For the interval (one-dimensional box) case, we define", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + }, + { + "type": "interline_equation", + "bbox": [ + 244, + 660, + 367, + 689 + ], + "lines": [ + { + "bbox": [ + 244, + 660, + 367, + 689 + ], + "spans": [ + { + "bbox": [ + 244, + 660, + 367, + 689 + ], + "score": 0.92, + "content": "\\begin{array} { c } { p ( \\mathbf { x } ) \\propto \\mathrm { s o f t } ( b - a ) } \\\\ { p ( \\mathbf { x } , \\mathbf { y } ) \\propto \\mathrm { s o f t } ( b \\wedge d - a \\vee c ) } \\end{array}", + "type": "interline_equation", + "image_path": "830b9a4b00c4c680073a6cb7e9ac7debbb7d84eebb88c836878a7c616164b43c.jpg" + } + ] + } + ], + "index": 39.5, + "virtual_lines": [ + { + "bbox": [ + 244, + 660, + 367, + 674.5 + ], + "spans": [], + "index": 39 + }, + { + "bbox": [ + 244, + 674.5, + 367, + 689.0 + ], + "spans": [], + "index": 40 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 692, + 354, + 705 + ], + "lines": [ + { + "bbox": [ + 105, + 691, + 351, + 707 + ], + "spans": [ + { + "bbox": [ + 105, + 691, + 289, + 707 + ], + "score": 1.0, + "content": "which satisfies the idempotency requirement,", + "type": "text" + }, + { + "bbox": [ + 289, + 693, + 351, + 705 + ], + "score": 0.92, + "content": "p ( \\mathbf { x } ) = p ( \\mathbf { x } , \\mathbf { x } )", + "type": "inline_equation" + } + ], + "index": 41 + } + ], + "index": 41 + }, + { + "type": "text", + "bbox": [ + 105, + 709, + 506, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "Because softplus upper-bounds the hinge function, it is capable of outputting values that are greater", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 719, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 505, + 734 + ], + "score": 1.0, + "content": "than 1, and therefore must be normalized. In our experiments, we use two different approaches to", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 760 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 310, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 310, + 762 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "interline_equation", + "bbox": [ + 155, + 96, + 455, + 131 + ], + "lines": [ + { + "bbox": [ + 155, + 96, + 455, + 131 + ], + "spans": [ + { + "bbox": [ + 155, + 96, + 455, + 131 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { p _ { \\phi } ( \\mathbf { x } \\wedge \\mathbf { y } ) = \\sigma \\left( m _ { \\Phi } ( \\frac { b - c } { \\sigma } ) + m _ { \\Phi } ( \\frac { a - d } { \\sigma } ) - m _ { \\Phi } ( \\frac { b - d } { \\sigma } ) - m _ { \\Phi } ( \\frac { a - c } { \\sigma } ) \\right) } \\\\ & { \\qquad \\approx \\left( \\rho \\operatorname { s o f t } ( \\frac { b - c } { \\rho } ) + \\rho \\operatorname { s o f t } ( \\frac { a - d } { \\rho } ) \\right) - \\left( \\rho \\operatorname { s o f t } ( \\frac { b - d } { \\rho } ) + \\rho \\operatorname { s o f t } ( \\frac { a - c } { \\rho } ) \\right) } \\end{array}", + "type": "interline_equation", + "image_path": "dfced9ab50f8fb9bf2009489d2a8081d61924d11e4adaa5c2093964d30174ff3.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 155, + 96, + 455, + 107.66666666666667 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 155, + 107.66666666666667, + 455, + 119.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 155, + 119.33333333333334, + 455, + 131.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 141, + 506, + 168 + ], + "lines": [ + { + "bbox": [ + 105, + 140, + 506, + 159 + ], + "spans": [ + { + "bbox": [ + 105, + 140, + 133, + 159 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 141, + 196, + 155 + ], + "score": 0.9, + "content": "\\sigma = \\sqrt { \\sigma _ { 1 } ^ { 2 } + \\sigma _ { 2 } ^ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 140, + 200, + 159 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 201, + 142, + 311, + 155 + ], + "score": 0.87, + "content": "\\operatorname { s o f t } ( x ) = \\log ( 1 + \\exp ( x ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 140, + 506, + 159 + ], + "score": 1.0, + "content": "is the softplus function, the antiderivative of the", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 152, + 237, + 169 + ], + "spans": [ + { + "bbox": [ + 105, + 152, + 237, + 169 + ], + "score": 1.0, + "content": "logistic sigmoid, and ρ = σ1.702 .", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5, + "bbox_fs": [ + 105, + 140, + 506, + 169 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 177, + 504, + 200 + ], + "lines": [ + { + "bbox": [ + 106, + 177, + 505, + 190 + ], + "spans": [ + { + "bbox": [ + 106, + 177, + 505, + 190 + ], + "score": 1.0, + "content": "Proof. The first line is proved in Appendix A, the second approximation follows from the approxi-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 188, + 505, + 201 + ], + "spans": [ + { + "bbox": [ + 106, + 188, + 147, + 201 + ], + "score": 1.0, + "content": "mation of", + "type": "text" + }, + { + "bbox": [ + 148, + 189, + 156, + 198 + ], + "score": 0.84, + "content": "\\Phi", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 188, + 367, + 201 + ], + "score": 1.0, + "content": "by a logistic sigmoid given in Bowling et al. (2009).", + "type": "text" + }, + { + "bbox": [ + 495, + 190, + 505, + 199 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 106, + 177, + 505, + 201 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 212, + 434, + 224 + ], + "lines": [ + { + "bbox": [ + 105, + 211, + 434, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 211, + 277, + 225 + ], + "score": 1.0, + "content": "Note that, in the zero-temperature limit, as", + "type": "text" + }, + { + "bbox": [ + 278, + 214, + 285, + 224 + ], + "score": 0.81, + "content": "\\rho", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 211, + 434, + 225 + ], + "score": 1.0, + "content": "goes to zero, we recover the formula", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7, + "bbox_fs": [ + 105, + 211, + 434, + 225 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 146, + 226, + 465, + 279 + ], + "lines": [ + { + "bbox": [ + 146, + 226, + 465, + 279 + ], + "spans": [ + { + "bbox": [ + 146, + 226, + 465, + 279 + ], + "score": 0.92, + "content": "{ \\begin{array} { l } { p _ { \\phi } ( \\mathbf { x } \\wedge \\mathbf { y } ) = \\operatorname* { l i m } _ { \\rho \\to 0 } \\left( \\rho { \\mathrm { s o f t } } ( { \\frac { b - c } { \\rho } } ) + \\rho { \\mathrm { s o f t } } ( { \\frac { a - d } { \\rho } } ) \\right) - \\left( \\rho { \\mathrm { s o f t } } ( { \\frac { b - d } { \\rho } } ) + \\rho { \\mathrm { s o f t } } ( { \\frac { a - c } { \\rho } } ) \\right) } \\\\ { \\qquad = \\left( m _ { h } ( b - c ) + m _ { h } ( a - d ) \\right) - \\left( m _ { h } ( b - d ) + m _ { h } ( a - c ) \\right) } \\\\ { \\qquad = m _ { h } ( b \\wedge d - a \\vee c ) } \\end{array} }", + "type": "interline_equation", + "image_path": "03b772feac1e3a638d4c6b0c9b9b017d4732d377a5de9395086d7fcff5ac59f8.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 146, + 226, + 465, + 243.66666666666666 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 146, + 243.66666666666666, + 465, + 261.3333333333333 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 146, + 261.3333333333333, + 465, + 279.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 281, + 503, + 326 + ], + "lines": [ + { + "bbox": [ + 106, + 281, + 505, + 294 + ], + "spans": [ + { + "bbox": [ + 106, + 281, + 262, + 294 + ], + "score": 1.0, + "content": "with equality in the last line because", + "type": "text" + }, + { + "bbox": [ + 262, + 281, + 285, + 294 + ], + "score": 0.92, + "content": "( a , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 281, + 306, + 294 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 306, + 281, + 328, + 294 + ], + "score": 0.92, + "content": "( c , d )", + "type": "inline_equation" + }, + { + "bbox": [ + 329, + 281, + 505, + 294 + ], + "score": 1.0, + "content": "are intervals. This last line is exactly our", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 293, + 505, + 304 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 505, + 304 + ], + "score": 1.0, + "content": "original equation equation 1, which is expected from convolution with a zero-bandwidth kernel (a", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 304, + 505, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 304, + 505, + 316 + ], + "score": 1.0, + "content": "Dirac delta function, the identity element under convolution). This is true for both the exact formula", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 314, + 304, + 328 + ], + "spans": [ + { + "bbox": [ + 106, + 315, + 131, + 328 + ], + "score": 1.0, + "content": "using", + "type": "text" + }, + { + "bbox": [ + 131, + 314, + 172, + 327 + ], + "score": 0.93, + "content": "\\textstyle \\int \\Phi ( x ) d x", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 315, + 304, + 328 + ], + "score": 1.0, + "content": ", and the softplus approximation.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 281, + 505, + 328 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 331, + 505, + 365 + ], + "lines": [ + { + "bbox": [ + 106, + 330, + 506, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 330, + 198, + 344 + ], + "score": 1.0, + "content": "Unfortunately, for any", + "type": "text" + }, + { + "bbox": [ + 198, + 332, + 223, + 343 + ], + "score": 0.9, + "content": "\\rho > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 330, + 506, + 344 + ], + "score": 1.0, + "content": ", multiplication of Gaussian-smoothed indicators does not give a valid", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 342, + 506, + 356 + ], + "spans": [ + { + "bbox": [ + 105, + 342, + 352, + 356 + ], + "score": 1.0, + "content": "meet operation on a function lattice, for the simple reason that", + "type": "text" + }, + { + "bbox": [ + 352, + 342, + 382, + 354 + ], + "score": 0.92, + "content": "f ^ { 2 } \\neq f", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 342, + 506, + 356 + ], + "score": 1.0, + "content": ", except in the case of indicator", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 353, + 367, + 366 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 367, + 366 + ], + "score": 1.0, + "content": "functions, violating the idempotency requirement of Section 3.1.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 330, + 506, + 366 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 370, + 504, + 393 + ], + "lines": [ + { + "bbox": [ + 105, + 369, + 506, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 425, + 384 + ], + "score": 1.0, + "content": "More importantly, for practical considerations, if we are to treat the outputs of", + "type": "text" + }, + { + "bbox": [ + 426, + 372, + 437, + 383 + ], + "score": 0.87, + "content": "p _ { \\phi }", + "type": "inline_equation" + }, + { + "bbox": [ + 438, + 369, + 506, + 384 + ], + "score": 1.0, + "content": "as probabilities,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 381, + 184, + 393 + ], + "spans": [ + { + "bbox": [ + 106, + 381, + 184, + 393 + ], + "score": 1.0, + "content": "the consequence is", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 369, + 506, + 393 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 397, + 386, + 424 + ], + "lines": [ + { + "bbox": [ + 225, + 397, + 386, + 424 + ], + "spans": [ + { + "bbox": [ + 225, + 397, + 386, + 424 + ], + "score": 0.94, + "content": "p _ { \\phi } ( \\mathbf { x } | \\mathbf { x } ) = \\frac { p _ { \\phi } ( \\mathbf { x } , \\mathbf { x } ) } { p _ { \\phi } ( \\mathbf { x } ) } = \\frac { p _ { \\phi } ( \\mathbf { x } \\wedge \\mathbf { x } ) } { p _ { \\phi } ( \\mathbf { x } ) } \\neq 1", + "type": "interline_equation", + "image_path": "876a5a3f61060e6111fc2326df5ec8f61afae5c9aa38ad91752cad5d0e64273d.jpg" + } + ] + } + ], + "index": 20.5, + "virtual_lines": [ + { + "bbox": [ + 225, + 397, + 386, + 410.5 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 225, + 410.5, + 386, + 424.0 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 427, + 505, + 461 + ], + "lines": [ + { + "bbox": [ + 106, + 428, + 504, + 440 + ], + "spans": [ + { + "bbox": [ + 106, + 428, + 504, + 440 + ], + "score": 1.0, + "content": "which complicates our applications that train on conditional probabilities. However, by a modifica-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 438, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 280, + 452 + ], + "score": 1.0, + "content": "tion of equation 3, we can obtain a function", + "type": "text" + }, + { + "bbox": [ + 280, + 441, + 287, + 450 + ], + "score": 0.81, + "content": "p", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 438, + 326, + 452 + ], + "score": 1.0, + "content": "such that", + "type": "text" + }, + { + "bbox": [ + 326, + 438, + 393, + 451 + ], + "score": 0.94, + "content": "p ( \\mathbf { x } \\wedge \\mathbf { \\bar { x } } ) = p ( \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 438, + 505, + 452 + ], + "score": 1.0, + "content": ", while retaining the smooth", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 450, + 296, + 461 + ], + "spans": [ + { + "bbox": [ + 106, + 450, + 296, + 461 + ], + "score": 1.0, + "content": "optimization properties of the Gaussian model.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 428, + 505, + 461 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 466, + 432, + 478 + ], + "lines": [ + { + "bbox": [ + 106, + 465, + 432, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 241, + 479 + ], + "score": 1.0, + "content": "Recall that for the hinge function", + "type": "text" + }, + { + "bbox": [ + 241, + 469, + 256, + 478 + ], + "score": 0.87, + "content": "m _ { h }", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 465, + 329, + 479 + ], + "score": 1.0, + "content": "and two intervals", + "type": "text" + }, + { + "bbox": [ + 329, + 466, + 351, + 479 + ], + "score": 0.92, + "content": "( a , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 465, + 370, + 479 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 370, + 466, + 392, + 479 + ], + "score": 0.93, + "content": "( c , d )", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 465, + 432, + 479 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25, + "bbox_fs": [ + 106, + 465, + 432, + 479 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 147, + 483, + 465, + 498 + ], + "lines": [ + { + "bbox": [ + 147, + 483, + 465, + 498 + ], + "spans": [ + { + "bbox": [ + 147, + 483, + 465, + 498 + ], + "score": 0.89, + "content": "\\bigl ( m _ { h } ( b - c ) + m _ { h } ( a - d ) \\bigr ) - \\bigl ( m _ { h } ( b - d ) + m _ { h } ( a - c ) \\bigr ) = m _ { h } ( b \\wedge d - a \\vee c )", + "type": "interline_equation", + "image_path": "ded1db3dd855abd51ac121bf2cea0a3cbbb86c25e931235f4837f6642e38cec3.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 147, + 483, + 465, + 498 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 501, + 503, + 524 + ], + "lines": [ + { + "bbox": [ + 106, + 501, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 106, + 501, + 505, + 514 + ], + "score": 1.0, + "content": "where the left hand side is the zero-temperature limit of the Gaussian model from equation 3. This", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 512, + 388, + 524 + ], + "spans": [ + { + "bbox": [ + 106, + 512, + 252, + 524 + ], + "score": 1.0, + "content": "identity is true of the hinge function", + "type": "text" + }, + { + "bbox": [ + 252, + 514, + 267, + 523 + ], + "score": 0.87, + "content": "m _ { h }", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 512, + 388, + 524 + ], + "score": 1.0, + "content": ", but not the softplus function.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5, + "bbox_fs": [ + 106, + 501, + 505, + 524 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 528, + 505, + 573 + ], + "lines": [ + { + "bbox": [ + 106, + 529, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 529, + 505, + 541 + ], + "score": 1.0, + "content": "However, an equation with a similar functional form as equation 6 (on both the left- and right-hand", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 540, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 540, + 505, + 552 + ], + "score": 1.0, + "content": "sides) is true not only of the hinge function from the unsmoothed model, but also true of the softplus.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 550, + 505, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 177, + 564 + ], + "score": 1.0, + "content": "For two intervals", + "type": "text" + }, + { + "bbox": [ + 177, + 551, + 221, + 563 + ], + "score": 0.93, + "content": "\\mathbf { x } = ( a , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 550, + 234, + 564 + ], + "score": 1.0, + "content": "an", + "type": "text" + }, + { + "bbox": [ + 234, + 551, + 278, + 563 + ], + "score": 0.93, + "content": "\\mathbf { y } = ( c , d )", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 550, + 505, + 564 + ], + "score": 1.0, + "content": ", by the commutativity of min and max with monotonic", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 561, + 185, + 574 + ], + "spans": [ + { + "bbox": [ + 105, + 561, + 185, + 574 + ], + "score": 1.0, + "content": "functions, we have", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 529, + 505, + 574 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 142, + 577, + 470, + 592 + ], + "lines": [ + { + "bbox": [ + 142, + 577, + 470, + 592 + ], + "spans": [ + { + "bbox": [ + 142, + 577, + 470, + 592 + ], + "score": 0.89, + "content": "{ \\bigl ( } \\operatorname { s o f t } ( b - c ) \\lor \\operatorname { s o f t } ( a - d ) { \\bigr ) } \\land { \\bigl ( } \\operatorname { s o f t } ( b - d ) \\lor \\operatorname { s o f t } ( a - c ) { \\bigr ) } = \\operatorname { s o f t } ( b \\land d - a \\lor c )", + "type": "interline_equation", + "image_path": "a6ac4e6db7c1097f3f21faf20c6218ee82ade51dc605203fb45b069ff7c2d83d.jpg" + } + ] + } + ], + "index": 33, + "virtual_lines": [ + { + "bbox": [ + 142, + 577, + 470, + 592 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 595, + 503, + 630 + ], + "lines": [ + { + "bbox": [ + 105, + 595, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 595, + 505, + 608 + ], + "score": 1.0, + "content": "In the zero-temperature limit, all terms in equations 3 and 7 are equivalent. However, outside of this,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 606, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 224, + 620 + ], + "score": 1.0, + "content": "equation 7 is idempotent for", + "type": "text" + }, + { + "bbox": [ + 225, + 606, + 328, + 618 + ], + "score": 0.92, + "content": "\\mathbf { x } = \\mathbf { y } = ( { a } , { \\bar { b } } ) = ( { c } , { d } )", + "type": "inline_equation" + }, + { + "bbox": [ + 328, + 606, + 505, + 620 + ], + "score": 1.0, + "content": "(when considered as a measure of overlap,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 618, + 350, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 618, + 350, + 630 + ], + "score": 1.0, + "content": "made precise in the next paragraph), while equation 3 is not.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35, + "bbox_fs": [ + 105, + 595, + 505, + 630 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 633, + 503, + 657 + ], + "lines": [ + { + "bbox": [ + 106, + 633, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 633, + 273, + 648 + ], + "score": 1.0, + "content": "This inspires us to define the probabilities", + "type": "text" + }, + { + "bbox": [ + 273, + 634, + 293, + 646 + ], + "score": 0.92, + "content": "p ( \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 633, + 310, + 648 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 311, + 634, + 341, + 646 + ], + "score": 0.92, + "content": "p ( \\mathbf { x } , \\mathbf { y } )", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 633, + 505, + 648 + ], + "score": 1.0, + "content": "using a normalized version of equation 7", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 645, + 418, + 658 + ], + "spans": [ + { + "bbox": [ + 105, + 645, + 418, + 658 + ], + "score": 1.0, + "content": "in place of equation 3. For the interval (one-dimensional box) case, we define", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 633, + 505, + 658 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 244, + 660, + 367, + 689 + ], + "lines": [ + { + "bbox": [ + 244, + 660, + 367, + 689 + ], + "spans": [ + { + "bbox": [ + 244, + 660, + 367, + 689 + ], + "score": 0.92, + "content": "\\begin{array} { c } { p ( \\mathbf { x } ) \\propto \\mathrm { s o f t } ( b - a ) } \\\\ { p ( \\mathbf { x } , \\mathbf { y } ) \\propto \\mathrm { s o f t } ( b \\wedge d - a \\vee c ) } \\end{array}", + "type": "interline_equation", + "image_path": "830b9a4b00c4c680073a6cb7e9ac7debbb7d84eebb88c836878a7c616164b43c.jpg" + } + ] + } + ], + "index": 39.5, + "virtual_lines": [ + { + "bbox": [ + 244, + 660, + 367, + 674.5 + ], + "spans": [], + "index": 39 + }, + { + "bbox": [ + 244, + 674.5, + 367, + 689.0 + ], + "spans": [], + "index": 40 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 692, + 354, + 705 + ], + "lines": [ + { + "bbox": [ + 105, + 691, + 351, + 707 + ], + "spans": [ + { + "bbox": [ + 105, + 691, + 289, + 707 + ], + "score": 1.0, + "content": "which satisfies the idempotency requirement,", + "type": "text" + }, + { + "bbox": [ + 289, + 693, + 351, + 705 + ], + "score": 0.92, + "content": "p ( \\mathbf { x } ) = p ( \\mathbf { x } , \\mathbf { x } )", + "type": "inline_equation" + } + ], + "index": 41 + } + ], + "index": 41, + "bbox_fs": [ + 105, + 691, + 351, + 707 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 709, + 506, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "Because softplus upper-bounds the hinge function, it is capable of outputting values that are greater", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 719, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 719, + 505, + 734 + ], + "score": 1.0, + "content": "than 1, and therefore must be normalized. In our experiments, we use two different approaches to", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42.5, + "bbox_fs": [ + 105, + 709, + 505, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 119 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "normalization. For experiments with a relatively small number of entities (all besides Flickr), we", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "allow the boxes to learn unconstrained, and divide each dimension by the measured size of the global", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 101, + 102, + 328, + 125 + ], + "spans": [ + { + "bbox": [ + 101, + 102, + 207, + 125 + ], + "score": 1.0, + "content": "minimum and maximum", + "type": "text" + }, + { + "bbox": [ + 208, + 104, + 255, + 119 + ], + "score": 0.93, + "content": "( G _ { m } ^ { ( i ) } , G _ { M } ^ { ( i ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 102, + 328, + 125 + ], + "score": 1.0, + "content": "at that dimension", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "interline_equation", + "bbox": [ + 250, + 123, + 361, + 156 + ], + "lines": [ + { + "bbox": [ + 250, + 123, + 361, + 156 + ], + "spans": [ + { + "bbox": [ + 250, + 123, + 361, + 156 + ], + "score": 0.95, + "content": "m _ { \\mathrm { s o f t } } ^ { ( i ) } ( x ) = \\frac { \\mathrm { s o f t } ( \\frac { x } { \\rho } ) } { \\mathrm { s o f t } ( \\frac { G _ { m } - G _ { m } } { \\rho } ) }", + "type": "interline_equation", + "image_path": "7832bb55f999b930ff4fadee85ea8ed40b25c367e1a2878bf4bf1b1e7f01025d.jpg" + } + ] + } + ], + "index": 3.5, + "virtual_lines": [ + { + "bbox": [ + 250, + 123, + 361, + 139.5 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 250, + 139.5, + 361, + 156.0 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 160, + 504, + 183 + ], + "lines": [ + { + "bbox": [ + 106, + 160, + 505, + 173 + ], + "spans": [ + { + "bbox": [ + 106, + 160, + 505, + 173 + ], + "score": 1.0, + "content": "For data where computing these values repeatedly is infeasible, we project onto the unit hypercube", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 171, + 486, + 184 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 178, + 184 + ], + "score": 1.0, + "content": "and normalize by", + "type": "text" + }, + { + "bbox": [ + 178, + 171, + 214, + 183 + ], + "score": 0.91, + "content": "m _ { \\mathrm { { s o f t } } } ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 171, + 303, + 184 + ], + "score": 1.0, + "content": ". The final probability", + "type": "text" + }, + { + "bbox": [ + 303, + 171, + 322, + 183 + ], + "score": 0.92, + "content": "p ( \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 171, + 486, + 184 + ], + "score": 1.0, + "content": "is given by the product over dimensions", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "interline_equation", + "bbox": [ + 207, + 187, + 405, + 242 + ], + "lines": [ + { + "bbox": [ + 207, + 187, + 405, + 242 + ], + "spans": [ + { + "bbox": [ + 207, + 187, + 405, + 242 + ], + "score": 0.93, + "content": "\\begin{array} { c } { { p ( { \\bf x } ) = \\displaystyle \\prod _ { i } m _ { \\mathrm { s o f t } } ^ { ( i ) } ( x _ { M , i } - x _ { m , i } ) } } \\\\ { { p ( { \\bf x } , { \\bf y } ) = \\displaystyle \\prod _ { i } m _ { \\mathrm { s o f t } } ^ { ( i ) } ( x _ { M , i } \\wedge y _ { M , i } - x _ { m , i } \\vee y _ { m , i } ) } } \\end{array}", + "type": "interline_equation", + "image_path": "3533456f48eebcfb6e9a669fb9cfd30dc2c84b7f6b994cf199199f9e8b42fd38.jpg" + } + ] + } + ], + "index": 8.5, + "virtual_lines": [ + { + "bbox": [ + 207, + 187, + 405, + 200.75 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 207, + 200.75, + 405, + 214.5 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 207, + 214.5, + 405, + 228.25 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 207, + 228.25, + 405, + 242.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 245, + 505, + 334 + ], + "lines": [ + { + "bbox": [ + 105, + 245, + 505, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 505, + 258 + ], + "score": 1.0, + "content": "Note that, while equivalent in the zero temperature limit to the standard uniform probability measure", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 256, + 505, + 269 + ], + "spans": [ + { + "bbox": [ + 105, + 256, + 505, + 269 + ], + "score": 1.0, + "content": "of the box model, this function, like the Gaussian model, is not a valid probability measure on the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 267, + 505, + 280 + ], + "spans": [ + { + "bbox": [ + 106, + 267, + 505, + 280 + ], + "score": 1.0, + "content": "entire joint space of events (the lattice). However, neither is factorization of a conditional probability", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 279, + 506, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 506, + 291 + ], + "score": 1.0, + "content": "table using a logistic sigmoid link function, which is commonly used for the similar tasks. Our", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 289, + 506, + 302 + ], + "spans": [ + { + "bbox": [ + 104, + 289, + 506, + 302 + ], + "score": 1.0, + "content": "approach retains the inductive bias of the original box model, is equivalent in the limit, and satisfies", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 300, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 221, + 313 + ], + "score": 1.0, + "content": "the necessary condition that", + "type": "text" + }, + { + "bbox": [ + 222, + 300, + 285, + 313 + ], + "score": 0.93, + "content": "p ( \\mathbf { x } , \\mathbf { x } ) = p ( \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 300, + 505, + 313 + ], + "score": 1.0, + "content": ". A comparison of the 3 different functions is given in", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 312, + 505, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 312, + 505, + 324 + ], + "score": 1.0, + "content": "Figure 3, with the softplus overlap showing much better behavior for highly disjoint boxes than the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 322, + 338, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 338, + 335 + ], + "score": 1.0, + "content": "Gaussian model, while also preserving the meet property.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 14.5 + }, + { + "type": "image", + "bbox": [ + 111, + 361, + 498, + 451 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 361, + 498, + 451 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 111, + 361, + 498, + 451 + ], + "spans": [ + { + "bbox": [ + 111, + 361, + 498, + 451 + ], + "score": 0.965, + "type": "image", + "image_path": "9cc21fb6cb8ac7eae24bc49999bdb4ae1bfab3ab3d06a128598a726986167a50.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 111, + 361, + 498, + 391.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 111, + 391.0, + 498, + 421.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 111, + 421.0, + 498, + 451.0 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 459, + 505, + 493 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 459, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 505, + 471 + ], + "score": 1.0, + "content": "Figure 3: Comparison of different overlap functions for two boxes of width 0.3 as a function of their", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 471, + 505, + 482 + ], + "spans": [ + { + "bbox": [ + 106, + 471, + 505, + 482 + ], + "score": 1.0, + "content": "centers. Note that in order to achieve high overlap, the Gaussian model must drastically lower its", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 482, + 319, + 494 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 319, + 494 + ], + "score": 1.0, + "content": "temperature, causing vanishing gradients in the tails.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23 + } + ], + "index": 21.5 + }, + { + "type": "title", + "bbox": [ + 108, + 516, + 200, + 528 + ], + "lines": [ + { + "bbox": [ + 104, + 514, + 201, + 531 + ], + "spans": [ + { + "bbox": [ + 104, + 514, + 201, + 531 + ], + "score": 1.0, + "content": "5 EXPERIMENTS", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "title", + "bbox": [ + 107, + 540, + 178, + 552 + ], + "lines": [ + { + "bbox": [ + 105, + 539, + 180, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 180, + 554 + ], + "score": 1.0, + "content": "5.1 WORDNET", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "table", + "bbox": [ + 228, + 563, + 381, + 654 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 228, + 563, + 381, + 654 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 228, + 563, + 381, + 654 + ], + "spans": [ + { + "bbox": [ + 228, + 563, + 381, + 654 + ], + "score": 0.955, + "html": "
MethodTest Accuracy %
transitive88.2
word2gauss86.6
OE90.6
Li et al. (2017)91.3
POE91.6
Box92.2
Smoothed Box92.0
", + "type": "table", + "image_path": "62f03052434b9f36bbe4fd8c1adedbdf8998958456ce8de09945a681c37bf5a4.jpg" + } + ] + } + ], + "index": 27.5, + "virtual_lines": [ + { + "bbox": [ + 228, + 563, + 381, + 608.5 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 228, + 608.5, + 381, + 654.0 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 199, + 662, + 411, + 673 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 198, + 660, + 412, + 675 + ], + "spans": [ + { + "bbox": [ + 198, + 660, + 412, + 675 + ], + "score": 1.0, + "content": "Table 4: Classification accuracy on WordNet test set.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + } + ], + "index": 28.25 + }, + { + "type": "text", + "bbox": [ + 106, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "We perform experiments on the WordNet hypernym prediction task in order to evaluate the per-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "formance of these improvements in practice. The WordNet hypernym hierarchy contains 837,888-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "edges after performing the transitive closure on the direct edges in WordNet. We used the same", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "train/dev/test split as in Vendrov et al. (2016). Positive examples are randomly chosen from the", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31.5 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 750, + 309, + 762 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 119 + ], + "lines": [ + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "normalization. For experiments with a relatively small number of entities (all besides Flickr), we", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "allow the boxes to learn unconstrained, and divide each dimension by the measured size of the global", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 101, + 102, + 328, + 125 + ], + "spans": [ + { + "bbox": [ + 101, + 102, + 207, + 125 + ], + "score": 1.0, + "content": "minimum and maximum", + "type": "text" + }, + { + "bbox": [ + 208, + 104, + 255, + 119 + ], + "score": 0.93, + "content": "( G _ { m } ^ { ( i ) } , G _ { M } ^ { ( i ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 102, + 328, + 125 + ], + "score": 1.0, + "content": "at that dimension", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 101, + 82, + 506, + 125 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 250, + 123, + 361, + 156 + ], + "lines": [ + { + "bbox": [ + 250, + 123, + 361, + 156 + ], + "spans": [ + { + "bbox": [ + 250, + 123, + 361, + 156 + ], + "score": 0.95, + "content": "m _ { \\mathrm { s o f t } } ^ { ( i ) } ( x ) = \\frac { \\mathrm { s o f t } ( \\frac { x } { \\rho } ) } { \\mathrm { s o f t } ( \\frac { G _ { m } - G _ { m } } { \\rho } ) }", + "type": "interline_equation", + "image_path": "7832bb55f999b930ff4fadee85ea8ed40b25c367e1a2878bf4bf1b1e7f01025d.jpg" + } + ] + } + ], + "index": 3.5, + "virtual_lines": [ + { + "bbox": [ + 250, + 123, + 361, + 139.5 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 250, + 139.5, + 361, + 156.0 + ], + "spans": [], + "index": 4 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 160, + 504, + 183 + ], + "lines": [ + { + "bbox": [ + 106, + 160, + 505, + 173 + ], + "spans": [ + { + "bbox": [ + 106, + 160, + 505, + 173 + ], + "score": 1.0, + "content": "For data where computing these values repeatedly is infeasible, we project onto the unit hypercube", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 171, + 486, + 184 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 178, + 184 + ], + "score": 1.0, + "content": "and normalize by", + "type": "text" + }, + { + "bbox": [ + 178, + 171, + 214, + 183 + ], + "score": 0.91, + "content": "m _ { \\mathrm { { s o f t } } } ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 171, + 303, + 184 + ], + "score": 1.0, + "content": ". The final probability", + "type": "text" + }, + { + "bbox": [ + 303, + 171, + 322, + 183 + ], + "score": 0.92, + "content": "p ( \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 171, + 486, + 184 + ], + "score": 1.0, + "content": "is given by the product over dimensions", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5, + "bbox_fs": [ + 106, + 160, + 505, + 184 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 207, + 187, + 405, + 242 + ], + "lines": [ + { + "bbox": [ + 207, + 187, + 405, + 242 + ], + "spans": [ + { + "bbox": [ + 207, + 187, + 405, + 242 + ], + "score": 0.93, + "content": "\\begin{array} { c } { { p ( { \\bf x } ) = \\displaystyle \\prod _ { i } m _ { \\mathrm { s o f t } } ^ { ( i ) } ( x _ { M , i } - x _ { m , i } ) } } \\\\ { { p ( { \\bf x } , { \\bf y } ) = \\displaystyle \\prod _ { i } m _ { \\mathrm { s o f t } } ^ { ( i ) } ( x _ { M , i } \\wedge y _ { M , i } - x _ { m , i } \\vee y _ { m , i } ) } } \\end{array}", + "type": "interline_equation", + "image_path": "3533456f48eebcfb6e9a669fb9cfd30dc2c84b7f6b994cf199199f9e8b42fd38.jpg" + } + ] + } + ], + "index": 8.5, + "virtual_lines": [ + { + "bbox": [ + 207, + 187, + 405, + 200.75 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 207, + 200.75, + 405, + 214.5 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 207, + 214.5, + 405, + 228.25 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 207, + 228.25, + 405, + 242.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 245, + 505, + 334 + ], + "lines": [ + { + "bbox": [ + 105, + 245, + 505, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 505, + 258 + ], + "score": 1.0, + "content": "Note that, while equivalent in the zero temperature limit to the standard uniform probability measure", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 256, + 505, + 269 + ], + "spans": [ + { + "bbox": [ + 105, + 256, + 505, + 269 + ], + "score": 1.0, + "content": "of the box model, this function, like the Gaussian model, is not a valid probability measure on the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 267, + 505, + 280 + ], + "spans": [ + { + "bbox": [ + 106, + 267, + 505, + 280 + ], + "score": 1.0, + "content": "entire joint space of events (the lattice). However, neither is factorization of a conditional probability", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 279, + 506, + 291 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 506, + 291 + ], + "score": 1.0, + "content": "table using a logistic sigmoid link function, which is commonly used for the similar tasks. Our", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 289, + 506, + 302 + ], + "spans": [ + { + "bbox": [ + 104, + 289, + 506, + 302 + ], + "score": 1.0, + "content": "approach retains the inductive bias of the original box model, is equivalent in the limit, and satisfies", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 300, + 505, + 313 + ], + "spans": [ + { + "bbox": [ + 105, + 300, + 221, + 313 + ], + "score": 1.0, + "content": "the necessary condition that", + "type": "text" + }, + { + "bbox": [ + 222, + 300, + 285, + 313 + ], + "score": 0.93, + "content": "p ( \\mathbf { x } , \\mathbf { x } ) = p ( \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 285, + 300, + 505, + 313 + ], + "score": 1.0, + "content": ". A comparison of the 3 different functions is given in", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 312, + 505, + 324 + ], + "spans": [ + { + "bbox": [ + 105, + 312, + 505, + 324 + ], + "score": 1.0, + "content": "Figure 3, with the softplus overlap showing much better behavior for highly disjoint boxes than the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 322, + 338, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 322, + 338, + 335 + ], + "score": 1.0, + "content": "Gaussian model, while also preserving the meet property.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 14.5, + "bbox_fs": [ + 104, + 245, + 506, + 335 + ] + }, + { + "type": "image", + "bbox": [ + 111, + 361, + 498, + 451 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 361, + 498, + 451 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 111, + 361, + 498, + 451 + ], + "spans": [ + { + "bbox": [ + 111, + 361, + 498, + 451 + ], + "score": 0.965, + "type": "image", + "image_path": "9cc21fb6cb8ac7eae24bc49999bdb4ae1bfab3ab3d06a128598a726986167a50.jpg" + } + ] + } + ], + "index": 20, + "virtual_lines": [ + { + "bbox": [ + 111, + 361, + 498, + 391.0 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 111, + 391.0, + 498, + 421.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 111, + 421.0, + 498, + 451.0 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 459, + 505, + 493 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 459, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 459, + 505, + 471 + ], + "score": 1.0, + "content": "Figure 3: Comparison of different overlap functions for two boxes of width 0.3 as a function of their", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 471, + 505, + 482 + ], + "spans": [ + { + "bbox": [ + 106, + 471, + 505, + 482 + ], + "score": 1.0, + "content": "centers. Note that in order to achieve high overlap, the Gaussian model must drastically lower its", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 482, + 319, + 494 + ], + "spans": [ + { + "bbox": [ + 106, + 482, + 319, + 494 + ], + "score": 1.0, + "content": "temperature, causing vanishing gradients in the tails.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23 + } + ], + "index": 21.5 + }, + { + "type": "title", + "bbox": [ + 108, + 516, + 200, + 528 + ], + "lines": [ + { + "bbox": [ + 104, + 514, + 201, + 531 + ], + "spans": [ + { + "bbox": [ + 104, + 514, + 201, + 531 + ], + "score": 1.0, + "content": "5 EXPERIMENTS", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "title", + "bbox": [ + 107, + 540, + 178, + 552 + ], + "lines": [ + { + "bbox": [ + 105, + 539, + 180, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 180, + 554 + ], + "score": 1.0, + "content": "5.1 WORDNET", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "table", + "bbox": [ + 228, + 563, + 381, + 654 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 228, + 563, + 381, + 654 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 228, + 563, + 381, + 654 + ], + "spans": [ + { + "bbox": [ + 228, + 563, + 381, + 654 + ], + "score": 0.955, + "html": "
MethodTest Accuracy %
transitive88.2
word2gauss86.6
OE90.6
Li et al. (2017)91.3
POE91.6
Box92.2
Smoothed Box92.0
", + "type": "table", + "image_path": "62f03052434b9f36bbe4fd8c1adedbdf8998958456ce8de09945a681c37bf5a4.jpg" + } + ] + } + ], + "index": 27.5, + "virtual_lines": [ + { + "bbox": [ + 228, + 563, + 381, + 608.5 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 228, + 608.5, + 381, + 654.0 + ], + "spans": [], + "index": 28 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 199, + 662, + 411, + 673 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 198, + 660, + 412, + 675 + ], + "spans": [ + { + "bbox": [ + 198, + 660, + 412, + 675 + ], + "score": 1.0, + "content": "Table 4: Classification accuracy on WordNet test set.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + } + ], + "index": 28.25 + }, + { + "type": "text", + "bbox": [ + 106, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 106, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "We perform experiments on the WordNet hypernym prediction task in order to evaluate the per-", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "formance of these improvements in practice. The WordNet hypernym hierarchy contains 837,888-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 709, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 709, + 505, + 723 + ], + "score": 1.0, + "content": "edges after performing the transitive closure on the direct edges in WordNet. We used the same", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 733 + ], + "score": 1.0, + "content": "train/dev/test split as in Vendrov et al. (2016). Positive examples are randomly chosen from the", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 128, + 93 + ], + "score": 0.33, + "content": "{ } ^ { 8 3 7 \\mathrm { k } }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 128, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "edges, while negative examples are generated by swapping one of the terms to a random word", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 372, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 372, + 106 + ], + "score": 1.0, + "content": "in the dictionary. Experimental details are given in Appendix D.1.", + "type": "text", + "cross_page": true + } + ], + "index": 1 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 687, + 505, + 733 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 128, + 93 + ], + "score": 0.33, + "content": "{ } ^ { 8 3 7 \\mathrm { k } }", + "type": "inline_equation" + }, + { + "bbox": [ + 128, + 82, + 506, + 96 + ], + "score": 1.0, + "content": "edges, while negative examples are generated by swapping one of the terms to a random word", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 372, + 106 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 372, + 106 + ], + "score": 1.0, + "content": "in the dictionary. Experimental details are given in Appendix D.1.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 107, + 110, + 504, + 155 + ], + "lines": [ + { + "bbox": [ + 105, + 110, + 506, + 123 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 506, + 123 + ], + "score": 1.0, + "content": "The smoothed box model performs nearly as well as the original box lattice in terms of test ac-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 120, + 506, + 135 + ], + "spans": [ + { + "bbox": [ + 105, + 120, + 506, + 135 + ], + "score": 1.0, + "content": "curacy1. While our model requires less hyper-parameter tuning than the original, we suspect that", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 132, + 506, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 132, + 506, + 145 + ], + "score": 1.0, + "content": "our performance would be increased on a task with a higher degree of sparsity than the 50/50 posi-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 143, + 452, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 143, + 452, + 156 + ], + "score": 1.0, + "content": "tive/negative split of the standard WordNet data, which we explore in the next section.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3.5 + }, + { + "type": "title", + "bbox": [ + 108, + 171, + 239, + 182 + ], + "lines": [ + { + "bbox": [ + 105, + 170, + 241, + 184 + ], + "spans": [ + { + "bbox": [ + 105, + 170, + 241, + 184 + ], + "score": 1.0, + "content": "5.2 IMBALANCED WORDNET", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 193, + 505, + 270 + ], + "lines": [ + { + "bbox": [ + 105, + 192, + 505, + 206 + ], + "spans": [ + { + "bbox": [ + 105, + 192, + 505, + 206 + ], + "score": 1.0, + "content": "In order to confirm our intuition that the smoothed box model performs better in the sparse regime,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 204, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 505, + 217 + ], + "score": 1.0, + "content": "we perform further experiments using different numbers of positive and negative examples from the", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 214, + 505, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 505, + 228 + ], + "score": 1.0, + "content": "WordNet mammal subset, comparing the box lattice, our smoothed approach, and order embeddings", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 226, + 505, + 238 + ], + "spans": [ + { + "bbox": [ + 106, + 226, + 505, + 238 + ], + "score": 1.0, + "content": "(OE) as a baseline. The training data is the transitive reduction of this subset of the mammal Word-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 237, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 505, + 249 + ], + "score": 1.0, + "content": "Net, while the dev/test is the transitive closure of the training data. The training data contains 1,176", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 248, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 104, + 248, + 505, + 261 + ], + "score": 1.0, + "content": "positive examples, and the dev and test sets contain 209 positive examples. Negative examples are", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 259, + 323, + 271 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 323, + 271 + ], + "score": 1.0, + "content": "generated randomly using the ratio stated in the table.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 275, + 505, + 331 + ], + "lines": [ + { + "bbox": [ + 106, + 276, + 504, + 287 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 504, + 287 + ], + "score": 1.0, + "content": "As we can see from the table, with balanced data, all models include OE baseline, Box, Smoothed", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 285, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 505, + 300 + ], + "score": 1.0, + "content": "Box models nearly match the full transitive closure. As the number of negative examples increases,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 298, + 505, + 310 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 505, + 310 + ], + "score": 1.0, + "content": "the performance drops for the original box model, but Smoothed Box still outperforms OE and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 308, + 505, + 322 + ], + "spans": [ + { + "bbox": [ + 106, + 308, + 505, + 322 + ], + "score": 1.0, + "content": "Box in all setting. This superior performance on imbalanced data is important for e.g. real-world", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 320, + 462, + 332 + ], + "spans": [ + { + "bbox": [ + 106, + 320, + 462, + 332 + ], + "score": 1.0, + "content": "entailment graph learning, where the number of negatives greatly outweigh the positives.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16 + }, + { + "type": "table", + "bbox": [ + 187, + 344, + 423, + 403 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 187, + 344, + 423, + 403 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 187, + 344, + 423, + 403 + ], + "spans": [ + { + "bbox": [ + 187, + 344, + 423, + 403 + ], + "score": 0.976, + "html": "
Positive:NegativeBoxOESmoothed Box
1:10.99050.99761.0
1:20.89820.91391.0
1:60.66800.66400.9561
1:100.54950.58970.8800
", + "type": "table", + "image_path": "bd6fa0527bebce8c0735fbabacd735d261e9921ef0dca15a9cc8c9c53d41a008.jpg" + } + ] + } + ], + "index": 20.5, + "virtual_lines": [ + { + "bbox": [ + 187, + 344, + 423, + 358.75 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 187, + 358.75, + 423, + 373.5 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 187, + 373.5, + 423, + 388.25 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 187, + 388.25, + 423, + 403.0 + ], + "spans": [], + "index": 22 + } + ] + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 104, + 411, + 504, + 434 + ], + "lines": [ + { + "bbox": [ + 106, + 410, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 106, + 410, + 505, + 423 + ], + "score": 1.0, + "content": "Table 5: F1 scores of the box lattice, order embeddings, and our smoothed model, for different levels", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 423, + 315, + 434 + ], + "spans": [ + { + "bbox": [ + 106, + 423, + 315, + 434 + ], + "score": 1.0, + "content": "of label imbalance on the WordNet mammal subset.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "title", + "bbox": [ + 107, + 460, + 164, + 471 + ], + "lines": [ + { + "bbox": [ + 106, + 459, + 165, + 473 + ], + "spans": [ + { + "bbox": [ + 106, + 459, + 165, + 473 + ], + "score": 1.0, + "content": "5.3 FLICKR", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 481, + 505, + 547 + ], + "lines": [ + { + "bbox": [ + 105, + 481, + 505, + 494 + ], + "spans": [ + { + "bbox": [ + 105, + 481, + 505, + 494 + ], + "score": 1.0, + "content": "We conduct experiments on the Flickr entailment dataset. Flickr is a large-scale caption entailment", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 492, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 505, + 506 + ], + "score": 1.0, + "content": "dataset containing of 45 million image caption pairs. In order to perform an apples-to-apples com-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 504, + 505, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 505, + 516 + ], + "score": 1.0, + "content": "parison with existing results we use the exact same dataset from Vilnis et al. (2018). In this case, we", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 514, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 505, + 527 + ], + "score": 1.0, + "content": "do constrain the boxes to the unit cube, using the same experimental setup as Vilnis et al. (2018),", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 526, + 505, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 526, + 505, + 538 + ], + "score": 1.0, + "content": "except we apply the softplus function before calculating the volume of the boxes. Experimental", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 537, + 245, + 549 + ], + "spans": [ + { + "bbox": [ + 106, + 537, + 245, + 549 + ], + "score": 1.0, + "content": "details are given in Appendix D.3.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 107, + 553, + 505, + 598 + ], + "lines": [ + { + "bbox": [ + 105, + 553, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 505, + 565 + ], + "score": 1.0, + "content": "We report KL divergence and Pearson correlation on the full test data, unseen pairs (caption pairs", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 564, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 505, + 577 + ], + "score": 1.0, + "content": "which are never occur in training data) and unseen captions (captions which are never occur in", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "score": 1.0, + "content": "training data). As shown in Table 6, we see a slight performance gain compared to the original", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 586, + 372, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 372, + 599 + ], + "score": 1.0, + "content": "model, with improvements most concentrated on unseen captions.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33.5 + }, + { + "type": "title", + "bbox": [ + 107, + 614, + 185, + 625 + ], + "lines": [ + { + "bbox": [ + 106, + 613, + 186, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 613, + 186, + 627 + ], + "score": 1.0, + "content": "5.4 MOVIELENS", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 107, + 636, + 505, + 707 + ], + "lines": [ + { + "bbox": [ + 106, + 636, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 636, + 505, + 648 + ], + "score": 1.0, + "content": "We apply our method to a market-basket task constructed using the MovieLens dataset. Here, the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 647, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 106, + 647, + 505, + 659 + ], + "score": 1.0, + "content": "task is to predict users’ preference for movie A given that they liked movie B. We first collect", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 657, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 505, + 671 + ], + "score": 1.0, + "content": "all pairs of user-movie ratings higher than 4 points (strong preference) from the MovieLens-20M", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 669, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 106, + 669, + 506, + 681 + ], + "score": 1.0, + "content": "dataset. From this we further prune to just a subset of movies which have more than 100 user", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 680, + 506, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 506, + 692 + ], + "score": 1.0, + "content": "ratings to make sure that counting statistics are significant enough. This leads to 8545 movies in our", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 691, + 506, + 708 + ], + "spans": [ + { + "bbox": [ + 106, + 693, + 301, + 705 + ], + "score": 1.0, + "content": "dataset. We calculate the conditional probability", + "type": "text" + }, + { + "bbox": [ + 301, + 691, + 485, + 708 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\overline { { P } } ( A | B ) = \\frac { \\overline { { P ( A , B ) } } } { \\overline { { P ( B ) } } } = \\frac { \\# r a t i n g ( A , B ) _ { > 4 } / \\# u s e r s } { \\# r a t i n g ( B ) _ { > 4 } / \\# u s e r s } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 693, + 506, + 705 + ], + "score": 1.0, + "content": "We", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 39.5 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 117, + 721, + 458, + 732 + ], + "lines": [ + { + "bbox": [ + 119, + 719, + 460, + 734 + ], + "spans": [ + { + "bbox": [ + 119, + 719, + 460, + 734 + ], + "score": 1.0, + "content": "1Accuracy is calculated by applying the same threshold which maximized accuracy in dev set.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 308, + 760 + ], + "lines": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "spans": [ + { + "bbox": [ + 300, + 750, + 309, + 761 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 82, + 504, + 105 + ], + "lines": [], + "index": 0.5, + "bbox_fs": [ + 105, + 82, + 506, + 106 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 110, + 504, + 155 + ], + "lines": [ + { + "bbox": [ + 105, + 110, + 506, + 123 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 506, + 123 + ], + "score": 1.0, + "content": "The smoothed box model performs nearly as well as the original box lattice in terms of test ac-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 120, + 506, + 135 + ], + "spans": [ + { + "bbox": [ + 105, + 120, + 506, + 135 + ], + "score": 1.0, + "content": "curacy1. While our model requires less hyper-parameter tuning than the original, we suspect that", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 132, + 506, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 132, + 506, + 145 + ], + "score": 1.0, + "content": "our performance would be increased on a task with a higher degree of sparsity than the 50/50 posi-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 143, + 452, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 143, + 452, + 156 + ], + "score": 1.0, + "content": "tive/negative split of the standard WordNet data, which we explore in the next section.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3.5, + "bbox_fs": [ + 105, + 110, + 506, + 156 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 171, + 239, + 182 + ], + "lines": [ + { + "bbox": [ + 105, + 170, + 241, + 184 + ], + "spans": [ + { + "bbox": [ + 105, + 170, + 241, + 184 + ], + "score": 1.0, + "content": "5.2 IMBALANCED WORDNET", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 193, + 505, + 270 + ], + "lines": [ + { + "bbox": [ + 105, + 192, + 505, + 206 + ], + "spans": [ + { + "bbox": [ + 105, + 192, + 505, + 206 + ], + "score": 1.0, + "content": "In order to confirm our intuition that the smoothed box model performs better in the sparse regime,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 204, + 505, + 217 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 505, + 217 + ], + "score": 1.0, + "content": "we perform further experiments using different numbers of positive and negative examples from the", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 214, + 505, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 505, + 228 + ], + "score": 1.0, + "content": "WordNet mammal subset, comparing the box lattice, our smoothed approach, and order embeddings", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 226, + 505, + 238 + ], + "spans": [ + { + "bbox": [ + 106, + 226, + 505, + 238 + ], + "score": 1.0, + "content": "(OE) as a baseline. The training data is the transitive reduction of this subset of the mammal Word-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 237, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 237, + 505, + 249 + ], + "score": 1.0, + "content": "Net, while the dev/test is the transitive closure of the training data. The training data contains 1,176", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 248, + 505, + 261 + ], + "spans": [ + { + "bbox": [ + 104, + 248, + 505, + 261 + ], + "score": 1.0, + "content": "positive examples, and the dev and test sets contain 209 positive examples. Negative examples are", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 259, + 323, + 271 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 323, + 271 + ], + "score": 1.0, + "content": "generated randomly using the ratio stated in the table.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 10, + "bbox_fs": [ + 104, + 192, + 505, + 271 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 275, + 505, + 331 + ], + "lines": [ + { + "bbox": [ + 106, + 276, + 504, + 287 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 504, + 287 + ], + "score": 1.0, + "content": "As we can see from the table, with balanced data, all models include OE baseline, Box, Smoothed", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 285, + 505, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 505, + 300 + ], + "score": 1.0, + "content": "Box models nearly match the full transitive closure. As the number of negative examples increases,", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 298, + 505, + 310 + ], + "spans": [ + { + "bbox": [ + 106, + 298, + 505, + 310 + ], + "score": 1.0, + "content": "the performance drops for the original box model, but Smoothed Box still outperforms OE and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 308, + 505, + 322 + ], + "spans": [ + { + "bbox": [ + 106, + 308, + 505, + 322 + ], + "score": 1.0, + "content": "Box in all setting. This superior performance on imbalanced data is important for e.g. real-world", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 320, + 462, + 332 + ], + "spans": [ + { + "bbox": [ + 106, + 320, + 462, + 332 + ], + "score": 1.0, + "content": "entailment graph learning, where the number of negatives greatly outweigh the positives.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 276, + 505, + 332 + ] + }, + { + "type": "table", + "bbox": [ + 187, + 344, + 423, + 403 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 187, + 344, + 423, + 403 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 187, + 344, + 423, + 403 + ], + "spans": [ + { + "bbox": [ + 187, + 344, + 423, + 403 + ], + "score": 0.976, + "html": "
Positive:NegativeBoxOESmoothed Box
1:10.99050.99761.0
1:20.89820.91391.0
1:60.66800.66400.9561
1:100.54950.58970.8800
", + "type": "table", + "image_path": "bd6fa0527bebce8c0735fbabacd735d261e9921ef0dca15a9cc8c9c53d41a008.jpg" + } + ] + } + ], + "index": 20.5, + "virtual_lines": [ + { + "bbox": [ + 187, + 344, + 423, + 358.75 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 187, + 358.75, + 423, + 373.5 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 187, + 373.5, + 423, + 388.25 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 187, + 388.25, + 423, + 403.0 + ], + "spans": [], + "index": 22 + } + ] + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 104, + 411, + 504, + 434 + ], + "lines": [ + { + "bbox": [ + 106, + 410, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 106, + 410, + 505, + 423 + ], + "score": 1.0, + "content": "Table 5: F1 scores of the box lattice, order embeddings, and our smoothed model, for different levels", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 423, + 315, + 434 + ], + "spans": [ + { + "bbox": [ + 106, + 423, + 315, + 434 + ], + "score": 1.0, + "content": "of label imbalance on the WordNet mammal subset.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 106, + 410, + 505, + 434 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 460, + 164, + 471 + ], + "lines": [ + { + "bbox": [ + 106, + 459, + 165, + 473 + ], + "spans": [ + { + "bbox": [ + 106, + 459, + 165, + 473 + ], + "score": 1.0, + "content": "5.3 FLICKR", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 481, + 505, + 547 + ], + "lines": [ + { + "bbox": [ + 105, + 481, + 505, + 494 + ], + "spans": [ + { + "bbox": [ + 105, + 481, + 505, + 494 + ], + "score": 1.0, + "content": "We conduct experiments on the Flickr entailment dataset. Flickr is a large-scale caption entailment", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 492, + 505, + 506 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 505, + 506 + ], + "score": 1.0, + "content": "dataset containing of 45 million image caption pairs. In order to perform an apples-to-apples com-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 504, + 505, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 504, + 505, + 516 + ], + "score": 1.0, + "content": "parison with existing results we use the exact same dataset from Vilnis et al. (2018). In this case, we", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 514, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 505, + 527 + ], + "score": 1.0, + "content": "do constrain the boxes to the unit cube, using the same experimental setup as Vilnis et al. (2018),", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 526, + 505, + 538 + ], + "spans": [ + { + "bbox": [ + 105, + 526, + 505, + 538 + ], + "score": 1.0, + "content": "except we apply the softplus function before calculating the volume of the boxes. Experimental", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 537, + 245, + 549 + ], + "spans": [ + { + "bbox": [ + 106, + 537, + 245, + 549 + ], + "score": 1.0, + "content": "details are given in Appendix D.3.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 481, + 505, + 549 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 553, + 505, + 598 + ], + "lines": [ + { + "bbox": [ + 105, + 553, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 505, + 565 + ], + "score": 1.0, + "content": "We report KL divergence and Pearson correlation on the full test data, unseen pairs (caption pairs", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 564, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 505, + 577 + ], + "score": 1.0, + "content": "which are never occur in training data) and unseen captions (captions which are never occur in", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 576, + 505, + 588 + ], + "score": 1.0, + "content": "training data). As shown in Table 6, we see a slight performance gain compared to the original", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 586, + 372, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 586, + 372, + 599 + ], + "score": 1.0, + "content": "model, with improvements most concentrated on unseen captions.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 553, + 505, + 599 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 614, + 185, + 625 + ], + "lines": [ + { + "bbox": [ + 106, + 613, + 186, + 627 + ], + "spans": [ + { + "bbox": [ + 106, + 613, + 186, + 627 + ], + "score": 1.0, + "content": "5.4 MOVIELENS", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 107, + 636, + 505, + 707 + ], + "lines": [ + { + "bbox": [ + 106, + 636, + 505, + 648 + ], + "spans": [ + { + "bbox": [ + 106, + 636, + 505, + 648 + ], + "score": 1.0, + "content": "We apply our method to a market-basket task constructed using the MovieLens dataset. Here, the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 647, + 505, + 659 + ], + "spans": [ + { + "bbox": [ + 106, + 647, + 505, + 659 + ], + "score": 1.0, + "content": "task is to predict users’ preference for movie A given that they liked movie B. We first collect", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 657, + 505, + 671 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 505, + 671 + ], + "score": 1.0, + "content": "all pairs of user-movie ratings higher than 4 points (strong preference) from the MovieLens-20M", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 669, + 506, + 681 + ], + "spans": [ + { + "bbox": [ + 106, + 669, + 506, + 681 + ], + "score": 1.0, + "content": "dataset. From this we further prune to just a subset of movies which have more than 100 user", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 680, + 506, + 692 + ], + "spans": [ + { + "bbox": [ + 105, + 680, + 506, + 692 + ], + "score": 1.0, + "content": "ratings to make sure that counting statistics are significant enough. This leads to 8545 movies in our", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 691, + 506, + 708 + ], + "spans": [ + { + "bbox": [ + 106, + 693, + 301, + 705 + ], + "score": 1.0, + "content": "dataset. We calculate the conditional probability", + "type": "text" + }, + { + "bbox": [ + 301, + 691, + 485, + 708 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\overline { { P } } ( A | B ) = \\frac { \\overline { { P ( A , B ) } } } { \\overline { { P ( B ) } } } = \\frac { \\# r a t i n g ( A , B ) _ { > 4 } / \\# u s e r s } { \\# r a t i n g ( B ) _ { > 4 } / \\# u s e r s } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 693, + 506, + 705 + ], + "score": 1.0, + "content": "We", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 302, + 505, + 314 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 505, + 314 + ], + "score": 1.0, + "content": "randomly pick 100K conditional probabilities for training data and 10k probabilities for dev and test", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 311, + 135, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 135, + 325 + ], + "score": 1.0, + "content": "data 2.", + "type": "text", + "cross_page": true + } + ], + "index": 15 + } + ], + "index": 39.5, + "bbox_fs": [ + 105, + 636, + 506, + 708 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 217, + 80, + 393, + 260 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 217, + 80, + 393, + 260 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 217, + 80, + 393, + 260 + ], + "spans": [ + { + "bbox": [ + 217, + 80, + 393, + 260 + ], + "score": 0.977, + "html": "
P(xly)
Full test data POEKL 0.031Pearson R 0.949
POE* Box0.031 0.0200.949 0.967
Smoothed Box0.0180.969
Unseen pairs POE0.0480.920 0.925
POE* Box Smoothed Box0.046 0.025 0.0240.957 0.957
Unseen captions
POE0.1270.696
POE*0.0840.854
Box Smoothed Box0.050 0.0360.900 0.917
", + "type": "table", + "image_path": "ed4366fc3a34a30a7f83b854427a16678abc8d80af3ded5a560277b258db13af.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 217, + 80, + 393, + 93.84615384615384 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 217, + 93.84615384615384, + 393, + 107.69230769230768 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 217, + 107.69230769230768, + 393, + 121.53846153846152 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 217, + 121.53846153846152, + 393, + 135.38461538461536 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 217, + 135.38461538461536, + 393, + 149.2307692307692 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 217, + 149.2307692307692, + 393, + 163.07692307692304 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 217, + 163.07692307692304, + 393, + 176.92307692307688 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 217, + 176.92307692307688, + 393, + 190.76923076923072 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 217, + 190.76923076923072, + 393, + 204.61538461538456 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 217, + 204.61538461538456, + 393, + 218.4615384615384 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 217, + 218.4615384615384, + 393, + 232.30769230769224 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 217, + 232.30769230769224, + 393, + 246.15384615384608 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 217, + 246.15384615384608, + 393, + 259.99999999999994 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 157, + 268, + 452, + 280 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 157, + 266, + 454, + 283 + ], + "spans": [ + { + "bbox": [ + 157, + 266, + 454, + 283 + ], + "score": 1.0, + "content": "Table 6: KL and Pearson correlation between model and gold probability.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 106, + 302, + 504, + 324 + ], + "lines": [ + { + "bbox": [ + 105, + 302, + 505, + 314 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 505, + 314 + ], + "score": 1.0, + "content": "randomly pick 100K conditional probabilities for training data and 10k probabilities for dev and test", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 311, + 135, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 311, + 135, + 325 + ], + "score": 1.0, + "content": "data 2.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 106, + 330, + 505, + 407 + ], + "lines": [ + { + "bbox": [ + 106, + 329, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 505, + 342 + ], + "score": 1.0, + "content": "We compare with several baselines: low-rank matrix factorization, complex bilinear factoriza-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 340, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 505, + 353 + ], + "score": 1.0, + "content": "tion (Trouillon et al., 2016), and two hierarchical embedding methods, POE (Lai & Hockenmaier,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 352, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 505, + 364 + ], + "score": 1.0, + "content": "2017) and the Box Lattice (Vilnis et al., 2018). Since the training matrix is asymmetric, we used", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 362, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 506, + 375 + ], + "score": 1.0, + "content": "separate embeddings for target and conditioned movies. For the complex bilinear model, we added", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 373, + 505, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 505, + 387 + ], + "score": 1.0, + "content": "one additional vector of parameters to capture the “imply” relation. We evaluate on the test set using", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 384, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 505, + 398 + ], + "score": 1.0, + "content": "KL divergence, Pearson correlation, and Spearman correlation with the ground truth probabilities.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 396, + 302, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 302, + 408 + ], + "score": 1.0, + "content": "Experimental details are given in Appendix D.4.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 108, + 412, + 504, + 457 + ], + "lines": [ + { + "bbox": [ + 105, + 412, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 505, + 425 + ], + "score": 1.0, + "content": "From the results in Table 7, we can see that our smoothed box embedding method outperforms the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 423, + 506, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 506, + 436 + ], + "score": 1.0, + "content": "original box lattice as well as all other baselines’ performances, especially in Spearman correlation,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 435, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 435, + 505, + 447 + ], + "score": 1.0, + "content": "the most relevant metric for recommendation, a ranking task. We perform an additional study on the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 446, + 415, + 457 + ], + "spans": [ + { + "bbox": [ + 106, + 446, + 415, + 457 + ], + "score": 1.0, + "content": "robustness of the smoothed model to initialization conditions in Appendix C.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5 + }, + { + "type": "table", + "bbox": [ + 158, + 468, + 452, + 538 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 158, + 468, + 452, + 538 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 158, + 468, + 452, + 538 + ], + "spans": [ + { + "bbox": [ + 158, + 468, + 452, + 538 + ], + "score": 0.979, + "html": "
KLPearson RSpearman R
Matrix Factorization0.01730.85490.8374
Complex Bilinear Factorization0.01410.87710.8636
POE0.01700.85480.8511
Box0.01470.87750.8768
Smoothed Box0.01380.89850.8977
", + "type": "table", + "image_path": "1daf555dd90eed27e67a4846f34576407424d24f1ba9de76ade32d78a28579ea.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 158, + 468, + 452, + 491.3333333333333 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 158, + 491.3333333333333, + 452, + 514.6666666666666 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 158, + 514.6666666666666, + 452, + 538.0 + ], + "spans": [], + "index": 29 + } + ] + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 546, + 504, + 568 + ], + "lines": [ + { + "bbox": [ + 105, + 545, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 505, + 559 + ], + "score": 1.0, + "content": "Table 7: Performance of the smoothed model, the original box model, and several baselines on", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 555, + 157, + 570 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 157, + 570 + ], + "score": 1.0, + "content": "MovieLens.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + }, + { + "type": "title", + "bbox": [ + 107, + 592, + 303, + 605 + ], + "lines": [ + { + "bbox": [ + 105, + 591, + 304, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 591, + 304, + 608 + ], + "score": 1.0, + "content": "6 CONCLUSION AND FUTURE WORK", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 617, + 505, + 673 + ], + "lines": [ + { + "bbox": [ + 106, + 618, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 618, + 505, + 630 + ], + "score": 1.0, + "content": "We presented an approach to smoothing the energy and optimization landscape of probabilistic box", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 629, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 629, + 505, + 641 + ], + "score": 1.0, + "content": "embeddings and provided a theoretical justification for the smoothing. Due to a decreased number", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 639, + 505, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 505, + 653 + ], + "score": 1.0, + "content": "of hyper-parameters this model is easier to train, and, furthermore, met or surpassed current state-of-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 650, + 505, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 505, + 664 + ], + "score": 1.0, + "content": "the-art results on several interesting datasets. We further demonstrated that this model is particularly", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 661, + 398, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 398, + 675 + ], + "score": 1.0, + "content": "effective in the case of sparse data and more robust to poor initialization.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 108, + 678, + 504, + 712 + ], + "lines": [ + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "score": 1.0, + "content": "Tackling the learning problems presented by rich, geometrically-inspired embedding models is an", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 690, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 690, + 506, + 702 + ], + "score": 1.0, + "content": "open and challenging area of research, which this work is far from the last word on. This task will", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 700, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 506, + 713 + ], + "score": 1.0, + "content": "become even more pressing as the embedding structures become more complex, such as unions of", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 39 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 114, + 721, + 484, + 732 + ], + "lines": [ + { + "bbox": [ + 118, + 719, + 485, + 734 + ], + "spans": [ + { + "bbox": [ + 118, + 719, + 293, + 734 + ], + "score": 1.0, + "content": "2In the dev and test data, we also remove all the", + "type": "text" + }, + { + "bbox": [ + 293, + 722, + 325, + 732 + ], + "score": 0.9, + "content": "P ( A | B )", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 719, + 350, + 734 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 351, + 722, + 382, + 732 + ], + "score": 0.9, + "content": "P ( B | A )", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 719, + 485, + 734 + ], + "score": 1.0, + "content": "appears in the training data.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 751, + 309, + 759 + ], + "lines": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "spans": [ + { + "bbox": [ + 302, + 751, + 309, + 762 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 217, + 80, + 393, + 260 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 217, + 80, + 393, + 260 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 217, + 80, + 393, + 260 + ], + "spans": [ + { + "bbox": [ + 217, + 80, + 393, + 260 + ], + "score": 0.977, + "html": "
P(xly)
Full test data POEKL 0.031Pearson R 0.949
POE* Box0.031 0.0200.949 0.967
Smoothed Box0.0180.969
Unseen pairs POE0.0480.920 0.925
POE* Box Smoothed Box0.046 0.025 0.0240.957 0.957
Unseen captions
POE0.1270.696
POE*0.0840.854
Box Smoothed Box0.050 0.0360.900 0.917
", + "type": "table", + "image_path": "ed4366fc3a34a30a7f83b854427a16678abc8d80af3ded5a560277b258db13af.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 217, + 80, + 393, + 93.84615384615384 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 217, + 93.84615384615384, + 393, + 107.69230769230768 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 217, + 107.69230769230768, + 393, + 121.53846153846152 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 217, + 121.53846153846152, + 393, + 135.38461538461536 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 217, + 135.38461538461536, + 393, + 149.2307692307692 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 217, + 149.2307692307692, + 393, + 163.07692307692304 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 217, + 163.07692307692304, + 393, + 176.92307692307688 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 217, + 176.92307692307688, + 393, + 190.76923076923072 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 217, + 190.76923076923072, + 393, + 204.61538461538456 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 217, + 204.61538461538456, + 393, + 218.4615384615384 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 217, + 218.4615384615384, + 393, + 232.30769230769224 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 217, + 232.30769230769224, + 393, + 246.15384615384608 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 217, + 246.15384615384608, + 393, + 259.99999999999994 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 157, + 268, + 452, + 280 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 157, + 266, + 454, + 283 + ], + "spans": [ + { + "bbox": [ + 157, + 266, + 454, + 283 + ], + "score": 1.0, + "content": "Table 6: KL and Pearson correlation between model and gold probability.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 106, + 302, + 504, + 324 + ], + "lines": [], + "index": 14.5, + "bbox_fs": [ + 105, + 302, + 505, + 325 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 330, + 505, + 407 + ], + "lines": [ + { + "bbox": [ + 106, + 329, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 505, + 342 + ], + "score": 1.0, + "content": "We compare with several baselines: low-rank matrix factorization, complex bilinear factoriza-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 340, + 505, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 340, + 505, + 353 + ], + "score": 1.0, + "content": "tion (Trouillon et al., 2016), and two hierarchical embedding methods, POE (Lai & Hockenmaier,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 352, + 505, + 364 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 505, + 364 + ], + "score": 1.0, + "content": "2017) and the Box Lattice (Vilnis et al., 2018). Since the training matrix is asymmetric, we used", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 362, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 362, + 506, + 375 + ], + "score": 1.0, + "content": "separate embeddings for target and conditioned movies. For the complex bilinear model, we added", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 373, + 505, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 505, + 387 + ], + "score": 1.0, + "content": "one additional vector of parameters to capture the “imply” relation. We evaluate on the test set using", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 384, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 505, + 398 + ], + "score": 1.0, + "content": "KL divergence, Pearson correlation, and Spearman correlation with the ground truth probabilities.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 396, + 302, + 408 + ], + "spans": [ + { + "bbox": [ + 106, + 396, + 302, + 408 + ], + "score": 1.0, + "content": "Experimental details are given in Appendix D.4.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 329, + 506, + 408 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 412, + 504, + 457 + ], + "lines": [ + { + "bbox": [ + 105, + 412, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 505, + 425 + ], + "score": 1.0, + "content": "From the results in Table 7, we can see that our smoothed box embedding method outperforms the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 423, + 506, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 423, + 506, + 436 + ], + "score": 1.0, + "content": "original box lattice as well as all other baselines’ performances, especially in Spearman correlation,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 435, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 106, + 435, + 505, + 447 + ], + "score": 1.0, + "content": "the most relevant metric for recommendation, a ranking task. We perform an additional study on the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 446, + 415, + 457 + ], + "spans": [ + { + "bbox": [ + 106, + 446, + 415, + 457 + ], + "score": 1.0, + "content": "robustness of the smoothed model to initialization conditions in Appendix C.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5, + "bbox_fs": [ + 105, + 412, + 506, + 457 + ] + }, + { + "type": "table", + "bbox": [ + 158, + 468, + 452, + 538 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 158, + 468, + 452, + 538 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 158, + 468, + 452, + 538 + ], + "spans": [ + { + "bbox": [ + 158, + 468, + 452, + 538 + ], + "score": 0.979, + "html": "
KLPearson RSpearman R
Matrix Factorization0.01730.85490.8374
Complex Bilinear Factorization0.01410.87710.8636
POE0.01700.85480.8511
Box0.01470.87750.8768
Smoothed Box0.01380.89850.8977
", + "type": "table", + "image_path": "1daf555dd90eed27e67a4846f34576407424d24f1ba9de76ade32d78a28579ea.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 158, + 468, + 452, + 491.3333333333333 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 158, + 491.3333333333333, + 452, + 514.6666666666666 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 158, + 514.6666666666666, + 452, + 538.0 + ], + "spans": [], + "index": 29 + } + ] + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 546, + 504, + 568 + ], + "lines": [ + { + "bbox": [ + 105, + 545, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 505, + 559 + ], + "score": 1.0, + "content": "Table 7: Performance of the smoothed model, the original box model, and several baselines on", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 555, + 157, + 570 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 157, + 570 + ], + "score": 1.0, + "content": "MovieLens.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 545, + 505, + 570 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 592, + 303, + 605 + ], + "lines": [ + { + "bbox": [ + 105, + 591, + 304, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 591, + 304, + 608 + ], + "score": 1.0, + "content": "6 CONCLUSION AND FUTURE WORK", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 107, + 617, + 505, + 673 + ], + "lines": [ + { + "bbox": [ + 106, + 618, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 106, + 618, + 505, + 630 + ], + "score": 1.0, + "content": "We presented an approach to smoothing the energy and optimization landscape of probabilistic box", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 629, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 629, + 505, + 641 + ], + "score": 1.0, + "content": "embeddings and provided a theoretical justification for the smoothing. Due to a decreased number", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 639, + 505, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 505, + 653 + ], + "score": 1.0, + "content": "of hyper-parameters this model is easier to train, and, furthermore, met or surpassed current state-of-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 650, + 505, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 505, + 664 + ], + "score": 1.0, + "content": "the-art results on several interesting datasets. We further demonstrated that this model is particularly", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 661, + 398, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 398, + 675 + ], + "score": 1.0, + "content": "effective in the case of sparse data and more robust to poor initialization.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 35, + "bbox_fs": [ + 105, + 618, + 505, + 675 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 678, + 504, + 712 + ], + "lines": [ + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "score": 1.0, + "content": "Tackling the learning problems presented by rich, geometrically-inspired embedding models is an", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 690, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 690, + 506, + 702 + ], + "score": 1.0, + "content": "open and challenging area of research, which this work is far from the last word on. This task will", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 700, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 506, + 713 + ], + "score": 1.0, + "content": "become even more pressing as the embedding structures become more complex, such as unions of", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "boxes or other non-convex objects. To this end, we will continue to explore both function lattices,", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 286, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 286, + 107 + ], + "score": 1.0, + "content": "and constraint-based approaches to learning.", + "type": "text", + "cross_page": true + } + ], + "index": 1 + } + ], + "index": 39, + "bbox_fs": [ + 105, + 677, + 506, + 713 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 105 + ], + "lines": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 96 + ], + "score": 1.0, + "content": "boxes or other non-convex objects. To this end, we will continue to explore both function lattices,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 93, + 286, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 286, + 107 + ], + "score": 1.0, + "content": "and constraint-based approaches to learning.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "title", + "bbox": [ + 108, + 122, + 237, + 134 + ], + "lines": [ + { + "bbox": [ + 105, + 121, + 238, + 136 + ], + "spans": [ + { + "bbox": [ + 105, + 121, + 238, + 136 + ], + "score": 1.0, + "content": "7 ACKNOWLEDGMENTS", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 107, + 147, + 505, + 225 + ], + "lines": [ + { + "bbox": [ + 106, + 146, + 505, + 159 + ], + "spans": [ + { + "bbox": [ + 106, + 146, + 505, + 159 + ], + "score": 1.0, + "content": "We thank Travis Wolfe, Colin Evans, Rob Zinkov, Ben Poole, and Laurent Dinh for helpful dis-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 159, + 505, + 170 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 505, + 170 + ], + "score": 1.0, + "content": "cussions. We also thank the anonymous reviewers for their constructive feedback. This work was", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 169, + 506, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 506, + 182 + ], + "score": 1.0, + "content": "supported in part by the Center for Intelligent Information Retrieval and the Center for Data Science,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 180, + 505, + 193 + ], + "spans": [ + { + "bbox": [ + 105, + 180, + 505, + 193 + ], + "score": 1.0, + "content": "in part by the Chan Zuckerberg Initiative under the project Scientific Knowledge Base Construction,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 191, + 505, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 505, + 203 + ], + "score": 1.0, + "content": "and in part by the National Science Foundation under Grant No. IIS-1514053. Any opinions, find-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 203, + 505, + 214 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 505, + 214 + ], + "score": 1.0, + "content": "ings and conclusions or recommendations expressed in this material are those of the authors and do", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 214, + 280, + 225 + ], + "spans": [ + { + "bbox": [ + 106, + 214, + 280, + 225 + ], + "score": 1.0, + "content": "not necessarily reflect those of the sponsor.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6 + }, + { + "type": "title", + "bbox": [ + 107, + 241, + 175, + 254 + ], + "lines": [ + { + "bbox": [ + 106, + 242, + 176, + 255 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 176, + 255 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 260, + 487, + 272 + ], + "lines": [ + { + "bbox": [ + 106, + 260, + 487, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 260, + 487, + 273 + ], + "score": 1.0, + "content": "Ben Athiwaratkun and Andrew Gordon Wilson. Multimodal word distributions. In ACL, 2017.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11 + }, + { + "type": "text", + "bbox": [ + 108, + 280, + 504, + 314 + ], + "lines": [ + { + "bbox": [ + 105, + 279, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 505, + 295 + ], + "score": 1.0, + "content": "Ben Athiwaratkun and Andrew Gordon Wilson. Hierarchical density order embeddings. In Interna-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 291, + 505, + 305 + ], + "spans": [ + { + "bbox": [ + 115, + 291, + 505, + 305 + ], + "score": 1.0, + "content": "tional Conference on Learning Representations, 2018. URL https://openreview.net/", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 117, + 303, + 229, + 315 + ], + "spans": [ + { + "bbox": [ + 117, + 303, + 164, + 315 + ], + "score": 1.0, + "content": "forum?id", + "type": "text" + }, + { + "bbox": [ + 165, + 304, + 172, + 313 + ], + "score": 0.61, + "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 303, + 229, + 315 + ], + "score": 1.0, + "content": "HJCXZQbAZ.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 106, + 323, + 503, + 346 + ], + "lines": [ + { + "bbox": [ + 106, + 323, + 505, + 336 + ], + "spans": [ + { + "bbox": [ + 106, + 323, + 505, + 336 + ], + "score": 1.0, + "content": "Yoshua Bengio, Rejean Ducharme, Pascal Vincent, and Christian Jauvin. A neural probabilistic ´", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 335, + 445, + 347 + ], + "spans": [ + { + "bbox": [ + 115, + 335, + 445, + 347 + ], + "score": 1.0, + "content": "language model. Journal of machine learning research, 3(Feb):1137–1155, 2003.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 107, + 353, + 505, + 388 + ], + "lines": [ + { + "bbox": [ + 106, + 354, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 354, + 505, + 367 + ], + "score": 1.0, + "content": "Shannon R Bowling, Mohammad T Khasawneh, Sittichai Kaewkuekool, and Byung R Cho. A", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 364, + 505, + 379 + ], + "spans": [ + { + "bbox": [ + 115, + 364, + 505, + 379 + ], + "score": 1.0, + "content": "logistic approximation to the cumulative normal distribution. Journal of Industrial Engineering", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 376, + 237, + 388 + ], + "spans": [ + { + "bbox": [ + 115, + 376, + 237, + 388 + ], + "score": 1.0, + "content": "and Management, 2(1), 2009.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 396, + 504, + 430 + ], + "lines": [ + { + "bbox": [ + 105, + 396, + 504, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 504, + 410 + ], + "score": 1.0, + "content": "Scott Deerwester, Susan T Dumais, George W Furnas, Thomas K Landauer, and Richard Harshman.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 407, + 504, + 421 + ], + "spans": [ + { + "bbox": [ + 115, + 407, + 504, + 421 + ], + "score": 1.0, + "content": "Indexing by latent semantic analysis. Journal of the American society for information science, 41", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 419, + 196, + 430 + ], + "spans": [ + { + "bbox": [ + 116, + 419, + 196, + 430 + ], + "score": 1.0, + "content": "(6):391–407, 1990.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 106, + 438, + 504, + 462 + ], + "lines": [ + { + "bbox": [ + 106, + 438, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 505, + 452 + ], + "score": 1.0, + "content": "Octavian-Eugen Ganea, Gary Becigneul, and Thomas Hofmann. Hyperbolic entailment cones for ´", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 450, + 309, + 462 + ], + "spans": [ + { + "bbox": [ + 115, + 450, + 309, + 462 + ], + "score": 1.0, + "content": "learning hierarchical embeddings. ICML, 2018.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 106, + 470, + 503, + 493 + ], + "lines": [ + { + "bbox": [ + 106, + 469, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 469, + 505, + 484 + ], + "score": 1.0, + "content": "Caglar Gulcehre, Marcin Moczulski, Misha Denil, and Yoshua Bengio. Noisy activation functions.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 481, + 416, + 493 + ], + "spans": [ + { + "bbox": [ + 115, + 481, + 416, + 493 + ], + "score": 1.0, + "content": "In International Conference on Machine Learning, pp. 3059–3068, 2016a.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 107, + 501, + 503, + 524 + ], + "lines": [ + { + "bbox": [ + 106, + 500, + 504, + 514 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 504, + 514 + ], + "score": 1.0, + "content": "Caglar Gulcehre, Marcin Moczulski, Francesco Visin, and Yoshua Bengio. Mollifying networks.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 513, + 283, + 524 + ], + "spans": [ + { + "bbox": [ + 116, + 513, + 283, + 524 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1608.04980, 2016b.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 107, + 532, + 503, + 555 + ], + "lines": [ + { + "bbox": [ + 105, + 532, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 505, + 546 + ], + "score": 1.0, + "content": "Tony Jebara, Risi Kondor, and Andrew Howard. Probability product kernels. Journal of Machine", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 116, + 543, + 288, + 557 + ], + "spans": [ + { + "bbox": [ + 116, + 543, + 288, + 557 + ], + "score": 1.0, + "content": "Learning Research, 5(Jul):819–844, 2004.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 106, + 564, + 503, + 587 + ], + "lines": [ + { + "bbox": [ + 105, + 562, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 562, + 505, + 577 + ], + "score": 1.0, + "content": "Alice Lai and Julia Hockenmaier. Learning to predict denotational probabilities for modeling en-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 116, + 575, + 220, + 586 + ], + "spans": [ + { + "bbox": [ + 116, + 575, + 220, + 586 + ], + "score": 1.0, + "content": "tailment. In EACL, 2017.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 106, + 595, + 502, + 608 + ], + "lines": [ + { + "bbox": [ + 106, + 595, + 503, + 608 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 503, + 608 + ], + "score": 1.0, + "content": "Solomon Leader. Measures on semilattices. Pacific Journal of Mathematics, 39(2):407–423, 1971.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 105, + 615, + 503, + 639 + ], + "lines": [ + { + "bbox": [ + 105, + 614, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 505, + 630 + ], + "score": 1.0, + "content": "Xiang Li, Luke Vilnis, and Andrew McCallum. Improved representation learning for predicting", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 627, + 417, + 639 + ], + "spans": [ + { + "bbox": [ + 115, + 627, + 417, + 639 + ], + "score": 1.0, + "content": "commonsense ontologies. NIPS Workshop on Structured Prediction, 2017.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 106, + 646, + 504, + 670 + ], + "lines": [ + { + "bbox": [ + 106, + 646, + 504, + 660 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 504, + 660 + ], + "score": 1.0, + "content": "Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean. Distributed representa-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 658, + 396, + 670 + ], + "spans": [ + { + "bbox": [ + 116, + 658, + 396, + 670 + ], + "score": 1.0, + "content": "tions of words and phrases and their compositionality. In NIPS, 2013.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 105, + 678, + 504, + 701 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 504, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 504, + 690 + ], + "score": 1.0, + "content": "Andriy Mnih and Geoffrey E Hinton. A scalable hierarchical distributed language model. In Ad-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 690, + 405, + 702 + ], + "spans": [ + { + "bbox": [ + 116, + 690, + 405, + 702 + ], + "score": 1.0, + "content": "vances in neural information processing systems, pp. 1081–1088, 2009.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5 + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 504, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "score": 1.0, + "content": "Hossein Mobahi. Training recurrent neural networks by diffusion. arXiv preprint arXiv:1601.04114,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 114, + 719, + 143, + 732 + ], + "spans": [ + { + "bbox": [ + 114, + 719, + 143, + 732 + ], + "score": 1.0, + "content": "2016.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 301, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 82, + 504, + 105 + ], + "lines": [], + "index": 0.5, + "bbox_fs": [ + 105, + 81, + 505, + 107 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 108, + 122, + 237, + 134 + ], + "lines": [ + { + "bbox": [ + 105, + 121, + 238, + 136 + ], + "spans": [ + { + "bbox": [ + 105, + 121, + 238, + 136 + ], + "score": 1.0, + "content": "7 ACKNOWLEDGMENTS", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 107, + 147, + 505, + 225 + ], + "lines": [ + { + "bbox": [ + 106, + 146, + 505, + 159 + ], + "spans": [ + { + "bbox": [ + 106, + 146, + 505, + 159 + ], + "score": 1.0, + "content": "We thank Travis Wolfe, Colin Evans, Rob Zinkov, Ben Poole, and Laurent Dinh for helpful dis-", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 159, + 505, + 170 + ], + "spans": [ + { + "bbox": [ + 106, + 159, + 505, + 170 + ], + "score": 1.0, + "content": "cussions. We also thank the anonymous reviewers for their constructive feedback. This work was", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 169, + 506, + 182 + ], + "spans": [ + { + "bbox": [ + 105, + 169, + 506, + 182 + ], + "score": 1.0, + "content": "supported in part by the Center for Intelligent Information Retrieval and the Center for Data Science,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 180, + 505, + 193 + ], + "spans": [ + { + "bbox": [ + 105, + 180, + 505, + 193 + ], + "score": 1.0, + "content": "in part by the Chan Zuckerberg Initiative under the project Scientific Knowledge Base Construction,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 191, + 505, + 203 + ], + "spans": [ + { + "bbox": [ + 105, + 191, + 505, + 203 + ], + "score": 1.0, + "content": "and in part by the National Science Foundation under Grant No. IIS-1514053. Any opinions, find-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 203, + 505, + 214 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 505, + 214 + ], + "score": 1.0, + "content": "ings and conclusions or recommendations expressed in this material are those of the authors and do", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 214, + 280, + 225 + ], + "spans": [ + { + "bbox": [ + 106, + 214, + 280, + 225 + ], + "score": 1.0, + "content": "not necessarily reflect those of the sponsor.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6, + "bbox_fs": [ + 105, + 146, + 506, + 225 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 241, + 175, + 254 + ], + "lines": [ + { + "bbox": [ + 106, + 242, + 176, + 255 + ], + "spans": [ + { + "bbox": [ + 106, + 242, + 176, + 255 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 106, + 260, + 487, + 272 + ], + "lines": [ + { + "bbox": [ + 106, + 260, + 487, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 260, + 487, + 273 + ], + "score": 1.0, + "content": "Ben Athiwaratkun and Andrew Gordon Wilson. Multimodal word distributions. In ACL, 2017.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 11, + "bbox_fs": [ + 106, + 260, + 487, + 273 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 280, + 504, + 314 + ], + "lines": [ + { + "bbox": [ + 105, + 279, + 505, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 279, + 505, + 295 + ], + "score": 1.0, + "content": "Ben Athiwaratkun and Andrew Gordon Wilson. Hierarchical density order embeddings. In Interna-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 115, + 291, + 505, + 305 + ], + "spans": [ + { + "bbox": [ + 115, + 291, + 505, + 305 + ], + "score": 1.0, + "content": "tional Conference on Learning Representations, 2018. URL https://openreview.net/", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 117, + 303, + 229, + 315 + ], + "spans": [ + { + "bbox": [ + 117, + 303, + 164, + 315 + ], + "score": 1.0, + "content": "forum?id", + "type": "text" + }, + { + "bbox": [ + 165, + 304, + 172, + 313 + ], + "score": 0.61, + "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 303, + 229, + 315 + ], + "score": 1.0, + "content": "HJCXZQbAZ.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13, + "bbox_fs": [ + 105, + 279, + 505, + 315 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 323, + 503, + 346 + ], + "lines": [ + { + "bbox": [ + 106, + 323, + 505, + 336 + ], + "spans": [ + { + "bbox": [ + 106, + 323, + 505, + 336 + ], + "score": 1.0, + "content": "Yoshua Bengio, Rejean Ducharme, Pascal Vincent, and Christian Jauvin. A neural probabilistic ´", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 115, + 335, + 445, + 347 + ], + "spans": [ + { + "bbox": [ + 115, + 335, + 445, + 347 + ], + "score": 1.0, + "content": "language model. Journal of machine learning research, 3(Feb):1137–1155, 2003.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 106, + 323, + 505, + 347 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 353, + 505, + 388 + ], + "lines": [ + { + "bbox": [ + 106, + 354, + 505, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 354, + 505, + 367 + ], + "score": 1.0, + "content": "Shannon R Bowling, Mohammad T Khasawneh, Sittichai Kaewkuekool, and Byung R Cho. A", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 115, + 364, + 505, + 379 + ], + "spans": [ + { + "bbox": [ + 115, + 364, + 505, + 379 + ], + "score": 1.0, + "content": "logistic approximation to the cumulative normal distribution. Journal of Industrial Engineering", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 115, + 376, + 237, + 388 + ], + "spans": [ + { + "bbox": [ + 115, + 376, + 237, + 388 + ], + "score": 1.0, + "content": "and Management, 2(1), 2009.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18, + "bbox_fs": [ + 106, + 354, + 505, + 388 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 396, + 504, + 430 + ], + "lines": [ + { + "bbox": [ + 105, + 396, + 504, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 504, + 410 + ], + "score": 1.0, + "content": "Scott Deerwester, Susan T Dumais, George W Furnas, Thomas K Landauer, and Richard Harshman.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 115, + 407, + 504, + 421 + ], + "spans": [ + { + "bbox": [ + 115, + 407, + 504, + 421 + ], + "score": 1.0, + "content": "Indexing by latent semantic analysis. Journal of the American society for information science, 41", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 116, + 419, + 196, + 430 + ], + "spans": [ + { + "bbox": [ + 116, + 419, + 196, + 430 + ], + "score": 1.0, + "content": "(6):391–407, 1990.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21, + "bbox_fs": [ + 105, + 396, + 504, + 430 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 438, + 504, + 462 + ], + "lines": [ + { + "bbox": [ + 106, + 438, + 505, + 452 + ], + "spans": [ + { + "bbox": [ + 106, + 438, + 505, + 452 + ], + "score": 1.0, + "content": "Octavian-Eugen Ganea, Gary Becigneul, and Thomas Hofmann. Hyperbolic entailment cones for ´", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 115, + 450, + 309, + 462 + ], + "spans": [ + { + "bbox": [ + 115, + 450, + 309, + 462 + ], + "score": 1.0, + "content": "learning hierarchical embeddings. ICML, 2018.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5, + "bbox_fs": [ + 106, + 438, + 505, + 462 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 470, + 503, + 493 + ], + "lines": [ + { + "bbox": [ + 106, + 469, + 505, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 469, + 505, + 484 + ], + "score": 1.0, + "content": "Caglar Gulcehre, Marcin Moczulski, Misha Denil, and Yoshua Bengio. Noisy activation functions.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 115, + 481, + 416, + 493 + ], + "spans": [ + { + "bbox": [ + 115, + 481, + 416, + 493 + ], + "score": 1.0, + "content": "In International Conference on Machine Learning, pp. 3059–3068, 2016a.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 106, + 469, + 505, + 493 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 501, + 503, + 524 + ], + "lines": [ + { + "bbox": [ + 106, + 500, + 504, + 514 + ], + "spans": [ + { + "bbox": [ + 106, + 500, + 504, + 514 + ], + "score": 1.0, + "content": "Caglar Gulcehre, Marcin Moczulski, Francesco Visin, and Yoshua Bengio. Mollifying networks.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 116, + 513, + 283, + 524 + ], + "spans": [ + { + "bbox": [ + 116, + 513, + 283, + 524 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1608.04980, 2016b.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5, + "bbox_fs": [ + 106, + 500, + 504, + 524 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 532, + 503, + 555 + ], + "lines": [ + { + "bbox": [ + 105, + 532, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 505, + 546 + ], + "score": 1.0, + "content": "Tony Jebara, Risi Kondor, and Andrew Howard. Probability product kernels. Journal of Machine", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 116, + 543, + 288, + 557 + ], + "spans": [ + { + "bbox": [ + 116, + 543, + 288, + 557 + ], + "score": 1.0, + "content": "Learning Research, 5(Jul):819–844, 2004.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 532, + 505, + 557 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 564, + 503, + 587 + ], + "lines": [ + { + "bbox": [ + 105, + 562, + 505, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 562, + 505, + 577 + ], + "score": 1.0, + "content": "Alice Lai and Julia Hockenmaier. Learning to predict denotational probabilities for modeling en-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 116, + 575, + 220, + 586 + ], + "spans": [ + { + "bbox": [ + 116, + 575, + 220, + 586 + ], + "score": 1.0, + "content": "tailment. In EACL, 2017.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 105, + 562, + 505, + 586 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 595, + 502, + 608 + ], + "lines": [ + { + "bbox": [ + 106, + 595, + 503, + 608 + ], + "spans": [ + { + "bbox": [ + 106, + 595, + 503, + 608 + ], + "score": 1.0, + "content": "Solomon Leader. Measures on semilattices. Pacific Journal of Mathematics, 39(2):407–423, 1971.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33, + "bbox_fs": [ + 106, + 595, + 503, + 608 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 615, + 503, + 639 + ], + "lines": [ + { + "bbox": [ + 105, + 614, + 505, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 614, + 505, + 630 + ], + "score": 1.0, + "content": "Xiang Li, Luke Vilnis, and Andrew McCallum. Improved representation learning for predicting", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 115, + 627, + 417, + 639 + ], + "spans": [ + { + "bbox": [ + 115, + 627, + 417, + 639 + ], + "score": 1.0, + "content": "commonsense ontologies. NIPS Workshop on Structured Prediction, 2017.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5, + "bbox_fs": [ + 105, + 614, + 505, + 639 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 646, + 504, + 670 + ], + "lines": [ + { + "bbox": [ + 106, + 646, + 504, + 660 + ], + "spans": [ + { + "bbox": [ + 106, + 646, + 504, + 660 + ], + "score": 1.0, + "content": "Tomas Mikolov, Ilya Sutskever, Kai Chen, Greg S Corrado, and Jeff Dean. Distributed representa-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 116, + 658, + 396, + 670 + ], + "spans": [ + { + "bbox": [ + 116, + 658, + 396, + 670 + ], + "score": 1.0, + "content": "tions of words and phrases and their compositionality. In NIPS, 2013.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36.5, + "bbox_fs": [ + 106, + 646, + 504, + 670 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 678, + 504, + 701 + ], + "lines": [ + { + "bbox": [ + 106, + 677, + 504, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 504, + 690 + ], + "score": 1.0, + "content": "Andriy Mnih and Geoffrey E Hinton. A scalable hierarchical distributed language model. In Ad-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 116, + 690, + 405, + 702 + ], + "spans": [ + { + "bbox": [ + 116, + 690, + 405, + 702 + ], + "score": 1.0, + "content": "vances in neural information processing systems, pp. 1081–1088, 2009.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 38.5, + "bbox_fs": [ + 106, + 677, + 504, + 702 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 709, + 504, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 708, + 505, + 722 + ], + "score": 1.0, + "content": "Hossein Mobahi. Training recurrent neural networks by diffusion. arXiv preprint arXiv:1601.04114,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 114, + 719, + 143, + 732 + ], + "spans": [ + { + "bbox": [ + 114, + 719, + 143, + 732 + ], + "score": 1.0, + "content": "2016.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 708, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Arvind Neelakantan, Luke Vilnis, Quoc V Le, Ilya Sutskever, Lukasz Kaiser, Karol Kurach, and", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "James Martens. Adding gradient noise improves learning for very deep networks. arXiv preprint", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 105, + 220, + 115 + ], + "spans": [ + { + "bbox": [ + 116, + 105, + 220, + 115 + ], + "score": 1.0, + "content": "arXiv:1511.06807, 2015.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 106, + 123, + 518, + 190 + ], + "lines": [ + { + "bbox": [ + 105, + 122, + 505, + 137 + ], + "spans": [ + { + "bbox": [ + 105, + 122, + 505, + 137 + ], + "score": 1.0, + "content": "Maximillian Nickel and Douwe Kiela. Poincare embeddings for learning hierarchical rep- ´", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 134, + 505, + 147 + ], + "spans": [ + { + "bbox": [ + 115, + 134, + 505, + 147 + ], + "score": 1.0, + "content": "resentations. In I. Guyon, U. V. Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vish-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 145, + 505, + 158 + ], + "spans": [ + { + "bbox": [ + 115, + 145, + 505, + 158 + ], + "score": 1.0, + "content": "wanathan, and R. Garnett (eds.), Advances in Neural Information Processing Systems 30,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 155, + 505, + 169 + ], + "spans": [ + { + "bbox": [ + 115, + 155, + 505, + 169 + ], + "score": 1.0, + "content": "pp. 6338–6347. Curran Associates, Inc., 2017. URL http://papers.nips.cc/paper/", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 166, + 521, + 180 + ], + "spans": [ + { + "bbox": [ + 115, + 166, + 521, + 180 + ], + "score": 1.0, + "content": "7213-poincare-embeddings-for-learning-hierarchical-representations.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 113, + 178, + 141, + 190 + ], + "spans": [ + { + "bbox": [ + 113, + 178, + 141, + 190 + ], + "score": 1.0, + "content": "pdf.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 108, + 197, + 504, + 220 + ], + "lines": [ + { + "bbox": [ + 106, + 195, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 106, + 195, + 505, + 210 + ], + "score": 1.0, + "content": "Gerard Salton, Anita Wong, and Chung-Shu Yang. A vector space model for automatic indexing.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 117, + 208, + 331, + 220 + ], + "spans": [ + { + "bbox": [ + 117, + 208, + 331, + 220 + ], + "score": 1.0, + "content": "Communications of the ACM, 18(11):613–620, 1975.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5 + }, + { + "type": "text", + "bbox": [ + 107, + 226, + 505, + 249 + ], + "lines": [ + { + "bbox": [ + 105, + 226, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 505, + 240 + ], + "score": 1.0, + "content": "Sandeep Subramanian and Soumen Chakrabarti. New embedded representations and evaluation", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 238, + 365, + 250 + ], + "spans": [ + { + "bbox": [ + 115, + 238, + 365, + 250 + ], + "score": 1.0, + "content": "protocols for inferring transitive relations. SIGIR 2018, 2018.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 106, + 257, + 505, + 291 + ], + "lines": [ + { + "bbox": [ + 105, + 257, + 505, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 505, + 270 + ], + "score": 1.0, + "content": "Theo Trouillon, Johannes Welbl, Sebastian Riedel, ´ Eric Gaussier, and Guillaume Bouchard. Com- ´", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 268, + 506, + 282 + ], + "spans": [ + { + "bbox": [ + 115, + 268, + 506, + 282 + ], + "score": 1.0, + "content": "plex embeddings for simple link prediction. In International Conference on Machine Learning,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 280, + 207, + 291 + ], + "spans": [ + { + "bbox": [ + 115, + 280, + 207, + 291 + ], + "score": 1.0, + "content": "pp. 2071–2080, 2016.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 105, + 298, + 504, + 321 + ], + "lines": [ + { + "bbox": [ + 105, + 297, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 105, + 297, + 505, + 312 + ], + "score": 1.0, + "content": "Ivan Vendrov, Ryan Kiros, Sanja Fidler, and Raquel Urtasun. Order-embeddings of images and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 309, + 221, + 321 + ], + "spans": [ + { + "bbox": [ + 115, + 309, + 221, + 321 + ], + "score": 1.0, + "content": "language. In ICLR, 2016.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 106, + 328, + 503, + 341 + ], + "lines": [ + { + "bbox": [ + 105, + 328, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 505, + 342 + ], + "score": 1.0, + "content": "Luke Vilnis and Andrew McCallum. Word representations via gaussian embedding. In ICLR, 2015.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 105, + 347, + 504, + 370 + ], + "lines": [ + { + "bbox": [ + 105, + 347, + 505, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 505, + 360 + ], + "score": 1.0, + "content": "Luke Vilnis, Xiang Li, Shikhar Murty, and Andrew McCallum. Probabilistic embedding of knowl-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 358, + 505, + 372 + ], + "spans": [ + { + "bbox": [ + 115, + 358, + 505, + 372 + ], + "score": 1.0, + "content": "edge graphs with box lattice measures. In ACL. Association for Computational Linguistics, 2018.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 106, + 377, + 504, + 400 + ], + "lines": [ + { + "bbox": [ + 105, + 375, + 506, + 392 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 506, + 392 + ], + "score": 1.0, + "content": "Adriaan C. Zaanen. Introduction to Operator Theory in Riesz Spaces. Springer Berlin Heidelberg,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 117, + 389, + 237, + 400 + ], + "spans": [ + { + "bbox": [ + 117, + 389, + 237, + 400 + ], + "score": 1.0, + "content": "1997. ISBN 9783642644870.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 310, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 312, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 82, + 505, + 116 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 505, + 95 + ], + "score": 1.0, + "content": "Arvind Neelakantan, Luke Vilnis, Quoc V Le, Ilya Sutskever, Lukasz Kaiser, Karol Kurach, and", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "spans": [ + { + "bbox": [ + 115, + 93, + 505, + 106 + ], + "score": 1.0, + "content": "James Martens. Adding gradient noise improves learning for very deep networks. arXiv preprint", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 116, + 105, + 220, + 115 + ], + "spans": [ + { + "bbox": [ + 116, + 105, + 220, + 115 + ], + "score": 1.0, + "content": "arXiv:1511.06807, 2015.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 106, + 82, + 505, + 115 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 123, + 518, + 190 + ], + "lines": [ + { + "bbox": [ + 105, + 122, + 505, + 137 + ], + "spans": [ + { + "bbox": [ + 105, + 122, + 505, + 137 + ], + "score": 1.0, + "content": "Maximillian Nickel and Douwe Kiela. Poincare embeddings for learning hierarchical rep- ´", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 115, + 134, + 505, + 147 + ], + "spans": [ + { + "bbox": [ + 115, + 134, + 505, + 147 + ], + "score": 1.0, + "content": "resentations. In I. Guyon, U. V. Luxburg, S. Bengio, H. Wallach, R. Fergus, S. Vish-", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 115, + 145, + 505, + 158 + ], + "spans": [ + { + "bbox": [ + 115, + 145, + 505, + 158 + ], + "score": 1.0, + "content": "wanathan, and R. Garnett (eds.), Advances in Neural Information Processing Systems 30,", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 115, + 155, + 505, + 169 + ], + "spans": [ + { + "bbox": [ + 115, + 155, + 505, + 169 + ], + "score": 1.0, + "content": "pp. 6338–6347. Curran Associates, Inc., 2017. URL http://papers.nips.cc/paper/", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 115, + 166, + 521, + 180 + ], + "spans": [ + { + "bbox": [ + 115, + 166, + 521, + 180 + ], + "score": 1.0, + "content": "7213-poincare-embeddings-for-learning-hierarchical-representations.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 113, + 178, + 141, + 190 + ], + "spans": [ + { + "bbox": [ + 113, + 178, + 141, + 190 + ], + "score": 1.0, + "content": "pdf.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 5.5, + "bbox_fs": [ + 105, + 122, + 521, + 190 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 197, + 504, + 220 + ], + "lines": [ + { + "bbox": [ + 106, + 195, + 505, + 210 + ], + "spans": [ + { + "bbox": [ + 106, + 195, + 505, + 210 + ], + "score": 1.0, + "content": "Gerard Salton, Anita Wong, and Chung-Shu Yang. A vector space model for automatic indexing.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 117, + 208, + 331, + 220 + ], + "spans": [ + { + "bbox": [ + 117, + 208, + 331, + 220 + ], + "score": 1.0, + "content": "Communications of the ACM, 18(11):613–620, 1975.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9.5, + "bbox_fs": [ + 106, + 195, + 505, + 220 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 226, + 505, + 249 + ], + "lines": [ + { + "bbox": [ + 105, + 226, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 505, + 240 + ], + "score": 1.0, + "content": "Sandeep Subramanian and Soumen Chakrabarti. New embedded representations and evaluation", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 115, + 238, + 365, + 250 + ], + "spans": [ + { + "bbox": [ + 115, + 238, + 365, + 250 + ], + "score": 1.0, + "content": "protocols for inferring transitive relations. SIGIR 2018, 2018.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 226, + 505, + 250 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 257, + 505, + 291 + ], + "lines": [ + { + "bbox": [ + 105, + 257, + 505, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 505, + 270 + ], + "score": 1.0, + "content": "Theo Trouillon, Johannes Welbl, Sebastian Riedel, ´ Eric Gaussier, and Guillaume Bouchard. Com- ´", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 115, + 268, + 506, + 282 + ], + "spans": [ + { + "bbox": [ + 115, + 268, + 506, + 282 + ], + "score": 1.0, + "content": "plex embeddings for simple link prediction. In International Conference on Machine Learning,", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 115, + 280, + 207, + 291 + ], + "spans": [ + { + "bbox": [ + 115, + 280, + 207, + 291 + ], + "score": 1.0, + "content": "pp. 2071–2080, 2016.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 14, + "bbox_fs": [ + 105, + 257, + 506, + 291 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 298, + 504, + 321 + ], + "lines": [ + { + "bbox": [ + 105, + 297, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 105, + 297, + 505, + 312 + ], + "score": 1.0, + "content": "Ivan Vendrov, Ryan Kiros, Sanja Fidler, and Raquel Urtasun. Order-embeddings of images and", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 115, + 309, + 221, + 321 + ], + "spans": [ + { + "bbox": [ + 115, + 309, + 221, + 321 + ], + "score": 1.0, + "content": "language. In ICLR, 2016.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 105, + 297, + 505, + 321 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 328, + 503, + 341 + ], + "lines": [ + { + "bbox": [ + 105, + 328, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 505, + 342 + ], + "score": 1.0, + "content": "Luke Vilnis and Andrew McCallum. Word representations via gaussian embedding. In ICLR, 2015.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 328, + 505, + 342 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 347, + 504, + 370 + ], + "lines": [ + { + "bbox": [ + 105, + 347, + 505, + 360 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 505, + 360 + ], + "score": 1.0, + "content": "Luke Vilnis, Xiang Li, Shikhar Murty, and Andrew McCallum. Probabilistic embedding of knowl-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 115, + 358, + 505, + 372 + ], + "spans": [ + { + "bbox": [ + 115, + 358, + 505, + 372 + ], + "score": 1.0, + "content": "edge graphs with box lattice measures. In ACL. Association for Computational Linguistics, 2018.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 105, + 347, + 505, + 372 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 377, + 504, + 400 + ], + "lines": [ + { + "bbox": [ + 105, + 375, + 506, + 392 + ], + "spans": [ + { + "bbox": [ + 105, + 375, + 506, + 392 + ], + "score": 1.0, + "content": "Adriaan C. Zaanen. Introduction to Operator Theory in Riesz Spaces. Springer Berlin Heidelberg,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 117, + 389, + 237, + 400 + ], + "spans": [ + { + "bbox": [ + 117, + 389, + 237, + 400 + ], + "score": 1.0, + "content": "1997. ISBN 9783642644870.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 375, + 506, + 400 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 235, + 79, + 376, + 95 + ], + "lines": [ + { + "bbox": [ + 233, + 78, + 378, + 96 + ], + "spans": [ + { + "bbox": [ + 233, + 78, + 378, + 96 + ], + "score": 1.0, + "content": "Supplementary Material", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 106, + 109, + 345, + 123 + ], + "lines": [ + { + "bbox": [ + 106, + 109, + 346, + 125 + ], + "spans": [ + { + "bbox": [ + 106, + 109, + 346, + 125 + ], + "score": 1.0, + "content": "A PROOF OF GAUSSIAN OVERLAP FORMULA", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 104, + 134, + 504, + 147 + ], + "lines": [ + { + "bbox": [ + 105, + 132, + 501, + 149 + ], + "spans": [ + { + "bbox": [ + 105, + 132, + 287, + 149 + ], + "score": 1.0, + "content": "We wish to evaluate, for two lattice elements", + "type": "text" + }, + { + "bbox": [ + 287, + 137, + 294, + 145 + ], + "score": 0.68, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 132, + 312, + 149 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 312, + 137, + 320, + 146 + ], + "score": 0.71, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 132, + 470, + 149 + ], + "score": 1.0, + "content": ", with associated smoothed indicators", + "type": "text" + }, + { + "bbox": [ + 470, + 135, + 477, + 146 + ], + "score": 0.86, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 132, + 495, + 149 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 495, + 137, + 501, + 146 + ], + "score": 0.76, + "content": "g", + "type": "inline_equation" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "interline_equation", + "bbox": [ + 124, + 151, + 486, + 207 + ], + "lines": [ + { + "bbox": [ + 124, + 151, + 486, + 207 + ], + "spans": [ + { + "bbox": [ + 124, + 151, + 486, + 207 + ], + "score": 0.94, + "content": "\\begin{array} { l } { f ( x ; a , b , \\sigma ^ { 2 } ) = \\mathbb { 1 } _ { [ a , b ] } ( x ) * \\phi ( x ; \\sigma ^ { 2 } ) = \\displaystyle \\int _ { \\mathbb { R } } \\mathbb { 1 } _ { [ a , b ] } ( z ) \\phi ( x - z ; \\sigma ^ { 2 } ) d z = \\displaystyle \\int _ { a } ^ { b } \\phi ( x - z ; \\sigma ^ { 2 } ) d z } \\\\ { p _ { \\phi } ( \\mathbf x \\wedge \\mathbf y ) = \\displaystyle \\int _ { \\mathbb { R } } f ( x ; a , b , \\sigma _ { 1 } ^ { 2 } ) g ( x ; c , d , \\sigma _ { 2 } ^ { 2 } ) d x } \\end{array}", + "type": "interline_equation", + "image_path": "9939d5026840e9d3e9f00811d4525330523b09dc954f3760c021d8571049199c.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 124, + 151, + 486, + 169.66666666666666 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 124, + 169.66666666666666, + 486, + 188.33333333333331 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 124, + 188.33333333333331, + 486, + 206.99999999999997 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 209, + 505, + 232 + ], + "lines": [ + { + "bbox": [ + 106, + 209, + 505, + 223 + ], + "spans": [ + { + "bbox": [ + 106, + 209, + 505, + 223 + ], + "score": 1.0, + "content": "Since the Gaussian kernel is normalized to have total integral equal to 1, so as not to change the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 220, + 307, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 220, + 307, + 233 + ], + "score": 1.0, + "content": "overall areas of the boxes, the concrete formula is", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5 + }, + { + "type": "interline_equation", + "bbox": [ + 257, + 236, + 353, + 262 + ], + "lines": [ + { + "bbox": [ + 257, + 236, + 353, + 262 + ], + "spans": [ + { + "bbox": [ + 257, + 236, + 353, + 262 + ], + "score": 0.93, + "content": "\\phi ( z ; \\sigma ^ { 2 } ) = { \\frac { 1 } { \\sigma { \\sqrt { 2 \\pi } } } } e ^ { { \\frac { - z ^ { 2 } } { 2 \\sigma ^ { 2 } } } }", + "type": "interline_equation", + "image_path": "54334b640accaf278981391caf44001ed9fddae84c10a477cda430b2b32f0492.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 257, + 236, + 353, + 262 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 266, + 505, + 300 + ], + "lines": [ + { + "bbox": [ + 106, + 266, + 505, + 279 + ], + "spans": [ + { + "bbox": [ + 106, + 266, + 224, + 279 + ], + "score": 1.0, + "content": "Since the antiderivative of", + "type": "text" + }, + { + "bbox": [ + 224, + 267, + 232, + 278 + ], + "score": 0.84, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 266, + 505, + 279 + ], + "score": 1.0, + "content": "is the normal CDF, this may be recognized as the difference", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 277, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 107, + 277, + 208, + 289 + ], + "score": 0.91, + "content": "\\Phi ( x ; a , \\sigma ^ { 2 } ) - \\Phi ( x ; b , \\sigma ^ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 277, + 505, + 289 + ], + "score": 1.0, + "content": ", but this does not allow us to easily evaluate the integral of interest, which", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 288, + 312, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 312, + 300 + ], + "score": 1.0, + "content": "is the integral of the product of two such functions.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 103, + 304, + 465, + 317 + ], + "lines": [ + { + "bbox": [ + 105, + 304, + 465, + 319 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 465, + 319 + ], + "score": 1.0, + "content": "To evaluate equation 8, recall the identity (Jebara et al., 2004; Vilnis & McCallum, 2015)", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "interline_equation", + "bbox": [ + 186, + 321, + 424, + 348 + ], + "lines": [ + { + "bbox": [ + 186, + 321, + 424, + 348 + ], + "spans": [ + { + "bbox": [ + 186, + 321, + 424, + 348 + ], + "score": 0.92, + "content": "\\int _ { \\mathbb { R } } \\phi ( x - \\mu _ { 1 } ; \\sigma _ { 1 } ^ { 2 } ) \\phi ( x - \\mu _ { 2 } ; \\sigma _ { 2 } ^ { 2 } ) d x = \\phi ( \\mu _ { 1 } - \\mu _ { 2 } ; \\sigma _ { 1 } ^ { 2 } + \\sigma _ { 2 } ^ { 2 } )", + "type": "interline_equation", + "image_path": "e97734126bb94aad7df0ccb009a3f51d700c688f537e3e9d7bdec98a7b895b9d.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 186, + 321, + 424, + 348 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 352, + 483, + 371 + ], + "lines": [ + { + "bbox": [ + 106, + 352, + 484, + 371 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 189, + 366 + ], + "score": 1.0, + "content": "For convenience, let", + "type": "text" + }, + { + "bbox": [ + 190, + 352, + 246, + 371 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\tau : = \\frac { 1 } { \\sqrt { \\sigma _ { 1 } ^ { 2 } + \\sigma _ { 2 } ^ { 2 } } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 353, + 484, + 365 + ], + "score": 1.0, + "content": "Applying Fubini’s theorem and using equation 9, we have", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "interline_equation", + "bbox": [ + 130, + 375, + 479, + 518 + ], + "lines": [ + { + "bbox": [ + 130, + 375, + 479, + 518 + ], + "spans": [ + { + "bbox": [ + 130, + 375, + 479, + 518 + ], + "score": 0.95, + "content": "\\begin{array} { l } { \\displaystyle p _ { \\phi } ( \\mathbf { x } \\cdot \\mathbf { y } ) = \\int _ { \\mathbb { R } } \\int _ { a } ^ { b } \\phi ( x - y ; \\sigma _ { 1 } ^ { 2 } ) d y \\int _ { c } ^ { d } \\phi ( x - z ; \\sigma _ { 2 } ^ { 2 } ) d z d x } \\\\ { \\displaystyle = \\int _ { c } ^ { d } \\int _ { a } ^ { b } \\phi ( y - z ; \\tau ^ { - 2 } ) d y d z } \\\\ { \\displaystyle = \\int _ { c } ^ { d } \\int _ { a } ^ { b } \\Phi ^ { \\prime } ( \\tau ( y - z ) ) \\tau d y d z } \\\\ { \\displaystyle = \\int _ { c } ^ { d } \\Phi ( \\tau ( b - z ) ) - \\Phi ( \\tau ( a - z ) ) d z } \\\\ { \\displaystyle = \\frac { - 1 } { \\tau } ( m _ { \\Phi } ( \\tau ( b - d ) ) - m _ { \\Phi } ( \\tau ( a - d ) ) - m _ { \\Phi } ( \\tau ( b - c ) ) + m _ { \\Phi } ( \\tau ( a - c ) ) ) } \\end{array}", + "type": "interline_equation", + "image_path": "7777b148c0b64a58829cf16ec162b689127407fff77bf52c4dbfa44a3b80fcd9.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 130, + 375, + 479, + 422.6666666666667 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 130, + 422.6666666666667, + 479, + 470.33333333333337 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 130, + 470.33333333333337, + 479, + 518.0 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 525, + 224, + 538 + ], + "lines": [ + { + "bbox": [ + 105, + 524, + 221, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 184, + 540 + ], + "score": 1.0, + "content": "and therefore, with", + "type": "text" + }, + { + "bbox": [ + 185, + 526, + 221, + 536 + ], + "score": 0.91, + "content": "\\sigma = \\tau ^ { - 1 }", + "type": "inline_equation" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "interline_equation", + "bbox": [ + 172, + 541, + 438, + 558 + ], + "lines": [ + { + "bbox": [ + 172, + 541, + 438, + 558 + ], + "spans": [ + { + "bbox": [ + 172, + 541, + 438, + 558 + ], + "score": 0.88, + "content": "\\begin{array} { r } { p _ { \\phi } ( \\mathbf { x } \\wedge \\mathbf { y } ) = \\sigma \\left( m _ { \\Phi } ( \\frac { b - c } { \\sigma } ) + m _ { \\Phi } ( \\frac { a - d } { \\sigma } ) - m _ { \\Phi } ( \\frac { b - d } { \\sigma } ) - m _ { \\Phi } ( \\frac { a - c } { \\sigma } ) \\right) } \\end{array}", + "type": "interline_equation", + "image_path": "2e066e99d90fd858a406ab98f2f908da6dfaa83e3e7929f11743f706f9392d86.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 172, + 541, + 438, + 558 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 561, + 150, + 572 + ], + "lines": [ + { + "bbox": [ + 105, + 560, + 151, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 151, + 573 + ], + "score": 1.0, + "content": "as desired.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "title", + "bbox": [ + 108, + 588, + 285, + 601 + ], + "lines": [ + { + "bbox": [ + 105, + 588, + 287, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 287, + 603 + ], + "score": 1.0, + "content": "B MOVIELENS PSEUDOSPARSITY", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 613, + 505, + 647 + ], + "lines": [ + { + "bbox": [ + 105, + 613, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 505, + 627 + ], + "score": 1.0, + "content": "The MovieLens dataset, while not truly sparse, has a large proportion of small probabilities which", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 624, + 506, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 506, + 637 + ], + "score": 1.0, + "content": "make it especially suitable for optimization by the smoothed model. The rough distribution of", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 636, + 344, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 344, + 648 + ], + "score": 1.0, + "content": "probabilities, in buckets of width 0.1, is shown in Figure 1.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23 + }, + { + "type": "title", + "bbox": [ + 106, + 662, + 346, + 676 + ], + "lines": [ + { + "bbox": [ + 106, + 662, + 347, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 662, + 347, + 677 + ], + "score": 1.0, + "content": "C MOVIELENS INITIALIZATION SENSITIVITY", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "score": 1.0, + "content": "We perform an additional set of experiments to determine the robustness of the smoothed box model", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 699, + 505, + 710 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 710 + ], + "score": 1.0, + "content": "to initialization. While the model is normally initialized randomly so that each box is a product of", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "intervals that almost always overlaps with the other boxes, we would like to determine the models ro-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "score": 1.0, + "content": "bustness to disjoint boxes in a principled way. While we can control initialization, we cannot always", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 26, + 294, + 38 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 765 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 765 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 235, + 79, + 376, + 95 + ], + "lines": [ + { + "bbox": [ + 233, + 78, + 378, + 96 + ], + "spans": [ + { + "bbox": [ + 233, + 78, + 378, + 96 + ], + "score": 1.0, + "content": "Supplementary Material", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 106, + 109, + 345, + 123 + ], + "lines": [ + { + "bbox": [ + 106, + 109, + 346, + 125 + ], + "spans": [ + { + "bbox": [ + 106, + 109, + 346, + 125 + ], + "score": 1.0, + "content": "A PROOF OF GAUSSIAN OVERLAP FORMULA", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 104, + 134, + 504, + 147 + ], + "lines": [ + { + "bbox": [ + 105, + 132, + 501, + 149 + ], + "spans": [ + { + "bbox": [ + 105, + 132, + 287, + 149 + ], + "score": 1.0, + "content": "We wish to evaluate, for two lattice elements", + "type": "text" + }, + { + "bbox": [ + 287, + 137, + 294, + 145 + ], + "score": 0.68, + "content": "\\mathbf { x }", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 132, + 312, + 149 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 312, + 137, + 320, + 146 + ], + "score": 0.71, + "content": "\\mathbf { y }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 132, + 470, + 149 + ], + "score": 1.0, + "content": ", with associated smoothed indicators", + "type": "text" + }, + { + "bbox": [ + 470, + 135, + 477, + 146 + ], + "score": 0.86, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 132, + 495, + 149 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 495, + 137, + 501, + 146 + ], + "score": 0.76, + "content": "g", + "type": "inline_equation" + } + ], + "index": 2 + } + ], + "index": 2, + "bbox_fs": [ + 105, + 132, + 501, + 149 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 124, + 151, + 486, + 207 + ], + "lines": [ + { + "bbox": [ + 124, + 151, + 486, + 207 + ], + "spans": [ + { + "bbox": [ + 124, + 151, + 486, + 207 + ], + "score": 0.94, + "content": "\\begin{array} { l } { f ( x ; a , b , \\sigma ^ { 2 } ) = \\mathbb { 1 } _ { [ a , b ] } ( x ) * \\phi ( x ; \\sigma ^ { 2 } ) = \\displaystyle \\int _ { \\mathbb { R } } \\mathbb { 1 } _ { [ a , b ] } ( z ) \\phi ( x - z ; \\sigma ^ { 2 } ) d z = \\displaystyle \\int _ { a } ^ { b } \\phi ( x - z ; \\sigma ^ { 2 } ) d z } \\\\ { p _ { \\phi } ( \\mathbf x \\wedge \\mathbf y ) = \\displaystyle \\int _ { \\mathbb { R } } f ( x ; a , b , \\sigma _ { 1 } ^ { 2 } ) g ( x ; c , d , \\sigma _ { 2 } ^ { 2 } ) d x } \\end{array}", + "type": "interline_equation", + "image_path": "9939d5026840e9d3e9f00811d4525330523b09dc954f3760c021d8571049199c.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 124, + 151, + 486, + 169.66666666666666 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 124, + 169.66666666666666, + 486, + 188.33333333333331 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 124, + 188.33333333333331, + 486, + 206.99999999999997 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 209, + 505, + 232 + ], + "lines": [ + { + "bbox": [ + 106, + 209, + 505, + 223 + ], + "spans": [ + { + "bbox": [ + 106, + 209, + 505, + 223 + ], + "score": 1.0, + "content": "Since the Gaussian kernel is normalized to have total integral equal to 1, so as not to change the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 220, + 307, + 233 + ], + "spans": [ + { + "bbox": [ + 105, + 220, + 307, + 233 + ], + "score": 1.0, + "content": "overall areas of the boxes, the concrete formula is", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6.5, + "bbox_fs": [ + 105, + 209, + 505, + 233 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 257, + 236, + 353, + 262 + ], + "lines": [ + { + "bbox": [ + 257, + 236, + 353, + 262 + ], + "spans": [ + { + "bbox": [ + 257, + 236, + 353, + 262 + ], + "score": 0.93, + "content": "\\phi ( z ; \\sigma ^ { 2 } ) = { \\frac { 1 } { \\sigma { \\sqrt { 2 \\pi } } } } e ^ { { \\frac { - z ^ { 2 } } { 2 \\sigma ^ { 2 } } } }", + "type": "interline_equation", + "image_path": "54334b640accaf278981391caf44001ed9fddae84c10a477cda430b2b32f0492.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 257, + 236, + 353, + 262 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 266, + 505, + 300 + ], + "lines": [ + { + "bbox": [ + 106, + 266, + 505, + 279 + ], + "spans": [ + { + "bbox": [ + 106, + 266, + 224, + 279 + ], + "score": 1.0, + "content": "Since the antiderivative of", + "type": "text" + }, + { + "bbox": [ + 224, + 267, + 232, + 278 + ], + "score": 0.84, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 266, + 505, + 279 + ], + "score": 1.0, + "content": "is the normal CDF, this may be recognized as the difference", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 277, + 505, + 289 + ], + "spans": [ + { + "bbox": [ + 107, + 277, + 208, + 289 + ], + "score": 0.91, + "content": "\\Phi ( x ; a , \\sigma ^ { 2 } ) - \\Phi ( x ; b , \\sigma ^ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 277, + 505, + 289 + ], + "score": 1.0, + "content": ", but this does not allow us to easily evaluate the integral of interest, which", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 288, + 312, + 300 + ], + "spans": [ + { + "bbox": [ + 105, + 288, + 312, + 300 + ], + "score": 1.0, + "content": "is the integral of the product of two such functions.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 105, + 266, + 505, + 300 + ] + }, + { + "type": "text", + "bbox": [ + 103, + 304, + 465, + 317 + ], + "lines": [ + { + "bbox": [ + 105, + 304, + 465, + 319 + ], + "spans": [ + { + "bbox": [ + 105, + 304, + 465, + 319 + ], + "score": 1.0, + "content": "To evaluate equation 8, recall the identity (Jebara et al., 2004; Vilnis & McCallum, 2015)", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12, + "bbox_fs": [ + 105, + 304, + 465, + 319 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 186, + 321, + 424, + 348 + ], + "lines": [ + { + "bbox": [ + 186, + 321, + 424, + 348 + ], + "spans": [ + { + "bbox": [ + 186, + 321, + 424, + 348 + ], + "score": 0.92, + "content": "\\int _ { \\mathbb { R } } \\phi ( x - \\mu _ { 1 } ; \\sigma _ { 1 } ^ { 2 } ) \\phi ( x - \\mu _ { 2 } ; \\sigma _ { 2 } ^ { 2 } ) d x = \\phi ( \\mu _ { 1 } - \\mu _ { 2 } ; \\sigma _ { 1 } ^ { 2 } + \\sigma _ { 2 } ^ { 2 } )", + "type": "interline_equation", + "image_path": "e97734126bb94aad7df0ccb009a3f51d700c688f537e3e9d7bdec98a7b895b9d.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 186, + 321, + 424, + 348 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 352, + 483, + 371 + ], + "lines": [ + { + "bbox": [ + 106, + 352, + 484, + 371 + ], + "spans": [ + { + "bbox": [ + 106, + 352, + 189, + 366 + ], + "score": 1.0, + "content": "For convenience, let", + "type": "text" + }, + { + "bbox": [ + 190, + 352, + 246, + 371 + ], + "score": 0.93, + "content": "\\begin{array} { r } { \\tau : = \\frac { 1 } { \\sqrt { \\sigma _ { 1 } ^ { 2 } + \\sigma _ { 2 } ^ { 2 } } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 353, + 484, + 365 + ], + "score": 1.0, + "content": "Applying Fubini’s theorem and using equation 9, we have", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14, + "bbox_fs": [ + 106, + 352, + 484, + 371 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 130, + 375, + 479, + 518 + ], + "lines": [ + { + "bbox": [ + 130, + 375, + 479, + 518 + ], + "spans": [ + { + "bbox": [ + 130, + 375, + 479, + 518 + ], + "score": 0.95, + "content": "\\begin{array} { l } { \\displaystyle p _ { \\phi } ( \\mathbf { x } \\cdot \\mathbf { y } ) = \\int _ { \\mathbb { R } } \\int _ { a } ^ { b } \\phi ( x - y ; \\sigma _ { 1 } ^ { 2 } ) d y \\int _ { c } ^ { d } \\phi ( x - z ; \\sigma _ { 2 } ^ { 2 } ) d z d x } \\\\ { \\displaystyle = \\int _ { c } ^ { d } \\int _ { a } ^ { b } \\phi ( y - z ; \\tau ^ { - 2 } ) d y d z } \\\\ { \\displaystyle = \\int _ { c } ^ { d } \\int _ { a } ^ { b } \\Phi ^ { \\prime } ( \\tau ( y - z ) ) \\tau d y d z } \\\\ { \\displaystyle = \\int _ { c } ^ { d } \\Phi ( \\tau ( b - z ) ) - \\Phi ( \\tau ( a - z ) ) d z } \\\\ { \\displaystyle = \\frac { - 1 } { \\tau } ( m _ { \\Phi } ( \\tau ( b - d ) ) - m _ { \\Phi } ( \\tau ( a - d ) ) - m _ { \\Phi } ( \\tau ( b - c ) ) + m _ { \\Phi } ( \\tau ( a - c ) ) ) } \\end{array}", + "type": "interline_equation", + "image_path": "7777b148c0b64a58829cf16ec162b689127407fff77bf52c4dbfa44a3b80fcd9.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 130, + 375, + 479, + 422.6666666666667 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 130, + 422.6666666666667, + 479, + 470.33333333333337 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 130, + 470.33333333333337, + 479, + 518.0 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 525, + 224, + 538 + ], + "lines": [ + { + "bbox": [ + 105, + 524, + 221, + 540 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 184, + 540 + ], + "score": 1.0, + "content": "and therefore, with", + "type": "text" + }, + { + "bbox": [ + 185, + 526, + 221, + 536 + ], + "score": 0.91, + "content": "\\sigma = \\tau ^ { - 1 }", + "type": "inline_equation" + } + ], + "index": 18 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 524, + 221, + 540 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 172, + 541, + 438, + 558 + ], + "lines": [ + { + "bbox": [ + 172, + 541, + 438, + 558 + ], + "spans": [ + { + "bbox": [ + 172, + 541, + 438, + 558 + ], + "score": 0.88, + "content": "\\begin{array} { r } { p _ { \\phi } ( \\mathbf { x } \\wedge \\mathbf { y } ) = \\sigma \\left( m _ { \\Phi } ( \\frac { b - c } { \\sigma } ) + m _ { \\Phi } ( \\frac { a - d } { \\sigma } ) - m _ { \\Phi } ( \\frac { b - d } { \\sigma } ) - m _ { \\Phi } ( \\frac { a - c } { \\sigma } ) \\right) } \\end{array}", + "type": "interline_equation", + "image_path": "2e066e99d90fd858a406ab98f2f908da6dfaa83e3e7929f11743f706f9392d86.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 172, + 541, + 438, + 558 + ], + "spans": [], + "index": 19 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 561, + 150, + 572 + ], + "lines": [ + { + "bbox": [ + 105, + 560, + 151, + 573 + ], + "spans": [ + { + "bbox": [ + 105, + 560, + 151, + 573 + ], + "score": 1.0, + "content": "as desired.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20, + "bbox_fs": [ + 105, + 560, + 151, + 573 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 588, + 285, + 601 + ], + "lines": [ + { + "bbox": [ + 105, + 588, + 287, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 287, + 603 + ], + "score": 1.0, + "content": "B MOVIELENS PSEUDOSPARSITY", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 107, + 613, + 505, + 647 + ], + "lines": [ + { + "bbox": [ + 105, + 613, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 613, + 505, + 627 + ], + "score": 1.0, + "content": "The MovieLens dataset, while not truly sparse, has a large proportion of small probabilities which", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 624, + 506, + 637 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 506, + 637 + ], + "score": 1.0, + "content": "make it especially suitable for optimization by the smoothed model. The rough distribution of", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 636, + 344, + 648 + ], + "spans": [ + { + "bbox": [ + 105, + 636, + 344, + 648 + ], + "score": 1.0, + "content": "probabilities, in buckets of width 0.1, is shown in Figure 1.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23, + "bbox_fs": [ + 105, + 613, + 506, + 648 + ] + }, + { + "type": "title", + "bbox": [ + 106, + 662, + 346, + 676 + ], + "lines": [ + { + "bbox": [ + 106, + 662, + 347, + 677 + ], + "spans": [ + { + "bbox": [ + 106, + 662, + 347, + 677 + ], + "score": 1.0, + "content": "C MOVIELENS INITIALIZATION SENSITIVITY", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 107, + 687, + 505, + 732 + ], + "lines": [ + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "spans": [ + { + "bbox": [ + 106, + 688, + 505, + 699 + ], + "score": 1.0, + "content": "We perform an additional set of experiments to determine the robustness of the smoothed box model", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 699, + 505, + 710 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 710 + ], + "score": 1.0, + "content": "to initialization. While the model is normally initialized randomly so that each box is a product of", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 505, + 722 + ], + "score": 1.0, + "content": "intervals that almost always overlaps with the other boxes, we would like to determine the models ro-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "spans": [ + { + "bbox": [ + 105, + 720, + 505, + 734 + ], + "score": 1.0, + "content": "bustness to disjoint boxes in a principled way. While we can control initialization, we cannot always", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 253, + 505, + 265 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 505, + 265 + ], + "score": 1.0, + "content": "control the intermediate results of optimization, which may drive boxes to be disjoint, a condition", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "score": 1.0, + "content": "from which the original, hard-edged box model may have difficulty recovering. So, parametriz-", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 274, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 274, + 505, + 286 + ], + "score": 1.0, + "content": "ing the initial distribution of boxes with a minimum coordinate and a positive width, we adjust the", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 286, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 286, + 263, + 298 + ], + "score": 1.0, + "content": "width parameter so that approximately", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 263, + 286, + 277, + 296 + ], + "score": 0.84, + "content": "0 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 278, + 286, + 281, + 298 + ], + "score": 1.0, + "content": ",", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 281, + 286, + 301, + 296 + ], + "score": 0.84, + "content": "20 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 301, + 286, + 304, + 298 + ], + "score": 1.0, + "content": ",", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 304, + 286, + 324, + 296 + ], + "score": 0.86, + "content": "50 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 325, + 286, + 345, + 298 + ], + "score": 1.0, + "content": ", and", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 345, + 286, + 370, + 296 + ], + "score": 0.88, + "content": "100 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 370, + 286, + 505, + 298 + ], + "score": 1.0, + "content": "of boxes are disjoint at initializa-", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 297, + 504, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 297, + 504, + 308 + ], + "score": 1.0, + "content": "tion before learning on the MovieLens dataset as usual. These results are presented in table 8. The", + "type": "text", + "cross_page": true + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 307, + 506, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 307, + 506, + 320 + ], + "score": 1.0, + "content": "smoothed model does not seem to suffer at all from disjoint initialization, while the performance of", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 317, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 505, + 331 + ], + "score": 1.0, + "content": "the original box model degrades significantly. From this we can speculate that part of the strength", + "type": "text", + "cross_page": true + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 329, + 444, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 444, + 342 + ], + "score": 1.0, + "content": "of the smoothed box model is its ability to smoothly optimize in the disjoint regime.", + "type": "text", + "cross_page": true + } + ], + "index": 16 + } + ], + "index": 27.5, + "bbox_fs": [ + 105, + 688, + 505, + 734 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 210, + 86, + 401, + 200 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 210, + 86, + 401, + 200 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 210, + 86, + 401, + 200 + ], + "spans": [ + { + "bbox": [ + 210, + 86, + 401, + 200 + ], + "score": 0.954, + "type": "image", + "image_path": "262dd36934bbf5f0c1eafe924b8ad5d60a4fadb6282c801a3dd744a0a5c70830.jpg" + } + ] + } + ], + "index": 3.5, + "virtual_lines": [ + { + "bbox": [ + 210, + 86, + 401, + 100.25 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 210, + 100.25, + 401, + 114.5 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 210, + 114.5, + 401, + 128.75 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 210, + 128.75, + 401, + 143.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 210, + 143.0, + 401, + 157.25 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 210, + 157.25, + 401, + 171.5 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 210, + 171.5, + 401, + 185.75 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 210, + 185.75, + 401, + 200.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 182, + 213, + 428, + 225 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 182, + 213, + 428, + 226 + ], + "spans": [ + { + "bbox": [ + 182, + 213, + 428, + 226 + ], + "score": 1.0, + "content": "Figure 1: Distribution of probabilities in MovieLens Dataset.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + } + ], + "index": 5.75 + }, + { + "type": "text", + "bbox": [ + 106, + 252, + 505, + 341 + ], + "lines": [ + { + "bbox": [ + 106, + 253, + 505, + 265 + ], + "spans": [ + { + "bbox": [ + 106, + 253, + 505, + 265 + ], + "score": 1.0, + "content": "control the intermediate results of optimization, which may drive boxes to be disjoint, a condition", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "spans": [ + { + "bbox": [ + 105, + 263, + 505, + 276 + ], + "score": 1.0, + "content": "from which the original, hard-edged box model may have difficulty recovering. So, parametriz-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 274, + 505, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 274, + 505, + 286 + ], + "score": 1.0, + "content": "ing the initial distribution of boxes with a minimum coordinate and a positive width, we adjust the", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 286, + 505, + 298 + ], + "spans": [ + { + "bbox": [ + 106, + 286, + 263, + 298 + ], + "score": 1.0, + "content": "width parameter so that approximately", + "type": "text" + }, + { + "bbox": [ + 263, + 286, + 277, + 296 + ], + "score": 0.84, + "content": "0 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 286, + 281, + 298 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 281, + 286, + 301, + 296 + ], + "score": 0.84, + "content": "20 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 301, + 286, + 304, + 298 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 304, + 286, + 324, + 296 + ], + "score": 0.86, + "content": "50 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 286, + 345, + 298 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 345, + 286, + 370, + 296 + ], + "score": 0.88, + "content": "100 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 286, + 505, + 298 + ], + "score": 1.0, + "content": "of boxes are disjoint at initializa-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 297, + 504, + 308 + ], + "spans": [ + { + "bbox": [ + 106, + 297, + 504, + 308 + ], + "score": 1.0, + "content": "tion before learning on the MovieLens dataset as usual. These results are presented in table 8. The", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 307, + 506, + 320 + ], + "spans": [ + { + "bbox": [ + 106, + 307, + 506, + 320 + ], + "score": 1.0, + "content": "smoothed model does not seem to suffer at all from disjoint initialization, while the performance of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 317, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 505, + 331 + ], + "score": 1.0, + "content": "the original box model degrades significantly. From this we can speculate that part of the strength", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 329, + 444, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 444, + 342 + ], + "score": 1.0, + "content": "of the smoothed box model is its ability to smoothly optimize in the disjoint regime.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 12.5 + }, + { + "type": "table", + "bbox": [ + 135, + 357, + 477, + 428 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 135, + 357, + 477, + 428 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 135, + 357, + 477, + 428 + ], + "spans": [ + { + "bbox": [ + 135, + 357, + 477, + 428 + ], + "score": 0.978, + "html": "
Approx. % DisjointKLPearsonSpearman
BoxSmoothBoxSmoothBoxSmooth
0%0.01470.01380.87750.89850.87680.8977
20%0.01720.01410.86680.89170.86080.8898
50%0.01820.01410.86130.89080.85510.8910
100%0.03460.01420.84010.89210.81670.8947
", + "type": "table", + "image_path": "7da55663fb997de6a414f9e9494f47e39188b56c7b8d36c675195f21e96278c9.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 135, + 357, + 477, + 380.6666666666667 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 135, + 380.6666666666667, + 477, + 404.33333333333337 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 135, + 404.33333333333337, + 477, + 428.00000000000006 + ], + "spans": [], + "index": 19 + } + ] + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 435, + 506, + 458 + ], + "lines": [ + { + "bbox": [ + 105, + 434, + 506, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 506, + 448 + ], + "score": 1.0, + "content": "Table 8: Performance of the original box model and smoothed box model on MovieLens, as a", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 446, + 360, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 360, + 459 + ], + "score": 1.0, + "content": "function of different degrees of disjointness upon initialization.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5 + }, + { + "type": "title", + "bbox": [ + 107, + 494, + 241, + 507 + ], + "lines": [ + { + "bbox": [ + 105, + 492, + 242, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 242, + 509 + ], + "score": 1.0, + "content": "D MODEL PARAMETERS", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 523, + 505, + 556 + ], + "lines": [ + { + "bbox": [ + 105, + 522, + 505, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 505, + 536 + ], + "score": 1.0, + "content": "We give a brief overview of our methodology and hyperparameter selection methods for each ex-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 534, + 506, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 534, + 506, + 546 + ], + "score": 1.0, + "content": "periment. Detailed hyperparameter settings and code to reproduce experiments can be found at", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 544, + 429, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 544, + 429, + 558 + ], + "score": 1.0, + "content": "https://github.com/Lorraine333/smoothed_box_embedding.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24 + }, + { + "type": "title", + "bbox": [ + 107, + 576, + 241, + 587 + ], + "lines": [ + { + "bbox": [ + 106, + 576, + 242, + 589 + ], + "spans": [ + { + "bbox": [ + 106, + 576, + 242, + 589 + ], + "score": 1.0, + "content": "D.1 WORDNET PARAMETERS", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 600, + 505, + 644 + ], + "lines": [ + { + "bbox": [ + 105, + 599, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 599, + 505, + 613 + ], + "score": 1.0, + "content": "For the WordNet experiments, the model is evaluated every epoch on the development set for a large", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 611, + 505, + 622 + ], + "spans": [ + { + "bbox": [ + 106, + 611, + 505, + 622 + ], + "score": 1.0, + "content": "fixed number of epochs, and the best development model is used to score the test set. Baseline", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 621, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 505, + 635 + ], + "score": 1.0, + "content": "models are trained using the parameters of Vilnis et al. (2018), with the smoothed model using", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 633, + 320, + 645 + ], + "spans": [ + { + "bbox": [ + 106, + 633, + 320, + 645 + ], + "score": 1.0, + "content": "hyperparameters determined on the development set.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 28.5 + }, + { + "type": "title", + "bbox": [ + 109, + 664, + 302, + 675 + ], + "lines": [ + { + "bbox": [ + 107, + 664, + 303, + 677 + ], + "spans": [ + { + "bbox": [ + 107, + 664, + 303, + 677 + ], + "score": 1.0, + "content": "D.2 IMBALANCED WORDNET PARAMETERS", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 108, + 687, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "We follow the same routine as the WordNet experiments section to select best parameters. For the", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "12 experiments we conducted in this section, negative examples are generated randomly based on", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "the ratio for each batch of positive examples. We do a parameter sweep for all models then choose", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 721, + 300, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 300, + 732 + ], + "score": 1.0, + "content": "the best result for each model as our final result.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33.5 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 26, + 293, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 301, + 751, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 298, + 750, + 312, + 763 + ], + "spans": [ + { + "bbox": [ + 298, + 750, + 312, + 763 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 13, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 210, + 86, + 401, + 200 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 210, + 86, + 401, + 200 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 210, + 86, + 401, + 200 + ], + "spans": [ + { + "bbox": [ + 210, + 86, + 401, + 200 + ], + "score": 0.954, + "type": "image", + "image_path": "262dd36934bbf5f0c1eafe924b8ad5d60a4fadb6282c801a3dd744a0a5c70830.jpg" + } + ] + } + ], + "index": 3.5, + "virtual_lines": [ + { + "bbox": [ + 210, + 86, + 401, + 100.25 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 210, + 100.25, + 401, + 114.5 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 210, + 114.5, + 401, + 128.75 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 210, + 128.75, + 401, + 143.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 210, + 143.0, + 401, + 157.25 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 210, + 157.25, + 401, + 171.5 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 210, + 171.5, + 401, + 185.75 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 210, + 185.75, + 401, + 200.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 182, + 213, + 428, + 225 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 182, + 213, + 428, + 226 + ], + "spans": [ + { + "bbox": [ + 182, + 213, + 428, + 226 + ], + "score": 1.0, + "content": "Figure 1: Distribution of probabilities in MovieLens Dataset.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + } + ], + "index": 5.75 + }, + { + "type": "text", + "bbox": [ + 106, + 252, + 505, + 341 + ], + "lines": [], + "index": 12.5, + "bbox_fs": [ + 105, + 253, + 506, + 342 + ], + "lines_deleted": true + }, + { + "type": "table", + "bbox": [ + 135, + 357, + 477, + 428 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 135, + 357, + 477, + 428 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 135, + 357, + 477, + 428 + ], + "spans": [ + { + "bbox": [ + 135, + 357, + 477, + 428 + ], + "score": 0.978, + "html": "
Approx. % DisjointKLPearsonSpearman
BoxSmoothBoxSmoothBoxSmooth
0%0.01470.01380.87750.89850.87680.8977
20%0.01720.01410.86680.89170.86080.8898
50%0.01820.01410.86130.89080.85510.8910
100%0.03460.01420.84010.89210.81670.8947
", + "type": "table", + "image_path": "7da55663fb997de6a414f9e9494f47e39188b56c7b8d36c675195f21e96278c9.jpg" + } + ] + } + ], + "index": 18, + "virtual_lines": [ + { + "bbox": [ + 135, + 357, + 477, + 380.6666666666667 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 135, + 380.6666666666667, + 477, + 404.33333333333337 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 135, + 404.33333333333337, + 477, + 428.00000000000006 + ], + "spans": [], + "index": 19 + } + ] + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 435, + 506, + 458 + ], + "lines": [ + { + "bbox": [ + 105, + 434, + 506, + 448 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 506, + 448 + ], + "score": 1.0, + "content": "Table 8: Performance of the original box model and smoothed box model on MovieLens, as a", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 446, + 360, + 459 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 360, + 459 + ], + "score": 1.0, + "content": "function of different degrees of disjointness upon initialization.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5, + "bbox_fs": [ + 105, + 434, + 506, + 459 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 494, + 241, + 507 + ], + "lines": [ + { + "bbox": [ + 105, + 492, + 242, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 492, + 242, + 509 + ], + "score": 1.0, + "content": "D MODEL PARAMETERS", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "text", + "bbox": [ + 107, + 523, + 505, + 556 + ], + "lines": [ + { + "bbox": [ + 105, + 522, + 505, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 505, + 536 + ], + "score": 1.0, + "content": "We give a brief overview of our methodology and hyperparameter selection methods for each ex-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 534, + 506, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 534, + 506, + 546 + ], + "score": 1.0, + "content": "periment. Detailed hyperparameter settings and code to reproduce experiments can be found at", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 544, + 429, + 558 + ], + "spans": [ + { + "bbox": [ + 105, + 544, + 429, + 558 + ], + "score": 1.0, + "content": "https://github.com/Lorraine333/smoothed_box_embedding.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 522, + 506, + 558 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 576, + 241, + 587 + ], + "lines": [ + { + "bbox": [ + 106, + 576, + 242, + 589 + ], + "spans": [ + { + "bbox": [ + 106, + 576, + 242, + 589 + ], + "score": 1.0, + "content": "D.1 WORDNET PARAMETERS", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 600, + 505, + 644 + ], + "lines": [ + { + "bbox": [ + 105, + 599, + 505, + 613 + ], + "spans": [ + { + "bbox": [ + 105, + 599, + 505, + 613 + ], + "score": 1.0, + "content": "For the WordNet experiments, the model is evaluated every epoch on the development set for a large", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 611, + 505, + 622 + ], + "spans": [ + { + "bbox": [ + 106, + 611, + 505, + 622 + ], + "score": 1.0, + "content": "fixed number of epochs, and the best development model is used to score the test set. Baseline", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 621, + 505, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 505, + 635 + ], + "score": 1.0, + "content": "models are trained using the parameters of Vilnis et al. (2018), with the smoothed model using", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 633, + 320, + 645 + ], + "spans": [ + { + "bbox": [ + 106, + 633, + 320, + 645 + ], + "score": 1.0, + "content": "hyperparameters determined on the development set.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 28.5, + "bbox_fs": [ + 105, + 599, + 505, + 645 + ] + }, + { + "type": "title", + "bbox": [ + 109, + 664, + 302, + 675 + ], + "lines": [ + { + "bbox": [ + 107, + 664, + 303, + 677 + ], + "spans": [ + { + "bbox": [ + 107, + 664, + 303, + 677 + ], + "score": 1.0, + "content": "D.2 IMBALANCED WORDNET PARAMETERS", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 108, + 687, + 505, + 731 + ], + "lines": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 505, + 700 + ], + "score": 1.0, + "content": "We follow the same routine as the WordNet experiments section to select best parameters. For the", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 505, + 711 + ], + "score": 1.0, + "content": "12 experiments we conducted in this section, negative examples are generated randomly based on", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "spans": [ + { + "bbox": [ + 105, + 709, + 505, + 722 + ], + "score": 1.0, + "content": "the ratio for each batch of positive examples. We do a parameter sweep for all models then choose", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 721, + 300, + 732 + ], + "spans": [ + { + "bbox": [ + 106, + 721, + 300, + 732 + ], + "score": 1.0, + "content": "the best result for each model as our final result.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 687, + 505, + 732 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 227, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 228, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 228, + 94 + ], + "score": 1.0, + "content": "D.3 FLICKR PARAMETERS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 102, + 505, + 169 + ], + "lines": [ + { + "bbox": [ + 105, + 102, + 506, + 115 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 506, + 115 + ], + "score": 1.0, + "content": "The experimental setup uses the same architecture as Vilnis et al. (2018) and Lai & Hockenmaier", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 114, + 505, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 505, + 127 + ], + "score": 1.0, + "content": "(2017), a single-layer LSTM that reads captions and produces a box embedding parameterized by", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 125, + 505, + 137 + ], + "spans": [ + { + "bbox": [ + 105, + 125, + 505, + 137 + ], + "score": 1.0, + "content": "min and delta. Embeddings are produced by feedforward networks on the output of the LSTM.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 135, + 506, + 149 + ], + "spans": [ + { + "bbox": [ + 105, + 135, + 506, + 149 + ], + "score": 1.0, + "content": "The model is trained for a large fixed number of epochs, and tested on the development data at", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 146, + 506, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 146, + 506, + 160 + ], + "score": 1.0, + "content": "each epoch. The best development model is used to report test set score. Hyperparameters were", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 158, + 250, + 170 + ], + "spans": [ + { + "bbox": [ + 106, + 158, + 250, + 170 + ], + "score": 1.0, + "content": "determined on the development set.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3.5 + }, + { + "type": "title", + "bbox": [ + 108, + 182, + 247, + 194 + ], + "lines": [ + { + "bbox": [ + 106, + 182, + 248, + 195 + ], + "spans": [ + { + "bbox": [ + 106, + 182, + 248, + 195 + ], + "score": 1.0, + "content": "D.4 MOVIELENS PARAMETERS", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 107, + 203, + 505, + 236 + ], + "lines": [ + { + "bbox": [ + 106, + 203, + 505, + 216 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 505, + 216 + ], + "score": 1.0, + "content": "For all MovieLens experiments, the model is evaluated every 50 steps on the development set, and", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 214, + 506, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 506, + 227 + ], + "score": 1.0, + "content": "optimization is stopped if the best development set score fails to improve after 200 steps. The best", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 226, + 298, + 237 + ], + "spans": [ + { + "bbox": [ + 106, + 226, + 298, + 237 + ], + "score": 1.0, + "content": "development model is used to score the test set.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 108, + 27, + 293, + 37 + ], + "lines": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "spans": [ + { + "bbox": [ + 106, + 25, + 294, + 38 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 300, + 752, + 311, + 760 + ], + "lines": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "spans": [ + { + "bbox": [ + 299, + 750, + 313, + 764 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 108, + 82, + 227, + 93 + ], + "lines": [ + { + "bbox": [ + 106, + 82, + 228, + 94 + ], + "spans": [ + { + "bbox": [ + 106, + 82, + 228, + 94 + ], + "score": 1.0, + "content": "D.3 FLICKR PARAMETERS", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 107, + 102, + 505, + 169 + ], + "lines": [ + { + "bbox": [ + 105, + 102, + 506, + 115 + ], + "spans": [ + { + "bbox": [ + 105, + 102, + 506, + 115 + ], + "score": 1.0, + "content": "The experimental setup uses the same architecture as Vilnis et al. (2018) and Lai & Hockenmaier", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 114, + 505, + 127 + ], + "spans": [ + { + "bbox": [ + 105, + 114, + 505, + 127 + ], + "score": 1.0, + "content": "(2017), a single-layer LSTM that reads captions and produces a box embedding parameterized by", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 125, + 505, + 137 + ], + "spans": [ + { + "bbox": [ + 105, + 125, + 505, + 137 + ], + "score": 1.0, + "content": "min and delta. Embeddings are produced by feedforward networks on the output of the LSTM.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 135, + 506, + 149 + ], + "spans": [ + { + "bbox": [ + 105, + 135, + 506, + 149 + ], + "score": 1.0, + "content": "The model is trained for a large fixed number of epochs, and tested on the development data at", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 146, + 506, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 146, + 506, + 160 + ], + "score": 1.0, + "content": "each epoch. The best development model is used to report test set score. Hyperparameters were", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 106, + 158, + 250, + 170 + ], + "spans": [ + { + "bbox": [ + 106, + 158, + 250, + 170 + ], + "score": 1.0, + "content": "determined on the development set.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 3.5, + "bbox_fs": [ + 105, + 102, + 506, + 170 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 182, + 247, + 194 + ], + "lines": [ + { + "bbox": [ + 106, + 182, + 248, + 195 + ], + "spans": [ + { + "bbox": [ + 106, + 182, + 248, + 195 + ], + "score": 1.0, + "content": "D.4 MOVIELENS PARAMETERS", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 107, + 203, + 505, + 236 + ], + "lines": [ + { + "bbox": [ + 106, + 203, + 505, + 216 + ], + "spans": [ + { + "bbox": [ + 106, + 203, + 505, + 216 + ], + "score": 1.0, + "content": "For all MovieLens experiments, the model is evaluated every 50 steps on the development set, and", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 214, + 506, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 214, + 506, + 227 + ], + "score": 1.0, + "content": "optimization is stopped if the best development set score fails to improve after 200 steps. The best", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 226, + 298, + 237 + ], + "spans": [ + { + "bbox": [ + 106, + 226, + 298, + 237 + ], + "score": 1.0, + "content": "development model is used to score the test set.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 203, + 506, + 237 + ] + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/train/HJg2b0VYDr/images/132f67ab277aed9b5761873a2bc40f7486ba93c2fb1139d4bfa053c786577b6c.jpg b/parse/train/HJg2b0VYDr/images/132f67ab277aed9b5761873a2bc40f7486ba93c2fb1139d4bfa053c786577b6c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d2d5fe4d116718f66d6ec3d01673dcc2fe335240 --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/132f67ab277aed9b5761873a2bc40f7486ba93c2fb1139d4bfa053c786577b6c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:45a81898374937aac5c7828e0a7eda21213fab82322872cf6a56735e78056a5f +size 63157 diff --git a/parse/train/HJg2b0VYDr/images/135e9546426565cde5bd7a6ab04f2046397f08e0e5d1709924e3918d3884456f.jpg b/parse/train/HJg2b0VYDr/images/135e9546426565cde5bd7a6ab04f2046397f08e0e5d1709924e3918d3884456f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f00c851ec2b2c94982c15f183b287d74c05f8554 --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/135e9546426565cde5bd7a6ab04f2046397f08e0e5d1709924e3918d3884456f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec65e1e07ec752d8ac54973f046e9aead5cb0e0a4053e0a6cd7c04c3423e9fcd +size 380645 diff --git a/parse/train/HJg2b0VYDr/images/138630cdb4e9c9c3582fa3078e08c53f338f71e3d3693865e5b5f85d7acd411f.jpg b/parse/train/HJg2b0VYDr/images/138630cdb4e9c9c3582fa3078e08c53f338f71e3d3693865e5b5f85d7acd411f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b8f78c73233750df9d9b05d0b6dca932928194a1 --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/138630cdb4e9c9c3582fa3078e08c53f338f71e3d3693865e5b5f85d7acd411f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:afc4207ba00052b91e6d0676974150b2c9fb01af1a93249d30b6910b9182a048 +size 450608 diff --git a/parse/train/HJg2b0VYDr/images/173edc63bfe701570c058f678fa25430e14ff9cd87060ba20aef9a32024be975.jpg b/parse/train/HJg2b0VYDr/images/173edc63bfe701570c058f678fa25430e14ff9cd87060ba20aef9a32024be975.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d0d88c55a5ae397157f3ea4176250a24bef5db7c --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/173edc63bfe701570c058f678fa25430e14ff9cd87060ba20aef9a32024be975.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7deb130fd8745707c0339f32d21f82d5f7807eb751c1570306d8712cfd2a1084 +size 34845 diff --git a/parse/train/HJg2b0VYDr/images/17b58d06b1b9a06faffd136bdf6bb9df925ce7e3e38ecd46a3ba390ea6ae0586.jpg b/parse/train/HJg2b0VYDr/images/17b58d06b1b9a06faffd136bdf6bb9df925ce7e3e38ecd46a3ba390ea6ae0586.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b5f290de47c8982010db2c93eb13e4564e3b083d --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/17b58d06b1b9a06faffd136bdf6bb9df925ce7e3e38ecd46a3ba390ea6ae0586.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d93189d4f257bd39bd0286d0bd988132521c3d6813932ff63ede92a2cf7158e6 +size 53031 diff --git a/parse/train/HJg2b0VYDr/images/18d9cbd79fa9a9e1786c42a0757521a06fc8b6b1783dabbd7f6e54ae2b6bd016.jpg b/parse/train/HJg2b0VYDr/images/18d9cbd79fa9a9e1786c42a0757521a06fc8b6b1783dabbd7f6e54ae2b6bd016.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3efd25b753ab3272a89fc447983244e08e11c7f0 --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/18d9cbd79fa9a9e1786c42a0757521a06fc8b6b1783dabbd7f6e54ae2b6bd016.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3bd6df66fbee2ec8035ddabb058e3bfaccc6345e7a9a0c25792e1579f08866c5 +size 167914 diff --git a/parse/train/HJg2b0VYDr/images/1c301851468f5d4bdc70480ed16bff416cb7242b5705a6efa5d1899e2c07d5c5.jpg b/parse/train/HJg2b0VYDr/images/1c301851468f5d4bdc70480ed16bff416cb7242b5705a6efa5d1899e2c07d5c5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0c2e620ba8eac2defd7ed88cd991525405d040ce --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/1c301851468f5d4bdc70480ed16bff416cb7242b5705a6efa5d1899e2c07d5c5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:24ed4a3f0b8516ae1ddc485232cce056b129872ec39941705a7d5aab93a80cf0 +size 22449 diff --git a/parse/train/HJg2b0VYDr/images/315e58532f8a07d413f1dc5a9d0ff5c21d6ec5e0aca5d8401ab7d7ace2f3d6a3.jpg b/parse/train/HJg2b0VYDr/images/315e58532f8a07d413f1dc5a9d0ff5c21d6ec5e0aca5d8401ab7d7ace2f3d6a3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8b2dd8585d5c13e9091f5aee8a1da179320a3a1b --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/315e58532f8a07d413f1dc5a9d0ff5c21d6ec5e0aca5d8401ab7d7ace2f3d6a3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3b2f2ddc0931fd017dfb7b14c463e1eb5ec305acc22b75808d0596ff51799e4b +size 24976 diff --git a/parse/train/HJg2b0VYDr/images/3acbaf560ec46f3f87160eb3d7438049069a4886e072ff9816174357c3a21310.jpg b/parse/train/HJg2b0VYDr/images/3acbaf560ec46f3f87160eb3d7438049069a4886e072ff9816174357c3a21310.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4b1674873f8de9c98ee0dee76e732789594dd165 --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/3acbaf560ec46f3f87160eb3d7438049069a4886e072ff9816174357c3a21310.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:04685f1ed9aa364b9cebf537b8a2d2a7a57f973046dcb548c3e4637a880b3c40 +size 23444 diff --git a/parse/train/HJg2b0VYDr/images/42536c76ac3983ea601745efa48ccc0a3d08ee32fd495e411fa23a8b8b3a6fba.jpg b/parse/train/HJg2b0VYDr/images/42536c76ac3983ea601745efa48ccc0a3d08ee32fd495e411fa23a8b8b3a6fba.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6b9887b2838131d3eaf6e267a2e926de2a55b781 --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/42536c76ac3983ea601745efa48ccc0a3d08ee32fd495e411fa23a8b8b3a6fba.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d8eaabcf0b4750884a5a1dd14ccecf8629eda1d2050cc56eb4d67d67af54f120 +size 5481 diff --git a/parse/train/HJg2b0VYDr/images/4913bb8962175fe0c05bf59ec64b03dc1f4973bfcc5ad6c1f7bb3e1fee923c59.jpg b/parse/train/HJg2b0VYDr/images/4913bb8962175fe0c05bf59ec64b03dc1f4973bfcc5ad6c1f7bb3e1fee923c59.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f6cff6e764eed6ac76a5ba5d6661de039edeffd8 --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/4913bb8962175fe0c05bf59ec64b03dc1f4973bfcc5ad6c1f7bb3e1fee923c59.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2a07792aa9ded5725850f3b26bfcac8ad9b7fecc3a5a2768bb76822396c096d4 +size 124401 diff --git a/parse/train/HJg2b0VYDr/images/5bca31bbd02679d6f14e18fd2d064d6f3903113bbb9991e2df61c29d47876987.jpg b/parse/train/HJg2b0VYDr/images/5bca31bbd02679d6f14e18fd2d064d6f3903113bbb9991e2df61c29d47876987.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fbba0cb1f939708a6b24ba3fcc186f7a8e47dbbc --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/5bca31bbd02679d6f14e18fd2d064d6f3903113bbb9991e2df61c29d47876987.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:beca7d3b4dcec448f9c1388e72b5157260f974fbf4e948b598d4fd93d28d4253 +size 55312 diff --git a/parse/train/HJg2b0VYDr/images/6183d318cb8298d337f477b1a65da6888298a1d31637ce1761aa5fb9ae5c24c5.jpg b/parse/train/HJg2b0VYDr/images/6183d318cb8298d337f477b1a65da6888298a1d31637ce1761aa5fb9ae5c24c5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9f906c8b3df478d77924097ced64fe02fdcee622 --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/6183d318cb8298d337f477b1a65da6888298a1d31637ce1761aa5fb9ae5c24c5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:62ebc5d7b3d60b14d1e2c3d51e4d6a5c15d1254d20c003cd4f3edb3b5ca405e9 +size 11027 diff --git a/parse/train/HJg2b0VYDr/images/7711d439e9b790460097cf03c072c47dad645b677428d43caecee95376340fc3.jpg b/parse/train/HJg2b0VYDr/images/7711d439e9b790460097cf03c072c47dad645b677428d43caecee95376340fc3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b739f3e6402d2a91c933eea78e1e7df0934145af --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/7711d439e9b790460097cf03c072c47dad645b677428d43caecee95376340fc3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8de93ccd9a1b2ec28317e7facd2c64733d66755aca9eb9b7d8b8984d173684b7 +size 68193 diff --git a/parse/train/HJg2b0VYDr/images/8b7d2cdf692dc9a0476aa1a93c28940d8f4ad9cca3eb1f3355648005dc9e1475.jpg b/parse/train/HJg2b0VYDr/images/8b7d2cdf692dc9a0476aa1a93c28940d8f4ad9cca3eb1f3355648005dc9e1475.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3062db700366cfaf6517bac55d850f7246826cee --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/8b7d2cdf692dc9a0476aa1a93c28940d8f4ad9cca3eb1f3355648005dc9e1475.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4206f8e1156eeeeb7e9940d6fcf18963b728ce325df21724af0d99a2b30bbc27 +size 26888 diff --git a/parse/train/HJg2b0VYDr/images/8ebfd91596117c390532fa258ff604630bf72a6c36741c6d59e61955b8b9f786.jpg b/parse/train/HJg2b0VYDr/images/8ebfd91596117c390532fa258ff604630bf72a6c36741c6d59e61955b8b9f786.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e5ef1fe995921fa6bd7c4ae0fb620f87b48eb71a --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/8ebfd91596117c390532fa258ff604630bf72a6c36741c6d59e61955b8b9f786.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:17d29ca025b949ed9bcd03a82f0a24559e31d3dbadcb941c08daf218497f8602 +size 51380 diff --git a/parse/train/HJg2b0VYDr/images/95e477ebc2fba5ea1aaae7d6b8f184ab81b1a1ad1cd21325059a6d1fb0e6d6b8.jpg b/parse/train/HJg2b0VYDr/images/95e477ebc2fba5ea1aaae7d6b8f184ab81b1a1ad1cd21325059a6d1fb0e6d6b8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e0d73bed5ddf62950df4e0f09a6de848f64daa11 --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/95e477ebc2fba5ea1aaae7d6b8f184ab81b1a1ad1cd21325059a6d1fb0e6d6b8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e378a9b037486a00da1705b7efb05b5fab89983da64c89a1049d6f844dc805ee +size 222268 diff --git a/parse/train/HJg2b0VYDr/images/a8eeb085fe80812913a2815f676e7f520e5d5c7dd526169e447c3ec424fb628e.jpg b/parse/train/HJg2b0VYDr/images/a8eeb085fe80812913a2815f676e7f520e5d5c7dd526169e447c3ec424fb628e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a68e31af0089d8aa3a3628deb06f69dc5d57900d --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/a8eeb085fe80812913a2815f676e7f520e5d5c7dd526169e447c3ec424fb628e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:93d746935aa43bb24d0d26bd9befc368f8c6e6205f832030ed17adea86e55ebd +size 54440 diff --git a/parse/train/HJg2b0VYDr/images/a93e35e30b426284c7bf1f182b5c22e6d60788926d10b3cd32e666b878479450.jpg b/parse/train/HJg2b0VYDr/images/a93e35e30b426284c7bf1f182b5c22e6d60788926d10b3cd32e666b878479450.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fbfdbb4334e9f16ff0dd08115a34ca75f6428220 --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/a93e35e30b426284c7bf1f182b5c22e6d60788926d10b3cd32e666b878479450.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9fbde4855462480be76f91f6158f3a5dee8908714837049d6ba22cab9fd435c2 +size 7831 diff --git a/parse/train/HJg2b0VYDr/images/b42767111d4bb37e1564616acb98610ea044d53840f77df96333ea5f31fe6e6c.jpg b/parse/train/HJg2b0VYDr/images/b42767111d4bb37e1564616acb98610ea044d53840f77df96333ea5f31fe6e6c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..07523d671033d00c86c21d3d108e107c43a294b5 --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/b42767111d4bb37e1564616acb98610ea044d53840f77df96333ea5f31fe6e6c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:474f29216f5fe3e017367f126a694da4e6af93334034811dd381e10c69681063 +size 50015 diff --git a/parse/train/HJg2b0VYDr/images/b49ae592f5c3826a11ab7bd55028f05d951b6ce39fb9d4ece92d471131b27969.jpg b/parse/train/HJg2b0VYDr/images/b49ae592f5c3826a11ab7bd55028f05d951b6ce39fb9d4ece92d471131b27969.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8d45bbd122d7594a22818c4025a69c556cb50c3b --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/b49ae592f5c3826a11ab7bd55028f05d951b6ce39fb9d4ece92d471131b27969.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9935d78899dbdce68f066ba61005cbf3d3aaafd7ca4af4dca9579a6ad932af10 +size 51915 diff --git a/parse/train/HJg2b0VYDr/images/c06a45b3d70a1a5e6ac6b8b1d3a904de73b1882f983b50d1cce7255327dcff5c.jpg b/parse/train/HJg2b0VYDr/images/c06a45b3d70a1a5e6ac6b8b1d3a904de73b1882f983b50d1cce7255327dcff5c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..30f139b7a0633e508b0b8fed2733227f9f977643 --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/c06a45b3d70a1a5e6ac6b8b1d3a904de73b1882f983b50d1cce7255327dcff5c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2a6ef91037de5142bb841ded10aacb3bcb439d287209ac05dd3a936adb0b153c +size 92859 diff --git a/parse/train/HJg2b0VYDr/images/d2e1999295459ae03026407e0764655a52cb8c17d5862d15837f858eeff535c2.jpg b/parse/train/HJg2b0VYDr/images/d2e1999295459ae03026407e0764655a52cb8c17d5862d15837f858eeff535c2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3bdb4e436a8f7a4fd89636dc0162bc00941a2aa5 --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/d2e1999295459ae03026407e0764655a52cb8c17d5862d15837f858eeff535c2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:df301503971d8afe3119828428792ec8171eb0c94e539773cc39df0dd625cf8c +size 55529 diff --git a/parse/train/HJg2b0VYDr/images/da512b9f1360ec15eaa2a45a8cfb6271ad24aae9865a39d6745ec60fbc2572ac.jpg b/parse/train/HJg2b0VYDr/images/da512b9f1360ec15eaa2a45a8cfb6271ad24aae9865a39d6745ec60fbc2572ac.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0e33a360a4fdf2826b090144235ff5c417123395 --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/da512b9f1360ec15eaa2a45a8cfb6271ad24aae9865a39d6745ec60fbc2572ac.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:32a002ced9208eea8ed12b5005aa816989284e62a446e3a4006e8f7982a946e8 +size 262041 diff --git a/parse/train/HJg2b0VYDr/images/ede7a5f78d4a88f314e7e1d700f01a1d4cb31394bdc50fb2e937cb986e8875d1.jpg b/parse/train/HJg2b0VYDr/images/ede7a5f78d4a88f314e7e1d700f01a1d4cb31394bdc50fb2e937cb986e8875d1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..91975c6cbd91f871a22adbe6c2af94ce041a807b --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/ede7a5f78d4a88f314e7e1d700f01a1d4cb31394bdc50fb2e937cb986e8875d1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:44aef2b3c1db25c6fdd0a91a4d4ce485aafdfb15b7081cb6d2ff43d9c08b1dd0 +size 141311 diff --git a/parse/train/HJg2b0VYDr/images/f1a632959169aba60e389816cc6a4b99b464cd5fcabd37fd672f7abbff6562f7.jpg b/parse/train/HJg2b0VYDr/images/f1a632959169aba60e389816cc6a4b99b464cd5fcabd37fd672f7abbff6562f7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2e9ecf5cd5bf62e4b1d7f5da25b2aa4b5ecc85da --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/f1a632959169aba60e389816cc6a4b99b464cd5fcabd37fd672f7abbff6562f7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec931860f6ba3ec9fa9e1cf7761919c69eb2a83f6750f41228ed9293f59ad86d +size 54951 diff --git a/parse/train/HJg2b0VYDr/images/f5f2e7977dcdf73da31b9ffe9da46484851bc70bb27e292913d649d864ae0bca.jpg b/parse/train/HJg2b0VYDr/images/f5f2e7977dcdf73da31b9ffe9da46484851bc70bb27e292913d649d864ae0bca.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4c023cf15d82f2d6c7c83d9bebd7632abb723d61 --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/f5f2e7977dcdf73da31b9ffe9da46484851bc70bb27e292913d649d864ae0bca.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6bf43d05c5fef9e143811d670985cae45a3c8645220297582a11656d99a71cdc +size 209463 diff --git a/parse/train/HJg2b0VYDr/images/f65c0b1db3859b42857dc9d512dbe27e33676ca4e652f7e62deaa6ca9a078f66.jpg b/parse/train/HJg2b0VYDr/images/f65c0b1db3859b42857dc9d512dbe27e33676ca4e652f7e62deaa6ca9a078f66.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3ffb00afeefe99129aba9056ccb96ceb752c0c04 --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/f65c0b1db3859b42857dc9d512dbe27e33676ca4e652f7e62deaa6ca9a078f66.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:014bb9cf2882a47918bac22a270644fe7466dc1a73bd9010653adeff6260b5f5 +size 96841 diff --git a/parse/train/HJg2b0VYDr/images/fe8fdd4f7bb83ccbbfc0b6352f9c83fcf3ebeb84100179861c915170427ac3a7.jpg b/parse/train/HJg2b0VYDr/images/fe8fdd4f7bb83ccbbfc0b6352f9c83fcf3ebeb84100179861c915170427ac3a7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6a53668f1bf98d5dcc348a1093975092db65b1b6 --- /dev/null +++ b/parse/train/HJg2b0VYDr/images/fe8fdd4f7bb83ccbbfc0b6352f9c83fcf3ebeb84100179861c915170427ac3a7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:76f572f7dfeebe6e33ab52f75fd8249a2ee2d2f12ad7339e222a1bbdd8c518d4 +size 69967 diff --git a/parse/train/HJgK0h4Ywr/images/0116cc06620fddf06a375189afb1efb1866f29924bb73e2ed28074b6f566cb41.jpg b/parse/train/HJgK0h4Ywr/images/0116cc06620fddf06a375189afb1efb1866f29924bb73e2ed28074b6f566cb41.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5217d01c687a7eea1ef3ba0d5517890873bcd78a --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/0116cc06620fddf06a375189afb1efb1866f29924bb73e2ed28074b6f566cb41.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:160c39a0d39528306b27b7db2db02f8f05b7b781e171cab32dd68e028f4f7be0 +size 4474 diff --git a/parse/train/HJgK0h4Ywr/images/02b1f4965a60bffa1ab3a40c1e82841017dfd0c4ebfc3ce0090d157c873f28ca.jpg b/parse/train/HJgK0h4Ywr/images/02b1f4965a60bffa1ab3a40c1e82841017dfd0c4ebfc3ce0090d157c873f28ca.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d0d41ec362662a1eb0666a5f7ad73119468a07d9 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/02b1f4965a60bffa1ab3a40c1e82841017dfd0c4ebfc3ce0090d157c873f28ca.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e31732ef7d76e7ce621c976d669a19da37c2d8c17cd4a3b2ac860910cb8b7081 +size 5774 diff --git a/parse/train/HJgK0h4Ywr/images/0d0bbb8e13f71e7e397323c9418453cd4b710c6d29e48be50753d5546062afef.jpg b/parse/train/HJgK0h4Ywr/images/0d0bbb8e13f71e7e397323c9418453cd4b710c6d29e48be50753d5546062afef.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0e1ea502e95812c2b82d29fd43b612bf39f06de0 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/0d0bbb8e13f71e7e397323c9418453cd4b710c6d29e48be50753d5546062afef.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c1938443f1bc65bc4032933f9e10cfa98183d12b43e92bbfce05b264b9273104 +size 5101 diff --git a/parse/train/HJgK0h4Ywr/images/0ef412e554c5c637f0f166a76c4f62b7e9912093a8a0cf57d606f8982ef50ad3.jpg b/parse/train/HJgK0h4Ywr/images/0ef412e554c5c637f0f166a76c4f62b7e9912093a8a0cf57d606f8982ef50ad3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6c0f7982dfe5966c8d89e0f4185991b59be43e76 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/0ef412e554c5c637f0f166a76c4f62b7e9912093a8a0cf57d606f8982ef50ad3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d0728e00b0cc311f174db12d19b42583405a37a78fac8d4fb7b2835c0d6b91f5 +size 18104 diff --git a/parse/train/HJgK0h4Ywr/images/0fe8148d7c426d62aa60546fdad01511eb5c471f19e34194f7abfd1fee21a143.jpg b/parse/train/HJgK0h4Ywr/images/0fe8148d7c426d62aa60546fdad01511eb5c471f19e34194f7abfd1fee21a143.jpg new file mode 100644 index 0000000000000000000000000000000000000000..541ca7d03b64de24f3dbad0fbdde6c7c7abd3be1 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/0fe8148d7c426d62aa60546fdad01511eb5c471f19e34194f7abfd1fee21a143.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:574c3c7eb0d0ddb374a19d03c54009372df291860ee0696d6e2260629c368f5e +size 19180 diff --git a/parse/train/HJgK0h4Ywr/images/0fed44cf92464a9464d1440548c9095519bf83c134a8c6a05c687eb54d38f43a.jpg b/parse/train/HJgK0h4Ywr/images/0fed44cf92464a9464d1440548c9095519bf83c134a8c6a05c687eb54d38f43a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b7af68a4d025c92d3560e89cd37b339726840d48 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/0fed44cf92464a9464d1440548c9095519bf83c134a8c6a05c687eb54d38f43a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a87af956e2cadf83857efe16e926a05e8e4c438d659b8eb8e53b5e49f36c6952 +size 5418 diff --git a/parse/train/HJgK0h4Ywr/images/10dd40eaeb42b4fb67e0da5fb0e5f5633db06042d8c2bbda6c809eadbcd64eda.jpg b/parse/train/HJgK0h4Ywr/images/10dd40eaeb42b4fb67e0da5fb0e5f5633db06042d8c2bbda6c809eadbcd64eda.jpg new file mode 100644 index 0000000000000000000000000000000000000000..89343a87d76f73c3c4b6ad2636c313bbd7b1b696 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/10dd40eaeb42b4fb67e0da5fb0e5f5633db06042d8c2bbda6c809eadbcd64eda.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:de430e546bca5495bdabff93cbf3d9262062487f2e82ffa06c471963a9cb7d2a +size 78957 diff --git a/parse/train/HJgK0h4Ywr/images/13a338f0f49d7bfb370994522b2d0c6d55b0fb052d47bcf36e184c3843f16823.jpg b/parse/train/HJgK0h4Ywr/images/13a338f0f49d7bfb370994522b2d0c6d55b0fb052d47bcf36e184c3843f16823.jpg new file mode 100644 index 0000000000000000000000000000000000000000..55b3774b0d3c7049334beb8e4f77931988ab8ac4 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/13a338f0f49d7bfb370994522b2d0c6d55b0fb052d47bcf36e184c3843f16823.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:198c3258fbd43844795cdd4fc605fec2560b8988003940a8e22a96edb160ed20 +size 41607 diff --git a/parse/train/HJgK0h4Ywr/images/14220a30eda7d8a5c6cb5a41c149b75f186d910aef2ded08ad714680a76bd0b4.jpg b/parse/train/HJgK0h4Ywr/images/14220a30eda7d8a5c6cb5a41c149b75f186d910aef2ded08ad714680a76bd0b4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9dfebfa2bc82eabf6a2705425079366de6d4d50a --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/14220a30eda7d8a5c6cb5a41c149b75f186d910aef2ded08ad714680a76bd0b4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:979b11f079655fc54f8e0bc0d8c4eb2ada0e295b1a215079009ccbffd1118c5d +size 5049 diff --git a/parse/train/HJgK0h4Ywr/images/156b0126ebd9db6abab06e8de11ed44a7986864db49f9b75fea723c1b75fd3b3.jpg b/parse/train/HJgK0h4Ywr/images/156b0126ebd9db6abab06e8de11ed44a7986864db49f9b75fea723c1b75fd3b3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cafb44f24c709b9aad5d2d3b5dc182d5c4954aec --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/156b0126ebd9db6abab06e8de11ed44a7986864db49f9b75fea723c1b75fd3b3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:216bf0f4b9ceb713230bb94a95f3fa17b53a5c599c73c9bce00cc2e889d9d97a +size 4943 diff --git a/parse/train/HJgK0h4Ywr/images/15b080bbc4388ffa68fc5bfebff3c88f07cb10b59f1c092dafa2c1818701fbe3.jpg b/parse/train/HJgK0h4Ywr/images/15b080bbc4388ffa68fc5bfebff3c88f07cb10b59f1c092dafa2c1818701fbe3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f62a3f7924eb4c8d3d5e9f91dc7674b224d54935 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/15b080bbc4388ffa68fc5bfebff3c88f07cb10b59f1c092dafa2c1818701fbe3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:811ff366a7df38c7ec24ab89cdface4689a2ba88819c259146bcf770e1bd5aa4 +size 12271 diff --git a/parse/train/HJgK0h4Ywr/images/15d55a0091469a56ade9a9fdf759f9d1b9633ef66c426a63931f029293abd312.jpg b/parse/train/HJgK0h4Ywr/images/15d55a0091469a56ade9a9fdf759f9d1b9633ef66c426a63931f029293abd312.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6f257f244c49c99cd9d16bd86949d8bc5d511603 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/15d55a0091469a56ade9a9fdf759f9d1b9633ef66c426a63931f029293abd312.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:95f8a5d8d75eedc3b4c3e8c58190190ced1d14476b142967895611c227629f7f +size 75929 diff --git a/parse/train/HJgK0h4Ywr/images/1c5c87bcf5a1dc5a309a793a4efd117282a1f5c5fdd4fd4168d0e65cc0fd8f7e.jpg b/parse/train/HJgK0h4Ywr/images/1c5c87bcf5a1dc5a309a793a4efd117282a1f5c5fdd4fd4168d0e65cc0fd8f7e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e93456268d1770351c060865722ac4e64e4e4b85 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/1c5c87bcf5a1dc5a309a793a4efd117282a1f5c5fdd4fd4168d0e65cc0fd8f7e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:92aa98e64a88f218addd12b3b6f5e7dc70b48f9f568f2eacef936b75a87befb5 +size 30137 diff --git a/parse/train/HJgK0h4Ywr/images/202c3e86423c17a497b0ad223acb4a45303034419768f72e578f064d5f356278.jpg b/parse/train/HJgK0h4Ywr/images/202c3e86423c17a497b0ad223acb4a45303034419768f72e578f064d5f356278.jpg new file mode 100644 index 0000000000000000000000000000000000000000..894aecd25bd7f3bfabaef65c88b6aa4126596eec --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/202c3e86423c17a497b0ad223acb4a45303034419768f72e578f064d5f356278.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a7e2c7afef444a2d01c13dfb2b14deef54a78effefc5e7973351e678cb46cd59 +size 6361 diff --git a/parse/train/HJgK0h4Ywr/images/2222b04ef6a59d0fe63ea413b3b3d8ab6f1be3957ff1b05d5806b930e7e75e51.jpg b/parse/train/HJgK0h4Ywr/images/2222b04ef6a59d0fe63ea413b3b3d8ab6f1be3957ff1b05d5806b930e7e75e51.jpg new file mode 100644 index 0000000000000000000000000000000000000000..059c2a6f6916b35deaa032e1347ef725b972022c --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/2222b04ef6a59d0fe63ea413b3b3d8ab6f1be3957ff1b05d5806b930e7e75e51.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1119a1274a44778258fc475e78778885c783e4ce945bb4b52f7d357d2a958b86 +size 1956 diff --git a/parse/train/HJgK0h4Ywr/images/27a956ab7689c6af42ccaadedeb297bc1b2dd8158fafda4180ffb15db66bfe91.jpg b/parse/train/HJgK0h4Ywr/images/27a956ab7689c6af42ccaadedeb297bc1b2dd8158fafda4180ffb15db66bfe91.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d69cc14526adfa27a4c0091038f56f7154dcc1c5 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/27a956ab7689c6af42ccaadedeb297bc1b2dd8158fafda4180ffb15db66bfe91.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5563cf827d04a5c2698a63381c928695317da3fb2e7e75b5b746624328441e04 +size 23005 diff --git a/parse/train/HJgK0h4Ywr/images/2ba0f4d5d16d2b384109e8c1beaeed41cbb49168e79afa968b3260ebba755fbb.jpg b/parse/train/HJgK0h4Ywr/images/2ba0f4d5d16d2b384109e8c1beaeed41cbb49168e79afa968b3260ebba755fbb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..23587eec574bc5498c6439df6666c4670078d092 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/2ba0f4d5d16d2b384109e8c1beaeed41cbb49168e79afa968b3260ebba755fbb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3acca11c09fe9e64df956a1629b867369f3dffa69e85ae3fd54a8dc92d9f2be7 +size 45906 diff --git a/parse/train/HJgK0h4Ywr/images/2ce914cd6efb59d915e922876067580cfe70bf4bd4ab6d48ced9a6a443bf4a74.jpg b/parse/train/HJgK0h4Ywr/images/2ce914cd6efb59d915e922876067580cfe70bf4bd4ab6d48ced9a6a443bf4a74.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ef68e5060bf0ab3eb637d460a663497f90c50501 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/2ce914cd6efb59d915e922876067580cfe70bf4bd4ab6d48ced9a6a443bf4a74.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5033baaca4cf4cdcf82b336ee70ec6e06ef24c6290e770ec9035e2e51252e477 +size 39246 diff --git a/parse/train/HJgK0h4Ywr/images/2e6caaff72bb050303e2bc9c72a45414fbca6f0ca7a4f22ecdba64fb653b7e94.jpg b/parse/train/HJgK0h4Ywr/images/2e6caaff72bb050303e2bc9c72a45414fbca6f0ca7a4f22ecdba64fb653b7e94.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d68e15186e8b4d28c41b8c77e7100d74a4c377e6 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/2e6caaff72bb050303e2bc9c72a45414fbca6f0ca7a4f22ecdba64fb653b7e94.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2c0ce50de59ebf5e7f3e67c3382ed6e2356b3dcf92d7a0767fced68056d96138 +size 21043 diff --git a/parse/train/HJgK0h4Ywr/images/31ac49a7e07784ba60245f3e8e6c4c5d9b67b5446ba86f03513ad4f2a6f63866.jpg b/parse/train/HJgK0h4Ywr/images/31ac49a7e07784ba60245f3e8e6c4c5d9b67b5446ba86f03513ad4f2a6f63866.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7d4da0378b12fd2b12e69233a63edd1d6f380d86 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/31ac49a7e07784ba60245f3e8e6c4c5d9b67b5446ba86f03513ad4f2a6f63866.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0df9346f7b102bfb62ede39f973cace43ea2f11ac1ed8b37e9bfc9c9f3b5e586 +size 5465 diff --git a/parse/train/HJgK0h4Ywr/images/32eb53272d2f70256ce3437d03fdcd64cfddf683f973854e8cb8b8662196289d.jpg b/parse/train/HJgK0h4Ywr/images/32eb53272d2f70256ce3437d03fdcd64cfddf683f973854e8cb8b8662196289d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b89c81d1d55803a394b5eb8e752d29a2a815528b --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/32eb53272d2f70256ce3437d03fdcd64cfddf683f973854e8cb8b8662196289d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:340d15cfba3175cd82c0d26a78a038c63c840f1d91ab1f11c59981fd509f6e26 +size 76006 diff --git a/parse/train/HJgK0h4Ywr/images/35a559a79dcf8d906788a600d6dcce1b9e4061e0ba22925948deb3452cefe594.jpg b/parse/train/HJgK0h4Ywr/images/35a559a79dcf8d906788a600d6dcce1b9e4061e0ba22925948deb3452cefe594.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3aca3f750aacc5fb0821b0ed0c1390e3b90bb887 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/35a559a79dcf8d906788a600d6dcce1b9e4061e0ba22925948deb3452cefe594.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d27055157eb940454cd2c240511955b43c3b40e68e310ed41c8b6b89e0c941a3 +size 4976 diff --git a/parse/train/HJgK0h4Ywr/images/38a62684499d1a8097de5a1c4d82bb50250559fbf4b331fed535a6b26e4f494d.jpg b/parse/train/HJgK0h4Ywr/images/38a62684499d1a8097de5a1c4d82bb50250559fbf4b331fed535a6b26e4f494d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..61385eacf06d1b82745b2fdd4614b24decf5d267 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/38a62684499d1a8097de5a1c4d82bb50250559fbf4b331fed535a6b26e4f494d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b4ac9acdbfe186abb6fa05d53e627deb0818dc7e53b413df1860a3b886062bf7 +size 7781 diff --git a/parse/train/HJgK0h4Ywr/images/4147cc623201a135e5901b47e231ce6cef58aa7431da42ffaef158c2ce894d5c.jpg b/parse/train/HJgK0h4Ywr/images/4147cc623201a135e5901b47e231ce6cef58aa7431da42ffaef158c2ce894d5c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a9e2674127e217d5b6fc2ce96f871a32f6d8748f --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/4147cc623201a135e5901b47e231ce6cef58aa7431da42ffaef158c2ce894d5c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec75d9c7c7ae95b6ee5c0f6b8ba9b9d6729df4138b15fa1c4e720ae663c7d5a9 +size 7830 diff --git a/parse/train/HJgK0h4Ywr/images/43d1c581a315f624bc425342a964fa40581159e9f4db5d6becbbedac3dab8f52.jpg b/parse/train/HJgK0h4Ywr/images/43d1c581a315f624bc425342a964fa40581159e9f4db5d6becbbedac3dab8f52.jpg new file mode 100644 index 0000000000000000000000000000000000000000..adbb9840fd8198b854a6eefd5bbbc0aad205a20a --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/43d1c581a315f624bc425342a964fa40581159e9f4db5d6becbbedac3dab8f52.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:52cc817cd95b80df86d71602a7e5410730170bb3e46e82fd37c8b84e14419098 +size 76862 diff --git a/parse/train/HJgK0h4Ywr/images/4621b45330e09a0b7108bbf0d6eb47c8c0f9d7256f14316a178f8b81a1daa61c.jpg b/parse/train/HJgK0h4Ywr/images/4621b45330e09a0b7108bbf0d6eb47c8c0f9d7256f14316a178f8b81a1daa61c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7cd04e73fbfe1784fa37d2ce95dfae857ad6614f --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/4621b45330e09a0b7108bbf0d6eb47c8c0f9d7256f14316a178f8b81a1daa61c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cceca8b61d9f4c19f892e4bd97c05b8be99f65d0486ceb649f7a683d60bb06ce +size 26941 diff --git a/parse/train/HJgK0h4Ywr/images/49546410c8b203d4e06d2f2c4216c71303a0327f251dc668c135c0b794dea121.jpg b/parse/train/HJgK0h4Ywr/images/49546410c8b203d4e06d2f2c4216c71303a0327f251dc668c135c0b794dea121.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2f602dd0e3bb81c52729e09428fd854ccfb9d4a4 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/49546410c8b203d4e06d2f2c4216c71303a0327f251dc668c135c0b794dea121.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6b8c3cbed244cc16a289e9c504b3609ccc1adfd9438fdb8d3fc2a69c8060afe4 +size 36930 diff --git a/parse/train/HJgK0h4Ywr/images/4a2a842d53cb6d66b8e808824490792cf590d16243356040c2d401506c1ffaad.jpg b/parse/train/HJgK0h4Ywr/images/4a2a842d53cb6d66b8e808824490792cf590d16243356040c2d401506c1ffaad.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dc2f7d9a7fb6acea5887607e32375b659afe6ef8 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/4a2a842d53cb6d66b8e808824490792cf590d16243356040c2d401506c1ffaad.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:02e534bd990b4d9e6f70de43b1295683c5ec55fda0109f0ac3560e9f3495c8c3 +size 15142 diff --git a/parse/train/HJgK0h4Ywr/images/5121a18d4b8d1643672d86c387197e66bf78efe9550d619a88782c1d6d80a5f2.jpg b/parse/train/HJgK0h4Ywr/images/5121a18d4b8d1643672d86c387197e66bf78efe9550d619a88782c1d6d80a5f2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c490ff2947bbb1b363c17e5f6b1f3adb92786f2d --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/5121a18d4b8d1643672d86c387197e66bf78efe9550d619a88782c1d6d80a5f2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:87824ccbcbc272a3baf8a7f9c048d1f481838f4a41410278c2ef9533ca773c22 +size 7596 diff --git a/parse/train/HJgK0h4Ywr/images/56f2f6be83d6f1fce60747497b87b918323aaa43acf0d9965925cf68e53cd2b1.jpg b/parse/train/HJgK0h4Ywr/images/56f2f6be83d6f1fce60747497b87b918323aaa43acf0d9965925cf68e53cd2b1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8f584bec4f09c5ee7d85d1177a0623872d6abb2b --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/56f2f6be83d6f1fce60747497b87b918323aaa43acf0d9965925cf68e53cd2b1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:53e5cd3fd21c269a72e5b79ea66c3abb87a7a51dd1a3025a068e3a5efbb53acd +size 15879 diff --git a/parse/train/HJgK0h4Ywr/images/57b0fd291ae8ce620cf9d7d7a84ee38592b4c4f82f92f3240a6e44f035175902.jpg b/parse/train/HJgK0h4Ywr/images/57b0fd291ae8ce620cf9d7d7a84ee38592b4c4f82f92f3240a6e44f035175902.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3a4d878167300bac4821c846d39eeefe2b3e8e2f --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/57b0fd291ae8ce620cf9d7d7a84ee38592b4c4f82f92f3240a6e44f035175902.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cefe94c4ae9b93a42215eff875a6624f3696b78a8e8218d495ff03a766a0f473 +size 10668 diff --git a/parse/train/HJgK0h4Ywr/images/5c5a317b50c0b71e0f36f3298e35e1e41d3cf2a5decbf79b0fd2c4b004d0e2ee.jpg b/parse/train/HJgK0h4Ywr/images/5c5a317b50c0b71e0f36f3298e35e1e41d3cf2a5decbf79b0fd2c4b004d0e2ee.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f83addb93bb1a6b00a6d69f1af0088923502a201 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/5c5a317b50c0b71e0f36f3298e35e1e41d3cf2a5decbf79b0fd2c4b004d0e2ee.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ce55d16de47a4be3dfd65d06bb54fa10612db206060004f30b5c9701dc67c68e +size 69677 diff --git a/parse/train/HJgK0h4Ywr/images/6102459cfe8b4bea5d13644dc769bfdd67b907e5c6e27ed2a29a5893e9af56ba.jpg b/parse/train/HJgK0h4Ywr/images/6102459cfe8b4bea5d13644dc769bfdd67b907e5c6e27ed2a29a5893e9af56ba.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e5033fc18b966a5230650950a2d58f17372322bf --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/6102459cfe8b4bea5d13644dc769bfdd67b907e5c6e27ed2a29a5893e9af56ba.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:766906bf54429e502c43ed80b7f45f4a145f1c33d21fa02adfd9c2b0f25b16e6 +size 30558 diff --git a/parse/train/HJgK0h4Ywr/images/6169ca36d0f3a993effc4148e9838d72cc00625db5cad8e733852538aa7e82a2.jpg b/parse/train/HJgK0h4Ywr/images/6169ca36d0f3a993effc4148e9838d72cc00625db5cad8e733852538aa7e82a2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..94a43fbed86f4569c9470970a9e6172457c9b702 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/6169ca36d0f3a993effc4148e9838d72cc00625db5cad8e733852538aa7e82a2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fc7b0314c7b012dc45c706af51a2df673798580891ec7a43f89f8da23d6b5ae8 +size 3630 diff --git a/parse/train/HJgK0h4Ywr/images/647a6a0aeec0169a10f04dc5fd7cbb27fe9b514064e479a88c6d692f48fc1d6f.jpg b/parse/train/HJgK0h4Ywr/images/647a6a0aeec0169a10f04dc5fd7cbb27fe9b514064e479a88c6d692f48fc1d6f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0e68150cc0a92181fa4484add810bff966f22191 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/647a6a0aeec0169a10f04dc5fd7cbb27fe9b514064e479a88c6d692f48fc1d6f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a8067638128b22c89d261989eaf96877120499f04f0fb3a3feada60782258ff2 +size 4761 diff --git a/parse/train/HJgK0h4Ywr/images/65203b49c3c917820bddddf7f237f76729c595504f43155baaaad98ae402a69b.jpg b/parse/train/HJgK0h4Ywr/images/65203b49c3c917820bddddf7f237f76729c595504f43155baaaad98ae402a69b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0cf958cc65a055f0fb7151c784cace2b79e8e242 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/65203b49c3c917820bddddf7f237f76729c595504f43155baaaad98ae402a69b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7f06ab689ec7fbe85b0d23d81061843d2afb9e4ad0c02ac3dc145b9ea9072b6b +size 72099 diff --git a/parse/train/HJgK0h4Ywr/images/69aa415cf8b0f009b63e022af7548970d6580e241475826e116c7fd123604e09.jpg b/parse/train/HJgK0h4Ywr/images/69aa415cf8b0f009b63e022af7548970d6580e241475826e116c7fd123604e09.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3d9426ca68a1958642dca8270bf17587ac685305 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/69aa415cf8b0f009b63e022af7548970d6580e241475826e116c7fd123604e09.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ea8e4e5f14aa6de6062759e293887bd74df1f925b9f5b8789da6e477e49854d2 +size 22542 diff --git a/parse/train/HJgK0h4Ywr/images/6b599ac8ce7c991df21bf38571455d12bc8e970eae5494650229a4a0d9a5e971.jpg b/parse/train/HJgK0h4Ywr/images/6b599ac8ce7c991df21bf38571455d12bc8e970eae5494650229a4a0d9a5e971.jpg new file mode 100644 index 0000000000000000000000000000000000000000..714e6d813ea2982684d36a78923d25caa016266b --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/6b599ac8ce7c991df21bf38571455d12bc8e970eae5494650229a4a0d9a5e971.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b5a9535378930521816f196b848a155df13127692035685b515a30447b041bba +size 8020 diff --git a/parse/train/HJgK0h4Ywr/images/72ff9dd9b5995f94637d58de9d00bf676c03e184764caae6d0723160f61146f3.jpg b/parse/train/HJgK0h4Ywr/images/72ff9dd9b5995f94637d58de9d00bf676c03e184764caae6d0723160f61146f3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bf74884ac4242b3685b74596d99545ebd2d0acb2 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/72ff9dd9b5995f94637d58de9d00bf676c03e184764caae6d0723160f61146f3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4b88a5581a78a4db946fbc88b6e24dd473aed64ea85fb51b7f3aafc91e755ac7 +size 150897 diff --git a/parse/train/HJgK0h4Ywr/images/75c82c26aaac7f41c08b6d558889f5ff8ccdea4fab923f66520ed6d758006654.jpg b/parse/train/HJgK0h4Ywr/images/75c82c26aaac7f41c08b6d558889f5ff8ccdea4fab923f66520ed6d758006654.jpg new file mode 100644 index 0000000000000000000000000000000000000000..97866b8bca484d04e48eaa1722dd5baa719de0bb --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/75c82c26aaac7f41c08b6d558889f5ff8ccdea4fab923f66520ed6d758006654.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2bc466217341d60c7a98b95c2d3dbaf9b148f330e5230c5155a83e66e392c7db +size 17221 diff --git a/parse/train/HJgK0h4Ywr/images/82ee9a8f7651a89e381857b436d51aa432a30950a2f8a118e7dbcbe38ea77744.jpg b/parse/train/HJgK0h4Ywr/images/82ee9a8f7651a89e381857b436d51aa432a30950a2f8a118e7dbcbe38ea77744.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ef749e8c03edbba959ac7f73e5f9ed0f784c1e6d --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/82ee9a8f7651a89e381857b436d51aa432a30950a2f8a118e7dbcbe38ea77744.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3b8cf7b3ce13a55a4b45d4333f6fb406b6cdac4a6351bc1ad81f98b78103dcd8 +size 53798 diff --git a/parse/train/HJgK0h4Ywr/images/840125de866003d49d4bbf9007a0b8f000aae7debbb83c12d169855bc612ca9f.jpg b/parse/train/HJgK0h4Ywr/images/840125de866003d49d4bbf9007a0b8f000aae7debbb83c12d169855bc612ca9f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1c2a96cc3251ce273e6f2630e7da26d030435408 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/840125de866003d49d4bbf9007a0b8f000aae7debbb83c12d169855bc612ca9f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f46b92f7786ae9831c0747ed38f50fe32eecd8b931db078921e9de18756602a7 +size 83616 diff --git a/parse/train/HJgK0h4Ywr/images/86de28cb8b230d6b23dcd3cfb65e94db1d2d52eb1a592d1d6c42f560955b7afd.jpg b/parse/train/HJgK0h4Ywr/images/86de28cb8b230d6b23dcd3cfb65e94db1d2d52eb1a592d1d6c42f560955b7afd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d8754d832db13fd666ca2ad1df4503c2f24f64c4 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/86de28cb8b230d6b23dcd3cfb65e94db1d2d52eb1a592d1d6c42f560955b7afd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:25f070b53d87747913faf351b04694566f7b592183214318e89bd5711b5b9cd9 +size 25506 diff --git a/parse/train/HJgK0h4Ywr/images/87bcc074864eb9ebf3867988e38f621402f5902b4318cd335abef659a8f38a12.jpg b/parse/train/HJgK0h4Ywr/images/87bcc074864eb9ebf3867988e38f621402f5902b4318cd335abef659a8f38a12.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7b8deb8f26aef3be7fa6f7f77b83f43098a75e65 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/87bcc074864eb9ebf3867988e38f621402f5902b4318cd335abef659a8f38a12.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:912adbb4cb37f9cd468830b8293952762052ac386f184026f098b5186695dec1 +size 8523 diff --git a/parse/train/HJgK0h4Ywr/images/87d5e78417a066c07972dfa376244910b7e9940783f4fc653bd49f532c20d93a.jpg b/parse/train/HJgK0h4Ywr/images/87d5e78417a066c07972dfa376244910b7e9940783f4fc653bd49f532c20d93a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ee0e0279a016b0e7a61f025f0908324e93c47497 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/87d5e78417a066c07972dfa376244910b7e9940783f4fc653bd49f532c20d93a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0e72edb6d60263ed7431a0a6ca4754355b8780a1e564beb607a50524eda7c3c8 +size 66375 diff --git a/parse/train/HJgK0h4Ywr/images/8eb40d06f0990a4344803bbe3f2034ca85c62a9510447bff4a005c33df87f027.jpg b/parse/train/HJgK0h4Ywr/images/8eb40d06f0990a4344803bbe3f2034ca85c62a9510447bff4a005c33df87f027.jpg new file mode 100644 index 0000000000000000000000000000000000000000..124a07ff129752ca205aa2f26fa37b22263a9578 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/8eb40d06f0990a4344803bbe3f2034ca85c62a9510447bff4a005c33df87f027.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:761988603d79c6441c89707f71b9bdd36f8a796e12a83fcf84f931f27ad76012 +size 28582 diff --git a/parse/train/HJgK0h4Ywr/images/9156aef7cb4bfd4ec2d77ea86180827c5474885f2ca78ea40b46c0cf2df06f02.jpg b/parse/train/HJgK0h4Ywr/images/9156aef7cb4bfd4ec2d77ea86180827c5474885f2ca78ea40b46c0cf2df06f02.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d73585b055df8ebb8756f6ccfe0fa5cbacaff31a --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/9156aef7cb4bfd4ec2d77ea86180827c5474885f2ca78ea40b46c0cf2df06f02.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1aa98d9e0b6433886e46c91bfb642f9806ff1e98005cfbf41681a69c151164aa +size 38610 diff --git a/parse/train/HJgK0h4Ywr/images/91adf93a1f5cf890f89fa245e315ffb811d8160b471ac8186078b605821b9c31.jpg b/parse/train/HJgK0h4Ywr/images/91adf93a1f5cf890f89fa245e315ffb811d8160b471ac8186078b605821b9c31.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6f83bbcf45bc99cb0be833a9b9044b02d073fe85 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/91adf93a1f5cf890f89fa245e315ffb811d8160b471ac8186078b605821b9c31.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:74d7c6aa6ca36208e4027e68972d687cd4c8af768288f96bcc41a0f9cc227832 +size 39518 diff --git a/parse/train/HJgK0h4Ywr/images/937891fc0866f4582d29f9ffaed6eb7b0d0d890b9da2d7431f0e8641d413dd48.jpg b/parse/train/HJgK0h4Ywr/images/937891fc0866f4582d29f9ffaed6eb7b0d0d890b9da2d7431f0e8641d413dd48.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6586b4f1e2d5c5e2d167439819257cd31a14518f --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/937891fc0866f4582d29f9ffaed6eb7b0d0d890b9da2d7431f0e8641d413dd48.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a6185383cc3899f9853559c857fbdc62eb81db28ba85d3f1e149f2ec219c65d4 +size 30401 diff --git a/parse/train/HJgK0h4Ywr/images/963f5631f66ee8d3cf6ce49ce8f23dfa99a2fc28dfb3d204386005d15e412de1.jpg b/parse/train/HJgK0h4Ywr/images/963f5631f66ee8d3cf6ce49ce8f23dfa99a2fc28dfb3d204386005d15e412de1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..480c9369a34d494ed9698832df7b6637db7adaba --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/963f5631f66ee8d3cf6ce49ce8f23dfa99a2fc28dfb3d204386005d15e412de1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:267e79183d1a5770f5c86e93a00eace5760ae18cdb9c8a013647065aaf8a3d9d +size 24262 diff --git a/parse/train/HJgK0h4Ywr/images/9b2647745f13e534693d5b749fe66ea49fcf7ee15f40946093ebafe41585a0ff.jpg b/parse/train/HJgK0h4Ywr/images/9b2647745f13e534693d5b749fe66ea49fcf7ee15f40946093ebafe41585a0ff.jpg new file mode 100644 index 0000000000000000000000000000000000000000..19636a569b80466a6efec9e781ef09237e9d5fdb --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/9b2647745f13e534693d5b749fe66ea49fcf7ee15f40946093ebafe41585a0ff.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:96c5a14cbb6bcaeb13d2a0c0fdc57389c9db154862e3e692895c40538186724c +size 5110 diff --git a/parse/train/HJgK0h4Ywr/images/9b3c8e9d3c6760157f852ac29c69e437c3a6fc71baa2864ccdef22afa4e64c90.jpg b/parse/train/HJgK0h4Ywr/images/9b3c8e9d3c6760157f852ac29c69e437c3a6fc71baa2864ccdef22afa4e64c90.jpg new file mode 100644 index 0000000000000000000000000000000000000000..09da5107b2dfa3721736d70559cc93cb5924c7bd --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/9b3c8e9d3c6760157f852ac29c69e437c3a6fc71baa2864ccdef22afa4e64c90.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:98155d2fe4690ad6db593630eb550931cc7013a7b10a3337f88316b96eb8aea8 +size 31765 diff --git a/parse/train/HJgK0h4Ywr/images/9c87c440370eddae06369bfce5f93c4feb29afb8f6f0175155e5f8c508822e30.jpg b/parse/train/HJgK0h4Ywr/images/9c87c440370eddae06369bfce5f93c4feb29afb8f6f0175155e5f8c508822e30.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9945219f0295cac2307ae763d45329749858995d --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/9c87c440370eddae06369bfce5f93c4feb29afb8f6f0175155e5f8c508822e30.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a960cd66c875a0ee30ee0a131d2e5cf421d0d1435f46c2c7c6fb0b49c0c20a02 +size 4512 diff --git a/parse/train/HJgK0h4Ywr/images/9e5de9bffb566c7d9deecc007c855f5eb469f3a6c73ae76d2f49842644f39253.jpg b/parse/train/HJgK0h4Ywr/images/9e5de9bffb566c7d9deecc007c855f5eb469f3a6c73ae76d2f49842644f39253.jpg new file mode 100644 index 0000000000000000000000000000000000000000..533c19ca77015138124e885ac89f095635dce0f2 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/9e5de9bffb566c7d9deecc007c855f5eb469f3a6c73ae76d2f49842644f39253.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6274cf328445286fd24b4a6ab8d0c7a03bc970882a2e8ac36f195a38ab6d8732 +size 4966 diff --git a/parse/train/HJgK0h4Ywr/images/a0e8a257760cdcc4b57ca07d8d1b7008b124bd0c40ac62125af611937c3b64a2.jpg b/parse/train/HJgK0h4Ywr/images/a0e8a257760cdcc4b57ca07d8d1b7008b124bd0c40ac62125af611937c3b64a2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..40c37004cb5ac29a3d6d2921b5ddbb3e0f0a2770 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/a0e8a257760cdcc4b57ca07d8d1b7008b124bd0c40ac62125af611937c3b64a2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fea58e72b054426b5b9219741191130dea1d40b04534d94000ebf8f6574bae39 +size 4632 diff --git a/parse/train/HJgK0h4Ywr/images/a10838d425e7c9fac2e0788d6c3643557fb55b5b6041cf225d34e17f2419786b.jpg b/parse/train/HJgK0h4Ywr/images/a10838d425e7c9fac2e0788d6c3643557fb55b5b6041cf225d34e17f2419786b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..71b39144220a9d0d97d3d3f9f929bf1c30f9c61b --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/a10838d425e7c9fac2e0788d6c3643557fb55b5b6041cf225d34e17f2419786b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cbb92ab77be5103339beddfaaa5768b69403e458fa0dbadcdc896a4ee1d27d9c +size 103910 diff --git a/parse/train/HJgK0h4Ywr/images/a5792a52ef9c965156ae9d79e041db566ac016426a11ab113972afe258de71dc.jpg b/parse/train/HJgK0h4Ywr/images/a5792a52ef9c965156ae9d79e041db566ac016426a11ab113972afe258de71dc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0e7862af37c816a8b45770dbba75b7fdc6f015b3 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/a5792a52ef9c965156ae9d79e041db566ac016426a11ab113972afe258de71dc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cbedbd14eb9398d8a1c98e78af16b86874102836cd94181682fb7abfda46cea7 +size 72354 diff --git a/parse/train/HJgK0h4Ywr/images/a8488bcf7f19de51de14f5b245b6adec438dfbdc6cb2ad7896d45cebaf8d70cb.jpg b/parse/train/HJgK0h4Ywr/images/a8488bcf7f19de51de14f5b245b6adec438dfbdc6cb2ad7896d45cebaf8d70cb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a8299a29bdfd68de32a31d1d6b2d3d25d4dadab4 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/a8488bcf7f19de51de14f5b245b6adec438dfbdc6cb2ad7896d45cebaf8d70cb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0e1def68645363b66e5ee77d492690bb554440405d34f28408d7cb82efa3374d +size 2180 diff --git a/parse/train/HJgK0h4Ywr/images/a9de644351be11c68df7e45d43c4146fc1ad302034450357f38fe3d9dd8b1e60.jpg b/parse/train/HJgK0h4Ywr/images/a9de644351be11c68df7e45d43c4146fc1ad302034450357f38fe3d9dd8b1e60.jpg new file mode 100644 index 0000000000000000000000000000000000000000..aa7d579ac16c26ead8b3274b7cdf76bb15c0bbad --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/a9de644351be11c68df7e45d43c4146fc1ad302034450357f38fe3d9dd8b1e60.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:16dee4a05492ae0397e28845d72c1e72d6a47a46594997f44d8d9839ee856ad0 +size 7319 diff --git a/parse/train/HJgK0h4Ywr/images/aa6c496a3c137d2b302de7d6fa373e5659dda70c3b9c62fb7eae06e2038d857b.jpg b/parse/train/HJgK0h4Ywr/images/aa6c496a3c137d2b302de7d6fa373e5659dda70c3b9c62fb7eae06e2038d857b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cde2d4fc2762b69f980bdf4882d13ca4cb652cb1 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/aa6c496a3c137d2b302de7d6fa373e5659dda70c3b9c62fb7eae06e2038d857b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:48fd418a66b1ae026f6c02aeb5d6a390096a68476c12cc1621dfa9b42237e3f7 +size 7269 diff --git a/parse/train/HJgK0h4Ywr/images/ab277c3709d9005a8cd7f03a9c9e903a95d8abbbc712aa90bc39d5f228b8f1de.jpg b/parse/train/HJgK0h4Ywr/images/ab277c3709d9005a8cd7f03a9c9e903a95d8abbbc712aa90bc39d5f228b8f1de.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2c3c77cff3d725b92c5b38761fbd8fef5d6b7642 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/ab277c3709d9005a8cd7f03a9c9e903a95d8abbbc712aa90bc39d5f228b8f1de.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:880fd741b25bb48f04b1ff6790e7468130fb17923755b695ed9105fc9fe8fbeb +size 4428 diff --git a/parse/train/HJgK0h4Ywr/images/b55c13edc8762c82b4f9909bada6c5927899ecdd49a8ece9dcfe940d98e591ac.jpg b/parse/train/HJgK0h4Ywr/images/b55c13edc8762c82b4f9909bada6c5927899ecdd49a8ece9dcfe940d98e591ac.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e24b8484239e13dd6260f5b6d6b8aabc2503339b --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/b55c13edc8762c82b4f9909bada6c5927899ecdd49a8ece9dcfe940d98e591ac.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:19135253b07d64dadeaccaa8da2acd0f2c2b899bb21c0bab979f4be37d3f08bd +size 13053 diff --git a/parse/train/HJgK0h4Ywr/images/b5fa71feed0f422bb570b0720e9daede1c317cbadf32dda32af564d618bc2fd1.jpg b/parse/train/HJgK0h4Ywr/images/b5fa71feed0f422bb570b0720e9daede1c317cbadf32dda32af564d618bc2fd1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..49b82067f26a73eb07f8469d6c8cf625742f6385 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/b5fa71feed0f422bb570b0720e9daede1c317cbadf32dda32af564d618bc2fd1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3b5a9d48fcaf3ba646b4ab6d0e4b1f08eed30c360b4a98a833fa5e7872d60f5a +size 142482 diff --git a/parse/train/HJgK0h4Ywr/images/b70db9867d49d00bd452ea495216cc17d0fdc263054b8eb833c8326c50516b1a.jpg b/parse/train/HJgK0h4Ywr/images/b70db9867d49d00bd452ea495216cc17d0fdc263054b8eb833c8326c50516b1a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..13bd4193279257938dd1c9509c3063b77613930e --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/b70db9867d49d00bd452ea495216cc17d0fdc263054b8eb833c8326c50516b1a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:99b8620a2b676ebf01ff5a41f5f8dfb10a4cbef474113e20eeb0012cdcd09349 +size 80343 diff --git a/parse/train/HJgK0h4Ywr/images/b767b2f1c3a5f9b1ccd4df9d8e03f09e404dfc42a4aa1a586f49e418eabc6413.jpg b/parse/train/HJgK0h4Ywr/images/b767b2f1c3a5f9b1ccd4df9d8e03f09e404dfc42a4aa1a586f49e418eabc6413.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a6a35646e228a355469330294e9cfb52bf328441 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/b767b2f1c3a5f9b1ccd4df9d8e03f09e404dfc42a4aa1a586f49e418eabc6413.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6ebda3e89fe7b6a6a03808f97c2e6db0bfe216fea0a20fcd135bea4e75319620 +size 8347 diff --git a/parse/train/HJgK0h4Ywr/images/bf83d5ad8023529dbd238b55539522a8fe0c86abb49ea727b9fe2f48cc8c6600.jpg b/parse/train/HJgK0h4Ywr/images/bf83d5ad8023529dbd238b55539522a8fe0c86abb49ea727b9fe2f48cc8c6600.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0fee56ea533bc8296c39db4cf844f3b75335e31a --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/bf83d5ad8023529dbd238b55539522a8fe0c86abb49ea727b9fe2f48cc8c6600.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fd5a0dc4124a814a1472d4056e72b4052cdb9e82388d6df3024da16fa374e72d +size 4539 diff --git a/parse/train/HJgK0h4Ywr/images/c3ddc2556a743ee120354d27522833c4204bf69e653d3936a526603d8f129dcd.jpg b/parse/train/HJgK0h4Ywr/images/c3ddc2556a743ee120354d27522833c4204bf69e653d3936a526603d8f129dcd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c7a686df1b95ff410ba3996858a40572290e9617 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/c3ddc2556a743ee120354d27522833c4204bf69e653d3936a526603d8f129dcd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:af7513663f444b902abcf3bd059abfeadf2f6c71d980ccb4796e1915df9f33ac +size 31063 diff --git a/parse/train/HJgK0h4Ywr/images/c587147a988e7002a8a270e83b4d1dc0f59f31306ed51654c3517a12c987e36b.jpg b/parse/train/HJgK0h4Ywr/images/c587147a988e7002a8a270e83b4d1dc0f59f31306ed51654c3517a12c987e36b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c8a1acf895d9ce7e63bf4949fed11b66fe075573 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/c587147a988e7002a8a270e83b4d1dc0f59f31306ed51654c3517a12c987e36b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2cdc87138a718fda5806cc281218c4de37914e3b1571e7151b8cca5b0c0a82d8 +size 7502 diff --git a/parse/train/HJgK0h4Ywr/images/d1b2d468a360845dbbcda444ac5e7fda8ec1ab9c63b94c47c3450f87503e3286.jpg b/parse/train/HJgK0h4Ywr/images/d1b2d468a360845dbbcda444ac5e7fda8ec1ab9c63b94c47c3450f87503e3286.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e1c3c69940f9f4d05d163e865e47126a4f67c558 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/d1b2d468a360845dbbcda444ac5e7fda8ec1ab9c63b94c47c3450f87503e3286.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1f322a7eb108c21d3ce2c47bf58e67d02bde33481bcbbf177cd9da0de3fbc91b +size 4322 diff --git a/parse/train/HJgK0h4Ywr/images/d3739417837722647e7f66042569522f5f4c5aad468dddbabbf86d84cca84ad8.jpg b/parse/train/HJgK0h4Ywr/images/d3739417837722647e7f66042569522f5f4c5aad468dddbabbf86d84cca84ad8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fd7d3aa6a52616c88792dfeb4cce7f1c348adaee --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/d3739417837722647e7f66042569522f5f4c5aad468dddbabbf86d84cca84ad8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:516cdb42fd60b29f0c1156f3fe686a75eb247da5c879e6316f904e08731b119e +size 37027 diff --git a/parse/train/HJgK0h4Ywr/images/d5b2c58435ba51a4726918919c2206ff39b552d37b1a6e883a2ec676a574ccfa.jpg b/parse/train/HJgK0h4Ywr/images/d5b2c58435ba51a4726918919c2206ff39b552d37b1a6e883a2ec676a574ccfa.jpg new file mode 100644 index 0000000000000000000000000000000000000000..db7b7e613850262671fd72a2d9eeb4c222223b6a --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/d5b2c58435ba51a4726918919c2206ff39b552d37b1a6e883a2ec676a574ccfa.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0a0f7696c8ecd9a94cf036069231fe6b4a8509e0f93a507f3921f5a2b1ab3212 +size 91243 diff --git a/parse/train/HJgK0h4Ywr/images/d5fb1490ce9b60cfef79adff8fe848608ed56e5df02a8a1e628c93a0f7ef57da.jpg b/parse/train/HJgK0h4Ywr/images/d5fb1490ce9b60cfef79adff8fe848608ed56e5df02a8a1e628c93a0f7ef57da.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9d4f08c1ec2b8057f783609f5b513dac71c17cbd --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/d5fb1490ce9b60cfef79adff8fe848608ed56e5df02a8a1e628c93a0f7ef57da.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:333ddbe37561b02396a87615f71815c66da8e94f57bbb7fdb550010be82d6a11 +size 6789 diff --git a/parse/train/HJgK0h4Ywr/images/d83dd64d96063af6cbbec0cacd7a2b9fbc870e7cb4c1a074d01b5bf7c17692d5.jpg b/parse/train/HJgK0h4Ywr/images/d83dd64d96063af6cbbec0cacd7a2b9fbc870e7cb4c1a074d01b5bf7c17692d5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7faed46fea33e8829f59a5e6ee19cb06bc0f6f6b --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/d83dd64d96063af6cbbec0cacd7a2b9fbc870e7cb4c1a074d01b5bf7c17692d5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:45a8d55f878fd405bd2f86404e72e66dab32be8330d1afebfe64dd20d060d1eb +size 10973 diff --git a/parse/train/HJgK0h4Ywr/images/dd28d992b5817c98bea452e60b471745abe56d8db24fa02761193c7b314b4907.jpg b/parse/train/HJgK0h4Ywr/images/dd28d992b5817c98bea452e60b471745abe56d8db24fa02761193c7b314b4907.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b67baceb403dd588d0d2b1946ea3aa759c9ce70e --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/dd28d992b5817c98bea452e60b471745abe56d8db24fa02761193c7b314b4907.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:38a42caa9b218fe2eba1d5bc1697267b9e0800ff9931aadfa3b5426d27e3b209 +size 3980 diff --git a/parse/train/HJgK0h4Ywr/images/dd3a41a850518f19100d44b9083c3c6c1d0b288c926c6ebafa74d388d1a445d4.jpg b/parse/train/HJgK0h4Ywr/images/dd3a41a850518f19100d44b9083c3c6c1d0b288c926c6ebafa74d388d1a445d4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8a5ea2c5a7a0abd85e96f9a7dfdfcc7c13eb94c7 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/dd3a41a850518f19100d44b9083c3c6c1d0b288c926c6ebafa74d388d1a445d4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9c974b09f665067df8ce41df1448caf6144624173d675c990e9707ec6fe0bdcc +size 37034 diff --git a/parse/train/HJgK0h4Ywr/images/e3da3edab6200b7e2e6dfd36f409dca37389ce7bd08c433f0a65c110376076e2.jpg b/parse/train/HJgK0h4Ywr/images/e3da3edab6200b7e2e6dfd36f409dca37389ce7bd08c433f0a65c110376076e2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bc2250d31e12263e505bf3522e28f788d4251dc9 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/e3da3edab6200b7e2e6dfd36f409dca37389ce7bd08c433f0a65c110376076e2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fa0fbfd184980cb113548796ef8ebd3feedaed556e9e6cf155d4bce18968e80c +size 19377 diff --git a/parse/train/HJgK0h4Ywr/images/e40b8ab0bd8def787c5680fc1980cde332c139441925d1f30c567103cedb46e1.jpg b/parse/train/HJgK0h4Ywr/images/e40b8ab0bd8def787c5680fc1980cde332c139441925d1f30c567103cedb46e1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2f3e73c18bdc8ae7c91f3bc310f37acc9c1e7d40 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/e40b8ab0bd8def787c5680fc1980cde332c139441925d1f30c567103cedb46e1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4c116551a044c02c3abe97d697050541790205150dd3e724b935c9f89d7db7db +size 53912 diff --git a/parse/train/HJgK0h4Ywr/images/e967ee171891c93e9f92f86b354c9e6a2928c921dcc9d0ad99abece0552c6158.jpg b/parse/train/HJgK0h4Ywr/images/e967ee171891c93e9f92f86b354c9e6a2928c921dcc9d0ad99abece0552c6158.jpg new file mode 100644 index 0000000000000000000000000000000000000000..18c510c409bb60b4dbefd1fab284d7333a8858ad --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/e967ee171891c93e9f92f86b354c9e6a2928c921dcc9d0ad99abece0552c6158.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b2de53d83eaf074d9e240a1e44e660b68126fe16b9f82780cd2dc8b0ac731647 +size 43906 diff --git a/parse/train/HJgK0h4Ywr/images/f3e86dd5d1d114e9c6e641865138f0b21a85d0020d22087c6dda6f48fd0af584.jpg b/parse/train/HJgK0h4Ywr/images/f3e86dd5d1d114e9c6e641865138f0b21a85d0020d22087c6dda6f48fd0af584.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6de2797fe2c33fb99e63f7175642c6af118bdee9 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/f3e86dd5d1d114e9c6e641865138f0b21a85d0020d22087c6dda6f48fd0af584.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1a023b99746e8fae4aed680f69e4ff59cb58b0e688b1c4994e533b449cd800fb +size 8076 diff --git a/parse/train/HJgK0h4Ywr/images/fe12ac6b165a0a827c1af1691916b9dae56e4bb8083df497ac353aeb3ebfb639.jpg b/parse/train/HJgK0h4Ywr/images/fe12ac6b165a0a827c1af1691916b9dae56e4bb8083df497ac353aeb3ebfb639.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2f1bcdc6ae620e2d17f6b3dbc052ad5259a37558 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/fe12ac6b165a0a827c1af1691916b9dae56e4bb8083df497ac353aeb3ebfb639.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a8be98fc692fb939b69675f3b315563cd7caf5ffc5615228b2b2a53b77915110 +size 14554 diff --git a/parse/train/HJgK0h4Ywr/images/fea7812cb2bcec55475f1401faab96cd7d705091304b0fe2ae02b3aca225d716.jpg b/parse/train/HJgK0h4Ywr/images/fea7812cb2bcec55475f1401faab96cd7d705091304b0fe2ae02b3aca225d716.jpg new file mode 100644 index 0000000000000000000000000000000000000000..31b65b1f509ab0a83d01e6bd39c83b0ff9f6ddc9 --- /dev/null +++ b/parse/train/HJgK0h4Ywr/images/fea7812cb2bcec55475f1401faab96cd7d705091304b0fe2ae02b3aca225d716.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:525dbd271367fe26d318d3919c763d9792cd135045dd0fb4780d381c1fe762d8 +size 85462 diff --git a/parse/train/HJlNpoA5YQ/images/00d9d60a782825d7dcbb6529c38dddbbc517d1cb07dbd137d1b7b8bec1bd1671.jpg b/parse/train/HJlNpoA5YQ/images/00d9d60a782825d7dcbb6529c38dddbbc517d1cb07dbd137d1b7b8bec1bd1671.jpg new file mode 100644 index 0000000000000000000000000000000000000000..013dad289bd1e4193761ffd7f1d2535af1758b8e --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/00d9d60a782825d7dcbb6529c38dddbbc517d1cb07dbd137d1b7b8bec1bd1671.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:15f0cfd48a4cdcd8bbf8e183737150ba1f35624829638facac090f2a3a6dfb0c +size 11738 diff --git a/parse/train/HJlNpoA5YQ/images/03647ebae88427ee1fae5c52957e8d364f51506e485cf7566722f97d240f6ee2.jpg b/parse/train/HJlNpoA5YQ/images/03647ebae88427ee1fae5c52957e8d364f51506e485cf7566722f97d240f6ee2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8b968ed1640b00067cafde4e369abcb549025865 --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/03647ebae88427ee1fae5c52957e8d364f51506e485cf7566722f97d240f6ee2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:78e3691c7e3637af9c8a0ad1f8bfe537dcdaaee2916d9810f52b216b8b686525 +size 47906 diff --git a/parse/train/HJlNpoA5YQ/images/0e74a3bb37b7e0c357bc5c8744415c861e7620376b7d800b33efb19e72ea7858.jpg b/parse/train/HJlNpoA5YQ/images/0e74a3bb37b7e0c357bc5c8744415c861e7620376b7d800b33efb19e72ea7858.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c35c613e812eee57a725a32c6d0105e45f8b4687 --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/0e74a3bb37b7e0c357bc5c8744415c861e7620376b7d800b33efb19e72ea7858.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:02d7917679f72603a4043df6739d8559e071c6f33c372ed35de536efe1985026 +size 11751 diff --git a/parse/train/HJlNpoA5YQ/images/16a86539d4d085524d77f3d181f1963ece00a8765615262f0983e45debc166f2.jpg b/parse/train/HJlNpoA5YQ/images/16a86539d4d085524d77f3d181f1963ece00a8765615262f0983e45debc166f2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b20a5bcbb5dfec4755aa5f2eeab5e95526a0cb96 --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/16a86539d4d085524d77f3d181f1963ece00a8765615262f0983e45debc166f2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:22051ac9abd4e0799a76e07f9e7ac0e444ceaa6b1e78810ba02d226cb4ca8032 +size 25848 diff --git a/parse/train/HJlNpoA5YQ/images/18f5f8d021a1e35a61d1987a02e6452b241dd3f72c1042381688fc247436157b.jpg b/parse/train/HJlNpoA5YQ/images/18f5f8d021a1e35a61d1987a02e6452b241dd3f72c1042381688fc247436157b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..411389ce0e24b7c10f6604db9a5fce690db1f79f --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/18f5f8d021a1e35a61d1987a02e6452b241dd3f72c1042381688fc247436157b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c0cda8de6d86d7215bde37cece3699270d9b16e8319498a1d3a93ab71512f610 +size 60488 diff --git a/parse/train/HJlNpoA5YQ/images/1bd56f42b09756030aa6785847dce94ec1ce055b5e9a1d73a0a86fe583d5e36a.jpg b/parse/train/HJlNpoA5YQ/images/1bd56f42b09756030aa6785847dce94ec1ce055b5e9a1d73a0a86fe583d5e36a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..10fa115b3a596e469c823358cd24a7c474046e5a --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/1bd56f42b09756030aa6785847dce94ec1ce055b5e9a1d73a0a86fe583d5e36a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:22c2a282d4d111e7675e67ac0820a62b3b15b5829efd9e68e5cf77495775e5bd +size 8591 diff --git a/parse/train/HJlNpoA5YQ/images/217eb1982d095ab8e2958c1f7c8c811a60cd5ef8d1acbaf4f454cb7a725d2b8c.jpg b/parse/train/HJlNpoA5YQ/images/217eb1982d095ab8e2958c1f7c8c811a60cd5ef8d1acbaf4f454cb7a725d2b8c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b0408bbdb27aae57d7eec54ee0412c63dd768473 --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/217eb1982d095ab8e2958c1f7c8c811a60cd5ef8d1acbaf4f454cb7a725d2b8c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:db2e77eb781cbabb53a6bb56ac49a2167346dc244dc2160e52ae8414a2d3055f +size 8427 diff --git a/parse/train/HJlNpoA5YQ/images/3003d13a2d2a50d4ebaf2b706d80fbe3548995d3bec04a56393f6f41a16bb4ba.jpg b/parse/train/HJlNpoA5YQ/images/3003d13a2d2a50d4ebaf2b706d80fbe3548995d3bec04a56393f6f41a16bb4ba.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9813e8a62afd981163a7571347d7da8b5261e704 --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/3003d13a2d2a50d4ebaf2b706d80fbe3548995d3bec04a56393f6f41a16bb4ba.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f6618a27946c5e9e1c96d5056a1766a231e673a6bfa1b71a79c3cc0814e9fed1 +size 8285 diff --git a/parse/train/HJlNpoA5YQ/images/3c407de6b9bf8ea9db6fa8a29f2a2bad3877847e05622cdb3024aca15d767062.jpg b/parse/train/HJlNpoA5YQ/images/3c407de6b9bf8ea9db6fa8a29f2a2bad3877847e05622cdb3024aca15d767062.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3378aa93c31d60000e54805d5cad8a9ba41fdad3 --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/3c407de6b9bf8ea9db6fa8a29f2a2bad3877847e05622cdb3024aca15d767062.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1370967b673860f09a6bf83e515aff0c7dfae477f11ad7c905c26d38650a4753 +size 19965 diff --git a/parse/train/HJlNpoA5YQ/images/3dad8e3bbd65d973163be73e053ed8d73f71a5e9e959ebfb1f80749cf45e3fbc.jpg b/parse/train/HJlNpoA5YQ/images/3dad8e3bbd65d973163be73e053ed8d73f71a5e9e959ebfb1f80749cf45e3fbc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..478d53d6670bdcd507bf2db9b7c5b37fc7c2fa6c --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/3dad8e3bbd65d973163be73e053ed8d73f71a5e9e959ebfb1f80749cf45e3fbc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1bd43d3b3d29830592253206d4ae32aadbfc9b1fde3998dc1f1e25ca1317ea0f +size 25690 diff --git a/parse/train/HJlNpoA5YQ/images/44e779fe9151c92697a4b88aaa10af9a96534542f2019273da2e8172688af5c3.jpg b/parse/train/HJlNpoA5YQ/images/44e779fe9151c92697a4b88aaa10af9a96534542f2019273da2e8172688af5c3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..76e48e572284ae62024ea27f2295b27f49ff3c2e --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/44e779fe9151c92697a4b88aaa10af9a96534542f2019273da2e8172688af5c3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7a2560407fb2fbb11ae692b5c33068d50a9d8a5c1fb2aff6f59a31bdb9b9821a +size 14205 diff --git a/parse/train/HJlNpoA5YQ/images/46049dcadff383607ed824cefdeed075ac32849b227858955474e64b49a578d1.jpg b/parse/train/HJlNpoA5YQ/images/46049dcadff383607ed824cefdeed075ac32849b227858955474e64b49a578d1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7d90c90855216ed6450688014fc2d44624957503 --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/46049dcadff383607ed824cefdeed075ac32849b227858955474e64b49a578d1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3643219ced082f00ecc04965e7c6c44bc6cc446e6e3408ddfcfeb9726115c76d +size 6426 diff --git a/parse/train/HJlNpoA5YQ/images/4b87087c926e30e60944a446221ed96515d4d4a582692da1fbce0ede62bb223c.jpg b/parse/train/HJlNpoA5YQ/images/4b87087c926e30e60944a446221ed96515d4d4a582692da1fbce0ede62bb223c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8571975aa4f599bf21bfd25b24d4591d642abb70 --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/4b87087c926e30e60944a446221ed96515d4d4a582692da1fbce0ede62bb223c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c4b60d290be14f8dc86586dc123435162e1acba497aa56aac1b9219f2bba9bb7 +size 80611 diff --git a/parse/train/HJlNpoA5YQ/images/54239f5eade6f469b0ad09db52818bfae4c3cea802a8af332851dfdaa0f002fd.jpg b/parse/train/HJlNpoA5YQ/images/54239f5eade6f469b0ad09db52818bfae4c3cea802a8af332851dfdaa0f002fd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1ac0be1d7c4fb46a7c66588241f7515383b67264 --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/54239f5eade6f469b0ad09db52818bfae4c3cea802a8af332851dfdaa0f002fd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b1a769b11e646650fdb5492a4dc596cb45f91c1bf23ef8bc20c457daa9615f76 +size 24866 diff --git a/parse/train/HJlNpoA5YQ/images/6d439cc29b5ec6a7c661ffaee468c0df8779b54bb0ea8f8f163a30515775aadd.jpg b/parse/train/HJlNpoA5YQ/images/6d439cc29b5ec6a7c661ffaee468c0df8779b54bb0ea8f8f163a30515775aadd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f0c1fafa41fa2cd7df4391974a02eea469d4cdfe --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/6d439cc29b5ec6a7c661ffaee468c0df8779b54bb0ea8f8f163a30515775aadd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:86783fe9f1be49469a620ce55c4d23b4e351500aaece5787aedbaa3bdc032c64 +size 5193 diff --git a/parse/train/HJlNpoA5YQ/images/71305745afe273bef9dba34bfd812c624b02a8df4968eca2e0ed0ad6ee0fc168.jpg b/parse/train/HJlNpoA5YQ/images/71305745afe273bef9dba34bfd812c624b02a8df4968eca2e0ed0ad6ee0fc168.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fd6ad748fd1c96f345996ba4c928c137a5b0dda5 --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/71305745afe273bef9dba34bfd812c624b02a8df4968eca2e0ed0ad6ee0fc168.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e2bd95a2f4067a2593c0a7fe296efea475249839db264b716339a805ded5c907 +size 4621 diff --git a/parse/train/HJlNpoA5YQ/images/72f66f491cc8fc63aa30146871b2e8a6ef324060109aab4e067de5f141fc0ab9.jpg b/parse/train/HJlNpoA5YQ/images/72f66f491cc8fc63aa30146871b2e8a6ef324060109aab4e067de5f141fc0ab9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c47ef4787ccce93e7df8f75902da978654312f5d --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/72f66f491cc8fc63aa30146871b2e8a6ef324060109aab4e067de5f141fc0ab9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e2fef7b911441972eedcdde465a325450815e34846edf9435e0fbac27653e1c0 +size 4721 diff --git a/parse/train/HJlNpoA5YQ/images/7a0600115125e4c0f1bff626331de76122d8fc107608124445632c9151154d89.jpg b/parse/train/HJlNpoA5YQ/images/7a0600115125e4c0f1bff626331de76122d8fc107608124445632c9151154d89.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f31f13d1dac5010be0723a2c2b9b3adfc083e900 --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/7a0600115125e4c0f1bff626331de76122d8fc107608124445632c9151154d89.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:33a7ffb98d304abf6adc4d49f7437167fdf915f5c650f14723be24ee167b26ed +size 4385 diff --git a/parse/train/HJlNpoA5YQ/images/867ab7e43d6d59b8a874412d68c56a26c7ca278152613aeee65b5b52f3a22aa7.jpg b/parse/train/HJlNpoA5YQ/images/867ab7e43d6d59b8a874412d68c56a26c7ca278152613aeee65b5b52f3a22aa7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7fccfc2cf83b0248728c1e7aa3d9cf4bf7b6ed3f --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/867ab7e43d6d59b8a874412d68c56a26c7ca278152613aeee65b5b52f3a22aa7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a90e58f75af26a32f9ffc20fda99e8df2e4861f28f715c0b4e7a28b007d946c3 +size 35514 diff --git a/parse/train/HJlNpoA5YQ/images/8de7f3392b800ea490a6c7e1bca2083e02c02fce13fac39c1d8bd8fc93ea8b6f.jpg b/parse/train/HJlNpoA5YQ/images/8de7f3392b800ea490a6c7e1bca2083e02c02fce13fac39c1d8bd8fc93ea8b6f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8043644dad3c5536fd64d0336c5f9b91973ce38b --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/8de7f3392b800ea490a6c7e1bca2083e02c02fce13fac39c1d8bd8fc93ea8b6f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1e4e5f6b79c175da21bd07c227d82ab6851c93b1a4518a90e1e550bc6b618443 +size 8189 diff --git a/parse/train/HJlNpoA5YQ/images/938e726dc9ff511cf1f5451e3f56ec7939fa19bfe1a420e4934c626b29151710.jpg b/parse/train/HJlNpoA5YQ/images/938e726dc9ff511cf1f5451e3f56ec7939fa19bfe1a420e4934c626b29151710.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0d95eb3c1be6f079d105c40f8fdc7451530808fa --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/938e726dc9ff511cf1f5451e3f56ec7939fa19bfe1a420e4934c626b29151710.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1463913d12a9784632c4178df080a195ee1773374b6186a1d6417a097c157d5b +size 8373 diff --git a/parse/train/HJlNpoA5YQ/images/9da9a923f71e32475e4a5f37aa05a2b39c29e12f1458bd43a20e33b148377f9a.jpg b/parse/train/HJlNpoA5YQ/images/9da9a923f71e32475e4a5f37aa05a2b39c29e12f1458bd43a20e33b148377f9a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..890232c6f1b81e5737c292485e696175e2d94ac3 --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/9da9a923f71e32475e4a5f37aa05a2b39c29e12f1458bd43a20e33b148377f9a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1f58c45f3cc6546816e9f8dfbbdb8cd176b223926219ea0b7a4edb3fee8122df +size 16769 diff --git a/parse/train/HJlNpoA5YQ/images/e3cb58a6d51eb4bb4eb36e27c034cb9f9df79a9a07f3daedada5677d38873def.jpg b/parse/train/HJlNpoA5YQ/images/e3cb58a6d51eb4bb4eb36e27c034cb9f9df79a9a07f3daedada5677d38873def.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5e206bd8fc4eefdcf1eaf4fa9ab82958e33f5351 --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/e3cb58a6d51eb4bb4eb36e27c034cb9f9df79a9a07f3daedada5677d38873def.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:071127ae2118200801186e64351ec89a5a2fbf8b281385b610c1141e38e0c494 +size 65090 diff --git a/parse/train/HJlNpoA5YQ/images/f360a467546dec966d4365ad238b31f3b6c053c5d8ae3227faae93f2abaaf189.jpg b/parse/train/HJlNpoA5YQ/images/f360a467546dec966d4365ad238b31f3b6c053c5d8ae3227faae93f2abaaf189.jpg new file mode 100644 index 0000000000000000000000000000000000000000..549ab3db440b95740983b703d3d968f9194c036e --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/f360a467546dec966d4365ad238b31f3b6c053c5d8ae3227faae93f2abaaf189.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c3da89c395be03853665f4e28ad766f394f02ac0b2e0f5e206b91f71a4613f18 +size 10114 diff --git a/parse/train/HJlNpoA5YQ/images/f453de5a5fe28013cc32fb70f664abb37c22f88e370edadcab610e1bd0aea217.jpg b/parse/train/HJlNpoA5YQ/images/f453de5a5fe28013cc32fb70f664abb37c22f88e370edadcab610e1bd0aea217.jpg new file mode 100644 index 0000000000000000000000000000000000000000..026c25be8f47162a3e8862f3d3e9e1e99baaf463 --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/f453de5a5fe28013cc32fb70f664abb37c22f88e370edadcab610e1bd0aea217.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cec74142b0535ed48d31c7596b635ea4768a3f8f41a14b77436d351df2c105a5 +size 11978 diff --git a/parse/train/HJlNpoA5YQ/images/f4543058c43d30cd26d95bf7b48cd90c7967939caeb6f78120fe4752a50cb726.jpg b/parse/train/HJlNpoA5YQ/images/f4543058c43d30cd26d95bf7b48cd90c7967939caeb6f78120fe4752a50cb726.jpg new file mode 100644 index 0000000000000000000000000000000000000000..98816d65ae6d03e0c7fd19132e7a78442ec5db09 --- /dev/null +++ b/parse/train/HJlNpoA5YQ/images/f4543058c43d30cd26d95bf7b48cd90c7967939caeb6f78120fe4752a50cb726.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ea7500968718485eacd699c057b9cf3653b5d42a2c66efa81a891649ba1237f5 +size 7255 diff --git a/parse/train/HyGySsAct7/images/0c186411e689d3f9078d269f5673bb099d4a54d7f7766152c3f7edb521c5496f.jpg b/parse/train/HyGySsAct7/images/0c186411e689d3f9078d269f5673bb099d4a54d7f7766152c3f7edb521c5496f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7a548b0f588deeab8ae6b912cd27584ee149fbbb --- /dev/null +++ b/parse/train/HyGySsAct7/images/0c186411e689d3f9078d269f5673bb099d4a54d7f7766152c3f7edb521c5496f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e40d31acb511769b52bb9b18740ee18caa8832f7f5e90cf9138f776b78ac9be3 +size 7143 diff --git a/parse/train/HyGySsAct7/images/2059e59de774fa523712cc2081fa1bf0391f4f59372c42dbbd7ed4e23abdcb3b.jpg b/parse/train/HyGySsAct7/images/2059e59de774fa523712cc2081fa1bf0391f4f59372c42dbbd7ed4e23abdcb3b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9cab330faf51c81651891ef3d31624b902a4de6f --- /dev/null +++ b/parse/train/HyGySsAct7/images/2059e59de774fa523712cc2081fa1bf0391f4f59372c42dbbd7ed4e23abdcb3b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1808a6db0acd6ff354da79de6fc4d94aab81e81e21c12bb03c7b159d49a352a5 +size 19876 diff --git a/parse/train/HyGySsAct7/images/4a9f86fa45f8d6ae01b9e4d9fa9c535d790f43224090e2ead706a3c784d56609.jpg b/parse/train/HyGySsAct7/images/4a9f86fa45f8d6ae01b9e4d9fa9c535d790f43224090e2ead706a3c784d56609.jpg new file mode 100644 index 0000000000000000000000000000000000000000..226bd8955e88ed112bfd5cbd1353bcf7c66aab9e --- /dev/null +++ b/parse/train/HyGySsAct7/images/4a9f86fa45f8d6ae01b9e4d9fa9c535d790f43224090e2ead706a3c784d56609.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:92081a5b417d173d9d63765698617bf4af35da321788c7c92ffd51f6c878c4c0 +size 126529 diff --git a/parse/train/HyGySsAct7/images/656910746ad0201bc30ede92cf518b0aad14fb183b75004c9a113d0dac6a7194.jpg b/parse/train/HyGySsAct7/images/656910746ad0201bc30ede92cf518b0aad14fb183b75004c9a113d0dac6a7194.jpg new file mode 100644 index 0000000000000000000000000000000000000000..68d5fffff56161845420bbce43f989b027532786 --- /dev/null +++ b/parse/train/HyGySsAct7/images/656910746ad0201bc30ede92cf518b0aad14fb183b75004c9a113d0dac6a7194.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4b779ef59fc8fc4aa44c37a413c119bc914a0bd21aeb98075c9f17b61126cfac +size 14890 diff --git a/parse/train/HyGySsAct7/images/6fb46dae308f8b2a2cc5d3ba1937ddca82b4ce42badf393284f45b8a330d3b62.jpg b/parse/train/HyGySsAct7/images/6fb46dae308f8b2a2cc5d3ba1937ddca82b4ce42badf393284f45b8a330d3b62.jpg new file mode 100644 index 0000000000000000000000000000000000000000..af5039002192bbf55012f51bb854b678bf86bb96 --- /dev/null +++ b/parse/train/HyGySsAct7/images/6fb46dae308f8b2a2cc5d3ba1937ddca82b4ce42badf393284f45b8a330d3b62.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8fb3722ef08fa99e1891a3cc6bba18120f0bb122d43ab702f2ae9bfad7fc4b47 +size 5301 diff --git a/parse/train/HyGySsAct7/images/702a79e13def8c6dba6e31287f0a41fee5a20c11dbb53617509b9193ada57cd8.jpg b/parse/train/HyGySsAct7/images/702a79e13def8c6dba6e31287f0a41fee5a20c11dbb53617509b9193ada57cd8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..21f8760f1337dbbe6d988d908223372edb43010a --- /dev/null +++ b/parse/train/HyGySsAct7/images/702a79e13def8c6dba6e31287f0a41fee5a20c11dbb53617509b9193ada57cd8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4cddf8babec69135553d11cfa8431499cbb48e1537106ecda3ad1d5cf4c50428 +size 9915 diff --git a/parse/train/HyGySsAct7/images/73371a04dc01d589dedd50963db02395c3c1c2164f92a9725fef55d67d3744ff.jpg b/parse/train/HyGySsAct7/images/73371a04dc01d589dedd50963db02395c3c1c2164f92a9725fef55d67d3744ff.jpg new file mode 100644 index 0000000000000000000000000000000000000000..af544c7f78e5c48bbbc1405f38046fb8015f5e83 --- /dev/null +++ b/parse/train/HyGySsAct7/images/73371a04dc01d589dedd50963db02395c3c1c2164f92a9725fef55d67d3744ff.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a281ed474874657067a5d22ba51e0d3182d35a3ae719c3b4861fa63988d44a2b +size 14147 diff --git a/parse/train/HyGySsAct7/images/80048329b24ecc6daea8dd9975d7ef756926647b32dcd1da0388d76937134845.jpg b/parse/train/HyGySsAct7/images/80048329b24ecc6daea8dd9975d7ef756926647b32dcd1da0388d76937134845.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dae37863943862938066f9d13084a0c96897bd89 --- /dev/null +++ b/parse/train/HyGySsAct7/images/80048329b24ecc6daea8dd9975d7ef756926647b32dcd1da0388d76937134845.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:de146769d32c10ea8935e20a54d38e5761e19a81ca26b8e281be8873a454f788 +size 8593 diff --git a/parse/train/HyGySsAct7/images/8c6122a88db2dcefe101a33d3619580a3eac07f621d6ba72f356cdd967b32445.jpg b/parse/train/HyGySsAct7/images/8c6122a88db2dcefe101a33d3619580a3eac07f621d6ba72f356cdd967b32445.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7200e811f9b0f2d473b6a365fb01232695c65d32 --- /dev/null +++ b/parse/train/HyGySsAct7/images/8c6122a88db2dcefe101a33d3619580a3eac07f621d6ba72f356cdd967b32445.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7531c715c17430682ba0e33e2eed24326b12293d43734ae1a1f94ea9e32a7868 +size 31389 diff --git a/parse/train/HyGySsAct7/images/9d7976e786dc0d63013b39a37ae469f4b92ff09f0c7a7bbf063e382fcb5cc71d.jpg b/parse/train/HyGySsAct7/images/9d7976e786dc0d63013b39a37ae469f4b92ff09f0c7a7bbf063e382fcb5cc71d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1edf3c49ce33bd543de08f352b1b4e80a56e0092 --- /dev/null +++ b/parse/train/HyGySsAct7/images/9d7976e786dc0d63013b39a37ae469f4b92ff09f0c7a7bbf063e382fcb5cc71d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ac3d288509db73b67797aad8192bc0c28d53c4a061661946f02c942c6622873f +size 38610 diff --git a/parse/train/HyGySsAct7/images/d14bf5a129c3fd17d74d1e72d9ac3af24b1909a1d0705d938ec3ea5122f0bcd8.jpg b/parse/train/HyGySsAct7/images/d14bf5a129c3fd17d74d1e72d9ac3af24b1909a1d0705d938ec3ea5122f0bcd8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d2fe6d337f30d27a09c8a6702663246a02127e3d --- /dev/null +++ b/parse/train/HyGySsAct7/images/d14bf5a129c3fd17d74d1e72d9ac3af24b1909a1d0705d938ec3ea5122f0bcd8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2b8c2d81c48c3b171f59990c6a728e6891cd9c48afc2e3f0d1147c5d07d530fd +size 15859 diff --git a/parse/train/HyGySsAct7/images/e8d0b09def98fd41837b2eed3da420825b7ea1dd39e0f5de32688811363e6393.jpg b/parse/train/HyGySsAct7/images/e8d0b09def98fd41837b2eed3da420825b7ea1dd39e0f5de32688811363e6393.jpg new file mode 100644 index 0000000000000000000000000000000000000000..02a578b9314f1304c025476dea9affa929243be6 --- /dev/null +++ b/parse/train/HyGySsAct7/images/e8d0b09def98fd41837b2eed3da420825b7ea1dd39e0f5de32688811363e6393.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8cb67bf0ea680cda480b6263172bcc2e5444eec2bdd5df806101ad6d8abb2168 +size 58929 diff --git a/parse/train/HyGySsAct7/images/ee6ffc52bace3a4c44f95198b3c050ae97a740e1b816076a94e00330c6a8be92.jpg b/parse/train/HyGySsAct7/images/ee6ffc52bace3a4c44f95198b3c050ae97a740e1b816076a94e00330c6a8be92.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9f4d43435606554c2244c1764d03a49942145c6f --- /dev/null +++ b/parse/train/HyGySsAct7/images/ee6ffc52bace3a4c44f95198b3c050ae97a740e1b816076a94e00330c6a8be92.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:db2f556f0bbafee4ab8403fb0bbec71e3e49376caa58f2c2bf5b134ee922cdc0 +size 7096 diff --git a/parse/train/J28lNO4p3ki/J28lNO4p3ki.md b/parse/train/J28lNO4p3ki/J28lNO4p3ki.md new file mode 100644 index 0000000000000000000000000000000000000000..aa83c6a55d7e7e7b270f63bb314779819a87a90b --- /dev/null +++ b/parse/train/J28lNO4p3ki/J28lNO4p3ki.md @@ -0,0 +1,294 @@ +# STEM: A Stochastic Two-Sided Momentum Algorithm Achieving Near-Optimal Sample and Communication Complexities for Federated Learning + +Prashant Khanduri University of Minnesota khand095@umn.edu + +Pranay Sharma Carnegie Mellon University pranaysh@andrew.cmu.edu + +Haibo Yang The Ohio State University yang.5952@buckeyemail.osu.edu + +Mingyi Hong⇤ University of Minnesota mhong@umn.edu + +Jia Liu The Ohio State University liu@ece.osu.edu + +Ketan Rajawat Indian Institute of Technology Kanpur ketan@iitk.ac.in + +Pramod K. Varshney Syracuse University varshney@syr.edu + +# Abstract + +Federated Learning (FL) refers to the paradigm where multiple worker nodes (WNs) build a joint model by using local data. Despite extensive research, for a generic non-convex FL problem, it is not clear, how to choose the WNs’ and the server’s update directions, the minibatch sizes, and the number of local updates, so that the WNs use the minimum number of samples and communication rounds to achieve the desired solution. This work addresses the above question and considers a class of stochastic algorithms where the WNs perform a few local updates before communication. We show that when both the WN’s and the server’s directions are chosen based on certain stochastic momentum estimator, the algorithm requires $\tilde { \mathcal O } ( \epsilon ^ { - 3 / 2 } )$ samples and $\tilde { \mathcal { O } } ( \epsilon ^ { - 1 } )$ communication rounds to compute an $\epsilon$ -stationary solution. To the best of our knowledge, this is the first FL algorithm that achieves such near-optimal sample and communication complexities simultaneously. Further, we show that there is a trade-off curve between the number of local updates and the minibatch sizes, on which the above sample and communication complexities can be maintained. Finally, we show that for the classical FedAvg (a.k.a. Local SGD, which is a momentum-less special case of the STEM), a similar trade-off curve exists, albeit with worse sample and communication complexities. Our insights on this trade-off provides guidelines for choosing the four important design elements for FL algorithms, the number of local updates, WNs’ and server’s update directions, and minibatch sizes to achieve the best performance. + +# 1 Introduction + +In Federated Learning (FL), multiple worker nodes (WNs) collaborate with the goal of learning a joint model, by only using local data. Therefore it has become popular for machine learning problems where datasets are massively distributed [1]. In FL, the data is often collected at or off-loaded to multiple WNs which in collaboration with a server node (SN) jointly aim to learn a centralized model [2, 3]. The local WNs share the computational load and since the data is local to each WN, FL also provides some level of data privacy $\mathbb { \lVert \rVert }$ . A classical distributed optimization problem that $K$ WNs aim to solve: + +![](images/8b109c70673d05d4fb4863253d56729a5c2b37b6a73924b30a5e660d7b6d1078.jpg) +Figure 1: The 3D surface in (a) plots the communication complexity of the proposed STEM for different minibatch sizes and number of local updates. The surface is generated such that each point represents STEM with a particular choice of $( b , I )$ , so that it requires $\tilde { \mathcal { O } } ( \epsilon ^ { - 3 / 2 } )$ samples to achieve $\epsilon$ -stationarity. Plot (b) shows the optimal trade off between the minibatch sizes and the number of local updates at each WN (i.e., achieving the lowest communication and sample complexities). Both plots are generated for an accuracy of $\epsilon = 1 0 ^ { - 3 }$ and all the constants dependent on system parameters (variance of stochastic gradients, heterogeneity parameter, optimality gap, Lipschitz constants, etc.) are assumed to be 1. Fed STEM is a special case of STEM where $\mathcal { O } ( 1 )$ minibatch is used; Minibatch STEM is a special case of STEM where $\mathcal { O } ( 1 )$ local updates are used. + +$$ +\operatorname* { m i n } _ { x \in \mathbb { R } ^ { d } } \bigg \{ f ( x ) : = \frac { 1 } { K } \sum _ { k = 1 } ^ { K } f ^ { ( k ) } ( x ) : = \frac { 1 } { K } \sum _ { k = 1 } ^ { K } \mathbb { E } _ { \xi ^ { ( k ) } \sim \mathcal { D } ^ { ( k ) } } \big [ f ^ { ( k ) } ( x ; \xi ^ { ( k ) } ) \big ] \bigg \} . +$$ + +where $f ^ { ( k ) } : \mathbb { R } ^ { d } \mathbb { R }$ denotes the smooth (possibly non-convex) objective function and $\xi ^ { ( k ) } \sim \mathcal { D } ^ { ( k ) }$ represents the sample/s drawn from distribution $\mathcal { D } ^ { ( k ) }$ at the $k ^ { \mathrm { { t h } } }$ WN with $k \in [ K ]$ . When the distributions $\mathcal { D } ^ { ( k ) }$ are different across the WNs, it is referred to as the heterogeneous data setting. + +The optimization performance of non-convex FL algorithms is typically measured by the total number of samples accessed (cf. Definition $2 . 2 )$ and the total rounds of communication (cf. Definition $2 . 3 )$ required by each WN to achieve an $\epsilon$ -stationary solution (cf. Definition $2 . 1 )$ . To minimize the sample and the communication complexities, FL algorithms rely on the following four key design elements: (i) the WNs’ local model update directions, (ii) Minibatch size to compute each local direction, (iii) the number of local updates before WNs share their parameters, and (iv) the SN’s update direction. How to find effective FL algorithms by (optimally) designing these parameters has received significant research interest recently. + +Contributions. The main contributions of this work are listed below: + +1) We propose the Stochastic Two-Sided Momentum (STEM) algorithm, that utilizes certain momentum-assisted stochastic gradient directions for both the WNs and SN updates. We show that there exists an optimal trade off between the minibatch sizes and number of local updates, such that on the trade-off curve STEM requires $\underline { { \tilde { \mathcal { O } } } } ( \epsilon ^ { - 3 / 2 } ) \underline { { \left[ \frac { 2 } { } \right] } }$ samples and $\tilde { \mathcal { O } } ( \epsilon ^ { - 1 } )$ communication rounds to reach an $\epsilon$ -stationary solution; see Figure $\bigstar$ for an illustration. These complexity results are the best achievable for first-order stochastic FL algorithms (under certain assumptions, cf. Assumption $\bigstar \bigstar$ ; see $\pm \boxed { 5 } \boxed { 8 } \parallel$ and $\mathbb { B } \mathbb { n o }$ , as well as Remark $\checkmark$ of this paper for discussions regarding optimality. To the best of our knowledge, STEM is the first algorithm which – (i) simultaneously achieves the optimal sample and communication complexities for FL and (ii) can optimally trade off the minibatch sizes and the number of local updates. + +2) A momentum-less special case of our STEM result further reveals some interesting insights of the classical FedAvg algorithm (a.k.a. the Local SGD) [11–13]. Specifically, we show that for FedAvg, there also exists a trade-off between the minibatch sizes and the number of local updates, such that it requires $\mathcal { O } ( \epsilon ^ { - 2 } )$ samples and $\mathcal { O } ( \epsilon ^ { - 3 / 2 } )$ communication rounds to achieve an $\epsilon$ -stationary solution. + +
AlgorithmWorkSampleComm.Minibatch (b)Local Updates (I) /round
FedAvg国园 国国0(€-2)0(c-3/2) 0(c-2)0(1) 0(1) 2(1-v)0(c-1/2) 0(1) 3v
SCAFFOLD*this work 国0(c-2)O(e-3/2) 0(c-2)O(c 4-v) 0(1)O(c−2(4-D)) 0(1)
FedPD/FedProx*四/□O(c-2)0(e-1)0(1)0(e-1)
MIME†/FedGLOMO/80(c-3/2)O(€-3/2)0(1)0(1)
STEM Fed STEM Minibatch STEM* this workO(€-3/2)O(e-1)( 0(1) O(e-1/2)O(∈−(3)) O(∈-1/2) 0(1)
+ +Table 1: Comparison of FedAvg and STEM with different FL algorithms for various choices of the minibatch sizes $( b )$ and the number of per node local updates between two rounds of communication $( I )$ . $^ \circ \nu \in [ 0 , 1 ]$ trades off $^ { b }$ and $I$ ; $\nu = 1$ (resp. $\nu = 0$ ) uses multiple (resp. $\mathcal { O } ( 1 ) .$ ) local updates and $\mathcal { O } ( 1 )$ (resp. multiple) samples. Fed STEM and Minibatch STEM are two variants of the proposed STEM. ‡The data heterogeneity assumption is weaker than Assumption $2$ (please see $\bigstar \bigstar$ for details). †Requires bounded Hessian dissimilarity to model data heterogeneity across WNs. ⇤Guarantees for Minibatch STEM with $I = 1$ and SCAFFOLD are independent of the data heterogeneity. + +Collectively, our insights on the trade-offs provide practical guidelines for choosing different design elements for FL algorithms. + +Related Works. FL algorithms were first proposed in the form of FedAvg [11], where the local update directions at each WN were chosen to be the SGD updates. Earlier works analyzed these algorithms in the homogeneous data setting [19–25], while many recent studies have focused on designing new algorithms to deal with heterogeneous data settings, as well as problems where the local loss functions are non-convex [9, 10, 12–16, 18, 26–32]. In $\bar { \mathbb { E } 2 } \mathbb { I }$ , the authors showed that Parallel Restarted SGD (Local SGD or FedAvg [11]) achieves linear speed up while requiring $\mathcal { O } ( \epsilon ^ { - 2 } )$ samples and $\mathcal { O } ( \epsilon ^ { - 3 / 2 } )$ rounds of communication to reach an $\epsilon$ -stationary solution. In $[ \bar { \lVert { 4 } } ]$ , a Momentum SGD was proposed, which achieved the same sample and communication complexities as Parallel Restarted SGD $\mathbb { \lVert 1 2 \rVert }$ , without requiring that the second moments of the gradients be bounded. Further, it was shown that under the homogeneous data setting, the communication complexity can be improved to $\mathcal { O } ( \epsilon ^ { - 1 } )$ while maintaining the same sample complexity. The works in $\boxed { 1 5 } , \boxed { 1 6 }$ conducted tighter analysis for FedAvg with partial WN participation with $\mathcal { O } ( 1 )$ local updates and batch sizes. Their analysis showed that FedAvg’s sample and communication complexities are both $\mathcal { O } ( \epsilon ^ { - 2 } )$ . Additionally, SCAFFOLD was proposed in $\bar { \| 1 5 \| }$ , which utilized variance reduction based local update directions $\pmb { \mathbb { B 3 } }$ to achieve the same sample and communication complexities as FedAvg. Similarly, VRL-SGD proposed in $\left[ \left[ 2 9 \right] \right]$ also utilized variance reduction and showed improved communication complexity of $\mathcal { O } ( \epsilon ^ { - 1 } )$ , while requiring the same computations as FedAvg. Importantly, both SCAFFOLD and VRL-SGD’s guarantees were independent of the data heterogeneity. The FedProx proposed in $\mathbb { \ m }$ used a penalty based method to improve the communication complexity of FedAvg (i.e., the Parallel Restarted and Momentum SGD [14, $\mathbb { L } 2 \mathbb { I }$ ) to $\mathcal { O } ( \epsilon ^ { - 1 } )$ . FedProx used a gradient similarity assumption to model data heterogeneity which can be stringent for many practical applications. This assumption was relaxed by FedPD proposed in $\pmb { \mathbb { B } }$ . + +Recently, the works [17, 18] proposed to utilize hybrid momentum gradient estimators [7, 8]. The MIME algorithm $\mathbb { \lVert 1 7 \rVert }$ matched the optimal sample complexity (under certain smoothness assumptions) of $\mathcal { O } ( \epsilon ^ { - 3 / 2 } )$ of the centralized non-convex stochastic optimization algorithms [5–8]. Similarly, Fed-GLOMO $[ \overline { { 1 8 } } ]$ achieved the same sample complexity while employing compression to further reduce communication. Both MIME and Fed-GLOMO required $\mathcal { O } ( \epsilon ^ { - 3 / 2 } )$ communication rounds to achieve an $\epsilon$ -stationary solution. Please see Table $^ 1$ for a summary of the above discussion. + +The comparison of Local SGD (FedAvg) to Minibatch SGD for convex and strongly convex problems with homogeneous data setting was first conducted in $\mathbb { \lVert 1 9 \rVert }$ and later extended to heterogeneous setting in $\mathbb { \lVert \rVert 3 \rVert }$ . It was shown that Minibatch SGD almost always dominates the Local SGD. In contrast, it was shown in $\pmb { \Vert 2 4 \Vert }$ that Local SGD dominates Minibatch SGD in terms of generalization performance. Although existing FL results are rich, but they are somehow ad hoc and there is a lack of principled understanding of the algorithms. We note that the proposed STEM algorithmic framework provides a theoretical framework that unifies all existing $\mathrm { F L }$ results on sample and communication complexities. + +Notations. The expected value of a random variable $X$ is denoted by $\mathbb { E } [ X ]$ and its expectation conditioned on an Event $A$ is denoted as $\mathbb { E } [ X | \mathrm { E v e n t ~ } A ]$ . We denote by $\mathbb { R }$ (and $\mathbb { R } ^ { d }$ ) the real line (and the $d$ -dimensional Euclidean space). The set of natural numbers is denoted by $\mathbb { N }$ . Given a positive integer $K \in \mathbb N$ , we denote $[ K ] \triangleq \{ 1 , 2 , \dots , K \}$ . Notation $\| \cdot \|$ denotes the $\ell _ { 2 }$ -norm and $\langle \cdot , \cdot \rangle$ the Euclidean inner product. For a discrete set $\boldsymbol { B }$ , $| B |$ denotes the cardinality of the set. Uniform distribution over a discrete set $\{ 1 , \ldots , T \}$ is denoted as ${ \dot { \mathcal { U } } } \{ 1 , \dots , T \}$ . + +# 2 Preliminaries + +Before we proceed to the algorithms, we make the following assumptions about problem $( 1 )$ . + +Assumption 1 (Sample Gradient Lipschitz Smoothness). The stochastic functions $f ^ { ( k ) } ( \cdot , \xi ^ { ( k ) } )$ with $\xi ^ { ( k ) } \sim \mathcal { D } ^ { ( k ) }$ for all $k \in [ K ]$ , satisfy the mean squared smoothness property, i.e, we have + +$$ +\begin{array} { r } { \mathbb { E } \| \nabla f ^ { ( k ) } ( x ; \xi ^ { ( k ) } ) - \nabla f ^ { ( k ) } ( y ; \xi ^ { ( k ) } ) \| ^ { 2 } \leq L ^ { 2 } \| x - y \| ^ { 2 } \mathrm { ~ f o r ~ a l l ~ } x , y \in \mathbb { R } ^ { d } . } \end{array} +$$ + +Assumption 2 (Unbiased gradient and Variance Bounds). (i) Unbiased Gradient. The stochastic gradients computed at each WN are unbiased + +$$ +\mathbb { E } [ \nabla f ^ { ( k ) } ( x ; \xi ^ { ( k ) } ) ] = \nabla f ^ { ( k ) } ( x ) , \forall \xi ^ { ( k ) } \sim \mathcal { D } ^ { ( k ) } , \forall k \in [ K ] . +$$ + +(ii) Intra- and inter- node Variance Bound. The following bounds hold: + +$$ +\begin{array} { r } { \mathbb { \tilde { z } } \| \nabla f ^ { ( k ) } ( x ; \xi ^ { ( k ) } ) - \nabla f ^ { ( k ) } ( x ) \| ^ { 2 } \leq \sigma ^ { 2 } , \| \nabla f ^ { ( k ) } ( x ) - \nabla f ^ { ( \ell ) } ( x ) \| ^ { 2 } \leq \zeta ^ { 2 } , \forall \xi ^ { ( k ) } \sim \mathcal { D } ^ { ( k ) } , \forall k , \ell \in [ K ] . } \end{array} +$$ + +Note that Assumption $\boxed { 1 }$ is stronger than directly assuming $f ^ { ( k ) }$ ’s are Lipschitz smooth (which we will refer to as the averaged gradient Lipschitz smooth condition), but it is still a rather standard assumption in SGD analysis. For example it has been used in analyzing centralized SGD algorithms such as SPIDER $ { \mathbb { I } }$ , SNVRG $\pmb { \Vert 6 \Vert }$ , STORM $\mathbb { [ [ \big ] ] }$ (and many others) as well as in FL algorithms such as MIME $ { \mathbb { I } } ^ { [ 1 2 ] }$ and Fed-GLOMO $ { \mathbb { I } } { \mathrm { 1 8 } } { \Vert }$ . The second relation in Assumption $2 \cdot$ (ii) quantifies the data heterogeneity, and we call $\zeta > 0$ as the heterogeneity parameter. This is a typical assumption required to evaluate the performance of FL algorithms. If data distributions across individual WNs are identical, i.e., $\mathcal { D } ^ { ( k ) } \stackrel { = } { = } \mathcal { D } ^ { ( \ell ) }$ for all $k , \ell \in [ K ]$ then we have $\zeta = 0$ . + +Next, we define the $\epsilon$ -stationary solution for non-convex optimization problems, as well as quantify the computation and communication complexities to achieve an $\epsilon$ -stationary point. + +Definition 2.1 $\epsilon$ -Stationary Point). A point $x$ is called $\epsilon$ -stationary if $\| \nabla f ( x ) \| ^ { 2 } \leq \epsilon$ . Moreover, a stochastic algorithm is said to achieve an $\epsilon$ -stationary point in $t$ iterations if $\begin{array} { r } { \ddot { \mathbb { E } } [ \| \nabla f ( x _ { t } ) \| ^ { 2 } ] \le \epsilon . } \end{array}$ where the expectation is over the stochasticity of the algorithm until time instant $t$ . + +Definition 2.2 (Sample complexity). We assume an Incremental First-order Oracle (IFO) framework $\textcircled { \lVert { 3 4 } \rVert }$ where, given a sample $\xi ^ { ( k ) } \sim \mathcal { D } ^ { ( k ) }$ at the $k ^ { \mathrm { { t h } } }$ node and iterate $x$ , the oracle returns $( f ^ { ( k ) } ( x ; \xi ^ { ( k ) } ) , \nabla f ^ { ( k ) } ( x ; \xi ^ { ( k ) } ) )$ . Each access to the oracle is counted as a single IFO operation. We measure the sample (and computational) complexity in terms of the total number of calls to the IFO by all WNs to achieve an $\epsilon$ -stationary point given in Definition 2.1. + +Definition 2.3 (Communication complexity). We define a communication round as a one back-andforth sharing of parameters between the WNs and the SN. Then the communication complexity is defined to be the total number of communication rounds between any WN and the SN required to achieve an $\epsilon$ -stationary point given in Definition 2.1. + +# 3 The STEM algorithm and the trade-off analysis + +In this section, we discuss the proposed algorithm and present the main results. The key in the algorithm design is to carefully balance all the four design elements mentioned in Sec. $^ { 1 , }$ so that sufficient and useful progress can be made between two rounds of communication. + +Let us discuss the key steps of STEM, listed in Algorithm $^ { 1 . }$ In Step 10, each node locally updates its model parameters using the local direction $d _ { t } ^ { k }$ , computed by using $b$ stochastic gradients at two + +1: Input: Parameters: $c > 0$ , the number of local updates $I$ , batch size $b$ , stepsizes $\{ \eta _ { t } \}$ . +2: Initialize: Iterate $\begin{array} { r } { x _ { 1 } ^ { ( k ) } = \bar { x } _ { 1 } = \frac { 1 } { K } \sum _ { k = 1 } ^ { K } x _ { 1 } ^ { ( k ) } } \end{array}$ , descent direction $\begin{array} { r } { d _ { 1 } ^ { ( k ) } = \bar { d } _ { 1 } = \frac { 1 } { K } \sum _ { k = 1 } ^ { K } d _ { 1 } ^ { ( k ) } } \end{array}$ +with $\begin{array} { r } { d _ { 1 } ^ { ( k ) } = \frac { 1 } { B } \sum _ { \xi _ { 1 } ^ { ( k ) } \in \mathcal { B } _ { 1 } ^ { ( k ) } } \nabla f ^ { ( k ) } ( x _ { 1 } ^ { ( k ) } ; \xi _ { 1 } ^ { ( k ) } ) } \end{array}$ and $| B _ { 1 } ^ { ( k ) } | = B$ for $k \in [ K ]$ . +3: Perform: $x _ { 2 } ^ { ( k ) } = x _ { 1 } ^ { k } - \eta _ { 1 } d _ { 1 } ^ { ( k ) }$ , $\forall k \in [ K ]$ +4: for $t = 1$ to $T$ do +5: for $k = 1$ to $K$ do #at the WN +6: $\mathcal { d } _ { t + 1 } ^ { ( k ) } = \frac { 1 } { b } \sum _ { \xi _ { t + 1 } ^ { ( k ) } \in \mathcal { B } _ { t + 1 } ^ { ( k ) } } ^ { \pi \Delta } \nabla f ^ { ( k ) } ( x _ { t + 1 } ^ { ( k ) } ; \xi _ { t + 1 } ^ { ( k ) } ) + \left( 1 - a _ { t + 1 } \right) \bigg ( d _ { t } ^ { ( k ) } - \frac { 1 } { b } \sum _ { \xi _ { t + 1 } ^ { ( k ) } \in \mathcal { B } _ { t + 1 } ^ { ( k ) } } \nabla f ^ { ( k ) } ( x _ { t } ^ { ( k ) } ; \xi _ { t + 1 } ^ { ( k ) } ) \bigg )$ +where we choose $| B _ { t + 1 } ^ { ( k ) } | = b$ , and $a _ { t + 1 } = c \cdot \eta _ { t } ^ { 2 }$ ; +7: 8: if $t$ $I = 0$ #at the SN +$\begin{array} { r } { d _ { t + 1 } ^ { ( k ) } = \bar { d } _ { t + 1 } : = \frac { 1 } { K } \sum _ { k = 1 } ^ { K } d _ { t + 1 } ^ { ( k ) } } \end{array}$ +9: 10: e $\begin{array} { r l } & { x _ { t + 2 } ^ { ( k ) ^ { \bot } } : = \bar { x } _ { t + 1 } - \eta _ { t + 1 } \bar { d } _ { t + 1 } = \frac { 1 } { K } \sum _ { k = 1 } ^ { K } x _ { t + 1 } ^ { ( k ) } - \eta _ { t + 1 } \bar { d } _ { t + 1 } } \end{array}$ $x _ { t + 2 } ^ { ( k ) } = x _ { t + 1 } ^ { ( k ) } - \eta _ { t + 1 } d _ { t + 1 } ^ { ( k ) }$ #server-side momentum +11: end if +12: end for +13: end for +14: Return: $\scriptstyle { \bar { x } } _ { a }$ where $a \sim \mathcal { U } \{ 1 , . . . , T \}$ . + +consecutive iterates $x _ { t + 1 } ^ { ( k ) }$ and $x _ { t } ^ { ( k ) }$ . After every $I$ local steps, the WNs share their current local models $\{ x _ { t + 1 } ^ { ( k ) } \} _ { k = 1 } ^ { K }$ and directions $\{ d _ { t + 1 } ^ { ( k ) } \} _ { k = 1 } ^ { K }$ with the SN. The SN aggregates these quantities, and performs a server-side momentum step, before returning $\bar { x } _ { t + 1 }$ and $\bar { d } _ { t + 1 }$ to all the WNs. Because both the WNs and the SN perform momentum based updates, we call the algorithm a stochastic two-sided momentum algorithm. The key parameters are: $b$ the minibatch size, $I$ the local update steps between two communication rounds, $\eta _ { t }$ the stepsizes, and $a _ { t }$ the momentum parameters. + +One key technical innovation of our algorithm design is to identify the most suitable way to incorporate momentum based directions in FL algorithms. Although the momentum-based gradient estimator itself is not new and has been used in the literature before (see e.g., in $\mathbb { \left[ \bigstar \bigstar \right] }$ and $\overline { { \mathbb { D } \mathbb { Z } \mathbb { G } \mathbb { S } } }$ to improve the sample complexities of centralized and decentralized stochastic optimization problems, respectively), it is by no means clear if and how it can contribute to improve the communication complexity of FL algorithms. We show that in the FL setting, the local directions together with the local models have to be aggregated by the SN so to avoid being influenced too much by the local data. More importantly, besides the WNs, the SN also needs to perform updates using the (aggregated) momentum directions. Finally, such two-sided momentum updates have to be done carefully with the correct choice of minibatch size $b$ , and the number of local updates $I$ . Overall, it is the judicious choice of all these design elements that results in the optimal sample and communication complexities. + +Next, we present the convergence guarantees of the STEM algorithm. + +# 3.1 Main results: convergence guarantees for STEM + +In this section, we analyze the performance of STEM. We first present our main result, and then provide discussions about a few parameter choices. In the next subsection, we discuss a special case of STEM related to the classical FedAvg and minibatch SGD algorithms. + +Theorem 3.1. Under the Assumptions 1 and 2, suppose the stepsize sequence is chosen as: + +$$ +\eta _ { t } = \frac { \bar { \kappa } } { ( w _ { t } + \sigma ^ { 2 } t ) ^ { 1 / 3 } } , +$$ + +where we define : + +$$ +\bar { \kappa } = \frac { ( b K ) ^ { 2 / 3 } \sigma ^ { 2 / 3 } } { L } , \quad w _ { t } = \operatorname * { m a x } \bigg \{ 2 \sigma ^ { 2 } , 4 0 9 6 L ^ { 3 } I ^ { 3 } \bar { \kappa } ^ { 3 } - \sigma ^ { 2 } t , \frac { c ^ { 3 } \bar { \kappa } ^ { 3 } } { 4 0 9 6 L ^ { 3 } I ^ { 3 } } \bigg \} . +$$ + +Further, let us set $\begin{array} { r } { c = \frac { 6 4 L ^ { 2 } } { b K } + \frac { \sigma ^ { 2 } } { 2 4 \bar { \kappa } ^ { 3 } L I } = L ^ { 2 } \bigg ( \frac { 6 4 } { b K } + \frac { 1 } { 2 4 ( b K ) ^ { 2 } I } \bigg ) } \end{array}$ and set the initial batch size as $B = b I$ ; set the local updates $I$ and minibatch size b as follows: + +$$ +I = \mathcal { O } \big ( ( T / K ^ { 2 } ) ^ { \nu / 3 } \big ) , \quad b = \mathcal { O } \big ( ( T / K ^ { 2 } ) ^ { 1 / 2 - \nu / 2 } \big ) +$$ + +where $\nu$ satisfies $\nu \in [ 0 , 1 ]$ . Then for STEM the following holds: + +(i) For $\scriptstyle { \bar { x } } _ { a }$ chosen according to Algorithm $\boldsymbol { l } ,$ we have: + +$$ +\mathbb { E } \Vert \nabla f ( \bar { x } _ { a } ) \Vert ^ { 2 } = \mathcal { O } \Bigg ( \frac { f ( \bar { x } _ { 1 } ) - f ^ { * } } { K ^ { 2 \nu / 3 } T ^ { 1 - \nu / 3 } } \Bigg ) + \tilde { \mathcal { O } } \Bigg ( \frac { \sigma ^ { 2 } } { K ^ { 2 \nu / 3 } T ^ { 1 - \nu / 3 } } \Bigg ) + \tilde { \mathcal { O } } \Bigg ( \frac { \zeta ^ { 2 } } { K ^ { 2 \nu / 3 } T ^ { 1 - \nu / 3 } } \Bigg ) . +$$ + +(ii) For any $\nu \in [ 0 , 1 ]$ , we have + +Sample Complexity: The sample complexity of STEM is $\tilde { \mathcal { O } } ( \epsilon ^ { - 3 / 2 } )$ . This implies that each WN requires at most $\tilde { \mathcal { O } } ( K ^ { - 1 } \epsilon ^ { - 3 / 2 } )$ gradient computations, thereby achieving linear speedup with the number of WNs present in the network. + +Communication Complexity: The communication complexity of STEM is $\tilde { \mathcal { O } } ( \epsilon ^ { - 1 } )$ . + +The proof of this result is relegated to the Supplemental Material. A few remarks are in order. + +Remark 1 (Near-Optimal sample and communication complexities). Theorem $3 . 1$ suggests that when $I$ and $b$ are selected appropriately, then STEM achieves $\tilde { \mathcal { O } } ( \epsilon ^ { - 3 / 2 } )$ and $\tilde { \mathcal { O } } ( \overline { { \epsilon ^ { - 1 } } } )$ sample and communication complexities. Taking them separately, these complexity bounds are the best achievable by the existing FL algorithms (upto logarithmic factors regardless of sample or batch Lipschitz smooth assumption) $[ [ 3 5 ] ]$ ; see Table $1 .$ We note that the $\mathcal { O } ( \epsilon ^ { - 3 / 2 } )$ complexity is the best possible that can be achieved by centralized SGD with the sample Lipschitz gradient assumption; see $ { \mathbb { I } } ^ { { \left[ 5 \right] } }$ . On the other hand, the $\bar { \mathcal { O } } ( \epsilon ^ { - 1 } )$ complexity bound is also likely to be the optimal, since in $\mathbb { P }$ the authors showed that even when the local steps use a class of (deterministic) first-order algorithms, $\mathcal { O } ( \epsilon ^ { - 1 } )$ is the best achievable communication complexity. The only difference is that $\bigstar \bigstar$ does not explicitly assume the inter-node variance bound (i.e., the second relation in Assumption $2 \cdot$ -(ii)). We leave the precise characterization of the communication lower bound with inter-node variance as future work. □ + +Remark 2 (Large Batch Sizes and/or Local Updates). At first glance, it may seem that the requirement of STEM to compute large mini-batches and/or local updates (cf. Table $^ { 1 ) }$ to achieve this (near) optimal performance is a drawback, however, we note that it is in fact an advantage of STEM that it allows the WNs to perform larger number of local updates (or compute large minibatches) without communicating often. This follows from the fact that irrespective of the number of local updates (or batch sizes) STEM achieves near-optimal communication complexity while attaining optimal overall sample complexity. Moreover, note that even with $b = I = \mathcal { O } ( 1 )$ (i.e., $b$ and $I$ are chosen as constants), STEM achieves the same (optimal) sample and communication complexities as achieved by FedGLOMO [18] and MIME [17]. We further note that to the best of our knowledge the algorithms that achieve the communication complexity of $\mathcal { O } ( \epsilon ^ { - 1 } )$ either require the number of local updates or the batch-sizes that depend on the solution accuracy $\epsilon$ . For example, FedProx $\mathbb { I O } ]$ , FedPD $\bigstar \bigstar$ , and FedDyn $\pmb { \Vert 3 6 \Vert }$ rely on solving the “local problems" to achieve an $\epsilon$ -accuracy, which implies that the number of local updates (or the batch sizes) implicitly depends on the desired solution accuracy $\epsilon$ , as is the case for STEM. Similarly, as shown in $\bar { \mathbb { E } 2 } \mathbb { I }$ and $\bar { \textregistered 4 } \bar { 1 }$ the communication complexity of FedAvg and its momentum version can be improved from $\mathcal { O } ( \epsilon ^ { - 2 } )$ to $\mathcal { O } ( \epsilon ^ { - 3 / 2 } )$ when the number of local updates (or batch size) is chosen as $\mathcal { O } ( \epsilon ^ { - 1 / 2 } )$ (cf. Section $3 . 2$ for a more detailed discussion). + +Remark 3 (The Optimal Batch Sizes and Local Updates Trade-off). The parameter $\nu \in [ 0 , 1 ]$ is used to balance the local minibatch sizes $b$ , and the number of local updates $I$ . Eqs. in $\textcircled { 3 }$ suggest that when $\nu$ increases from 0 to 1, $b$ decreases and $I$ increases. Specifically, if $\nu = 1$ , then $b$ is a constant but $I = \mathcal { O } ( T ^ { 1 / 3 } / K ^ { 2 / 3 } )$ . In this case, each WN chooses a small minibatch while executing multiple local updates, and STEM resembles a FedAvg (a.k.a. Local SGD) algorithm but with double-sided momentum update directions, and is referred to as Fed STEM. In contrast, if $\nu = 0$ , then $b = \mathcal { O } ( T ^ { 1 / 2 } / K )$ but $I$ is a constant. In this case, each WN chooses a large batch size while executing only a few, or even one, local updates, and STEM resembles the Minibatch SGD, but again with different update directions, and is referred to as Minibatch STEM. Such a trade-off can be seen in Fig. 1b. Due to space limitation, these two special cases will be precisely stated in the supplementary materials as corollaries of Theorem 3.1. □ + +# Algorithm 2 The FedAvg Algorithm + +1: Input: $\{ \eta _ { t } \} _ { t = 0 } ^ { T } ; I$ , the # of local updates per communication round; $b$ , the minibatch sizes. +2: for $t = 1$ to $T$ do +3: 4: 5: for $\begin{array} { r l } & { \mathcal { \kappa } _ { t } ^ { = } \stackrel { \mathrm { ~ L ~ U ~ O ~ } \Lambda } { = } \mathbf { 0 } } \\ & { d _ { t } ^ { ( k ) } = \frac { 1 } { b } \sum _ { \xi _ { t } ^ { ( k ) } \in \mathcal { B } _ { t } ^ { ( k ) } } \nabla f ^ { ( k ) } ( x _ { t } ^ { ( k ) } ; \xi _ { t } ^ { ( k ) } ) \mathrm { ~ w i t h ~ } | \mathcal { B } _ { t } ^ { ( k ) } | = b } \\ & { x _ { t + 1 } ^ { ( k ) } = x _ { t } ^ { ( k ) } - \eta _ { t } d _ { t } ^ { ( k ) } } \\ & { \mathbf { i f } t \operatorname* { m o d } I = 0 \mathbf { \Lambda } \mathbf { t h e n } } \\ & { ~ x _ { t + 1 } ^ { ( k ) } = \bar { x } _ { t + 1 } = \frac { 1 } { K } \sum _ { k = 1 } ^ { K } x _ { t + 1 } ^ { ( k ) } } \\ & { \mathbf { e n d } \mathbf { \Phi } \mathbf { i f } } \end{array}$ $k = 1$ $K$ +6: +7: +8: +9: end for +10: end for +11: Return: $\scriptstyle { \bar { x } } _ { a }$ where $a \sim \mathcal { U } \{ 1 , . . . , T \}$ . + +Remark 4 (The Sub-Optimal Batch Sizes and Local Updates Trade-off). From our proof (Theorem $\left| \overline { { \mathbf { C . 1 0 } } } \right|$ included in the supplemental material), we can see that STEM requires $\bar { \tilde { O } } ( \operatorname* { m a x } \big \{ ( b \cdot$ $I ) \epsilon ^ { - 1 } , K ^ { - 1 } \epsilon ^ { - 3 / 2 } \rbrace )$ samples and $\tilde { \mathcal { O } } \big ( \operatorname* { m a x } \big \{ \epsilon ^ { - 1 } , ( b \cdot I ) ^ { - 1 } K ^ { - 1 } \epsilon ^ { - 3 / 2 } \big \} \big )$ and communication rounds. According to the above expressions, if $b \cdot I$ increases beyond $\mathcal { O } ( K ^ { - 1 } \epsilon ^ { - 1 / 2 } )$ , then the sample complexity will increase from the optimal $\tilde { \mathcal { O } } ( \epsilon ^ { - 3 / 2 } )$ ; otherwise, the optimal sample complexity $\tilde { \mathcal O } ( \epsilon ^ { - 3 / 2 } )$ is maintained. On the other hand, if $b \cdot I$ decreases beyond $\mathcal { O } ( K ^ { - 1 } \epsilon ^ { - 1 / 2 } )$ , the communication complexity increases from $\tilde { \mathcal { O } } ( \epsilon ^ { - 1 } )$ . For instance, if we choose $b = \mathcal { O } ( 1 )$ and $I = { \mathcal { O } } ( 1 )$ the communication complexity becomes $\tilde { \mathcal { O } } ( \epsilon ^ { - 3 / 2 } )$ while the optimal sample complexity $\tilde { \mathcal { O } } ( \epsilon ^ { - 3 / 2 } )$ is maintained. This trade-off is illustrated in Figure $\boxed { 1 \mathrm { a } }$ where we maintain the optimal sample complexity, while changing $b$ and $I$ to generate the trade-off surface. □ + +Remark 5 (Data Heterogeneity). The term $\begin{array} { r } { \tilde { \mathcal { O } } \biggl ( \frac { \zeta ^ { 2 } } { K ^ { 2 \nu / 3 } T ^ { 1 - \nu / 3 } } \biggr ) } \end{array}$ in the gradient bound $\textcircled{4}$ captures the effect of the heterogeneity of data across WNs, where $\zeta$ is the parameter characterizing the intra-node variance and has been defined in Assumption $\bigstar$ (ii). Highly heterogeneous data with large $\zeta ^ { 2 }$ can adversely impact the performance of STEM. Note that such a dependency on $\zeta$ also appears in other existing FL algorithms, such as $[ \bigcirc , \bigcirc , \bigcirc , \bigcirc , \bigcirc ]$ . However, there is one special case of STEM that does not depend on the parameter $\zeta$ . This is the case where $I = 1$ , i.e., the minibatch SGD counterpart of STEM where only a single local iteration is performed between two communication rounds. We have the following corollary. □ + +Corollary 1 (Minibatch STEM). Under Assumptions $\bigstar \bigstar \bigstar | \bigstar |$ , and choose the algorithm parameters as in Theorem $3 . I .$ At each WN, choose $I = 1$ , $b = ( T / K ^ { 2 } ) ^ { 1 / 2 }$ , and the initial batch size $B = \boldsymbol { b } \cdot \boldsymbol { I }$ . Then STEM satisfies: + +(i) For $\bar { x } _ { a }$ chosen according to Algorithm $\boldsymbol { l } ,$ we have + +$$ +\mathbb { E } \| \nabla f ( \bar { x } _ { a } ) \| ^ { 2 } = \mathcal { O } \Big ( \frac { f ( \bar { x } _ { 1 } ) - f ^ { * } } { T } \Big ) + \tilde { \mathcal { O } } \Big ( \frac { \sigma ^ { 2 } } { T } \Big ) . +$$ + +(ii) Minibatch STEM achieves $\tilde { \mathcal O } ( \epsilon ^ { - 3 / 2 } )$ sample and $\tilde { \mathcal { O } } ( \epsilon ^ { - 1 } )$ communication complexity. + +Next, we show that FedAvg also exhibits a trade-off similar to that of STEM but with worse sample and communication complexities. + +# 3.2 Special cases: The FedAvg algorithm + +We briefly discuss another interesting special case of STEM, where the local momentum update is replaced by the conventional SGD (i.e., $a _ { t } = 1 , ~ \forall ~ t )$ , while the server does not perform the momentum update (i.e., $\bar { d } _ { t } = 0 , \forall t )$ . This is essentially the classical FedAvg algorithm, just that it balances the number of local updates $I$ and the minibatch size $b$ . We show that this algorithm also exhibits a trade-off between $b$ and $I$ and on the trade-off curve it achieves $\mathcal { O } ( \epsilon ^ { - 2 } )$ sample complexity and $\mathcal { O } ( \epsilon ^ { - 3 / 2 } )$ communication complexity. + +
Algorithm Training Acc.Testing Acc.
FedAvg78.274.1
FedProx79.274.8
FedDyn68.966.0
SCAFFOLD71.974.0
MIME82.676.8
FedGLOMO76.172.8
STEM80.178.8
+ +(a) Mild heterogeneity, $b = 6 4$ , and $I = 7$ . + +
AlgorithmTraining Acc.Testing Acc.
FedAvg73.675.4
FedProx80.075.2
FedDyn76.171.3
SCAFFOLD72.573.7
MIME61.558.6
FedGLOMO10.010.0
STEM81.178.5
+ +(b) Moderate heterogeneity, $b = 8$ , and $I = 6 1$ + +Table 2: Training and testing accuracy of different algorithms on CIFAR-10 dataset for different batch-sizes, number of local updates, and heteregeneity settings. + +Theorem 3.2 (The FedAvg Algorithm). Under Assumptions 1 and 2, suppose the stepsize is chosen as: $\begin{array} { r } { \eta = \sqrt { \frac { b K } { T } } } \end{array}$ ; Let us set: + +$$ +I = \mathcal { O } \big ( ( T / K ^ { 3 } ) ^ { \nu / 4 } \big ) , \quad b = \mathcal { O } \big ( ( T / K ^ { 3 } ) ^ { 1 / 3 - \nu / 3 } \big ) +$$ + +where $\nu \in [ 0 , 1 ]$ is a constant. Then for FedAvg with $T \geq 8 1 L ^ { 2 } I ^ { 2 } b K$ , the following holds + +(i) For $\scriptstyle { \bar { x } } _ { a }$ chosen according to Algorithm $\perp$ we have + +$$ +\mathbb { E } \Vert \nabla f ( \bar { x } _ { a } ) \Vert ^ { 2 } = \mathcal { O } \Bigg ( \frac { f ( \bar { x } _ { 1 } ) - f ^ { * } } { K ^ { \nu / 2 } T ^ { 2 / 3 - \nu / 6 } } \Bigg ) + \mathcal { O } \Bigg ( \frac { \sigma ^ { 2 } } { K ^ { \nu / 2 } T ^ { 2 / 3 - \nu / 6 } } \Bigg ) + \mathcal { O } \Bigg ( \frac { \zeta ^ { 2 } } { K ^ { \nu / 2 } T ^ { 2 / 3 - \nu / 6 } } \Bigg ) . +$$ + +(ii) For any choice of $\nu \in [ 0 , 1 ]$ we have: + +Sample Complexity: The sample complexity of FedAvg is $\mathcal { O } ( \epsilon ^ { - 2 } )$ . This implies that each WN requires at most $\tilde { \mathcal { O } } ( K ^ { - 1 } \epsilon ^ { - 2 } )$ gradient computations, thereby achieving linear speedup with the number of WNs in the network. + +Communication Complexity: The communication complexity of FedAvg is $\mathcal { O } ( \epsilon ^ { - 3 / 2 } )$ . + +Note that the requirement on $T$ being lower bounded is only relevant for theoretical purposes, a similar requirement was also imposed in $[ \mathbb { 1 4 } ]$ to prove convergence. Again, the parameter $\nu \in [ 0 , 1 ]$ in the statement of Theorem $3 . 2$ balances $I$ and $b$ at each WN while maintaining state-of-the-art sample and communication complexities; please see Table $^ 1$ for a comparison of those bounds with existing FedAvg bounds. For $\nu = 1$ , FedAvg (cf. Theorem $\textcircled { 3 . 2 }$ reduces to FedAvg proposed in [12, 14] and for $\nu = 0$ , the algorithm can be viewed as a large batch FedAvg with constant local updates [15, 16]. Note that similar to STEM, it is known that for $I = 1$ , the Minibatch SGD’s performance is independent of the heterogeneity parameter, $\zeta \equiv \mathbb { I I } 3 \mathbb { I }$ . We also point out that if Algorithm $^ 1$ uses Nesterov’s or Polyak’s momentum $[ \textcircled { 1 4 } ]$ at local WNs instead of the recursive momentum estimator we get the same guarantees as in Theorem $3 . 2 .$ + +In summary, this section established that once the WN’s and the SN’s update directions (SGD in FedAvg and momentum based directions in STEM) are fixed, there exists a sequence of optimal choices of the number of local updates $I$ , and the batch sizes $b$ , which guarantees the best possible sample and communication complexities for the particular algorithm. The trade-off analysis presented in this section provides some useful guidelines for how to best select $b$ and $I$ in practice. Our subsequent numerical results will also verify that if $b$ or $I$ are not chosen judiciously, then the practical performance of the algorithms can degrade significantly. + +# 4 Numerical results + +In this section, we validate the proposed STEM algorithm and compare its performance with the de facto standard FedAvg [11], and the algorithms stated in Table $^ { 1 . }$ Note that instead of FedPD we include the performance comparison with FedDyn $\pmb { \mathbb { B } } 6 \|$ since they are known to be very closely related. The goal of our experiments are three-fold: (1) To show that STEM performs on par, if not better, compared to other algorithms in different heterogeneity settings, (2) there are multiple ways to reach the desired solution accuracy, one can either choose a large batch size and perform only a few local updates or select a smaller batch size and perform multiple local updates, and finally, (3) if the local updates and the batch sizes are not chosen appropriately, the WNs might need to perform excessive computations to achieve the desired solution accuracy, thereby slowing down convergence. + +Table 3: Training and testing accuracy on CIFAR-10 dataset for high heterogeneity, $b =$ 128 and $I = 6$ . + +
AlgorithmTraining Acc.Testing Acc.
FedAvg57.657.1
FedProx59.158.5
FedDyn51.251.3
SCAFFOLD53.154.7
MIME56.155.1
FedGLOMO56.856.1
STEM58.557.4
+ +Table 4: Training and testing accuracy on Shakespeare dataset. + +
AlgorithmTraining Acc.Testing Acc.
FedAvg40.139.2
FedProx43.543.2
FedDyn43.743.2
SCAFFOLD40.341.3
MIME32.132.1
FedGLOMO40.340.1
STEM44.543.8
+ +![](images/1a3649642909a08bbdd852d92b8ec00817216ffd484e8ea8263d27aa9fb977f4.jpg) +Figure 2: Training loss and the testing accuracy for classification on MNIST data set against the number of samples accessed at each WN for moderate heterogeneity setting with $b = 8$ . + +Data and Parameter Settings: We compare the algorithms for image classification tasks on CIFAR10 and MNIST data sets with 100 WNs, and for next character prediction task on Shakespeare dataset $\pmb { \Vert 3 7 } \Vert$ with 143 WNs in the network. For both CIFAR-10 and MNIST, each WN implements a two-hidden-layer convolutional neural network (CNN) architecture followed by three linear layers for CIFAR-10 and two for MNIST. For CIFAR-10 (and MNIST) datset, we consider three settings with mild, moderate and high heterogeneity. For all the three settings, the data is partitioned into disjoint sets among the WNs. In the mild heterogeneity setting, the WNs have access to partitioned data from all the classes. In the moderate (resp. high) heterogeneity setting the data is partitioned such that each WN can access data from only 5 (resp. 2) out of 10 classes. For CIFAR-10 (resp. MNIST), each WN has access to 490 (resp. 540) samples for training and 90 (resp. 80) samples for testing purposes. + +We also compare the performance of algorithms on a popular FL benchmarking dataset, Shakespeare dataset $ { \mathbb { I } } ^ { \smash { \sum } }$ . For this task, we adopt the settings from $\dot { \left[ \left| 1 0 \right| \right] }$ and utilize a 2-Layer LSTM network with 100 hidden units and an 8-D embedding layer at each WN. Each WN has access to 3616 samples on average, and the samples are randomly split into an $80 \%$ training set and a $20 \%$ testing set. We randomly sample 10 nodes out of 143 for the training purpose. All the experiments are implemented on a single NVIDIA Quadro RTX 5000 GPU. More details are provided in appendix. + +For the proposed STEM algorithm, recall that the step-size is $\eta _ { t } = \bar { \kappa } / ( w _ { t } + \sigma ^ { 2 } t ) ^ { 1 / 3 }$ with momentum parameter defined as $a _ { t } = { c } \eta _ { t } ^ { 2 }$ . The step-size is used to update the iterates while the momentum parameter is used to construct the stochastic gradient estimate (cf. Algorithm 1 and Theorem $\boxed { 3 . 1 }$ . For the experiments, we set $w _ { t } = \sigma ^ { 2 } = 1$ and $c \doteq \bar { c } / \bar { \kappa } ^ { 2 }$ and tune for $\bar { \kappa } \in [ 1 0 ^ { - 1 } , \overline { { 1 0 } } ^ { - 2 } ]$ for the CIFAR-10 dataset and for ${ \bar { \kappa } } \in \{ 1 0 ^ { 1 } , 1 0 ^ { 0 } , 1 0 ^ { - 1 } , 1 0 ^ { - 2 } \}$ for the Shakespeare dataset. For both the datasets we tune for $\bar { c }$ in the range [1, 10]. For FedProx $[ \equiv ]$ and FedDyn $\lVert \bar { 3 6 } \rVert$ we choose the regularization constant to be 0.1. The momentum parameters for FedGLOMO $\pm \textcircled { 1 8 } \textcircled { 1 }$ and MIME $\mathbb { \lVert 1 7 \rVert }$ are set based on the choices given in the respective papers. Specifically, for FedGLOMO we choose the parameter $\beta _ { k } = 0 . 2$ and design the momentum gradient using a damping factor given in Appendix A.4 of FedGLOMO $\mathbb { \left[ \left[ 8 \right] \right] }$ . Moreover, for MIME we choose the momentum parameter as 0.9. For the rest of the algorithms (including FedAvg and SCAFFOLD), the step-size is tuned from the set $\{ 1 0 ^ { 1 } , 1 0 ^ { 0 } , 1 0 ^ { - 1 } , \hat { 1 0 } ^ { - 2 } \}$ . + +Discussion: We evaluate the training and testing performance of STEM against multiple algorithms for different heterogeneity settings, minibatch sizes, and number of local updates. In Tables $\bigstar$ $\boxed { 2 \mathbf { b } }$ and $\bigstar ,$ we compare the training and testing accuracy of STEM to that of other algorithms on the CIFAR-10 dataset. Specific, heterogeneity settings, the choices of minibatches, and number of local updates are stated along with the tables. Note that STEM performs uniformly well under all the conditions. Moreover, note from Table $\bigstar$ that FedGLOMO diverges once the number of local updates are high. Also, note from Table $\textcircled { 3 }$ that FedProx and STEM adapt well to high heterogeneity. Finally, with the next set of experiments we emphasize the importance of choosing $b$ and $I$ carefully. In Figure $\bigstar ,$ we compare the training and testing performance of STEM, FedAvg and SCAFFOLD, against the number of samples accessed at each WN for the classification task on MNIST dataset with moderate heterogeneity. We fix $b = 8$ and conduct experiments under two settings, one with $I = 6 7$ , and the other with $I = 5 3 6$ local updates at each WN. Note that although a large number of local updates might lead to fewer communication rounds but it can make the sample complexity extremely high as is demonstrated by Figure $2 .$ For example, Figure $\bigtriangledown$ shows that to reach testing accuracy of $9 6 - 9 7 \%$ with $I = 6 7$ , STEM requires approximately $5 0 0 0 - 6 0 0 0$ samples, in contrast with $I = 5 3 6$ it requires more than 25000 samples at each WN. Similar behavior can be observed if we fix $I > 1$ and increase the local batch sizes. This implies not choosing the local updates and the batch sizes judiciously might lead to increased sample complexity. Additional experiments are included in the supplementary material to further evaluate the performance of the proposed algorithms. + +# Conclusion + +In this work, we proposed a novel algorithm STEM, for distributed stochastic non-convex optimization with applications to FL. We showed that STEM reaches an $\epsilon$ -stationary point with $\tilde { \mathcal O } ( \epsilon ^ { - 3 / 2 } )$ sample complexity while achieving linear speed-up with the number of WNs. Moreover, the algorithm achieves a communication complexity of $\tilde { \mathcal { O } } ( \epsilon ^ { - 1 } )$ . We established a (optimal) trade-off that allows interpolation between varying choices of local updates and the batch sizes at each WN while maintaining (near optimal) sample and communication complexities. We showed that FedAvg (a.k.a LocalSGD) also exhibits a similar trade-off while achieving worse complexities. Our results provide guidelines to carefully choose the number of local updates, update directions, and minibatch sizes to achieve the best performance. The future directions of this work include developing lower bounds on communication complexity that establishes the tightness of the analysis conducted in this work. + +# Acknowledgement + +We thank the anonymous reviewers for their valuable comments and suggestions. The work of Prashant Khanduri and Mingyi Hong was supported by NSF grant CMMI-1727757, AFOSR grant 19RT0424 and ARO grant W911NF-19-1-0247. The work of Mingyi Hong was also supported by an IBM Faculty Research award. The work of Jia Liu has been supported in part by NSF grants CAREER CNS-2110259, CNS-2112471, CNS-2102233, CCF-2110252, ECCS-2140277, and a Google Faculty Research Award. + +References +[1] J. Konecnˇ y, H. B. McMahan, D. Ramage, and P. Richtárik, “Federated optimization: Distributed \` machine learning for on-device intelligence,” arXiv preprint arXiv:1610.02527, 2016. +[2] M. Li, D. G. Andersen, A. J. Smola, and K. Yu, “Communication efficient distributed machine learning with the parameter server,” in Advances in Neural Information Processing Systems 27, Z. Ghahramani, M. Welling, C. Cortes, N. D. Lawrence, and K. Q. Weinberger, Eds. Curran Associates, Inc., 2014, pp. 19–27. +[3] J. Dean, G. Corrado, R. Monga, K. Chen, M. Devin, M. Mao, M. Ranzato, A. Senior, P. Tucker, K. Yang et al., “Large scale distributed deep networks,” in Advances in neural information processing systems, 2012, pp. 1223–1231. +[4] T. Léauté and B. Faltings, “Protecting privacy through distributed computation in multi-agent decision making,” Journal of Artificial Intelligence Research, vol. 47, pp. 649–695, 2013. +[5] C. Fang, C. J. Li, Z. Lin, and T. Zhang, “Spider: Near-optimal non-convex optimization via stochastic path-integrated differential estimator,” in Advances in Neural Information Processing Systems, 2018, pp. 689–699. +[6] D. Zhou, P. Xu, and Q. Gu, “Stochastic nested variance reduction for nonconvex optimization,” arXiv preprint arXiv:1806.07811, 2018. +[7] A. Cutkosky and F. Orabona, “Momentum-based variance reduction in non-convex SGD,” in Advances in Neural Information Processing Systems 32. Curran Associates, Inc., 2019, pp. 15 236–15 245. +[8] Q. Tran-Dinh, N. H. Pham, D. T. Phan, and L. M. Nguyen, “Hybrid stochastic gradient descent algorithms for stochastic nonconvex optimization,” arXiv preprint arXiv:1905.05920, 2019. +[9] X. Zhang, M. Hong, S. Dhople, W. Yin, and Y. Liu, “Fedpd: A federated learning framework with adaptivity to non-iid data,” IEEE Transactions on Signal Processing, pp. 1–1, 2021. +[10] T. Li, A. K. Sahu, M. Zaheer, M. Sanjabi, A. Talwalkar, and V. Smith, “Federated optimization in heterogeneous networks,” arXiv preprint arXiv:1812.06127, 2018. +[11] B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, “Communication-efficient learning of deep networks from decentralized data,” in Artificial Intelligence and Statistics. PMLR, 2017, pp. 1273–1282. +[12] H. Yu, S. Yang, and S. Zhu, “Parallel restarted sgd with faster convergence and less communication: Demystifying why model averaging works for deep learning,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 33, no. 01, 2019, pp. 5693–5700. +[13] B. Woodworth, K. K. Patel, and N. Srebro, “Minibatch vs local sgd for heterogeneous distributed learning,” arXiv preprint arXiv:2006.04735, 2020. +[14] H. Yu, R. Jin, and S. Yang, “On the linear speedup analysis of communication efficient momentum sgd for distributed non-convex optimization,” in International Conference on Machine Learning. PMLR, 2019, pp. 7184–7193. +[15] S. P. Karimireddy, S. Kale, M. Mohri, S. Reddi, S. Stich, and A. T. Suresh, “Scaffold: Stochastic controlled averaging for federated learning,” in International Conference on Machine Learning. PMLR, 2020, pp. 5132–5143. +[16] H. Yang, M. Fang, and J. Liu, “Achieving linear speedup with partial worker participation in non-iid federated learning,” arXiv preprint arXiv:2101.11203, 2021. +[17] S. P. Karimireddy, M. Jaggi, S. Kale, M. Mohri, S. J. Reddi, S. U. Stich, and A. T. Suresh, “Mime: Mimicking centralized stochastic algorithms in federated learning,” arXiv preprint arXiv:2008.03606, 2020. +[18] R. Das, A. Hashemi, S. Sanghavi, and I. S. Dhillon, “Improved convergence rates for non-convex federated learning with compression,” arXiv preprint arXiv:2012.04061, 2020. +[19] B. Woodworth, K. K. Patel, S. U. Stich, Z. Dai, B. Bullins, H. B. McMahan, O. Shamir, and N. Srebro, “Is local sgd better than minibatch sgd?” arXiv preprint arXiv:2002.07839, 2020. +[20] H. Yu and R. Jin, “On the computation and communication complexity of parallel sgd with dynamic batch sizes for stochastic non-convex optimization,” in International Conference on Machine Learning. PMLR, 2019, pp. 7174–7183. +[21] J. Wang and G. Joshi, “Cooperative sgd: A unified framework for the design and analysis of local-update sgd algorithms,” Journal of Machine Learning Research, vol. 22, no. 213, pp. 1–50, 2021. +[22] A. Khaled, K. Mishchenko, and P. Richtárik, “Better communication complexity for local sgd,” arXiv, 2019. +[23] S. U. Stich, “Local sgd converges fast and communicates little,” arXiv preprint arXiv:1805.09767, 2018. +[24] T. Lin, S. U. Stich, K. K. Patel, and M. Jaggi, “Don’t use large mini-batches, use local sgd,” in International Conference on Learning Representations, 2020. +[25] F. Zhou and G. Cong, “On the convergence properties of a k-step averaging stochastic gradient descent algorithm for nonconvex optimization,” in Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence, IJCAI-18, 7 2018, pp. 3219–3227. +[26] F. Sattler, S. Wiedemann, K.-R. Müller, and W. Samek, “Robust and communication-efficient federated learning from non-iid data,” IEEE transactions on neural networks and learning systems, vol. 31, no. 9, pp. 3400–3413, 2019. +[27] Y. Zhao, M. Li, L. Lai, N. Suda, D. Civin, and V. Chandra, “Federated learning with non-iid data,” arXiv preprint arXiv:1806.00582, 2018. +[28] J. Wang, V. Tantia, N. Ballas, and M. Rabbat, “Slowmo: Improving communication-efficient distributed sgd with slow momentum,” arXiv preprint arXiv:1910.00643, 2019. +[29] X. Liang, S. Shen, J. Liu, Z. Pan, E. Chen, and Y. Cheng, “Variance reduced local sgd with lower communication complexity,” arXiv preprint arXiv:1912.12844, 2019. +[30] P. Sharma, P. Khanduri, S. Bulusu, K. Rajawat, and P. K. Varshney, “Parallel restarted SPIDER – communication efficient distributed nonconvex optimization with optimal computation complexity,” arXiv preprint arXiv:1912.06036, 2019. +[31] S. J. Reddi, S. Kale, and S. Kumar, “On the convergence of adam and beyond,” arXiv preprint arXiv:1904.09237, 2019. +[32] A. Koloskova, N. Loizou, S. Boreiri, M. Jaggi, and S. Stich, “A unified theory of decentralized sgd with changing topology and local updates,” in International Conference on Machine Learning. PMLR, 2020, pp. 5381–5393. +[33] R. Johnson and T. Zhang, “Accelerating stochastic gradient descent using predictive variance reduction,” in Advances in Neural Information Processing Systems 26. Curran Associates, Inc., 2013, pp. 315–323. +[34] L. Bottou, F. E. Curtis, and J. Nocedal, “Optimization methods for large-scale machine learning,” SIAM Review, vol. 60, no. 2, pp. 223–311, 2018. +[35] Y. Drori and O. Shamir, “The complexity of finding stationary points with stochastic gradient descent,” in International Conference on Machine Learning. PMLR, 2020, pp. 2658–2667. +[36] D. A. E. Acar, Y. Zhao, R. Matas, M. Mattina, P. Whatmough, and V. Saligrama, “Federated learning based on dynamic regularization,” in International Conference on Learning Representations, 2020. +[37] S. Caldas, S. M. K. Duddu, P. Wu, T. Li, J. Konecnˇ ý, H. B. McMahan, V. Smith, and A. Talwalkar, “Leaf: A benchmark for federated settings,” 2019. \ No newline at end of file diff --git a/parse/train/J28lNO4p3ki/J28lNO4p3ki_content_list.json b/parse/train/J28lNO4p3ki/J28lNO4p3ki_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..09b005298f844ff63df32ee9321440dca1cb9d2a --- /dev/null +++ b/parse/train/J28lNO4p3ki/J28lNO4p3ki_content_list.json @@ -0,0 +1,1234 @@ +[ + { + "type": "text", + "text": "STEM: A Stochastic Two-Sided Momentum Algorithm Achieving Near-Optimal Sample and Communication Complexities for Federated Learning ", + "text_level": 1, + "bbox": [ + 174, + 122, + 823, + 198 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Prashant Khanduri University of Minnesota khand095@umn.edu ", + "bbox": [ + 269, + 251, + 429, + 292 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Pranay Sharma Carnegie Mellon University pranaysh@andrew.cmu.edu ", + "bbox": [ + 529, + 251, + 730, + 292 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Haibo Yang The Ohio State University yang.5952@buckeyemail.osu.edu ", + "bbox": [ + 186, + 314, + 436, + 356 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Mingyi Hong⇤ University of Minnesota mhong@umn.edu ", + "bbox": [ + 455, + 314, + 617, + 356 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Jia Liu The Ohio State University liu@ece.osu.edu ", + "bbox": [ + 637, + 314, + 812, + 356 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Ketan Rajawat Indian Institute of Technology Kanpur ketan@iitk.ac.in ", + "bbox": [ + 253, + 377, + 508, + 419 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Pramod K. Varshney Syracuse University varshney@syr.edu ", + "bbox": [ + 593, + 377, + 743, + 420 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Abstract ", + "text_level": 1, + "bbox": [ + 462, + 455, + 535, + 472 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Federated Learning (FL) refers to the paradigm where multiple worker nodes (WNs) build a joint model by using local data. Despite extensive research, for a generic non-convex FL problem, it is not clear, how to choose the WNs’ and the server’s update directions, the minibatch sizes, and the number of local updates, so that the WNs use the minimum number of samples and communication rounds to achieve the desired solution. This work addresses the above question and considers a class of stochastic algorithms where the WNs perform a few local updates before communication. We show that when both the WN’s and the server’s directions are chosen based on certain stochastic momentum estimator, the algorithm requires $\\tilde { \\mathcal O } ( \\epsilon ^ { - 3 / 2 } )$ samples and $\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )$ communication rounds to compute an $\\epsilon$ -stationary solution. To the best of our knowledge, this is the first FL algorithm that achieves such near-optimal sample and communication complexities simultaneously. Further, we show that there is a trade-off curve between the number of local updates and the minibatch sizes, on which the above sample and communication complexities can be maintained. Finally, we show that for the classical FedAvg (a.k.a. Local SGD, which is a momentum-less special case of the STEM), a similar trade-off curve exists, albeit with worse sample and communication complexities. Our insights on this trade-off provides guidelines for choosing the four important design elements for FL algorithms, the number of local updates, WNs’ and server’s update directions, and minibatch sizes to achieve the best performance. ", + "bbox": [ + 232, + 486, + 766, + 765 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 Introduction ", + "text_level": 1, + "bbox": [ + 174, + 790, + 310, + 808 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "In Federated Learning (FL), multiple worker nodes (WNs) collaborate with the goal of learning a joint model, by only using local data. Therefore it has become popular for machine learning problems where datasets are massively distributed [1]. In FL, the data is often collected at or off-loaded to multiple WNs which in collaboration with a server node (SN) jointly aim to learn a centralized model [2, 3]. The local WNs share the computational load and since the data is local to each WN, FL also provides some level of data privacy $\\mathbb { \\lVert \\rVert }$ . A classical distributed optimization problem that $K$ WNs aim to solve: ", + "bbox": [ + 174, + 821, + 823, + 877 + ], + "page_idx": 0 + }, + { + "type": "image", + "img_path": "images/8b109c70673d05d4fb4863253d56729a5c2b37b6a73924b30a5e660d7b6d1078.jpg", + "image_caption": [ + "Figure 1: The 3D surface in (a) plots the communication complexity of the proposed STEM for different minibatch sizes and number of local updates. The surface is generated such that each point represents STEM with a particular choice of $( b , I )$ , so that it requires $\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )$ samples to achieve $\\epsilon$ -stationarity. Plot (b) shows the optimal trade off between the minibatch sizes and the number of local updates at each WN (i.e., achieving the lowest communication and sample complexities). Both plots are generated for an accuracy of $\\epsilon = 1 0 ^ { - 3 }$ and all the constants dependent on system parameters (variance of stochastic gradients, heterogeneity parameter, optimality gap, Lipschitz constants, etc.) are assumed to be 1. Fed STEM is a special case of STEM where $\\mathcal { O } ( 1 )$ minibatch is used; Minibatch STEM is a special case of STEM where $\\mathcal { O } ( 1 )$ local updates are used. " + ], + "image_footnote": [], + "bbox": [ + 205, + 92, + 772, + 261 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 398, + 826, + 440 + ], + "page_idx": 1 + }, + { + "type": "equation", + "img_path": "images/2805a58b3a852bb789307ddca3e7e2aea04c24beb34df4fd0d38807ed0628944.jpg", + "text": "$$\n\\operatorname* { m i n } _ { x \\in \\mathbb { R } ^ { d } } \\bigg \\{ f ( x ) : = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } f ^ { ( k ) } ( x ) : = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } \\mathbb { E } _ { \\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) } } \\big [ f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) \\big ] \\bigg \\} .\n$$", + "text_format": "latex", + "bbox": [ + 266, + 444, + 730, + 488 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "where $f ^ { ( k ) } : \\mathbb { R } ^ { d } \\mathbb { R }$ denotes the smooth (possibly non-convex) objective function and $\\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) }$ represents the sample/s drawn from distribution $\\mathcal { D } ^ { ( k ) }$ at the $k ^ { \\mathrm { { t h } } }$ WN with $k \\in [ K ]$ . When the distributions $\\mathcal { D } ^ { ( k ) }$ are different across the WNs, it is referred to as the heterogeneous data setting. ", + "bbox": [ + 174, + 493, + 825, + 539 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "The optimization performance of non-convex FL algorithms is typically measured by the total number of samples accessed (cf. Definition $2 . 2 )$ and the total rounds of communication (cf. Definition $2 . 3 )$ required by each WN to achieve an $\\epsilon$ -stationary solution (cf. Definition $2 . 1 )$ . To minimize the sample and the communication complexities, FL algorithms rely on the following four key design elements: (i) the WNs’ local model update directions, (ii) Minibatch size to compute each local direction, (iii) the number of local updates before WNs share their parameters, and (iv) the SN’s update direction. How to find effective FL algorithms by (optimally) designing these parameters has received significant research interest recently. ", + "bbox": [ + 173, + 545, + 825, + 657 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Contributions. The main contributions of this work are listed below: ", + "bbox": [ + 174, + 662, + 629, + 676 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "1) We propose the Stochastic Two-Sided Momentum (STEM) algorithm, that utilizes certain momentum-assisted stochastic gradient directions for both the WNs and SN updates. We show that there exists an optimal trade off between the minibatch sizes and number of local updates, such that on the trade-off curve STEM requires $\\underline { { \\tilde { \\mathcal { O } } } } ( \\epsilon ^ { - 3 / 2 } ) \\underline { { \\left[ \\frac { 2 } { } \\right] } }$ samples and $\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )$ communication rounds to reach an $\\epsilon$ -stationary solution; see Figure $\\bigstar$ for an illustration. These complexity results are the best achievable for first-order stochastic FL algorithms (under certain assumptions, cf. Assumption $\\bigstar \\bigstar$ ; see $\\pm \\boxed { 5 } \\boxed { 8 } \\parallel$ and $\\mathbb { B } \\mathbb { n o }$ , as well as Remark $\\checkmark$ of this paper for discussions regarding optimality. To the best of our knowledge, STEM is the first algorithm which – (i) simultaneously achieves the optimal sample and communication complexities for FL and (ii) can optimally trade off the minibatch sizes and the number of local updates. ", + "bbox": [ + 173, + 683, + 825, + 824 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2) A momentum-less special case of our STEM result further reveals some interesting insights of the classical FedAvg algorithm (a.k.a. the Local SGD) [11–13]. Specifically, we show that for FedAvg, there also exists a trade-off between the minibatch sizes and the number of local updates, such that it requires $\\mathcal { O } ( \\epsilon ^ { - 2 } )$ samples and $\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )$ communication rounds to achieve an $\\epsilon$ -stationary solution. ", + "bbox": [ + 174, + 830, + 825, + 887 + ], + "page_idx": 1 + }, + { + "type": "table", + "img_path": "images/c128a8f8c828914de2ebd0394fe1679cfb1eb64efae4ae9cdf73ecbb43acff9d.jpg", + "table_caption": [], + "table_footnote": [ + "Table 1: Comparison of FedAvg and STEM with different FL algorithms for various choices of the minibatch sizes $( b )$ and the number of per node local updates between two rounds of communication $( I )$ . $^ \\circ \\nu \\in [ 0 , 1 ]$ trades off $^ { b }$ and $I$ ; $\\nu = 1$ (resp. $\\nu = 0$ ) uses multiple (resp. $\\mathcal { O } ( 1 ) .$ ) local updates and $\\mathcal { O } ( 1 )$ (resp. multiple) samples. Fed STEM and Minibatch STEM are two variants of the proposed STEM. ‡The data heterogeneity assumption is weaker than Assumption $2$ (please see $\\bigstar \\bigstar$ for details). †Requires bounded Hessian dissimilarity to model data heterogeneity across WNs. ⇤Guarantees for Minibatch STEM with $I = 1$ and SCAFFOLD are independent of the data heterogeneity. " + ], + "table_body": "
AlgorithmWorkSampleComm.Minibatch (b)Local Updates (I) /round
FedAvg国园 国国0(€-2)0(c-3/2) 0(c-2)0(1) 0(1) 2(1-v)0(c-1/2) 0(1) 3v
SCAFFOLD*this work 国0(c-2)O(e-3/2) 0(c-2)O(c 4-v) 0(1)O(c−2(4-D)) 0(1)
FedPD/FedProx*四/□O(c-2)0(e-1)0(1)0(e-1)
MIME†/FedGLOMO/80(c-3/2)O(€-3/2)0(1)0(1)
STEM Fed STEM Minibatch STEM* this workO(€-3/2)O(e-1)( 0(1) O(e-1/2)O(∈−(3)) O(∈-1/2) 0(1)
", + "bbox": [ + 173, + 88, + 830, + 266 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Collectively, our insights on the trade-offs provide practical guidelines for choosing different design elements for FL algorithms. ", + "bbox": [ + 176, + 405, + 823, + 433 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Related Works. FL algorithms were first proposed in the form of FedAvg [11], where the local update directions at each WN were chosen to be the SGD updates. Earlier works analyzed these algorithms in the homogeneous data setting [19–25], while many recent studies have focused on designing new algorithms to deal with heterogeneous data settings, as well as problems where the local loss functions are non-convex [9, 10, 12–16, 18, 26–32]. In $\\bar { \\mathbb { E } 2 } \\mathbb { I }$ , the authors showed that Parallel Restarted SGD (Local SGD or FedAvg [11]) achieves linear speed up while requiring $\\mathcal { O } ( \\epsilon ^ { - 2 } )$ samples and $\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )$ rounds of communication to reach an $\\epsilon$ -stationary solution. In $[ \\bar { \\lVert { 4 } } ]$ , a Momentum SGD was proposed, which achieved the same sample and communication complexities as Parallel Restarted SGD $\\mathbb { \\lVert 1 2 \\rVert }$ , without requiring that the second moments of the gradients be bounded. Further, it was shown that under the homogeneous data setting, the communication complexity can be improved to $\\mathcal { O } ( \\epsilon ^ { - 1 } )$ while maintaining the same sample complexity. The works in $\\boxed { 1 5 } , \\boxed { 1 6 }$ conducted tighter analysis for FedAvg with partial WN participation with $\\mathcal { O } ( 1 )$ local updates and batch sizes. Their analysis showed that FedAvg’s sample and communication complexities are both $\\mathcal { O } ( \\epsilon ^ { - 2 } )$ . Additionally, SCAFFOLD was proposed in $\\bar { \\| 1 5 \\| }$ , which utilized variance reduction based local update directions $\\pmb { \\mathbb { B 3 } }$ to achieve the same sample and communication complexities as FedAvg. Similarly, VRL-SGD proposed in $\\left[ \\left[ 2 9 \\right] \\right]$ also utilized variance reduction and showed improved communication complexity of $\\mathcal { O } ( \\epsilon ^ { - 1 } )$ , while requiring the same computations as FedAvg. Importantly, both SCAFFOLD and VRL-SGD’s guarantees were independent of the data heterogeneity. The FedProx proposed in $\\mathbb { \\ m }$ used a penalty based method to improve the communication complexity of FedAvg (i.e., the Parallel Restarted and Momentum SGD [14, $\\mathbb { L } 2 \\mathbb { I }$ ) to $\\mathcal { O } ( \\epsilon ^ { - 1 } )$ . FedProx used a gradient similarity assumption to model data heterogeneity which can be stringent for many practical applications. This assumption was relaxed by FedPD proposed in $\\pmb { \\mathbb { B } }$ . ", + "bbox": [ + 173, + 439, + 825, + 743 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Recently, the works [17, 18] proposed to utilize hybrid momentum gradient estimators [7, 8]. The MIME algorithm $\\mathbb { \\lVert 1 7 \\rVert }$ matched the optimal sample complexity (under certain smoothness assumptions) of $\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )$ of the centralized non-convex stochastic optimization algorithms [5–8]. Similarly, Fed-GLOMO $[ \\overline { { 1 8 } } ]$ achieved the same sample complexity while employing compression to further reduce communication. Both MIME and Fed-GLOMO required $\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )$ communication rounds to achieve an $\\epsilon$ -stationary solution. Please see Table $^ 1$ for a summary of the above discussion. ", + "bbox": [ + 174, + 748, + 825, + 837 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "The comparison of Local SGD (FedAvg) to Minibatch SGD for convex and strongly convex problems with homogeneous data setting was first conducted in $\\mathbb { \\lVert 1 9 \\rVert }$ and later extended to heterogeneous setting in $\\mathbb { \\lVert \\rVert 3 \\rVert }$ . It was shown that Minibatch SGD almost always dominates the Local SGD. In contrast, it was shown in $\\pmb { \\Vert 2 4 \\Vert }$ that Local SGD dominates Minibatch SGD in terms of generalization performance. Although existing FL results are rich, but they are somehow ad hoc and there is a lack of principled understanding of the algorithms. We note that the proposed STEM algorithmic framework provides a theoretical framework that unifies all existing $\\mathrm { F L }$ results on sample and communication complexities. ", + "bbox": [ + 174, + 842, + 825, + 911 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 90, + 825, + 119 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Notations. The expected value of a random variable $X$ is denoted by $\\mathbb { E } [ X ]$ and its expectation conditioned on an Event $A$ is denoted as $\\mathbb { E } [ X | \\mathrm { E v e n t ~ } A ]$ . We denote by $\\mathbb { R }$ (and $\\mathbb { R } ^ { d }$ ) the real line (and the $d$ -dimensional Euclidean space). The set of natural numbers is denoted by $\\mathbb { N }$ . Given a positive integer $K \\in \\mathbb N$ , we denote $[ K ] \\triangleq \\{ 1 , 2 , \\dots , K \\}$ . Notation $\\| \\cdot \\|$ denotes the $\\ell _ { 2 }$ -norm and $\\langle \\cdot , \\cdot \\rangle$ the Euclidean inner product. For a discrete set $\\boldsymbol { B }$ , $| B |$ denotes the cardinality of the set. Uniform distribution over a discrete set $\\{ 1 , \\ldots , T \\}$ is denoted as ${ \\dot { \\mathcal { U } } } \\{ 1 , \\dots , T \\}$ . ", + "bbox": [ + 173, + 126, + 825, + 213 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "2 Preliminaries ", + "text_level": 1, + "bbox": [ + 174, + 231, + 318, + 248 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Before we proceed to the algorithms, we make the following assumptions about problem $( 1 )$ . ", + "bbox": [ + 171, + 262, + 782, + 277 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Assumption 1 (Sample Gradient Lipschitz Smoothness). The stochastic functions $f ^ { ( k ) } ( \\cdot , \\xi ^ { ( k ) } )$ with $\\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) }$ for all $k \\in [ K ]$ , satisfy the mean squared smoothness property, i.e, we have ", + "bbox": [ + 171, + 281, + 823, + 313 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/a3cfe57005b8e58ae1bfc3b55d7cca26c0bbb487296b4468bd38bfdc42e09c0e.jpg", + "text": "$$\n\\begin{array} { r } { \\mathbb { E } \\| \\nabla f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) - \\nabla f ^ { ( k ) } ( y ; \\xi ^ { ( k ) } ) \\| ^ { 2 } \\leq L ^ { 2 } \\| x - y \\| ^ { 2 } \\mathrm { ~ f o r ~ a l l ~ } x , y \\in \\mathbb { R } ^ { d } . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 259, + 318, + 736, + 338 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Assumption 2 (Unbiased gradient and Variance Bounds). (i) Unbiased Gradient. The stochastic gradients computed at each WN are unbiased ", + "bbox": [ + 176, + 343, + 823, + 372 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/f06c119c8d26c099d56bae28e2fa443fdee4af7121660da108f532b9c8355105.jpg", + "text": "$$\n\\mathbb { E } [ \\nabla f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) ] = \\nabla f ^ { ( k ) } ( x ) , \\forall \\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) } , \\forall k \\in [ K ] .\n$$", + "text_format": "latex", + "bbox": [ + 302, + 377, + 696, + 396 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "(ii) Intra- and inter- node Variance Bound. The following bounds hold: ", + "bbox": [ + 171, + 400, + 638, + 415 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/b27d78c5e7a746fe9622599d0a5137acd7faea00d13d76004e33686cafb79bc5.jpg", + "text": "$$\n\\begin{array} { r } { \\mathbb { \\tilde { z } } \\| \\nabla f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) - \\nabla f ^ { ( k ) } ( x ) \\| ^ { 2 } \\leq \\sigma ^ { 2 } , \\| \\nabla f ^ { ( k ) } ( x ) - \\nabla f ^ { ( \\ell ) } ( x ) \\| ^ { 2 } \\leq \\zeta ^ { 2 } , \\forall \\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) } , \\forall k , \\ell \\in [ K ] . } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 181, + 420, + 834, + 441 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Note that Assumption $\\boxed { 1 }$ is stronger than directly assuming $f ^ { ( k ) }$ ’s are Lipschitz smooth (which we will refer to as the averaged gradient Lipschitz smooth condition), but it is still a rather standard assumption in SGD analysis. For example it has been used in analyzing centralized SGD algorithms such as SPIDER $ { \\mathbb { I } }$ , SNVRG $\\pmb { \\Vert 6 \\Vert }$ , STORM $\\mathbb { [ [ \\big ] ] }$ (and many others) as well as in FL algorithms such as MIME $ { \\mathbb { I } } ^ { [ 1 2 ] }$ and Fed-GLOMO $ { \\mathbb { I } } { \\mathrm { 1 8 } } { \\Vert }$ . The second relation in Assumption $2 \\cdot$ (ii) quantifies the data heterogeneity, and we call $\\zeta > 0$ as the heterogeneity parameter. This is a typical assumption required to evaluate the performance of FL algorithms. If data distributions across individual WNs are identical, i.e., $\\mathcal { D } ^ { ( k ) } \\stackrel { = } { = } \\mathcal { D } ^ { ( \\ell ) }$ for all $k , \\ell \\in [ K ]$ then we have $\\zeta = 0$ . ", + "bbox": [ + 173, + 453, + 826, + 569 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Next, we define the $\\epsilon$ -stationary solution for non-convex optimization problems, as well as quantify the computation and communication complexities to achieve an $\\epsilon$ -stationary point. ", + "bbox": [ + 174, + 574, + 821, + 603 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Definition 2.1 $\\epsilon$ -Stationary Point). A point $x$ is called $\\epsilon$ -stationary if $\\| \\nabla f ( x ) \\| ^ { 2 } \\leq \\epsilon$ . Moreover, a stochastic algorithm is said to achieve an $\\epsilon$ -stationary point in $t$ iterations if $\\begin{array} { r } { \\ddot { \\mathbb { E } } [ \\| \\nabla f ( x _ { t } ) \\| ^ { 2 } ] \\le \\epsilon . } \\end{array}$ where the expectation is over the stochasticity of the algorithm until time instant $t$ . ", + "bbox": [ + 173, + 606, + 826, + 648 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Definition 2.2 (Sample complexity). We assume an Incremental First-order Oracle (IFO) framework $\\textcircled { \\lVert { 3 4 } \\rVert }$ where, given a sample $\\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) }$ at the $k ^ { \\mathrm { { t h } } }$ node and iterate $x$ , the oracle returns $( f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) , \\nabla f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) )$ . Each access to the oracle is counted as a single IFO operation. We measure the sample (and computational) complexity in terms of the total number of calls to the IFO by all WNs to achieve an $\\epsilon$ -stationary point given in Definition 2.1. ", + "bbox": [ + 173, + 652, + 825, + 727 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Definition 2.3 (Communication complexity). We define a communication round as a one back-andforth sharing of parameters between the WNs and the SN. Then the communication complexity is defined to be the total number of communication rounds between any WN and the SN required to achieve an $\\epsilon$ -stationary point given in Definition 2.1. ", + "bbox": [ + 173, + 728, + 825, + 786 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "3 The STEM algorithm and the trade-off analysis ", + "text_level": 1, + "bbox": [ + 174, + 803, + 602, + 821 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "In this section, we discuss the proposed algorithm and present the main results. The key in the algorithm design is to carefully balance all the four design elements mentioned in Sec. $^ { 1 , }$ so that sufficient and useful progress can be made between two rounds of communication. ", + "bbox": [ + 174, + 835, + 825, + 877 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Let us discuss the key steps of STEM, listed in Algorithm $^ { 1 . }$ In Step 10, each node locally updates its model parameters using the local direction $d _ { t } ^ { k }$ , computed by using $b$ stochastic gradients at two ", + "bbox": [ + 173, + 882, + 821, + 912 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "1: Input: Parameters: $c > 0$ , the number of local updates $I$ , batch size $b$ , stepsizes $\\{ \\eta _ { t } \\}$ . \n2: Initialize: Iterate $\\begin{array} { r } { x _ { 1 } ^ { ( k ) } = \\bar { x } _ { 1 } = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } x _ { 1 } ^ { ( k ) } } \\end{array}$ , descent direction $\\begin{array} { r } { d _ { 1 } ^ { ( k ) } = \\bar { d } _ { 1 } = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } d _ { 1 } ^ { ( k ) } } \\end{array}$ \nwith $\\begin{array} { r } { d _ { 1 } ^ { ( k ) } = \\frac { 1 } { B } \\sum _ { \\xi _ { 1 } ^ { ( k ) } \\in \\mathcal { B } _ { 1 } ^ { ( k ) } } \\nabla f ^ { ( k ) } ( x _ { 1 } ^ { ( k ) } ; \\xi _ { 1 } ^ { ( k ) } ) } \\end{array}$ and $| B _ { 1 } ^ { ( k ) } | = B$ for $k \\in [ K ]$ . \n3: Perform: $x _ { 2 } ^ { ( k ) } = x _ { 1 } ^ { k } - \\eta _ { 1 } d _ { 1 } ^ { ( k ) }$ , $\\forall k \\in [ K ]$ \n4: for $t = 1$ to $T$ do \n5: for $k = 1$ to $K$ do #at the WN \n6: $\\mathcal { d } _ { t + 1 } ^ { ( k ) } = \\frac { 1 } { b } \\sum _ { \\xi _ { t + 1 } ^ { ( k ) } \\in \\mathcal { B } _ { t + 1 } ^ { ( k ) } } ^ { \\pi \\Delta } \\nabla f ^ { ( k ) } ( x _ { t + 1 } ^ { ( k ) } ; \\xi _ { t + 1 } ^ { ( k ) } ) + \\left( 1 - a _ { t + 1 } \\right) \\bigg ( d _ { t } ^ { ( k ) } - \\frac { 1 } { b } \\sum _ { \\xi _ { t + 1 } ^ { ( k ) } \\in \\mathcal { B } _ { t + 1 } ^ { ( k ) } } \\nabla f ^ { ( k ) } ( x _ { t } ^ { ( k ) } ; \\xi _ { t + 1 } ^ { ( k ) } ) \\bigg )$ \nwhere we choose $| B _ { t + 1 } ^ { ( k ) } | = b$ , and $a _ { t + 1 } = c \\cdot \\eta _ { t } ^ { 2 }$ ; \n7: 8: if $t$ $I = 0$ #at the SN \n$\\begin{array} { r } { d _ { t + 1 } ^ { ( k ) } = \\bar { d } _ { t + 1 } : = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } d _ { t + 1 } ^ { ( k ) } } \\end{array}$ \n9: 10: e $\\begin{array} { r l } & { x _ { t + 2 } ^ { ( k ) ^ { \\bot } } : = \\bar { x } _ { t + 1 } - \\eta _ { t + 1 } \\bar { d } _ { t + 1 } = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } x _ { t + 1 } ^ { ( k ) } - \\eta _ { t + 1 } \\bar { d } _ { t + 1 } } \\end{array}$ $x _ { t + 2 } ^ { ( k ) } = x _ { t + 1 } ^ { ( k ) } - \\eta _ { t + 1 } d _ { t + 1 } ^ { ( k ) }$ #server-side momentum \n11: end if \n12: end for \n13: end for \n14: Return: $\\scriptstyle { \\bar { x } } _ { a }$ where $a \\sim \\mathcal { U } \\{ 1 , . . . , T \\}$ . ", + "bbox": [ + 178, + 108, + 826, + 393 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "consecutive iterates $x _ { t + 1 } ^ { ( k ) }$ and $x _ { t } ^ { ( k ) }$ . After every $I$ local steps, the WNs share their current local models $\\{ x _ { t + 1 } ^ { ( k ) } \\} _ { k = 1 } ^ { K }$ and directions $\\{ d _ { t + 1 } ^ { ( k ) } \\} _ { k = 1 } ^ { K }$ with the SN. The SN aggregates these quantities, and performs a server-side momentum step, before returning $\\bar { x } _ { t + 1 }$ and $\\bar { d } _ { t + 1 }$ to all the WNs. Because both the WNs and the SN perform momentum based updates, we call the algorithm a stochastic two-sided momentum algorithm. The key parameters are: $b$ the minibatch size, $I$ the local update steps between two communication rounds, $\\eta _ { t }$ the stepsizes, and $a _ { t }$ the momentum parameters. ", + "bbox": [ + 173, + 420, + 825, + 515 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "One key technical innovation of our algorithm design is to identify the most suitable way to incorporate momentum based directions in FL algorithms. Although the momentum-based gradient estimator itself is not new and has been used in the literature before (see e.g., in $\\mathbb { \\left[ \\bigstar \\bigstar \\right] }$ and $\\overline { { \\mathbb { D } \\mathbb { Z } \\mathbb { G } \\mathbb { S } } }$ to improve the sample complexities of centralized and decentralized stochastic optimization problems, respectively), it is by no means clear if and how it can contribute to improve the communication complexity of FL algorithms. We show that in the FL setting, the local directions together with the local models have to be aggregated by the SN so to avoid being influenced too much by the local data. More importantly, besides the WNs, the SN also needs to perform updates using the (aggregated) momentum directions. Finally, such two-sided momentum updates have to be done carefully with the correct choice of minibatch size $b$ , and the number of local updates $I$ . Overall, it is the judicious choice of all these design elements that results in the optimal sample and communication complexities. ", + "bbox": [ + 173, + 520, + 826, + 672 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Next, we present the convergence guarantees of the STEM algorithm. ", + "bbox": [ + 174, + 678, + 630, + 693 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "3.1 Main results: convergence guarantees for STEM ", + "text_level": 1, + "bbox": [ + 174, + 712, + 549, + 727 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "In this section, we analyze the performance of STEM. We first present our main result, and then provide discussions about a few parameter choices. In the next subsection, we discuss a special case of STEM related to the classical FedAvg and minibatch SGD algorithms. ", + "bbox": [ + 174, + 738, + 825, + 781 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Theorem 3.1. Under the Assumptions 1 and 2, suppose the stepsize sequence is chosen as: ", + "bbox": [ + 168, + 785, + 769, + 801 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/c7b9d663b7230014554f849f65c327948834a1aadf2fddbf4fd4727068678b71.jpg", + "text": "$$\n\\eta _ { t } = \\frac { \\bar { \\kappa } } { ( w _ { t } + \\sigma ^ { 2 } t ) ^ { 1 / 3 } } ,\n$$", + "text_format": "latex", + "bbox": [ + 428, + 809, + 568, + 842 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "where we define : ", + "bbox": [ + 173, + 851, + 289, + 864 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/1ab3f1b7115347d40a7bc1f3343dd7707b65537827e4b8370ab97442863fcfd1.jpg", + "text": "$$\n\\bar { \\kappa } = \\frac { ( b K ) ^ { 2 / 3 } \\sigma ^ { 2 / 3 } } { L } , \\quad w _ { t } = \\operatorname * { m a x } \\bigg \\{ 2 \\sigma ^ { 2 } , 4 0 9 6 L ^ { 3 } I ^ { 3 } \\bar { \\kappa } ^ { 3 } - \\sigma ^ { 2 } t , \\frac { c ^ { 3 } \\bar { \\kappa } ^ { 3 } } { 4 0 9 6 L ^ { 3 } I ^ { 3 } } \\bigg \\} .\n$$", + "text_format": "latex", + "bbox": [ + 253, + 873, + 738, + 910 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Further, let us set $\\begin{array} { r } { c = \\frac { 6 4 L ^ { 2 } } { b K } + \\frac { \\sigma ^ { 2 } } { 2 4 \\bar { \\kappa } ^ { 3 } L I } = L ^ { 2 } \\bigg ( \\frac { 6 4 } { b K } + \\frac { 1 } { 2 4 ( b K ) ^ { 2 } I } \\bigg ) } \\end{array}$ and set the initial batch size as $B = b I$ ; set the local updates $I$ and minibatch size b as follows: ", + "bbox": [ + 173, + 88, + 825, + 133 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/571cd8b7f6de9fecec247a32a77a5fc7ff44b9b1a1446281dd4865af343e8ec8.jpg", + "text": "$$\nI = \\mathcal { O } \\big ( ( T / K ^ { 2 } ) ^ { \\nu / 3 } \\big ) , \\quad b = \\mathcal { O } \\big ( ( T / K ^ { 2 } ) ^ { 1 / 2 - \\nu / 2 } \\big )\n$$", + "text_format": "latex", + "bbox": [ + 336, + 141, + 660, + 161 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "where $\\nu$ satisfies $\\nu \\in [ 0 , 1 ]$ . Then for STEM the following holds: ", + "bbox": [ + 173, + 167, + 607, + 185 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "(i) For $\\scriptstyle { \\bar { x } } _ { a }$ chosen according to Algorithm $\\boldsymbol { l } ,$ we have: ", + "bbox": [ + 176, + 195, + 532, + 212 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/e2c146df205c691e32ce24c2ee693fb97ea38edc895ba31bbeb511db58c5a452.jpg", + "text": "$$\n\\mathbb { E } \\Vert \\nabla f ( \\bar { x } _ { a } ) \\Vert ^ { 2 } = \\mathcal { O } \\Bigg ( \\frac { f ( \\bar { x } _ { 1 } ) - f ^ { * } } { K ^ { 2 \\nu / 3 } T ^ { 1 - \\nu / 3 } } \\Bigg ) + \\tilde { \\mathcal { O } } \\Bigg ( \\frac { \\sigma ^ { 2 } } { K ^ { 2 \\nu / 3 } T ^ { 1 - \\nu / 3 } } \\Bigg ) + \\tilde { \\mathcal { O } } \\Bigg ( \\frac { \\zeta ^ { 2 } } { K ^ { 2 \\nu / 3 } T ^ { 1 - \\nu / 3 } } \\Bigg ) .\n$$", + "text_format": "latex", + "bbox": [ + 223, + 218, + 782, + 253 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "(ii) For any $\\nu \\in [ 0 , 1 ]$ , we have ", + "bbox": [ + 173, + 267, + 382, + 284 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Sample Complexity: The sample complexity of STEM is $\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )$ . This implies that each WN requires at most $\\tilde { \\mathcal { O } } ( K ^ { - 1 } \\epsilon ^ { - 3 / 2 } )$ gradient computations, thereby achieving linear speedup with the number of WNs present in the network. ", + "bbox": [ + 176, + 284, + 825, + 329 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Communication Complexity: The communication complexity of STEM is $\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )$ . ", + "bbox": [ + 202, + 329, + 745, + 344 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "The proof of this result is relegated to the Supplemental Material. A few remarks are in order. ", + "bbox": [ + 174, + 356, + 785, + 371 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Remark 1 (Near-Optimal sample and communication complexities). Theorem $3 . 1$ suggests that when $I$ and $b$ are selected appropriately, then STEM achieves $\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )$ and $\\tilde { \\mathcal { O } } ( \\overline { { \\epsilon ^ { - 1 } } } )$ sample and communication complexities. Taking them separately, these complexity bounds are the best achievable by the existing FL algorithms (upto logarithmic factors regardless of sample or batch Lipschitz smooth assumption) $[ [ 3 5 ] ]$ ; see Table $1 .$ We note that the $\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )$ complexity is the best possible that can be achieved by centralized SGD with the sample Lipschitz gradient assumption; see $ { \\mathbb { I } } ^ { { \\left[ 5 \\right] } }$ . On the other hand, the $\\bar { \\mathcal { O } } ( \\epsilon ^ { - 1 } )$ complexity bound is also likely to be the optimal, since in $\\mathbb { P }$ the authors showed that even when the local steps use a class of (deterministic) first-order algorithms, $\\mathcal { O } ( \\epsilon ^ { - 1 } )$ is the best achievable communication complexity. The only difference is that $\\bigstar \\bigstar$ does not explicitly assume the inter-node variance bound (i.e., the second relation in Assumption $2 \\cdot$ -(ii)). We leave the precise characterization of the communication lower bound with inter-node variance as future work. □ ", + "bbox": [ + 173, + 372, + 825, + 527 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Remark 2 (Large Batch Sizes and/or Local Updates). At first glance, it may seem that the requirement of STEM to compute large mini-batches and/or local updates (cf. Table $^ { 1 ) }$ to achieve this (near) optimal performance is a drawback, however, we note that it is in fact an advantage of STEM that it allows the WNs to perform larger number of local updates (or compute large minibatches) without communicating often. This follows from the fact that irrespective of the number of local updates (or batch sizes) STEM achieves near-optimal communication complexity while attaining optimal overall sample complexity. Moreover, note that even with $b = I = \\mathcal { O } ( 1 )$ (i.e., $b$ and $I$ are chosen as constants), STEM achieves the same (optimal) sample and communication complexities as achieved by FedGLOMO [18] and MIME [17]. We further note that to the best of our knowledge the algorithms that achieve the communication complexity of $\\mathcal { O } ( \\epsilon ^ { - 1 } )$ either require the number of local updates or the batch-sizes that depend on the solution accuracy $\\epsilon$ . For example, FedProx $\\mathbb { I O } ]$ , FedPD $\\bigstar \\bigstar$ , and FedDyn $\\pmb { \\Vert 3 6 \\Vert }$ rely on solving the “local problems\" to achieve an $\\epsilon$ -accuracy, which implies that the number of local updates (or the batch sizes) implicitly depends on the desired solution accuracy $\\epsilon$ , as is the case for STEM. Similarly, as shown in $\\bar { \\mathbb { E } 2 } \\mathbb { I }$ and $\\bar { \\textregistered 4 } \\bar { 1 }$ the communication complexity of FedAvg and its momentum version can be improved from $\\mathcal { O } ( \\epsilon ^ { - 2 } )$ to $\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )$ when the number of local updates (or batch size) is chosen as $\\mathcal { O } ( \\epsilon ^ { - 1 / 2 } )$ (cf. Section $3 . 2$ for a more detailed discussion). ", + "bbox": [ + 173, + 531, + 825, + 753 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Remark 3 (The Optimal Batch Sizes and Local Updates Trade-off). The parameter $\\nu \\in [ 0 , 1 ]$ is used to balance the local minibatch sizes $b$ , and the number of local updates $I$ . Eqs. in $\\textcircled { 3 }$ suggest that when $\\nu$ increases from 0 to 1, $b$ decreases and $I$ increases. Specifically, if $\\nu = 1$ , then $b$ is a constant but $I = \\mathcal { O } ( T ^ { 1 / 3 } / K ^ { 2 / 3 } )$ . In this case, each WN chooses a small minibatch while executing multiple local updates, and STEM resembles a FedAvg (a.k.a. Local SGD) algorithm but with double-sided momentum update directions, and is referred to as Fed STEM. In contrast, if $\\nu = 0$ , then $b = \\mathcal { O } ( T ^ { 1 / 2 } / K )$ but $I$ is a constant. In this case, each WN chooses a large batch size while executing only a few, or even one, local updates, and STEM resembles the Minibatch SGD, but again with different update directions, and is referred to as Minibatch STEM. Such a trade-off can be seen in Fig. 1b. Due to space limitation, these two special cases will be precisely stated in the supplementary materials as corollaries of Theorem 3.1. □ ", + "bbox": [ + 173, + 756, + 825, + 912 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Algorithm 2 The FedAvg Algorithm ", + "text_level": 1, + "bbox": [ + 174, + 90, + 416, + 106 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "1: Input: $\\{ \\eta _ { t } \\} _ { t = 0 } ^ { T } ; I$ , the # of local updates per communication round; $b$ , the minibatch sizes. \n2: for $t = 1$ to $T$ do \n3: 4: 5: for $\\begin{array} { r l } & { \\mathcal { \\kappa } _ { t } ^ { = } \\stackrel { \\mathrm { ~ L ~ U ~ O ~ } \\Lambda } { = } \\mathbf { 0 } } \\\\ & { d _ { t } ^ { ( k ) } = \\frac { 1 } { b } \\sum _ { \\xi _ { t } ^ { ( k ) } \\in \\mathcal { B } _ { t } ^ { ( k ) } } \\nabla f ^ { ( k ) } ( x _ { t } ^ { ( k ) } ; \\xi _ { t } ^ { ( k ) } ) \\mathrm { ~ w i t h ~ } | \\mathcal { B } _ { t } ^ { ( k ) } | = b } \\\\ & { x _ { t + 1 } ^ { ( k ) } = x _ { t } ^ { ( k ) } - \\eta _ { t } d _ { t } ^ { ( k ) } } \\\\ & { \\mathbf { i f } t \\operatorname* { m o d } I = 0 \\mathbf { \\Lambda } \\mathbf { t h e n } } \\\\ & { ~ x _ { t + 1 } ^ { ( k ) } = \\bar { x } _ { t + 1 } = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } x _ { t + 1 } ^ { ( k ) } } \\\\ & { \\mathbf { e n d } \\mathbf { \\Phi } \\mathbf { i f } } \\end{array}$ $k = 1$ $K$ \n6: \n7: \n8: \n9: end for \n10: end for \n11: Return: $\\scriptstyle { \\bar { x } } _ { a }$ where $a \\sim \\mathcal { U } \\{ 1 , . . . , T \\}$ . ", + "bbox": [ + 178, + 111, + 795, + 276 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Remark 4 (The Sub-Optimal Batch Sizes and Local Updates Trade-off). From our proof (Theorem $\\left| \\overline { { \\mathbf { C . 1 0 } } } \\right|$ included in the supplemental material), we can see that STEM requires $\\bar { \\tilde { O } } ( \\operatorname* { m a x } \\big \\{ ( b \\cdot$ $I ) \\epsilon ^ { - 1 } , K ^ { - 1 } \\epsilon ^ { - 3 / 2 } \\rbrace )$ samples and $\\tilde { \\mathcal { O } } \\big ( \\operatorname* { m a x } \\big \\{ \\epsilon ^ { - 1 } , ( b \\cdot I ) ^ { - 1 } K ^ { - 1 } \\epsilon ^ { - 3 / 2 } \\big \\} \\big )$ \u0000 and communication rounds. According to the above expressions, if $b \\cdot I$ increases beyond $\\mathcal { O } ( K ^ { - 1 } \\epsilon ^ { - 1 / 2 } )$ , then the sample complexity will increase from the optimal $\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )$ ; otherwise, the optimal sample complexity $\\tilde { \\mathcal O } ( \\epsilon ^ { - 3 / 2 } )$ is maintained. On the other hand, if $b \\cdot I$ decreases beyond $\\mathcal { O } ( K ^ { - 1 } \\epsilon ^ { - 1 / 2 } )$ , the communication complexity increases from $\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )$ . For instance, if we choose $b = \\mathcal { O } ( 1 )$ and $I = { \\mathcal { O } } ( 1 )$ the communication complexity becomes $\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )$ while the optimal sample complexity $\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )$ is maintained. This trade-off is illustrated in Figure $\\boxed { 1 \\mathrm { a } }$ where we maintain the optimal sample complexity, while changing $b$ and $I$ to generate the trade-off surface. □ ", + "bbox": [ + 173, + 306, + 826, + 463 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Remark 5 (Data Heterogeneity). The term $\\begin{array} { r } { \\tilde { \\mathcal { O } } \\biggl ( \\frac { \\zeta ^ { 2 } } { K ^ { 2 \\nu / 3 } T ^ { 1 - \\nu / 3 } } \\biggr ) } \\end{array}$ in the gradient bound $\\textcircled{4}$ captures the effect of the heterogeneity of data across WNs, where $\\zeta$ is the parameter characterizing the intra-node variance and has been defined in Assumption $\\bigstar$ (ii). Highly heterogeneous data with large $\\zeta ^ { 2 }$ can adversely impact the performance of STEM. Note that such a dependency on $\\zeta$ also appears in other existing FL algorithms, such as $[ \\bigcirc , \\bigcirc , \\bigcirc , \\bigcirc , \\bigcirc ]$ . However, there is one special case of STEM that does not depend on the parameter $\\zeta$ . This is the case where $I = 1$ , i.e., the minibatch SGD counterpart of STEM where only a single local iteration is performed between two communication rounds. We have the following corollary. □ ", + "bbox": [ + 173, + 465, + 826, + 585 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Corollary 1 (Minibatch STEM). Under Assumptions $\\bigstar \\bigstar \\bigstar | \\bigstar |$ , and choose the algorithm parameters as in Theorem $3 . I .$ At each WN, choose $I = 1$ , $b = ( T / K ^ { 2 } ) ^ { 1 / 2 }$ , and the initial batch size $B = \\boldsymbol { b } \\cdot \\boldsymbol { I }$ . Then STEM satisfies: ", + "bbox": [ + 174, + 589, + 826, + 636 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "(i) For $\\bar { x } _ { a }$ chosen according to Algorithm $\\boldsymbol { l } ,$ we have ", + "bbox": [ + 176, + 646, + 529, + 662 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/fe27c5ca3c886c333d5add660b12f630bb6023b83b7450c21e97035dfce670f9.jpg", + "text": "$$\n\\mathbb { E } \\| \\nabla f ( \\bar { x } _ { a } ) \\| ^ { 2 } = \\mathcal { O } \\Big ( \\frac { f ( \\bar { x } _ { 1 } ) - f ^ { * } } { T } \\Big ) + \\tilde { \\mathcal { O } } \\Big ( \\frac { \\sigma ^ { 2 } } { T } \\Big ) .\n$$", + "text_format": "latex", + "bbox": [ + 359, + 671, + 665, + 704 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "(ii) Minibatch STEM achieves $\\tilde { \\mathcal O } ( \\epsilon ^ { - 3 / 2 } )$ sample and $\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )$ communication complexity. ", + "bbox": [ + 168, + 719, + 761, + 738 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Next, we show that FedAvg also exhibits a trade-off similar to that of STEM but with worse sample and communication complexities. ", + "bbox": [ + 173, + 748, + 826, + 777 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "3.2 Special cases: The FedAvg algorithm ", + "text_level": 1, + "bbox": [ + 174, + 796, + 470, + 813 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "We briefly discuss another interesting special case of STEM, where the local momentum update is replaced by the conventional SGD (i.e., $a _ { t } = 1 , ~ \\forall ~ t )$ , while the server does not perform the momentum update (i.e., $\\bar { d } _ { t } = 0 , \\forall t )$ . This is essentially the classical FedAvg algorithm, just that it balances the number of local updates $I$ and the minibatch size $b$ . We show that this algorithm also exhibits a trade-off between $b$ and $I$ and on the trade-off curve it achieves $\\mathcal { O } ( \\epsilon ^ { - 2 } )$ sample complexity and $\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )$ communication complexity. ", + "bbox": [ + 174, + 823, + 825, + 910 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/6bf0f1ad1737bddbd1a909e1c507f78edccdbe2e2cf8c63393f03d6b14335b18.jpg", + "table_caption": [], + "table_footnote": [ + "(a) Mild heterogeneity, $b = 6 4$ , and $I = 7$ . " + ], + "table_body": "
Algorithm Training Acc.Testing Acc.
FedAvg78.274.1
FedProx79.274.8
FedDyn68.966.0
SCAFFOLD71.974.0
MIME82.676.8
FedGLOMO76.172.8
STEM80.178.8
", + "bbox": [ + 173, + 88, + 500, + 233 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/b0513e337be144b5f857c24b8a10281f3fd33f9a14e65f9ee342f421e0353331.jpg", + "table_caption": [], + "table_footnote": [ + "(b) Moderate heterogeneity, $b = 8$ , and $I = 6 1$ " + ], + "table_body": "
AlgorithmTraining Acc.Testing Acc.
FedAvg73.675.4
FedProx80.075.2
FedDyn76.171.3
SCAFFOLD72.573.7
MIME61.558.6
FedGLOMO10.010.0
STEM81.178.5
", + "bbox": [ + 524, + 88, + 849, + 233 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Table 2: Training and testing accuracy of different algorithms on CIFAR-10 dataset for different batch-sizes, number of local updates, and heteregeneity settings. ", + "bbox": [ + 173, + 261, + 823, + 290 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Theorem 3.2 (The FedAvg Algorithm). Under Assumptions 1 and 2, suppose the stepsize is chosen as: $\\begin{array} { r } { \\eta = \\sqrt { \\frac { b K } { T } } } \\end{array}$ ; Let us set: ", + "bbox": [ + 173, + 324, + 825, + 362 + ], + "page_idx": 7 + }, + { + "type": "equation", + "img_path": "images/614827bc2783d2a130099b5dcc25a039f7d6ca4f991f27f94b6b9f6e8cc8d82c.jpg", + "text": "$$\nI = \\mathcal { O } \\big ( ( T / K ^ { 3 } ) ^ { \\nu / 4 } \\big ) , \\quad b = \\mathcal { O } \\big ( ( T / K ^ { 3 } ) ^ { 1 / 3 - \\nu / 3 } \\big )\n$$", + "text_format": "latex", + "bbox": [ + 336, + 369, + 660, + 391 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "where $\\nu \\in [ 0 , 1 ]$ is a constant. Then for FedAvg with $T \\geq 8 1 L ^ { 2 } I ^ { 2 } b K$ , the following holds ", + "bbox": [ + 171, + 397, + 759, + 414 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "(i) For $\\scriptstyle { \\bar { x } } _ { a }$ chosen according to Algorithm $\\perp$ we have ", + "bbox": [ + 176, + 424, + 527, + 440 + ], + "page_idx": 7 + }, + { + "type": "equation", + "img_path": "images/fcbbd1eccd40989ffc376397736fd5950f8008a4dccca06a87727fb72022e909.jpg", + "text": "$$\n\\mathbb { E } \\Vert \\nabla f ( \\bar { x } _ { a } ) \\Vert ^ { 2 } = \\mathcal { O } \\Bigg ( \\frac { f ( \\bar { x } _ { 1 } ) - f ^ { * } } { K ^ { \\nu / 2 } T ^ { 2 / 3 - \\nu / 6 } } \\Bigg ) + \\mathcal { O } \\Bigg ( \\frac { \\sigma ^ { 2 } } { K ^ { \\nu / 2 } T ^ { 2 / 3 - \\nu / 6 } } \\Bigg ) + \\mathcal { O } \\Bigg ( \\frac { \\zeta ^ { 2 } } { K ^ { \\nu / 2 } T ^ { 2 / 3 - \\nu / 6 } } \\Bigg ) .\n$$", + "text_format": "latex", + "bbox": [ + 223, + 446, + 802, + 482 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "(ii) For any choice of $\\nu \\in [ 0 , 1 ]$ we have: ", + "bbox": [ + 173, + 494, + 446, + 510 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Sample Complexity: The sample complexity of FedAvg is $\\mathcal { O } ( \\epsilon ^ { - 2 } )$ . This implies that each WN requires at most $\\tilde { \\mathcal { O } } ( K ^ { - 1 } \\epsilon ^ { - 2 } )$ gradient computations, thereby achieving linear speedup with the number of WNs in the network. ", + "bbox": [ + 196, + 508, + 825, + 549 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Communication Complexity: The communication complexity of FedAvg is $\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )$ . ", + "bbox": [ + 202, + 550, + 761, + 565 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Note that the requirement on $T$ being lower bounded is only relevant for theoretical purposes, a similar requirement was also imposed in $[ \\mathbb { 1 4 } ]$ to prove convergence. Again, the parameter $\\nu \\in [ 0 , 1 ]$ in the statement of Theorem $3 . 2$ balances $I$ and $b$ at each WN while maintaining state-of-the-art sample and communication complexities; please see Table $^ 1$ for a comparison of those bounds with existing FedAvg bounds. For $\\nu = 1$ , FedAvg (cf. Theorem $\\textcircled { 3 . 2 }$ reduces to FedAvg proposed in [12, 14] and for $\\nu = 0$ , the algorithm can be viewed as a large batch FedAvg with constant local updates [15, 16]. Note that similar to STEM, it is known that for $I = 1$ , the Minibatch SGD’s performance is independent of the heterogeneity parameter, $\\zeta \\equiv \\mathbb { I I } 3 \\mathbb { I }$ . We also point out that if Algorithm $^ 1$ uses Nesterov’s or Polyak’s momentum $[ \\textcircled { 1 4 } ]$ at local WNs instead of the recursive momentum estimator we get the same guarantees as in Theorem $3 . 2 .$ ", + "bbox": [ + 173, + 577, + 825, + 717 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "In summary, this section established that once the WN’s and the SN’s update directions (SGD in FedAvg and momentum based directions in STEM) are fixed, there exists a sequence of optimal choices of the number of local updates $I$ , and the batch sizes $b$ , which guarantees the best possible sample and communication complexities for the particular algorithm. The trade-off analysis presented in this section provides some useful guidelines for how to best select $b$ and $I$ in practice. Our subsequent numerical results will also verify that if $b$ or $I$ are not chosen judiciously, then the practical performance of the algorithms can degrade significantly. ", + "bbox": [ + 174, + 722, + 825, + 819 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "4 Numerical results ", + "text_level": 1, + "bbox": [ + 176, + 838, + 352, + 854 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "In this section, we validate the proposed STEM algorithm and compare its performance with the de facto standard FedAvg [11], and the algorithms stated in Table $^ { 1 . }$ Note that instead of FedPD we include the performance comparison with FedDyn $\\pmb { \\mathbb { B } } 6 \\|$ since they are known to be very closely related. The goal of our experiments are three-fold: (1) To show that STEM performs on par, if not better, compared to other algorithms in different heterogeneity settings, (2) there are multiple ways to reach the desired solution accuracy, one can either choose a large batch size and perform only a few local updates or select a smaller batch size and perform multiple local updates, and finally, (3) if the local updates and the batch sizes are not chosen appropriately, the WNs might need to perform excessive computations to achieve the desired solution accuracy, thereby slowing down convergence. ", + "bbox": [ + 174, + 868, + 823, + 911 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/dcdc37cb3965642ea8fadecde08dc3a78e268deb8860ba74a94a052f25763bcd.jpg", + "table_caption": [ + "Table 3: Training and testing accuracy on CIFAR-10 dataset for high heterogeneity, $b =$ 128 and $I = 6$ . " + ], + "table_footnote": [], + "table_body": "
AlgorithmTraining Acc.Testing Acc.
FedAvg57.657.1
FedProx59.158.5
FedDyn51.251.3
SCAFFOLD53.154.7
MIME56.155.1
FedGLOMO56.856.1
STEM58.557.4
", + "bbox": [ + 173, + 88, + 500, + 233 + ], + "page_idx": 8 + }, + { + "type": "table", + "img_path": "images/a8ee7b8ff04d3f8e652b073870044e3954f021e10e2ae94f2132f7ce72b56ed4.jpg", + "table_caption": [ + "Table 4: Training and testing accuracy on Shakespeare dataset. " + ], + "table_footnote": [], + "table_body": "
AlgorithmTraining Acc.Testing Acc.
FedAvg40.139.2
FedProx43.543.2
FedDyn43.743.2
SCAFFOLD40.341.3
MIME32.132.1
FedGLOMO40.340.1
STEM44.543.8
", + "bbox": [ + 527, + 88, + 854, + 233 + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/1a3649642909a08bbdd852d92b8ec00817216ffd484e8ea8263d27aa9fb977f4.jpg", + "image_caption": [ + "Figure 2: Training loss and the testing accuracy for classification on MNIST data set against the number of samples accessed at each WN for moderate heterogeneity setting with $b = 8$ . " + ], + "image_footnote": [], + "bbox": [ + 186, + 320, + 790, + 491 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 558, + 825, + 641 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Data and Parameter Settings: We compare the algorithms for image classification tasks on CIFAR10 and MNIST data sets with 100 WNs, and for next character prediction task on Shakespeare dataset $\\pmb { \\Vert 3 7 } \\Vert$ with 143 WNs in the network. For both CIFAR-10 and MNIST, each WN implements a two-hidden-layer convolutional neural network (CNN) architecture followed by three linear layers for CIFAR-10 and two for MNIST. For CIFAR-10 (and MNIST) datset, we consider three settings with mild, moderate and high heterogeneity. For all the three settings, the data is partitioned into disjoint sets among the WNs. In the mild heterogeneity setting, the WNs have access to partitioned data from all the classes. In the moderate (resp. high) heterogeneity setting the data is partitioned such that each WN can access data from only 5 (resp. 2) out of 10 classes. For CIFAR-10 (resp. MNIST), each WN has access to 490 (resp. 540) samples for training and 90 (resp. 80) samples for testing purposes. ", + "bbox": [ + 173, + 647, + 825, + 786 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "We also compare the performance of algorithms on a popular FL benchmarking dataset, Shakespeare dataset $ { \\mathbb { I } } ^ { \\smash { \\sum } }$ . For this task, we adopt the settings from $\\dot { \\left[ \\left| 1 0 \\right| \\right] }$ and utilize a 2-Layer LSTM network with 100 hidden units and an 8-D embedding layer at each WN. Each WN has access to 3616 samples on average, and the samples are randomly split into an $80 \\%$ training set and a $20 \\%$ testing set. We randomly sample 10 nodes out of 143 for the training purpose. All the experiments are implemented on a single NVIDIA Quadro RTX 5000 GPU. More details are provided in appendix. ", + "bbox": [ + 174, + 792, + 825, + 876 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "For the proposed STEM algorithm, recall that the step-size is $\\eta _ { t } = \\bar { \\kappa } / ( w _ { t } + \\sigma ^ { 2 } t ) ^ { 1 / 3 }$ with momentum parameter defined as $a _ { t } = { c } \\eta _ { t } ^ { 2 }$ . The step-size is used to update the iterates while the momentum parameter is used to construct the stochastic gradient estimate (cf. Algorithm 1 and Theorem $\\boxed { 3 . 1 }$ . For the experiments, we set $w _ { t } = \\sigma ^ { 2 } = 1$ and $c \\doteq \\bar { c } / \\bar { \\kappa } ^ { 2 }$ and tune for $\\bar { \\kappa } \\in [ 1 0 ^ { - 1 } , \\overline { { 1 0 } } ^ { - 2 } ]$ for the CIFAR-10 dataset and for ${ \\bar { \\kappa } } \\in \\{ 1 0 ^ { 1 } , 1 0 ^ { 0 } , 1 0 ^ { - 1 } , 1 0 ^ { - 2 } \\}$ for the Shakespeare dataset. For both the datasets we tune for $\\bar { c }$ in the range [1, 10]. For FedProx $[ \\equiv ]$ and FedDyn $\\lVert \\bar { 3 6 } \\rVert$ we choose the regularization constant to be 0.1. The momentum parameters for FedGLOMO $\\pm \\textcircled { 1 8 } \\textcircled { 1 }$ and MIME $\\mathbb { \\lVert 1 7 \\rVert }$ are set based on the choices given in the respective papers. Specifically, for FedGLOMO we choose the parameter $\\beta _ { k } = 0 . 2$ and design the momentum gradient using a damping factor given in Appendix A.4 of FedGLOMO $\\mathbb { \\left[ \\left[ 8 \\right] \\right] }$ . Moreover, for MIME we choose the momentum parameter as 0.9. For the rest of the algorithms (including FedAvg and SCAFFOLD), the step-size is tuned from the set $\\{ 1 0 ^ { 1 } , 1 0 ^ { 0 } , 1 0 ^ { - 1 } , \\hat { 1 0 } ^ { - 2 } \\}$ . ", + "bbox": [ + 174, + 882, + 821, + 911 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 90, + 825, + 217 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Discussion: We evaluate the training and testing performance of STEM against multiple algorithms for different heterogeneity settings, minibatch sizes, and number of local updates. In Tables $\\bigstar$ $\\boxed { 2 \\mathbf { b } }$ and $\\bigstar ,$ we compare the training and testing accuracy of STEM to that of other algorithms on the CIFAR-10 dataset. Specific, heterogeneity settings, the choices of minibatches, and number of local updates are stated along with the tables. Note that STEM performs uniformly well under all the conditions. Moreover, note from Table $\\bigstar$ that FedGLOMO diverges once the number of local updates are high. Also, note from Table $\\textcircled { 3 }$ that FedProx and STEM adapt well to high heterogeneity. Finally, with the next set of experiments we emphasize the importance of choosing $b$ and $I$ carefully. In Figure $\\bigstar ,$ we compare the training and testing performance of STEM, FedAvg and SCAFFOLD, against the number of samples accessed at each WN for the classification task on MNIST dataset with moderate heterogeneity. We fix $b = 8$ and conduct experiments under two settings, one with $I = 6 7$ , and the other with $I = 5 3 6$ local updates at each WN. Note that although a large number of local updates might lead to fewer communication rounds but it can make the sample complexity extremely high as is demonstrated by Figure $2 .$ For example, Figure $\\bigtriangledown$ shows that to reach testing accuracy of $9 6 - 9 7 \\%$ with $I = 6 7$ , STEM requires approximately $5 0 0 0 - 6 0 0 0$ samples, in contrast with $I = 5 3 6$ it requires more than 25000 samples at each WN. Similar behavior can be observed if we fix $I > 1$ and increase the local batch sizes. This implies not choosing the local updates and the batch sizes judiciously might lead to increased sample complexity. Additional experiments are included in the supplementary material to further evaluate the performance of the proposed algorithms. ", + "bbox": [ + 173, + 229, + 826, + 492 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Conclusion ", + "text_level": 1, + "bbox": [ + 174, + 512, + 269, + 529 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "In this work, we proposed a novel algorithm STEM, for distributed stochastic non-convex optimization with applications to FL. We showed that STEM reaches an $\\epsilon$ -stationary point with $\\tilde { \\mathcal O } ( \\epsilon ^ { - 3 / 2 } )$ sample complexity while achieving linear speed-up with the number of WNs. Moreover, the algorithm achieves a communication complexity of $\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )$ . We established a (optimal) trade-off that allows interpolation between varying choices of local updates and the batch sizes at each WN while maintaining (near optimal) sample and communication complexities. We showed that FedAvg (a.k.a LocalSGD) also exhibits a similar trade-off while achieving worse complexities. Our results provide guidelines to carefully choose the number of local updates, update directions, and minibatch sizes to achieve the best performance. The future directions of this work include developing lower bounds on communication complexity that establishes the tightness of the analysis conducted in this work. ", + "bbox": [ + 173, + 542, + 826, + 685 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Acknowledgement ", + "text_level": 1, + "bbox": [ + 176, + 704, + 330, + 722 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "We thank the anonymous reviewers for their valuable comments and suggestions. The work of Prashant Khanduri and Mingyi Hong was supported by NSF grant CMMI-1727757, AFOSR grant 19RT0424 and ARO grant W911NF-19-1-0247. The work of Mingyi Hong was also supported by an IBM Faculty Research award. The work of Jia Liu has been supported in part by NSF grants CAREER CNS-2110259, CNS-2112471, CNS-2102233, CCF-2110252, ECCS-2140277, and a Google Faculty Research Award. ", + "bbox": [ + 174, + 736, + 825, + 819 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "References \n[1] J. Konecnˇ y, H. B. McMahan, D. Ramage, and P. Richtárik, “Federated optimization: Distributed \\` machine learning for on-device intelligence,” arXiv preprint arXiv:1610.02527, 2016. \n[2] M. Li, D. G. Andersen, A. J. Smola, and K. Yu, “Communication efficient distributed machine learning with the parameter server,” in Advances in Neural Information Processing Systems 27, Z. Ghahramani, M. Welling, C. Cortes, N. D. Lawrence, and K. Q. Weinberger, Eds. Curran Associates, Inc., 2014, pp. 19–27. \n[3] J. Dean, G. Corrado, R. Monga, K. Chen, M. Devin, M. Mao, M. Ranzato, A. Senior, P. Tucker, K. Yang et al., “Large scale distributed deep networks,” in Advances in neural information processing systems, 2012, pp. 1223–1231. \n[4] T. Léauté and B. Faltings, “Protecting privacy through distributed computation in multi-agent decision making,” Journal of Artificial Intelligence Research, vol. 47, pp. 649–695, 2013. \n[5] C. Fang, C. J. Li, Z. Lin, and T. Zhang, “Spider: Near-optimal non-convex optimization via stochastic path-integrated differential estimator,” in Advances in Neural Information Processing Systems, 2018, pp. 689–699. \n[6] D. Zhou, P. Xu, and Q. Gu, “Stochastic nested variance reduction for nonconvex optimization,” arXiv preprint arXiv:1806.07811, 2018. \n[7] A. Cutkosky and F. Orabona, “Momentum-based variance reduction in non-convex SGD,” in Advances in Neural Information Processing Systems 32. Curran Associates, Inc., 2019, pp. 15 236–15 245. \n[8] Q. Tran-Dinh, N. H. Pham, D. T. Phan, and L. M. Nguyen, “Hybrid stochastic gradient descent algorithms for stochastic nonconvex optimization,” arXiv preprint arXiv:1905.05920, 2019. \n[9] X. Zhang, M. Hong, S. Dhople, W. Yin, and Y. Liu, “Fedpd: A federated learning framework with adaptivity to non-iid data,” IEEE Transactions on Signal Processing, pp. 1–1, 2021. \n[10] T. Li, A. K. Sahu, M. Zaheer, M. Sanjabi, A. Talwalkar, and V. Smith, “Federated optimization in heterogeneous networks,” arXiv preprint arXiv:1812.06127, 2018. \n[11] B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, “Communication-efficient learning of deep networks from decentralized data,” in Artificial Intelligence and Statistics. PMLR, 2017, pp. 1273–1282. \n[12] H. Yu, S. Yang, and S. Zhu, “Parallel restarted sgd with faster convergence and less communication: Demystifying why model averaging works for deep learning,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 33, no. 01, 2019, pp. 5693–5700. \n[13] B. Woodworth, K. K. Patel, and N. Srebro, “Minibatch vs local sgd for heterogeneous distributed learning,” arXiv preprint arXiv:2006.04735, 2020. \n[14] H. Yu, R. Jin, and S. Yang, “On the linear speedup analysis of communication efficient momentum sgd for distributed non-convex optimization,” in International Conference on Machine Learning. PMLR, 2019, pp. 7184–7193. \n[15] S. P. Karimireddy, S. Kale, M. Mohri, S. Reddi, S. Stich, and A. T. Suresh, “Scaffold: Stochastic controlled averaging for federated learning,” in International Conference on Machine Learning. PMLR, 2020, pp. 5132–5143. \n[16] H. Yang, M. Fang, and J. Liu, “Achieving linear speedup with partial worker participation in non-iid federated learning,” arXiv preprint arXiv:2101.11203, 2021. \n[17] S. P. Karimireddy, M. Jaggi, S. Kale, M. Mohri, S. J. Reddi, S. U. Stich, and A. T. Suresh, “Mime: Mimicking centralized stochastic algorithms in federated learning,” arXiv preprint arXiv:2008.03606, 2020. \n[18] R. Das, A. Hashemi, S. Sanghavi, and I. S. Dhillon, “Improved convergence rates for non-convex federated learning with compression,” arXiv preprint arXiv:2012.04061, 2020. \n[19] B. Woodworth, K. K. Patel, S. U. Stich, Z. Dai, B. Bullins, H. B. McMahan, O. Shamir, and N. Srebro, “Is local sgd better than minibatch sgd?” arXiv preprint arXiv:2002.07839, 2020. \n[20] H. Yu and R. Jin, “On the computation and communication complexity of parallel sgd with dynamic batch sizes for stochastic non-convex optimization,” in International Conference on Machine Learning. PMLR, 2019, pp. 7174–7183. \n[21] J. Wang and G. Joshi, “Cooperative sgd: A unified framework for the design and analysis of local-update sgd algorithms,” Journal of Machine Learning Research, vol. 22, no. 213, pp. 1–50, 2021. \n[22] A. Khaled, K. Mishchenko, and P. Richtárik, “Better communication complexity for local sgd,” arXiv, 2019. \n[23] S. U. Stich, “Local sgd converges fast and communicates little,” arXiv preprint arXiv:1805.09767, 2018. \n[24] T. Lin, S. U. Stich, K. K. Patel, and M. Jaggi, “Don’t use large mini-batches, use local sgd,” in International Conference on Learning Representations, 2020. \n[25] F. Zhou and G. Cong, “On the convergence properties of a k-step averaging stochastic gradient descent algorithm for nonconvex optimization,” in Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence, IJCAI-18, 7 2018, pp. 3219–3227. \n[26] F. Sattler, S. Wiedemann, K.-R. Müller, and W. Samek, “Robust and communication-efficient federated learning from non-iid data,” IEEE transactions on neural networks and learning systems, vol. 31, no. 9, pp. 3400–3413, 2019. \n[27] Y. Zhao, M. Li, L. Lai, N. Suda, D. Civin, and V. Chandra, “Federated learning with non-iid data,” arXiv preprint arXiv:1806.00582, 2018. \n[28] J. Wang, V. Tantia, N. Ballas, and M. Rabbat, “Slowmo: Improving communication-efficient distributed sgd with slow momentum,” arXiv preprint arXiv:1910.00643, 2019. \n[29] X. Liang, S. Shen, J. Liu, Z. Pan, E. Chen, and Y. Cheng, “Variance reduced local sgd with lower communication complexity,” arXiv preprint arXiv:1912.12844, 2019. \n[30] P. Sharma, P. Khanduri, S. Bulusu, K. Rajawat, and P. K. Varshney, “Parallel restarted SPIDER – communication efficient distributed nonconvex optimization with optimal computation complexity,” arXiv preprint arXiv:1912.06036, 2019. \n[31] S. J. Reddi, S. Kale, and S. Kumar, “On the convergence of adam and beyond,” arXiv preprint arXiv:1904.09237, 2019. \n[32] A. Koloskova, N. Loizou, S. Boreiri, M. Jaggi, and S. Stich, “A unified theory of decentralized sgd with changing topology and local updates,” in International Conference on Machine Learning. PMLR, 2020, pp. 5381–5393. \n[33] R. Johnson and T. Zhang, “Accelerating stochastic gradient descent using predictive variance reduction,” in Advances in Neural Information Processing Systems 26. Curran Associates, Inc., 2013, pp. 315–323. \n[34] L. Bottou, F. E. Curtis, and J. Nocedal, “Optimization methods for large-scale machine learning,” SIAM Review, vol. 60, no. 2, pp. 223–311, 2018. \n[35] Y. Drori and O. Shamir, “The complexity of finding stationary points with stochastic gradient descent,” in International Conference on Machine Learning. PMLR, 2020, pp. 2658–2667. \n[36] D. A. E. Acar, Y. Zhao, R. Matas, M. Mattina, P. Whatmough, and V. Saligrama, “Federated learning based on dynamic regularization,” in International Conference on Learning Representations, 2020. \n[37] S. Caldas, S. M. K. Duddu, P. Wu, T. Li, J. Konecnˇ ý, H. B. McMahan, V. Smith, and A. Talwalkar, “Leaf: A benchmark for federated settings,” 2019. ", + "bbox": [ + 173, + 90, + 828, + 920 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 68, + 828, + 909 + ], + "page_idx": 11 + } +] \ No newline at end of file diff --git a/parse/train/J28lNO4p3ki/J28lNO4p3ki_middle.json b/parse/train/J28lNO4p3ki/J28lNO4p3ki_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..62206b7534c05262e73e183979e7d1b12053b195 --- /dev/null +++ b/parse/train/J28lNO4p3ki/J28lNO4p3ki_middle.json @@ -0,0 +1,43614 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 97, + 504, + 157 + ], + "lines": [ + { + "bbox": [ + 142, + 97, + 468, + 116 + ], + "spans": [ + { + "bbox": [ + 142, + 97, + 468, + 116 + ], + "score": 1.0, + "content": "STEM: A Stochastic Two-Sided Momentum", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 127, + 117, + 484, + 138 + ], + "spans": [ + { + "bbox": [ + 127, + 117, + 484, + 138 + ], + "score": 1.0, + "content": "Algorithm Achieving Near-Optimal Sample and", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 136, + 506, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 136, + 506, + 160 + ], + "score": 1.0, + "content": "Communication Complexities for Federated Learning", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 165, + 199, + 263, + 232 + ], + "lines": [ + { + "bbox": [ + 171, + 200, + 256, + 210 + ], + "spans": [ + { + "bbox": [ + 171, + 200, + 256, + 210 + ], + "score": 1.0, + "content": "Prashant Khanduri", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 164, + 210, + 263, + 222 + ], + "spans": [ + { + "bbox": [ + 164, + 210, + 263, + 222 + ], + "score": 1.0, + "content": "University of Minnesota", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 170, + 221, + 257, + 232 + ], + "spans": [ + { + "bbox": [ + 170, + 221, + 257, + 232 + ], + "score": 1.0, + "content": "khand095@umn.edu", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 324, + 199, + 447, + 232 + ], + "lines": [ + { + "bbox": [ + 350, + 199, + 421, + 212 + ], + "spans": [ + { + "bbox": [ + 350, + 199, + 421, + 212 + ], + "score": 1.0, + "content": "Pranay Sharma", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 328, + 208, + 443, + 225 + ], + "spans": [ + { + "bbox": [ + 328, + 208, + 443, + 225 + ], + "score": 1.0, + "content": "Carnegie Mellon University", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 323, + 221, + 447, + 234 + ], + "spans": [ + { + "bbox": [ + 323, + 221, + 447, + 234 + ], + "score": 1.0, + "content": "pranaysh@andrew.cmu.edu", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 114, + 249, + 267, + 282 + ], + "lines": [ + { + "bbox": [ + 164, + 247, + 219, + 262 + ], + "spans": [ + { + "bbox": [ + 164, + 247, + 219, + 262 + ], + "score": 1.0, + "content": "Haibo Yang", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 137, + 259, + 245, + 273 + ], + "spans": [ + { + "bbox": [ + 137, + 259, + 245, + 273 + ], + "score": 1.0, + "content": "The Ohio State University", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 112, + 271, + 269, + 284 + ], + "spans": [ + { + "bbox": [ + 112, + 271, + 269, + 284 + ], + "score": 1.0, + "content": "yang.5952@buckeyemail.osu.edu", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 279, + 249, + 378, + 282 + ], + "lines": [ + { + "bbox": [ + 299, + 248, + 361, + 262 + ], + "spans": [ + { + "bbox": [ + 299, + 248, + 361, + 262 + ], + "score": 1.0, + "content": "Mingyi Hong⇤", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 279, + 259, + 379, + 273 + ], + "spans": [ + { + "bbox": [ + 279, + 259, + 379, + 273 + ], + "score": 1.0, + "content": "University of Minnesota", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 293, + 271, + 365, + 284 + ], + "spans": [ + { + "bbox": [ + 293, + 271, + 365, + 284 + ], + "score": 1.0, + "content": "mhong@umn.edu", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 390, + 249, + 497, + 282 + ], + "lines": [ + { + "bbox": [ + 425, + 248, + 461, + 261 + ], + "spans": [ + { + "bbox": [ + 425, + 248, + 461, + 261 + ], + "score": 1.0, + "content": "Jia Liu", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 389, + 259, + 497, + 273 + ], + "spans": [ + { + "bbox": [ + 389, + 259, + 497, + 273 + ], + "score": 1.0, + "content": "The Ohio State University", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 403, + 271, + 484, + 283 + ], + "spans": [ + { + "bbox": [ + 403, + 271, + 484, + 283 + ], + "score": 1.0, + "content": "liu@ece.osu.edu", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14 + }, + { + "type": "text", + "bbox": [ + 155, + 299, + 311, + 332 + ], + "lines": [ + { + "bbox": [ + 199, + 299, + 267, + 311 + ], + "spans": [ + { + "bbox": [ + 199, + 299, + 267, + 311 + ], + "score": 1.0, + "content": "Ketan Rajawat", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 154, + 309, + 311, + 323 + ], + "spans": [ + { + "bbox": [ + 154, + 309, + 311, + 323 + ], + "score": 1.0, + "content": "Indian Institute of Technology Kanpur", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 190, + 322, + 276, + 332 + ], + "spans": [ + { + "bbox": [ + 190, + 322, + 276, + 332 + ], + "score": 1.0, + "content": "ketan@iitk.ac.in", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 363, + 299, + 455, + 333 + ], + "lines": [ + { + "bbox": [ + 363, + 297, + 456, + 313 + ], + "spans": [ + { + "bbox": [ + 363, + 297, + 456, + 313 + ], + "score": 1.0, + "content": "Pramod K. Varshney", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 367, + 309, + 451, + 323 + ], + "spans": [ + { + "bbox": [ + 367, + 309, + 451, + 323 + ], + "score": 1.0, + "content": "Syracuse University", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 366, + 322, + 453, + 334 + ], + "spans": [ + { + "bbox": [ + 366, + 322, + 453, + 334 + ], + "score": 1.0, + "content": "varshney@syr.edu", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22 + }, + { + "type": "title", + "bbox": [ + 283, + 361, + 328, + 374 + ], + "lines": [ + { + "bbox": [ + 281, + 360, + 331, + 375 + ], + "spans": [ + { + "bbox": [ + 281, + 360, + 331, + 375 + ], + "score": 1.0, + "content": "Abstract", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 142, + 385, + 469, + 606 + ], + "lines": [ + { + "bbox": [ + 141, + 386, + 469, + 398 + ], + "spans": [ + { + "bbox": [ + 141, + 386, + 469, + 398 + ], + "score": 1.0, + "content": "Federated Learning (FL) refers to the paradigm where multiple worker nodes", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 396, + 470, + 409 + ], + "spans": [ + { + "bbox": [ + 141, + 396, + 470, + 409 + ], + "score": 1.0, + "content": "(WNs) build a joint model by using local data. Despite extensive research, for a", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 141, + 408, + 470, + 419 + ], + "spans": [ + { + "bbox": [ + 141, + 408, + 470, + 419 + ], + "score": 1.0, + "content": "generic non-convex FL problem, it is not clear, how to choose the WNs’ and the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 141, + 419, + 469, + 431 + ], + "spans": [ + { + "bbox": [ + 141, + 419, + 469, + 431 + ], + "score": 1.0, + "content": "server’s update directions, the minibatch sizes, and the number of local updates, so", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 141, + 429, + 469, + 442 + ], + "spans": [ + { + "bbox": [ + 141, + 429, + 469, + 442 + ], + "score": 1.0, + "content": "that the WNs use the minimum number of samples and communication rounds to", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 141, + 440, + 469, + 452 + ], + "spans": [ + { + "bbox": [ + 141, + 440, + 469, + 452 + ], + "score": 1.0, + "content": "achieve the desired solution. This work addresses the above question and considers", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 142, + 452, + 469, + 463 + ], + "spans": [ + { + "bbox": [ + 142, + 452, + 469, + 463 + ], + "score": 1.0, + "content": "a class of stochastic algorithms where the WNs perform a few local updates before", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 141, + 462, + 469, + 475 + ], + "spans": [ + { + "bbox": [ + 141, + 462, + 469, + 475 + ], + "score": 1.0, + "content": "communication. We show that when both the WN’s and the server’s directions are", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 142, + 473, + 469, + 486 + ], + "spans": [ + { + "bbox": [ + 142, + 473, + 469, + 486 + ], + "score": 1.0, + "content": "chosen based on certain stochastic momentum estimator, the algorithm requires", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 142, + 483, + 470, + 501 + ], + "spans": [ + { + "bbox": [ + 142, + 484, + 182, + 497 + ], + "score": 0.92, + "content": "\\tilde { \\mathcal O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 483, + 235, + 501 + ], + "score": 1.0, + "content": "samples and", + "type": "text" + }, + { + "bbox": [ + 235, + 484, + 267, + 497 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 483, + 420, + 501 + ], + "score": 1.0, + "content": "communication rounds to compute an", + "type": "text" + }, + { + "bbox": [ + 420, + 488, + 426, + 496 + ], + "score": 0.6, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 483, + 470, + 501 + ], + "score": 1.0, + "content": "-stationary", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 141, + 496, + 470, + 509 + ], + "spans": [ + { + "bbox": [ + 141, + 496, + 470, + 509 + ], + "score": 1.0, + "content": "solution. To the best of our knowledge, this is the first FL algorithm that achieves", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 142, + 508, + 470, + 519 + ], + "spans": [ + { + "bbox": [ + 142, + 508, + 470, + 519 + ], + "score": 1.0, + "content": "such near-optimal sample and communication complexities simultaneously. Fur-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 141, + 518, + 469, + 531 + ], + "spans": [ + { + "bbox": [ + 141, + 518, + 469, + 531 + ], + "score": 1.0, + "content": "ther, we show that there is a trade-off curve between the number of local updates", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 141, + 529, + 471, + 542 + ], + "spans": [ + { + "bbox": [ + 141, + 529, + 471, + 542 + ], + "score": 1.0, + "content": "and the minibatch sizes, on which the above sample and communication complexi-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 141, + 540, + 470, + 552 + ], + "spans": [ + { + "bbox": [ + 141, + 540, + 470, + 552 + ], + "score": 1.0, + "content": "ties can be maintained. Finally, we show that for the classical FedAvg (a.k.a. Local", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 141, + 551, + 470, + 563 + ], + "spans": [ + { + "bbox": [ + 141, + 551, + 470, + 563 + ], + "score": 1.0, + "content": "SGD, which is a momentum-less special case of the STEM), a similar trade-off", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 141, + 563, + 470, + 574 + ], + "spans": [ + { + "bbox": [ + 141, + 563, + 470, + 574 + ], + "score": 1.0, + "content": "curve exists, albeit with worse sample and communication complexities. Our", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 141, + 573, + 469, + 585 + ], + "spans": [ + { + "bbox": [ + 141, + 573, + 469, + 585 + ], + "score": 1.0, + "content": "insights on this trade-off provides guidelines for choosing the four important design", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 142, + 583, + 469, + 595 + ], + "spans": [ + { + "bbox": [ + 142, + 583, + 469, + 595 + ], + "score": 1.0, + "content": "elements for FL algorithms, the number of local updates, WNs’ and server’s update", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 141, + 594, + 399, + 607 + ], + "spans": [ + { + "bbox": [ + 141, + 594, + 399, + 607 + ], + "score": 1.0, + "content": "directions, and minibatch sizes to achieve the best performance.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 34.5 + }, + { + "type": "title", + "bbox": [ + 107, + 626, + 190, + 640 + ], + "lines": [ + { + "bbox": [ + 105, + 624, + 192, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 192, + 642 + ], + "score": 1.0, + "content": "1 Introduction", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 45 + }, + { + "type": "text", + "bbox": [ + 107, + 651, + 504, + 695 + ], + "lines": [ + { + "bbox": [ + 105, + 650, + 506, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 506, + 664 + ], + "score": 1.0, + "content": "In Federated Learning (FL), multiple worker nodes (WNs) collaborate with the goal of learning a", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 663, + 506, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 663, + 506, + 674 + ], + "score": 1.0, + "content": "joint model, by only using local data. Therefore it has become popular for machine learning problems", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 673, + 506, + 685 + ], + "spans": [ + { + "bbox": [ + 105, + 673, + 506, + 685 + ], + "score": 1.0, + "content": "where datasets are massively distributed [1]. In FL, the data is often collected at or off-loaded to", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 684, + 506, + 696 + ], + "spans": [ + { + "bbox": [ + 105, + 684, + 506, + 696 + ], + "score": 1.0, + "content": "multiple WNs which in collaboration with a server node (SN) jointly aim to learn a centralized model", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 47.5 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 119, + 703, + 260, + 713 + ], + "lines": [ + { + "bbox": [ + 118, + 700, + 262, + 716 + ], + "spans": [ + { + "bbox": [ + 118, + 700, + 262, + 716 + ], + "score": 1.0, + "content": "⇤Corresponding Author: Mingyi Hong.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 106, + 731, + 385, + 742 + ], + "lines": [ + { + "bbox": [ + 105, + 730, + 386, + 743 + ], + "spans": [ + { + "bbox": [ + 105, + 730, + 386, + 743 + ], + "score": 1.0, + "content": "35th Conference on Neural Information Processing Systems (NeurIPS 2021).", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 97, + 504, + 157 + ], + "lines": [ + { + "bbox": [ + 142, + 97, + 468, + 116 + ], + "spans": [ + { + "bbox": [ + 142, + 97, + 468, + 116 + ], + "score": 1.0, + "content": "STEM: A Stochastic Two-Sided Momentum", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 127, + 117, + 484, + 138 + ], + "spans": [ + { + "bbox": [ + 127, + 117, + 484, + 138 + ], + "score": 1.0, + "content": "Algorithm Achieving Near-Optimal Sample and", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 136, + 506, + 160 + ], + "spans": [ + { + "bbox": [ + 105, + 136, + 506, + 160 + ], + "score": 1.0, + "content": "Communication Complexities for Federated Learning", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 165, + 199, + 263, + 232 + ], + "lines": [ + { + "bbox": [ + 171, + 200, + 256, + 210 + ], + "spans": [ + { + "bbox": [ + 171, + 200, + 256, + 210 + ], + "score": 1.0, + "content": "Prashant Khanduri", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 164, + 210, + 263, + 222 + ], + "spans": [ + { + "bbox": [ + 164, + 210, + 263, + 222 + ], + "score": 1.0, + "content": "University of Minnesota", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 170, + 221, + 257, + 232 + ], + "spans": [ + { + "bbox": [ + 170, + 221, + 257, + 232 + ], + "score": 1.0, + "content": "khand095@umn.edu", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 5, + "bbox_fs": [ + 164, + 200, + 263, + 232 + ] + }, + { + "type": "text", + "bbox": [ + 324, + 199, + 447, + 232 + ], + "lines": [ + { + "bbox": [ + 350, + 199, + 421, + 212 + ], + "spans": [ + { + "bbox": [ + 350, + 199, + 421, + 212 + ], + "score": 1.0, + "content": "Pranay Sharma", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 328, + 208, + 443, + 225 + ], + "spans": [ + { + "bbox": [ + 328, + 208, + 443, + 225 + ], + "score": 1.0, + "content": "Carnegie Mellon University", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 323, + 221, + 447, + 234 + ], + "spans": [ + { + "bbox": [ + 323, + 221, + 447, + 234 + ], + "score": 1.0, + "content": "pranaysh@andrew.cmu.edu", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 6, + "bbox_fs": [ + 323, + 199, + 447, + 234 + ] + }, + { + "type": "text", + "bbox": [ + 114, + 249, + 267, + 282 + ], + "lines": [ + { + "bbox": [ + 164, + 247, + 219, + 262 + ], + "spans": [ + { + "bbox": [ + 164, + 247, + 219, + 262 + ], + "score": 1.0, + "content": "Haibo Yang", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 137, + 259, + 245, + 273 + ], + "spans": [ + { + "bbox": [ + 137, + 259, + 245, + 273 + ], + "score": 1.0, + "content": "The Ohio State University", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 112, + 271, + 269, + 284 + ], + "spans": [ + { + "bbox": [ + 112, + 271, + 269, + 284 + ], + "score": 1.0, + "content": "yang.5952@buckeyemail.osu.edu", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12, + "bbox_fs": [ + 112, + 247, + 269, + 284 + ] + }, + { + "type": "text", + "bbox": [ + 279, + 249, + 378, + 282 + ], + "lines": [ + { + "bbox": [ + 299, + 248, + 361, + 262 + ], + "spans": [ + { + "bbox": [ + 299, + 248, + 361, + 262 + ], + "score": 1.0, + "content": "Mingyi Hong⇤", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 279, + 259, + 379, + 273 + ], + "spans": [ + { + "bbox": [ + 279, + 259, + 379, + 273 + ], + "score": 1.0, + "content": "University of Minnesota", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 293, + 271, + 365, + 284 + ], + "spans": [ + { + "bbox": [ + 293, + 271, + 365, + 284 + ], + "score": 1.0, + "content": "mhong@umn.edu", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 13, + "bbox_fs": [ + 279, + 248, + 379, + 284 + ] + }, + { + "type": "text", + "bbox": [ + 390, + 249, + 497, + 282 + ], + "lines": [ + { + "bbox": [ + 425, + 248, + 461, + 261 + ], + "spans": [ + { + "bbox": [ + 425, + 248, + 461, + 261 + ], + "score": 1.0, + "content": "Jia Liu", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 389, + 259, + 497, + 273 + ], + "spans": [ + { + "bbox": [ + 389, + 259, + 497, + 273 + ], + "score": 1.0, + "content": "The Ohio State University", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 403, + 271, + 484, + 283 + ], + "spans": [ + { + "bbox": [ + 403, + 271, + 484, + 283 + ], + "score": 1.0, + "content": "liu@ece.osu.edu", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14, + "bbox_fs": [ + 389, + 248, + 497, + 283 + ] + }, + { + "type": "text", + "bbox": [ + 155, + 299, + 311, + 332 + ], + "lines": [ + { + "bbox": [ + 199, + 299, + 267, + 311 + ], + "spans": [ + { + "bbox": [ + 199, + 299, + 267, + 311 + ], + "score": 1.0, + "content": "Ketan Rajawat", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 154, + 309, + 311, + 323 + ], + "spans": [ + { + "bbox": [ + 154, + 309, + 311, + 323 + ], + "score": 1.0, + "content": "Indian Institute of Technology Kanpur", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 190, + 322, + 276, + 332 + ], + "spans": [ + { + "bbox": [ + 190, + 322, + 276, + 332 + ], + "score": 1.0, + "content": "ketan@iitk.ac.in", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19, + "bbox_fs": [ + 154, + 299, + 311, + 332 + ] + }, + { + "type": "text", + "bbox": [ + 363, + 299, + 455, + 333 + ], + "lines": [ + { + "bbox": [ + 363, + 297, + 456, + 313 + ], + "spans": [ + { + "bbox": [ + 363, + 297, + 456, + 313 + ], + "score": 1.0, + "content": "Pramod K. Varshney", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 367, + 309, + 451, + 323 + ], + "spans": [ + { + "bbox": [ + 367, + 309, + 451, + 323 + ], + "score": 1.0, + "content": "Syracuse University", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 366, + 322, + 453, + 334 + ], + "spans": [ + { + "bbox": [ + 366, + 322, + 453, + 334 + ], + "score": 1.0, + "content": "varshney@syr.edu", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22, + "bbox_fs": [ + 363, + 297, + 456, + 334 + ] + }, + { + "type": "title", + "bbox": [ + 283, + 361, + 328, + 374 + ], + "lines": [ + { + "bbox": [ + 281, + 360, + 331, + 375 + ], + "spans": [ + { + "bbox": [ + 281, + 360, + 331, + 375 + ], + "score": 1.0, + "content": "Abstract", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 142, + 385, + 469, + 606 + ], + "lines": [ + { + "bbox": [ + 141, + 386, + 469, + 398 + ], + "spans": [ + { + "bbox": [ + 141, + 386, + 469, + 398 + ], + "score": 1.0, + "content": "Federated Learning (FL) refers to the paradigm where multiple worker nodes", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 141, + 396, + 470, + 409 + ], + "spans": [ + { + "bbox": [ + 141, + 396, + 470, + 409 + ], + "score": 1.0, + "content": "(WNs) build a joint model by using local data. Despite extensive research, for a", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 141, + 408, + 470, + 419 + ], + "spans": [ + { + "bbox": [ + 141, + 408, + 470, + 419 + ], + "score": 1.0, + "content": "generic non-convex FL problem, it is not clear, how to choose the WNs’ and the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 141, + 419, + 469, + 431 + ], + "spans": [ + { + "bbox": [ + 141, + 419, + 469, + 431 + ], + "score": 1.0, + "content": "server’s update directions, the minibatch sizes, and the number of local updates, so", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 141, + 429, + 469, + 442 + ], + "spans": [ + { + "bbox": [ + 141, + 429, + 469, + 442 + ], + "score": 1.0, + "content": "that the WNs use the minimum number of samples and communication rounds to", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 141, + 440, + 469, + 452 + ], + "spans": [ + { + "bbox": [ + 141, + 440, + 469, + 452 + ], + "score": 1.0, + "content": "achieve the desired solution. This work addresses the above question and considers", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 142, + 452, + 469, + 463 + ], + "spans": [ + { + "bbox": [ + 142, + 452, + 469, + 463 + ], + "score": 1.0, + "content": "a class of stochastic algorithms where the WNs perform a few local updates before", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 141, + 462, + 469, + 475 + ], + "spans": [ + { + "bbox": [ + 141, + 462, + 469, + 475 + ], + "score": 1.0, + "content": "communication. We show that when both the WN’s and the server’s directions are", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 142, + 473, + 469, + 486 + ], + "spans": [ + { + "bbox": [ + 142, + 473, + 469, + 486 + ], + "score": 1.0, + "content": "chosen based on certain stochastic momentum estimator, the algorithm requires", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 142, + 483, + 470, + 501 + ], + "spans": [ + { + "bbox": [ + 142, + 484, + 182, + 497 + ], + "score": 0.92, + "content": "\\tilde { \\mathcal O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 483, + 235, + 501 + ], + "score": 1.0, + "content": "samples and", + "type": "text" + }, + { + "bbox": [ + 235, + 484, + 267, + 497 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 483, + 420, + 501 + ], + "score": 1.0, + "content": "communication rounds to compute an", + "type": "text" + }, + { + "bbox": [ + 420, + 488, + 426, + 496 + ], + "score": 0.6, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 483, + 470, + 501 + ], + "score": 1.0, + "content": "-stationary", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 141, + 496, + 470, + 509 + ], + "spans": [ + { + "bbox": [ + 141, + 496, + 470, + 509 + ], + "score": 1.0, + "content": "solution. To the best of our knowledge, this is the first FL algorithm that achieves", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 142, + 508, + 470, + 519 + ], + "spans": [ + { + "bbox": [ + 142, + 508, + 470, + 519 + ], + "score": 1.0, + "content": "such near-optimal sample and communication complexities simultaneously. Fur-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 141, + 518, + 469, + 531 + ], + "spans": [ + { + "bbox": [ + 141, + 518, + 469, + 531 + ], + "score": 1.0, + "content": "ther, we show that there is a trade-off curve between the number of local updates", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 141, + 529, + 471, + 542 + ], + "spans": [ + { + "bbox": [ + 141, + 529, + 471, + 542 + ], + "score": 1.0, + "content": "and the minibatch sizes, on which the above sample and communication complexi-", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 141, + 540, + 470, + 552 + ], + "spans": [ + { + "bbox": [ + 141, + 540, + 470, + 552 + ], + "score": 1.0, + "content": "ties can be maintained. Finally, we show that for the classical FedAvg (a.k.a. Local", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 141, + 551, + 470, + 563 + ], + "spans": [ + { + "bbox": [ + 141, + 551, + 470, + 563 + ], + "score": 1.0, + "content": "SGD, which is a momentum-less special case of the STEM), a similar trade-off", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 141, + 563, + 470, + 574 + ], + "spans": [ + { + "bbox": [ + 141, + 563, + 470, + 574 + ], + "score": 1.0, + "content": "curve exists, albeit with worse sample and communication complexities. Our", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 141, + 573, + 469, + 585 + ], + "spans": [ + { + "bbox": [ + 141, + 573, + 469, + 585 + ], + "score": 1.0, + "content": "insights on this trade-off provides guidelines for choosing the four important design", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 142, + 583, + 469, + 595 + ], + "spans": [ + { + "bbox": [ + 142, + 583, + 469, + 595 + ], + "score": 1.0, + "content": "elements for FL algorithms, the number of local updates, WNs’ and server’s update", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 141, + 594, + 399, + 607 + ], + "spans": [ + { + "bbox": [ + 141, + 594, + 399, + 607 + ], + "score": 1.0, + "content": "directions, and minibatch sizes to achieve the best performance.", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 34.5, + "bbox_fs": [ + 141, + 386, + 471, + 607 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 626, + 190, + 640 + ], + "lines": [ + { + "bbox": [ + 105, + 624, + 192, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 624, + 192, + 642 + ], + "score": 1.0, + "content": "1 Introduction", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 45 + }, + { + "type": "text", + "bbox": [ + 107, + 651, + 504, + 695 + ], + "lines": [ + { + "bbox": [ + 105, + 650, + 506, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 506, + 664 + ], + "score": 1.0, + "content": "In Federated Learning (FL), multiple worker nodes (WNs) collaborate with the goal of learning a", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 663, + 506, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 663, + 506, + 674 + ], + "score": 1.0, + "content": "joint model, by only using local data. Therefore it has become popular for machine learning problems", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 673, + 506, + 685 + ], + "spans": [ + { + "bbox": [ + 105, + 673, + 506, + 685 + ], + "score": 1.0, + "content": "where datasets are massively distributed [1]. In FL, the data is often collected at or off-loaded to", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 684, + 506, + 696 + ], + "spans": [ + { + "bbox": [ + 105, + 684, + 506, + 696 + ], + "score": 1.0, + "content": "multiple WNs which in collaboration with a server node (SN) jointly aim to learn a centralized model", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 316, + 505, + 328 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 505, + 328 + ], + "score": 1.0, + "content": "[2, 3]. The local WNs share the computational load and since the data is local to each WN, FL also", + "type": "text", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 328, + 505, + 339 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 253, + 339 + ], + "score": 1.0, + "content": "provides some level of data privacy", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 253, + 328, + 265, + 339 + ], + "score": 0.5, + "content": "\\mathbb { \\lVert \\rVert }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 266, + 328, + 470, + 339 + ], + "score": 1.0, + "content": ". A classical distributed optimization problem that", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 470, + 328, + 480, + 338 + ], + "score": 0.79, + "content": "K", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 481, + 328, + 505, + 339 + ], + "score": 1.0, + "content": "WNs", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 339, + 160, + 350 + ], + "spans": [ + { + "bbox": [ + 106, + 339, + 160, + 350 + ], + "score": 1.0, + "content": "aim to solve:", + "type": "text", + "cross_page": true + } + ], + "index": 13 + } + ], + "index": 47.5, + "bbox_fs": [ + 105, + 650, + 506, + 696 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 126, + 73, + 473, + 207 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 126, + 73, + 473, + 207 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 126, + 73, + 473, + 207 + ], + "spans": [ + { + "bbox": [ + 126, + 73, + 473, + 207 + ], + "score": 0.969, + "type": "image", + "image_path": "8b109c70673d05d4fb4863253d56729a5c2b37b6a73924b30a5e660d7b6d1078.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 126, + 73, + 473, + 117.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 126, + 117.66666666666666, + 473, + 162.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 126, + 162.33333333333331, + 473, + 206.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 214, + 505, + 297 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 213, + 506, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 506, + 227 + ], + "score": 1.0, + "content": "Figure 1: The 3D surface in (a) plots the communication complexity of the proposed STEM for different", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 224, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 106, + 224, + 505, + 236 + ], + "score": 1.0, + "content": "minibatch sizes and number of local updates. The surface is generated such that each point represents STEM with", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 233, + 506, + 249 + ], + "spans": [ + { + "bbox": [ + 104, + 233, + 186, + 249 + ], + "score": 1.0, + "content": "a particular choice of", + "type": "text" + }, + { + "bbox": [ + 186, + 236, + 207, + 246 + ], + "score": 0.91, + "content": "( b , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 233, + 276, + 249 + ], + "score": 1.0, + "content": ", so that it requires", + "type": "text" + }, + { + "bbox": [ + 276, + 234, + 312, + 247 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 233, + 384, + 249 + ], + "score": 1.0, + "content": "samples to achieve", + "type": "text" + }, + { + "bbox": [ + 385, + 238, + 389, + 245 + ], + "score": 0.72, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 233, + 506, + 249 + ], + "score": 1.0, + "content": "-stationarity. Plot (b) shows the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 245, + 506, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 506, + 258 + ], + "score": 1.0, + "content": "optimal trade off between the minibatch sizes and the number of local updates at each WN (i.e., achieving the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 254, + 506, + 268 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 451, + 268 + ], + "score": 1.0, + "content": "lowest communication and sample complexities). Both plots are generated for an accuracy of", + "type": "text" + }, + { + "bbox": [ + 451, + 255, + 488, + 265 + ], + "score": 0.91, + "content": "\\epsilon = 1 0 ^ { - 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 488, + 254, + 506, + 268 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 264, + 506, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 506, + 279 + ], + "score": 1.0, + "content": "all the constants dependent on system parameters (variance of stochastic gradients, heterogeneity parameter,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 275, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 505, + 288 + ], + "score": 1.0, + "content": "optimality gap, Lipschitz constants, etc.) are assumed to be 1. Fed STEM is a special case of STEM where", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 285, + 488, + 298 + ], + "spans": [ + { + "bbox": [ + 107, + 286, + 127, + 297 + ], + "score": 0.85, + "content": "\\mathcal { O } ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 285, + 382, + 298 + ], + "score": 1.0, + "content": "minibatch is used; Minibatch STEM is a special case of STEM where", + "type": "text" + }, + { + "bbox": [ + 383, + 286, + 402, + 297 + ], + "score": 0.88, + "content": "\\mathcal { O } ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 285, + 488, + 298 + ], + "score": 1.0, + "content": "local updates are used.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 6.5 + } + ], + "index": 3.75 + }, + { + "type": "text", + "bbox": [ + 107, + 316, + 506, + 349 + ], + "lines": [ + { + "bbox": [ + 106, + 316, + 505, + 328 + ], + "spans": [ + { + "bbox": [ + 106, + 316, + 505, + 328 + ], + "score": 1.0, + "content": "[2, 3]. The local WNs share the computational load and since the data is local to each WN, FL also", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 328, + 505, + 339 + ], + "spans": [ + { + "bbox": [ + 105, + 328, + 253, + 339 + ], + "score": 1.0, + "content": "provides some level of data privacy", + "type": "text" + }, + { + "bbox": [ + 253, + 328, + 265, + 339 + ], + "score": 0.5, + "content": "\\mathbb { \\lVert \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 328, + 470, + 339 + ], + "score": 1.0, + "content": ". A classical distributed optimization problem that", + "type": "text" + }, + { + "bbox": [ + 470, + 328, + 480, + 338 + ], + "score": 0.79, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 481, + 328, + 505, + 339 + ], + "score": 1.0, + "content": "WNs", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 339, + 160, + 350 + ], + "spans": [ + { + "bbox": [ + 106, + 339, + 160, + 350 + ], + "score": 1.0, + "content": "aim to solve:", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12 + }, + { + "type": "interline_equation", + "bbox": [ + 163, + 352, + 447, + 387 + ], + "lines": [ + { + "bbox": [ + 163, + 352, + 447, + 387 + ], + "spans": [ + { + "bbox": [ + 163, + 352, + 447, + 387 + ], + "score": 0.95, + "content": "\\operatorname* { m i n } _ { x \\in \\mathbb { R } ^ { d } } \\bigg \\{ f ( x ) : = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } f ^ { ( k ) } ( x ) : = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } \\mathbb { E } _ { \\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) } } \\big [ f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) \\big ] \\bigg \\} .", + "type": "interline_equation", + "image_path": "2805a58b3a852bb789307ddca3e7e2aea04c24beb34df4fd0d38807ed0628944.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 163, + 352, + 447, + 363.6666666666667 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 163, + 363.6666666666667, + 447, + 375.33333333333337 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 163, + 375.33333333333337, + 447, + 387.00000000000006 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 391, + 505, + 427 + ], + "lines": [ + { + "bbox": [ + 104, + 389, + 504, + 405 + ], + "spans": [ + { + "bbox": [ + 104, + 389, + 133, + 405 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 390, + 194, + 403 + ], + "score": 0.93, + "content": "f ^ { ( k ) } : \\mathbb { R } ^ { d } \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 389, + 455, + 405 + ], + "score": 1.0, + "content": "denotes the smooth (possibly non-convex) objective function and", + "type": "text" + }, + { + "bbox": [ + 456, + 390, + 504, + 403 + ], + "score": 0.92, + "content": "\\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) }", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 402, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 309, + 416 + ], + "score": 1.0, + "content": "represents the sample/s drawn from distribution", + "type": "text" + }, + { + "bbox": [ + 309, + 403, + 329, + 414 + ], + "score": 0.9, + "content": "\\mathcal { D } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 402, + 358, + 416 + ], + "score": 1.0, + "content": "at the", + "type": "text" + }, + { + "bbox": [ + 359, + 403, + 372, + 414 + ], + "score": 0.85, + "content": "k ^ { \\mathrm { { t h } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 402, + 416, + 416 + ], + "score": 1.0, + "content": "WN with", + "type": "text" + }, + { + "bbox": [ + 416, + 403, + 454, + 416 + ], + "score": 0.92, + "content": "k \\in [ K ]", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 402, + 505, + 416 + ], + "score": 1.0, + "content": ". When the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 414, + 497, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 159, + 430 + ], + "score": 1.0, + "content": "distributions", + "type": "text" + }, + { + "bbox": [ + 159, + 415, + 178, + 426 + ], + "score": 0.91, + "content": "\\mathcal { D } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 414, + 497, + 430 + ], + "score": 1.0, + "content": "are different across the WNs, it is referred to as the heterogeneous data setting.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 432, + 505, + 521 + ], + "lines": [ + { + "bbox": [ + 105, + 431, + 506, + 445 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 506, + 445 + ], + "score": 1.0, + "content": "The optimization performance of non-convex FL algorithms is typically measured by the total number", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 443, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 256, + 456 + ], + "score": 1.0, + "content": "of samples accessed (cf. Definition", + "type": "text" + }, + { + "bbox": [ + 256, + 443, + 275, + 456 + ], + "score": 0.89, + "content": "2 . 2 )", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 444, + 505, + 455 + ], + "score": 1.0, + "content": "and the total rounds of communication (cf. Definition", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 454, + 505, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 124, + 467 + ], + "score": 0.79, + "content": "2 . 3 )", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 454, + 269, + 467 + ], + "score": 1.0, + "content": "required by each WN to achieve an", + "type": "text" + }, + { + "bbox": [ + 270, + 457, + 275, + 464 + ], + "score": 0.67, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 454, + 414, + 467 + ], + "score": 1.0, + "content": "-stationary solution (cf. Definition", + "type": "text" + }, + { + "bbox": [ + 414, + 454, + 432, + 467 + ], + "score": 0.81, + "content": "2 . 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 454, + 505, + 467 + ], + "score": 1.0, + "content": ". To minimize the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "score": 1.0, + "content": "sample and the communication complexities, FL algorithms rely on the following four key design", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 477, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 505, + 488 + ], + "score": 1.0, + "content": "elements: (i) the WNs’ local model update directions, (ii) Minibatch size to compute each local", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 486, + 506, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 506, + 500 + ], + "score": 1.0, + "content": "direction, (iii) the number of local updates before WNs share their parameters, and (iv) the SN’s", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 498, + 506, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 506, + 511 + ], + "score": 1.0, + "content": "update direction. How to find effective FL algorithms by (optimally) designing these parameters has", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 509, + 290, + 522 + ], + "spans": [ + { + "bbox": [ + 106, + 509, + 290, + 522 + ], + "score": 1.0, + "content": "received significant research interest recently.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 107, + 525, + 385, + 536 + ], + "lines": [ + { + "bbox": [ + 106, + 524, + 385, + 538 + ], + "spans": [ + { + "bbox": [ + 106, + 524, + 385, + 538 + ], + "score": 1.0, + "content": "Contributions. The main contributions of this work are listed below:", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 541, + 505, + 653 + ], + "lines": [ + { + "bbox": [ + 105, + 541, + 506, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 506, + 554 + ], + "score": 1.0, + "content": "1) We propose the Stochastic Two-Sided Momentum (STEM) algorithm, that utilizes certain", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 553, + 505, + 564 + ], + "spans": [ + { + "bbox": [ + 106, + 553, + 505, + 564 + ], + "score": 1.0, + "content": "momentum-assisted stochastic gradient directions for both the WNs and SN updates. We show", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "score": 1.0, + "content": "that there exists an optimal trade off between the minibatch sizes and number of local updates, such", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 573, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 274, + 588 + ], + "score": 1.0, + "content": "that on the trade-off curve STEM requires", + "type": "text" + }, + { + "bbox": [ + 275, + 573, + 320, + 588 + ], + "score": 0.91, + "content": "\\underline { { \\tilde { \\mathcal { O } } } } ( \\epsilon ^ { - 3 / 2 } ) \\underline { { \\left[ \\frac { 2 } { } \\right] } }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 575, + 370, + 588 + ], + "score": 1.0, + "content": "samples and", + "type": "text" + }, + { + "bbox": [ + 370, + 574, + 401, + 588 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 575, + 505, + 588 + ], + "score": 1.0, + "content": "communication rounds to", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 587, + 506, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 142, + 600 + ], + "score": 1.0, + "content": "reach an", + "type": "text" + }, + { + "bbox": [ + 142, + 589, + 148, + 597 + ], + "score": 0.57, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 587, + 271, + 600 + ], + "score": 1.0, + "content": "-stationary solution; see Figure", + "type": "text" + }, + { + "bbox": [ + 272, + 587, + 281, + 599 + ], + "score": 0.57, + "content": "\\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 587, + 506, + 600 + ], + "score": 1.0, + "content": "for an illustration. These complexity results are the best", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 597, + 505, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 490, + 610 + ], + "score": 1.0, + "content": "achievable for first-order stochastic FL algorithms (under certain assumptions, cf. Assumption", + "type": "text" + }, + { + "bbox": [ + 491, + 597, + 502, + 611 + ], + "score": 0.43, + "content": "\\bigstar \\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 597, + 505, + 610 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 607, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 607, + 122, + 621 + ], + "score": 1.0, + "content": "see", + "type": "text" + }, + { + "bbox": [ + 122, + 608, + 145, + 620 + ], + "score": 0.42, + "content": "\\pm \\boxed { 5 } \\boxed { 8 } \\parallel", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 607, + 163, + 621 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 163, + 608, + 190, + 620 + ], + "score": 0.35, + "content": "\\mathbb { B } \\mathbb { n o }", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 607, + 268, + 621 + ], + "score": 1.0, + "content": ", as well as Remark", + "type": "text" + }, + { + "bbox": [ + 268, + 608, + 277, + 621 + ], + "score": 0.45, + "content": "\\checkmark", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 607, + 505, + 621 + ], + "score": 1.0, + "content": "of this paper for discussions regarding optimality. To the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 620, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 106, + 620, + 505, + 631 + ], + "score": 1.0, + "content": "best of our knowledge, STEM is the first algorithm which – (i) simultaneously achieves the optimal", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 631, + 505, + 642 + ], + "spans": [ + { + "bbox": [ + 106, + 631, + 505, + 642 + ], + "score": 1.0, + "content": "sample and communication complexities for FL and (ii) can optimally trade off the minibatch sizes", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 641, + 239, + 654 + ], + "spans": [ + { + "bbox": [ + 106, + 641, + 239, + 654 + ], + "score": 1.0, + "content": "and the number of local updates.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 107, + 658, + 505, + 703 + ], + "lines": [ + { + "bbox": [ + 105, + 657, + 506, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 506, + 670 + ], + "score": 1.0, + "content": "2) A momentum-less special case of our STEM result further reveals some interesting insights of the", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 667, + 507, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 507, + 682 + ], + "score": 1.0, + "content": "classical FedAvg algorithm (a.k.a. the Local SGD) [11–13]. Specifically, we show that for FedAvg,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 680, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 106, + 680, + 506, + 691 + ], + "score": 1.0, + "content": "there also exists a trade-off between the minibatch sizes and the number of local updates, such that it", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 690, + 503, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 690, + 141, + 705 + ], + "score": 1.0, + "content": "requires", + "type": "text" + }, + { + "bbox": [ + 141, + 691, + 173, + 704 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 690, + 226, + 705 + ], + "score": 1.0, + "content": "samples and", + "type": "text" + }, + { + "bbox": [ + 226, + 690, + 266, + 704 + ], + "score": 0.92, + "content": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 690, + 415, + 705 + ], + "score": 1.0, + "content": "communication rounds to achieve an", + "type": "text" + }, + { + "bbox": [ + 416, + 694, + 421, + 702 + ], + "score": 0.74, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 690, + 503, + 705 + ], + "score": 1.0, + "content": "-stationary solution.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 119, + 711, + 295, + 722 + ], + "lines": [ + { + "bbox": [ + 118, + 708, + 298, + 724 + ], + "spans": [ + { + "bbox": [ + 118, + 708, + 171, + 724 + ], + "score": 1.0, + "content": "2 The notation", + "type": "text" + }, + { + "bbox": [ + 171, + 710, + 189, + 722 + ], + "score": 0.88, + "content": "\\tilde { \\mathcal { O } } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 708, + 298, + 724 + ], + "score": 1.0, + "content": "hides the logarithmic factors.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 753 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 753 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 126, + 73, + 473, + 207 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 126, + 73, + 473, + 207 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 126, + 73, + 473, + 207 + ], + "spans": [ + { + "bbox": [ + 126, + 73, + 473, + 207 + ], + "score": 0.969, + "type": "image", + "image_path": "8b109c70673d05d4fb4863253d56729a5c2b37b6a73924b30a5e660d7b6d1078.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 126, + 73, + 473, + 117.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 126, + 117.66666666666666, + 473, + 162.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 126, + 162.33333333333331, + 473, + 206.99999999999997 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 214, + 505, + 297 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 213, + 506, + 227 + ], + "spans": [ + { + "bbox": [ + 105, + 213, + 506, + 227 + ], + "score": 1.0, + "content": "Figure 1: The 3D surface in (a) plots the communication complexity of the proposed STEM for different", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 224, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 106, + 224, + 505, + 236 + ], + "score": 1.0, + "content": "minibatch sizes and number of local updates. The surface is generated such that each point represents STEM with", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 233, + 506, + 249 + ], + "spans": [ + { + "bbox": [ + 104, + 233, + 186, + 249 + ], + "score": 1.0, + "content": "a particular choice of", + "type": "text" + }, + { + "bbox": [ + 186, + 236, + 207, + 246 + ], + "score": 0.91, + "content": "( b , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 233, + 276, + 249 + ], + "score": 1.0, + "content": ", so that it requires", + "type": "text" + }, + { + "bbox": [ + 276, + 234, + 312, + 247 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 233, + 384, + 249 + ], + "score": 1.0, + "content": "samples to achieve", + "type": "text" + }, + { + "bbox": [ + 385, + 238, + 389, + 245 + ], + "score": 0.72, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 233, + 506, + 249 + ], + "score": 1.0, + "content": "-stationarity. Plot (b) shows the", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 245, + 506, + 258 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 506, + 258 + ], + "score": 1.0, + "content": "optimal trade off between the minibatch sizes and the number of local updates at each WN (i.e., achieving the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 254, + 506, + 268 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 451, + 268 + ], + "score": 1.0, + "content": "lowest communication and sample complexities). Both plots are generated for an accuracy of", + "type": "text" + }, + { + "bbox": [ + 451, + 255, + 488, + 265 + ], + "score": 0.91, + "content": "\\epsilon = 1 0 ^ { - 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 488, + 254, + 506, + 268 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 264, + 506, + 279 + ], + "spans": [ + { + "bbox": [ + 105, + 264, + 506, + 279 + ], + "score": 1.0, + "content": "all the constants dependent on system parameters (variance of stochastic gradients, heterogeneity parameter,", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 105, + 275, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 105, + 275, + 505, + 288 + ], + "score": 1.0, + "content": "optimality gap, Lipschitz constants, etc.) are assumed to be 1. Fed STEM is a special case of STEM where", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 107, + 285, + 488, + 298 + ], + "spans": [ + { + "bbox": [ + 107, + 286, + 127, + 297 + ], + "score": 0.85, + "content": "\\mathcal { O } ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 285, + 382, + 298 + ], + "score": 1.0, + "content": "minibatch is used; Minibatch STEM is a special case of STEM where", + "type": "text" + }, + { + "bbox": [ + 383, + 286, + 402, + 297 + ], + "score": 0.88, + "content": "\\mathcal { O } ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 285, + 488, + 298 + ], + "score": 1.0, + "content": "local updates are used.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 6.5 + } + ], + "index": 3.75 + }, + { + "type": "text", + "bbox": [ + 107, + 316, + 506, + 349 + ], + "lines": [], + "index": 12, + "bbox_fs": [ + 105, + 316, + 505, + 350 + ], + "lines_deleted": true + }, + { + "type": "interline_equation", + "bbox": [ + 163, + 352, + 447, + 387 + ], + "lines": [ + { + "bbox": [ + 163, + 352, + 447, + 387 + ], + "spans": [ + { + "bbox": [ + 163, + 352, + 447, + 387 + ], + "score": 0.95, + "content": "\\operatorname* { m i n } _ { x \\in \\mathbb { R } ^ { d } } \\bigg \\{ f ( x ) : = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } f ^ { ( k ) } ( x ) : = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } \\mathbb { E } _ { \\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) } } \\big [ f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) \\big ] \\bigg \\} .", + "type": "interline_equation", + "image_path": "2805a58b3a852bb789307ddca3e7e2aea04c24beb34df4fd0d38807ed0628944.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 163, + 352, + 447, + 363.6666666666667 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 163, + 363.6666666666667, + 447, + 375.33333333333337 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 163, + 375.33333333333337, + 447, + 387.00000000000006 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 107, + 391, + 505, + 427 + ], + "lines": [ + { + "bbox": [ + 104, + 389, + 504, + 405 + ], + "spans": [ + { + "bbox": [ + 104, + 389, + 133, + 405 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 390, + 194, + 403 + ], + "score": 0.93, + "content": "f ^ { ( k ) } : \\mathbb { R } ^ { d } \\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 389, + 455, + 405 + ], + "score": 1.0, + "content": "denotes the smooth (possibly non-convex) objective function and", + "type": "text" + }, + { + "bbox": [ + 456, + 390, + 504, + 403 + ], + "score": 0.92, + "content": "\\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) }", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 402, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 402, + 309, + 416 + ], + "score": 1.0, + "content": "represents the sample/s drawn from distribution", + "type": "text" + }, + { + "bbox": [ + 309, + 403, + 329, + 414 + ], + "score": 0.9, + "content": "\\mathcal { D } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 402, + 358, + 416 + ], + "score": 1.0, + "content": "at the", + "type": "text" + }, + { + "bbox": [ + 359, + 403, + 372, + 414 + ], + "score": 0.85, + "content": "k ^ { \\mathrm { { t h } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 402, + 416, + 416 + ], + "score": 1.0, + "content": "WN with", + "type": "text" + }, + { + "bbox": [ + 416, + 403, + 454, + 416 + ], + "score": 0.92, + "content": "k \\in [ K ]", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 402, + 505, + 416 + ], + "score": 1.0, + "content": ". When the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 414, + 497, + 430 + ], + "spans": [ + { + "bbox": [ + 105, + 414, + 159, + 430 + ], + "score": 1.0, + "content": "distributions", + "type": "text" + }, + { + "bbox": [ + 159, + 415, + 178, + 426 + ], + "score": 0.91, + "content": "\\mathcal { D } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 414, + 497, + 430 + ], + "score": 1.0, + "content": "are different across the WNs, it is referred to as the heterogeneous data setting.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18, + "bbox_fs": [ + 104, + 389, + 505, + 430 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 432, + 505, + 521 + ], + "lines": [ + { + "bbox": [ + 105, + 431, + 506, + 445 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 506, + 445 + ], + "score": 1.0, + "content": "The optimization performance of non-convex FL algorithms is typically measured by the total number", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 443, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 256, + 456 + ], + "score": 1.0, + "content": "of samples accessed (cf. Definition", + "type": "text" + }, + { + "bbox": [ + 256, + 443, + 275, + 456 + ], + "score": 0.89, + "content": "2 . 2 )", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 444, + 505, + 455 + ], + "score": 1.0, + "content": "and the total rounds of communication (cf. Definition", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 454, + 505, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 124, + 467 + ], + "score": 0.79, + "content": "2 . 3 )", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 454, + 269, + 467 + ], + "score": 1.0, + "content": "required by each WN to achieve an", + "type": "text" + }, + { + "bbox": [ + 270, + 457, + 275, + 464 + ], + "score": 0.67, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 454, + 414, + 467 + ], + "score": 1.0, + "content": "-stationary solution (cf. Definition", + "type": "text" + }, + { + "bbox": [ + 414, + 454, + 432, + 467 + ], + "score": 0.81, + "content": "2 . 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 454, + 505, + 467 + ], + "score": 1.0, + "content": ". To minimize the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 505, + 478 + ], + "score": 1.0, + "content": "sample and the communication complexities, FL algorithms rely on the following four key design", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 477, + 505, + 488 + ], + "spans": [ + { + "bbox": [ + 105, + 477, + 505, + 488 + ], + "score": 1.0, + "content": "elements: (i) the WNs’ local model update directions, (ii) Minibatch size to compute each local", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 486, + 506, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 506, + 500 + ], + "score": 1.0, + "content": "direction, (iii) the number of local updates before WNs share their parameters, and (iv) the SN’s", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 498, + 506, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 506, + 511 + ], + "score": 1.0, + "content": "update direction. How to find effective FL algorithms by (optimally) designing these parameters has", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 509, + 290, + 522 + ], + "spans": [ + { + "bbox": [ + 106, + 509, + 290, + 522 + ], + "score": 1.0, + "content": "received significant research interest recently.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 431, + 506, + 522 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 525, + 385, + 536 + ], + "lines": [ + { + "bbox": [ + 106, + 524, + 385, + 538 + ], + "spans": [ + { + "bbox": [ + 106, + 524, + 385, + 538 + ], + "score": 1.0, + "content": "Contributions. The main contributions of this work are listed below:", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28, + "bbox_fs": [ + 106, + 524, + 385, + 538 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 541, + 505, + 653 + ], + "lines": [ + { + "bbox": [ + 105, + 541, + 506, + 554 + ], + "spans": [ + { + "bbox": [ + 105, + 541, + 506, + 554 + ], + "score": 1.0, + "content": "1) We propose the Stochastic Two-Sided Momentum (STEM) algorithm, that utilizes certain", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 553, + 505, + 564 + ], + "spans": [ + { + "bbox": [ + 106, + 553, + 505, + 564 + ], + "score": 1.0, + "content": "momentum-assisted stochastic gradient directions for both the WNs and SN updates. We show", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "spans": [ + { + "bbox": [ + 105, + 563, + 505, + 576 + ], + "score": 1.0, + "content": "that there exists an optimal trade off between the minibatch sizes and number of local updates, such", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 573, + 505, + 588 + ], + "spans": [ + { + "bbox": [ + 105, + 575, + 274, + 588 + ], + "score": 1.0, + "content": "that on the trade-off curve STEM requires", + "type": "text" + }, + { + "bbox": [ + 275, + 573, + 320, + 588 + ], + "score": 0.91, + "content": "\\underline { { \\tilde { \\mathcal { O } } } } ( \\epsilon ^ { - 3 / 2 } ) \\underline { { \\left[ \\frac { 2 } { } \\right] } }", + "type": "inline_equation" + }, + { + "bbox": [ + 320, + 575, + 370, + 588 + ], + "score": 1.0, + "content": "samples and", + "type": "text" + }, + { + "bbox": [ + 370, + 574, + 401, + 588 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 575, + 505, + 588 + ], + "score": 1.0, + "content": "communication rounds to", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 587, + 506, + 600 + ], + "spans": [ + { + "bbox": [ + 105, + 587, + 142, + 600 + ], + "score": 1.0, + "content": "reach an", + "type": "text" + }, + { + "bbox": [ + 142, + 589, + 148, + 597 + ], + "score": 0.57, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 587, + 271, + 600 + ], + "score": 1.0, + "content": "-stationary solution; see Figure", + "type": "text" + }, + { + "bbox": [ + 272, + 587, + 281, + 599 + ], + "score": 0.57, + "content": "\\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 281, + 587, + 506, + 600 + ], + "score": 1.0, + "content": "for an illustration. These complexity results are the best", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 597, + 505, + 611 + ], + "spans": [ + { + "bbox": [ + 105, + 597, + 490, + 610 + ], + "score": 1.0, + "content": "achievable for first-order stochastic FL algorithms (under certain assumptions, cf. Assumption", + "type": "text" + }, + { + "bbox": [ + 491, + 597, + 502, + 611 + ], + "score": 0.43, + "content": "\\bigstar \\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 597, + 505, + 610 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 607, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 607, + 122, + 621 + ], + "score": 1.0, + "content": "see", + "type": "text" + }, + { + "bbox": [ + 122, + 608, + 145, + 620 + ], + "score": 0.42, + "content": "\\pm \\boxed { 5 } \\boxed { 8 } \\parallel", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 607, + 163, + 621 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 163, + 608, + 190, + 620 + ], + "score": 0.35, + "content": "\\mathbb { B } \\mathbb { n o }", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 607, + 268, + 621 + ], + "score": 1.0, + "content": ", as well as Remark", + "type": "text" + }, + { + "bbox": [ + 268, + 608, + 277, + 621 + ], + "score": 0.45, + "content": "\\checkmark", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 607, + 505, + 621 + ], + "score": 1.0, + "content": "of this paper for discussions regarding optimality. To the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 620, + 505, + 631 + ], + "spans": [ + { + "bbox": [ + 106, + 620, + 505, + 631 + ], + "score": 1.0, + "content": "best of our knowledge, STEM is the first algorithm which – (i) simultaneously achieves the optimal", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 631, + 505, + 642 + ], + "spans": [ + { + "bbox": [ + 106, + 631, + 505, + 642 + ], + "score": 1.0, + "content": "sample and communication complexities for FL and (ii) can optimally trade off the minibatch sizes", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 641, + 239, + 654 + ], + "spans": [ + { + "bbox": [ + 106, + 641, + 239, + 654 + ], + "score": 1.0, + "content": "and the number of local updates.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 541, + 506, + 654 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 658, + 505, + 703 + ], + "lines": [ + { + "bbox": [ + 105, + 657, + 506, + 670 + ], + "spans": [ + { + "bbox": [ + 105, + 657, + 506, + 670 + ], + "score": 1.0, + "content": "2) A momentum-less special case of our STEM result further reveals some interesting insights of the", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 667, + 507, + 682 + ], + "spans": [ + { + "bbox": [ + 105, + 667, + 507, + 682 + ], + "score": 1.0, + "content": "classical FedAvg algorithm (a.k.a. the Local SGD) [11–13]. Specifically, we show that for FedAvg,", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 680, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 106, + 680, + 506, + 691 + ], + "score": 1.0, + "content": "there also exists a trade-off between the minibatch sizes and the number of local updates, such that it", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 690, + 503, + 705 + ], + "spans": [ + { + "bbox": [ + 105, + 690, + 141, + 705 + ], + "score": 1.0, + "content": "requires", + "type": "text" + }, + { + "bbox": [ + 141, + 691, + 173, + 704 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 690, + 226, + 705 + ], + "score": 1.0, + "content": "samples and", + "type": "text" + }, + { + "bbox": [ + 226, + 690, + 266, + 704 + ], + "score": 0.92, + "content": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 690, + 415, + 705 + ], + "score": 1.0, + "content": "communication rounds to achieve an", + "type": "text" + }, + { + "bbox": [ + 416, + 694, + 421, + 702 + ], + "score": 0.74, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 690, + 503, + 705 + ], + "score": 1.0, + "content": "-stationary solution.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 657, + 507, + 705 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 70, + 508, + 211 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 70, + 508, + 211 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 70, + 508, + 211 + ], + "spans": [ + { + "bbox": [ + 106, + 70, + 508, + 211 + ], + "score": 0.984, + "html": "
AlgorithmWorkSampleComm.Minibatch (b)Local Updates (I) /round
FedAvg国园 国国0(€-2)0(c-3/2) 0(c-2)0(1) 0(1) 2(1-v)0(c-1/2) 0(1) 3v
SCAFFOLD*this work 国0(c-2)O(e-3/2) 0(c-2)O(c 4-v) 0(1)O(c−2(4-D)) 0(1)
FedPD/FedProx*四/□O(c-2)0(e-1)0(1)0(e-1)
MIME†/FedGLOMO/80(c-3/2)O(€-3/2)0(1)0(1)
STEM Fed STEM Minibatch STEM* this workO(€-3/2)O(e-1)( 0(1) O(e-1/2)O(∈−(3)) O(∈-1/2) 0(1)
", + "type": "table", + "image_path": "c128a8f8c828914de2ebd0394fe1679cfb1eb64efae4ae9cdf73ecbb43acff9d.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 70, + 508, + 117.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 117.0, + 508, + 164.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 164.0, + 508, + 211.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 106, + 217, + 506, + 288 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 216, + 506, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 216, + 506, + 230 + ], + "score": 1.0, + "content": "Table 1: Comparison of FedAvg and STEM with different FL algorithms for various choices of the minibatch", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 227, + 447, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 227, + 126, + 239 + ], + "score": 1.0, + "content": "sizes", + "type": "text" + }, + { + "bbox": [ + 127, + 228, + 138, + 238 + ], + "score": 0.48, + "content": "( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 227, + 432, + 239 + ], + "score": 1.0, + "content": "and the number of per node local updates between two rounds of communication", + "type": "text" + }, + { + "bbox": [ + 433, + 228, + 445, + 239 + ], + "score": 0.62, + "content": "( I )", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 227, + 447, + 239 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 235, + 507, + 250 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 147, + 248 + ], + "score": 0.9, + "content": "^ \\circ \\nu \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 235, + 186, + 250 + ], + "score": 1.0, + "content": "trades off", + "type": "text" + }, + { + "bbox": [ + 186, + 238, + 191, + 246 + ], + "score": 0.69, + "content": "^ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 235, + 208, + 250 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 208, + 238, + 214, + 246 + ], + "score": 0.64, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 235, + 217, + 250 + ], + "score": 1.0, + "content": ";", + "type": "text" + }, + { + "bbox": [ + 218, + 238, + 242, + 247 + ], + "score": 0.88, + "content": "\\nu = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 235, + 267, + 250 + ], + "score": 1.0, + "content": "(resp.", + "type": "text" + }, + { + "bbox": [ + 267, + 238, + 291, + 247 + ], + "score": 0.87, + "content": "\\nu = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 235, + 370, + 250 + ], + "score": 1.0, + "content": ") uses multiple (resp.", + "type": "text" + }, + { + "bbox": [ + 371, + 237, + 392, + 248 + ], + "score": 0.75, + "content": "\\mathcal { O } ( 1 ) .", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 235, + 462, + 250 + ], + "score": 1.0, + "content": ") local updates and", + "type": "text" + }, + { + "bbox": [ + 462, + 237, + 482, + 248 + ], + "score": 0.86, + "content": "\\mathcal { O } ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 235, + 507, + 250 + ], + "score": 1.0, + "content": "(resp.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 246, + 448, + 259 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 448, + 259 + ], + "score": 1.0, + "content": "multiple) samples. Fed STEM and Minibatch STEM are two variants of the proposed STEM.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 255, + 442, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 255, + 336, + 270 + ], + "score": 1.0, + "content": "‡The data heterogeneity assumption is weaker than Assumption", + "type": "text" + }, + { + "bbox": [ + 336, + 256, + 345, + 269 + ], + "score": 0.46, + "content": "2", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 255, + 385, + 270 + ], + "score": 1.0, + "content": "(please see", + "type": "text" + }, + { + "bbox": [ + 386, + 257, + 397, + 268 + ], + "score": 0.71, + "content": "\\bigstar \\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 255, + 442, + 270 + ], + "score": 1.0, + "content": "for details).", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 265, + 406, + 280 + ], + "spans": [ + { + "bbox": [ + 105, + 265, + 406, + 280 + ], + "score": 1.0, + "content": "†Requires bounded Hessian dissimilarity to model data heterogeneity across WNs.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 276, + 491, + 289 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 252, + 289 + ], + "score": 1.0, + "content": "⇤Guarantees for Minibatch STEM with", + "type": "text" + }, + { + "bbox": [ + 253, + 278, + 276, + 286 + ], + "score": 0.89, + "content": "I = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 276, + 491, + 289 + ], + "score": 1.0, + "content": "and SCAFFOLD are independent of the data heterogeneity.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 108, + 321, + 504, + 343 + ], + "lines": [ + { + "bbox": [ + 105, + 320, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 505, + 335 + ], + "score": 1.0, + "content": "Collectively, our insights on the trade-offs provide practical guidelines for choosing different design", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 332, + 221, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 221, + 344 + ], + "score": 1.0, + "content": "elements for FL algorithms.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 106, + 348, + 505, + 589 + ], + "lines": [ + { + "bbox": [ + 105, + 348, + 505, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 348, + 505, + 361 + ], + "score": 1.0, + "content": "Related Works. FL algorithms were first proposed in the form of FedAvg [11], where the local update", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 360, + 504, + 371 + ], + "spans": [ + { + "bbox": [ + 106, + 360, + 504, + 371 + ], + "score": 1.0, + "content": "directions at each WN were chosen to be the SGD updates. Earlier works analyzed these algorithms", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 370, + 505, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 383 + ], + "score": 1.0, + "content": "in the homogeneous data setting [19–25], while many recent studies have focused on designing new", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 381, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 106, + 381, + 505, + 394 + ], + "score": 1.0, + "content": "algorithms to deal with heterogeneous data settings, as well as problems where the local loss functions", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 391, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 289, + 404 + ], + "score": 1.0, + "content": "are non-convex [9, 10, 12–16, 18, 26–32]. In", + "type": "text" + }, + { + "bbox": [ + 289, + 392, + 307, + 403 + ], + "score": 0.54, + "content": "\\bar { \\mathbb { E } 2 } \\mathbb { I }", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 391, + 505, + 404 + ], + "score": 1.0, + "content": ", the authors showed that Parallel Restarted SGD", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 402, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 104, + 402, + 381, + 416 + ], + "score": 1.0, + "content": "(Local SGD or FedAvg [11]) achieves linear speed up while requiring", + "type": "text" + }, + { + "bbox": [ + 382, + 402, + 413, + 415 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 402, + 465, + 416 + ], + "score": 1.0, + "content": "samples and", + "type": "text" + }, + { + "bbox": [ + 465, + 402, + 505, + 415 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 413, + 506, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 413, + 254, + 426 + ], + "score": 1.0, + "content": "rounds of communication to reach an", + "type": "text" + }, + { + "bbox": [ + 254, + 416, + 260, + 424 + ], + "score": 0.51, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 413, + 350, + 426 + ], + "score": 1.0, + "content": "-stationary solution. In", + "type": "text" + }, + { + "bbox": [ + 351, + 414, + 367, + 425 + ], + "score": 0.81, + "content": "[ \\bar { \\lVert { 4 } } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 413, + 506, + 426 + ], + "score": 1.0, + "content": ", a Momentum SGD was proposed,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 424, + 506, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 485, + 437 + ], + "score": 1.0, + "content": "which achieved the same sample and communication complexities as Parallel Restarted SGD", + "type": "text" + }, + { + "bbox": [ + 485, + 424, + 502, + 436 + ], + "score": 0.81, + "content": "\\mathbb { \\lVert 1 2 \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 424, + 506, + 437 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 435, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 506, + 447 + ], + "score": 1.0, + "content": "without requiring that the second moments of the gradients be bounded. Further, it was shown that", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 446, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 473, + 460 + ], + "score": 1.0, + "content": "under the homogeneous data setting, the communication complexity can be improved to", + "type": "text" + }, + { + "bbox": [ + 473, + 446, + 505, + 459 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 457, + 506, + 470 + ], + "spans": [ + { + "bbox": [ + 106, + 457, + 352, + 470 + ], + "score": 1.0, + "content": "while maintaining the same sample complexity. The works in", + "type": "text" + }, + { + "bbox": [ + 352, + 457, + 384, + 469 + ], + "score": 0.79, + "content": "\\boxed { 1 5 } , \\boxed { 1 6 }", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 457, + 506, + 470 + ], + "score": 1.0, + "content": "conducted tighter analysis for", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 468, + 506, + 481 + ], + "spans": [ + { + "bbox": [ + 106, + 468, + 276, + 481 + ], + "score": 1.0, + "content": "FedAvg with partial WN participation with", + "type": "text" + }, + { + "bbox": [ + 276, + 469, + 298, + 480 + ], + "score": 0.88, + "content": "\\mathcal { O } ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 468, + 506, + 481 + ], + "score": 1.0, + "content": "local updates and batch sizes. Their analysis showed", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 479, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 479, + 362, + 492 + ], + "score": 1.0, + "content": "that FedAvg’s sample and communication complexities are both", + "type": "text" + }, + { + "bbox": [ + 362, + 479, + 393, + 491 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 479, + 506, + 492 + ], + "score": 1.0, + "content": ". Additionally, SCAFFOLD", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 489, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 104, + 489, + 172, + 503 + ], + "score": 1.0, + "content": "was proposed in", + "type": "text" + }, + { + "bbox": [ + 173, + 490, + 190, + 502 + ], + "score": 0.72, + "content": "\\bar { \\| 1 5 \\| }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 489, + 444, + 503 + ], + "score": 1.0, + "content": ", which utilized variance reduction based local update directions", + "type": "text" + }, + { + "bbox": [ + 444, + 490, + 461, + 501 + ], + "score": 0.85, + "content": "\\pmb { \\mathbb { B 3 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 489, + 505, + 503 + ], + "score": 1.0, + "content": "to achieve", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 501, + 506, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 506, + 514 + ], + "score": 1.0, + "content": "the same sample and communication complexities as FedAvg. Similarly, VRL-SGD proposed in", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 511, + 506, + 525 + ], + "spans": [ + { + "bbox": [ + 106, + 512, + 124, + 523 + ], + "score": 0.5, + "content": "\\left[ \\left[ 2 9 \\right] \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 125, + 511, + 471, + 525 + ], + "score": 1.0, + "content": "also utilized variance reduction and showed improved communication complexity of", + "type": "text" + }, + { + "bbox": [ + 471, + 511, + 502, + 524 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 511, + 506, + 525 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 522, + 506, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 506, + 536 + ], + "score": 1.0, + "content": "while requiring the same computations as FedAvg. Importantly, both SCAFFOLD and VRL-SGD’s", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 534, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 534, + 427, + 546 + ], + "score": 1.0, + "content": "guarantees were independent of the data heterogeneity. The FedProx proposed in", + "type": "text" + }, + { + "bbox": [ + 428, + 534, + 446, + 545 + ], + "score": 0.75, + "content": "\\mathbb { \\ m }", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 534, + 505, + 546 + ], + "score": 1.0, + "content": "used a penalty", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 545, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 505, + 557 + ], + "score": 1.0, + "content": "based method to improve the communication complexity of FedAvg (i.e., the Parallel Restarted and", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 555, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 194, + 569 + ], + "score": 1.0, + "content": "Momentum SGD [14,", + "type": "text" + }, + { + "bbox": [ + 195, + 555, + 211, + 567 + ], + "score": 0.38, + "content": "\\mathbb { L } 2 \\mathbb { I }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 555, + 225, + 569 + ], + "score": 1.0, + "content": ") to", + "type": "text" + }, + { + "bbox": [ + 226, + 555, + 257, + 568 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 555, + 505, + 569 + ], + "score": 1.0, + "content": ". FedProx used a gradient similarity assumption to model data", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 566, + 506, + 580 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 506, + 580 + ], + "score": 1.0, + "content": "heterogeneity which can be stringent for many practical applications. This assumption was relaxed", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 577, + 214, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 198, + 590 + ], + "score": 1.0, + "content": "by FedPD proposed in", + "type": "text" + }, + { + "bbox": [ + 199, + 577, + 211, + 589 + ], + "score": 0.42, + "content": "\\pmb { \\mathbb { B } }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 577, + 214, + 590 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 107, + 593, + 505, + 663 + ], + "lines": [ + { + "bbox": [ + 105, + 593, + 506, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 593, + 506, + 606 + ], + "score": 1.0, + "content": "Recently, the works [17, 18] proposed to utilize hybrid momentum gradient estimators [7, 8]. The", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 604, + 506, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 178, + 618 + ], + "score": 1.0, + "content": "MIME algorithm", + "type": "text" + }, + { + "bbox": [ + 179, + 605, + 196, + 616 + ], + "score": 0.71, + "content": "\\mathbb { \\lVert 1 7 \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 604, + 506, + 618 + ], + "score": 1.0, + "content": "matched the optimal sample complexity (under certain smoothness assump-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 613, + 507, + 631 + ], + "spans": [ + { + "bbox": [ + 104, + 613, + 142, + 631 + ], + "score": 1.0, + "content": "tions) of", + "type": "text" + }, + { + "bbox": [ + 142, + 616, + 182, + 628 + ], + "score": 0.91, + "content": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 613, + 507, + 631 + ], + "score": 1.0, + "content": "of the centralized non-convex stochastic optimization algorithms [5–8]. Similarly,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 627, + 506, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 165, + 641 + ], + "score": 1.0, + "content": "Fed-GLOMO", + "type": "text" + }, + { + "bbox": [ + 165, + 628, + 183, + 639 + ], + "score": 0.27, + "content": "[ \\overline { { 1 8 } } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 627, + 506, + 641 + ], + "score": 1.0, + "content": "achieved the same sample complexity while employing compression to further", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 638, + 506, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 361, + 653 + ], + "score": 1.0, + "content": "reduce communication. Both MIME and Fed-GLOMO required", + "type": "text" + }, + { + "bbox": [ + 361, + 639, + 401, + 652 + ], + "score": 0.96, + "content": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 638, + 506, + 653 + ], + "score": 1.0, + "content": "communication rounds to", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 650, + 471, + 664 + ], + "spans": [ + { + "bbox": [ + 106, + 651, + 151, + 663 + ], + "score": 1.0, + "content": "achieve an", + "type": "text" + }, + { + "bbox": [ + 151, + 653, + 156, + 660 + ], + "score": 0.61, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 651, + 304, + 663 + ], + "score": 1.0, + "content": "-stationary solution. Please see Table", + "type": "text" + }, + { + "bbox": [ + 304, + 650, + 313, + 664 + ], + "score": 0.56, + "content": "^ 1", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 651, + 471, + 663 + ], + "score": 1.0, + "content": "for a summary of the above discussion.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 107, + 667, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 667, + 505, + 679 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 505, + 679 + ], + "score": 1.0, + "content": "The comparison of Local SGD (FedAvg) to Minibatch SGD for convex and strongly convex problems", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 676, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 316, + 691 + ], + "score": 1.0, + "content": "with homogeneous data setting was first conducted in", + "type": "text" + }, + { + "bbox": [ + 317, + 679, + 333, + 689 + ], + "score": 0.83, + "content": "\\mathbb { \\lVert 1 9 \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 676, + 505, + 691 + ], + "score": 1.0, + "content": "and later extended to heterogeneous setting", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 687, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 117, + 702 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 117, + 688, + 135, + 700 + ], + "score": 0.8, + "content": "\\mathbb { \\lVert \\rVert 3 \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 687, + 506, + 702 + ], + "score": 1.0, + "content": ". It was shown that Minibatch SGD almost always dominates the Local SGD. In contrast, it", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 161, + 712 + ], + "score": 1.0, + "content": "was shown in", + "type": "text" + }, + { + "bbox": [ + 162, + 699, + 179, + 711 + ], + "score": 0.85, + "content": "\\pmb { \\Vert 2 4 \\Vert }", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 699, + 506, + 712 + ], + "score": 1.0, + "content": "that Local SGD dominates Minibatch SGD in terms of generalization performance.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "score": 1.0, + "content": "Although existing FL results are rich, but they are somehow ad hoc and there is a lack of principled", + "type": "text" + } + ], + "index": 44 + } + ], + "index": 42 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 70, + 508, + 211 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 70, + 508, + 211 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 70, + 508, + 211 + ], + "spans": [ + { + "bbox": [ + 106, + 70, + 508, + 211 + ], + "score": 0.984, + "html": "
AlgorithmWorkSampleComm.Minibatch (b)Local Updates (I) /round
FedAvg国园 国国0(€-2)0(c-3/2) 0(c-2)0(1) 0(1) 2(1-v)0(c-1/2) 0(1) 3v
SCAFFOLD*this work 国0(c-2)O(e-3/2) 0(c-2)O(c 4-v) 0(1)O(c−2(4-D)) 0(1)
FedPD/FedProx*四/□O(c-2)0(e-1)0(1)0(e-1)
MIME†/FedGLOMO/80(c-3/2)O(€-3/2)0(1)0(1)
STEM Fed STEM Minibatch STEM* this workO(€-3/2)O(e-1)( 0(1) O(e-1/2)O(∈−(3)) O(∈-1/2) 0(1)
", + "type": "table", + "image_path": "c128a8f8c828914de2ebd0394fe1679cfb1eb64efae4ae9cdf73ecbb43acff9d.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 106, + 70, + 508, + 117.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 117.0, + 508, + 164.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 164.0, + 508, + 211.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 106, + 217, + 506, + 288 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 216, + 506, + 230 + ], + "spans": [ + { + "bbox": [ + 105, + 216, + 506, + 230 + ], + "score": 1.0, + "content": "Table 1: Comparison of FedAvg and STEM with different FL algorithms for various choices of the minibatch", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 227, + 447, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 227, + 126, + 239 + ], + "score": 1.0, + "content": "sizes", + "type": "text" + }, + { + "bbox": [ + 127, + 228, + 138, + 238 + ], + "score": 0.48, + "content": "( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 227, + 432, + 239 + ], + "score": 1.0, + "content": "and the number of per node local updates between two rounds of communication", + "type": "text" + }, + { + "bbox": [ + 433, + 228, + 445, + 239 + ], + "score": 0.62, + "content": "( I )", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 227, + 447, + 239 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 235, + 507, + 250 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 147, + 248 + ], + "score": 0.9, + "content": "^ \\circ \\nu \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 235, + 186, + 250 + ], + "score": 1.0, + "content": "trades off", + "type": "text" + }, + { + "bbox": [ + 186, + 238, + 191, + 246 + ], + "score": 0.69, + "content": "^ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 235, + 208, + 250 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 208, + 238, + 214, + 246 + ], + "score": 0.64, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 235, + 217, + 250 + ], + "score": 1.0, + "content": ";", + "type": "text" + }, + { + "bbox": [ + 218, + 238, + 242, + 247 + ], + "score": 0.88, + "content": "\\nu = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 243, + 235, + 267, + 250 + ], + "score": 1.0, + "content": "(resp.", + "type": "text" + }, + { + "bbox": [ + 267, + 238, + 291, + 247 + ], + "score": 0.87, + "content": "\\nu = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 292, + 235, + 370, + 250 + ], + "score": 1.0, + "content": ") uses multiple (resp.", + "type": "text" + }, + { + "bbox": [ + 371, + 237, + 392, + 248 + ], + "score": 0.75, + "content": "\\mathcal { O } ( 1 ) .", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 235, + 462, + 250 + ], + "score": 1.0, + "content": ") local updates and", + "type": "text" + }, + { + "bbox": [ + 462, + 237, + 482, + 248 + ], + "score": 0.86, + "content": "\\mathcal { O } ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 235, + 507, + 250 + ], + "score": 1.0, + "content": "(resp.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 246, + 448, + 259 + ], + "spans": [ + { + "bbox": [ + 105, + 246, + 448, + 259 + ], + "score": 1.0, + "content": "multiple) samples. Fed STEM and Minibatch STEM are two variants of the proposed STEM.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 255, + 442, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 255, + 336, + 270 + ], + "score": 1.0, + "content": "‡The data heterogeneity assumption is weaker than Assumption", + "type": "text" + }, + { + "bbox": [ + 336, + 256, + 345, + 269 + ], + "score": 0.46, + "content": "2", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 255, + 385, + 270 + ], + "score": 1.0, + "content": "(please see", + "type": "text" + }, + { + "bbox": [ + 386, + 257, + 397, + 268 + ], + "score": 0.71, + "content": "\\bigstar \\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 255, + 442, + 270 + ], + "score": 1.0, + "content": "for details).", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 265, + 406, + 280 + ], + "spans": [ + { + "bbox": [ + 105, + 265, + 406, + 280 + ], + "score": 1.0, + "content": "†Requires bounded Hessian dissimilarity to model data heterogeneity across WNs.", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 276, + 491, + 289 + ], + "spans": [ + { + "bbox": [ + 106, + 276, + 252, + 289 + ], + "score": 1.0, + "content": "⇤Guarantees for Minibatch STEM with", + "type": "text" + }, + { + "bbox": [ + 253, + 278, + 276, + 286 + ], + "score": 0.89, + "content": "I = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 276, + 491, + 289 + ], + "score": 1.0, + "content": "and SCAFFOLD are independent of the data heterogeneity.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 6 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 108, + 321, + 504, + 343 + ], + "lines": [ + { + "bbox": [ + 105, + 320, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 505, + 335 + ], + "score": 1.0, + "content": "Collectively, our insights on the trade-offs provide practical guidelines for choosing different design", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 332, + 221, + 344 + ], + "spans": [ + { + "bbox": [ + 106, + 332, + 221, + 344 + ], + "score": 1.0, + "content": "elements for FL algorithms.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 320, + 505, + 344 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 348, + 505, + 589 + ], + "lines": [ + { + "bbox": [ + 105, + 348, + 505, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 348, + 505, + 361 + ], + "score": 1.0, + "content": "Related Works. FL algorithms were first proposed in the form of FedAvg [11], where the local update", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 360, + 504, + 371 + ], + "spans": [ + { + "bbox": [ + 106, + 360, + 504, + 371 + ], + "score": 1.0, + "content": "directions at each WN were chosen to be the SGD updates. Earlier works analyzed these algorithms", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 370, + 505, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 370, + 505, + 383 + ], + "score": 1.0, + "content": "in the homogeneous data setting [19–25], while many recent studies have focused on designing new", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 381, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 106, + 381, + 505, + 394 + ], + "score": 1.0, + "content": "algorithms to deal with heterogeneous data settings, as well as problems where the local loss functions", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 391, + 505, + 404 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 289, + 404 + ], + "score": 1.0, + "content": "are non-convex [9, 10, 12–16, 18, 26–32]. In", + "type": "text" + }, + { + "bbox": [ + 289, + 392, + 307, + 403 + ], + "score": 0.54, + "content": "\\bar { \\mathbb { E } 2 } \\mathbb { I }", + "type": "inline_equation" + }, + { + "bbox": [ + 307, + 391, + 505, + 404 + ], + "score": 1.0, + "content": ", the authors showed that Parallel Restarted SGD", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 402, + 505, + 416 + ], + "spans": [ + { + "bbox": [ + 104, + 402, + 381, + 416 + ], + "score": 1.0, + "content": "(Local SGD or FedAvg [11]) achieves linear speed up while requiring", + "type": "text" + }, + { + "bbox": [ + 382, + 402, + 413, + 415 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 402, + 465, + 416 + ], + "score": 1.0, + "content": "samples and", + "type": "text" + }, + { + "bbox": [ + 465, + 402, + 505, + 415 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 413, + 506, + 426 + ], + "spans": [ + { + "bbox": [ + 105, + 413, + 254, + 426 + ], + "score": 1.0, + "content": "rounds of communication to reach an", + "type": "text" + }, + { + "bbox": [ + 254, + 416, + 260, + 424 + ], + "score": 0.51, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 260, + 413, + 350, + 426 + ], + "score": 1.0, + "content": "-stationary solution. In", + "type": "text" + }, + { + "bbox": [ + 351, + 414, + 367, + 425 + ], + "score": 0.81, + "content": "[ \\bar { \\lVert { 4 } } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 368, + 413, + 506, + 426 + ], + "score": 1.0, + "content": ", a Momentum SGD was proposed,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 424, + 506, + 437 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 485, + 437 + ], + "score": 1.0, + "content": "which achieved the same sample and communication complexities as Parallel Restarted SGD", + "type": "text" + }, + { + "bbox": [ + 485, + 424, + 502, + 436 + ], + "score": 0.81, + "content": "\\mathbb { \\lVert 1 2 \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 424, + 506, + 437 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 435, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 435, + 506, + 447 + ], + "score": 1.0, + "content": "without requiring that the second moments of the gradients be bounded. Further, it was shown that", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 446, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 446, + 473, + 460 + ], + "score": 1.0, + "content": "under the homogeneous data setting, the communication complexity can be improved to", + "type": "text" + }, + { + "bbox": [ + 473, + 446, + 505, + 459 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 457, + 506, + 470 + ], + "spans": [ + { + "bbox": [ + 106, + 457, + 352, + 470 + ], + "score": 1.0, + "content": "while maintaining the same sample complexity. The works in", + "type": "text" + }, + { + "bbox": [ + 352, + 457, + 384, + 469 + ], + "score": 0.79, + "content": "\\boxed { 1 5 } , \\boxed { 1 6 }", + "type": "inline_equation" + }, + { + "bbox": [ + 385, + 457, + 506, + 470 + ], + "score": 1.0, + "content": "conducted tighter analysis for", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 468, + 506, + 481 + ], + "spans": [ + { + "bbox": [ + 106, + 468, + 276, + 481 + ], + "score": 1.0, + "content": "FedAvg with partial WN participation with", + "type": "text" + }, + { + "bbox": [ + 276, + 469, + 298, + 480 + ], + "score": 0.88, + "content": "\\mathcal { O } ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 468, + 506, + 481 + ], + "score": 1.0, + "content": "local updates and batch sizes. Their analysis showed", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 479, + 506, + 492 + ], + "spans": [ + { + "bbox": [ + 106, + 479, + 362, + 492 + ], + "score": 1.0, + "content": "that FedAvg’s sample and communication complexities are both", + "type": "text" + }, + { + "bbox": [ + 362, + 479, + 393, + 491 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 479, + 506, + 492 + ], + "score": 1.0, + "content": ". Additionally, SCAFFOLD", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 489, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 104, + 489, + 172, + 503 + ], + "score": 1.0, + "content": "was proposed in", + "type": "text" + }, + { + "bbox": [ + 173, + 490, + 190, + 502 + ], + "score": 0.72, + "content": "\\bar { \\| 1 5 \\| }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 489, + 444, + 503 + ], + "score": 1.0, + "content": ", which utilized variance reduction based local update directions", + "type": "text" + }, + { + "bbox": [ + 444, + 490, + 461, + 501 + ], + "score": 0.85, + "content": "\\pmb { \\mathbb { B 3 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 489, + 505, + 503 + ], + "score": 1.0, + "content": "to achieve", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 501, + 506, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 506, + 514 + ], + "score": 1.0, + "content": "the same sample and communication complexities as FedAvg. Similarly, VRL-SGD proposed in", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 511, + 506, + 525 + ], + "spans": [ + { + "bbox": [ + 106, + 512, + 124, + 523 + ], + "score": 0.5, + "content": "\\left[ \\left[ 2 9 \\right] \\right]", + "type": "inline_equation" + }, + { + "bbox": [ + 125, + 511, + 471, + 525 + ], + "score": 1.0, + "content": "also utilized variance reduction and showed improved communication complexity of", + "type": "text" + }, + { + "bbox": [ + 471, + 511, + 502, + 524 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 511, + 506, + 525 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 522, + 506, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 522, + 506, + 536 + ], + "score": 1.0, + "content": "while requiring the same computations as FedAvg. Importantly, both SCAFFOLD and VRL-SGD’s", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 534, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 534, + 427, + 546 + ], + "score": 1.0, + "content": "guarantees were independent of the data heterogeneity. The FedProx proposed in", + "type": "text" + }, + { + "bbox": [ + 428, + 534, + 446, + 545 + ], + "score": 0.75, + "content": "\\mathbb { \\ m }", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 534, + 505, + 546 + ], + "score": 1.0, + "content": "used a penalty", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 545, + 505, + 557 + ], + "spans": [ + { + "bbox": [ + 105, + 545, + 505, + 557 + ], + "score": 1.0, + "content": "based method to improve the communication complexity of FedAvg (i.e., the Parallel Restarted and", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 555, + 505, + 569 + ], + "spans": [ + { + "bbox": [ + 105, + 555, + 194, + 569 + ], + "score": 1.0, + "content": "Momentum SGD [14,", + "type": "text" + }, + { + "bbox": [ + 195, + 555, + 211, + 567 + ], + "score": 0.38, + "content": "\\mathbb { L } 2 \\mathbb { I }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 555, + 225, + 569 + ], + "score": 1.0, + "content": ") to", + "type": "text" + }, + { + "bbox": [ + 226, + 555, + 257, + 568 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 555, + 505, + 569 + ], + "score": 1.0, + "content": ". FedProx used a gradient similarity assumption to model data", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 566, + 506, + 580 + ], + "spans": [ + { + "bbox": [ + 105, + 566, + 506, + 580 + ], + "score": 1.0, + "content": "heterogeneity which can be stringent for many practical applications. This assumption was relaxed", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 577, + 214, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 198, + 590 + ], + "score": 1.0, + "content": "by FedPD proposed in", + "type": "text" + }, + { + "bbox": [ + 199, + 577, + 211, + 589 + ], + "score": 0.42, + "content": "\\pmb { \\mathbb { B } }", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 577, + 214, + 590 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 22.5, + "bbox_fs": [ + 104, + 348, + 506, + 590 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 593, + 505, + 663 + ], + "lines": [ + { + "bbox": [ + 105, + 593, + 506, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 593, + 506, + 606 + ], + "score": 1.0, + "content": "Recently, the works [17, 18] proposed to utilize hybrid momentum gradient estimators [7, 8]. The", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 604, + 506, + 618 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 178, + 618 + ], + "score": 1.0, + "content": "MIME algorithm", + "type": "text" + }, + { + "bbox": [ + 179, + 605, + 196, + 616 + ], + "score": 0.71, + "content": "\\mathbb { \\lVert 1 7 \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 604, + 506, + 618 + ], + "score": 1.0, + "content": "matched the optimal sample complexity (under certain smoothness assump-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 104, + 613, + 507, + 631 + ], + "spans": [ + { + "bbox": [ + 104, + 613, + 142, + 631 + ], + "score": 1.0, + "content": "tions) of", + "type": "text" + }, + { + "bbox": [ + 142, + 616, + 182, + 628 + ], + "score": 0.91, + "content": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 613, + 507, + 631 + ], + "score": 1.0, + "content": "of the centralized non-convex stochastic optimization algorithms [5–8]. Similarly,", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 627, + 506, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 165, + 641 + ], + "score": 1.0, + "content": "Fed-GLOMO", + "type": "text" + }, + { + "bbox": [ + 165, + 628, + 183, + 639 + ], + "score": 0.27, + "content": "[ \\overline { { 1 8 } } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 183, + 627, + 506, + 641 + ], + "score": 1.0, + "content": "achieved the same sample complexity while employing compression to further", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 638, + 506, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 361, + 653 + ], + "score": 1.0, + "content": "reduce communication. Both MIME and Fed-GLOMO required", + "type": "text" + }, + { + "bbox": [ + 361, + 639, + 401, + 652 + ], + "score": 0.96, + "content": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 638, + 506, + 653 + ], + "score": 1.0, + "content": "communication rounds to", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 650, + 471, + 664 + ], + "spans": [ + { + "bbox": [ + 106, + 651, + 151, + 663 + ], + "score": 1.0, + "content": "achieve an", + "type": "text" + }, + { + "bbox": [ + 151, + 653, + 156, + 660 + ], + "score": 0.61, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 651, + 304, + 663 + ], + "score": 1.0, + "content": "-stationary solution. Please see Table", + "type": "text" + }, + { + "bbox": [ + 304, + 650, + 313, + 664 + ], + "score": 0.56, + "content": "^ 1", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 651, + 471, + 663 + ], + "score": 1.0, + "content": "for a summary of the above discussion.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 36.5, + "bbox_fs": [ + 104, + 593, + 507, + 664 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 667, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 667, + 505, + 679 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 505, + 679 + ], + "score": 1.0, + "content": "The comparison of Local SGD (FedAvg) to Minibatch SGD for convex and strongly convex problems", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 676, + 505, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 676, + 316, + 691 + ], + "score": 1.0, + "content": "with homogeneous data setting was first conducted in", + "type": "text" + }, + { + "bbox": [ + 317, + 679, + 333, + 689 + ], + "score": 0.83, + "content": "\\mathbb { \\lVert 1 9 \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 334, + 676, + 505, + 691 + ], + "score": 1.0, + "content": "and later extended to heterogeneous setting", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 687, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 687, + 117, + 702 + ], + "score": 1.0, + "content": "in", + "type": "text" + }, + { + "bbox": [ + 117, + 688, + 135, + 700 + ], + "score": 0.8, + "content": "\\mathbb { \\lVert \\rVert 3 \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 687, + 506, + 702 + ], + "score": 1.0, + "content": ". It was shown that Minibatch SGD almost always dominates the Local SGD. In contrast, it", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 161, + 712 + ], + "score": 1.0, + "content": "was shown in", + "type": "text" + }, + { + "bbox": [ + 162, + 699, + 179, + 711 + ], + "score": 0.85, + "content": "\\pmb { \\Vert 2 4 \\Vert }", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 699, + 506, + 712 + ], + "score": 1.0, + "content": "that Local SGD dominates Minibatch SGD in terms of generalization performance.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "score": 1.0, + "content": "Although existing FL results are rich, but they are somehow ad hoc and there is a lack of principled", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 72, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 72, + 506, + 86 + ], + "score": 1.0, + "content": "understanding of the algorithms. We note that the proposed STEM algorithmic framework provides a", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 83, + 507, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 287, + 96 + ], + "score": 1.0, + "content": "theoretical framework that unifies all existing", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 287, + 84, + 300, + 94 + ], + "score": 0.28, + "content": "\\mathrm { F L }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 300, + 83, + 507, + 96 + ], + "score": 1.0, + "content": "results on sample and communication complexities.", + "type": "text", + "cross_page": true + } + ], + "index": 1 + } + ], + "index": 42, + "bbox_fs": [ + 105, + 667, + 506, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 72, + 505, + 95 + ], + "lines": [ + { + "bbox": [ + 106, + 72, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 72, + 506, + 86 + ], + "score": 1.0, + "content": "understanding of the algorithms. We note that the proposed STEM algorithmic framework provides a", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 83, + 507, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 287, + 96 + ], + "score": 1.0, + "content": "theoretical framework that unifies all existing", + "type": "text" + }, + { + "bbox": [ + 287, + 84, + 300, + 94 + ], + "score": 0.28, + "content": "\\mathrm { F L }", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 83, + 507, + 96 + ], + "score": 1.0, + "content": "results on sample and communication complexities.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 106, + 100, + 505, + 169 + ], + "lines": [ + { + "bbox": [ + 106, + 100, + 505, + 112 + ], + "spans": [ + { + "bbox": [ + 106, + 100, + 329, + 112 + ], + "score": 1.0, + "content": "Notations. The expected value of a random variable", + "type": "text" + }, + { + "bbox": [ + 329, + 101, + 339, + 110 + ], + "score": 0.83, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 100, + 400, + 112 + ], + "score": 1.0, + "content": "is denoted by", + "type": "text" + }, + { + "bbox": [ + 400, + 100, + 423, + 112 + ], + "score": 0.91, + "content": "\\mathbb { E } [ X ]", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 100, + 505, + 112 + ], + "score": 1.0, + "content": "and its expectation", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 110, + 505, + 124 + ], + "spans": [ + { + "bbox": [ + 106, + 110, + 210, + 124 + ], + "score": 1.0, + "content": "conditioned on an Event", + "type": "text" + }, + { + "bbox": [ + 210, + 113, + 219, + 122 + ], + "score": 0.76, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 110, + 277, + 124 + ], + "score": 1.0, + "content": "is denoted as", + "type": "text" + }, + { + "bbox": [ + 278, + 112, + 337, + 124 + ], + "score": 0.49, + "content": "\\mathbb { E } [ X | \\mathrm { E v e n t ~ } A ]", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 110, + 403, + 124 + ], + "score": 1.0, + "content": ". We denote by", + "type": "text" + }, + { + "bbox": [ + 403, + 113, + 412, + 122 + ], + "score": 0.83, + "content": "\\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 110, + 434, + 124 + ], + "score": 1.0, + "content": "(and", + "type": "text" + }, + { + "bbox": [ + 435, + 111, + 448, + 122 + ], + "score": 0.84, + "content": "\\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 110, + 505, + 124 + ], + "score": 1.0, + "content": ") the real line", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 122, + 506, + 136 + ], + "spans": [ + { + "bbox": [ + 105, + 122, + 143, + 136 + ], + "score": 1.0, + "content": "(and the", + "type": "text" + }, + { + "bbox": [ + 143, + 123, + 150, + 133 + ], + "score": 0.81, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 122, + 455, + 136 + ], + "score": 1.0, + "content": "-dimensional Euclidean space). The set of natural numbers is denoted by", + "type": "text" + }, + { + "bbox": [ + 455, + 123, + 464, + 133 + ], + "score": 0.69, + "content": "\\mathbb { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 122, + 506, + 136 + ], + "score": 1.0, + "content": ". Given a", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 133, + 506, + 148 + ], + "spans": [ + { + "bbox": [ + 105, + 133, + 172, + 148 + ], + "score": 1.0, + "content": "positive integer", + "type": "text" + }, + { + "bbox": [ + 172, + 135, + 203, + 145 + ], + "score": 0.9, + "content": "K \\in \\mathbb N", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 133, + 252, + 148 + ], + "score": 1.0, + "content": ", we denote", + "type": "text" + }, + { + "bbox": [ + 252, + 134, + 338, + 147 + ], + "score": 0.92, + "content": "[ K ] \\triangleq \\{ 1 , 2 , \\dots , K \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 133, + 382, + 148 + ], + "score": 1.0, + "content": ". Notation", + "type": "text" + }, + { + "bbox": [ + 382, + 135, + 401, + 147 + ], + "score": 0.91, + "content": "\\| \\cdot \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 133, + 451, + 148 + ], + "score": 1.0, + "content": "denotes the", + "type": "text" + }, + { + "bbox": [ + 451, + 135, + 461, + 146 + ], + "score": 0.88, + "content": "\\ell _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 133, + 506, + 148 + ], + "score": 1.0, + "content": "-norm and", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 107, + 145, + 505, + 158 + ], + "spans": [ + { + "bbox": [ + 107, + 146, + 125, + 158 + ], + "score": 0.91, + "content": "\\langle \\cdot , \\cdot \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 126, + 145, + 309, + 158 + ], + "score": 1.0, + "content": "the Euclidean inner product. For a discrete set", + "type": "text" + }, + { + "bbox": [ + 309, + 147, + 318, + 156 + ], + "score": 0.72, + "content": "\\boldsymbol { B }", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 145, + 322, + 158 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 322, + 146, + 335, + 158 + ], + "score": 0.85, + "content": "| B |", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 145, + 505, + 158 + ], + "score": 1.0, + "content": "denotes the cardinality of the set. Uniform", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 156, + 387, + 170 + ], + "spans": [ + { + "bbox": [ + 105, + 156, + 229, + 170 + ], + "score": 1.0, + "content": "distribution over a discrete set", + "type": "text" + }, + { + "bbox": [ + 229, + 157, + 274, + 169 + ], + "score": 0.94, + "content": "\\{ 1 , \\ldots , T \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 156, + 329, + 170 + ], + "score": 1.0, + "content": "is denoted as", + "type": "text" + }, + { + "bbox": [ + 329, + 157, + 382, + 169 + ], + "score": 0.94, + "content": "{ \\dot { \\mathcal { U } } } \\{ 1 , \\dots , T \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 382, + 156, + 387, + 170 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 4.5 + }, + { + "type": "title", + "bbox": [ + 107, + 183, + 195, + 197 + ], + "lines": [ + { + "bbox": [ + 104, + 182, + 196, + 199 + ], + "spans": [ + { + "bbox": [ + 104, + 182, + 196, + 199 + ], + "score": 1.0, + "content": "2 Preliminaries", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 105, + 208, + 479, + 220 + ], + "lines": [ + { + "bbox": [ + 104, + 206, + 478, + 222 + ], + "spans": [ + { + "bbox": [ + 104, + 206, + 462, + 222 + ], + "score": 1.0, + "content": "Before we proceed to the algorithms, we make the following assumptions about problem", + "type": "text" + }, + { + "bbox": [ + 463, + 208, + 475, + 221 + ], + "score": 0.72, + "content": "( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 206, + 478, + 222 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 105, + 223, + 504, + 248 + ], + "lines": [ + { + "bbox": [ + 105, + 221, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 221, + 435, + 237 + ], + "score": 1.0, + "content": "Assumption 1 (Sample Gradient Lipschitz Smoothness). The stochastic functions", + "type": "text" + }, + { + "bbox": [ + 435, + 222, + 483, + 236 + ], + "score": 0.93, + "content": "f ^ { ( k ) } ( \\cdot , \\xi ^ { ( k ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 221, + 506, + 237 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 107, + 232, + 460, + 251 + ], + "spans": [ + { + "bbox": [ + 107, + 235, + 155, + 248 + ], + "score": 0.93, + "content": "\\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 232, + 184, + 251 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 184, + 236, + 217, + 248 + ], + "score": 0.93, + "content": "k \\in [ K ]", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 232, + 460, + 251 + ], + "score": 1.0, + "content": ", satisfy the mean squared smoothness property, i.e, we have", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "interline_equation", + "bbox": [ + 159, + 252, + 451, + 268 + ], + "lines": [ + { + "bbox": [ + 159, + 252, + 451, + 268 + ], + "spans": [ + { + "bbox": [ + 159, + 252, + 451, + 268 + ], + "score": 0.89, + "content": "\\begin{array} { r } { \\mathbb { E } \\| \\nabla f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) - \\nabla f ^ { ( k ) } ( y ; \\xi ^ { ( k ) } ) \\| ^ { 2 } \\leq L ^ { 2 } \\| x - y \\| ^ { 2 } \\mathrm { ~ f o r ~ a l l ~ } x , y \\in \\mathbb { R } ^ { d } . } \\end{array}", + "type": "interline_equation", + "image_path": "a3cfe57005b8e58ae1bfc3b55d7cca26c0bbb487296b4468bd38bfdc42e09c0e.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 159, + 252, + 451, + 268 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 272, + 504, + 295 + ], + "lines": [ + { + "bbox": [ + 106, + 271, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 505, + 284 + ], + "score": 1.0, + "content": "Assumption 2 (Unbiased gradient and Variance Bounds). (i) Unbiased Gradient. The stochastic", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 283, + 289, + 295 + ], + "spans": [ + { + "bbox": [ + 106, + 283, + 289, + 295 + ], + "score": 1.0, + "content": "gradients computed at each WN are unbiased", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "interline_equation", + "bbox": [ + 185, + 299, + 426, + 314 + ], + "lines": [ + { + "bbox": [ + 185, + 299, + 426, + 314 + ], + "spans": [ + { + "bbox": [ + 185, + 299, + 426, + 314 + ], + "score": 0.9, + "content": "\\mathbb { E } [ \\nabla f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) ] = \\nabla f ^ { ( k ) } ( x ) , \\forall \\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) } , \\forall k \\in [ K ] .", + "type": "interline_equation", + "image_path": "f06c119c8d26c099d56bae28e2fa443fdee4af7121660da108f532b9c8355105.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 185, + 299, + 426, + 314 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 317, + 391, + 329 + ], + "lines": [ + { + "bbox": [ + 105, + 317, + 391, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 391, + 331 + ], + "score": 1.0, + "content": "(ii) Intra- and inter- node Variance Bound. The following bounds hold:", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 333, + 511, + 350 + ], + "lines": [ + { + "bbox": [ + 111, + 333, + 511, + 350 + ], + "spans": [ + { + "bbox": [ + 111, + 333, + 511, + 350 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\mathbb { \\tilde { z } } \\| \\nabla f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) - \\nabla f ^ { ( k ) } ( x ) \\| ^ { 2 } \\leq \\sigma ^ { 2 } , \\| \\nabla f ^ { ( k ) } ( x ) - \\nabla f ^ { ( \\ell ) } ( x ) \\| ^ { 2 } \\leq \\zeta ^ { 2 } , \\forall \\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) } , \\forall k , \\ell \\in [ K ] . } \\end{array}", + "type": "interline_equation", + "image_path": "b27d78c5e7a746fe9622599d0a5137acd7faea00d13d76004e33686cafb79bc5.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 111, + 333, + 511, + 350 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 359, + 506, + 451 + ], + "lines": [ + { + "bbox": [ + 104, + 359, + 506, + 374 + ], + "spans": [ + { + "bbox": [ + 104, + 360, + 196, + 374 + ], + "score": 1.0, + "content": "Note that Assumption", + "type": "text" + }, + { + "bbox": [ + 197, + 359, + 207, + 374 + ], + "score": 0.77, + "content": "\\boxed { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 360, + 345, + 374 + ], + "score": 1.0, + "content": "is stronger than directly assuming", + "type": "text" + }, + { + "bbox": [ + 346, + 360, + 363, + 373 + ], + "score": 0.92, + "content": "f ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 360, + 506, + 374 + ], + "score": 1.0, + "content": "’s are Lipschitz smooth (which we", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 372, + 506, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 506, + 384 + ], + "score": 1.0, + "content": "will refer to as the averaged gradient Lipschitz smooth condition), but it is still a rather standard", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 384, + 505, + 395 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 505, + 395 + ], + "score": 1.0, + "content": "assumption in SGD analysis. For example it has been used in analyzing centralized SGD algorithms", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 393, + 506, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 176, + 407 + ], + "score": 1.0, + "content": "such as SPIDER", + "type": "text" + }, + { + "bbox": [ + 176, + 394, + 188, + 405 + ], + "score": 0.5, + "content": " { \\mathbb { I } }", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 393, + 229, + 407 + ], + "score": 1.0, + "content": ", SNVRG", + "type": "text" + }, + { + "bbox": [ + 229, + 394, + 241, + 405 + ], + "score": 0.47, + "content": "\\pmb { \\Vert 6 \\Vert }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 393, + 283, + 407 + ], + "score": 1.0, + "content": ", STORM", + "type": "text" + }, + { + "bbox": [ + 283, + 394, + 295, + 406 + ], + "score": 0.35, + "content": "\\mathbb { [ [ \\big ] ] }", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 393, + 506, + 407 + ], + "score": 1.0, + "content": "(and many others) as well as in FL algorithms such", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 403, + 506, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 147, + 418 + ], + "score": 1.0, + "content": "as MIME", + "type": "text" + }, + { + "bbox": [ + 147, + 405, + 165, + 416 + ], + "score": 0.74, + "content": " { \\mathbb { I } } ^ { [ 1 2 ] }", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 403, + 241, + 418 + ], + "score": 1.0, + "content": "and Fed-GLOMO", + "type": "text" + }, + { + "bbox": [ + 241, + 405, + 259, + 416 + ], + "score": 0.71, + "content": " { \\mathbb { I } } { \\mathrm { 1 8 } } { \\Vert }", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 403, + 405, + 418 + ], + "score": 1.0, + "content": ". The second relation in Assumption", + "type": "text" + }, + { + "bbox": [ + 405, + 405, + 415, + 418 + ], + "score": 0.79, + "content": "2 \\cdot", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 403, + 506, + 418 + ], + "score": 1.0, + "content": "(ii) quantifies the data", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 416, + 505, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 219, + 429 + ], + "score": 1.0, + "content": "heterogeneity, and we call", + "type": "text" + }, + { + "bbox": [ + 219, + 416, + 247, + 428 + ], + "score": 0.92, + "content": "\\zeta > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 247, + 416, + 505, + 429 + ], + "score": 1.0, + "content": "as the heterogeneity parameter. This is a typical assumption", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 427, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 427, + 506, + 439 + ], + "score": 1.0, + "content": "required to evaluate the performance of FL algorithms. If data distributions across individual WNs", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 437, + 384, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 177, + 452 + ], + "score": 1.0, + "content": "are identical, i.e.,", + "type": "text" + }, + { + "bbox": [ + 178, + 438, + 229, + 449 + ], + "score": 0.92, + "content": "\\mathcal { D } ^ { ( k ) } \\stackrel { = } { = } \\mathcal { D } ^ { ( \\ell ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 437, + 257, + 452 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 257, + 439, + 299, + 451 + ], + "score": 0.93, + "content": "k , \\ell \\in [ K ]", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 437, + 355, + 452 + ], + "score": 1.0, + "content": "then we have", + "type": "text" + }, + { + "bbox": [ + 355, + 439, + 380, + 451 + ], + "score": 0.91, + "content": "\\zeta = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 437, + 384, + 452 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 107, + 455, + 503, + 478 + ], + "lines": [ + { + "bbox": [ + 105, + 454, + 504, + 468 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 186, + 468 + ], + "score": 1.0, + "content": "Next, we define the", + "type": "text" + }, + { + "bbox": [ + 187, + 458, + 192, + 465 + ], + "score": 0.66, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 454, + 504, + 468 + ], + "score": 1.0, + "content": "-stationary solution for non-convex optimization problems, as well as quantify", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 465, + 437, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 362, + 479 + ], + "score": 1.0, + "content": "the computation and communication complexities to achieve an", + "type": "text" + }, + { + "bbox": [ + 363, + 469, + 368, + 476 + ], + "score": 0.72, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 465, + 437, + 479 + ], + "score": 1.0, + "content": "-stationary point.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 106, + 480, + 506, + 514 + ], + "lines": [ + { + "bbox": [ + 105, + 479, + 507, + 494 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 170, + 494 + ], + "score": 1.0, + "content": "Definition 2.1", + "type": "text" + }, + { + "bbox": [ + 171, + 483, + 176, + 491 + ], + "score": 0.39, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 479, + 289, + 494 + ], + "score": 1.0, + "content": "-Stationary Point). A point", + "type": "text" + }, + { + "bbox": [ + 289, + 483, + 296, + 491 + ], + "score": 0.77, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 479, + 334, + 494 + ], + "score": 1.0, + "content": "is called", + "type": "text" + }, + { + "bbox": [ + 335, + 483, + 340, + 491 + ], + "score": 0.72, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 479, + 394, + 494 + ], + "score": 1.0, + "content": "-stationary if", + "type": "text" + }, + { + "bbox": [ + 394, + 480, + 456, + 492 + ], + "score": 0.91, + "content": "\\| \\nabla f ( x ) \\| ^ { 2 } \\leq \\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 479, + 507, + 494 + ], + "score": 1.0, + "content": ". Moreover,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 491, + 503, + 505 + ], + "spans": [ + { + "bbox": [ + 104, + 491, + 284, + 505 + ], + "score": 1.0, + "content": "a stochastic algorithm is said to achieve an", + "type": "text" + }, + { + "bbox": [ + 284, + 494, + 290, + 502 + ], + "score": 0.68, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 491, + 368, + 505 + ], + "score": 1.0, + "content": "-stationary point in", + "type": "text" + }, + { + "bbox": [ + 369, + 493, + 374, + 502 + ], + "score": 0.75, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 491, + 425, + 505 + ], + "score": 1.0, + "content": "iterations if", + "type": "text" + }, + { + "bbox": [ + 425, + 491, + 503, + 504 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\ddot { \\mathbb { E } } [ \\| \\nabla f ( x _ { t } ) \\| ^ { 2 } ] \\le \\epsilon . } \\end{array}", + "type": "inline_equation" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 502, + 438, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 428, + 516 + ], + "score": 1.0, + "content": "where the expectation is over the stochasticity of the algorithm until time instant", + "type": "text" + }, + { + "bbox": [ + 429, + 504, + 433, + 513 + ], + "score": 0.74, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 502, + 438, + 516 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 106, + 517, + 505, + 576 + ], + "lines": [ + { + "bbox": [ + 105, + 516, + 506, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 506, + 530 + ], + "score": 1.0, + "content": "Definition 2.2 (Sample complexity). We assume an Incremental First-order Oracle (IFO) frame-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 527, + 506, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 530, + 130, + 540 + ], + "score": 1.0, + "content": "work", + "type": "text" + }, + { + "bbox": [ + 131, + 529, + 148, + 540 + ], + "score": 0.79, + "content": "\\textcircled { \\lVert { 3 4 } \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 527, + 250, + 542 + ], + "score": 1.0, + "content": "where, given a sample", + "type": "text" + }, + { + "bbox": [ + 251, + 528, + 303, + 541 + ], + "score": 0.93, + "content": "\\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 527, + 333, + 542 + ], + "score": 1.0, + "content": "at the", + "type": "text" + }, + { + "bbox": [ + 333, + 528, + 346, + 540 + ], + "score": 0.87, + "content": "k ^ { \\mathrm { { t h } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 527, + 418, + 542 + ], + "score": 1.0, + "content": "node and iterate", + "type": "text" + }, + { + "bbox": [ + 418, + 532, + 425, + 540 + ], + "score": 0.71, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 527, + 506, + 542 + ], + "score": 1.0, + "content": ", the oracle returns", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 538, + 507, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 228, + 553 + ], + "score": 0.9, + "content": "( f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) , \\nabla f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 538, + 507, + 554 + ], + "score": 1.0, + "content": ". Each access to the oracle is counted as a single IFO operation. We", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 552, + 505, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 505, + 564 + ], + "score": 1.0, + "content": "measure the sample (and computational) complexity in terms of the total number of calls to the IFO", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 564, + 374, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 209, + 576 + ], + "score": 1.0, + "content": "by all WNs to achieve an", + "type": "text" + }, + { + "bbox": [ + 209, + 565, + 214, + 573 + ], + "score": 0.65, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 564, + 374, + 576 + ], + "score": 1.0, + "content": "-stationary point given in Definition 2.1.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 106, + 577, + 505, + 623 + ], + "lines": [ + { + "bbox": [ + 105, + 577, + 507, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 507, + 590 + ], + "score": 1.0, + "content": "Definition 2.3 (Communication complexity). We define a communication round as a one back-and-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 588, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 505, + 601 + ], + "score": 1.0, + "content": "forth sharing of parameters between the WNs and the SN. Then the communication complexity is", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 600, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 505, + 612 + ], + "score": 1.0, + "content": "defined to be the total number of communication rounds between any WN and the SN required to", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 611, + 317, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 611, + 151, + 623 + ], + "score": 1.0, + "content": "achieve an", + "type": "text" + }, + { + "bbox": [ + 151, + 613, + 156, + 620 + ], + "score": 0.6, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 611, + 317, + 623 + ], + "score": 1.0, + "content": "-stationary point given in Definition 2.1.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37.5 + }, + { + "type": "title", + "bbox": [ + 107, + 636, + 369, + 651 + ], + "lines": [ + { + "bbox": [ + 104, + 635, + 370, + 654 + ], + "spans": [ + { + "bbox": [ + 104, + 635, + 370, + 654 + ], + "score": 1.0, + "content": "3 The STEM algorithm and the trade-off analysis", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 107, + 662, + 505, + 695 + ], + "lines": [ + { + "bbox": [ + 105, + 661, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 506, + 675 + ], + "score": 1.0, + "content": "In this section, we discuss the proposed algorithm and present the main results. The key in the", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 672, + 506, + 685 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 463, + 685 + ], + "score": 1.0, + "content": "algorithm design is to carefully balance all the four design elements mentioned in Sec.", + "type": "text" + }, + { + "bbox": [ + 464, + 672, + 474, + 685 + ], + "score": 0.66, + "content": "^ { 1 , }", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 672, + 506, + 685 + ], + "score": 1.0, + "content": "so that", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 684, + 438, + 696 + ], + "spans": [ + { + "bbox": [ + 106, + 684, + 438, + 696 + ], + "score": 1.0, + "content": "sufficient and useful progress can be made between two rounds of communication.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42 + }, + { + "type": "text", + "bbox": [ + 106, + 699, + 503, + 723 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 341, + 713 + ], + "score": 1.0, + "content": "Let us discuss the key steps of STEM, listed in Algorithm", + "type": "text" + }, + { + "bbox": [ + 341, + 700, + 351, + 713 + ], + "score": 0.77, + "content": "^ { 1 . }", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 698, + 505, + 713 + ], + "score": 1.0, + "content": "In Step 10, each node locally updates", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 710, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 294, + 724 + ], + "score": 1.0, + "content": "its model parameters using the local direction", + "type": "text" + }, + { + "bbox": [ + 294, + 711, + 304, + 723 + ], + "score": 0.91, + "content": "d _ { t } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 710, + 388, + 724 + ], + "score": 1.0, + "content": ", computed by using", + "type": "text" + }, + { + "bbox": [ + 389, + 711, + 394, + 721 + ], + "score": 0.68, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 710, + 505, + 724 + ], + "score": 1.0, + "content": "stochastic gradients at two", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 741, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 741, + 310, + 752 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 11, + "width": 9 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 72, + 505, + 95 + ], + "lines": [], + "index": 0.5, + "bbox_fs": [ + 105, + 72, + 507, + 96 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 100, + 505, + 169 + ], + "lines": [ + { + "bbox": [ + 106, + 100, + 505, + 112 + ], + "spans": [ + { + "bbox": [ + 106, + 100, + 329, + 112 + ], + "score": 1.0, + "content": "Notations. The expected value of a random variable", + "type": "text" + }, + { + "bbox": [ + 329, + 101, + 339, + 110 + ], + "score": 0.83, + "content": "X", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 100, + 400, + 112 + ], + "score": 1.0, + "content": "is denoted by", + "type": "text" + }, + { + "bbox": [ + 400, + 100, + 423, + 112 + ], + "score": 0.91, + "content": "\\mathbb { E } [ X ]", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 100, + 505, + 112 + ], + "score": 1.0, + "content": "and its expectation", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 110, + 505, + 124 + ], + "spans": [ + { + "bbox": [ + 106, + 110, + 210, + 124 + ], + "score": 1.0, + "content": "conditioned on an Event", + "type": "text" + }, + { + "bbox": [ + 210, + 113, + 219, + 122 + ], + "score": 0.76, + "content": "A", + "type": "inline_equation" + }, + { + "bbox": [ + 220, + 110, + 277, + 124 + ], + "score": 1.0, + "content": "is denoted as", + "type": "text" + }, + { + "bbox": [ + 278, + 112, + 337, + 124 + ], + "score": 0.49, + "content": "\\mathbb { E } [ X | \\mathrm { E v e n t ~ } A ]", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 110, + 403, + 124 + ], + "score": 1.0, + "content": ". We denote by", + "type": "text" + }, + { + "bbox": [ + 403, + 113, + 412, + 122 + ], + "score": 0.83, + "content": "\\mathbb { R }", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 110, + 434, + 124 + ], + "score": 1.0, + "content": "(and", + "type": "text" + }, + { + "bbox": [ + 435, + 111, + 448, + 122 + ], + "score": 0.84, + "content": "\\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 110, + 505, + 124 + ], + "score": 1.0, + "content": ") the real line", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 122, + 506, + 136 + ], + "spans": [ + { + "bbox": [ + 105, + 122, + 143, + 136 + ], + "score": 1.0, + "content": "(and the", + "type": "text" + }, + { + "bbox": [ + 143, + 123, + 150, + 133 + ], + "score": 0.81, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 122, + 455, + 136 + ], + "score": 1.0, + "content": "-dimensional Euclidean space). The set of natural numbers is denoted by", + "type": "text" + }, + { + "bbox": [ + 455, + 123, + 464, + 133 + ], + "score": 0.69, + "content": "\\mathbb { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 122, + 506, + 136 + ], + "score": 1.0, + "content": ". Given a", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 133, + 506, + 148 + ], + "spans": [ + { + "bbox": [ + 105, + 133, + 172, + 148 + ], + "score": 1.0, + "content": "positive integer", + "type": "text" + }, + { + "bbox": [ + 172, + 135, + 203, + 145 + ], + "score": 0.9, + "content": "K \\in \\mathbb N", + "type": "inline_equation" + }, + { + "bbox": [ + 203, + 133, + 252, + 148 + ], + "score": 1.0, + "content": ", we denote", + "type": "text" + }, + { + "bbox": [ + 252, + 134, + 338, + 147 + ], + "score": 0.92, + "content": "[ K ] \\triangleq \\{ 1 , 2 , \\dots , K \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 133, + 382, + 148 + ], + "score": 1.0, + "content": ". Notation", + "type": "text" + }, + { + "bbox": [ + 382, + 135, + 401, + 147 + ], + "score": 0.91, + "content": "\\| \\cdot \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 133, + 451, + 148 + ], + "score": 1.0, + "content": "denotes the", + "type": "text" + }, + { + "bbox": [ + 451, + 135, + 461, + 146 + ], + "score": 0.88, + "content": "\\ell _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 133, + 506, + 148 + ], + "score": 1.0, + "content": "-norm and", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 107, + 145, + 505, + 158 + ], + "spans": [ + { + "bbox": [ + 107, + 146, + 125, + 158 + ], + "score": 0.91, + "content": "\\langle \\cdot , \\cdot \\rangle", + "type": "inline_equation" + }, + { + "bbox": [ + 126, + 145, + 309, + 158 + ], + "score": 1.0, + "content": "the Euclidean inner product. For a discrete set", + "type": "text" + }, + { + "bbox": [ + 309, + 147, + 318, + 156 + ], + "score": 0.72, + "content": "\\boldsymbol { B }", + "type": "inline_equation" + }, + { + "bbox": [ + 318, + 145, + 322, + 158 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 322, + 146, + 335, + 158 + ], + "score": 0.85, + "content": "| B |", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 145, + 505, + 158 + ], + "score": 1.0, + "content": "denotes the cardinality of the set. Uniform", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 156, + 387, + 170 + ], + "spans": [ + { + "bbox": [ + 105, + 156, + 229, + 170 + ], + "score": 1.0, + "content": "distribution over a discrete set", + "type": "text" + }, + { + "bbox": [ + 229, + 157, + 274, + 169 + ], + "score": 0.94, + "content": "\\{ 1 , \\ldots , T \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 274, + 156, + 329, + 170 + ], + "score": 1.0, + "content": "is denoted as", + "type": "text" + }, + { + "bbox": [ + 329, + 157, + 382, + 169 + ], + "score": 0.94, + "content": "{ \\dot { \\mathcal { U } } } \\{ 1 , \\dots , T \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 382, + 156, + 387, + 170 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 4.5, + "bbox_fs": [ + 105, + 100, + 506, + 170 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 183, + 195, + 197 + ], + "lines": [ + { + "bbox": [ + 104, + 182, + 196, + 199 + ], + "spans": [ + { + "bbox": [ + 104, + 182, + 196, + 199 + ], + "score": 1.0, + "content": "2 Preliminaries", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 105, + 208, + 479, + 220 + ], + "lines": [ + { + "bbox": [ + 104, + 206, + 478, + 222 + ], + "spans": [ + { + "bbox": [ + 104, + 206, + 462, + 222 + ], + "score": 1.0, + "content": "Before we proceed to the algorithms, we make the following assumptions about problem", + "type": "text" + }, + { + "bbox": [ + 463, + 208, + 475, + 221 + ], + "score": 0.72, + "content": "( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 206, + 478, + 222 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9, + "bbox_fs": [ + 104, + 206, + 478, + 222 + ] + }, + { + "type": "text", + "bbox": [ + 105, + 223, + 504, + 248 + ], + "lines": [ + { + "bbox": [ + 105, + 221, + 506, + 237 + ], + "spans": [ + { + "bbox": [ + 105, + 221, + 435, + 237 + ], + "score": 1.0, + "content": "Assumption 1 (Sample Gradient Lipschitz Smoothness). The stochastic functions", + "type": "text" + }, + { + "bbox": [ + 435, + 222, + 483, + 236 + ], + "score": 0.93, + "content": "f ^ { ( k ) } ( \\cdot , \\xi ^ { ( k ) } )", + "type": "inline_equation" + }, + { + "bbox": [ + 484, + 221, + 506, + 237 + ], + "score": 1.0, + "content": "with", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 107, + 232, + 460, + 251 + ], + "spans": [ + { + "bbox": [ + 107, + 235, + 155, + 248 + ], + "score": 0.93, + "content": "\\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 232, + 184, + 251 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 184, + 236, + 217, + 248 + ], + "score": 0.93, + "content": "k \\in [ K ]", + "type": "inline_equation" + }, + { + "bbox": [ + 218, + 232, + 460, + 251 + ], + "score": 1.0, + "content": ", satisfy the mean squared smoothness property, i.e, we have", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5, + "bbox_fs": [ + 105, + 221, + 506, + 251 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 159, + 252, + 451, + 268 + ], + "lines": [ + { + "bbox": [ + 159, + 252, + 451, + 268 + ], + "spans": [ + { + "bbox": [ + 159, + 252, + 451, + 268 + ], + "score": 0.89, + "content": "\\begin{array} { r } { \\mathbb { E } \\| \\nabla f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) - \\nabla f ^ { ( k ) } ( y ; \\xi ^ { ( k ) } ) \\| ^ { 2 } \\leq L ^ { 2 } \\| x - y \\| ^ { 2 } \\mathrm { ~ f o r ~ a l l ~ } x , y \\in \\mathbb { R } ^ { d } . } \\end{array}", + "type": "interline_equation", + "image_path": "a3cfe57005b8e58ae1bfc3b55d7cca26c0bbb487296b4468bd38bfdc42e09c0e.jpg" + } + ] + } + ], + "index": 12, + "virtual_lines": [ + { + "bbox": [ + 159, + 252, + 451, + 268 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "text", + "bbox": [ + 108, + 272, + 504, + 295 + ], + "lines": [ + { + "bbox": [ + 106, + 271, + 505, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 271, + 505, + 284 + ], + "score": 1.0, + "content": "Assumption 2 (Unbiased gradient and Variance Bounds). (i) Unbiased Gradient. The stochastic", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 283, + 289, + 295 + ], + "spans": [ + { + "bbox": [ + 106, + 283, + 289, + 295 + ], + "score": 1.0, + "content": "gradients computed at each WN are unbiased", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5, + "bbox_fs": [ + 106, + 271, + 505, + 295 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 185, + 299, + 426, + 314 + ], + "lines": [ + { + "bbox": [ + 185, + 299, + 426, + 314 + ], + "spans": [ + { + "bbox": [ + 185, + 299, + 426, + 314 + ], + "score": 0.9, + "content": "\\mathbb { E } [ \\nabla f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) ] = \\nabla f ^ { ( k ) } ( x ) , \\forall \\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) } , \\forall k \\in [ K ] .", + "type": "interline_equation", + "image_path": "f06c119c8d26c099d56bae28e2fa443fdee4af7121660da108f532b9c8355105.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 185, + 299, + 426, + 314 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 317, + 391, + 329 + ], + "lines": [ + { + "bbox": [ + 105, + 317, + 391, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 391, + 331 + ], + "score": 1.0, + "content": "(ii) Intra- and inter- node Variance Bound. The following bounds hold:", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16, + "bbox_fs": [ + 105, + 317, + 391, + 331 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 111, + 333, + 511, + 350 + ], + "lines": [ + { + "bbox": [ + 111, + 333, + 511, + 350 + ], + "spans": [ + { + "bbox": [ + 111, + 333, + 511, + 350 + ], + "score": 0.91, + "content": "\\begin{array} { r } { \\mathbb { \\tilde { z } } \\| \\nabla f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) - \\nabla f ^ { ( k ) } ( x ) \\| ^ { 2 } \\leq \\sigma ^ { 2 } , \\| \\nabla f ^ { ( k ) } ( x ) - \\nabla f ^ { ( \\ell ) } ( x ) \\| ^ { 2 } \\leq \\zeta ^ { 2 } , \\forall \\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) } , \\forall k , \\ell \\in [ K ] . } \\end{array}", + "type": "interline_equation", + "image_path": "b27d78c5e7a746fe9622599d0a5137acd7faea00d13d76004e33686cafb79bc5.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 111, + 333, + 511, + 350 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 359, + 506, + 451 + ], + "lines": [ + { + "bbox": [ + 104, + 359, + 506, + 374 + ], + "spans": [ + { + "bbox": [ + 104, + 360, + 196, + 374 + ], + "score": 1.0, + "content": "Note that Assumption", + "type": "text" + }, + { + "bbox": [ + 197, + 359, + 207, + 374 + ], + "score": 0.77, + "content": "\\boxed { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 360, + 345, + 374 + ], + "score": 1.0, + "content": "is stronger than directly assuming", + "type": "text" + }, + { + "bbox": [ + 346, + 360, + 363, + 373 + ], + "score": 0.92, + "content": "f ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 360, + 506, + 374 + ], + "score": 1.0, + "content": "’s are Lipschitz smooth (which we", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 372, + 506, + 384 + ], + "spans": [ + { + "bbox": [ + 105, + 372, + 506, + 384 + ], + "score": 1.0, + "content": "will refer to as the averaged gradient Lipschitz smooth condition), but it is still a rather standard", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 384, + 505, + 395 + ], + "spans": [ + { + "bbox": [ + 106, + 384, + 505, + 395 + ], + "score": 1.0, + "content": "assumption in SGD analysis. For example it has been used in analyzing centralized SGD algorithms", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 393, + 506, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 176, + 407 + ], + "score": 1.0, + "content": "such as SPIDER", + "type": "text" + }, + { + "bbox": [ + 176, + 394, + 188, + 405 + ], + "score": 0.5, + "content": " { \\mathbb { I } }", + "type": "inline_equation" + }, + { + "bbox": [ + 189, + 393, + 229, + 407 + ], + "score": 1.0, + "content": ", SNVRG", + "type": "text" + }, + { + "bbox": [ + 229, + 394, + 241, + 405 + ], + "score": 0.47, + "content": "\\pmb { \\Vert 6 \\Vert }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 393, + 283, + 407 + ], + "score": 1.0, + "content": ", STORM", + "type": "text" + }, + { + "bbox": [ + 283, + 394, + 295, + 406 + ], + "score": 0.35, + "content": "\\mathbb { [ [ \\big ] ] }", + "type": "inline_equation" + }, + { + "bbox": [ + 296, + 393, + 506, + 407 + ], + "score": 1.0, + "content": "(and many others) as well as in FL algorithms such", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 403, + 506, + 418 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 147, + 418 + ], + "score": 1.0, + "content": "as MIME", + "type": "text" + }, + { + "bbox": [ + 147, + 405, + 165, + 416 + ], + "score": 0.74, + "content": " { \\mathbb { I } } ^ { [ 1 2 ] }", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 403, + 241, + 418 + ], + "score": 1.0, + "content": "and Fed-GLOMO", + "type": "text" + }, + { + "bbox": [ + 241, + 405, + 259, + 416 + ], + "score": 0.71, + "content": " { \\mathbb { I } } { \\mathrm { 1 8 } } { \\Vert }", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 403, + 405, + 418 + ], + "score": 1.0, + "content": ". The second relation in Assumption", + "type": "text" + }, + { + "bbox": [ + 405, + 405, + 415, + 418 + ], + "score": 0.79, + "content": "2 \\cdot", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 403, + 506, + 418 + ], + "score": 1.0, + "content": "(ii) quantifies the data", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 416, + 505, + 429 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 219, + 429 + ], + "score": 1.0, + "content": "heterogeneity, and we call", + "type": "text" + }, + { + "bbox": [ + 219, + 416, + 247, + 428 + ], + "score": 0.92, + "content": "\\zeta > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 247, + 416, + 505, + 429 + ], + "score": 1.0, + "content": "as the heterogeneity parameter. This is a typical assumption", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 427, + 506, + 439 + ], + "spans": [ + { + "bbox": [ + 105, + 427, + 506, + 439 + ], + "score": 1.0, + "content": "required to evaluate the performance of FL algorithms. If data distributions across individual WNs", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 437, + 384, + 452 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 177, + 452 + ], + "score": 1.0, + "content": "are identical, i.e.,", + "type": "text" + }, + { + "bbox": [ + 178, + 438, + 229, + 449 + ], + "score": 0.92, + "content": "\\mathcal { D } ^ { ( k ) } \\stackrel { = } { = } \\mathcal { D } ^ { ( \\ell ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 437, + 257, + 452 + ], + "score": 1.0, + "content": "for all", + "type": "text" + }, + { + "bbox": [ + 257, + 439, + 299, + 451 + ], + "score": 0.93, + "content": "k , \\ell \\in [ K ]", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 437, + 355, + 452 + ], + "score": 1.0, + "content": "then we have", + "type": "text" + }, + { + "bbox": [ + 355, + 439, + 380, + 451 + ], + "score": 0.91, + "content": "\\zeta = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 437, + 384, + 452 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 21.5, + "bbox_fs": [ + 104, + 359, + 506, + 452 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 455, + 503, + 478 + ], + "lines": [ + { + "bbox": [ + 105, + 454, + 504, + 468 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 186, + 468 + ], + "score": 1.0, + "content": "Next, we define the", + "type": "text" + }, + { + "bbox": [ + 187, + 458, + 192, + 465 + ], + "score": 0.66, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 454, + 504, + 468 + ], + "score": 1.0, + "content": "-stationary solution for non-convex optimization problems, as well as quantify", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 465, + 437, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 465, + 362, + 479 + ], + "score": 1.0, + "content": "the computation and communication complexities to achieve an", + "type": "text" + }, + { + "bbox": [ + 363, + 469, + 368, + 476 + ], + "score": 0.72, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 465, + 437, + 479 + ], + "score": 1.0, + "content": "-stationary point.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 454, + 504, + 479 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 480, + 506, + 514 + ], + "lines": [ + { + "bbox": [ + 105, + 479, + 507, + 494 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 170, + 494 + ], + "score": 1.0, + "content": "Definition 2.1", + "type": "text" + }, + { + "bbox": [ + 171, + 483, + 176, + 491 + ], + "score": 0.39, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 479, + 289, + 494 + ], + "score": 1.0, + "content": "-Stationary Point). A point", + "type": "text" + }, + { + "bbox": [ + 289, + 483, + 296, + 491 + ], + "score": 0.77, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 479, + 334, + 494 + ], + "score": 1.0, + "content": "is called", + "type": "text" + }, + { + "bbox": [ + 335, + 483, + 340, + 491 + ], + "score": 0.72, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 340, + 479, + 394, + 494 + ], + "score": 1.0, + "content": "-stationary if", + "type": "text" + }, + { + "bbox": [ + 394, + 480, + 456, + 492 + ], + "score": 0.91, + "content": "\\| \\nabla f ( x ) \\| ^ { 2 } \\leq \\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 479, + 507, + 494 + ], + "score": 1.0, + "content": ". Moreover,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 491, + 503, + 505 + ], + "spans": [ + { + "bbox": [ + 104, + 491, + 284, + 505 + ], + "score": 1.0, + "content": "a stochastic algorithm is said to achieve an", + "type": "text" + }, + { + "bbox": [ + 284, + 494, + 290, + 502 + ], + "score": 0.68, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 491, + 368, + 505 + ], + "score": 1.0, + "content": "-stationary point in", + "type": "text" + }, + { + "bbox": [ + 369, + 493, + 374, + 502 + ], + "score": 0.75, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 374, + 491, + 425, + 505 + ], + "score": 1.0, + "content": "iterations if", + "type": "text" + }, + { + "bbox": [ + 425, + 491, + 503, + 504 + ], + "score": 0.9, + "content": "\\begin{array} { r } { \\ddot { \\mathbb { E } } [ \\| \\nabla f ( x _ { t } ) \\| ^ { 2 } ] \\le \\epsilon . } \\end{array}", + "type": "inline_equation" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 502, + 438, + 516 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 428, + 516 + ], + "score": 1.0, + "content": "where the expectation is over the stochasticity of the algorithm until time instant", + "type": "text" + }, + { + "bbox": [ + 429, + 504, + 433, + 513 + ], + "score": 0.74, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 434, + 502, + 438, + 516 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29, + "bbox_fs": [ + 104, + 479, + 507, + 516 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 517, + 505, + 576 + ], + "lines": [ + { + "bbox": [ + 105, + 516, + 506, + 530 + ], + "spans": [ + { + "bbox": [ + 105, + 516, + 506, + 530 + ], + "score": 1.0, + "content": "Definition 2.2 (Sample complexity). We assume an Incremental First-order Oracle (IFO) frame-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 527, + 506, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 530, + 130, + 540 + ], + "score": 1.0, + "content": "work", + "type": "text" + }, + { + "bbox": [ + 131, + 529, + 148, + 540 + ], + "score": 0.79, + "content": "\\textcircled { \\lVert { 3 4 } \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 527, + 250, + 542 + ], + "score": 1.0, + "content": "where, given a sample", + "type": "text" + }, + { + "bbox": [ + 251, + 528, + 303, + 541 + ], + "score": 0.93, + "content": "\\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 527, + 333, + 542 + ], + "score": 1.0, + "content": "at the", + "type": "text" + }, + { + "bbox": [ + 333, + 528, + 346, + 540 + ], + "score": 0.87, + "content": "k ^ { \\mathrm { { t h } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 527, + 418, + 542 + ], + "score": 1.0, + "content": "node and iterate", + "type": "text" + }, + { + "bbox": [ + 418, + 532, + 425, + 540 + ], + "score": 0.71, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 527, + 506, + 542 + ], + "score": 1.0, + "content": ", the oracle returns", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 538, + 507, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 228, + 553 + ], + "score": 0.9, + "content": "( f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) , \\nabla f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) )", + "type": "inline_equation" + }, + { + "bbox": [ + 229, + 538, + 507, + 554 + ], + "score": 1.0, + "content": ". Each access to the oracle is counted as a single IFO operation. We", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 552, + 505, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 552, + 505, + 564 + ], + "score": 1.0, + "content": "measure the sample (and computational) complexity in terms of the total number of calls to the IFO", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 564, + 374, + 576 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 209, + 576 + ], + "score": 1.0, + "content": "by all WNs to achieve an", + "type": "text" + }, + { + "bbox": [ + 209, + 565, + 214, + 573 + ], + "score": 0.65, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 564, + 374, + 576 + ], + "score": 1.0, + "content": "-stationary point given in Definition 2.1.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33, + "bbox_fs": [ + 105, + 516, + 507, + 576 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 577, + 505, + 623 + ], + "lines": [ + { + "bbox": [ + 105, + 577, + 507, + 590 + ], + "spans": [ + { + "bbox": [ + 105, + 577, + 507, + 590 + ], + "score": 1.0, + "content": "Definition 2.3 (Communication complexity). We define a communication round as a one back-and-", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 588, + 505, + 601 + ], + "spans": [ + { + "bbox": [ + 105, + 588, + 505, + 601 + ], + "score": 1.0, + "content": "forth sharing of parameters between the WNs and the SN. Then the communication complexity is", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 600, + 505, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 600, + 505, + 612 + ], + "score": 1.0, + "content": "defined to be the total number of communication rounds between any WN and the SN required to", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 611, + 317, + 623 + ], + "spans": [ + { + "bbox": [ + 106, + 611, + 151, + 623 + ], + "score": 1.0, + "content": "achieve an", + "type": "text" + }, + { + "bbox": [ + 151, + 613, + 156, + 620 + ], + "score": 0.6, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 611, + 317, + 623 + ], + "score": 1.0, + "content": "-stationary point given in Definition 2.1.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 577, + 507, + 623 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 636, + 369, + 651 + ], + "lines": [ + { + "bbox": [ + 104, + 635, + 370, + 654 + ], + "spans": [ + { + "bbox": [ + 104, + 635, + 370, + 654 + ], + "score": 1.0, + "content": "3 The STEM algorithm and the trade-off analysis", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 40 + }, + { + "type": "text", + "bbox": [ + 107, + 662, + 505, + 695 + ], + "lines": [ + { + "bbox": [ + 105, + 661, + 506, + 675 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 506, + 675 + ], + "score": 1.0, + "content": "In this section, we discuss the proposed algorithm and present the main results. The key in the", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 672, + 506, + 685 + ], + "spans": [ + { + "bbox": [ + 105, + 672, + 463, + 685 + ], + "score": 1.0, + "content": "algorithm design is to carefully balance all the four design elements mentioned in Sec.", + "type": "text" + }, + { + "bbox": [ + 464, + 672, + 474, + 685 + ], + "score": 0.66, + "content": "^ { 1 , }", + "type": "inline_equation" + }, + { + "bbox": [ + 475, + 672, + 506, + 685 + ], + "score": 1.0, + "content": "so that", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 684, + 438, + 696 + ], + "spans": [ + { + "bbox": [ + 106, + 684, + 438, + 696 + ], + "score": 1.0, + "content": "sufficient and useful progress can be made between two rounds of communication.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42, + "bbox_fs": [ + 105, + 661, + 506, + 696 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 699, + 503, + 723 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 341, + 713 + ], + "score": 1.0, + "content": "Let us discuss the key steps of STEM, listed in Algorithm", + "type": "text" + }, + { + "bbox": [ + 341, + 700, + 351, + 713 + ], + "score": 0.77, + "content": "^ { 1 . }", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 698, + 505, + 713 + ], + "score": 1.0, + "content": "In Step 10, each node locally updates", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 710, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 294, + 724 + ], + "score": 1.0, + "content": "its model parameters using the local direction", + "type": "text" + }, + { + "bbox": [ + 294, + 711, + 304, + 723 + ], + "score": 0.91, + "content": "d _ { t } ^ { k }", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 710, + 388, + 724 + ], + "score": 1.0, + "content": ", computed by using", + "type": "text" + }, + { + "bbox": [ + 389, + 711, + 394, + 721 + ], + "score": 0.68, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 710, + 505, + 724 + ], + "score": 1.0, + "content": "stochastic gradients at two", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5, + "bbox_fs": [ + 105, + 698, + 505, + 724 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 109, + 86, + 506, + 312 + ], + "lines": [ + { + "bbox": [ + 110, + 87, + 467, + 102 + ], + "spans": [ + { + "bbox": [ + 110, + 87, + 203, + 102 + ], + "score": 1.0, + "content": "1: Input: Parameters:", + "type": "text" + }, + { + "bbox": [ + 203, + 90, + 227, + 99 + ], + "score": 0.89, + "content": "c > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 87, + 343, + 102 + ], + "score": 1.0, + "content": ", the number of local updates", + "type": "text" + }, + { + "bbox": [ + 344, + 90, + 350, + 99 + ], + "score": 0.74, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 87, + 396, + 102 + ], + "score": 1.0, + "content": ", batch size", + "type": "text" + }, + { + "bbox": [ + 396, + 90, + 402, + 99 + ], + "score": 0.74, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 87, + 443, + 102 + ], + "score": 1.0, + "content": ", stepsizes", + "type": "text" + }, + { + "bbox": [ + 443, + 89, + 462, + 101 + ], + "score": 0.9, + "content": "\\{ \\eta _ { t } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 463, + 87, + 467, + 102 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 109, + 96, + 503, + 120 + ], + "spans": [ + { + "bbox": [ + 109, + 96, + 197, + 120 + ], + "score": 1.0, + "content": "2: Initialize: Iterate", + "type": "text" + }, + { + "bbox": [ + 198, + 100, + 312, + 115 + ], + "score": 0.91, + "content": "\\begin{array} { r } { x _ { 1 } ^ { ( k ) } = \\bar { x } _ { 1 } = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } x _ { 1 } ^ { ( k ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 96, + 390, + 120 + ], + "score": 1.0, + "content": ", descent direction", + "type": "text" + }, + { + "bbox": [ + 390, + 100, + 503, + 116 + ], + "score": 0.92, + "content": "\\begin{array} { r } { d _ { 1 } ^ { ( k ) } = \\bar { d } _ { 1 } = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } d _ { 1 } ^ { ( k ) } } \\end{array}", + "type": "inline_equation" + } + ], + "index": 1 + }, + { + "bbox": [ + 122, + 111, + 420, + 138 + ], + "spans": [ + { + "bbox": [ + 122, + 111, + 143, + 138 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 144, + 115, + 302, + 133 + ], + "score": 0.91, + "content": "\\begin{array} { r } { d _ { 1 } ^ { ( k ) } = \\frac { 1 } { B } \\sum _ { \\xi _ { 1 } ^ { ( k ) } \\in \\mathcal { B } _ { 1 } ^ { ( k ) } } \\nabla f ^ { ( k ) } ( x _ { 1 } ^ { ( k ) } ; \\xi _ { 1 } ^ { ( k ) } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 111, + 320, + 138 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 320, + 115, + 366, + 130 + ], + "score": 0.93, + "content": "| B _ { 1 } ^ { ( k ) } | = B", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 111, + 381, + 138 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 381, + 117, + 415, + 129 + ], + "score": 0.91, + "content": "k \\in [ K ]", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 111, + 420, + 138 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 127, + 289, + 152 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 162, + 152 + ], + "score": 1.0, + "content": "3: Perform:", + "type": "text" + }, + { + "bbox": [ + 162, + 133, + 241, + 147 + ], + "score": 0.88, + "content": "x _ { 2 } ^ { ( k ) } = x _ { 1 } ^ { k } - \\eta _ { 1 } d _ { 1 } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 127, + 246, + 152 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 247, + 134, + 289, + 147 + ], + "score": 0.67, + "content": "\\forall k \\in [ K ]", + "type": "inline_equation" + } + ], + "index": 3 + }, + { + "bbox": [ + 109, + 145, + 197, + 158 + ], + "spans": [ + { + "bbox": [ + 109, + 145, + 138, + 158 + ], + "score": 1.0, + "content": "4: for", + "type": "text" + }, + { + "bbox": [ + 138, + 146, + 162, + 156 + ], + "score": 0.88, + "content": "t = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 145, + 173, + 158 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 173, + 147, + 182, + 156 + ], + "score": 0.71, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 145, + 197, + 158 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 110, + 157, + 505, + 169 + ], + "spans": [ + { + "bbox": [ + 110, + 157, + 122, + 169 + ], + "score": 1.0, + "content": "5:", + "type": "text" + }, + { + "bbox": [ + 137, + 157, + 153, + 168 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 154, + 157, + 178, + 167 + ], + "score": 0.84, + "content": "k = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 157, + 190, + 168 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 190, + 157, + 200, + 167 + ], + "score": 0.72, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 157, + 216, + 168 + ], + "score": 1.0, + "content": "do", + "type": "text" + }, + { + "bbox": [ + 449, + 157, + 505, + 168 + ], + "score": 1.0, + "content": "#at the WN", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 110, + 163, + 504, + 199 + ], + "spans": [ + { + "bbox": [ + 110, + 171, + 122, + 183 + ], + "score": 1.0, + "content": "6:", + "type": "text" + }, + { + "bbox": [ + 152, + 163, + 504, + 199 + ], + "score": 0.81, + "content": "\\mathcal { d } _ { t + 1 } ^ { ( k ) } = \\frac { 1 } { b } \\sum _ { \\xi _ { t + 1 } ^ { ( k ) } \\in \\mathcal { B } _ { t + 1 } ^ { ( k ) } } ^ { \\pi \\Delta } \\nabla f ^ { ( k ) } ( x _ { t + 1 } ^ { ( k ) } ; \\xi _ { t + 1 } ^ { ( k ) } ) + \\left( 1 - a _ { t + 1 } \\right) \\bigg ( d _ { t } ^ { ( k ) } - \\frac { 1 } { b } \\sum _ { \\xi _ { t + 1 } ^ { ( k ) } \\in \\mathcal { B } _ { t + 1 } ^ { ( k ) } } \\nabla f ^ { ( k ) } ( x _ { t } ^ { ( k ) } ; \\xi _ { t + 1 } ^ { ( k ) } ) \\bigg )", + "type": "inline_equation", + "image_path": "cb47ddefffada4ba7399d00907097a5fbdab3bb233d14f9fdff1de621faadef2.jpg" + } + ], + "index": 6 + }, + { + "bbox": [ + 123, + 194, + 321, + 218 + ], + "spans": [ + { + "bbox": [ + 123, + 194, + 196, + 218 + ], + "score": 1.0, + "content": "where we choose", + "type": "text" + }, + { + "bbox": [ + 197, + 199, + 241, + 214 + ], + "score": 0.89, + "content": "| B _ { t + 1 } ^ { ( k ) } | = b", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 194, + 262, + 218 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 262, + 201, + 316, + 213 + ], + "score": 0.92, + "content": "a _ { t + 1 } = c \\cdot \\eta _ { t } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 194, + 321, + 218 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 109, + 210, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 109, + 211, + 123, + 235 + ], + "score": 1.0, + "content": "7: 8:", + "type": "text" + }, + { + "bbox": [ + 151, + 210, + 162, + 240 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 162, + 213, + 167, + 222 + ], + "score": 0.66, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 213, + 213, + 222 + ], + "score": 0.83, + "content": "I = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 213, + 505, + 223 + ], + "score": 1.0, + "content": "#at the SN", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 163, + 221, + 288, + 236 + ], + "spans": [ + { + "bbox": [ + 163, + 221, + 288, + 236 + ], + "score": 0.39, + "content": "\\begin{array} { r } { d _ { t + 1 } ^ { ( k ) } = \\bar { d } _ { t + 1 } : = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } d _ { t + 1 } ^ { ( k ) } } \\end{array}", + "type": "inline_equation" + } + ], + "index": 9 + }, + { + "bbox": [ + 106, + 233, + 506, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 238, + 123, + 265 + ], + "score": 1.0, + "content": "9: 10:", + "type": "text" + }, + { + "bbox": [ + 146, + 234, + 159, + 273 + ], + "score": 1.0, + "content": "e", + "type": "text" + }, + { + "bbox": [ + 163, + 233, + 389, + 251 + ], + "score": 0.28, + "content": "\\begin{array} { r l } & { x _ { t + 2 } ^ { ( k ) ^ { \\bot } } : = \\bar { x } _ { t + 1 } - \\eta _ { t + 1 } \\bar { d } _ { t + 1 } = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } x _ { t + 1 } ^ { ( k ) } - \\eta _ { t + 1 } \\bar { d } _ { t + 1 } } \\end{array}", + "type": "inline_equation", + "image_path": "e67e65bdf0f2b24ad780e5ff967df0f8fef06b9743cde5d19e41839ae6127bab.jpg" + }, + { + "bbox": [ + 172, + 250, + 273, + 266 + ], + "score": 0.83, + "content": "x _ { t + 2 } ^ { ( k ) } = x _ { t + 1 } ^ { ( k ) } - \\eta _ { t + 1 } d _ { t + 1 } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 233, + 506, + 264 + ], + "score": 1.0, + "content": "#server-side momentum", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 263, + 181, + 275 + ], + "spans": [ + { + "bbox": [ + 106, + 263, + 122, + 275 + ], + "score": 1.0, + "content": "11:", + "type": "text" + }, + { + "bbox": [ + 151, + 263, + 181, + 274 + ], + "score": 1.0, + "content": "end if", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 274, + 172, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 274, + 122, + 286 + ], + "score": 1.0, + "content": "12:", + "type": "text" + }, + { + "bbox": [ + 137, + 274, + 172, + 286 + ], + "score": 1.0, + "content": "end for", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 285, + 158, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 158, + 298 + ], + "score": 1.0, + "content": "13: end for", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 295, + 269, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 160, + 309 + ], + "score": 1.0, + "content": "14: Return:", + "type": "text" + }, + { + "bbox": [ + 160, + 298, + 172, + 307 + ], + "score": 0.84, + "content": "\\scriptstyle { \\bar { x } } _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 295, + 200, + 309 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 200, + 296, + 266, + 309 + ], + "score": 0.91, + "content": "a \\sim \\mathcal { U } \\{ 1 , . . . , T \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 295, + 269, + 309 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 106, + 333, + 505, + 408 + ], + "lines": [ + { + "bbox": [ + 105, + 330, + 507, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 330, + 185, + 351 + ], + "score": 1.0, + "content": "consecutive iterates", + "type": "text" + }, + { + "bbox": [ + 185, + 333, + 205, + 348 + ], + "score": 0.92, + "content": "x _ { t + 1 } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 330, + 222, + 351 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 222, + 333, + 240, + 348 + ], + "score": 0.91, + "content": "x _ { t } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 330, + 291, + 351 + ], + "score": 1.0, + "content": ". After every", + "type": "text" + }, + { + "bbox": [ + 291, + 336, + 298, + 345 + ], + "score": 0.68, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 330, + 507, + 351 + ], + "score": 1.0, + "content": "local steps, the WNs share their current local models", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 107, + 342, + 510, + 370 + ], + "spans": [ + { + "bbox": [ + 107, + 347, + 152, + 364 + ], + "score": 0.93, + "content": "\\{ x _ { t + 1 } ^ { ( k ) } \\} _ { k = 1 } ^ { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 342, + 211, + 370 + ], + "score": 1.0, + "content": "and directions", + "type": "text" + }, + { + "bbox": [ + 211, + 348, + 255, + 363 + ], + "score": 0.94, + "content": "\\{ d _ { t + 1 } ^ { ( k ) } \\} _ { k = 1 } ^ { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 342, + 510, + 370 + ], + "score": 1.0, + "content": "with the SN. The SN aggregates these quantities, and performs", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 362, + 506, + 376 + ], + "spans": [ + { + "bbox": [ + 104, + 363, + 302, + 376 + ], + "score": 1.0, + "content": "a server-side momentum step, before returning", + "type": "text" + }, + { + "bbox": [ + 302, + 364, + 323, + 375 + ], + "score": 0.9, + "content": "\\bar { x } _ { t + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 363, + 342, + 376 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 342, + 362, + 362, + 375 + ], + "score": 0.92, + "content": "\\bar { d } _ { t + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 363, + 506, + 376 + ], + "score": 1.0, + "content": "to all the WNs. Because both the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 373, + 506, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 506, + 386 + ], + "score": 1.0, + "content": "WNs and the SN perform momentum based updates, we call the algorithm a stochastic two-sided", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 385, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 293, + 397 + ], + "score": 1.0, + "content": "momentum algorithm. The key parameters are:", + "type": "text" + }, + { + "bbox": [ + 294, + 385, + 299, + 395 + ], + "score": 0.72, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 385, + 376, + 397 + ], + "score": 1.0, + "content": "the minibatch size,", + "type": "text" + }, + { + "bbox": [ + 376, + 385, + 383, + 395 + ], + "score": 0.69, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 385, + 505, + 397 + ], + "score": 1.0, + "content": "the local update steps between", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 396, + 426, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 220, + 408 + ], + "score": 1.0, + "content": "two communication rounds,", + "type": "text" + }, + { + "bbox": [ + 221, + 398, + 230, + 408 + ], + "score": 0.84, + "content": "\\eta _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 396, + 303, + 408 + ], + "score": 1.0, + "content": "the stepsizes, and", + "type": "text" + }, + { + "bbox": [ + 303, + 397, + 313, + 407 + ], + "score": 0.85, + "content": "a _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 396, + 426, + 408 + ], + "score": 1.0, + "content": "the momentum parameters.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 106, + 412, + 506, + 533 + ], + "lines": [ + { + "bbox": [ + 106, + 412, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 106, + 412, + 505, + 425 + ], + "score": 1.0, + "content": "One key technical innovation of our algorithm design is to identify the most suitable way to incorporate", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 424, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 505, + 435 + ], + "score": 1.0, + "content": "momentum based directions in FL algorithms. Although the momentum-based gradient estimator", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 433, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 375, + 447 + ], + "score": 1.0, + "content": "itself is not new and has been used in the literature before (see e.g., in", + "type": "text" + }, + { + "bbox": [ + 375, + 434, + 397, + 446 + ], + "score": 0.27, + "content": "\\mathbb { \\left[ \\bigstar \\bigstar \\right] }", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 433, + 414, + 447 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 414, + 434, + 446, + 446 + ], + "score": 0.88, + "content": "\\overline { { \\mathbb { D } \\mathbb { Z } \\mathbb { G } \\mathbb { S } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 433, + 506, + 447 + ], + "score": 1.0, + "content": "to improve the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 444, + 506, + 459 + ], + "spans": [ + { + "bbox": [ + 104, + 444, + 506, + 459 + ], + "score": 1.0, + "content": "sample complexities of centralized and decentralized stochastic optimization problems, respectively),", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 455, + 506, + 468 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 506, + 468 + ], + "score": 1.0, + "content": "it is by no means clear if and how it can contribute to improve the communication complexity of FL", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 467, + 506, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 506, + 479 + ], + "score": 1.0, + "content": "algorithms. We show that in the FL setting, the local directions together with the local models have to", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 478, + 506, + 490 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 506, + 490 + ], + "score": 1.0, + "content": "be aggregated by the SN so to avoid being influenced too much by the local data. More importantly,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 488, + 507, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 507, + 501 + ], + "score": 1.0, + "content": "besides the WNs, the SN also needs to perform updates using the (aggregated) momentum directions.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 500, + 506, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 506, + 512 + ], + "score": 1.0, + "content": "Finally, such two-sided momentum updates have to be done carefully with the correct choice of", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 510, + 506, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 168, + 523 + ], + "score": 1.0, + "content": "minibatch size", + "type": "text" + }, + { + "bbox": [ + 168, + 511, + 173, + 520 + ], + "score": 0.46, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 510, + 310, + 523 + ], + "score": 1.0, + "content": ", and the number of local updates", + "type": "text" + }, + { + "bbox": [ + 310, + 511, + 316, + 520 + ], + "score": 0.41, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 510, + 506, + 523 + ], + "score": 1.0, + "content": ". Overall, it is the judicious choice of all these", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 522, + 445, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 522, + 445, + 534 + ], + "score": 1.0, + "content": "design elements that results in the optimal sample and communication complexities.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 107, + 537, + 386, + 549 + ], + "lines": [ + { + "bbox": [ + 105, + 536, + 388, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 536, + 388, + 552 + ], + "score": 1.0, + "content": "Next, we present the convergence guarantees of the STEM algorithm.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "title", + "bbox": [ + 107, + 564, + 336, + 576 + ], + "lines": [ + { + "bbox": [ + 105, + 564, + 337, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 337, + 577 + ], + "score": 1.0, + "content": "3.1 Main results: convergence guarantees for STEM", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 585, + 505, + 619 + ], + "lines": [ + { + "bbox": [ + 106, + 585, + 505, + 597 + ], + "spans": [ + { + "bbox": [ + 106, + 585, + 505, + 597 + ], + "score": 1.0, + "content": "In this section, we analyze the performance of STEM. We first present our main result, and then", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 596, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 505, + 608 + ], + "score": 1.0, + "content": "provide discussions about a few parameter choices. In the next subsection, we discuss a special case", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 606, + 403, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 403, + 619 + ], + "score": 1.0, + "content": "of STEM related to the classical FedAvg and minibatch SGD algorithms.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 103, + 622, + 471, + 635 + ], + "lines": [ + { + "bbox": [ + 104, + 621, + 474, + 637 + ], + "spans": [ + { + "bbox": [ + 104, + 621, + 474, + 637 + ], + "score": 1.0, + "content": "Theorem 3.1. Under the Assumptions 1 and 2, suppose the stepsize sequence is chosen as:", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + }, + { + "type": "interline_equation", + "bbox": [ + 262, + 641, + 348, + 667 + ], + "lines": [ + { + "bbox": [ + 262, + 641, + 348, + 667 + ], + "spans": [ + { + "bbox": [ + 262, + 641, + 348, + 667 + ], + "score": 0.96, + "content": "\\eta _ { t } = \\frac { \\bar { \\kappa } } { ( w _ { t } + \\sigma ^ { 2 } t ) ^ { 1 / 3 } } ,", + "type": "interline_equation", + "image_path": "c7b9d663b7230014554f849f65c327948834a1aadf2fddbf4fd4727068678b71.jpg" + } + ] + } + ], + "index": 38, + "virtual_lines": [ + { + "bbox": [ + 262, + 641, + 348, + 667 + ], + "spans": [], + "index": 38 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 674, + 177, + 685 + ], + "lines": [ + { + "bbox": [ + 106, + 673, + 180, + 686 + ], + "spans": [ + { + "bbox": [ + 106, + 673, + 180, + 686 + ], + "score": 1.0, + "content": "where we define :", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "interline_equation", + "bbox": [ + 155, + 692, + 452, + 721 + ], + "lines": [ + { + "bbox": [ + 155, + 692, + 452, + 721 + ], + "spans": [ + { + "bbox": [ + 155, + 692, + 452, + 721 + ], + "score": 0.93, + "content": "\\bar { \\kappa } = \\frac { ( b K ) ^ { 2 / 3 } \\sigma ^ { 2 / 3 } } { L } , \\quad w _ { t } = \\operatorname * { m a x } \\bigg \\{ 2 \\sigma ^ { 2 } , 4 0 9 6 L ^ { 3 } I ^ { 3 } \\bar { \\kappa } ^ { 3 } - \\sigma ^ { 2 } t , \\frac { c ^ { 3 } \\bar { \\kappa } ^ { 3 } } { 4 0 9 6 L ^ { 3 } I ^ { 3 } } \\bigg \\} .", + "type": "interline_equation", + "image_path": "1ab3f1b7115347d40a7bc1f3343dd7707b65537827e4b8370ab97442863fcfd1.jpg" + } + ] + } + ], + "index": 41, + "virtual_lines": [ + { + "bbox": [ + 155, + 692, + 452, + 701.6666666666666 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 155, + 701.6666666666666, + 452, + 711.3333333333333 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 155, + 711.3333333333333, + 452, + 720.9999999999999 + ], + "spans": [], + "index": 42 + } + ] + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 753 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 753 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 107, + 72, + 400, + 84 + ], + "lines": [ + { + "bbox": [ + 104, + 69, + 401, + 88 + ], + "spans": [ + { + "bbox": [ + 104, + 69, + 401, + 88 + ], + "score": 1.0, + "content": "Algorithm 1 The Stochastic Two-Sided Momemtum (STEM) Algorithm", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "index", + "bbox": [ + 109, + 86, + 506, + 312 + ], + "lines": [ + { + "bbox": [ + 110, + 87, + 467, + 102 + ], + "spans": [ + { + "bbox": [ + 110, + 87, + 203, + 102 + ], + "score": 1.0, + "content": "1: Input: Parameters:", + "type": "text" + }, + { + "bbox": [ + 203, + 90, + 227, + 99 + ], + "score": 0.89, + "content": "c > 0", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 87, + 343, + 102 + ], + "score": 1.0, + "content": ", the number of local updates", + "type": "text" + }, + { + "bbox": [ + 344, + 90, + 350, + 99 + ], + "score": 0.74, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 351, + 87, + 396, + 102 + ], + "score": 1.0, + "content": ", batch size", + "type": "text" + }, + { + "bbox": [ + 396, + 90, + 402, + 99 + ], + "score": 0.74, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 87, + 443, + 102 + ], + "score": 1.0, + "content": ", stepsizes", + "type": "text" + }, + { + "bbox": [ + 443, + 89, + 462, + 101 + ], + "score": 0.9, + "content": "\\{ \\eta _ { t } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 463, + 87, + 467, + 102 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 109, + 96, + 503, + 120 + ], + "spans": [ + { + "bbox": [ + 109, + 96, + 197, + 120 + ], + "score": 1.0, + "content": "2: Initialize: Iterate", + "type": "text" + }, + { + "bbox": [ + 198, + 100, + 312, + 115 + ], + "score": 0.91, + "content": "\\begin{array} { r } { x _ { 1 } ^ { ( k ) } = \\bar { x } _ { 1 } = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } x _ { 1 } ^ { ( k ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 96, + 390, + 120 + ], + "score": 1.0, + "content": ", descent direction", + "type": "text" + }, + { + "bbox": [ + 390, + 100, + 503, + 116 + ], + "score": 0.92, + "content": "\\begin{array} { r } { d _ { 1 } ^ { ( k ) } = \\bar { d } _ { 1 } = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } d _ { 1 } ^ { ( k ) } } \\end{array}", + "type": "inline_equation" + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 122, + 111, + 420, + 138 + ], + "spans": [ + { + "bbox": [ + 122, + 111, + 143, + 138 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 144, + 115, + 302, + 133 + ], + "score": 0.91, + "content": "\\begin{array} { r } { d _ { 1 } ^ { ( k ) } = \\frac { 1 } { B } \\sum _ { \\xi _ { 1 } ^ { ( k ) } \\in \\mathcal { B } _ { 1 } ^ { ( k ) } } \\nabla f ^ { ( k ) } ( x _ { 1 } ^ { ( k ) } ; \\xi _ { 1 } ^ { ( k ) } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 111, + 320, + 138 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 320, + 115, + 366, + 130 + ], + "score": 0.93, + "content": "| B _ { 1 } ^ { ( k ) } | = B", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 111, + 381, + 138 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 381, + 117, + 415, + 129 + ], + "score": 0.91, + "content": "k \\in [ K ]", + "type": "inline_equation" + }, + { + "bbox": [ + 415, + 111, + 420, + 138 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 105, + 127, + 289, + 152 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 162, + 152 + ], + "score": 1.0, + "content": "3: Perform:", + "type": "text" + }, + { + "bbox": [ + 162, + 133, + 241, + 147 + ], + "score": 0.88, + "content": "x _ { 2 } ^ { ( k ) } = x _ { 1 } ^ { k } - \\eta _ { 1 } d _ { 1 } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 127, + 246, + 152 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 247, + 134, + 289, + 147 + ], + "score": 0.67, + "content": "\\forall k \\in [ K ]", + "type": "inline_equation" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 109, + 145, + 197, + 158 + ], + "spans": [ + { + "bbox": [ + 109, + 145, + 138, + 158 + ], + "score": 1.0, + "content": "4: for", + "type": "text" + }, + { + "bbox": [ + 138, + 146, + 162, + 156 + ], + "score": 0.88, + "content": "t = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 145, + 173, + 158 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 173, + 147, + 182, + 156 + ], + "score": 0.71, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 145, + 197, + 158 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 157, + 505, + 169 + ], + "spans": [ + { + "bbox": [ + 110, + 157, + 122, + 169 + ], + "score": 1.0, + "content": "5:", + "type": "text" + }, + { + "bbox": [ + 137, + 157, + 153, + 168 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 154, + 157, + 178, + 167 + ], + "score": 0.84, + "content": "k = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 157, + 190, + 168 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 190, + 157, + 200, + 167 + ], + "score": 0.72, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 157, + 216, + 168 + ], + "score": 1.0, + "content": "do", + "type": "text" + }, + { + "bbox": [ + 449, + 157, + 505, + 168 + ], + "score": 1.0, + "content": "#at the WN", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 163, + 504, + 199 + ], + "spans": [ + { + "bbox": [ + 110, + 171, + 122, + 183 + ], + "score": 1.0, + "content": "6:", + "type": "text" + }, + { + "bbox": [ + 152, + 163, + 504, + 199 + ], + "score": 0.81, + "content": "\\mathcal { d } _ { t + 1 } ^ { ( k ) } = \\frac { 1 } { b } \\sum _ { \\xi _ { t + 1 } ^ { ( k ) } \\in \\mathcal { B } _ { t + 1 } ^ { ( k ) } } ^ { \\pi \\Delta } \\nabla f ^ { ( k ) } ( x _ { t + 1 } ^ { ( k ) } ; \\xi _ { t + 1 } ^ { ( k ) } ) + \\left( 1 - a _ { t + 1 } \\right) \\bigg ( d _ { t } ^ { ( k ) } - \\frac { 1 } { b } \\sum _ { \\xi _ { t + 1 } ^ { ( k ) } \\in \\mathcal { B } _ { t + 1 } ^ { ( k ) } } \\nabla f ^ { ( k ) } ( x _ { t } ^ { ( k ) } ; \\xi _ { t + 1 } ^ { ( k ) } ) \\bigg )", + "type": "inline_equation", + "image_path": "cb47ddefffada4ba7399d00907097a5fbdab3bb233d14f9fdff1de621faadef2.jpg" + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 123, + 194, + 321, + 218 + ], + "spans": [ + { + "bbox": [ + 123, + 194, + 196, + 218 + ], + "score": 1.0, + "content": "where we choose", + "type": "text" + }, + { + "bbox": [ + 197, + 199, + 241, + 214 + ], + "score": 0.89, + "content": "| B _ { t + 1 } ^ { ( k ) } | = b", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 194, + 262, + 218 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 262, + 201, + 316, + 213 + ], + "score": 0.92, + "content": "a _ { t + 1 } = c \\cdot \\eta _ { t } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 194, + 321, + 218 + ], + "score": 1.0, + "content": ";", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 109, + 210, + 505, + 240 + ], + "spans": [ + { + "bbox": [ + 109, + 211, + 123, + 235 + ], + "score": 1.0, + "content": "7: 8:", + "type": "text" + }, + { + "bbox": [ + 151, + 210, + 162, + 240 + ], + "score": 1.0, + "content": "if", + "type": "text" + }, + { + "bbox": [ + 162, + 213, + 167, + 222 + ], + "score": 0.66, + "content": "t", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 213, + 213, + 222 + ], + "score": 0.83, + "content": "I = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 213, + 505, + 223 + ], + "score": 1.0, + "content": "#at the SN", + "type": "text" + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 163, + 221, + 288, + 236 + ], + "spans": [ + { + "bbox": [ + 163, + 221, + 288, + 236 + ], + "score": 0.39, + "content": "\\begin{array} { r } { d _ { t + 1 } ^ { ( k ) } = \\bar { d } _ { t + 1 } : = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } d _ { t + 1 } ^ { ( k ) } } \\end{array}", + "type": "inline_equation" + } + ], + "index": 9, + "is_list_start_line": true + }, + { + "bbox": [ + 106, + 233, + 506, + 273 + ], + "spans": [ + { + "bbox": [ + 106, + 238, + 123, + 265 + ], + "score": 1.0, + "content": "9: 10:", + "type": "text" + }, + { + "bbox": [ + 146, + 234, + 159, + 273 + ], + "score": 1.0, + "content": "e", + "type": "text" + }, + { + "bbox": [ + 163, + 233, + 389, + 251 + ], + "score": 0.28, + "content": "\\begin{array} { r l } & { x _ { t + 2 } ^ { ( k ) ^ { \\bot } } : = \\bar { x } _ { t + 1 } - \\eta _ { t + 1 } \\bar { d } _ { t + 1 } = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } x _ { t + 1 } ^ { ( k ) } - \\eta _ { t + 1 } \\bar { d } _ { t + 1 } } \\end{array}", + "type": "inline_equation", + "image_path": "e67e65bdf0f2b24ad780e5ff967df0f8fef06b9743cde5d19e41839ae6127bab.jpg" + }, + { + "bbox": [ + 172, + 250, + 273, + 266 + ], + "score": 0.83, + "content": "x _ { t + 2 } ^ { ( k ) } = x _ { t + 1 } ^ { ( k ) } - \\eta _ { t + 1 } d _ { t + 1 } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 233, + 506, + 264 + ], + "score": 1.0, + "content": "#server-side momentum", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 106, + 263, + 181, + 275 + ], + "spans": [ + { + "bbox": [ + 106, + 263, + 122, + 275 + ], + "score": 1.0, + "content": "11:", + "type": "text" + }, + { + "bbox": [ + 151, + 263, + 181, + 274 + ], + "score": 1.0, + "content": "end if", + "type": "text" + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 106, + 274, + 172, + 286 + ], + "spans": [ + { + "bbox": [ + 106, + 274, + 122, + 286 + ], + "score": 1.0, + "content": "12:", + "type": "text" + }, + { + "bbox": [ + 137, + 274, + 172, + 286 + ], + "score": 1.0, + "content": "end for", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 105, + 285, + 158, + 298 + ], + "spans": [ + { + "bbox": [ + 105, + 285, + 158, + 298 + ], + "score": 1.0, + "content": "13: end for", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 105, + 295, + 269, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 295, + 160, + 309 + ], + "score": 1.0, + "content": "14: Return:", + "type": "text" + }, + { + "bbox": [ + 160, + 298, + 172, + 307 + ], + "score": 0.84, + "content": "\\scriptstyle { \\bar { x } } _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 295, + 200, + 309 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 200, + 296, + 266, + 309 + ], + "score": 0.91, + "content": "a \\sim \\mathcal { U } \\{ 1 , . . . , T \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 295, + 269, + 309 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 14, + "is_list_start_line": true + } + ], + "index": 7, + "bbox_fs": [ + 105, + 87, + 506, + 309 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 333, + 505, + 408 + ], + "lines": [ + { + "bbox": [ + 105, + 330, + 507, + 351 + ], + "spans": [ + { + "bbox": [ + 105, + 330, + 185, + 351 + ], + "score": 1.0, + "content": "consecutive iterates", + "type": "text" + }, + { + "bbox": [ + 185, + 333, + 205, + 348 + ], + "score": 0.92, + "content": "x _ { t + 1 } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 205, + 330, + 222, + 351 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 222, + 333, + 240, + 348 + ], + "score": 0.91, + "content": "x _ { t } ^ { ( k ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 330, + 291, + 351 + ], + "score": 1.0, + "content": ". After every", + "type": "text" + }, + { + "bbox": [ + 291, + 336, + 298, + 345 + ], + "score": 0.68, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 330, + 507, + 351 + ], + "score": 1.0, + "content": "local steps, the WNs share their current local models", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 107, + 342, + 510, + 370 + ], + "spans": [ + { + "bbox": [ + 107, + 347, + 152, + 364 + ], + "score": 0.93, + "content": "\\{ x _ { t + 1 } ^ { ( k ) } \\} _ { k = 1 } ^ { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 342, + 211, + 370 + ], + "score": 1.0, + "content": "and directions", + "type": "text" + }, + { + "bbox": [ + 211, + 348, + 255, + 363 + ], + "score": 0.94, + "content": "\\{ d _ { t + 1 } ^ { ( k ) } \\} _ { k = 1 } ^ { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 342, + 510, + 370 + ], + "score": 1.0, + "content": "with the SN. The SN aggregates these quantities, and performs", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 362, + 506, + 376 + ], + "spans": [ + { + "bbox": [ + 104, + 363, + 302, + 376 + ], + "score": 1.0, + "content": "a server-side momentum step, before returning", + "type": "text" + }, + { + "bbox": [ + 302, + 364, + 323, + 375 + ], + "score": 0.9, + "content": "\\bar { x } _ { t + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 363, + 342, + 376 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 342, + 362, + 362, + 375 + ], + "score": 0.92, + "content": "\\bar { d } _ { t + 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 363, + 506, + 376 + ], + "score": 1.0, + "content": "to all the WNs. Because both the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 373, + 506, + 386 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 506, + 386 + ], + "score": 1.0, + "content": "WNs and the SN perform momentum based updates, we call the algorithm a stochastic two-sided", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 385, + 505, + 397 + ], + "spans": [ + { + "bbox": [ + 105, + 385, + 293, + 397 + ], + "score": 1.0, + "content": "momentum algorithm. The key parameters are:", + "type": "text" + }, + { + "bbox": [ + 294, + 385, + 299, + 395 + ], + "score": 0.72, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 385, + 376, + 397 + ], + "score": 1.0, + "content": "the minibatch size,", + "type": "text" + }, + { + "bbox": [ + 376, + 385, + 383, + 395 + ], + "score": 0.69, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 385, + 505, + 397 + ], + "score": 1.0, + "content": "the local update steps between", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 396, + 426, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 396, + 220, + 408 + ], + "score": 1.0, + "content": "two communication rounds,", + "type": "text" + }, + { + "bbox": [ + 221, + 398, + 230, + 408 + ], + "score": 0.84, + "content": "\\eta _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 396, + 303, + 408 + ], + "score": 1.0, + "content": "the stepsizes, and", + "type": "text" + }, + { + "bbox": [ + 303, + 397, + 313, + 407 + ], + "score": 0.85, + "content": "a _ { t }", + "type": "inline_equation" + }, + { + "bbox": [ + 313, + 396, + 426, + 408 + ], + "score": 1.0, + "content": "the momentum parameters.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 17.5, + "bbox_fs": [ + 104, + 330, + 510, + 408 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 412, + 506, + 533 + ], + "lines": [ + { + "bbox": [ + 106, + 412, + 505, + 425 + ], + "spans": [ + { + "bbox": [ + 106, + 412, + 505, + 425 + ], + "score": 1.0, + "content": "One key technical innovation of our algorithm design is to identify the most suitable way to incorporate", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 424, + 505, + 435 + ], + "spans": [ + { + "bbox": [ + 106, + 424, + 505, + 435 + ], + "score": 1.0, + "content": "momentum based directions in FL algorithms. Although the momentum-based gradient estimator", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 433, + 506, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 375, + 447 + ], + "score": 1.0, + "content": "itself is not new and has been used in the literature before (see e.g., in", + "type": "text" + }, + { + "bbox": [ + 375, + 434, + 397, + 446 + ], + "score": 0.27, + "content": "\\mathbb { \\left[ \\bigstar \\bigstar \\right] }", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 433, + 414, + 447 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 414, + 434, + 446, + 446 + ], + "score": 0.88, + "content": "\\overline { { \\mathbb { D } \\mathbb { Z } \\mathbb { G } \\mathbb { S } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 433, + 506, + 447 + ], + "score": 1.0, + "content": "to improve the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 444, + 506, + 459 + ], + "spans": [ + { + "bbox": [ + 104, + 444, + 506, + 459 + ], + "score": 1.0, + "content": "sample complexities of centralized and decentralized stochastic optimization problems, respectively),", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 455, + 506, + 468 + ], + "spans": [ + { + "bbox": [ + 105, + 455, + 506, + 468 + ], + "score": 1.0, + "content": "it is by no means clear if and how it can contribute to improve the communication complexity of FL", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 467, + 506, + 479 + ], + "spans": [ + { + "bbox": [ + 105, + 467, + 506, + 479 + ], + "score": 1.0, + "content": "algorithms. We show that in the FL setting, the local directions together with the local models have to", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 478, + 506, + 490 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 506, + 490 + ], + "score": 1.0, + "content": "be aggregated by the SN so to avoid being influenced too much by the local data. More importantly,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 488, + 507, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 507, + 501 + ], + "score": 1.0, + "content": "besides the WNs, the SN also needs to perform updates using the (aggregated) momentum directions.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 500, + 506, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 506, + 512 + ], + "score": 1.0, + "content": "Finally, such two-sided momentum updates have to be done carefully with the correct choice of", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 510, + 506, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 168, + 523 + ], + "score": 1.0, + "content": "minibatch size", + "type": "text" + }, + { + "bbox": [ + 168, + 511, + 173, + 520 + ], + "score": 0.46, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 510, + 310, + 523 + ], + "score": 1.0, + "content": ", and the number of local updates", + "type": "text" + }, + { + "bbox": [ + 310, + 511, + 316, + 520 + ], + "score": 0.41, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 510, + 506, + 523 + ], + "score": 1.0, + "content": ". Overall, it is the judicious choice of all these", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 522, + 445, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 522, + 445, + 534 + ], + "score": 1.0, + "content": "design elements that results in the optimal sample and communication complexities.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 26, + "bbox_fs": [ + 104, + 412, + 507, + 534 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 537, + 386, + 549 + ], + "lines": [ + { + "bbox": [ + 105, + 536, + 388, + 552 + ], + "spans": [ + { + "bbox": [ + 105, + 536, + 388, + 552 + ], + "score": 1.0, + "content": "Next, we present the convergence guarantees of the STEM algorithm.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32, + "bbox_fs": [ + 105, + 536, + 388, + 552 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 564, + 336, + 576 + ], + "lines": [ + { + "bbox": [ + 105, + 564, + 337, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 337, + 577 + ], + "score": 1.0, + "content": "3.1 Main results: convergence guarantees for STEM", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 107, + 585, + 505, + 619 + ], + "lines": [ + { + "bbox": [ + 106, + 585, + 505, + 597 + ], + "spans": [ + { + "bbox": [ + 106, + 585, + 505, + 597 + ], + "score": 1.0, + "content": "In this section, we analyze the performance of STEM. We first present our main result, and then", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 596, + 505, + 608 + ], + "spans": [ + { + "bbox": [ + 105, + 596, + 505, + 608 + ], + "score": 1.0, + "content": "provide discussions about a few parameter choices. In the next subsection, we discuss a special case", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 606, + 403, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 403, + 619 + ], + "score": 1.0, + "content": "of STEM related to the classical FedAvg and minibatch SGD algorithms.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35, + "bbox_fs": [ + 105, + 585, + 505, + 619 + ] + }, + { + "type": "text", + "bbox": [ + 103, + 622, + 471, + 635 + ], + "lines": [ + { + "bbox": [ + 104, + 621, + 474, + 637 + ], + "spans": [ + { + "bbox": [ + 104, + 621, + 474, + 637 + ], + "score": 1.0, + "content": "Theorem 3.1. Under the Assumptions 1 and 2, suppose the stepsize sequence is chosen as:", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37, + "bbox_fs": [ + 104, + 621, + 474, + 637 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 262, + 641, + 348, + 667 + ], + "lines": [ + { + "bbox": [ + 262, + 641, + 348, + 667 + ], + "spans": [ + { + "bbox": [ + 262, + 641, + 348, + 667 + ], + "score": 0.96, + "content": "\\eta _ { t } = \\frac { \\bar { \\kappa } } { ( w _ { t } + \\sigma ^ { 2 } t ) ^ { 1 / 3 } } ,", + "type": "interline_equation", + "image_path": "c7b9d663b7230014554f849f65c327948834a1aadf2fddbf4fd4727068678b71.jpg" + } + ] + } + ], + "index": 38, + "virtual_lines": [ + { + "bbox": [ + 262, + 641, + 348, + 667 + ], + "spans": [], + "index": 38 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 674, + 177, + 685 + ], + "lines": [ + { + "bbox": [ + 106, + 673, + 180, + 686 + ], + "spans": [ + { + "bbox": [ + 106, + 673, + 180, + 686 + ], + "score": 1.0, + "content": "where we define :", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39, + "bbox_fs": [ + 106, + 673, + 180, + 686 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 155, + 692, + 452, + 721 + ], + "lines": [ + { + "bbox": [ + 155, + 692, + 452, + 721 + ], + "spans": [ + { + "bbox": [ + 155, + 692, + 452, + 721 + ], + "score": 0.93, + "content": "\\bar { \\kappa } = \\frac { ( b K ) ^ { 2 / 3 } \\sigma ^ { 2 / 3 } } { L } , \\quad w _ { t } = \\operatorname * { m a x } \\bigg \\{ 2 \\sigma ^ { 2 } , 4 0 9 6 L ^ { 3 } I ^ { 3 } \\bar { \\kappa } ^ { 3 } - \\sigma ^ { 2 } t , \\frac { c ^ { 3 } \\bar { \\kappa } ^ { 3 } } { 4 0 9 6 L ^ { 3 } I ^ { 3 } } \\bigg \\} .", + "type": "interline_equation", + "image_path": "1ab3f1b7115347d40a7bc1f3343dd7707b65537827e4b8370ab97442863fcfd1.jpg" + } + ] + } + ], + "index": 41, + "virtual_lines": [ + { + "bbox": [ + 155, + 692, + 452, + 701.6666666666666 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 155, + 701.6666666666666, + 452, + 711.3333333333333 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 155, + 711.3333333333333, + 452, + 720.9999999999999 + ], + "spans": [], + "index": 42 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 70, + 505, + 106 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 76, + 182, + 89 + ], + "score": 1.0, + "content": "Further, let us set", + "type": "text" + }, + { + "bbox": [ + 183, + 72, + 368, + 96 + ], + "score": 0.95, + "content": "\\begin{array} { r } { c = \\frac { 6 4 L ^ { 2 } } { b K } + \\frac { \\sigma ^ { 2 } } { 2 4 \\bar { \\kappa } ^ { 3 } L I } = L ^ { 2 } \\bigg ( \\frac { 6 4 } { b K } + \\frac { 1 } { 2 4 ( b K ) ^ { 2 } I } \\bigg ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 77, + 506, + 89 + ], + "score": 1.0, + "content": "and set the initial batch size as", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 107, + 93, + 365, + 106 + ], + "spans": [ + { + "bbox": [ + 107, + 95, + 138, + 105 + ], + "score": 0.88, + "content": "B = b I", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 93, + 227, + 106 + ], + "score": 1.0, + "content": "; set the local updates", + "type": "text" + }, + { + "bbox": [ + 227, + 95, + 234, + 104 + ], + "score": 0.58, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 93, + 365, + 106 + ], + "score": 1.0, + "content": "and minibatch size b as follows:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "interline_equation", + "bbox": [ + 206, + 112, + 404, + 128 + ], + "lines": [ + { + "bbox": [ + 206, + 112, + 404, + 128 + ], + "spans": [ + { + "bbox": [ + 206, + 112, + 404, + 128 + ], + "score": 0.91, + "content": "I = \\mathcal { O } \\big ( ( T / K ^ { 2 } ) ^ { \\nu / 3 } \\big ) , \\quad b = \\mathcal { O } \\big ( ( T / K ^ { 2 } ) ^ { 1 / 2 - \\nu / 2 } \\big )", + "type": "interline_equation", + "image_path": "571cd8b7f6de9fecec247a32a77a5fc7ff44b9b1a1446281dd4865af343e8ec8.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 206, + 112, + 404, + 128 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 133, + 372, + 147 + ], + "lines": [ + { + "bbox": [ + 106, + 133, + 372, + 148 + ], + "spans": [ + { + "bbox": [ + 106, + 133, + 133, + 148 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 137, + 140, + 144 + ], + "score": 0.56, + "content": "\\nu", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 133, + 175, + 148 + ], + "score": 1.0, + "content": "satisfies", + "type": "text" + }, + { + "bbox": [ + 175, + 135, + 214, + 147 + ], + "score": 0.92, + "content": "\\nu \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 133, + 372, + 148 + ], + "score": 1.0, + "content": ". Then for STEM the following holds:", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + }, + { + "type": "text", + "bbox": [ + 108, + 155, + 326, + 168 + ], + "lines": [ + { + "bbox": [ + 108, + 154, + 327, + 170 + ], + "spans": [ + { + "bbox": [ + 108, + 154, + 140, + 170 + ], + "score": 1.0, + "content": "(i) For", + "type": "text" + }, + { + "bbox": [ + 140, + 157, + 151, + 167 + ], + "score": 0.85, + "content": "\\scriptstyle { \\bar { x } } _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 154, + 277, + 170 + ], + "score": 1.0, + "content": "chosen according to Algorithm", + "type": "text" + }, + { + "bbox": [ + 277, + 155, + 287, + 169 + ], + "score": 0.5, + "content": "\\boldsymbol { l } ,", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 154, + 327, + 170 + ], + "score": 1.0, + "content": "we have:", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "interline_equation", + "bbox": [ + 137, + 173, + 479, + 201 + ], + "lines": [ + { + "bbox": [ + 137, + 173, + 479, + 201 + ], + "spans": [ + { + "bbox": [ + 137, + 173, + 479, + 201 + ], + "score": 0.92, + "content": "\\mathbb { E } \\Vert \\nabla f ( \\bar { x } _ { a } ) \\Vert ^ { 2 } = \\mathcal { O } \\Bigg ( \\frac { f ( \\bar { x } _ { 1 } ) - f ^ { * } } { K ^ { 2 \\nu / 3 } T ^ { 1 - \\nu / 3 } } \\Bigg ) + \\tilde { \\mathcal { O } } \\Bigg ( \\frac { \\sigma ^ { 2 } } { K ^ { 2 \\nu / 3 } T ^ { 1 - \\nu / 3 } } \\Bigg ) + \\tilde { \\mathcal { O } } \\Bigg ( \\frac { \\zeta ^ { 2 } } { K ^ { 2 \\nu / 3 } T ^ { 1 - \\nu / 3 } } \\Bigg ) .", + "type": "interline_equation", + "image_path": "e2c146df205c691e32ce24c2ee693fb97ea38edc895ba31bbeb511db58c5a452.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 137, + 173, + 479, + 182.33333333333334 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 137, + 182.33333333333334, + 479, + 191.66666666666669 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 137, + 191.66666666666669, + 479, + 201.00000000000003 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 212, + 234, + 225 + ], + "lines": [ + { + "bbox": [ + 105, + 212, + 235, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 156, + 226 + ], + "score": 1.0, + "content": "(ii) For any", + "type": "text" + }, + { + "bbox": [ + 157, + 213, + 195, + 225 + ], + "score": 0.91, + "content": "\\nu \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 212, + 235, + 226 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 108, + 225, + 505, + 261 + ], + "lines": [ + { + "bbox": [ + 122, + 223, + 506, + 239 + ], + "spans": [ + { + "bbox": [ + 122, + 223, + 353, + 239 + ], + "score": 1.0, + "content": "Sample Complexity: The sample complexity of STEM is", + "type": "text" + }, + { + "bbox": [ + 353, + 224, + 392, + 238 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 223, + 506, + 239 + ], + "score": 1.0, + "content": ". This implies that each WN", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 122, + 236, + 506, + 252 + ], + "spans": [ + { + "bbox": [ + 122, + 236, + 191, + 252 + ], + "score": 1.0, + "content": "requires at most", + "type": "text" + }, + { + "bbox": [ + 192, + 237, + 250, + 250 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal { O } } ( K ^ { - 1 } \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 251, + 236, + 506, + 252 + ], + "score": 1.0, + "content": "gradient computations, thereby achieving linear speedup with", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 123, + 249, + 295, + 262 + ], + "spans": [ + { + "bbox": [ + 123, + 249, + 295, + 262 + ], + "score": 1.0, + "content": "the number of WNs present in the network.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 124, + 261, + 456, + 273 + ], + "lines": [ + { + "bbox": [ + 122, + 259, + 456, + 275 + ], + "spans": [ + { + "bbox": [ + 122, + 260, + 421, + 275 + ], + "score": 1.0, + "content": "Communication Complexity: The communication complexity of STEM is", + "type": "text" + }, + { + "bbox": [ + 422, + 259, + 453, + 273 + ], + "score": 0.92, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 260, + 456, + 275 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 107, + 282, + 481, + 294 + ], + "lines": [ + { + "bbox": [ + 105, + 281, + 482, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 482, + 295 + ], + "score": 1.0, + "content": "The proof of this result is relegated to the Supplemental Material. A few remarks are in order.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 106, + 295, + 505, + 418 + ], + "lines": [ + { + "bbox": [ + 105, + 294, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 432, + 308 + ], + "score": 1.0, + "content": "Remark 1 (Near-Optimal sample and communication complexities). Theorem", + "type": "text" + }, + { + "bbox": [ + 433, + 294, + 450, + 307 + ], + "score": 0.69, + "content": "3 . 1", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 294, + 506, + 308 + ], + "score": 1.0, + "content": "suggests that", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 306, + 506, + 321 + ], + "spans": [ + { + "bbox": [ + 104, + 306, + 131, + 321 + ], + "score": 1.0, + "content": "when", + "type": "text" + }, + { + "bbox": [ + 131, + 308, + 138, + 318 + ], + "score": 0.67, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 306, + 156, + 321 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 157, + 308, + 163, + 318 + ], + "score": 0.71, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 306, + 363, + 321 + ], + "score": 1.0, + "content": "are selected appropriately, then STEM achieves", + "type": "text" + }, + { + "bbox": [ + 363, + 306, + 403, + 320 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 306, + 423, + 321 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 423, + 307, + 454, + 320 + ], + "score": 0.92, + "content": "\\tilde { \\mathcal { O } } ( \\overline { { \\epsilon ^ { - 1 } } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 306, + 506, + 321 + ], + "score": 1.0, + "content": "sample and", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 318, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 505, + 331 + ], + "score": 1.0, + "content": "communication complexities. Taking them separately, these complexity bounds are the best achievable", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 329, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 505, + 342 + ], + "score": 1.0, + "content": "by the existing FL algorithms (upto logarithmic factors regardless of sample or batch Lipschitz smooth", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 339, + 506, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 157, + 354 + ], + "score": 1.0, + "content": "assumption)", + "type": "text" + }, + { + "bbox": [ + 157, + 341, + 175, + 353 + ], + "score": 0.73, + "content": "[ [ 3 5 ] ]", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 339, + 217, + 354 + ], + "score": 1.0, + "content": "; see Table", + "type": "text" + }, + { + "bbox": [ + 217, + 340, + 227, + 354 + ], + "score": 0.32, + "content": "1 .", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 339, + 295, + 354 + ], + "score": 1.0, + "content": "We note that the", + "type": "text" + }, + { + "bbox": [ + 295, + 340, + 335, + 354 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 339, + 506, + 354 + ], + "score": 1.0, + "content": "complexity is the best possible that can be", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 352, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 435, + 365 + ], + "score": 1.0, + "content": "achieved by centralized SGD with the sample Lipschitz gradient assumption; see", + "type": "text" + }, + { + "bbox": [ + 435, + 352, + 448, + 364 + ], + "score": 0.75, + "content": " { \\mathbb { I } } ^ { { \\left[ 5 \\right] } }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 352, + 506, + 365 + ], + "score": 1.0, + "content": ". On the other", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 363, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 146, + 375 + ], + "score": 1.0, + "content": "hand, the", + "type": "text" + }, + { + "bbox": [ + 146, + 363, + 177, + 375 + ], + "score": 0.93, + "content": "\\bar { \\mathcal { O } } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 363, + 411, + 375 + ], + "score": 1.0, + "content": "complexity bound is also likely to be the optimal, since in", + "type": "text" + }, + { + "bbox": [ + 412, + 363, + 424, + 374 + ], + "score": 0.67, + "content": "\\mathbb { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 363, + 506, + 375 + ], + "score": 1.0, + "content": "the authors showed", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 374, + 506, + 386 + ], + "spans": [ + { + "bbox": [ + 106, + 374, + 430, + 386 + ], + "score": 1.0, + "content": "that even when the local steps use a class of (deterministic) first-order algorithms,", + "type": "text" + }, + { + "bbox": [ + 430, + 374, + 461, + 386 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 374, + 506, + 386 + ], + "score": 1.0, + "content": "is the best", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 384, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 380, + 398 + ], + "score": 1.0, + "content": "achievable communication complexity. The only difference is that", + "type": "text" + }, + { + "bbox": [ + 380, + 385, + 393, + 396 + ], + "score": 0.66, + "content": "\\bigstar \\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 384, + 505, + 398 + ], + "score": 1.0, + "content": "does not explicitly assume", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 395, + 506, + 409 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 389, + 409 + ], + "score": 1.0, + "content": "the inter-node variance bound (i.e., the second relation in Assumption", + "type": "text" + }, + { + "bbox": [ + 389, + 396, + 397, + 409 + ], + "score": 0.72, + "content": "2 \\cdot", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 395, + 506, + 409 + ], + "score": 1.0, + "content": "-(ii)). We leave the precise", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 407, + 504, + 418 + ], + "spans": [ + { + "bbox": [ + 106, + 407, + 478, + 418 + ], + "score": 1.0, + "content": "characterization of the communication lower bound with inter-node variance as future work.", + "type": "text" + }, + { + "bbox": [ + 494, + 408, + 504, + 417 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 421, + 505, + 597 + ], + "lines": [ + { + "bbox": [ + 106, + 420, + 506, + 432 + ], + "spans": [ + { + "bbox": [ + 106, + 420, + 506, + 432 + ], + "score": 1.0, + "content": "Remark 2 (Large Batch Sizes and/or Local Updates). At first glance, it may seem that the requirement", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 430, + 506, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 404, + 444 + ], + "score": 1.0, + "content": "of STEM to compute large mini-batches and/or local updates (cf. Table", + "type": "text" + }, + { + "bbox": [ + 404, + 430, + 415, + 443 + ], + "score": 0.63, + "content": "^ { 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 430, + 506, + 444 + ], + "score": 1.0, + "content": "to achieve this (near)", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 441, + 506, + 454 + ], + "spans": [ + { + "bbox": [ + 105, + 441, + 506, + 454 + ], + "score": 1.0, + "content": "optimal performance is a drawback, however, we note that it is in fact an advantage of STEM that it", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 453, + 505, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 453, + 505, + 465 + ], + "score": 1.0, + "content": "allows the WNs to perform larger number of local updates (or compute large minibatches) without", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 464, + 505, + 476 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 505, + 476 + ], + "score": 1.0, + "content": "communicating often. This follows from the fact that irrespective of the number of local updates", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 474, + 505, + 487 + ], + "spans": [ + { + "bbox": [ + 106, + 474, + 505, + 487 + ], + "score": 1.0, + "content": "(or batch sizes) STEM achieves near-optimal communication complexity while attaining optimal", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 486, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 339, + 498 + ], + "score": 1.0, + "content": "overall sample complexity. Moreover, note that even with", + "type": "text" + }, + { + "bbox": [ + 339, + 486, + 396, + 498 + ], + "score": 0.93, + "content": "b = I = \\mathcal { O } ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 486, + 418, + 498 + ], + "score": 1.0, + "content": "(i.e.,", + "type": "text" + }, + { + "bbox": [ + 418, + 486, + 424, + 496 + ], + "score": 0.71, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 486, + 442, + 498 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 442, + 486, + 448, + 496 + ], + "score": 0.69, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 486, + 505, + 498 + ], + "score": 1.0, + "content": "are chosen as", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 496, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 505, + 509 + ], + "score": 1.0, + "content": "constants), STEM achieves the same (optimal) sample and communication complexities as achieved", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 506, + 505, + 520 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 505, + 520 + ], + "score": 1.0, + "content": "by FedGLOMO [18] and MIME [17]. We further note that to the best of our knowledge the algorithms", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 518, + 506, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 293, + 531 + ], + "score": 1.0, + "content": "that achieve the communication complexity of", + "type": "text" + }, + { + "bbox": [ + 293, + 518, + 325, + 531 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 518, + 506, + 531 + ], + "score": 1.0, + "content": "either require the number of local updates or", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 529, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 529, + 317, + 541 + ], + "score": 1.0, + "content": "the batch-sizes that depend on the solution accuracy", + "type": "text" + }, + { + "bbox": [ + 318, + 532, + 324, + 539 + ], + "score": 0.4, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 529, + 419, + 541 + ], + "score": 1.0, + "content": ". For example, FedProx", + "type": "text" + }, + { + "bbox": [ + 420, + 529, + 437, + 541 + ], + "score": 0.78, + "content": "\\mathbb { I O } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 529, + 471, + 541 + ], + "score": 1.0, + "content": ", FedPD", + "type": "text" + }, + { + "bbox": [ + 472, + 529, + 484, + 541 + ], + "score": 0.67, + "content": "\\bigstar \\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 529, + 505, + 541 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 540, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 141, + 552 + ], + "score": 1.0, + "content": "FedDyn", + "type": "text" + }, + { + "bbox": [ + 142, + 540, + 160, + 551 + ], + "score": 0.63, + "content": "\\pmb { \\Vert 3 6 \\Vert }", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 540, + 365, + 552 + ], + "score": 1.0, + "content": "rely on solving the “local problems\" to achieve an", + "type": "text" + }, + { + "bbox": [ + 365, + 542, + 371, + 550 + ], + "score": 0.68, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 540, + 505, + 552 + ], + "score": 1.0, + "content": "-accuracy, which implies that the", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 550, + 506, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 485, + 564 + ], + "score": 1.0, + "content": "number of local updates (or the batch sizes) implicitly depends on the desired solution accuracy", + "type": "text" + }, + { + "bbox": [ + 485, + 553, + 491, + 561 + ], + "score": 0.42, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 491, + 550, + 506, + 564 + ], + "score": 1.0, + "content": ", as", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 560, + 505, + 575 + ], + "spans": [ + { + "bbox": [ + 104, + 560, + 283, + 575 + ], + "score": 1.0, + "content": "is the case for STEM. Similarly, as shown in", + "type": "text" + }, + { + "bbox": [ + 284, + 561, + 301, + 573 + ], + "score": 0.83, + "content": "\\bar { \\mathbb { E } 2 } \\mathbb { I }", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 560, + 319, + 575 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 319, + 561, + 336, + 573 + ], + "score": 0.81, + "content": "\\bar { \\textregistered 4 } \\bar { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 560, + 505, + 575 + ], + "score": 1.0, + "content": "the communication complexity of FedAvg", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 573, + 506, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 312, + 586 + ], + "score": 1.0, + "content": "and its momentum version can be improved from", + "type": "text" + }, + { + "bbox": [ + 312, + 574, + 343, + 586 + ], + "score": 0.91, + "content": "\\mathcal { O } ( \\epsilon ^ { - 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 573, + 356, + 586 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 356, + 573, + 396, + 586 + ], + "score": 0.92, + "content": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 573, + 506, + 586 + ], + "score": 1.0, + "content": "when the number of local", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 585, + 504, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 249, + 599 + ], + "score": 1.0, + "content": "updates (or batch size) is chosen as", + "type": "text" + }, + { + "bbox": [ + 249, + 585, + 289, + 598 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 1 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 585, + 337, + 599 + ], + "score": 1.0, + "content": "(cf. Section", + "type": "text" + }, + { + "bbox": [ + 338, + 585, + 355, + 599 + ], + "score": 0.33, + "content": "3 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 585, + 504, + 599 + ], + "score": 1.0, + "content": "for a more detailed discussion).", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 106, + 599, + 505, + 723 + ], + "lines": [ + { + "bbox": [ + 106, + 599, + 506, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 453, + 612 + ], + "score": 1.0, + "content": "Remark 3 (The Optimal Batch Sizes and Local Updates Trade-off). The parameter", + "type": "text" + }, + { + "bbox": [ + 453, + 599, + 494, + 611 + ], + "score": 0.91, + "content": "\\nu \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 599, + 506, + 612 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 609, + 506, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 274, + 624 + ], + "score": 1.0, + "content": "used to balance the local minibatch sizes", + "type": "text" + }, + { + "bbox": [ + 274, + 611, + 280, + 620 + ], + "score": 0.72, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 609, + 415, + 624 + ], + "score": 1.0, + "content": ", and the number of local updates", + "type": "text" + }, + { + "bbox": [ + 416, + 611, + 423, + 620 + ], + "score": 0.6, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 609, + 458, + 624 + ], + "score": 1.0, + "content": ". Eqs. in", + "type": "text" + }, + { + "bbox": [ + 458, + 610, + 471, + 622 + ], + "score": 0.85, + "content": "\\textcircled { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 609, + 506, + 624 + ], + "score": 1.0, + "content": "suggest", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 621, + 506, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 149, + 634 + ], + "score": 1.0, + "content": "that when", + "type": "text" + }, + { + "bbox": [ + 149, + 623, + 156, + 631 + ], + "score": 0.75, + "content": "\\nu", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 621, + 249, + 634 + ], + "score": 1.0, + "content": "increases from 0 to 1,", + "type": "text" + }, + { + "bbox": [ + 250, + 622, + 256, + 631 + ], + "score": 0.51, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 621, + 316, + 634 + ], + "score": 1.0, + "content": "decreases and", + "type": "text" + }, + { + "bbox": [ + 316, + 622, + 323, + 631 + ], + "score": 0.72, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 621, + 430, + 634 + ], + "score": 1.0, + "content": "increases. Specifically, if", + "type": "text" + }, + { + "bbox": [ + 430, + 622, + 456, + 632 + ], + "score": 0.9, + "content": "\\nu = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 621, + 481, + 634 + ], + "score": 1.0, + "content": ", then", + "type": "text" + }, + { + "bbox": [ + 481, + 622, + 487, + 631 + ], + "score": 0.7, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 621, + 506, + 634 + ], + "score": 1.0, + "content": "is a", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 104, + 630, + 506, + 648 + ], + "spans": [ + { + "bbox": [ + 104, + 630, + 157, + 648 + ], + "score": 1.0, + "content": "constant but", + "type": "text" + }, + { + "bbox": [ + 157, + 632, + 239, + 645 + ], + "score": 0.93, + "content": "I = \\mathcal { O } ( T ^ { 1 / 3 } / K ^ { 2 / 3 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 630, + 506, + 648 + ], + "score": 1.0, + "content": ". In this case, each WN chooses a small minibatch while executing", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 644, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 106, + 644, + 505, + 656 + ], + "score": 1.0, + "content": "multiple local updates, and STEM resembles a FedAvg (a.k.a. Local SGD) algorithm but with", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 654, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 477, + 667 + ], + "score": 1.0, + "content": "double-sided momentum update directions, and is referred to as Fed STEM. In contrast, if", + "type": "text" + }, + { + "bbox": [ + 477, + 655, + 502, + 665 + ], + "score": 0.89, + "content": "\\nu = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 654, + 506, + 667 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 665, + 506, + 679 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 127, + 679 + ], + "score": 1.0, + "content": "then", + "type": "text" + }, + { + "bbox": [ + 127, + 666, + 196, + 679 + ], + "score": 0.93, + "content": "b = \\mathcal { O } ( T ^ { 1 / 2 } / K )", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 665, + 213, + 679 + ], + "score": 1.0, + "content": "but", + "type": "text" + }, + { + "bbox": [ + 214, + 667, + 220, + 677 + ], + "score": 0.78, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 665, + 506, + 679 + ], + "score": 1.0, + "content": "is a constant. In this case, each WN chooses a large batch size while", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "score": 1.0, + "content": "executing only a few, or even one, local updates, and STEM resembles the Minibatch SGD, but", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 688, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 506, + 701 + ], + "score": 1.0, + "content": "again with different update directions, and is referred to as Minibatch STEM. Such a trade-off can", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "score": 1.0, + "content": "be seen in Fig. 1b. Due to space limitation, these two special cases will be precisely stated in the", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 327, + 723 + ], + "score": 1.0, + "content": "supplementary materials as corollaries of Theorem 3.1.", + "type": "text" + }, + { + "bbox": [ + 494, + 712, + 505, + 721 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 46 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 310, + 752 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 70, + 505, + 106 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 76, + 182, + 89 + ], + "score": 1.0, + "content": "Further, let us set", + "type": "text" + }, + { + "bbox": [ + 183, + 72, + 368, + 96 + ], + "score": 0.95, + "content": "\\begin{array} { r } { c = \\frac { 6 4 L ^ { 2 } } { b K } + \\frac { \\sigma ^ { 2 } } { 2 4 \\bar { \\kappa } ^ { 3 } L I } = L ^ { 2 } \\bigg ( \\frac { 6 4 } { b K } + \\frac { 1 } { 2 4 ( b K ) ^ { 2 } I } \\bigg ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 77, + 506, + 89 + ], + "score": 1.0, + "content": "and set the initial batch size as", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 107, + 93, + 365, + 106 + ], + "spans": [ + { + "bbox": [ + 107, + 95, + 138, + 105 + ], + "score": 0.88, + "content": "B = b I", + "type": "inline_equation" + }, + { + "bbox": [ + 139, + 93, + 227, + 106 + ], + "score": 1.0, + "content": "; set the local updates", + "type": "text" + }, + { + "bbox": [ + 227, + 95, + 234, + 104 + ], + "score": 0.58, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 93, + 365, + 106 + ], + "score": 1.0, + "content": "and minibatch size b as follows:", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5, + "bbox_fs": [ + 105, + 72, + 506, + 106 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 206, + 112, + 404, + 128 + ], + "lines": [ + { + "bbox": [ + 206, + 112, + 404, + 128 + ], + "spans": [ + { + "bbox": [ + 206, + 112, + 404, + 128 + ], + "score": 0.91, + "content": "I = \\mathcal { O } \\big ( ( T / K ^ { 2 } ) ^ { \\nu / 3 } \\big ) , \\quad b = \\mathcal { O } \\big ( ( T / K ^ { 2 } ) ^ { 1 / 2 - \\nu / 2 } \\big )", + "type": "interline_equation", + "image_path": "571cd8b7f6de9fecec247a32a77a5fc7ff44b9b1a1446281dd4865af343e8ec8.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 206, + 112, + 404, + 128 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 133, + 372, + 147 + ], + "lines": [ + { + "bbox": [ + 106, + 133, + 372, + 148 + ], + "spans": [ + { + "bbox": [ + 106, + 133, + 133, + 148 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 137, + 140, + 144 + ], + "score": 0.56, + "content": "\\nu", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 133, + 175, + 148 + ], + "score": 1.0, + "content": "satisfies", + "type": "text" + }, + { + "bbox": [ + 175, + 135, + 214, + 147 + ], + "score": 0.92, + "content": "\\nu \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 133, + 372, + 148 + ], + "score": 1.0, + "content": ". Then for STEM the following holds:", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3, + "bbox_fs": [ + 106, + 133, + 372, + 148 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 155, + 326, + 168 + ], + "lines": [ + { + "bbox": [ + 108, + 154, + 327, + 170 + ], + "spans": [ + { + "bbox": [ + 108, + 154, + 140, + 170 + ], + "score": 1.0, + "content": "(i) For", + "type": "text" + }, + { + "bbox": [ + 140, + 157, + 151, + 167 + ], + "score": 0.85, + "content": "\\scriptstyle { \\bar { x } } _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 154, + 277, + 170 + ], + "score": 1.0, + "content": "chosen according to Algorithm", + "type": "text" + }, + { + "bbox": [ + 277, + 155, + 287, + 169 + ], + "score": 0.5, + "content": "\\boldsymbol { l } ,", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 154, + 327, + 170 + ], + "score": 1.0, + "content": "we have:", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4, + "bbox_fs": [ + 108, + 154, + 327, + 170 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 137, + 173, + 479, + 201 + ], + "lines": [ + { + "bbox": [ + 137, + 173, + 479, + 201 + ], + "spans": [ + { + "bbox": [ + 137, + 173, + 479, + 201 + ], + "score": 0.92, + "content": "\\mathbb { E } \\Vert \\nabla f ( \\bar { x } _ { a } ) \\Vert ^ { 2 } = \\mathcal { O } \\Bigg ( \\frac { f ( \\bar { x } _ { 1 } ) - f ^ { * } } { K ^ { 2 \\nu / 3 } T ^ { 1 - \\nu / 3 } } \\Bigg ) + \\tilde { \\mathcal { O } } \\Bigg ( \\frac { \\sigma ^ { 2 } } { K ^ { 2 \\nu / 3 } T ^ { 1 - \\nu / 3 } } \\Bigg ) + \\tilde { \\mathcal { O } } \\Bigg ( \\frac { \\zeta ^ { 2 } } { K ^ { 2 \\nu / 3 } T ^ { 1 - \\nu / 3 } } \\Bigg ) .", + "type": "interline_equation", + "image_path": "e2c146df205c691e32ce24c2ee693fb97ea38edc895ba31bbeb511db58c5a452.jpg" + } + ] + } + ], + "index": 6, + "virtual_lines": [ + { + "bbox": [ + 137, + 173, + 479, + 182.33333333333334 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 137, + 182.33333333333334, + 479, + 191.66666666666669 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 137, + 191.66666666666669, + 479, + 201.00000000000003 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 212, + 234, + 225 + ], + "lines": [ + { + "bbox": [ + 105, + 212, + 235, + 226 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 156, + 226 + ], + "score": 1.0, + "content": "(ii) For any", + "type": "text" + }, + { + "bbox": [ + 157, + 213, + 195, + 225 + ], + "score": 0.91, + "content": "\\nu \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 212, + 235, + 226 + ], + "score": 1.0, + "content": ", we have", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8, + "bbox_fs": [ + 105, + 212, + 235, + 226 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 225, + 505, + 261 + ], + "lines": [ + { + "bbox": [ + 122, + 223, + 506, + 239 + ], + "spans": [ + { + "bbox": [ + 122, + 223, + 353, + 239 + ], + "score": 1.0, + "content": "Sample Complexity: The sample complexity of STEM is", + "type": "text" + }, + { + "bbox": [ + 353, + 224, + 392, + 238 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 223, + 506, + 239 + ], + "score": 1.0, + "content": ". This implies that each WN", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 122, + 236, + 506, + 252 + ], + "spans": [ + { + "bbox": [ + 122, + 236, + 191, + 252 + ], + "score": 1.0, + "content": "requires at most", + "type": "text" + }, + { + "bbox": [ + 192, + 237, + 250, + 250 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal { O } } ( K ^ { - 1 } \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 251, + 236, + 506, + 252 + ], + "score": 1.0, + "content": "gradient computations, thereby achieving linear speedup with", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 123, + 249, + 295, + 262 + ], + "spans": [ + { + "bbox": [ + 123, + 249, + 295, + 262 + ], + "score": 1.0, + "content": "the number of WNs present in the network.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 122, + 223, + 506, + 262 + ] + }, + { + "type": "text", + "bbox": [ + 124, + 261, + 456, + 273 + ], + "lines": [ + { + "bbox": [ + 122, + 259, + 456, + 275 + ], + "spans": [ + { + "bbox": [ + 122, + 260, + 421, + 275 + ], + "score": 1.0, + "content": "Communication Complexity: The communication complexity of STEM is", + "type": "text" + }, + { + "bbox": [ + 422, + 259, + 453, + 273 + ], + "score": 0.92, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 260, + 456, + 275 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12, + "bbox_fs": [ + 122, + 259, + 456, + 275 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 282, + 481, + 294 + ], + "lines": [ + { + "bbox": [ + 105, + 281, + 482, + 295 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 482, + 295 + ], + "score": 1.0, + "content": "The proof of this result is relegated to the Supplemental Material. A few remarks are in order.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 13, + "bbox_fs": [ + 105, + 281, + 482, + 295 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 295, + 505, + 418 + ], + "lines": [ + { + "bbox": [ + 105, + 294, + 506, + 308 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 432, + 308 + ], + "score": 1.0, + "content": "Remark 1 (Near-Optimal sample and communication complexities). Theorem", + "type": "text" + }, + { + "bbox": [ + 433, + 294, + 450, + 307 + ], + "score": 0.69, + "content": "3 . 1", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 294, + 506, + 308 + ], + "score": 1.0, + "content": "suggests that", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 306, + 506, + 321 + ], + "spans": [ + { + "bbox": [ + 104, + 306, + 131, + 321 + ], + "score": 1.0, + "content": "when", + "type": "text" + }, + { + "bbox": [ + 131, + 308, + 138, + 318 + ], + "score": 0.67, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 306, + 156, + 321 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 157, + 308, + 163, + 318 + ], + "score": 0.71, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 306, + 363, + 321 + ], + "score": 1.0, + "content": "are selected appropriately, then STEM achieves", + "type": "text" + }, + { + "bbox": [ + 363, + 306, + 403, + 320 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 403, + 306, + 423, + 321 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 423, + 307, + 454, + 320 + ], + "score": 0.92, + "content": "\\tilde { \\mathcal { O } } ( \\overline { { \\epsilon ^ { - 1 } } } )", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 306, + 506, + 321 + ], + "score": 1.0, + "content": "sample and", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 318, + 505, + 331 + ], + "spans": [ + { + "bbox": [ + 105, + 318, + 505, + 331 + ], + "score": 1.0, + "content": "communication complexities. Taking them separately, these complexity bounds are the best achievable", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 329, + 505, + 342 + ], + "spans": [ + { + "bbox": [ + 105, + 329, + 505, + 342 + ], + "score": 1.0, + "content": "by the existing FL algorithms (upto logarithmic factors regardless of sample or batch Lipschitz smooth", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 339, + 506, + 354 + ], + "spans": [ + { + "bbox": [ + 105, + 339, + 157, + 354 + ], + "score": 1.0, + "content": "assumption)", + "type": "text" + }, + { + "bbox": [ + 157, + 341, + 175, + 353 + ], + "score": 0.73, + "content": "[ [ 3 5 ] ]", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 339, + 217, + 354 + ], + "score": 1.0, + "content": "; see Table", + "type": "text" + }, + { + "bbox": [ + 217, + 340, + 227, + 354 + ], + "score": 0.32, + "content": "1 .", + "type": "inline_equation" + }, + { + "bbox": [ + 227, + 339, + 295, + 354 + ], + "score": 1.0, + "content": "We note that the", + "type": "text" + }, + { + "bbox": [ + 295, + 340, + 335, + 354 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 339, + 506, + 354 + ], + "score": 1.0, + "content": "complexity is the best possible that can be", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 352, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 352, + 435, + 365 + ], + "score": 1.0, + "content": "achieved by centralized SGD with the sample Lipschitz gradient assumption; see", + "type": "text" + }, + { + "bbox": [ + 435, + 352, + 448, + 364 + ], + "score": 0.75, + "content": " { \\mathbb { I } } ^ { { \\left[ 5 \\right] } }", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 352, + 506, + 365 + ], + "score": 1.0, + "content": ". On the other", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 363, + 506, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 146, + 375 + ], + "score": 1.0, + "content": "hand, the", + "type": "text" + }, + { + "bbox": [ + 146, + 363, + 177, + 375 + ], + "score": 0.93, + "content": "\\bar { \\mathcal { O } } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 178, + 363, + 411, + 375 + ], + "score": 1.0, + "content": "complexity bound is also likely to be the optimal, since in", + "type": "text" + }, + { + "bbox": [ + 412, + 363, + 424, + 374 + ], + "score": 0.67, + "content": "\\mathbb { P }", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 363, + 506, + 375 + ], + "score": 1.0, + "content": "the authors showed", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 374, + 506, + 386 + ], + "spans": [ + { + "bbox": [ + 106, + 374, + 430, + 386 + ], + "score": 1.0, + "content": "that even when the local steps use a class of (deterministic) first-order algorithms,", + "type": "text" + }, + { + "bbox": [ + 430, + 374, + 461, + 386 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 462, + 374, + 506, + 386 + ], + "score": 1.0, + "content": "is the best", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 384, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 105, + 384, + 380, + 398 + ], + "score": 1.0, + "content": "achievable communication complexity. The only difference is that", + "type": "text" + }, + { + "bbox": [ + 380, + 385, + 393, + 396 + ], + "score": 0.66, + "content": "\\bigstar \\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 393, + 384, + 505, + 398 + ], + "score": 1.0, + "content": "does not explicitly assume", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 395, + 506, + 409 + ], + "spans": [ + { + "bbox": [ + 105, + 395, + 389, + 409 + ], + "score": 1.0, + "content": "the inter-node variance bound (i.e., the second relation in Assumption", + "type": "text" + }, + { + "bbox": [ + 389, + 396, + 397, + 409 + ], + "score": 0.72, + "content": "2 \\cdot", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 395, + 506, + 409 + ], + "score": 1.0, + "content": "-(ii)). We leave the precise", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 407, + 504, + 418 + ], + "spans": [ + { + "bbox": [ + 106, + 407, + 478, + 418 + ], + "score": 1.0, + "content": "characterization of the communication lower bound with inter-node variance as future work.", + "type": "text" + }, + { + "bbox": [ + 494, + 408, + 504, + 417 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 19, + "bbox_fs": [ + 104, + 294, + 506, + 418 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 421, + 505, + 597 + ], + "lines": [ + { + "bbox": [ + 106, + 420, + 506, + 432 + ], + "spans": [ + { + "bbox": [ + 106, + 420, + 506, + 432 + ], + "score": 1.0, + "content": "Remark 2 (Large Batch Sizes and/or Local Updates). At first glance, it may seem that the requirement", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 430, + 506, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 404, + 444 + ], + "score": 1.0, + "content": "of STEM to compute large mini-batches and/or local updates (cf. Table", + "type": "text" + }, + { + "bbox": [ + 404, + 430, + 415, + 443 + ], + "score": 0.63, + "content": "^ { 1 ) }", + "type": "inline_equation" + }, + { + "bbox": [ + 416, + 430, + 506, + 444 + ], + "score": 1.0, + "content": "to achieve this (near)", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 441, + 506, + 454 + ], + "spans": [ + { + "bbox": [ + 105, + 441, + 506, + 454 + ], + "score": 1.0, + "content": "optimal performance is a drawback, however, we note that it is in fact an advantage of STEM that it", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 453, + 505, + 465 + ], + "spans": [ + { + "bbox": [ + 106, + 453, + 505, + 465 + ], + "score": 1.0, + "content": "allows the WNs to perform larger number of local updates (or compute large minibatches) without", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 464, + 505, + 476 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 505, + 476 + ], + "score": 1.0, + "content": "communicating often. This follows from the fact that irrespective of the number of local updates", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 474, + 505, + 487 + ], + "spans": [ + { + "bbox": [ + 106, + 474, + 505, + 487 + ], + "score": 1.0, + "content": "(or batch sizes) STEM achieves near-optimal communication complexity while attaining optimal", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 486, + 505, + 498 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 339, + 498 + ], + "score": 1.0, + "content": "overall sample complexity. Moreover, note that even with", + "type": "text" + }, + { + "bbox": [ + 339, + 486, + 396, + 498 + ], + "score": 0.93, + "content": "b = I = \\mathcal { O } ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 397, + 486, + 418, + 498 + ], + "score": 1.0, + "content": "(i.e.,", + "type": "text" + }, + { + "bbox": [ + 418, + 486, + 424, + 496 + ], + "score": 0.71, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 486, + 442, + 498 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 442, + 486, + 448, + 496 + ], + "score": 0.69, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 486, + 505, + 498 + ], + "score": 1.0, + "content": "are chosen as", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 496, + 505, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 505, + 509 + ], + "score": 1.0, + "content": "constants), STEM achieves the same (optimal) sample and communication complexities as achieved", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 506, + 505, + 520 + ], + "spans": [ + { + "bbox": [ + 105, + 506, + 505, + 520 + ], + "score": 1.0, + "content": "by FedGLOMO [18] and MIME [17]. We further note that to the best of our knowledge the algorithms", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 518, + 506, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 518, + 293, + 531 + ], + "score": 1.0, + "content": "that achieve the communication complexity of", + "type": "text" + }, + { + "bbox": [ + 293, + 518, + 325, + 531 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 518, + 506, + 531 + ], + "score": 1.0, + "content": "either require the number of local updates or", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 529, + 505, + 541 + ], + "spans": [ + { + "bbox": [ + 106, + 529, + 317, + 541 + ], + "score": 1.0, + "content": "the batch-sizes that depend on the solution accuracy", + "type": "text" + }, + { + "bbox": [ + 318, + 532, + 324, + 539 + ], + "score": 0.4, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 529, + 419, + 541 + ], + "score": 1.0, + "content": ". For example, FedProx", + "type": "text" + }, + { + "bbox": [ + 420, + 529, + 437, + 541 + ], + "score": 0.78, + "content": "\\mathbb { I O } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 529, + 471, + 541 + ], + "score": 1.0, + "content": ", FedPD", + "type": "text" + }, + { + "bbox": [ + 472, + 529, + 484, + 541 + ], + "score": 0.67, + "content": "\\bigstar \\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 529, + 505, + 541 + ], + "score": 1.0, + "content": ", and", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 540, + 505, + 552 + ], + "spans": [ + { + "bbox": [ + 106, + 540, + 141, + 552 + ], + "score": 1.0, + "content": "FedDyn", + "type": "text" + }, + { + "bbox": [ + 142, + 540, + 160, + 551 + ], + "score": 0.63, + "content": "\\pmb { \\Vert 3 6 \\Vert }", + "type": "inline_equation" + }, + { + "bbox": [ + 160, + 540, + 365, + 552 + ], + "score": 1.0, + "content": "rely on solving the “local problems\" to achieve an", + "type": "text" + }, + { + "bbox": [ + 365, + 542, + 371, + 550 + ], + "score": 0.68, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 540, + 505, + 552 + ], + "score": 1.0, + "content": "-accuracy, which implies that the", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 550, + 506, + 564 + ], + "spans": [ + { + "bbox": [ + 105, + 550, + 485, + 564 + ], + "score": 1.0, + "content": "number of local updates (or the batch sizes) implicitly depends on the desired solution accuracy", + "type": "text" + }, + { + "bbox": [ + 485, + 553, + 491, + 561 + ], + "score": 0.42, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 491, + 550, + 506, + 564 + ], + "score": 1.0, + "content": ", as", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 104, + 560, + 505, + 575 + ], + "spans": [ + { + "bbox": [ + 104, + 560, + 283, + 575 + ], + "score": 1.0, + "content": "is the case for STEM. Similarly, as shown in", + "type": "text" + }, + { + "bbox": [ + 284, + 561, + 301, + 573 + ], + "score": 0.83, + "content": "\\bar { \\mathbb { E } 2 } \\mathbb { I }", + "type": "inline_equation" + }, + { + "bbox": [ + 302, + 560, + 319, + 575 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 319, + 561, + 336, + 573 + ], + "score": 0.81, + "content": "\\bar { \\textregistered 4 } \\bar { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 560, + 505, + 575 + ], + "score": 1.0, + "content": "the communication complexity of FedAvg", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 573, + 506, + 586 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 312, + 586 + ], + "score": 1.0, + "content": "and its momentum version can be improved from", + "type": "text" + }, + { + "bbox": [ + 312, + 574, + 343, + 586 + ], + "score": 0.91, + "content": "\\mathcal { O } ( \\epsilon ^ { - 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 573, + 356, + 586 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 356, + 573, + 396, + 586 + ], + "score": 0.92, + "content": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 573, + 506, + 586 + ], + "score": 1.0, + "content": "when the number of local", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 585, + 504, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 249, + 599 + ], + "score": 1.0, + "content": "updates (or batch size) is chosen as", + "type": "text" + }, + { + "bbox": [ + 249, + 585, + 289, + 598 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 1 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 585, + 337, + 599 + ], + "score": 1.0, + "content": "(cf. Section", + "type": "text" + }, + { + "bbox": [ + 338, + 585, + 355, + 599 + ], + "score": 0.33, + "content": "3 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 585, + 504, + 599 + ], + "score": 1.0, + "content": "for a more detailed discussion).", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 32.5, + "bbox_fs": [ + 104, + 420, + 506, + 599 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 599, + 505, + 723 + ], + "lines": [ + { + "bbox": [ + 106, + 599, + 506, + 612 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 453, + 612 + ], + "score": 1.0, + "content": "Remark 3 (The Optimal Batch Sizes and Local Updates Trade-off). The parameter", + "type": "text" + }, + { + "bbox": [ + 453, + 599, + 494, + 611 + ], + "score": 0.91, + "content": "\\nu \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 599, + 506, + 612 + ], + "score": 1.0, + "content": "is", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 609, + 506, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 609, + 274, + 624 + ], + "score": 1.0, + "content": "used to balance the local minibatch sizes", + "type": "text" + }, + { + "bbox": [ + 274, + 611, + 280, + 620 + ], + "score": 0.72, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 609, + 415, + 624 + ], + "score": 1.0, + "content": ", and the number of local updates", + "type": "text" + }, + { + "bbox": [ + 416, + 611, + 423, + 620 + ], + "score": 0.6, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 609, + 458, + 624 + ], + "score": 1.0, + "content": ". Eqs. in", + "type": "text" + }, + { + "bbox": [ + 458, + 610, + 471, + 622 + ], + "score": 0.85, + "content": "\\textcircled { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 609, + 506, + 624 + ], + "score": 1.0, + "content": "suggest", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 621, + 506, + 634 + ], + "spans": [ + { + "bbox": [ + 105, + 621, + 149, + 634 + ], + "score": 1.0, + "content": "that when", + "type": "text" + }, + { + "bbox": [ + 149, + 623, + 156, + 631 + ], + "score": 0.75, + "content": "\\nu", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 621, + 249, + 634 + ], + "score": 1.0, + "content": "increases from 0 to 1,", + "type": "text" + }, + { + "bbox": [ + 250, + 622, + 256, + 631 + ], + "score": 0.51, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 256, + 621, + 316, + 634 + ], + "score": 1.0, + "content": "decreases and", + "type": "text" + }, + { + "bbox": [ + 316, + 622, + 323, + 631 + ], + "score": 0.72, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 621, + 430, + 634 + ], + "score": 1.0, + "content": "increases. Specifically, if", + "type": "text" + }, + { + "bbox": [ + 430, + 622, + 456, + 632 + ], + "score": 0.9, + "content": "\\nu = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 621, + 481, + 634 + ], + "score": 1.0, + "content": ", then", + "type": "text" + }, + { + "bbox": [ + 481, + 622, + 487, + 631 + ], + "score": 0.7, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 621, + 506, + 634 + ], + "score": 1.0, + "content": "is a", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 104, + 630, + 506, + 648 + ], + "spans": [ + { + "bbox": [ + 104, + 630, + 157, + 648 + ], + "score": 1.0, + "content": "constant but", + "type": "text" + }, + { + "bbox": [ + 157, + 632, + 239, + 645 + ], + "score": 0.93, + "content": "I = \\mathcal { O } ( T ^ { 1 / 3 } / K ^ { 2 / 3 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 239, + 630, + 506, + 648 + ], + "score": 1.0, + "content": ". In this case, each WN chooses a small minibatch while executing", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 644, + 505, + 656 + ], + "spans": [ + { + "bbox": [ + 106, + 644, + 505, + 656 + ], + "score": 1.0, + "content": "multiple local updates, and STEM resembles a FedAvg (a.k.a. Local SGD) algorithm but with", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 654, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 477, + 667 + ], + "score": 1.0, + "content": "double-sided momentum update directions, and is referred to as Fed STEM. In contrast, if", + "type": "text" + }, + { + "bbox": [ + 477, + 655, + 502, + 665 + ], + "score": 0.89, + "content": "\\nu = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 654, + 506, + 667 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 665, + 506, + 679 + ], + "spans": [ + { + "bbox": [ + 105, + 665, + 127, + 679 + ], + "score": 1.0, + "content": "then", + "type": "text" + }, + { + "bbox": [ + 127, + 666, + 196, + 679 + ], + "score": 0.93, + "content": "b = \\mathcal { O } ( T ^ { 1 / 2 } / K )", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 665, + 213, + 679 + ], + "score": 1.0, + "content": "but", + "type": "text" + }, + { + "bbox": [ + 214, + 667, + 220, + 677 + ], + "score": 0.78, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 665, + 506, + 679 + ], + "score": 1.0, + "content": "is a constant. In this case, each WN chooses a large batch size while", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 690 + ], + "score": 1.0, + "content": "executing only a few, or even one, local updates, and STEM resembles the Minibatch SGD, but", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 688, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 506, + 701 + ], + "score": 1.0, + "content": "again with different update directions, and is referred to as Minibatch STEM. Such a trade-off can", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 505, + 712 + ], + "score": 1.0, + "content": "be seen in Fig. 1b. Due to space limitation, these two special cases will be precisely stated in the", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 106, + 711, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 327, + 723 + ], + "score": 1.0, + "content": "supplementary materials as corollaries of Theorem 3.1.", + "type": "text" + }, + { + "bbox": [ + 494, + 712, + 505, + 721 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 51 + } + ], + "index": 46, + "bbox_fs": [ + 104, + 599, + 506, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 72, + 255, + 84 + ], + "lines": [ + { + "bbox": [ + 105, + 70, + 256, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 70, + 256, + 86 + ], + "score": 1.0, + "content": "Algorithm 2 The FedAvg Algorithm", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 109, + 88, + 487, + 219 + ], + "lines": [ + { + "bbox": [ + 110, + 87, + 488, + 102 + ], + "spans": [ + { + "bbox": [ + 110, + 87, + 153, + 102 + ], + "score": 1.0, + "content": "1: Input:", + "type": "text" + }, + { + "bbox": [ + 154, + 88, + 197, + 101 + ], + "score": 0.9, + "content": "\\{ \\eta _ { t } \\} _ { t = 0 } ^ { T } ; I", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 87, + 396, + 102 + ], + "score": 1.0, + "content": ", the # of local updates per communication round;", + "type": "text" + }, + { + "bbox": [ + 397, + 90, + 402, + 100 + ], + "score": 0.68, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 87, + 488, + 102 + ], + "score": 1.0, + "content": ", the minibatch sizes.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 110, + 99, + 197, + 112 + ], + "spans": [ + { + "bbox": [ + 110, + 99, + 138, + 112 + ], + "score": 1.0, + "content": "2: for", + "type": "text" + }, + { + "bbox": [ + 138, + 101, + 162, + 110 + ], + "score": 0.88, + "content": "t = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 99, + 173, + 112 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 173, + 101, + 182, + 110 + ], + "score": 0.73, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 99, + 197, + 112 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 110, + 110, + 374, + 183 + ], + "spans": [ + { + "bbox": [ + 110, + 111, + 122, + 135 + ], + "score": 1.0, + "content": "3: 4:", + "type": "text" + }, + { + "bbox": [ + 110, + 138, + 122, + 152 + ], + "score": 1.0, + "content": "5:", + "type": "text" + }, + { + "bbox": [ + 137, + 110, + 153, + 144 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 153, + 115, + 374, + 183 + ], + "score": 0.39, + "content": "\\begin{array} { r l } & { \\mathcal { \\kappa } _ { t } ^ { = } \\stackrel { \\mathrm { ~ L ~ U ~ O ~ } \\Lambda } { = } \\mathbf { 0 } } \\\\ & { d _ { t } ^ { ( k ) } = \\frac { 1 } { b } \\sum _ { \\xi _ { t } ^ { ( k ) } \\in \\mathcal { B } _ { t } ^ { ( k ) } } \\nabla f ^ { ( k ) } ( x _ { t } ^ { ( k ) } ; \\xi _ { t } ^ { ( k ) } ) \\mathrm { ~ w i t h ~ } | \\mathcal { B } _ { t } ^ { ( k ) } | = b } \\\\ & { x _ { t + 1 } ^ { ( k ) } = x _ { t } ^ { ( k ) } - \\eta _ { t } d _ { t } ^ { ( k ) } } \\\\ & { \\mathbf { i f } t \\operatorname* { m o d } I = 0 \\mathbf { \\Lambda } \\mathbf { t h e n } } \\\\ & { ~ x _ { t + 1 } ^ { ( k ) } = \\bar { x } _ { t + 1 } = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } x _ { t + 1 } ^ { ( k ) } } \\\\ & { \\mathbf { e n d } \\mathbf { \\Phi } \\mathbf { i f } } \\end{array}", + "type": "inline_equation", + "image_path": "b9c324b0196546714a22fb179bfa2a8770feda5532e5c2738427835774c820ac.jpg" + }, + { + "bbox": [ + 154, + 111, + 178, + 121 + ], + "score": 0.84, + "content": "k = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 111, + 200, + 121 + ], + "score": 0.73, + "content": "K", + "type": "inline_equation" + } + ], + "index": 6 + }, + { + "bbox": [ + 109, + 150, + 122, + 162 + ], + "spans": [ + { + "bbox": [ + 109, + 150, + 122, + 162 + ], + "score": 1.0, + "content": "6:", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 110, + 160, + 122, + 174 + ], + "spans": [ + { + "bbox": [ + 110, + 160, + 122, + 174 + ], + "score": 1.0, + "content": "7:", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 110, + 172, + 122, + 183 + ], + "spans": [ + { + "bbox": [ + 110, + 172, + 122, + 183 + ], + "score": 1.0, + "content": "8:", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 109, + 181, + 173, + 195 + ], + "spans": [ + { + "bbox": [ + 109, + 182, + 122, + 195 + ], + "score": 1.0, + "content": "9:", + "type": "text" + }, + { + "bbox": [ + 136, + 181, + 173, + 195 + ], + "score": 1.0, + "content": "end for", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 193, + 157, + 205 + ], + "spans": [ + { + "bbox": [ + 106, + 193, + 157, + 205 + ], + "score": 1.0, + "content": "10: end for", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 204, + 270, + 218 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 160, + 218 + ], + "score": 1.0, + "content": "11: Return:", + "type": "text" + }, + { + "bbox": [ + 160, + 206, + 172, + 216 + ], + "score": 0.85, + "content": "\\scriptstyle { \\bar { x } } _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 204, + 200, + 218 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 200, + 205, + 266, + 217 + ], + "score": 0.91, + "content": "a \\sim \\mathcal { U } \\{ 1 , . . . , T \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 204, + 270, + 218 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 106, + 243, + 506, + 367 + ], + "lines": [ + { + "bbox": [ + 105, + 242, + 506, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 506, + 257 + ], + "score": 1.0, + "content": "Remark 4 (The Sub-Optimal Batch Sizes and Local Updates Trade-off). From our proof (Theo-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 254, + 505, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 124, + 270 + ], + "score": 1.0, + "content": "rem", + "type": "text" + }, + { + "bbox": [ + 124, + 254, + 148, + 268 + ], + "score": 0.41, + "content": "\\left| \\overline { { \\mathbf { C . 1 0 } } } \\right|", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 254, + 448, + 270 + ], + "score": 1.0, + "content": "included in the supplemental material), we can see that STEM requires", + "type": "text" + }, + { + "bbox": [ + 449, + 254, + 505, + 269 + ], + "score": 0.91, + "content": "\\bar { \\tilde { O } } ( \\operatorname* { m a x } \\big \\{ ( b \\cdot", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 267, + 507, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 268, + 188, + 283 + ], + "score": 0.88, + "content": "I ) \\epsilon ^ { - 1 } , K ^ { - 1 } \\epsilon ^ { - 3 / 2 } \\rbrace )", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 267, + 243, + 284 + ], + "score": 1.0, + "content": "samples and", + "type": "text" + }, + { + "bbox": [ + 243, + 268, + 389, + 283 + ], + "score": 0.91, + "content": "\\tilde { \\mathcal { O } } \\big ( \\operatorname* { m a x } \\big \\{ \\epsilon ^ { - 1 } , ( b \\cdot I ) ^ { - 1 } K ^ { - 1 } \\epsilon ^ { - 3 / 2 } \\big \\} \\big )", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 267, + 507, + 284 + ], + "score": 1.0, + "content": "\u0000 and communication rounds.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 282, + 507, + 296 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 263, + 296 + ], + "score": 1.0, + "content": "According to the above expressions, if", + "type": "text" + }, + { + "bbox": [ + 263, + 284, + 281, + 293 + ], + "score": 0.88, + "content": "b \\cdot I", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 282, + 354, + 296 + ], + "score": 1.0, + "content": "increases beyond", + "type": "text" + }, + { + "bbox": [ + 354, + 282, + 413, + 295 + ], + "score": 0.92, + "content": "\\mathcal { O } ( K ^ { - 1 } \\epsilon ^ { - 1 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 282, + 507, + 296 + ], + "score": 1.0, + "content": ", then the sample com-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 294, + 505, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 255, + 309 + ], + "score": 1.0, + "content": "plexity will increase from the optimal", + "type": "text" + }, + { + "bbox": [ + 256, + 294, + 295, + 308 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 294, + 465, + 309 + ], + "score": 1.0, + "content": "; otherwise, the optimal sample complexity", + "type": "text" + }, + { + "bbox": [ + 465, + 294, + 505, + 308 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 306, + 506, + 322 + ], + "spans": [ + { + "bbox": [ + 104, + 306, + 261, + 322 + ], + "score": 1.0, + "content": "is maintained. On the other hand, if", + "type": "text" + }, + { + "bbox": [ + 262, + 309, + 281, + 318 + ], + "score": 0.86, + "content": "b \\cdot I", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 306, + 359, + 322 + ], + "score": 1.0, + "content": "decreases beyond", + "type": "text" + }, + { + "bbox": [ + 359, + 307, + 418, + 320 + ], + "score": 0.95, + "content": "\\mathcal { O } ( K ^ { - 1 } \\epsilon ^ { - 1 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 306, + 506, + 322 + ], + "score": 1.0, + "content": ", the communication", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 319, + 506, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 212, + 334 + ], + "score": 1.0, + "content": "complexity increases from", + "type": "text" + }, + { + "bbox": [ + 212, + 319, + 243, + 332 + ], + "score": 0.92, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 320, + 351, + 334 + ], + "score": 1.0, + "content": ". For instance, if we choose", + "type": "text" + }, + { + "bbox": [ + 351, + 321, + 391, + 333 + ], + "score": 0.92, + "content": "b = \\mathcal { O } ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 320, + 408, + 334 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 408, + 321, + 448, + 332 + ], + "score": 0.9, + "content": "I = { \\mathcal { O } } ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 320, + 506, + 334 + ], + "score": 1.0, + "content": "the communi-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 330, + 507, + 347 + ], + "spans": [ + { + "bbox": [ + 104, + 330, + 217, + 347 + ], + "score": 1.0, + "content": "cation complexity becomes", + "type": "text" + }, + { + "bbox": [ + 218, + 333, + 257, + 345 + ], + "score": 0.91, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 330, + 407, + 347 + ], + "score": 1.0, + "content": "while the optimal sample complexity", + "type": "text" + }, + { + "bbox": [ + 408, + 332, + 447, + 345 + ], + "score": 0.91, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 330, + 507, + 347 + ], + "score": 1.0, + "content": "is maintained.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 343, + 506, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 343, + 255, + 357 + ], + "score": 1.0, + "content": "This trade-off is illustrated in Figure", + "type": "text" + }, + { + "bbox": [ + 255, + 344, + 270, + 357 + ], + "score": 0.82, + "content": "\\boxed { 1 \\mathrm { a } }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 343, + 506, + 357 + ], + "score": 1.0, + "content": "where we maintain the optimal sample complexity, while", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 355, + 505, + 368 + ], + "spans": [ + { + "bbox": [ + 106, + 355, + 145, + 368 + ], + "score": 1.0, + "content": "changing", + "type": "text" + }, + { + "bbox": [ + 146, + 356, + 151, + 365 + ], + "score": 0.65, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 355, + 169, + 368 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 169, + 356, + 176, + 365 + ], + "score": 0.78, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 355, + 309, + 368 + ], + "score": 1.0, + "content": "to generate the trade-off surface.", + "type": "text" + }, + { + "bbox": [ + 494, + 356, + 505, + 367 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 106, + 369, + 506, + 464 + ], + "lines": [ + { + "bbox": [ + 105, + 368, + 506, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 368, + 281, + 386 + ], + "score": 1.0, + "content": "Remark 5 (Data Heterogeneity). The term", + "type": "text" + }, + { + "bbox": [ + 282, + 369, + 352, + 388 + ], + "score": 0.94, + "content": "\\begin{array} { r } { \\tilde { \\mathcal { O } } \\biggl ( \\frac { \\zeta ^ { 2 } } { K ^ { 2 \\nu / 3 } T ^ { 1 - \\nu / 3 } } \\biggr ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 371, + 441, + 387 + ], + "score": 1.0, + "content": "in the gradient bound", + "type": "text" + }, + { + "bbox": [ + 441, + 372, + 454, + 385 + ], + "score": 0.82, + "content": "\\textcircled{4}", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 371, + 506, + 387 + ], + "score": 1.0, + "content": "captures the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 387, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 106, + 387, + 318, + 399 + ], + "score": 1.0, + "content": "effect of the heterogeneity of data across WNs, where", + "type": "text" + }, + { + "bbox": [ + 318, + 388, + 325, + 398 + ], + "score": 0.8, + "content": "\\zeta", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 387, + 505, + 399 + ], + "score": 1.0, + "content": "is the parameter characterizing the intra-node", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 397, + 506, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 292, + 410 + ], + "score": 1.0, + "content": "variance and has been defined in Assumption", + "type": "text" + }, + { + "bbox": [ + 292, + 397, + 302, + 410 + ], + "score": 0.8, + "content": "\\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 398, + 476, + 410 + ], + "score": 1.0, + "content": "(ii). Highly heterogeneous data with large", + "type": "text" + }, + { + "bbox": [ + 477, + 398, + 487, + 410 + ], + "score": 0.87, + "content": "\\zeta ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 398, + 506, + 410 + ], + "score": 1.0, + "content": "can", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 408, + 506, + 421 + ], + "spans": [ + { + "bbox": [ + 106, + 408, + 414, + 421 + ], + "score": 1.0, + "content": "adversely impact the performance of STEM. Note that such a dependency on", + "type": "text" + }, + { + "bbox": [ + 414, + 409, + 420, + 420 + ], + "score": 0.83, + "content": "\\zeta", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 408, + 506, + 421 + ], + "score": 1.0, + "content": "also appears in other", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 419, + 506, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 419, + 235, + 431 + ], + "score": 1.0, + "content": "existing FL algorithms, such as", + "type": "text" + }, + { + "bbox": [ + 235, + 419, + 278, + 430 + ], + "score": 0.59, + "content": "[ \\bigcirc , \\bigcirc , \\bigcirc , \\bigcirc , \\bigcirc ]", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 419, + 506, + 431 + ], + "score": 1.0, + "content": ". However, there is one special case of STEM that does", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 429, + 225, + 442 + ], + "score": 1.0, + "content": "not depend on the parameter", + "type": "text" + }, + { + "bbox": [ + 225, + 431, + 231, + 442 + ], + "score": 0.82, + "content": "\\zeta", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 429, + 328, + 442 + ], + "score": 1.0, + "content": ". This is the case where", + "type": "text" + }, + { + "bbox": [ + 329, + 431, + 353, + 441 + ], + "score": 0.89, + "content": "I = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 429, + 505, + 442 + ], + "score": 1.0, + "content": ", i.e., the minibatch SGD counterpart", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 441, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 441, + 506, + 453 + ], + "score": 1.0, + "content": "of STEM where only a single local iteration is performed between two communication rounds. We", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 451, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 223, + 465 + ], + "score": 1.0, + "content": "have the following corollary.", + "type": "text" + }, + { + "bbox": [ + 494, + 452, + 506, + 464 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 107, + 467, + 506, + 504 + ], + "lines": [ + { + "bbox": [ + 106, + 467, + 506, + 481 + ], + "spans": [ + { + "bbox": [ + 106, + 467, + 319, + 481 + ], + "score": 1.0, + "content": "Corollary 1 (Minibatch STEM). Under Assumptions", + "type": "text" + }, + { + "bbox": [ + 319, + 467, + 352, + 480 + ], + "score": 0.7, + "content": "\\bigstar \\bigstar \\bigstar | \\bigstar |", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 467, + 506, + 481 + ], + "score": 1.0, + "content": ", and choose the algorithm parameters", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 478, + 508, + 494 + ], + "spans": [ + { + "bbox": [ + 104, + 478, + 164, + 494 + ], + "score": 1.0, + "content": "as in Theorem", + "type": "text" + }, + { + "bbox": [ + 164, + 478, + 182, + 493 + ], + "score": 0.74, + "content": "3 . I .", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 478, + 266, + 494 + ], + "score": 1.0, + "content": "At each WN, choose", + "type": "text" + }, + { + "bbox": [ + 267, + 481, + 291, + 491 + ], + "score": 0.86, + "content": "I = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 478, + 295, + 494 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 295, + 480, + 359, + 493 + ], + "score": 0.9, + "content": "b = ( T / K ^ { 2 } ) ^ { 1 / 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 478, + 464, + 494 + ], + "score": 1.0, + "content": ", and the initial batch size", + "type": "text" + }, + { + "bbox": [ + 464, + 481, + 503, + 491 + ], + "score": 0.88, + "content": "B = \\boldsymbol { b } \\cdot \\boldsymbol { I }", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 478, + 508, + 494 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 490, + 196, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 196, + 504 + ], + "score": 1.0, + "content": "Then STEM satisfies:", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 108, + 512, + 324, + 525 + ], + "lines": [ + { + "bbox": [ + 109, + 512, + 324, + 527 + ], + "spans": [ + { + "bbox": [ + 109, + 512, + 140, + 527 + ], + "score": 1.0, + "content": "(i) For", + "type": "text" + }, + { + "bbox": [ + 140, + 514, + 151, + 524 + ], + "score": 0.86, + "content": "\\bar { x } _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 512, + 277, + 527 + ], + "score": 1.0, + "content": "chosen according to Algorithm", + "type": "text" + }, + { + "bbox": [ + 277, + 513, + 287, + 526 + ], + "score": 0.69, + "content": "\\boldsymbol { l } ,", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 512, + 324, + 527 + ], + "score": 1.0, + "content": "we have", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31 + }, + { + "type": "interline_equation", + "bbox": [ + 220, + 532, + 407, + 558 + ], + "lines": [ + { + "bbox": [ + 220, + 532, + 407, + 558 + ], + "spans": [ + { + "bbox": [ + 220, + 532, + 407, + 558 + ], + "score": 0.92, + "content": "\\mathbb { E } \\| \\nabla f ( \\bar { x } _ { a } ) \\| ^ { 2 } = \\mathcal { O } \\Big ( \\frac { f ( \\bar { x } _ { 1 } ) - f ^ { * } } { T } \\Big ) + \\tilde { \\mathcal { O } } \\Big ( \\frac { \\sigma ^ { 2 } } { T } \\Big ) .", + "type": "interline_equation", + "image_path": "fe27c5ca3c886c333d5add660b12f630bb6023b83b7450c21e97035dfce670f9.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 220, + 532, + 407, + 558 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 103, + 570, + 466, + 585 + ], + "lines": [ + { + "bbox": [ + 104, + 568, + 468, + 587 + ], + "spans": [ + { + "bbox": [ + 104, + 568, + 234, + 587 + ], + "score": 1.0, + "content": "(ii) Minibatch STEM achieves", + "type": "text" + }, + { + "bbox": [ + 234, + 570, + 273, + 584 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 568, + 323, + 587 + ], + "score": 1.0, + "content": "sample and", + "type": "text" + }, + { + "bbox": [ + 323, + 570, + 354, + 584 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 568, + 468, + 587 + ], + "score": 1.0, + "content": "communication complexity.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 106, + 593, + 506, + 616 + ], + "lines": [ + { + "bbox": [ + 105, + 592, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 592, + 505, + 607 + ], + "score": 1.0, + "content": "Next, we show that FedAvg also exhibits a trade-off similar to that of STEM but with worse sample", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 605, + 243, + 616 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 243, + 616 + ], + "score": 1.0, + "content": "and communication complexities.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5 + }, + { + "type": "title", + "bbox": [ + 107, + 631, + 288, + 644 + ], + "lines": [ + { + "bbox": [ + 105, + 629, + 288, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 629, + 288, + 645 + ], + "score": 1.0, + "content": "3.2 Special cases: The FedAvg algorithm", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 107, + 652, + 505, + 721 + ], + "lines": [ + { + "bbox": [ + 105, + 652, + 506, + 665 + ], + "spans": [ + { + "bbox": [ + 105, + 652, + 506, + 665 + ], + "score": 1.0, + "content": "We briefly discuss another interesting special case of STEM, where the local momentum update", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 663, + 506, + 676 + ], + "spans": [ + { + "bbox": [ + 105, + 663, + 285, + 676 + ], + "score": 1.0, + "content": "is replaced by the conventional SGD (i.e.,", + "type": "text" + }, + { + "bbox": [ + 285, + 664, + 341, + 675 + ], + "score": 0.87, + "content": "a _ { t } = 1 , ~ \\forall ~ t )", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 663, + 506, + 676 + ], + "score": 1.0, + "content": ", while the server does not perform the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 673, + 506, + 687 + ], + "spans": [ + { + "bbox": [ + 105, + 674, + 205, + 687 + ], + "score": 1.0, + "content": "momentum update (i.e.,", + "type": "text" + }, + { + "bbox": [ + 205, + 673, + 252, + 685 + ], + "score": 0.91, + "content": "\\bar { d } _ { t } = 0 , \\forall t )", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 674, + 506, + 687 + ], + "score": 1.0, + "content": ". This is essentially the classical FedAvg algorithm, just that it", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 684, + 505, + 698 + ], + "spans": [ + { + "bbox": [ + 105, + 684, + 259, + 698 + ], + "score": 1.0, + "content": "balances the number of local updates", + "type": "text" + }, + { + "bbox": [ + 259, + 686, + 266, + 695 + ], + "score": 0.78, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 684, + 360, + 698 + ], + "score": 1.0, + "content": "and the minibatch size", + "type": "text" + }, + { + "bbox": [ + 360, + 686, + 366, + 695 + ], + "score": 0.53, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 684, + 505, + 698 + ], + "score": 1.0, + "content": ". We show that this algorithm also", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 694, + 505, + 710 + ], + "spans": [ + { + "bbox": [ + 105, + 694, + 217, + 710 + ], + "score": 1.0, + "content": "exhibits a trade-off between", + "type": "text" + }, + { + "bbox": [ + 218, + 697, + 224, + 706 + ], + "score": 0.71, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 694, + 241, + 710 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 241, + 697, + 248, + 706 + ], + "score": 0.77, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 694, + 396, + 710 + ], + "score": 1.0, + "content": "and on the trade-off curve it achieves", + "type": "text" + }, + { + "bbox": [ + 396, + 696, + 428, + 708 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 694, + 505, + 710 + ], + "score": 1.0, + "content": "sample complexity", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 104, + 705, + 279, + 723 + ], + "spans": [ + { + "bbox": [ + 104, + 705, + 123, + 723 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 707, + 163, + 721 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 705, + 279, + 723 + ], + "score": 1.0, + "content": "communication complexity.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 39.5 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 107, + 72, + 255, + 84 + ], + "lines": [ + { + "bbox": [ + 105, + 70, + 256, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 70, + 256, + 86 + ], + "score": 1.0, + "content": "Algorithm 2 The FedAvg Algorithm", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "index", + "bbox": [ + 109, + 88, + 487, + 219 + ], + "lines": [ + { + "bbox": [ + 110, + 87, + 488, + 102 + ], + "spans": [ + { + "bbox": [ + 110, + 87, + 153, + 102 + ], + "score": 1.0, + "content": "1: Input:", + "type": "text" + }, + { + "bbox": [ + 154, + 88, + 197, + 101 + ], + "score": 0.9, + "content": "\\{ \\eta _ { t } \\} _ { t = 0 } ^ { T } ; I", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 87, + 396, + 102 + ], + "score": 1.0, + "content": ", the # of local updates per communication round;", + "type": "text" + }, + { + "bbox": [ + 397, + 90, + 402, + 100 + ], + "score": 0.68, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 87, + 488, + 102 + ], + "score": 1.0, + "content": ", the minibatch sizes.", + "type": "text" + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 99, + 197, + 112 + ], + "spans": [ + { + "bbox": [ + 110, + 99, + 138, + 112 + ], + "score": 1.0, + "content": "2: for", + "type": "text" + }, + { + "bbox": [ + 138, + 101, + 162, + 110 + ], + "score": 0.88, + "content": "t = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 99, + 173, + 112 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 173, + 101, + 182, + 110 + ], + "score": 0.73, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 99, + 197, + 112 + ], + "score": 1.0, + "content": "do", + "type": "text" + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 110, + 374, + 183 + ], + "spans": [ + { + "bbox": [ + 110, + 111, + 122, + 135 + ], + "score": 1.0, + "content": "3: 4:", + "type": "text" + }, + { + "bbox": [ + 110, + 138, + 122, + 152 + ], + "score": 1.0, + "content": "5:", + "type": "text" + }, + { + "bbox": [ + 137, + 110, + 153, + 144 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 153, + 115, + 374, + 183 + ], + "score": 0.39, + "content": "\\begin{array} { r l } & { \\mathcal { \\kappa } _ { t } ^ { = } \\stackrel { \\mathrm { ~ L ~ U ~ O ~ } \\Lambda } { = } \\mathbf { 0 } } \\\\ & { d _ { t } ^ { ( k ) } = \\frac { 1 } { b } \\sum _ { \\xi _ { t } ^ { ( k ) } \\in \\mathcal { B } _ { t } ^ { ( k ) } } \\nabla f ^ { ( k ) } ( x _ { t } ^ { ( k ) } ; \\xi _ { t } ^ { ( k ) } ) \\mathrm { ~ w i t h ~ } | \\mathcal { B } _ { t } ^ { ( k ) } | = b } \\\\ & { x _ { t + 1 } ^ { ( k ) } = x _ { t } ^ { ( k ) } - \\eta _ { t } d _ { t } ^ { ( k ) } } \\\\ & { \\mathbf { i f } t \\operatorname* { m o d } I = 0 \\mathbf { \\Lambda } \\mathbf { t h e n } } \\\\ & { ~ x _ { t + 1 } ^ { ( k ) } = \\bar { x } _ { t + 1 } = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } x _ { t + 1 } ^ { ( k ) } } \\\\ & { \\mathbf { e n d } \\mathbf { \\Phi } \\mathbf { i f } } \\end{array}", + "type": "inline_equation", + "image_path": "b9c324b0196546714a22fb179bfa2a8770feda5532e5c2738427835774c820ac.jpg" + }, + { + "bbox": [ + 154, + 111, + 178, + 121 + ], + "score": 0.84, + "content": "k = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 111, + 200, + 121 + ], + "score": 0.73, + "content": "K", + "type": "inline_equation" + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 109, + 150, + 122, + 162 + ], + "spans": [ + { + "bbox": [ + 109, + 150, + 122, + 162 + ], + "score": 1.0, + "content": "6:", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 160, + 122, + 174 + ], + "spans": [ + { + "bbox": [ + 110, + 160, + 122, + 174 + ], + "score": 1.0, + "content": "7:", + "type": "text" + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 110, + 172, + 122, + 183 + ], + "spans": [ + { + "bbox": [ + 110, + 172, + 122, + 183 + ], + "score": 1.0, + "content": "8:", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 109, + 181, + 173, + 195 + ], + "spans": [ + { + "bbox": [ + 109, + 182, + 122, + 195 + ], + "score": 1.0, + "content": "9:", + "type": "text" + }, + { + "bbox": [ + 136, + 181, + 173, + 195 + ], + "score": 1.0, + "content": "end for", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 106, + 193, + 157, + 205 + ], + "spans": [ + { + "bbox": [ + 106, + 193, + 157, + 205 + ], + "score": 1.0, + "content": "10: end for", + "type": "text" + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 106, + 204, + 270, + 218 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 160, + 218 + ], + "score": 1.0, + "content": "11: Return:", + "type": "text" + }, + { + "bbox": [ + 160, + 206, + 172, + 216 + ], + "score": 0.85, + "content": "\\scriptstyle { \\bar { x } } _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 204, + 200, + 218 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 200, + 205, + 266, + 217 + ], + "score": 0.91, + "content": "a \\sim \\mathcal { U } \\{ 1 , . . . , T \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 204, + 270, + 218 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 9, + "is_list_start_line": true + } + ], + "index": 5, + "bbox_fs": [ + 106, + 87, + 488, + 218 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 243, + 506, + 367 + ], + "lines": [ + { + "bbox": [ + 105, + 242, + 506, + 257 + ], + "spans": [ + { + "bbox": [ + 105, + 242, + 506, + 257 + ], + "score": 1.0, + "content": "Remark 4 (The Sub-Optimal Batch Sizes and Local Updates Trade-off). From our proof (Theo-", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 254, + 505, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 254, + 124, + 270 + ], + "score": 1.0, + "content": "rem", + "type": "text" + }, + { + "bbox": [ + 124, + 254, + 148, + 268 + ], + "score": 0.41, + "content": "\\left| \\overline { { \\mathbf { C . 1 0 } } } \\right|", + "type": "inline_equation" + }, + { + "bbox": [ + 148, + 254, + 448, + 270 + ], + "score": 1.0, + "content": "included in the supplemental material), we can see that STEM requires", + "type": "text" + }, + { + "bbox": [ + 449, + 254, + 505, + 269 + ], + "score": 0.91, + "content": "\\bar { \\tilde { O } } ( \\operatorname* { m a x } \\big \\{ ( b \\cdot", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 267, + 507, + 284 + ], + "spans": [ + { + "bbox": [ + 106, + 268, + 188, + 283 + ], + "score": 0.88, + "content": "I ) \\epsilon ^ { - 1 } , K ^ { - 1 } \\epsilon ^ { - 3 / 2 } \\rbrace )", + "type": "inline_equation" + }, + { + "bbox": [ + 188, + 267, + 243, + 284 + ], + "score": 1.0, + "content": "samples and", + "type": "text" + }, + { + "bbox": [ + 243, + 268, + 389, + 283 + ], + "score": 0.91, + "content": "\\tilde { \\mathcal { O } } \\big ( \\operatorname* { m a x } \\big \\{ \\epsilon ^ { - 1 } , ( b \\cdot I ) ^ { - 1 } K ^ { - 1 } \\epsilon ^ { - 3 / 2 } \\big \\} \\big )", + "type": "inline_equation" + }, + { + "bbox": [ + 389, + 267, + 507, + 284 + ], + "score": 1.0, + "content": "\u0000 and communication rounds.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 282, + 507, + 296 + ], + "spans": [ + { + "bbox": [ + 105, + 282, + 263, + 296 + ], + "score": 1.0, + "content": "According to the above expressions, if", + "type": "text" + }, + { + "bbox": [ + 263, + 284, + 281, + 293 + ], + "score": 0.88, + "content": "b \\cdot I", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 282, + 354, + 296 + ], + "score": 1.0, + "content": "increases beyond", + "type": "text" + }, + { + "bbox": [ + 354, + 282, + 413, + 295 + ], + "score": 0.92, + "content": "\\mathcal { O } ( K ^ { - 1 } \\epsilon ^ { - 1 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 414, + 282, + 507, + 296 + ], + "score": 1.0, + "content": ", then the sample com-", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 294, + 505, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 294, + 255, + 309 + ], + "score": 1.0, + "content": "plexity will increase from the optimal", + "type": "text" + }, + { + "bbox": [ + 256, + 294, + 295, + 308 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 294, + 465, + 309 + ], + "score": 1.0, + "content": "; otherwise, the optimal sample complexity", + "type": "text" + }, + { + "bbox": [ + 465, + 294, + 505, + 308 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + } + ], + "index": 14 + }, + { + "bbox": [ + 104, + 306, + 506, + 322 + ], + "spans": [ + { + "bbox": [ + 104, + 306, + 261, + 322 + ], + "score": 1.0, + "content": "is maintained. On the other hand, if", + "type": "text" + }, + { + "bbox": [ + 262, + 309, + 281, + 318 + ], + "score": 0.86, + "content": "b \\cdot I", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 306, + 359, + 322 + ], + "score": 1.0, + "content": "decreases beyond", + "type": "text" + }, + { + "bbox": [ + 359, + 307, + 418, + 320 + ], + "score": 0.95, + "content": "\\mathcal { O } ( K ^ { - 1 } \\epsilon ^ { - 1 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 419, + 306, + 506, + 322 + ], + "score": 1.0, + "content": ", the communication", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 319, + 506, + 334 + ], + "spans": [ + { + "bbox": [ + 105, + 320, + 212, + 334 + ], + "score": 1.0, + "content": "complexity increases from", + "type": "text" + }, + { + "bbox": [ + 212, + 319, + 243, + 332 + ], + "score": 0.92, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 320, + 351, + 334 + ], + "score": 1.0, + "content": ". For instance, if we choose", + "type": "text" + }, + { + "bbox": [ + 351, + 321, + 391, + 333 + ], + "score": 0.92, + "content": "b = \\mathcal { O } ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 391, + 320, + 408, + 334 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 408, + 321, + 448, + 332 + ], + "score": 0.9, + "content": "I = { \\mathcal { O } } ( 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 320, + 506, + 334 + ], + "score": 1.0, + "content": "the communi-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 104, + 330, + 507, + 347 + ], + "spans": [ + { + "bbox": [ + 104, + 330, + 217, + 347 + ], + "score": 1.0, + "content": "cation complexity becomes", + "type": "text" + }, + { + "bbox": [ + 218, + 333, + 257, + 345 + ], + "score": 0.91, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 330, + 407, + 347 + ], + "score": 1.0, + "content": "while the optimal sample complexity", + "type": "text" + }, + { + "bbox": [ + 408, + 332, + 447, + 345 + ], + "score": 0.91, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 447, + 330, + 507, + 347 + ], + "score": 1.0, + "content": "is maintained.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 343, + 506, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 343, + 255, + 357 + ], + "score": 1.0, + "content": "This trade-off is illustrated in Figure", + "type": "text" + }, + { + "bbox": [ + 255, + 344, + 270, + 357 + ], + "score": 0.82, + "content": "\\boxed { 1 \\mathrm { a } }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 343, + 506, + 357 + ], + "score": 1.0, + "content": "where we maintain the optimal sample complexity, while", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 355, + 505, + 368 + ], + "spans": [ + { + "bbox": [ + 106, + 355, + 145, + 368 + ], + "score": 1.0, + "content": "changing", + "type": "text" + }, + { + "bbox": [ + 146, + 356, + 151, + 365 + ], + "score": 0.65, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 355, + 169, + 368 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 169, + 356, + 176, + 365 + ], + "score": 0.78, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 355, + 309, + 368 + ], + "score": 1.0, + "content": "to generate the trade-off surface.", + "type": "text" + }, + { + "bbox": [ + 494, + 356, + 505, + 367 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 14.5, + "bbox_fs": [ + 104, + 242, + 507, + 368 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 369, + 506, + 464 + ], + "lines": [ + { + "bbox": [ + 105, + 368, + 506, + 388 + ], + "spans": [ + { + "bbox": [ + 105, + 368, + 281, + 386 + ], + "score": 1.0, + "content": "Remark 5 (Data Heterogeneity). The term", + "type": "text" + }, + { + "bbox": [ + 282, + 369, + 352, + 388 + ], + "score": 0.94, + "content": "\\begin{array} { r } { \\tilde { \\mathcal { O } } \\biggl ( \\frac { \\zeta ^ { 2 } } { K ^ { 2 \\nu / 3 } T ^ { 1 - \\nu / 3 } } \\biggr ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 371, + 441, + 387 + ], + "score": 1.0, + "content": "in the gradient bound", + "type": "text" + }, + { + "bbox": [ + 441, + 372, + 454, + 385 + ], + "score": 0.82, + "content": "\\textcircled{4}", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 371, + 506, + 387 + ], + "score": 1.0, + "content": "captures the", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 387, + 505, + 399 + ], + "spans": [ + { + "bbox": [ + 106, + 387, + 318, + 399 + ], + "score": 1.0, + "content": "effect of the heterogeneity of data across WNs, where", + "type": "text" + }, + { + "bbox": [ + 318, + 388, + 325, + 398 + ], + "score": 0.8, + "content": "\\zeta", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 387, + 505, + 399 + ], + "score": 1.0, + "content": "is the parameter characterizing the intra-node", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 397, + 506, + 410 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 292, + 410 + ], + "score": 1.0, + "content": "variance and has been defined in Assumption", + "type": "text" + }, + { + "bbox": [ + 292, + 397, + 302, + 410 + ], + "score": 0.8, + "content": "\\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 398, + 476, + 410 + ], + "score": 1.0, + "content": "(ii). Highly heterogeneous data with large", + "type": "text" + }, + { + "bbox": [ + 477, + 398, + 487, + 410 + ], + "score": 0.87, + "content": "\\zeta ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 398, + 506, + 410 + ], + "score": 1.0, + "content": "can", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 408, + 506, + 421 + ], + "spans": [ + { + "bbox": [ + 106, + 408, + 414, + 421 + ], + "score": 1.0, + "content": "adversely impact the performance of STEM. Note that such a dependency on", + "type": "text" + }, + { + "bbox": [ + 414, + 409, + 420, + 420 + ], + "score": 0.83, + "content": "\\zeta", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 408, + 506, + 421 + ], + "score": 1.0, + "content": "also appears in other", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 419, + 506, + 431 + ], + "spans": [ + { + "bbox": [ + 105, + 419, + 235, + 431 + ], + "score": 1.0, + "content": "existing FL algorithms, such as", + "type": "text" + }, + { + "bbox": [ + 235, + 419, + 278, + 430 + ], + "score": 0.59, + "content": "[ \\bigcirc , \\bigcirc , \\bigcirc , \\bigcirc , \\bigcirc ]", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 419, + 506, + 431 + ], + "score": 1.0, + "content": ". However, there is one special case of STEM that does", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 429, + 505, + 442 + ], + "spans": [ + { + "bbox": [ + 105, + 429, + 225, + 442 + ], + "score": 1.0, + "content": "not depend on the parameter", + "type": "text" + }, + { + "bbox": [ + 225, + 431, + 231, + 442 + ], + "score": 0.82, + "content": "\\zeta", + "type": "inline_equation" + }, + { + "bbox": [ + 232, + 429, + 328, + 442 + ], + "score": 1.0, + "content": ". This is the case where", + "type": "text" + }, + { + "bbox": [ + 329, + 431, + 353, + 441 + ], + "score": 0.89, + "content": "I = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 429, + 505, + 442 + ], + "score": 1.0, + "content": ", i.e., the minibatch SGD counterpart", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 441, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 106, + 441, + 506, + 453 + ], + "score": 1.0, + "content": "of STEM where only a single local iteration is performed between two communication rounds. We", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 451, + 506, + 465 + ], + "spans": [ + { + "bbox": [ + 105, + 451, + 223, + 465 + ], + "score": 1.0, + "content": "have the following corollary.", + "type": "text" + }, + { + "bbox": [ + 494, + 452, + 506, + 464 + ], + "score": 0.999, + "content": "□", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 23.5, + "bbox_fs": [ + 105, + 368, + 506, + 465 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 467, + 506, + 504 + ], + "lines": [ + { + "bbox": [ + 106, + 467, + 506, + 481 + ], + "spans": [ + { + "bbox": [ + 106, + 467, + 319, + 481 + ], + "score": 1.0, + "content": "Corollary 1 (Minibatch STEM). Under Assumptions", + "type": "text" + }, + { + "bbox": [ + 319, + 467, + 352, + 480 + ], + "score": 0.7, + "content": "\\bigstar \\bigstar \\bigstar | \\bigstar |", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 467, + 506, + 481 + ], + "score": 1.0, + "content": ", and choose the algorithm parameters", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 104, + 478, + 508, + 494 + ], + "spans": [ + { + "bbox": [ + 104, + 478, + 164, + 494 + ], + "score": 1.0, + "content": "as in Theorem", + "type": "text" + }, + { + "bbox": [ + 164, + 478, + 182, + 493 + ], + "score": 0.74, + "content": "3 . I .", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 478, + 266, + 494 + ], + "score": 1.0, + "content": "At each WN, choose", + "type": "text" + }, + { + "bbox": [ + 267, + 481, + 291, + 491 + ], + "score": 0.86, + "content": "I = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 478, + 295, + 494 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 295, + 480, + 359, + 493 + ], + "score": 0.9, + "content": "b = ( T / K ^ { 2 } ) ^ { 1 / 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 478, + 464, + 494 + ], + "score": 1.0, + "content": ", and the initial batch size", + "type": "text" + }, + { + "bbox": [ + 464, + 481, + 503, + 491 + ], + "score": 0.88, + "content": "B = \\boldsymbol { b } \\cdot \\boldsymbol { I }", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 478, + 508, + 494 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 490, + 196, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 490, + 196, + 504 + ], + "score": 1.0, + "content": "Then STEM satisfies:", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29, + "bbox_fs": [ + 104, + 467, + 508, + 504 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 512, + 324, + 525 + ], + "lines": [ + { + "bbox": [ + 109, + 512, + 324, + 527 + ], + "spans": [ + { + "bbox": [ + 109, + 512, + 140, + 527 + ], + "score": 1.0, + "content": "(i) For", + "type": "text" + }, + { + "bbox": [ + 140, + 514, + 151, + 524 + ], + "score": 0.86, + "content": "\\bar { x } _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 512, + 277, + 527 + ], + "score": 1.0, + "content": "chosen according to Algorithm", + "type": "text" + }, + { + "bbox": [ + 277, + 513, + 287, + 526 + ], + "score": 0.69, + "content": "\\boldsymbol { l } ,", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 512, + 324, + 527 + ], + "score": 1.0, + "content": "we have", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 31, + "bbox_fs": [ + 109, + 512, + 324, + 527 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 220, + 532, + 407, + 558 + ], + "lines": [ + { + "bbox": [ + 220, + 532, + 407, + 558 + ], + "spans": [ + { + "bbox": [ + 220, + 532, + 407, + 558 + ], + "score": 0.92, + "content": "\\mathbb { E } \\| \\nabla f ( \\bar { x } _ { a } ) \\| ^ { 2 } = \\mathcal { O } \\Big ( \\frac { f ( \\bar { x } _ { 1 } ) - f ^ { * } } { T } \\Big ) + \\tilde { \\mathcal { O } } \\Big ( \\frac { \\sigma ^ { 2 } } { T } \\Big ) .", + "type": "interline_equation", + "image_path": "fe27c5ca3c886c333d5add660b12f630bb6023b83b7450c21e97035dfce670f9.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 220, + 532, + 407, + 558 + ], + "spans": [], + "index": 32 + } + ] + }, + { + "type": "text", + "bbox": [ + 103, + 570, + 466, + 585 + ], + "lines": [ + { + "bbox": [ + 104, + 568, + 468, + 587 + ], + "spans": [ + { + "bbox": [ + 104, + 568, + 234, + 587 + ], + "score": 1.0, + "content": "(ii) Minibatch STEM achieves", + "type": "text" + }, + { + "bbox": [ + 234, + 570, + 273, + 584 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 273, + 568, + 323, + 587 + ], + "score": 1.0, + "content": "sample and", + "type": "text" + }, + { + "bbox": [ + 323, + 570, + 354, + 584 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 568, + 468, + 587 + ], + "score": 1.0, + "content": "communication complexity.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 33, + "bbox_fs": [ + 104, + 568, + 468, + 587 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 593, + 506, + 616 + ], + "lines": [ + { + "bbox": [ + 105, + 592, + 505, + 607 + ], + "spans": [ + { + "bbox": [ + 105, + 592, + 505, + 607 + ], + "score": 1.0, + "content": "Next, we show that FedAvg also exhibits a trade-off similar to that of STEM but with worse sample", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 605, + 243, + 616 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 243, + 616 + ], + "score": 1.0, + "content": "and communication complexities.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5, + "bbox_fs": [ + 105, + 592, + 505, + 616 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 631, + 288, + 644 + ], + "lines": [ + { + "bbox": [ + 105, + 629, + 288, + 645 + ], + "spans": [ + { + "bbox": [ + 105, + 629, + 288, + 645 + ], + "score": 1.0, + "content": "3.2 Special cases: The FedAvg algorithm", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 107, + 652, + 505, + 721 + ], + "lines": [ + { + "bbox": [ + 105, + 652, + 506, + 665 + ], + "spans": [ + { + "bbox": [ + 105, + 652, + 506, + 665 + ], + "score": 1.0, + "content": "We briefly discuss another interesting special case of STEM, where the local momentum update", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 663, + 506, + 676 + ], + "spans": [ + { + "bbox": [ + 105, + 663, + 285, + 676 + ], + "score": 1.0, + "content": "is replaced by the conventional SGD (i.e.,", + "type": "text" + }, + { + "bbox": [ + 285, + 664, + 341, + 675 + ], + "score": 0.87, + "content": "a _ { t } = 1 , ~ \\forall ~ t )", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 663, + 506, + 676 + ], + "score": 1.0, + "content": ", while the server does not perform the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 673, + 506, + 687 + ], + "spans": [ + { + "bbox": [ + 105, + 674, + 205, + 687 + ], + "score": 1.0, + "content": "momentum update (i.e.,", + "type": "text" + }, + { + "bbox": [ + 205, + 673, + 252, + 685 + ], + "score": 0.91, + "content": "\\bar { d } _ { t } = 0 , \\forall t )", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 674, + 506, + 687 + ], + "score": 1.0, + "content": ". This is essentially the classical FedAvg algorithm, just that it", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 684, + 505, + 698 + ], + "spans": [ + { + "bbox": [ + 105, + 684, + 259, + 698 + ], + "score": 1.0, + "content": "balances the number of local updates", + "type": "text" + }, + { + "bbox": [ + 259, + 686, + 266, + 695 + ], + "score": 0.78, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 684, + 360, + 698 + ], + "score": 1.0, + "content": "and the minibatch size", + "type": "text" + }, + { + "bbox": [ + 360, + 686, + 366, + 695 + ], + "score": 0.53, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 366, + 684, + 505, + 698 + ], + "score": 1.0, + "content": ". We show that this algorithm also", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 694, + 505, + 710 + ], + "spans": [ + { + "bbox": [ + 105, + 694, + 217, + 710 + ], + "score": 1.0, + "content": "exhibits a trade-off between", + "type": "text" + }, + { + "bbox": [ + 218, + 697, + 224, + 706 + ], + "score": 0.71, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 224, + 694, + 241, + 710 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 241, + 697, + 248, + 706 + ], + "score": 0.77, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 248, + 694, + 396, + 710 + ], + "score": 1.0, + "content": "and on the trade-off curve it achieves", + "type": "text" + }, + { + "bbox": [ + 396, + 696, + 428, + 708 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 428, + 694, + 505, + 710 + ], + "score": 1.0, + "content": "sample complexity", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 104, + 705, + 279, + 723 + ], + "spans": [ + { + "bbox": [ + 104, + 705, + 123, + 723 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 124, + 707, + 163, + 721 + ], + "score": 0.93, + "content": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 705, + 279, + 723 + ], + "score": 1.0, + "content": "communication complexity.", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 39.5, + "bbox_fs": [ + 104, + 652, + 506, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 70, + 306, + 185 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 70, + 306, + 185 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 70, + 306, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 70, + 306, + 185 + ], + "score": 0.977, + "html": "
Algorithm Training Acc.Testing Acc.
FedAvg78.274.1
FedProx79.274.8
FedDyn68.966.0
SCAFFOLD71.974.0
MIME82.676.8
FedGLOMO76.172.8
STEM80.178.8
", + "type": "table", + "image_path": "6bf0f1ad1737bddbd1a909e1c507f78edccdbe2e2cf8c63393f03d6b14335b18.jpg" + } + ] + } + ], + "index": 3.5, + "virtual_lines": [ + { + "bbox": [ + 106, + 70, + 306, + 84.375 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 84.375, + 306, + 98.75 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 98.75, + 306, + 113.125 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 113.125, + 306, + 127.5 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 106, + 127.5, + 306, + 141.875 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 106, + 141.875, + 306, + 156.25 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 106, + 156.25, + 306, + 170.625 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 106, + 170.625, + 306, + 185.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 118, + 190, + 274, + 201 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 118, + 189, + 275, + 202 + ], + "spans": [ + { + "bbox": [ + 118, + 189, + 203, + 202 + ], + "score": 1.0, + "content": "(a) Mild heterogeneity,", + "type": "text" + }, + { + "bbox": [ + 203, + 190, + 230, + 200 + ], + "score": 0.88, + "content": "b = 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 189, + 248, + 202 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 248, + 190, + 271, + 200 + ], + "score": 0.77, + "content": "I = 7", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 189, + 275, + 202 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + } + ], + "index": 5.75 + }, + { + "type": "table", + "bbox": [ + 321, + 70, + 520, + 185 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 321, + 70, + 520, + 185 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 321, + 70, + 520, + 185 + ], + "spans": [ + { + "bbox": [ + 321, + 70, + 520, + 185 + ], + "score": 0.978, + "html": "
AlgorithmTraining Acc.Testing Acc.
FedAvg73.675.4
FedProx80.075.2
FedDyn76.171.3
SCAFFOLD72.573.7
MIME61.558.6
FedGLOMO10.010.0
STEM81.178.5
", + "type": "table", + "image_path": "b0513e337be144b5f857c24b8a10281f3fd33f9a14e65f9ee342f421e0353331.jpg" + } + ] + } + ], + "index": 12.5, + "virtual_lines": [ + { + "bbox": [ + 321, + 70, + 520, + 84.375 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 321, + 84.375, + 520, + 98.75 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 321, + 98.75, + 520, + 113.125 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 321, + 113.125, + 520, + 127.5 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 321, + 127.5, + 520, + 141.875 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 321, + 141.875, + 520, + 156.25 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 321, + 156.25, + 520, + 170.625 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 321, + 170.625, + 520, + 185.0 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 325, + 190, + 497, + 201 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 324, + 189, + 495, + 202 + ], + "spans": [ + { + "bbox": [ + 324, + 189, + 427, + 202 + ], + "score": 1.0, + "content": "(b) Moderate heterogeneity,", + "type": "text" + }, + { + "bbox": [ + 427, + 190, + 450, + 200 + ], + "score": 0.92, + "content": "b = 8", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 189, + 467, + 202 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 468, + 190, + 495, + 200 + ], + "score": 0.88, + "content": "I = 6 1", + "type": "inline_equation" + } + ], + "index": 17 + } + ], + "index": 17 + } + ], + "index": 14.75 + }, + { + "type": "text", + "bbox": [ + 106, + 207, + 504, + 230 + ], + "lines": [ + { + "bbox": [ + 106, + 206, + 506, + 219 + ], + "spans": [ + { + "bbox": [ + 106, + 206, + 506, + 219 + ], + "score": 1.0, + "content": "Table 2: Training and testing accuracy of different algorithms on CIFAR-10 dataset for different", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 217, + 365, + 231 + ], + "spans": [ + { + "bbox": [ + 105, + 217, + 365, + 231 + ], + "score": 1.0, + "content": "batch-sizes, number of local updates, and heteregeneity settings.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 106, + 257, + 505, + 287 + ], + "lines": [ + { + "bbox": [ + 105, + 256, + 506, + 271 + ], + "spans": [ + { + "bbox": [ + 105, + 256, + 506, + 271 + ], + "score": 1.0, + "content": "Theorem 3.2 (The FedAvg Algorithm). Under Assumptions 1 and 2, suppose the stepsize is chosen", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 268, + 215, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 122, + 289 + ], + "score": 1.0, + "content": "as:", + "type": "text" + }, + { + "bbox": [ + 122, + 269, + 164, + 289 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\eta = \\sqrt { \\frac { b K } { T } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 268, + 215, + 289 + ], + "score": 1.0, + "content": "; Let us set:", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5 + }, + { + "type": "interline_equation", + "bbox": [ + 206, + 293, + 404, + 310 + ], + "lines": [ + { + "bbox": [ + 206, + 293, + 404, + 310 + ], + "spans": [ + { + "bbox": [ + 206, + 293, + 404, + 310 + ], + "score": 0.89, + "content": "I = \\mathcal { O } \\big ( ( T / K ^ { 3 } ) ^ { \\nu / 4 } \\big ) , \\quad b = \\mathcal { O } \\big ( ( T / K ^ { 3 } ) ^ { 1 / 3 - \\nu / 3 } \\big )", + "type": "interline_equation", + "image_path": "614827bc2783d2a130099b5dcc25a039f7d6ca4f991f27f94b6b9f6e8cc8d82c.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 206, + 293, + 404, + 310 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 315, + 465, + 328 + ], + "lines": [ + { + "bbox": [ + 106, + 315, + 466, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 315, + 133, + 329 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 316, + 172, + 328 + ], + "score": 0.91, + "content": "\\nu \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 315, + 317, + 329 + ], + "score": 1.0, + "content": "is a constant. Then for FedAvg with", + "type": "text" + }, + { + "bbox": [ + 318, + 315, + 383, + 327 + ], + "score": 0.91, + "content": "T \\geq 8 1 L ^ { 2 } I ^ { 2 } b K", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 315, + 466, + 329 + ], + "score": 1.0, + "content": ", the following holds", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 108, + 336, + 323, + 349 + ], + "lines": [ + { + "bbox": [ + 108, + 336, + 324, + 351 + ], + "spans": [ + { + "bbox": [ + 108, + 336, + 140, + 351 + ], + "score": 1.0, + "content": "(i) For", + "type": "text" + }, + { + "bbox": [ + 140, + 339, + 151, + 348 + ], + "score": 0.85, + "content": "\\scriptstyle { \\bar { x } } _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 336, + 277, + 351 + ], + "score": 1.0, + "content": "chosen according to Algorithm", + "type": "text" + }, + { + "bbox": [ + 277, + 337, + 287, + 350 + ], + "score": 0.53, + "content": "\\perp", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 336, + 324, + 351 + ], + "score": 1.0, + "content": "we have", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "interline_equation", + "bbox": [ + 137, + 354, + 491, + 382 + ], + "lines": [ + { + "bbox": [ + 137, + 354, + 491, + 382 + ], + "spans": [ + { + "bbox": [ + 137, + 354, + 491, + 382 + ], + "score": 0.95, + "content": "\\mathbb { E } \\Vert \\nabla f ( \\bar { x } _ { a } ) \\Vert ^ { 2 } = \\mathcal { O } \\Bigg ( \\frac { f ( \\bar { x } _ { 1 } ) - f ^ { * } } { K ^ { \\nu / 2 } T ^ { 2 / 3 - \\nu / 6 } } \\Bigg ) + \\mathcal { O } \\Bigg ( \\frac { \\sigma ^ { 2 } } { K ^ { \\nu / 2 } T ^ { 2 / 3 - \\nu / 6 } } \\Bigg ) + \\mathcal { O } \\Bigg ( \\frac { \\zeta ^ { 2 } } { K ^ { \\nu / 2 } T ^ { 2 / 3 - \\nu / 6 } } \\Bigg ) .", + "type": "interline_equation", + "image_path": "fcbbd1eccd40989ffc376397736fd5950f8008a4dccca06a87727fb72022e909.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 137, + 354, + 491, + 363.3333333333333 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 137, + 363.3333333333333, + 491, + 372.66666666666663 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 137, + 372.66666666666663, + 491, + 381.99999999999994 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 392, + 273, + 404 + ], + "lines": [ + { + "bbox": [ + 105, + 391, + 274, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 195, + 406 + ], + "score": 1.0, + "content": "(ii) For any choice of", + "type": "text" + }, + { + "bbox": [ + 196, + 393, + 234, + 404 + ], + "score": 0.9, + "content": "\\nu \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 391, + 274, + 406 + ], + "score": 1.0, + "content": "we have:", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 120, + 403, + 505, + 435 + ], + "lines": [ + { + "bbox": [ + 123, + 402, + 506, + 416 + ], + "spans": [ + { + "bbox": [ + 123, + 402, + 358, + 416 + ], + "score": 1.0, + "content": "Sample Complexity: The sample complexity of FedAvg is", + "type": "text" + }, + { + "bbox": [ + 358, + 403, + 390, + 415 + ], + "score": 0.92, + "content": "\\mathcal { O } ( \\epsilon ^ { - 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 402, + 506, + 416 + ], + "score": 1.0, + "content": ". This implies that each WN", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 122, + 413, + 506, + 427 + ], + "spans": [ + { + "bbox": [ + 122, + 413, + 190, + 427 + ], + "score": 1.0, + "content": "requires at most", + "type": "text" + }, + { + "bbox": [ + 190, + 414, + 241, + 426 + ], + "score": 0.91, + "content": "\\tilde { \\mathcal { O } } ( K ^ { - 1 } \\epsilon ^ { - 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 413, + 506, + 427 + ], + "score": 1.0, + "content": "gradient computations, thereby achieving linear speedup with the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 122, + 425, + 249, + 437 + ], + "spans": [ + { + "bbox": [ + 122, + 425, + 249, + 437 + ], + "score": 1.0, + "content": "number of WNs in the network.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 124, + 436, + 466, + 448 + ], + "lines": [ + { + "bbox": [ + 122, + 433, + 467, + 450 + ], + "spans": [ + { + "bbox": [ + 122, + 433, + 423, + 450 + ], + "score": 1.0, + "content": "Communication Complexity: The communication complexity of FedAvg is", + "type": "text" + }, + { + "bbox": [ + 424, + 435, + 462, + 448 + ], + "score": 0.9, + "content": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 463, + 433, + 467, + 450 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 106, + 457, + 505, + 568 + ], + "lines": [ + { + "bbox": [ + 105, + 456, + 506, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 228, + 470 + ], + "score": 1.0, + "content": "Note that the requirement on", + "type": "text" + }, + { + "bbox": [ + 229, + 458, + 237, + 467 + ], + "score": 0.8, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 456, + 506, + 470 + ], + "score": 1.0, + "content": "being lower bounded is only relevant for theoretical purposes, a", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 467, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 467, + 265, + 480 + ], + "score": 1.0, + "content": "similar requirement was also imposed in", + "type": "text" + }, + { + "bbox": [ + 266, + 468, + 283, + 479 + ], + "score": 0.84, + "content": "[ \\mathbb { 1 4 } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 283, + 467, + 455, + 480 + ], + "score": 1.0, + "content": "to prove convergence. Again, the parameter", + "type": "text" + }, + { + "bbox": [ + 456, + 469, + 494, + 480 + ], + "score": 0.91, + "content": "\\nu \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 467, + 505, + 480 + ], + "score": 1.0, + "content": "in", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 478, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 208, + 492 + ], + "score": 1.0, + "content": "the statement of Theorem", + "type": "text" + }, + { + "bbox": [ + 208, + 479, + 225, + 492 + ], + "score": 0.76, + "content": "3 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 478, + 260, + 492 + ], + "score": 1.0, + "content": "balances", + "type": "text" + }, + { + "bbox": [ + 261, + 480, + 267, + 489 + ], + "score": 0.74, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 478, + 284, + 492 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 285, + 480, + 291, + 489 + ], + "score": 0.77, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 478, + 505, + 492 + ], + "score": 1.0, + "content": "at each WN while maintaining state-of-the-art sample", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 489, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 106, + 490, + 308, + 503 + ], + "score": 1.0, + "content": "and communication complexities; please see Table", + "type": "text" + }, + { + "bbox": [ + 308, + 489, + 317, + 502 + ], + "score": 0.72, + "content": "^ 1", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 490, + 505, + 503 + ], + "score": 1.0, + "content": "for a comparison of those bounds with existing", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 500, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 195, + 514 + ], + "score": 1.0, + "content": "FedAvg bounds. For", + "type": "text" + }, + { + "bbox": [ + 195, + 502, + 222, + 511 + ], + "score": 0.89, + "content": "\\nu = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 500, + 318, + 514 + ], + "score": 1.0, + "content": ", FedAvg (cf. Theorem", + "type": "text" + }, + { + "bbox": [ + 318, + 501, + 337, + 514 + ], + "score": 0.83, + "content": "\\textcircled { 3 . 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 500, + 505, + 514 + ], + "score": 1.0, + "content": "reduces to FedAvg proposed in [12, 14]", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 511, + 505, + 524 + ], + "spans": [ + { + "bbox": [ + 106, + 511, + 138, + 524 + ], + "score": 1.0, + "content": "and for", + "type": "text" + }, + { + "bbox": [ + 139, + 513, + 165, + 522 + ], + "score": 0.9, + "content": "\\nu = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 511, + 505, + 524 + ], + "score": 1.0, + "content": ", the algorithm can be viewed as a large batch FedAvg with constant local updates", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 522, + 505, + 535 + ], + "spans": [ + { + "bbox": [ + 106, + 522, + 335, + 535 + ], + "score": 1.0, + "content": "[15, 16]. Note that similar to STEM, it is known that for", + "type": "text" + }, + { + "bbox": [ + 335, + 523, + 360, + 533 + ], + "score": 0.9, + "content": "I = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 522, + 505, + 535 + ], + "score": 1.0, + "content": ", the Minibatch SGD’s performance", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 533, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 298, + 546 + ], + "score": 1.0, + "content": "is independent of the heterogeneity parameter,", + "type": "text" + }, + { + "bbox": [ + 299, + 533, + 325, + 545 + ], + "score": 0.38, + "content": "\\zeta \\equiv \\mathbb { I I } 3 \\mathbb { I }", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 533, + 475, + 546 + ], + "score": 1.0, + "content": ". We also point out that if Algorithm", + "type": "text" + }, + { + "bbox": [ + 476, + 533, + 486, + 546 + ], + "score": 0.56, + "content": "^ 1", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 533, + 505, + 546 + ], + "score": 1.0, + "content": "uses", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 544, + 506, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 544, + 248, + 557 + ], + "score": 1.0, + "content": "Nesterov’s or Polyak’s momentum", + "type": "text" + }, + { + "bbox": [ + 248, + 545, + 266, + 556 + ], + "score": 0.77, + "content": "[ \\textcircled { 1 4 } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 544, + 506, + 557 + ], + "score": 1.0, + "content": "at local WNs instead of the recursive momentum estimator", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 555, + 293, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 275, + 568 + ], + "score": 1.0, + "content": "we get the same guarantees as in Theorem", + "type": "text" + }, + { + "bbox": [ + 276, + 555, + 293, + 568 + ], + "score": 0.33, + "content": "3 . 2 .", + "type": "inline_equation" + } + ], + "index": 42 + } + ], + "index": 37.5 + }, + { + "type": "text", + "bbox": [ + 107, + 572, + 505, + 649 + ], + "lines": [ + { + "bbox": [ + 105, + 572, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 505, + 585 + ], + "score": 1.0, + "content": "In summary, this section established that once the WN’s and the SN’s update directions (SGD in", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 582, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 505, + 596 + ], + "score": 1.0, + "content": "FedAvg and momentum based directions in STEM) are fixed, there exists a sequence of optimal", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 593, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 593, + 264, + 606 + ], + "score": 1.0, + "content": "choices of the number of local updates", + "type": "text" + }, + { + "bbox": [ + 265, + 594, + 271, + 604 + ], + "score": 0.62, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 593, + 353, + 606 + ], + "score": 1.0, + "content": ", and the batch sizes", + "type": "text" + }, + { + "bbox": [ + 353, + 594, + 359, + 604 + ], + "score": 0.69, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 593, + 505, + 606 + ], + "score": 1.0, + "content": ", which guarantees the best possible", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "sample and communication complexities for the particular algorithm. The trade-off analysis presented", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 615, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 399, + 627 + ], + "score": 1.0, + "content": "in this section provides some useful guidelines for how to best select", + "type": "text" + }, + { + "bbox": [ + 400, + 616, + 406, + 626 + ], + "score": 0.61, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 615, + 425, + 627 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 426, + 616, + 433, + 626 + ], + "score": 0.72, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 615, + 505, + 627 + ], + "score": 1.0, + "content": "in practice. Our", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 325, + 640 + ], + "score": 1.0, + "content": "subsequent numerical results will also verify that if", + "type": "text" + }, + { + "bbox": [ + 325, + 627, + 331, + 636 + ], + "score": 0.72, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 626, + 344, + 640 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 345, + 627, + 352, + 636 + ], + "score": 0.75, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 626, + 505, + 640 + ], + "score": 1.0, + "content": "are not chosen judiciously, then the", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 638, + 370, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 370, + 650 + ], + "score": 1.0, + "content": "practical performance of the algorithms can degrade significantly.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 46 + }, + { + "type": "title", + "bbox": [ + 108, + 664, + 216, + 677 + ], + "lines": [ + { + "bbox": [ + 105, + 662, + 218, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 218, + 680 + ], + "score": 1.0, + "content": "4 Numerical results", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 50 + }, + { + "type": "text", + "bbox": [ + 107, + 688, + 504, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 689, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 689, + 506, + 701 + ], + "score": 1.0, + "content": "In this section, we validate the proposed STEM algorithm and compare its performance with the", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 377, + 711 + ], + "score": 1.0, + "content": "de facto standard FedAvg [11], and the algorithms stated in Table", + "type": "text" + }, + { + "bbox": [ + 378, + 700, + 388, + 712 + ], + "score": 0.71, + "content": "^ { 1 . }", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 699, + 506, + 711 + ], + "score": 1.0, + "content": "Note that instead of FedPD", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 711, + 506, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 326, + 724 + ], + "score": 1.0, + "content": "we include the performance comparison with FedDyn", + "type": "text" + }, + { + "bbox": [ + 326, + 711, + 343, + 722 + ], + "score": 0.64, + "content": "\\pmb { \\mathbb { B } } 6 \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 711, + 506, + 724 + ], + "score": 1.0, + "content": "since they are known to be very closely", + "type": "text" + } + ], + "index": 53 + } + ], + "index": 52 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 9 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 70, + 306, + 185 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 70, + 306, + 185 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 70, + 306, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 70, + 306, + 185 + ], + "score": 0.977, + "html": "
Algorithm Training Acc.Testing Acc.
FedAvg78.274.1
FedProx79.274.8
FedDyn68.966.0
SCAFFOLD71.974.0
MIME82.676.8
FedGLOMO76.172.8
STEM80.178.8
", + "type": "table", + "image_path": "6bf0f1ad1737bddbd1a909e1c507f78edccdbe2e2cf8c63393f03d6b14335b18.jpg" + } + ] + } + ], + "index": 3.5, + "virtual_lines": [ + { + "bbox": [ + 106, + 70, + 306, + 84.375 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 84.375, + 306, + 98.75 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 98.75, + 306, + 113.125 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 113.125, + 306, + 127.5 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 106, + 127.5, + 306, + 141.875 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 106, + 141.875, + 306, + 156.25 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 106, + 156.25, + 306, + 170.625 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 106, + 170.625, + 306, + 185.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 118, + 190, + 274, + 201 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 118, + 189, + 275, + 202 + ], + "spans": [ + { + "bbox": [ + 118, + 189, + 203, + 202 + ], + "score": 1.0, + "content": "(a) Mild heterogeneity,", + "type": "text" + }, + { + "bbox": [ + 203, + 190, + 230, + 200 + ], + "score": 0.88, + "content": "b = 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 189, + 248, + 202 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 248, + 190, + 271, + 200 + ], + "score": 0.77, + "content": "I = 7", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 189, + 275, + 202 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + } + ], + "index": 5.75 + }, + { + "type": "table", + "bbox": [ + 321, + 70, + 520, + 185 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 321, + 70, + 520, + 185 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 321, + 70, + 520, + 185 + ], + "spans": [ + { + "bbox": [ + 321, + 70, + 520, + 185 + ], + "score": 0.978, + "html": "
AlgorithmTraining Acc.Testing Acc.
FedAvg73.675.4
FedProx80.075.2
FedDyn76.171.3
SCAFFOLD72.573.7
MIME61.558.6
FedGLOMO10.010.0
STEM81.178.5
", + "type": "table", + "image_path": "b0513e337be144b5f857c24b8a10281f3fd33f9a14e65f9ee342f421e0353331.jpg" + } + ] + } + ], + "index": 12.5, + "virtual_lines": [ + { + "bbox": [ + 321, + 70, + 520, + 84.375 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 321, + 84.375, + 520, + 98.75 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 321, + 98.75, + 520, + 113.125 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 321, + 113.125, + 520, + 127.5 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 321, + 127.5, + 520, + 141.875 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 321, + 141.875, + 520, + 156.25 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 321, + 156.25, + 520, + 170.625 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 321, + 170.625, + 520, + 185.0 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 325, + 190, + 497, + 201 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 324, + 189, + 495, + 202 + ], + "spans": [ + { + "bbox": [ + 324, + 189, + 427, + 202 + ], + "score": 1.0, + "content": "(b) Moderate heterogeneity,", + "type": "text" + }, + { + "bbox": [ + 427, + 190, + 450, + 200 + ], + "score": 0.92, + "content": "b = 8", + "type": "inline_equation" + }, + { + "bbox": [ + 450, + 189, + 467, + 202 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 468, + 190, + 495, + 200 + ], + "score": 0.88, + "content": "I = 6 1", + "type": "inline_equation" + } + ], + "index": 17 + } + ], + "index": 17 + } + ], + "index": 14.75 + }, + { + "type": "text", + "bbox": [ + 106, + 207, + 504, + 230 + ], + "lines": [ + { + "bbox": [ + 106, + 206, + 506, + 219 + ], + "spans": [ + { + "bbox": [ + 106, + 206, + 506, + 219 + ], + "score": 1.0, + "content": "Table 2: Training and testing accuracy of different algorithms on CIFAR-10 dataset for different", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 217, + 365, + 231 + ], + "spans": [ + { + "bbox": [ + 105, + 217, + 365, + 231 + ], + "score": 1.0, + "content": "batch-sizes, number of local updates, and heteregeneity settings.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 206, + 506, + 231 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 257, + 505, + 287 + ], + "lines": [ + { + "bbox": [ + 105, + 256, + 506, + 271 + ], + "spans": [ + { + "bbox": [ + 105, + 256, + 506, + 271 + ], + "score": 1.0, + "content": "Theorem 3.2 (The FedAvg Algorithm). Under Assumptions 1 and 2, suppose the stepsize is chosen", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 268, + 215, + 289 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 122, + 289 + ], + "score": 1.0, + "content": "as:", + "type": "text" + }, + { + "bbox": [ + 122, + 269, + 164, + 289 + ], + "score": 0.92, + "content": "\\begin{array} { r } { \\eta = \\sqrt { \\frac { b K } { T } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 268, + 215, + 289 + ], + "score": 1.0, + "content": "; Let us set:", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20.5, + "bbox_fs": [ + 105, + 256, + 506, + 289 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 206, + 293, + 404, + 310 + ], + "lines": [ + { + "bbox": [ + 206, + 293, + 404, + 310 + ], + "spans": [ + { + "bbox": [ + 206, + 293, + 404, + 310 + ], + "score": 0.89, + "content": "I = \\mathcal { O } \\big ( ( T / K ^ { 3 } ) ^ { \\nu / 4 } \\big ) , \\quad b = \\mathcal { O } \\big ( ( T / K ^ { 3 } ) ^ { 1 / 3 - \\nu / 3 } \\big )", + "type": "interline_equation", + "image_path": "614827bc2783d2a130099b5dcc25a039f7d6ca4f991f27f94b6b9f6e8cc8d82c.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 206, + 293, + 404, + 310 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 105, + 315, + 465, + 328 + ], + "lines": [ + { + "bbox": [ + 106, + 315, + 466, + 329 + ], + "spans": [ + { + "bbox": [ + 106, + 315, + 133, + 329 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 316, + 172, + 328 + ], + "score": 0.91, + "content": "\\nu \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 315, + 317, + 329 + ], + "score": 1.0, + "content": "is a constant. Then for FedAvg with", + "type": "text" + }, + { + "bbox": [ + 318, + 315, + 383, + 327 + ], + "score": 0.91, + "content": "T \\geq 8 1 L ^ { 2 } I ^ { 2 } b K", + "type": "inline_equation" + }, + { + "bbox": [ + 384, + 315, + 466, + 329 + ], + "score": 1.0, + "content": ", the following holds", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23, + "bbox_fs": [ + 106, + 315, + 466, + 329 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 336, + 323, + 349 + ], + "lines": [ + { + "bbox": [ + 108, + 336, + 324, + 351 + ], + "spans": [ + { + "bbox": [ + 108, + 336, + 140, + 351 + ], + "score": 1.0, + "content": "(i) For", + "type": "text" + }, + { + "bbox": [ + 140, + 339, + 151, + 348 + ], + "score": 0.85, + "content": "\\scriptstyle { \\bar { x } } _ { a }", + "type": "inline_equation" + }, + { + "bbox": [ + 152, + 336, + 277, + 351 + ], + "score": 1.0, + "content": "chosen according to Algorithm", + "type": "text" + }, + { + "bbox": [ + 277, + 337, + 287, + 350 + ], + "score": 0.53, + "content": "\\perp", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 336, + 324, + 351 + ], + "score": 1.0, + "content": "we have", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24, + "bbox_fs": [ + 108, + 336, + 324, + 351 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 137, + 354, + 491, + 382 + ], + "lines": [ + { + "bbox": [ + 137, + 354, + 491, + 382 + ], + "spans": [ + { + "bbox": [ + 137, + 354, + 491, + 382 + ], + "score": 0.95, + "content": "\\mathbb { E } \\Vert \\nabla f ( \\bar { x } _ { a } ) \\Vert ^ { 2 } = \\mathcal { O } \\Bigg ( \\frac { f ( \\bar { x } _ { 1 } ) - f ^ { * } } { K ^ { \\nu / 2 } T ^ { 2 / 3 - \\nu / 6 } } \\Bigg ) + \\mathcal { O } \\Bigg ( \\frac { \\sigma ^ { 2 } } { K ^ { \\nu / 2 } T ^ { 2 / 3 - \\nu / 6 } } \\Bigg ) + \\mathcal { O } \\Bigg ( \\frac { \\zeta ^ { 2 } } { K ^ { \\nu / 2 } T ^ { 2 / 3 - \\nu / 6 } } \\Bigg ) .", + "type": "interline_equation", + "image_path": "fcbbd1eccd40989ffc376397736fd5950f8008a4dccca06a87727fb72022e909.jpg" + } + ] + } + ], + "index": 26, + "virtual_lines": [ + { + "bbox": [ + 137, + 354, + 491, + 363.3333333333333 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 137, + 363.3333333333333, + 491, + 372.66666666666663 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 137, + 372.66666666666663, + 491, + 381.99999999999994 + ], + "spans": [], + "index": 27 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 392, + 273, + 404 + ], + "lines": [ + { + "bbox": [ + 105, + 391, + 274, + 406 + ], + "spans": [ + { + "bbox": [ + 105, + 391, + 195, + 406 + ], + "score": 1.0, + "content": "(ii) For any choice of", + "type": "text" + }, + { + "bbox": [ + 196, + 393, + 234, + 404 + ], + "score": 0.9, + "content": "\\nu \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 235, + 391, + 274, + 406 + ], + "score": 1.0, + "content": "we have:", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28, + "bbox_fs": [ + 105, + 391, + 274, + 406 + ] + }, + { + "type": "text", + "bbox": [ + 120, + 403, + 505, + 435 + ], + "lines": [ + { + "bbox": [ + 123, + 402, + 506, + 416 + ], + "spans": [ + { + "bbox": [ + 123, + 402, + 358, + 416 + ], + "score": 1.0, + "content": "Sample Complexity: The sample complexity of FedAvg is", + "type": "text" + }, + { + "bbox": [ + 358, + 403, + 390, + 415 + ], + "score": 0.92, + "content": "\\mathcal { O } ( \\epsilon ^ { - 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 402, + 506, + 416 + ], + "score": 1.0, + "content": ". This implies that each WN", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 122, + 413, + 506, + 427 + ], + "spans": [ + { + "bbox": [ + 122, + 413, + 190, + 427 + ], + "score": 1.0, + "content": "requires at most", + "type": "text" + }, + { + "bbox": [ + 190, + 414, + 241, + 426 + ], + "score": 0.91, + "content": "\\tilde { \\mathcal { O } } ( K ^ { - 1 } \\epsilon ^ { - 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 413, + 506, + 427 + ], + "score": 1.0, + "content": "gradient computations, thereby achieving linear speedup with the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 122, + 425, + 249, + 437 + ], + "spans": [ + { + "bbox": [ + 122, + 425, + 249, + 437 + ], + "score": 1.0, + "content": "number of WNs in the network.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30, + "bbox_fs": [ + 122, + 402, + 506, + 437 + ] + }, + { + "type": "text", + "bbox": [ + 124, + 436, + 466, + 448 + ], + "lines": [ + { + "bbox": [ + 122, + 433, + 467, + 450 + ], + "spans": [ + { + "bbox": [ + 122, + 433, + 423, + 450 + ], + "score": 1.0, + "content": "Communication Complexity: The communication complexity of FedAvg is", + "type": "text" + }, + { + "bbox": [ + 424, + 435, + 462, + 448 + ], + "score": 0.9, + "content": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 463, + 433, + 467, + 450 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32, + "bbox_fs": [ + 122, + 433, + 467, + 450 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 457, + 505, + 568 + ], + "lines": [ + { + "bbox": [ + 105, + 456, + 506, + 470 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 228, + 470 + ], + "score": 1.0, + "content": "Note that the requirement on", + "type": "text" + }, + { + "bbox": [ + 229, + 458, + 237, + 467 + ], + "score": 0.8, + "content": "T", + "type": "inline_equation" + }, + { + "bbox": [ + 237, + 456, + 506, + 470 + ], + "score": 1.0, + "content": "being lower bounded is only relevant for theoretical purposes, a", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 467, + 505, + 480 + ], + "spans": [ + { + "bbox": [ + 106, + 467, + 265, + 480 + ], + "score": 1.0, + "content": "similar requirement was also imposed in", + "type": "text" + }, + { + "bbox": [ + 266, + 468, + 283, + 479 + ], + "score": 0.84, + "content": "[ \\mathbb { 1 4 } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 283, + 467, + 455, + 480 + ], + "score": 1.0, + "content": "to prove convergence. Again, the parameter", + "type": "text" + }, + { + "bbox": [ + 456, + 469, + 494, + 480 + ], + "score": 0.91, + "content": "\\nu \\in [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 467, + 505, + 480 + ], + "score": 1.0, + "content": "in", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 478, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 208, + 492 + ], + "score": 1.0, + "content": "the statement of Theorem", + "type": "text" + }, + { + "bbox": [ + 208, + 479, + 225, + 492 + ], + "score": 0.76, + "content": "3 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 478, + 260, + 492 + ], + "score": 1.0, + "content": "balances", + "type": "text" + }, + { + "bbox": [ + 261, + 480, + 267, + 489 + ], + "score": 0.74, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 478, + 284, + 492 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 285, + 480, + 291, + 489 + ], + "score": 0.77, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 291, + 478, + 505, + 492 + ], + "score": 1.0, + "content": "at each WN while maintaining state-of-the-art sample", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 489, + 505, + 503 + ], + "spans": [ + { + "bbox": [ + 106, + 490, + 308, + 503 + ], + "score": 1.0, + "content": "and communication complexities; please see Table", + "type": "text" + }, + { + "bbox": [ + 308, + 489, + 317, + 502 + ], + "score": 0.72, + "content": "^ 1", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 490, + 505, + 503 + ], + "score": 1.0, + "content": "for a comparison of those bounds with existing", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 500, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 500, + 195, + 514 + ], + "score": 1.0, + "content": "FedAvg bounds. For", + "type": "text" + }, + { + "bbox": [ + 195, + 502, + 222, + 511 + ], + "score": 0.89, + "content": "\\nu = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 500, + 318, + 514 + ], + "score": 1.0, + "content": ", FedAvg (cf. Theorem", + "type": "text" + }, + { + "bbox": [ + 318, + 501, + 337, + 514 + ], + "score": 0.83, + "content": "\\textcircled { 3 . 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 338, + 500, + 505, + 514 + ], + "score": 1.0, + "content": "reduces to FedAvg proposed in [12, 14]", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 511, + 505, + 524 + ], + "spans": [ + { + "bbox": [ + 106, + 511, + 138, + 524 + ], + "score": 1.0, + "content": "and for", + "type": "text" + }, + { + "bbox": [ + 139, + 513, + 165, + 522 + ], + "score": 0.9, + "content": "\\nu = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 166, + 511, + 505, + 524 + ], + "score": 1.0, + "content": ", the algorithm can be viewed as a large batch FedAvg with constant local updates", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 522, + 505, + 535 + ], + "spans": [ + { + "bbox": [ + 106, + 522, + 335, + 535 + ], + "score": 1.0, + "content": "[15, 16]. Note that similar to STEM, it is known that for", + "type": "text" + }, + { + "bbox": [ + 335, + 523, + 360, + 533 + ], + "score": 0.9, + "content": "I = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 522, + 505, + 535 + ], + "score": 1.0, + "content": ", the Minibatch SGD’s performance", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 533, + 505, + 546 + ], + "spans": [ + { + "bbox": [ + 105, + 533, + 298, + 546 + ], + "score": 1.0, + "content": "is independent of the heterogeneity parameter,", + "type": "text" + }, + { + "bbox": [ + 299, + 533, + 325, + 545 + ], + "score": 0.38, + "content": "\\zeta \\equiv \\mathbb { I I } 3 \\mathbb { I }", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 533, + 475, + 546 + ], + "score": 1.0, + "content": ". We also point out that if Algorithm", + "type": "text" + }, + { + "bbox": [ + 476, + 533, + 486, + 546 + ], + "score": 0.56, + "content": "^ 1", + "type": "inline_equation" + }, + { + "bbox": [ + 486, + 533, + 505, + 546 + ], + "score": 1.0, + "content": "uses", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 544, + 506, + 557 + ], + "spans": [ + { + "bbox": [ + 106, + 544, + 248, + 557 + ], + "score": 1.0, + "content": "Nesterov’s or Polyak’s momentum", + "type": "text" + }, + { + "bbox": [ + 248, + 545, + 266, + 556 + ], + "score": 0.77, + "content": "[ \\textcircled { 1 4 } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 266, + 544, + 506, + 557 + ], + "score": 1.0, + "content": "at local WNs instead of the recursive momentum estimator", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 555, + 293, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 555, + 275, + 568 + ], + "score": 1.0, + "content": "we get the same guarantees as in Theorem", + "type": "text" + }, + { + "bbox": [ + 276, + 555, + 293, + 568 + ], + "score": 0.33, + "content": "3 . 2 .", + "type": "inline_equation" + } + ], + "index": 42 + } + ], + "index": 37.5, + "bbox_fs": [ + 105, + 456, + 506, + 568 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 572, + 505, + 649 + ], + "lines": [ + { + "bbox": [ + 105, + 572, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 572, + 505, + 585 + ], + "score": 1.0, + "content": "In summary, this section established that once the WN’s and the SN’s update directions (SGD in", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 582, + 505, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 582, + 505, + 596 + ], + "score": 1.0, + "content": "FedAvg and momentum based directions in STEM) are fixed, there exists a sequence of optimal", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 593, + 505, + 606 + ], + "spans": [ + { + "bbox": [ + 105, + 593, + 264, + 606 + ], + "score": 1.0, + "content": "choices of the number of local updates", + "type": "text" + }, + { + "bbox": [ + 265, + 594, + 271, + 604 + ], + "score": 0.62, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 593, + 353, + 606 + ], + "score": 1.0, + "content": ", and the batch sizes", + "type": "text" + }, + { + "bbox": [ + 353, + 594, + 359, + 604 + ], + "score": 0.69, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 593, + 505, + 606 + ], + "score": 1.0, + "content": ", which guarantees the best possible", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "spans": [ + { + "bbox": [ + 106, + 605, + 505, + 617 + ], + "score": 1.0, + "content": "sample and communication complexities for the particular algorithm. The trade-off analysis presented", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 105, + 615, + 505, + 627 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 399, + 627 + ], + "score": 1.0, + "content": "in this section provides some useful guidelines for how to best select", + "type": "text" + }, + { + "bbox": [ + 400, + 616, + 406, + 626 + ], + "score": 0.61, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 615, + 425, + 627 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 426, + 616, + 433, + 626 + ], + "score": 0.72, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 433, + 615, + 505, + 627 + ], + "score": 1.0, + "content": "in practice. Our", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 105, + 626, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 325, + 640 + ], + "score": 1.0, + "content": "subsequent numerical results will also verify that if", + "type": "text" + }, + { + "bbox": [ + 325, + 627, + 331, + 636 + ], + "score": 0.72, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 331, + 626, + 344, + 640 + ], + "score": 1.0, + "content": "or", + "type": "text" + }, + { + "bbox": [ + 345, + 627, + 352, + 636 + ], + "score": 0.75, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 626, + 505, + 640 + ], + "score": 1.0, + "content": "are not chosen judiciously, then the", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 638, + 370, + 650 + ], + "spans": [ + { + "bbox": [ + 106, + 638, + 370, + 650 + ], + "score": 1.0, + "content": "practical performance of the algorithms can degrade significantly.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 46, + "bbox_fs": [ + 105, + 572, + 505, + 650 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 664, + 216, + 677 + ], + "lines": [ + { + "bbox": [ + 105, + 662, + 218, + 680 + ], + "spans": [ + { + "bbox": [ + 105, + 662, + 218, + 680 + ], + "score": 1.0, + "content": "4 Numerical results", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 50 + }, + { + "type": "text", + "bbox": [ + 107, + 688, + 504, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 689, + 506, + 701 + ], + "spans": [ + { + "bbox": [ + 105, + 689, + 506, + 701 + ], + "score": 1.0, + "content": "In this section, we validate the proposed STEM algorithm and compare its performance with the", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 105, + 699, + 506, + 712 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 377, + 711 + ], + "score": 1.0, + "content": "de facto standard FedAvg [11], and the algorithms stated in Table", + "type": "text" + }, + { + "bbox": [ + 378, + 700, + 388, + 712 + ], + "score": 0.71, + "content": "^ { 1 . }", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 699, + 506, + 711 + ], + "score": 1.0, + "content": "Note that instead of FedPD", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 711, + 506, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 326, + 724 + ], + "score": 1.0, + "content": "we include the performance comparison with FedDyn", + "type": "text" + }, + { + "bbox": [ + 326, + 711, + 343, + 722 + ], + "score": 0.64, + "content": "\\pmb { \\mathbb { B } } 6 \\|", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 711, + 506, + 724 + ], + "score": 1.0, + "content": "since they are known to be very closely", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 442, + 505, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 505, + 455 + ], + "score": 1.0, + "content": "related. The goal of our experiments are three-fold: (1) To show that STEM performs on par, if not", + "type": "text", + "cross_page": true + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 452, + 506, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 452, + 506, + 467 + ], + "score": 1.0, + "content": "better, compared to other algorithms in different heterogeneity settings, (2) there are multiple ways", + "type": "text", + "cross_page": true + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "score": 1.0, + "content": "to reach the desired solution accuracy, one can either choose a large batch size and perform only a", + "type": "text", + "cross_page": true + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 475, + 505, + 487 + ], + "spans": [ + { + "bbox": [ + 106, + 475, + 505, + 487 + ], + "score": 1.0, + "content": "few local updates or select a smaller batch size and perform multiple local updates, and finally, (3) if", + "type": "text", + "cross_page": true + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 486, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 505, + 500 + ], + "score": 1.0, + "content": "the local updates and the batch sizes are not chosen appropriately, the WNs might need to perform", + "type": "text", + "cross_page": true + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 496, + 506, + 510 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 506, + 510 + ], + "score": 1.0, + "content": "excessive computations to achieve the desired solution accuracy, thereby slowing down convergence.", + "type": "text", + "cross_page": true + } + ], + "index": 31 + } + ], + "index": 52, + "bbox_fs": [ + 105, + 689, + 506, + 724 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 70, + 306, + 185 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 70, + 306, + 185 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 70, + 306, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 70, + 306, + 185 + ], + "score": 0.977, + "html": "
AlgorithmTraining Acc.Testing Acc.
FedAvg57.657.1
FedProx59.158.5
FedDyn51.251.3
SCAFFOLD53.154.7
MIME56.155.1
FedGLOMO56.856.1
STEM58.557.4
", + "type": "table", + "image_path": "dcdc37cb3965642ea8fadecde08dc3a78e268deb8860ba74a94a052f25763bcd.jpg" + } + ] + } + ], + "index": 3.5, + "virtual_lines": [ + { + "bbox": [ + 106, + 70, + 306, + 84.375 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 84.375, + 306, + 98.75 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 98.75, + 306, + 113.125 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 113.125, + 306, + 127.5 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 106, + 127.5, + 306, + 141.875 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 106, + 141.875, + 306, + 156.25 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 106, + 156.25, + 306, + 170.625 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 106, + 170.625, + 306, + 185.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 106, + 193, + 287, + 226 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 192, + 287, + 206 + ], + "spans": [ + { + "bbox": [ + 106, + 192, + 287, + 206 + ], + "score": 1.0, + "content": "Table 3: Training and testing accuracy on", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 204, + 288, + 217 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 270, + 217 + ], + "score": 1.0, + "content": "CIFAR-10 dataset for high heterogeneity,", + "type": "text" + }, + { + "bbox": [ + 270, + 204, + 288, + 215 + ], + "score": 0.83, + "content": "b =", + "type": "inline_equation" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 215, + 169, + 227 + ], + "spans": [ + { + "bbox": [ + 106, + 215, + 140, + 227 + ], + "score": 1.0, + "content": "128 and", + "type": "text" + }, + { + "bbox": [ + 140, + 216, + 165, + 226 + ], + "score": 0.89, + "content": "I = 6", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 215, + 169, + 227 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + } + ], + "index": 6.25 + }, + { + "type": "table", + "bbox": [ + 323, + 70, + 523, + 185 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 323, + 70, + 523, + 185 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 323, + 70, + 523, + 185 + ], + "spans": [ + { + "bbox": [ + 323, + 70, + 523, + 185 + ], + "score": 0.977, + "html": "
AlgorithmTraining Acc.Testing Acc.
FedAvg40.139.2
FedProx43.543.2
FedDyn43.743.2
SCAFFOLD40.341.3
MIME32.132.1
FedGLOMO40.340.1
STEM44.543.8
", + "type": "table", + "image_path": "a8ee7b8ff04d3f8e652b073870044e3954f021e10e2ae94f2132f7ce72b56ed4.jpg" + } + ] + } + ], + "index": 14.0, + "virtual_lines": [ + { + "bbox": [ + 323, + 70, + 523, + 84.375 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 323, + 84.375, + 523, + 98.75 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 323, + 98.75, + 523, + 113.125 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 323, + 113.125, + 523, + 127.5 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 323, + 127.5, + 523, + 141.875 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 323, + 141.875, + 523, + 156.25 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 323, + 156.25, + 523, + 170.625 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 323, + 170.625, + 523, + 185.0 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 324, + 193, + 505, + 216 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 323, + 192, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 323, + 192, + 505, + 207 + ], + "score": 1.0, + "content": "Table 4: Training and testing accuracy on", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 325, + 204, + 409, + 216 + ], + "spans": [ + { + "bbox": [ + 325, + 204, + 409, + 216 + ], + "score": 1.0, + "content": "Shakespeare dataset.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + } + ], + "index": 16.75 + }, + { + "type": "image", + "bbox": [ + 114, + 254, + 484, + 389 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 114, + 254, + 484, + 389 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 114, + 254, + 484, + 389 + ], + "spans": [ + { + "bbox": [ + 114, + 254, + 484, + 389 + ], + "score": 0.967, + "type": "image", + "image_path": "1a3649642909a08bbdd852d92b8ec00817216ffd484e8ea8263d27aa9fb977f4.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 114, + 254, + 484, + 299.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 114, + 299.0, + 484, + 344.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 114, + 344.0, + 484, + 389.0 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 395, + 505, + 418 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 394, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 394, + 505, + 408 + ], + "score": 1.0, + "content": "Figure 2: Training loss and the testing accuracy for classification on MNIST data set against the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 406, + 459, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 431, + 419 + ], + "score": 1.0, + "content": "number of samples accessed at each WN for moderate heterogeneity setting with", + "type": "text" + }, + { + "bbox": [ + 431, + 407, + 455, + 416 + ], + "score": 0.89, + "content": "b = 8", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 406, + 459, + 419 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + } + ], + "index": 23.25 + }, + { + "type": "text", + "bbox": [ + 107, + 442, + 505, + 508 + ], + "lines": [ + { + "bbox": [ + 105, + 442, + 505, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 505, + 455 + ], + "score": 1.0, + "content": "related. The goal of our experiments are three-fold: (1) To show that STEM performs on par, if not", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 452, + 506, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 452, + 506, + 467 + ], + "score": 1.0, + "content": "better, compared to other algorithms in different heterogeneity settings, (2) there are multiple ways", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 105, + 464, + 506, + 477 + ], + "score": 1.0, + "content": "to reach the desired solution accuracy, one can either choose a large batch size and perform only a", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 475, + 505, + 487 + ], + "spans": [ + { + "bbox": [ + 106, + 475, + 505, + 487 + ], + "score": 1.0, + "content": "few local updates or select a smaller batch size and perform multiple local updates, and finally, (3) if", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 486, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 486, + 505, + 500 + ], + "score": 1.0, + "content": "the local updates and the batch sizes are not chosen appropriately, the WNs might need to perform", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 496, + 506, + 510 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 506, + 510 + ], + "score": 1.0, + "content": "excessive computations to achieve the desired solution accuracy, thereby slowing down convergence.", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 106, + 513, + 505, + 623 + ], + "lines": [ + { + "bbox": [ + 106, + 513, + 506, + 525 + ], + "spans": [ + { + "bbox": [ + 106, + 513, + 506, + 525 + ], + "score": 1.0, + "content": "Data and Parameter Settings: We compare the algorithms for image classification tasks on CIFAR-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 523, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 505, + 537 + ], + "score": 1.0, + "content": "10 and MNIST data sets with 100 WNs, and for next character prediction task on Shakespeare", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 534, + 506, + 548 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 136, + 548 + ], + "score": 1.0, + "content": "dataset", + "type": "text" + }, + { + "bbox": [ + 137, + 534, + 154, + 546 + ], + "score": 0.52, + "content": "\\pmb { \\Vert 3 7 } \\Vert", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 534, + 506, + 548 + ], + "score": 1.0, + "content": "with 143 WNs in the network. For both CIFAR-10 and MNIST, each WN implements a", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 546, + 505, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 505, + 558 + ], + "score": 1.0, + "content": "two-hidden-layer convolutional neural network (CNN) architecture followed by three linear layers for", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 556, + 504, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 556, + 504, + 569 + ], + "score": 1.0, + "content": "CIFAR-10 and two for MNIST. For CIFAR-10 (and MNIST) datset, we consider three settings with", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 568, + 505, + 581 + ], + "spans": [ + { + "bbox": [ + 106, + 568, + 505, + 581 + ], + "score": 1.0, + "content": "mild, moderate and high heterogeneity. For all the three settings, the data is partitioned into disjoint", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 578, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 505, + 591 + ], + "score": 1.0, + "content": "sets among the WNs. In the mild heterogeneity setting, the WNs have access to partitioned data from", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 590, + 506, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 506, + 603 + ], + "score": 1.0, + "content": "all the classes. In the moderate (resp. high) heterogeneity setting the data is partitioned such that", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 601, + 506, + 614 + ], + "spans": [ + { + "bbox": [ + 106, + 601, + 506, + 614 + ], + "score": 1.0, + "content": "each WN can access data from only 5 (resp. 2) out of 10 classes. For CIFAR-10 (resp. MNIST), each", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 610, + 507, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 507, + 624 + ], + "score": 1.0, + "content": "WN has access to 490 (resp. 540) samples for training and 90 (resp. 80) samples for testing purposes.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 36.5 + }, + { + "type": "text", + "bbox": [ + 107, + 628, + 505, + 694 + ], + "lines": [ + { + "bbox": [ + 105, + 627, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 505, + 640 + ], + "score": 1.0, + "content": "We also compare the performance of algorithms on a popular FL benchmarking dataset, Shakespeare", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 637, + 505, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 136, + 651 + ], + "score": 1.0, + "content": "dataset", + "type": "text" + }, + { + "bbox": [ + 137, + 638, + 154, + 650 + ], + "score": 0.76, + "content": " { \\mathbb { I } } ^ { \\smash { \\sum } }", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 637, + 318, + 651 + ], + "score": 1.0, + "content": ". For this task, we adopt the settings from", + "type": "text" + }, + { + "bbox": [ + 319, + 639, + 336, + 650 + ], + "score": 0.86, + "content": "\\dot { \\left[ \\left| 1 0 \\right| \\right] }", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 637, + 505, + 651 + ], + "score": 1.0, + "content": "and utilize a 2-Layer LSTM network with", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 650, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 106, + 650, + 505, + 661 + ], + "score": 1.0, + "content": "100 hidden units and an 8-D embedding layer at each WN. Each WN has access to 3616 samples", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 661, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 329, + 672 + ], + "score": 1.0, + "content": "on average, and the samples are randomly split into an", + "type": "text" + }, + { + "bbox": [ + 329, + 661, + 349, + 671 + ], + "score": 0.87, + "content": "80 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 661, + 421, + 672 + ], + "score": 1.0, + "content": "training set and a", + "type": "text" + }, + { + "bbox": [ + 422, + 661, + 442, + 671 + ], + "score": 0.87, + "content": "20 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 661, + 505, + 672 + ], + "score": 1.0, + "content": "testing set. We", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 671, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 505, + 684 + ], + "score": 1.0, + "content": "randomly sample 10 nodes out of 143 for the training purpose. All the experiments are implemented", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 104, + 682, + 448, + 695 + ], + "spans": [ + { + "bbox": [ + 104, + 682, + 448, + 695 + ], + "score": 1.0, + "content": "on a single NVIDIA Quadro RTX 5000 GPU. More details are provided in appendix.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 44.5 + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 503, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 347, + 713 + ], + "score": 1.0, + "content": "For the proposed STEM algorithm, recall that the step-size is", + "type": "text" + }, + { + "bbox": [ + 348, + 698, + 437, + 712 + ], + "score": 0.92, + "content": "\\eta _ { t } = \\bar { \\kappa } / ( w _ { t } + \\sigma ^ { 2 } t ) ^ { 1 / 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 698, + 505, + 713 + ], + "score": 1.0, + "content": "with momentum", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 710, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 194, + 723 + ], + "score": 1.0, + "content": "parameter defined as", + "type": "text" + }, + { + "bbox": [ + 194, + 710, + 233, + 723 + ], + "score": 0.93, + "content": "a _ { t } = { c } \\eta _ { t } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 233, + 711, + 505, + 723 + ], + "score": 1.0, + "content": ". The step-size is used to update the iterates while the momentum", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 48.5 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 106, + 70, + 306, + 185 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 106, + 70, + 306, + 185 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 70, + 306, + 185 + ], + "spans": [ + { + "bbox": [ + 106, + 70, + 306, + 185 + ], + "score": 0.977, + "html": "
AlgorithmTraining Acc.Testing Acc.
FedAvg57.657.1
FedProx59.158.5
FedDyn51.251.3
SCAFFOLD53.154.7
MIME56.155.1
FedGLOMO56.856.1
STEM58.557.4
", + "type": "table", + "image_path": "dcdc37cb3965642ea8fadecde08dc3a78e268deb8860ba74a94a052f25763bcd.jpg" + } + ] + } + ], + "index": 3.5, + "virtual_lines": [ + { + "bbox": [ + 106, + 70, + 306, + 84.375 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 106, + 84.375, + 306, + 98.75 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 106, + 98.75, + 306, + 113.125 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 106, + 113.125, + 306, + 127.5 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 106, + 127.5, + 306, + 141.875 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 106, + 141.875, + 306, + 156.25 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 106, + 156.25, + 306, + 170.625 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 106, + 170.625, + 306, + 185.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 106, + 193, + 287, + 226 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 192, + 287, + 206 + ], + "spans": [ + { + "bbox": [ + 106, + 192, + 287, + 206 + ], + "score": 1.0, + "content": "Table 3: Training and testing accuracy on", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 204, + 288, + 217 + ], + "spans": [ + { + "bbox": [ + 106, + 204, + 270, + 217 + ], + "score": 1.0, + "content": "CIFAR-10 dataset for high heterogeneity,", + "type": "text" + }, + { + "bbox": [ + 270, + 204, + 288, + 215 + ], + "score": 0.83, + "content": "b =", + "type": "inline_equation" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 215, + 169, + 227 + ], + "spans": [ + { + "bbox": [ + 106, + 215, + 140, + 227 + ], + "score": 1.0, + "content": "128 and", + "type": "text" + }, + { + "bbox": [ + 140, + 216, + 165, + 226 + ], + "score": 0.89, + "content": "I = 6", + "type": "inline_equation" + }, + { + "bbox": [ + 165, + 215, + 169, + 227 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + } + ], + "index": 6.25 + }, + { + "type": "table", + "bbox": [ + 323, + 70, + 523, + 185 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 323, + 70, + 523, + 185 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 323, + 70, + 523, + 185 + ], + "spans": [ + { + "bbox": [ + 323, + 70, + 523, + 185 + ], + "score": 0.977, + "html": "
AlgorithmTraining Acc.Testing Acc.
FedAvg40.139.2
FedProx43.543.2
FedDyn43.743.2
SCAFFOLD40.341.3
MIME32.132.1
FedGLOMO40.340.1
STEM44.543.8
", + "type": "table", + "image_path": "a8ee7b8ff04d3f8e652b073870044e3954f021e10e2ae94f2132f7ce72b56ed4.jpg" + } + ] + } + ], + "index": 14.0, + "virtual_lines": [ + { + "bbox": [ + 323, + 70, + 523, + 84.375 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 323, + 84.375, + 523, + 98.75 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 323, + 98.75, + 523, + 113.125 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 323, + 113.125, + 523, + 127.5 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 323, + 127.5, + 523, + 141.875 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 323, + 141.875, + 523, + 156.25 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 323, + 156.25, + 523, + 170.625 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 323, + 170.625, + 523, + 185.0 + ], + "spans": [], + "index": 18 + } + ] + }, + { + "type": "table_caption", + "bbox": [ + 324, + 193, + 505, + 216 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 323, + 192, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 323, + 192, + 505, + 207 + ], + "score": 1.0, + "content": "Table 4: Training and testing accuracy on", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 325, + 204, + 409, + 216 + ], + "spans": [ + { + "bbox": [ + 325, + 204, + 409, + 216 + ], + "score": 1.0, + "content": "Shakespeare dataset.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + } + ], + "index": 16.75 + }, + { + "type": "image", + "bbox": [ + 114, + 254, + 484, + 389 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 114, + 254, + 484, + 389 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 114, + 254, + 484, + 389 + ], + "spans": [ + { + "bbox": [ + 114, + 254, + 484, + 389 + ], + "score": 0.967, + "type": "image", + "image_path": "1a3649642909a08bbdd852d92b8ec00817216ffd484e8ea8263d27aa9fb977f4.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 114, + 254, + 484, + 299.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 114, + 299.0, + 484, + 344.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 114, + 344.0, + 484, + 389.0 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 395, + 505, + 418 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 394, + 505, + 408 + ], + "spans": [ + { + "bbox": [ + 105, + 394, + 505, + 408 + ], + "score": 1.0, + "content": "Figure 2: Training loss and the testing accuracy for classification on MNIST data set against the", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 406, + 459, + 419 + ], + "spans": [ + { + "bbox": [ + 106, + 406, + 431, + 419 + ], + "score": 1.0, + "content": "number of samples accessed at each WN for moderate heterogeneity setting with", + "type": "text" + }, + { + "bbox": [ + 431, + 407, + 455, + 416 + ], + "score": 0.89, + "content": "b = 8", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 406, + 459, + 419 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24.5 + } + ], + "index": 23.25 + }, + { + "type": "text", + "bbox": [ + 107, + 442, + 505, + 508 + ], + "lines": [], + "index": 28.5, + "bbox_fs": [ + 105, + 442, + 506, + 510 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 513, + 505, + 623 + ], + "lines": [ + { + "bbox": [ + 106, + 513, + 506, + 525 + ], + "spans": [ + { + "bbox": [ + 106, + 513, + 506, + 525 + ], + "score": 1.0, + "content": "Data and Parameter Settings: We compare the algorithms for image classification tasks on CIFAR-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 523, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 505, + 537 + ], + "score": 1.0, + "content": "10 and MNIST data sets with 100 WNs, and for next character prediction task on Shakespeare", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 534, + 506, + 548 + ], + "spans": [ + { + "bbox": [ + 106, + 534, + 136, + 548 + ], + "score": 1.0, + "content": "dataset", + "type": "text" + }, + { + "bbox": [ + 137, + 534, + 154, + 546 + ], + "score": 0.52, + "content": "\\pmb { \\Vert 3 7 } \\Vert", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 534, + 506, + 548 + ], + "score": 1.0, + "content": "with 143 WNs in the network. For both CIFAR-10 and MNIST, each WN implements a", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 546, + 505, + 558 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 505, + 558 + ], + "score": 1.0, + "content": "two-hidden-layer convolutional neural network (CNN) architecture followed by three linear layers for", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 556, + 504, + 569 + ], + "spans": [ + { + "bbox": [ + 106, + 556, + 504, + 569 + ], + "score": 1.0, + "content": "CIFAR-10 and two for MNIST. For CIFAR-10 (and MNIST) datset, we consider three settings with", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 568, + 505, + 581 + ], + "spans": [ + { + "bbox": [ + 106, + 568, + 505, + 581 + ], + "score": 1.0, + "content": "mild, moderate and high heterogeneity. For all the three settings, the data is partitioned into disjoint", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 578, + 505, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 578, + 505, + 591 + ], + "score": 1.0, + "content": "sets among the WNs. In the mild heterogeneity setting, the WNs have access to partitioned data from", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 590, + 506, + 603 + ], + "spans": [ + { + "bbox": [ + 105, + 590, + 506, + 603 + ], + "score": 1.0, + "content": "all the classes. In the moderate (resp. high) heterogeneity setting the data is partitioned such that", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 601, + 506, + 614 + ], + "spans": [ + { + "bbox": [ + 106, + 601, + 506, + 614 + ], + "score": 1.0, + "content": "each WN can access data from only 5 (resp. 2) out of 10 classes. For CIFAR-10 (resp. MNIST), each", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 610, + 507, + 624 + ], + "spans": [ + { + "bbox": [ + 105, + 610, + 507, + 624 + ], + "score": 1.0, + "content": "WN has access to 490 (resp. 540) samples for training and 90 (resp. 80) samples for testing purposes.", + "type": "text" + } + ], + "index": 41 + } + ], + "index": 36.5, + "bbox_fs": [ + 105, + 513, + 507, + 624 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 628, + 505, + 694 + ], + "lines": [ + { + "bbox": [ + 105, + 627, + 505, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 627, + 505, + 640 + ], + "score": 1.0, + "content": "We also compare the performance of algorithms on a popular FL benchmarking dataset, Shakespeare", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 637, + 505, + 651 + ], + "spans": [ + { + "bbox": [ + 105, + 637, + 136, + 651 + ], + "score": 1.0, + "content": "dataset", + "type": "text" + }, + { + "bbox": [ + 137, + 638, + 154, + 650 + ], + "score": 0.76, + "content": " { \\mathbb { I } } ^ { \\smash { \\sum } }", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 637, + 318, + 651 + ], + "score": 1.0, + "content": ". For this task, we adopt the settings from", + "type": "text" + }, + { + "bbox": [ + 319, + 639, + 336, + 650 + ], + "score": 0.86, + "content": "\\dot { \\left[ \\left| 1 0 \\right| \\right] }", + "type": "inline_equation" + }, + { + "bbox": [ + 337, + 637, + 505, + 651 + ], + "score": 1.0, + "content": "and utilize a 2-Layer LSTM network with", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 650, + 505, + 661 + ], + "spans": [ + { + "bbox": [ + 106, + 650, + 505, + 661 + ], + "score": 1.0, + "content": "100 hidden units and an 8-D embedding layer at each WN. Each WN has access to 3616 samples", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 661, + 505, + 672 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 329, + 672 + ], + "score": 1.0, + "content": "on average, and the samples are randomly split into an", + "type": "text" + }, + { + "bbox": [ + 329, + 661, + 349, + 671 + ], + "score": 0.87, + "content": "80 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 349, + 661, + 421, + 672 + ], + "score": 1.0, + "content": "training set and a", + "type": "text" + }, + { + "bbox": [ + 422, + 661, + 442, + 671 + ], + "score": 0.87, + "content": "20 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 661, + 505, + 672 + ], + "score": 1.0, + "content": "testing set. We", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 671, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 505, + 684 + ], + "score": 1.0, + "content": "randomly sample 10 nodes out of 143 for the training purpose. All the experiments are implemented", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 104, + 682, + 448, + 695 + ], + "spans": [ + { + "bbox": [ + 104, + 682, + 448, + 695 + ], + "score": 1.0, + "content": "on a single NVIDIA Quadro RTX 5000 GPU. More details are provided in appendix.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 44.5, + "bbox_fs": [ + 104, + 627, + 505, + 695 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 699, + 503, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 698, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 698, + 347, + 713 + ], + "score": 1.0, + "content": "For the proposed STEM algorithm, recall that the step-size is", + "type": "text" + }, + { + "bbox": [ + 348, + 698, + 437, + 712 + ], + "score": 0.92, + "content": "\\eta _ { t } = \\bar { \\kappa } / ( w _ { t } + \\sigma ^ { 2 } t ) ^ { 1 / 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 698, + 505, + 713 + ], + "score": 1.0, + "content": "with momentum", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 105, + 710, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 194, + 723 + ], + "score": 1.0, + "content": "parameter defined as", + "type": "text" + }, + { + "bbox": [ + 194, + 710, + 233, + 723 + ], + "score": 0.93, + "content": "a _ { t } = { c } \\eta _ { t } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 233, + 711, + 505, + 723 + ], + "score": 1.0, + "content": ". The step-size is used to update the iterates while the momentum", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 72, + 506, + 85 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 467, + 85 + ], + "score": 1.0, + "content": "parameter is used to construct the stochastic gradient estimate (cf. Algorithm 1 and Theorem", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 467, + 72, + 484, + 85 + ], + "score": 0.84, + "content": "\\boxed { 3 . 1 }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 485, + 72, + 506, + 85 + ], + "score": 1.0, + "content": ". For", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 200, + 95 + ], + "score": 1.0, + "content": "the experiments, we set", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 201, + 83, + 254, + 94 + ], + "score": 0.91, + "content": "w _ { t } = \\sigma ^ { 2 } = 1", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 255, + 82, + 272, + 95 + ], + "score": 1.0, + "content": "and", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 273, + 83, + 310, + 95 + ], + "score": 0.93, + "content": "c \\doteq \\bar { c } / \\bar { \\kappa } ^ { 2 }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 311, + 82, + 361, + 95 + ], + "score": 1.0, + "content": "and tune for", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 362, + 83, + 432, + 96 + ], + "score": 0.91, + "content": "\\bar { \\kappa } \\in [ 1 0 ^ { - 1 } , \\overline { { 1 0 } } ^ { - 2 } ]", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 432, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "for the CIFAR-10", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 92, + 506, + 108 + ], + "spans": [ + { + "bbox": [ + 105, + 92, + 165, + 108 + ], + "score": 1.0, + "content": "dataset and for", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 166, + 94, + 278, + 106 + ], + "score": 0.82, + "content": "{ \\bar { \\kappa } } \\in \\{ 1 0 ^ { 1 } , 1 0 ^ { 0 } , 1 0 ^ { - 1 } , 1 0 ^ { - 2 } \\}", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 278, + 92, + 506, + 108 + ], + "score": 1.0, + "content": "for the Shakespeare dataset. For both the datasets we tune", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 105, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 120, + 118 + ], + "score": 1.0, + "content": "for", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 120, + 106, + 127, + 115 + ], + "score": 0.52, + "content": "\\bar { c }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 127, + 105, + 257, + 118 + ], + "score": 1.0, + "content": "in the range [1, 10]. For FedProx", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 258, + 106, + 275, + 117 + ], + "score": 0.74, + "content": "[ \\equiv ]", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 275, + 105, + 327, + 118 + ], + "score": 1.0, + "content": "and FedDyn", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 327, + 105, + 344, + 117 + ], + "score": 0.84, + "content": "\\lVert \\bar { 3 6 } \\rVert", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 345, + 105, + 505, + 118 + ], + "score": 1.0, + "content": "we choose the regularization constant to", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 116, + 505, + 129 + ], + "spans": [ + { + "bbox": [ + 106, + 116, + 312, + 129 + ], + "score": 1.0, + "content": "be 0.1. The momentum parameters for FedGLOMO", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 312, + 117, + 330, + 127 + ], + "score": 0.75, + "content": "\\pm \\textcircled { 1 8 } \\textcircled { 1 }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 330, + 116, + 376, + 129 + ], + "score": 1.0, + "content": "and MIME", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 376, + 116, + 394, + 127 + ], + "score": 0.78, + "content": "\\mathbb { \\lVert 1 7 \\rVert }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 394, + 116, + 505, + 129 + ], + "score": 1.0, + "content": "are set based on the choices", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 127, + 506, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 449, + 139 + ], + "score": 1.0, + "content": "given in the respective papers. Specifically, for FedGLOMO we choose the parameter", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 449, + 127, + 487, + 138 + ], + "score": 0.91, + "content": "\\beta _ { k } = 0 . 2", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 488, + 127, + 506, + 139 + ], + "score": 1.0, + "content": "and", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 137, + 506, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 485, + 151 + ], + "score": 1.0, + "content": "design the momentum gradient using a damping factor given in Appendix A.4 of FedGLOMO", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 486, + 138, + 503, + 149 + ], + "score": 0.5, + "content": "\\mathbb { \\left[ \\left[ 8 \\right] \\right] }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 503, + 137, + 506, + 151 + ], + "score": 1.0, + "content": ".", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 149, + 505, + 162 + ], + "spans": [ + { + "bbox": [ + 105, + 149, + 505, + 162 + ], + "score": 1.0, + "content": "Moreover, for MIME we choose the momentum parameter as 0.9. For the rest of the algorithms", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 159, + 494, + 172 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 396, + 172 + ], + "score": 1.0, + "content": "(including FedAvg and SCAFFOLD), the step-size is tuned from the set", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 396, + 159, + 490, + 172 + ], + "score": 0.93, + "content": "\\{ 1 0 ^ { 1 } , 1 0 ^ { 0 } , 1 0 ^ { - 1 } , \\hat { 1 0 } ^ { - 2 } \\}", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 490, + 159, + 494, + 172 + ], + "score": 1.0, + "content": ".", + "type": "text", + "cross_page": true + } + ], + "index": 8 + } + ], + "index": 48.5, + "bbox_fs": [ + 105, + 698, + 505, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 72, + 505, + 172 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 506, + 85 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 467, + 85 + ], + "score": 1.0, + "content": "parameter is used to construct the stochastic gradient estimate (cf. Algorithm 1 and Theorem", + "type": "text" + }, + { + "bbox": [ + 467, + 72, + 484, + 85 + ], + "score": 0.84, + "content": "\\boxed { 3 . 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 485, + 72, + 506, + 85 + ], + "score": 1.0, + "content": ". For", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 82, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 82, + 200, + 95 + ], + "score": 1.0, + "content": "the experiments, we set", + "type": "text" + }, + { + "bbox": [ + 201, + 83, + 254, + 94 + ], + "score": 0.91, + "content": "w _ { t } = \\sigma ^ { 2 } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 82, + 272, + 95 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 273, + 83, + 310, + 95 + ], + "score": 0.93, + "content": "c \\doteq \\bar { c } / \\bar { \\kappa } ^ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 311, + 82, + 361, + 95 + ], + "score": 1.0, + "content": "and tune for", + "type": "text" + }, + { + "bbox": [ + 362, + 83, + 432, + 96 + ], + "score": 0.91, + "content": "\\bar { \\kappa } \\in [ 1 0 ^ { - 1 } , \\overline { { 1 0 } } ^ { - 2 } ]", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 82, + 506, + 95 + ], + "score": 1.0, + "content": "for the CIFAR-10", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 92, + 506, + 108 + ], + "spans": [ + { + "bbox": [ + 105, + 92, + 165, + 108 + ], + "score": 1.0, + "content": "dataset and for", + "type": "text" + }, + { + "bbox": [ + 166, + 94, + 278, + 106 + ], + "score": 0.82, + "content": "{ \\bar { \\kappa } } \\in \\{ 1 0 ^ { 1 } , 1 0 ^ { 0 } , 1 0 ^ { - 1 } , 1 0 ^ { - 2 } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 278, + 92, + 506, + 108 + ], + "score": 1.0, + "content": "for the Shakespeare dataset. For both the datasets we tune", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 105, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 105, + 120, + 118 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 120, + 106, + 127, + 115 + ], + "score": 0.52, + "content": "\\bar { c }", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 105, + 257, + 118 + ], + "score": 1.0, + "content": "in the range [1, 10]. For FedProx", + "type": "text" + }, + { + "bbox": [ + 258, + 106, + 275, + 117 + ], + "score": 0.74, + "content": "[ \\equiv ]", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 105, + 327, + 118 + ], + "score": 1.0, + "content": "and FedDyn", + "type": "text" + }, + { + "bbox": [ + 327, + 105, + 344, + 117 + ], + "score": 0.84, + "content": "\\lVert \\bar { 3 6 } \\rVert", + "type": "inline_equation" + }, + { + "bbox": [ + 345, + 105, + 505, + 118 + ], + "score": 1.0, + "content": "we choose the regularization constant to", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 116, + 505, + 129 + ], + "spans": [ + { + "bbox": [ + 106, + 116, + 312, + 129 + ], + "score": 1.0, + "content": "be 0.1. The momentum parameters for FedGLOMO", + "type": "text" + }, + { + "bbox": [ + 312, + 117, + 330, + 127 + ], + "score": 0.75, + "content": "\\pm \\textcircled { 1 8 } \\textcircled { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 116, + 376, + 129 + ], + "score": 1.0, + "content": "and MIME", + "type": "text" + }, + { + "bbox": [ + 376, + 116, + 394, + 127 + ], + "score": 0.78, + "content": "\\mathbb { \\lVert 1 7 \\rVert }", + "type": "inline_equation" + }, + { + "bbox": [ + 394, + 116, + 505, + 129 + ], + "score": 1.0, + "content": "are set based on the choices", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 127, + 506, + 139 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 449, + 139 + ], + "score": 1.0, + "content": "given in the respective papers. Specifically, for FedGLOMO we choose the parameter", + "type": "text" + }, + { + "bbox": [ + 449, + 127, + 487, + 138 + ], + "score": 0.91, + "content": "\\beta _ { k } = 0 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 488, + 127, + 506, + 139 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 137, + 506, + 151 + ], + "spans": [ + { + "bbox": [ + 105, + 137, + 485, + 151 + ], + "score": 1.0, + "content": "design the momentum gradient using a damping factor given in Appendix A.4 of FedGLOMO", + "type": "text" + }, + { + "bbox": [ + 486, + 138, + 503, + 149 + ], + "score": 0.5, + "content": "\\mathbb { \\left[ \\left[ 8 \\right] \\right] }", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 137, + 506, + 151 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 149, + 505, + 162 + ], + "spans": [ + { + "bbox": [ + 105, + 149, + 505, + 162 + ], + "score": 1.0, + "content": "Moreover, for MIME we choose the momentum parameter as 0.9. For the rest of the algorithms", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 159, + 494, + 172 + ], + "spans": [ + { + "bbox": [ + 105, + 159, + 396, + 172 + ], + "score": 1.0, + "content": "(including FedAvg and SCAFFOLD), the step-size is tuned from the set", + "type": "text" + }, + { + "bbox": [ + 396, + 159, + 490, + 172 + ], + "score": 0.93, + "content": "\\{ 1 0 ^ { 1 } , 1 0 ^ { 0 } , 1 0 ^ { - 1 } , \\hat { 1 0 } ^ { - 2 } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 490, + 159, + 494, + 172 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 106, + 182, + 506, + 390 + ], + "lines": [ + { + "bbox": [ + 105, + 182, + 506, + 196 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 506, + 196 + ], + "score": 1.0, + "content": "Discussion: We evaluate the training and testing performance of STEM against multiple algorithms", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 193, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 490, + 207 + ], + "score": 1.0, + "content": "for different heterogeneity settings, minibatch sizes, and number of local updates. In Tables", + "type": "text" + }, + { + "bbox": [ + 490, + 193, + 505, + 207 + ], + "score": 0.27, + "content": "\\bigstar", + "type": "inline_equation" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 204, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 105, + 204, + 119, + 217 + ], + "score": 0.33, + "content": "\\boxed { 2 \\mathbf { b } }", + "type": "inline_equation" + }, + { + "bbox": [ + 120, + 205, + 136, + 217 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 136, + 204, + 145, + 217 + ], + "score": 0.67, + "content": "\\bigstar ,", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 205, + 506, + 217 + ], + "score": 1.0, + "content": "we compare the training and testing accuracy of STEM to that of other algorithms on", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 215, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 104, + 215, + 506, + 228 + ], + "score": 1.0, + "content": "the CIFAR-10 dataset. Specific, heterogeneity settings, the choices of minibatches, and number of", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 226, + 505, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 505, + 239 + ], + "score": 1.0, + "content": "local updates are stated along with the tables. Note that STEM performs uniformly well under all", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 237, + 506, + 250 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 279, + 250 + ], + "score": 1.0, + "content": "the conditions. Moreover, note from Table", + "type": "text" + }, + { + "bbox": [ + 279, + 237, + 294, + 250 + ], + "score": 0.59, + "content": "\\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 237, + 506, + 250 + ], + "score": 1.0, + "content": "that FedGLOMO diverges once the number of local", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 248, + 506, + 262 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 265, + 262 + ], + "score": 1.0, + "content": "updates are high. Also, note from Table", + "type": "text" + }, + { + "bbox": [ + 266, + 248, + 275, + 261 + ], + "score": 0.29, + "content": "\\textcircled { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 248, + 506, + 262 + ], + "score": 1.0, + "content": "that FedProx and STEM adapt well to high heterogeneity.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 258, + 507, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 258, + 436, + 273 + ], + "score": 1.0, + "content": "Finally, with the next set of experiments we emphasize the importance of choosing", + "type": "text" + }, + { + "bbox": [ + 436, + 260, + 442, + 270 + ], + "score": 0.49, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 258, + 459, + 273 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 460, + 260, + 466, + 269 + ], + "score": 0.71, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 258, + 507, + 273 + ], + "score": 1.0, + "content": "carefully.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 269, + 506, + 283 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 144, + 283 + ], + "score": 1.0, + "content": "In Figure", + "type": "text" + }, + { + "bbox": [ + 145, + 269, + 155, + 283 + ], + "score": 0.63, + "content": "\\bigstar ,", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 270, + 506, + 283 + ], + "score": 1.0, + "content": "we compare the training and testing performance of STEM, FedAvg and SCAFFOLD,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 281, + 506, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 506, + 293 + ], + "score": 1.0, + "content": "against the number of samples accessed at each WN for the classification task on MNIST dataset with", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 292, + 506, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 232, + 305 + ], + "score": 1.0, + "content": "moderate heterogeneity. We fix", + "type": "text" + }, + { + "bbox": [ + 232, + 292, + 257, + 302 + ], + "score": 0.89, + "content": "b = 8", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 292, + 473, + 305 + ], + "score": 1.0, + "content": "and conduct experiments under two settings, one with", + "type": "text" + }, + { + "bbox": [ + 473, + 292, + 502, + 302 + ], + "score": 0.9, + "content": "I = 6 7", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 292, + 506, + 305 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 303, + 506, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 183, + 316 + ], + "score": 1.0, + "content": "and the other with", + "type": "text" + }, + { + "bbox": [ + 183, + 303, + 219, + 313 + ], + "score": 0.89, + "content": "I = 5 3 6", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 303, + 506, + 316 + ], + "score": 1.0, + "content": "local updates at each WN. Note that although a large number of local", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 314, + 506, + 327 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 506, + 327 + ], + "score": 1.0, + "content": "updates might lead to fewer communication rounds but it can make the sample complexity extremely", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 324, + 506, + 338 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 247, + 338 + ], + "score": 1.0, + "content": "high as is demonstrated by Figure", + "type": "text" + }, + { + "bbox": [ + 247, + 324, + 258, + 337 + ], + "score": 0.74, + "content": "2 .", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 325, + 345, + 338 + ], + "score": 1.0, + "content": "For example, Figure", + "type": "text" + }, + { + "bbox": [ + 345, + 325, + 355, + 337 + ], + "score": 0.73, + "content": "\\bigtriangledown", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 325, + 506, + 338 + ], + "score": 1.0, + "content": "shows that to reach testing accuracy", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 335, + 506, + 349 + ], + "spans": [ + { + "bbox": [ + 104, + 335, + 118, + 349 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 336, + 160, + 347 + ], + "score": 0.91, + "content": "9 6 - 9 7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 335, + 182, + 349 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 182, + 336, + 213, + 346 + ], + "score": 0.88, + "content": "I = 6 7", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 335, + 345, + 349 + ], + "score": 1.0, + "content": ", STEM requires approximately", + "type": "text" + }, + { + "bbox": [ + 345, + 336, + 399, + 347 + ], + "score": 0.83, + "content": "5 0 0 0 - 6 0 0 0", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 335, + 506, + 349 + ], + "score": 1.0, + "content": "samples, in contrast with", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 107, + 347, + 506, + 359 + ], + "spans": [ + { + "bbox": [ + 107, + 347, + 141, + 357 + ], + "score": 0.89, + "content": "I = 5 3 6", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 347, + 506, + 359 + ], + "score": 1.0, + "content": "it requires more than 25000 samples at each WN. Similar behavior can be observed if we fix", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 357, + 506, + 370 + ], + "spans": [ + { + "bbox": [ + 106, + 358, + 131, + 368 + ], + "score": 0.88, + "content": "I > 1", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 357, + 506, + 370 + ], + "score": 1.0, + "content": "and increase the local batch sizes. This implies not choosing the local updates and the batch", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 369, + 506, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 506, + 381 + ], + "score": 1.0, + "content": "sizes judiciously might lead to increased sample complexity. Additional experiments are included in", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 380, + 472, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 472, + 392 + ], + "score": 1.0, + "content": "the supplementary material to further evaluate the performance of the proposed algorithms.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 18 + }, + { + "type": "title", + "bbox": [ + 107, + 406, + 165, + 419 + ], + "lines": [ + { + "bbox": [ + 106, + 405, + 167, + 421 + ], + "spans": [ + { + "bbox": [ + 106, + 405, + 167, + 421 + ], + "score": 1.0, + "content": "Conclusion", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 430, + 506, + 543 + ], + "lines": [ + { + "bbox": [ + 105, + 430, + 507, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 507, + 444 + ], + "score": 1.0, + "content": "In this work, we proposed a novel algorithm STEM, for distributed stochastic non-convex optimiza-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 442, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 370, + 456 + ], + "score": 1.0, + "content": "tion with applications to FL. We showed that STEM reaches an", + "type": "text" + }, + { + "bbox": [ + 370, + 445, + 376, + 453 + ], + "score": 0.68, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 442, + 465, + 456 + ], + "score": 1.0, + "content": "-stationary point with", + "type": "text" + }, + { + "bbox": [ + 465, + 442, + 505, + 455 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 453, + 506, + 468 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 506, + 468 + ], + "score": 1.0, + "content": "sample complexity while achieving linear speed-up with the number of WNs. Moreover, the algo-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 465, + 506, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 301, + 479 + ], + "score": 1.0, + "content": "rithm achieves a communication complexity of", + "type": "text" + }, + { + "bbox": [ + 302, + 465, + 333, + 479 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 466, + 506, + 479 + ], + "score": 1.0, + "content": ". We established a (optimal) trade-off that", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 477, + 506, + 490 + ], + "spans": [ + { + "bbox": [ + 106, + 477, + 506, + 490 + ], + "score": 1.0, + "content": "allows interpolation between varying choices of local updates and the batch sizes at each WN while", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 488, + 506, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 506, + 501 + ], + "score": 1.0, + "content": "maintaining (near optimal) sample and communication complexities. We showed that FedAvg (a.k.a", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 499, + 506, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 499, + 506, + 512 + ], + "score": 1.0, + "content": "LocalSGD) also exhibits a similar trade-off while achieving worse complexities. Our results provide", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 510, + 506, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 506, + 523 + ], + "score": 1.0, + "content": "guidelines to carefully choose the number of local updates, update directions, and minibatch sizes to", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 520, + 506, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 520, + 506, + 534 + ], + "score": 1.0, + "content": "achieve the best performance. The future directions of this work include developing lower bounds on", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 532, + 489, + 544 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 489, + 544 + ], + "score": 1.0, + "content": "communication complexity that establishes the tightness of the analysis conducted in this work.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 33.5 + }, + { + "type": "title", + "bbox": [ + 108, + 558, + 202, + 572 + ], + "lines": [ + { + "bbox": [ + 106, + 557, + 204, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 557, + 204, + 574 + ], + "score": 1.0, + "content": "Acknowledgement", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 583, + 505, + 649 + ], + "lines": [ + { + "bbox": [ + 105, + 583, + 506, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 506, + 597 + ], + "score": 1.0, + "content": "We thank the anonymous reviewers for their valuable comments and suggestions. The work of", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 593, + 506, + 609 + ], + "spans": [ + { + "bbox": [ + 104, + 593, + 506, + 609 + ], + "score": 1.0, + "content": "Prashant Khanduri and Mingyi Hong was supported by NSF grant CMMI-1727757, AFOSR grant", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 604, + 506, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 506, + 619 + ], + "score": 1.0, + "content": "19RT0424 and ARO grant W911NF-19-1-0247. The work of Mingyi Hong was also supported by", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "score": 1.0, + "content": "an IBM Faculty Research award. The work of Jia Liu has been supported in part by NSF grants", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 627, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 506, + 640 + ], + "score": 1.0, + "content": "CAREER CNS-2110259, CNS-2112471, CNS-2102233, CCF-2110252, ECCS-2140277, and a", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 638, + 241, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 241, + 650 + ], + "score": 1.0, + "content": "Google Faculty Research Award.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 42.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 301, + 742, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 72, + 505, + 172 + ], + "lines": [], + "index": 4, + "bbox_fs": [ + 105, + 72, + 506, + 172 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 182, + 506, + 390 + ], + "lines": [ + { + "bbox": [ + 105, + 182, + 506, + 196 + ], + "spans": [ + { + "bbox": [ + 105, + 182, + 506, + 196 + ], + "score": 1.0, + "content": "Discussion: We evaluate the training and testing performance of STEM against multiple algorithms", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 193, + 505, + 207 + ], + "spans": [ + { + "bbox": [ + 105, + 194, + 490, + 207 + ], + "score": 1.0, + "content": "for different heterogeneity settings, minibatch sizes, and number of local updates. In Tables", + "type": "text" + }, + { + "bbox": [ + 490, + 193, + 505, + 207 + ], + "score": 0.27, + "content": "\\bigstar", + "type": "inline_equation" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 204, + 506, + 217 + ], + "spans": [ + { + "bbox": [ + 105, + 204, + 119, + 217 + ], + "score": 0.33, + "content": "\\boxed { 2 \\mathbf { b } }", + "type": "inline_equation" + }, + { + "bbox": [ + 120, + 205, + 136, + 217 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 136, + 204, + 145, + 217 + ], + "score": 0.67, + "content": "\\bigstar ,", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 205, + 506, + 217 + ], + "score": 1.0, + "content": "we compare the training and testing accuracy of STEM to that of other algorithms on", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 104, + 215, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 104, + 215, + 506, + 228 + ], + "score": 1.0, + "content": "the CIFAR-10 dataset. Specific, heterogeneity settings, the choices of minibatches, and number of", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 226, + 505, + 239 + ], + "spans": [ + { + "bbox": [ + 105, + 226, + 505, + 239 + ], + "score": 1.0, + "content": "local updates are stated along with the tables. Note that STEM performs uniformly well under all", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 237, + 506, + 250 + ], + "spans": [ + { + "bbox": [ + 106, + 237, + 279, + 250 + ], + "score": 1.0, + "content": "the conditions. Moreover, note from Table", + "type": "text" + }, + { + "bbox": [ + 279, + 237, + 294, + 250 + ], + "score": 0.59, + "content": "\\bigstar", + "type": "inline_equation" + }, + { + "bbox": [ + 294, + 237, + 506, + 250 + ], + "score": 1.0, + "content": "that FedGLOMO diverges once the number of local", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 248, + 506, + 262 + ], + "spans": [ + { + "bbox": [ + 105, + 248, + 265, + 262 + ], + "score": 1.0, + "content": "updates are high. Also, note from Table", + "type": "text" + }, + { + "bbox": [ + 266, + 248, + 275, + 261 + ], + "score": 0.29, + "content": "\\textcircled { 3 }", + "type": "inline_equation" + }, + { + "bbox": [ + 275, + 248, + 506, + 262 + ], + "score": 1.0, + "content": "that FedProx and STEM adapt well to high heterogeneity.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 258, + 507, + 273 + ], + "spans": [ + { + "bbox": [ + 105, + 258, + 436, + 273 + ], + "score": 1.0, + "content": "Finally, with the next set of experiments we emphasize the importance of choosing", + "type": "text" + }, + { + "bbox": [ + 436, + 260, + 442, + 270 + ], + "score": 0.49, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 442, + 258, + 459, + 273 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 460, + 260, + 466, + 269 + ], + "score": 0.71, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 258, + 507, + 273 + ], + "score": 1.0, + "content": "carefully.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 269, + 506, + 283 + ], + "spans": [ + { + "bbox": [ + 105, + 270, + 144, + 283 + ], + "score": 1.0, + "content": "In Figure", + "type": "text" + }, + { + "bbox": [ + 145, + 269, + 155, + 283 + ], + "score": 0.63, + "content": "\\bigstar ,", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 270, + 506, + 283 + ], + "score": 1.0, + "content": "we compare the training and testing performance of STEM, FedAvg and SCAFFOLD,", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 281, + 506, + 293 + ], + "spans": [ + { + "bbox": [ + 105, + 281, + 506, + 293 + ], + "score": 1.0, + "content": "against the number of samples accessed at each WN for the classification task on MNIST dataset with", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 292, + 506, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 292, + 232, + 305 + ], + "score": 1.0, + "content": "moderate heterogeneity. We fix", + "type": "text" + }, + { + "bbox": [ + 232, + 292, + 257, + 302 + ], + "score": 0.89, + "content": "b = 8", + "type": "inline_equation" + }, + { + "bbox": [ + 257, + 292, + 473, + 305 + ], + "score": 1.0, + "content": "and conduct experiments under two settings, one with", + "type": "text" + }, + { + "bbox": [ + 473, + 292, + 502, + 302 + ], + "score": 0.9, + "content": "I = 6 7", + "type": "inline_equation" + }, + { + "bbox": [ + 503, + 292, + 506, + 305 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 303, + 506, + 316 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 183, + 316 + ], + "score": 1.0, + "content": "and the other with", + "type": "text" + }, + { + "bbox": [ + 183, + 303, + 219, + 313 + ], + "score": 0.89, + "content": "I = 5 3 6", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 303, + 506, + 316 + ], + "score": 1.0, + "content": "local updates at each WN. Note that although a large number of local", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 314, + 506, + 327 + ], + "spans": [ + { + "bbox": [ + 105, + 314, + 506, + 327 + ], + "score": 1.0, + "content": "updates might lead to fewer communication rounds but it can make the sample complexity extremely", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 324, + 506, + 338 + ], + "spans": [ + { + "bbox": [ + 105, + 325, + 247, + 338 + ], + "score": 1.0, + "content": "high as is demonstrated by Figure", + "type": "text" + }, + { + "bbox": [ + 247, + 324, + 258, + 337 + ], + "score": 0.74, + "content": "2 .", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 325, + 345, + 338 + ], + "score": 1.0, + "content": "For example, Figure", + "type": "text" + }, + { + "bbox": [ + 345, + 325, + 355, + 337 + ], + "score": 0.73, + "content": "\\bigtriangledown", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 325, + 506, + 338 + ], + "score": 1.0, + "content": "shows that to reach testing accuracy", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 104, + 335, + 506, + 349 + ], + "spans": [ + { + "bbox": [ + 104, + 335, + 118, + 349 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 336, + 160, + 347 + ], + "score": 0.91, + "content": "9 6 - 9 7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 335, + 182, + 349 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 182, + 336, + 213, + 346 + ], + "score": 0.88, + "content": "I = 6 7", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 335, + 345, + 349 + ], + "score": 1.0, + "content": ", STEM requires approximately", + "type": "text" + }, + { + "bbox": [ + 345, + 336, + 399, + 347 + ], + "score": 0.83, + "content": "5 0 0 0 - 6 0 0 0", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 335, + 506, + 349 + ], + "score": 1.0, + "content": "samples, in contrast with", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 107, + 347, + 506, + 359 + ], + "spans": [ + { + "bbox": [ + 107, + 347, + 141, + 357 + ], + "score": 0.89, + "content": "I = 5 3 6", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 347, + 506, + 359 + ], + "score": 1.0, + "content": "it requires more than 25000 samples at each WN. Similar behavior can be observed if we fix", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 357, + 506, + 370 + ], + "spans": [ + { + "bbox": [ + 106, + 358, + 131, + 368 + ], + "score": 0.88, + "content": "I > 1", + "type": "inline_equation" + }, + { + "bbox": [ + 132, + 357, + 506, + 370 + ], + "score": 1.0, + "content": "and increase the local batch sizes. This implies not choosing the local updates and the batch", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 369, + 506, + 381 + ], + "spans": [ + { + "bbox": [ + 105, + 369, + 506, + 381 + ], + "score": 1.0, + "content": "sizes judiciously might lead to increased sample complexity. Additional experiments are included in", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 380, + 472, + 392 + ], + "spans": [ + { + "bbox": [ + 106, + 380, + 472, + 392 + ], + "score": 1.0, + "content": "the supplementary material to further evaluate the performance of the proposed algorithms.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 18, + "bbox_fs": [ + 104, + 182, + 507, + 392 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 406, + 165, + 419 + ], + "lines": [ + { + "bbox": [ + 106, + 405, + 167, + 421 + ], + "spans": [ + { + "bbox": [ + 106, + 405, + 167, + 421 + ], + "score": 1.0, + "content": "Conclusion", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 106, + 430, + 506, + 543 + ], + "lines": [ + { + "bbox": [ + 105, + 430, + 507, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 430, + 507, + 444 + ], + "score": 1.0, + "content": "In this work, we proposed a novel algorithm STEM, for distributed stochastic non-convex optimiza-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 442, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 370, + 456 + ], + "score": 1.0, + "content": "tion with applications to FL. We showed that STEM reaches an", + "type": "text" + }, + { + "bbox": [ + 370, + 445, + 376, + 453 + ], + "score": 0.68, + "content": "\\epsilon", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 442, + 465, + 456 + ], + "score": 1.0, + "content": "-stationary point with", + "type": "text" + }, + { + "bbox": [ + 465, + 442, + 505, + 455 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal O } ( \\epsilon ^ { - 3 / 2 } )", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 453, + 506, + 468 + ], + "spans": [ + { + "bbox": [ + 105, + 453, + 506, + 468 + ], + "score": 1.0, + "content": "sample complexity while achieving linear speed-up with the number of WNs. Moreover, the algo-", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 465, + 506, + 479 + ], + "spans": [ + { + "bbox": [ + 106, + 466, + 301, + 479 + ], + "score": 1.0, + "content": "rithm achieves a communication complexity of", + "type": "text" + }, + { + "bbox": [ + 302, + 465, + 333, + 479 + ], + "score": 0.93, + "content": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 466, + 506, + 479 + ], + "score": 1.0, + "content": ". We established a (optimal) trade-off that", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 477, + 506, + 490 + ], + "spans": [ + { + "bbox": [ + 106, + 477, + 506, + 490 + ], + "score": 1.0, + "content": "allows interpolation between varying choices of local updates and the batch sizes at each WN while", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 488, + 506, + 501 + ], + "spans": [ + { + "bbox": [ + 105, + 488, + 506, + 501 + ], + "score": 1.0, + "content": "maintaining (near optimal) sample and communication complexities. We showed that FedAvg (a.k.a", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 499, + 506, + 512 + ], + "spans": [ + { + "bbox": [ + 105, + 499, + 506, + 512 + ], + "score": 1.0, + "content": "LocalSGD) also exhibits a similar trade-off while achieving worse complexities. Our results provide", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 105, + 510, + 506, + 523 + ], + "spans": [ + { + "bbox": [ + 105, + 510, + 506, + 523 + ], + "score": 1.0, + "content": "guidelines to carefully choose the number of local updates, update directions, and minibatch sizes to", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 520, + 506, + 534 + ], + "spans": [ + { + "bbox": [ + 106, + 520, + 506, + 534 + ], + "score": 1.0, + "content": "achieve the best performance. The future directions of this work include developing lower bounds on", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 532, + 489, + 544 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 489, + 544 + ], + "score": 1.0, + "content": "communication complexity that establishes the tightness of the analysis conducted in this work.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 33.5, + "bbox_fs": [ + 105, + 430, + 507, + 544 + ] + }, + { + "type": "title", + "bbox": [ + 108, + 558, + 202, + 572 + ], + "lines": [ + { + "bbox": [ + 106, + 557, + 204, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 557, + 204, + 574 + ], + "score": 1.0, + "content": "Acknowledgement", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 583, + 505, + 649 + ], + "lines": [ + { + "bbox": [ + 105, + 583, + 506, + 597 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 506, + 597 + ], + "score": 1.0, + "content": "We thank the anonymous reviewers for their valuable comments and suggestions. The work of", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 104, + 593, + 506, + 609 + ], + "spans": [ + { + "bbox": [ + 104, + 593, + 506, + 609 + ], + "score": 1.0, + "content": "Prashant Khanduri and Mingyi Hong was supported by NSF grant CMMI-1727757, AFOSR grant", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 604, + 506, + 619 + ], + "spans": [ + { + "bbox": [ + 105, + 604, + 506, + 619 + ], + "score": 1.0, + "content": "19RT0424 and ARO grant W911NF-19-1-0247. The work of Mingyi Hong was also supported by", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "spans": [ + { + "bbox": [ + 105, + 615, + 505, + 629 + ], + "score": 1.0, + "content": "an IBM Faculty Research award. The work of Jia Liu has been supported in part by NSF grants", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 627, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 106, + 627, + 506, + 640 + ], + "score": 1.0, + "content": "CAREER CNS-2110259, CNS-2112471, CNS-2102233, CCF-2110252, ECCS-2140277, and a", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 638, + 241, + 650 + ], + "spans": [ + { + "bbox": [ + 105, + 638, + 241, + 650 + ], + "score": 1.0, + "content": "Google Faculty Research Award.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 42.5, + "bbox_fs": [ + 104, + 583, + 506, + 650 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 72, + 507, + 729 + ], + "lines": [ + { + "bbox": [ + 106, + 70, + 165, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 70, + 165, + 86 + ], + "score": 1.0, + "content": "References", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 108, + 89, + 506, + 103 + ], + "spans": [ + { + "bbox": [ + 108, + 89, + 506, + 103 + ], + "score": 1.0, + "content": "[1] J. Konecnˇ y, H. B. McMahan, D. Ramage, and P. Richtárik, “Federated optimization: Distributed `", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 127, + 101, + 473, + 115 + ], + "spans": [ + { + "bbox": [ + 127, + 101, + 473, + 115 + ], + "score": 1.0, + "content": "machine learning for on-device intelligence,” arXiv preprint arXiv:1610.02527, 2016.", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 109, + 118, + 506, + 133 + ], + "spans": [ + { + "bbox": [ + 109, + 118, + 506, + 133 + ], + "score": 1.0, + "content": "[2] M. Li, D. G. Andersen, A. J. Smola, and K. Yu, “Communication efficient distributed machine", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 127, + 131, + 506, + 143 + ], + "spans": [ + { + "bbox": [ + 127, + 131, + 506, + 143 + ], + "score": 1.0, + "content": "learning with the parameter server,” in Advances in Neural Information Processing Systems 27,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 127, + 141, + 506, + 155 + ], + "spans": [ + { + "bbox": [ + 127, + 141, + 506, + 155 + ], + "score": 1.0, + "content": "Z. Ghahramani, M. Welling, C. Cortes, N. D. Lawrence, and K. Q. Weinberger, Eds. Curran", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 127, + 152, + 266, + 166 + ], + "spans": [ + { + "bbox": [ + 127, + 152, + 266, + 166 + ], + "score": 1.0, + "content": "Associates, Inc., 2014, pp. 19–27.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 109, + 171, + 506, + 184 + ], + "spans": [ + { + "bbox": [ + 109, + 171, + 506, + 184 + ], + "score": 1.0, + "content": "[3] J. Dean, G. Corrado, R. Monga, K. Chen, M. Devin, M. Mao, M. Ranzato, A. Senior, P. Tucker,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 127, + 182, + 506, + 196 + ], + "spans": [ + { + "bbox": [ + 127, + 182, + 506, + 196 + ], + "score": 1.0, + "content": "K. Yang et al., “Large scale distributed deep networks,” in Advances in neural information", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 125, + 193, + 299, + 207 + ], + "spans": [ + { + "bbox": [ + 125, + 193, + 299, + 207 + ], + "score": 1.0, + "content": "processing systems, 2012, pp. 1223–1231.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 109, + 210, + 507, + 226 + ], + "spans": [ + { + "bbox": [ + 109, + 210, + 507, + 226 + ], + "score": 1.0, + "content": "[4] T. Léauté and B. Faltings, “Protecting privacy through distributed computation in multi-agent", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 127, + 222, + 487, + 236 + ], + "spans": [ + { + "bbox": [ + 127, + 222, + 487, + 236 + ], + "score": 1.0, + "content": "decision making,” Journal of Artificial Intelligence Research, vol. 47, pp. 649–695, 2013.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 110, + 240, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 110, + 240, + 506, + 254 + ], + "score": 1.0, + "content": "[5] C. Fang, C. J. Li, Z. Lin, and T. Zhang, “Spider: Near-optimal non-convex optimization via", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 125, + 250, + 506, + 267 + ], + "spans": [ + { + "bbox": [ + 125, + 250, + 506, + 267 + ], + "score": 1.0, + "content": "stochastic path-integrated differential estimator,” in Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 127, + 263, + 243, + 275 + ], + "spans": [ + { + "bbox": [ + 127, + 263, + 243, + 275 + ], + "score": 1.0, + "content": "Systems, 2018, pp. 689–699.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 110, + 281, + 506, + 294 + ], + "spans": [ + { + "bbox": [ + 110, + 281, + 506, + 294 + ], + "score": 1.0, + "content": "[6] D. Zhou, P. Xu, and Q. Gu, “Stochastic nested variance reduction for nonconvex optimization,”", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 127, + 293, + 290, + 304 + ], + "spans": [ + { + "bbox": [ + 127, + 293, + 290, + 304 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1806.07811, 2018.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 110, + 310, + 506, + 324 + ], + "spans": [ + { + "bbox": [ + 110, + 310, + 506, + 324 + ], + "score": 1.0, + "content": "[7] A. Cutkosky and F. Orabona, “Momentum-based variance reduction in non-convex SGD,” in", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 124, + 320, + 507, + 336 + ], + "spans": [ + { + "bbox": [ + 124, + 320, + 507, + 336 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems 32. Curran Associates, Inc., 2019, pp.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 127, + 332, + 191, + 345 + ], + "spans": [ + { + "bbox": [ + 127, + 332, + 191, + 345 + ], + "score": 1.0, + "content": "15 236–15 245.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 110, + 351, + 506, + 364 + ], + "spans": [ + { + "bbox": [ + 110, + 351, + 506, + 364 + ], + "score": 1.0, + "content": "[8] Q. Tran-Dinh, N. H. Pham, D. T. Phan, and L. M. Nguyen, “Hybrid stochastic gradient descent", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 127, + 362, + 496, + 376 + ], + "spans": [ + { + "bbox": [ + 127, + 362, + 496, + 376 + ], + "score": 1.0, + "content": "algorithms for stochastic nonconvex optimization,” arXiv preprint arXiv:1905.05920, 2019.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 110, + 380, + 506, + 394 + ], + "spans": [ + { + "bbox": [ + 110, + 380, + 506, + 394 + ], + "score": 1.0, + "content": "[9] X. Zhang, M. Hong, S. Dhople, W. Yin, and Y. Liu, “Fedpd: A federated learning framework", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 126, + 390, + 485, + 406 + ], + "spans": [ + { + "bbox": [ + 126, + 390, + 485, + 406 + ], + "score": 1.0, + "content": "with adaptivity to non-iid data,” IEEE Transactions on Signal Processing, pp. 1–1, 2021.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 411, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 106, + 411, + 505, + 422 + ], + "score": 1.0, + "content": "[10] T. Li, A. K. Sahu, M. Zaheer, M. Sanjabi, A. Talwalkar, and V. Smith, “Federated optimization", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 127, + 421, + 406, + 434 + ], + "spans": [ + { + "bbox": [ + 127, + 421, + 406, + 434 + ], + "score": 1.0, + "content": "in heterogeneous networks,” arXiv preprint arXiv:1812.06127, 2018.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 437, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 506, + 453 + ], + "score": 1.0, + "content": "[11] B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, “Communication-efficient", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 127, + 450, + 508, + 464 + ], + "spans": [ + { + "bbox": [ + 127, + 450, + 508, + 464 + ], + "score": 1.0, + "content": "learning of deep networks from decentralized data,” in Artificial Intelligence and Statistics.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 126, + 459, + 251, + 475 + ], + "spans": [ + { + "bbox": [ + 126, + 459, + 251, + 475 + ], + "score": 1.0, + "content": "PMLR, 2017, pp. 1273–1282.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 479, + 507, + 493 + ], + "spans": [ + { + "bbox": [ + 106, + 479, + 507, + 493 + ], + "score": 1.0, + "content": "[12] H. Yu, S. Yang, and S. Zhu, “Parallel restarted sgd with faster convergence and less communica-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 127, + 491, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 127, + 491, + 506, + 504 + ], + "score": 1.0, + "content": "tion: Demystifying why model averaging works for deep learning,” in Proceedings of the AAAI", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 127, + 501, + 429, + 515 + ], + "spans": [ + { + "bbox": [ + 127, + 501, + 429, + 515 + ], + "score": 1.0, + "content": "Conference on Artificial Intelligence, vol. 33, no. 01, 2019, pp. 5693–5700.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 104, + 519, + 506, + 534 + ], + "spans": [ + { + "bbox": [ + 104, + 519, + 506, + 534 + ], + "score": 1.0, + "content": "[13] B. Woodworth, K. K. Patel, and N. Srebro, “Minibatch vs local sgd for heterogeneous distributed", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 127, + 531, + 332, + 545 + ], + "spans": [ + { + "bbox": [ + 127, + 531, + 332, + 545 + ], + "score": 1.0, + "content": "learning,” arXiv preprint arXiv:2006.04735, 2020.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 104, + 547, + 508, + 564 + ], + "spans": [ + { + "bbox": [ + 104, + 547, + 508, + 564 + ], + "score": 1.0, + "content": "[14] H. Yu, R. Jin, and S. Yang, “On the linear speedup analysis of communication efficient mo-", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 126, + 559, + 506, + 575 + ], + "spans": [ + { + "bbox": [ + 126, + 559, + 506, + 575 + ], + "score": 1.0, + "content": "mentum sgd for distributed non-convex optimization,” in International Conference on Machine", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 125, + 570, + 299, + 585 + ], + "spans": [ + { + "bbox": [ + 125, + 570, + 299, + 585 + ], + "score": 1.0, + "content": "Learning. PMLR, 2019, pp. 7184–7193.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 588, + 506, + 604 + ], + "spans": [ + { + "bbox": [ + 104, + 588, + 506, + 604 + ], + "score": 1.0, + "content": "[15] S. P. Karimireddy, S. Kale, M. Mohri, S. Reddi, S. Stich, and A. T. Suresh, “Scaffold: Stochastic", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 127, + 601, + 507, + 614 + ], + "spans": [ + { + "bbox": [ + 127, + 601, + 507, + 614 + ], + "score": 1.0, + "content": "controlled averaging for federated learning,” in International Conference on Machine Learning.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 126, + 610, + 252, + 625 + ], + "spans": [ + { + "bbox": [ + 126, + 610, + 252, + 625 + ], + "score": 1.0, + "content": "PMLR, 2020, pp. 5132–5143.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 106, + 630, + 506, + 644 + ], + "spans": [ + { + "bbox": [ + 106, + 630, + 506, + 644 + ], + "score": 1.0, + "content": "[16] H. Yang, M. Fang, and J. Liu, “Achieving linear speedup with partial worker participation in", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 125, + 640, + 403, + 655 + ], + "spans": [ + { + "bbox": [ + 125, + 640, + 403, + 655 + ], + "score": 1.0, + "content": "non-iid federated learning,” arXiv preprint arXiv:2101.11203, 2021.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 659, + 506, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 506, + 673 + ], + "score": 1.0, + "content": "[17] S. P. Karimireddy, M. Jaggi, S. Kale, M. Mohri, S. J. Reddi, S. U. Stich, and A. T. Suresh,", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 127, + 669, + 506, + 685 + ], + "spans": [ + { + "bbox": [ + 127, + 669, + 506, + 685 + ], + "score": 1.0, + "content": "“Mime: Mimicking centralized stochastic algorithms in federated learning,” arXiv preprint", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 127, + 681, + 232, + 695 + ], + "spans": [ + { + "bbox": [ + 127, + 681, + 232, + 695 + ], + "score": 1.0, + "content": "arXiv:2008.03606, 2020.", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 700, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 506, + 713 + ], + "score": 1.0, + "content": "[18] R. Das, A. Hashemi, S. Sanghavi, and I. S. Dhillon, “Improved convergence rates for non-convex", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 127, + 711, + 444, + 724 + ], + "spans": [ + { + "bbox": [ + 127, + 711, + 444, + 724 + ], + "score": 1.0, + "content": "federated learning with compression,” arXiv preprint arXiv:2012.04061, 2020.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 23 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 310, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 312, + 755 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 312, + 755 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 106, + 72, + 507, + 729 + ], + "lines": [ + { + "bbox": [ + 106, + 70, + 165, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 70, + 165, + 86 + ], + "score": 1.0, + "content": "References", + "type": "text" + } + ], + "index": 0, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 108, + 89, + 506, + 103 + ], + "spans": [ + { + "bbox": [ + 108, + 89, + 506, + 103 + ], + "score": 1.0, + "content": "[1] J. Konecnˇ y, H. B. McMahan, D. Ramage, and P. Richtárik, “Federated optimization: Distributed `", + "type": "text" + } + ], + "index": 1, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 101, + 473, + 115 + ], + "spans": [ + { + "bbox": [ + 127, + 101, + 473, + 115 + ], + "score": 1.0, + "content": "machine learning for on-device intelligence,” arXiv preprint arXiv:1610.02527, 2016.", + "type": "text" + } + ], + "index": 2, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 118, + 506, + 133 + ], + "spans": [ + { + "bbox": [ + 109, + 118, + 506, + 133 + ], + "score": 1.0, + "content": "[2] M. Li, D. G. Andersen, A. J. Smola, and K. Yu, “Communication efficient distributed machine", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 131, + 506, + 143 + ], + "spans": [ + { + "bbox": [ + 127, + 131, + 506, + 143 + ], + "score": 1.0, + "content": "learning with the parameter server,” in Advances in Neural Information Processing Systems 27,", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 127, + 141, + 506, + 155 + ], + "spans": [ + { + "bbox": [ + 127, + 141, + 506, + 155 + ], + "score": 1.0, + "content": "Z. Ghahramani, M. Welling, C. Cortes, N. D. Lawrence, and K. Q. Weinberger, Eds. Curran", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 127, + 152, + 266, + 166 + ], + "spans": [ + { + "bbox": [ + 127, + 152, + 266, + 166 + ], + "score": 1.0, + "content": "Associates, Inc., 2014, pp. 19–27.", + "type": "text" + } + ], + "index": 6, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 171, + 506, + 184 + ], + "spans": [ + { + "bbox": [ + 109, + 171, + 506, + 184 + ], + "score": 1.0, + "content": "[3] J. Dean, G. Corrado, R. Monga, K. Chen, M. Devin, M. Mao, M. Ranzato, A. Senior, P. Tucker,", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 182, + 506, + 196 + ], + "spans": [ + { + "bbox": [ + 127, + 182, + 506, + 196 + ], + "score": 1.0, + "content": "K. Yang et al., “Large scale distributed deep networks,” in Advances in neural information", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 125, + 193, + 299, + 207 + ], + "spans": [ + { + "bbox": [ + 125, + 193, + 299, + 207 + ], + "score": 1.0, + "content": "processing systems, 2012, pp. 1223–1231.", + "type": "text" + } + ], + "index": 9, + "is_list_end_line": true + }, + { + "bbox": [ + 109, + 210, + 507, + 226 + ], + "spans": [ + { + "bbox": [ + 109, + 210, + 507, + 226 + ], + "score": 1.0, + "content": "[4] T. Léauté and B. Faltings, “Protecting privacy through distributed computation in multi-agent", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 222, + 487, + 236 + ], + "spans": [ + { + "bbox": [ + 127, + 222, + 487, + 236 + ], + "score": 1.0, + "content": "decision making,” Journal of Artificial Intelligence Research, vol. 47, pp. 649–695, 2013.", + "type": "text" + } + ], + "index": 11, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 240, + 506, + 254 + ], + "spans": [ + { + "bbox": [ + 110, + 240, + 506, + 254 + ], + "score": 1.0, + "content": "[5] C. Fang, C. J. Li, Z. Lin, and T. Zhang, “Spider: Near-optimal non-convex optimization via", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 250, + 506, + 267 + ], + "spans": [ + { + "bbox": [ + 125, + 250, + 506, + 267 + ], + "score": 1.0, + "content": "stochastic path-integrated differential estimator,” in Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 127, + 263, + 243, + 275 + ], + "spans": [ + { + "bbox": [ + 127, + 263, + 243, + 275 + ], + "score": 1.0, + "content": "Systems, 2018, pp. 689–699.", + "type": "text" + } + ], + "index": 14, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 281, + 506, + 294 + ], + "spans": [ + { + "bbox": [ + 110, + 281, + 506, + 294 + ], + "score": 1.0, + "content": "[6] D. Zhou, P. Xu, and Q. Gu, “Stochastic nested variance reduction for nonconvex optimization,”", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 293, + 290, + 304 + ], + "spans": [ + { + "bbox": [ + 127, + 293, + 290, + 304 + ], + "score": 1.0, + "content": "arXiv preprint arXiv:1806.07811, 2018.", + "type": "text" + } + ], + "index": 16, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 310, + 506, + 324 + ], + "spans": [ + { + "bbox": [ + 110, + 310, + 506, + 324 + ], + "score": 1.0, + "content": "[7] A. Cutkosky and F. Orabona, “Momentum-based variance reduction in non-convex SGD,” in", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 320, + 507, + 336 + ], + "spans": [ + { + "bbox": [ + 124, + 320, + 507, + 336 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems 32. Curran Associates, Inc., 2019, pp.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 127, + 332, + 191, + 345 + ], + "spans": [ + { + "bbox": [ + 127, + 332, + 191, + 345 + ], + "score": 1.0, + "content": "15 236–15 245.", + "type": "text" + } + ], + "index": 19, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 351, + 506, + 364 + ], + "spans": [ + { + "bbox": [ + 110, + 351, + 506, + 364 + ], + "score": 1.0, + "content": "[8] Q. Tran-Dinh, N. H. Pham, D. T. Phan, and L. M. Nguyen, “Hybrid stochastic gradient descent", + "type": "text" + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 362, + 496, + 376 + ], + "spans": [ + { + "bbox": [ + 127, + 362, + 496, + 376 + ], + "score": 1.0, + "content": "algorithms for stochastic nonconvex optimization,” arXiv preprint arXiv:1905.05920, 2019.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 110, + 380, + 506, + 394 + ], + "spans": [ + { + "bbox": [ + 110, + 380, + 506, + 394 + ], + "score": 1.0, + "content": "[9] X. Zhang, M. Hong, S. Dhople, W. Yin, and Y. Liu, “Fedpd: A federated learning framework", + "type": "text" + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 390, + 485, + 406 + ], + "spans": [ + { + "bbox": [ + 126, + 390, + 485, + 406 + ], + "score": 1.0, + "content": "with adaptivity to non-iid data,” IEEE Transactions on Signal Processing, pp. 1–1, 2021.", + "type": "text" + } + ], + "index": 23, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 411, + 505, + 422 + ], + "spans": [ + { + "bbox": [ + 106, + 411, + 505, + 422 + ], + "score": 1.0, + "content": "[10] T. Li, A. K. Sahu, M. Zaheer, M. Sanjabi, A. Talwalkar, and V. Smith, “Federated optimization", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 421, + 406, + 434 + ], + "spans": [ + { + "bbox": [ + 127, + 421, + 406, + 434 + ], + "score": 1.0, + "content": "in heterogeneous networks,” arXiv preprint arXiv:1812.06127, 2018.", + "type": "text" + } + ], + "index": 25, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 437, + 506, + 453 + ], + "spans": [ + { + "bbox": [ + 105, + 437, + 506, + 453 + ], + "score": 1.0, + "content": "[11] B. McMahan, E. Moore, D. Ramage, S. Hampson, and B. A. y Arcas, “Communication-efficient", + "type": "text" + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 450, + 508, + 464 + ], + "spans": [ + { + "bbox": [ + 127, + 450, + 508, + 464 + ], + "score": 1.0, + "content": "learning of deep networks from decentralized data,” in Artificial Intelligence and Statistics.", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 126, + 459, + 251, + 475 + ], + "spans": [ + { + "bbox": [ + 126, + 459, + 251, + 475 + ], + "score": 1.0, + "content": "PMLR, 2017, pp. 1273–1282.", + "type": "text" + } + ], + "index": 28, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 479, + 507, + 493 + ], + "spans": [ + { + "bbox": [ + 106, + 479, + 507, + 493 + ], + "score": 1.0, + "content": "[12] H. Yu, S. Yang, and S. Zhu, “Parallel restarted sgd with faster convergence and less communica-", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 491, + 506, + 504 + ], + "spans": [ + { + "bbox": [ + 127, + 491, + 506, + 504 + ], + "score": 1.0, + "content": "tion: Demystifying why model averaging works for deep learning,” in Proceedings of the AAAI", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 127, + 501, + 429, + 515 + ], + "spans": [ + { + "bbox": [ + 127, + 501, + 429, + 515 + ], + "score": 1.0, + "content": "Conference on Artificial Intelligence, vol. 33, no. 01, 2019, pp. 5693–5700.", + "type": "text" + } + ], + "index": 31, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 519, + 506, + 534 + ], + "spans": [ + { + "bbox": [ + 104, + 519, + 506, + 534 + ], + "score": 1.0, + "content": "[13] B. Woodworth, K. K. Patel, and N. Srebro, “Minibatch vs local sgd for heterogeneous distributed", + "type": "text" + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 531, + 332, + 545 + ], + "spans": [ + { + "bbox": [ + 127, + 531, + 332, + 545 + ], + "score": 1.0, + "content": "learning,” arXiv preprint arXiv:2006.04735, 2020.", + "type": "text" + } + ], + "index": 33, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 547, + 508, + 564 + ], + "spans": [ + { + "bbox": [ + 104, + 547, + 508, + 564 + ], + "score": 1.0, + "content": "[14] H. Yu, R. Jin, and S. Yang, “On the linear speedup analysis of communication efficient mo-", + "type": "text" + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 559, + 506, + 575 + ], + "spans": [ + { + "bbox": [ + 126, + 559, + 506, + 575 + ], + "score": 1.0, + "content": "mentum sgd for distributed non-convex optimization,” in International Conference on Machine", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 125, + 570, + 299, + 585 + ], + "spans": [ + { + "bbox": [ + 125, + 570, + 299, + 585 + ], + "score": 1.0, + "content": "Learning. PMLR, 2019, pp. 7184–7193.", + "type": "text" + } + ], + "index": 36, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 588, + 506, + 604 + ], + "spans": [ + { + "bbox": [ + 104, + 588, + 506, + 604 + ], + "score": 1.0, + "content": "[15] S. P. Karimireddy, S. Kale, M. Mohri, S. Reddi, S. Stich, and A. T. Suresh, “Scaffold: Stochastic", + "type": "text" + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 601, + 507, + 614 + ], + "spans": [ + { + "bbox": [ + 127, + 601, + 507, + 614 + ], + "score": 1.0, + "content": "controlled averaging for federated learning,” in International Conference on Machine Learning.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 126, + 610, + 252, + 625 + ], + "spans": [ + { + "bbox": [ + 126, + 610, + 252, + 625 + ], + "score": 1.0, + "content": "PMLR, 2020, pp. 5132–5143.", + "type": "text" + } + ], + "index": 39, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 630, + 506, + 644 + ], + "spans": [ + { + "bbox": [ + 106, + 630, + 506, + 644 + ], + "score": 1.0, + "content": "[16] H. Yang, M. Fang, and J. Liu, “Achieving linear speedup with partial worker participation in", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 640, + 403, + 655 + ], + "spans": [ + { + "bbox": [ + 125, + 640, + 403, + 655 + ], + "score": 1.0, + "content": "non-iid federated learning,” arXiv preprint arXiv:2101.11203, 2021.", + "type": "text" + } + ], + "index": 41, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 659, + 506, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 659, + 506, + 673 + ], + "score": 1.0, + "content": "[17] S. P. Karimireddy, M. Jaggi, S. Kale, M. Mohri, S. J. Reddi, S. U. Stich, and A. T. Suresh,", + "type": "text" + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 669, + 506, + 685 + ], + "spans": [ + { + "bbox": [ + 127, + 669, + 506, + 685 + ], + "score": 1.0, + "content": "“Mime: Mimicking centralized stochastic algorithms in federated learning,” arXiv preprint", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 127, + 681, + 232, + 695 + ], + "spans": [ + { + "bbox": [ + 127, + 681, + 232, + 695 + ], + "score": 1.0, + "content": "arXiv:2008.03606, 2020.", + "type": "text" + } + ], + "index": 44, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 700, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 106, + 700, + 506, + 713 + ], + "score": 1.0, + "content": "[18] R. Das, A. Hashemi, S. Sanghavi, and I. S. Dhillon, “Improved convergence rates for non-convex", + "type": "text" + } + ], + "index": 45, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 711, + 444, + 724 + ], + "spans": [ + { + "bbox": [ + 127, + 711, + 444, + 724 + ], + "score": 1.0, + "content": "federated learning with compression,” arXiv preprint arXiv:2012.04061, 2020.", + "type": "text" + } + ], + "index": 46, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 73, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 73, + 506, + 86 + ], + "score": 1.0, + "content": "[19] B. Woodworth, K. K. Patel, S. U. Stich, Z. Dai, B. Bullins, H. B. McMahan, O. Shamir, and", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 83, + 500, + 97 + ], + "spans": [ + { + "bbox": [ + 127, + 83, + 500, + 97 + ], + "score": 1.0, + "content": "N. Srebro, “Is local sgd better than minibatch sgd?” arXiv preprint arXiv:2002.07839, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 104, + 101, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 104, + 101, + 506, + 117 + ], + "score": 1.0, + "content": "[20] H. Yu and R. Jin, “On the computation and communication complexity of parallel sgd with", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 113, + 506, + 127 + ], + "spans": [ + { + "bbox": [ + 127, + 113, + 506, + 127 + ], + "score": 1.0, + "content": "dynamic batch sizes for stochastic non-convex optimization,” in International Conference on", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 126, + 124, + 336, + 138 + ], + "spans": [ + { + "bbox": [ + 126, + 124, + 336, + 138 + ], + "score": 1.0, + "content": "Machine Learning. PMLR, 2019, pp. 7174–7183.", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 141, + 508, + 158 + ], + "spans": [ + { + "bbox": [ + 104, + 141, + 508, + 158 + ], + "score": 1.0, + "content": "[21] J. Wang and G. Joshi, “Cooperative sgd: A unified framework for the design and analysis of", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 154, + 506, + 168 + ], + "spans": [ + { + "bbox": [ + 127, + 154, + 506, + 168 + ], + "score": 1.0, + "content": "local-update sgd algorithms,” Journal of Machine Learning Research, vol. 22, no. 213, pp. 1–50,", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 127, + 162, + 155, + 180 + ], + "spans": [ + { + "bbox": [ + 127, + 162, + 155, + 180 + ], + "score": 1.0, + "content": "2021.", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 184, + 507, + 198 + ], + "spans": [ + { + "bbox": [ + 106, + 184, + 507, + 198 + ], + "score": 1.0, + "content": "[22] A. Khaled, K. Mishchenko, and P. Richtárik, “Better communication complexity for local sgd,”", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 194, + 181, + 208 + ], + "spans": [ + { + "bbox": [ + 126, + 194, + 181, + 208 + ], + "score": 1.0, + "content": "arXiv, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 212, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 506, + 228 + ], + "score": 1.0, + "content": "[23] S. U. Stich, “Local sgd converges fast and communicates little,” arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 223, + 232, + 238 + ], + "spans": [ + { + "bbox": [ + 126, + 223, + 232, + 238 + ], + "score": 1.0, + "content": "arXiv:1805.09767, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 243, + 506, + 257 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 506, + 257 + ], + "score": 1.0, + "content": "[24] T. Lin, S. U. Stich, K. K. Patel, and M. Jaggi, “Don’t use large mini-batches, use local sgd,” in", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 254, + 377, + 268 + ], + "spans": [ + { + "bbox": [ + 126, + 254, + 377, + 268 + ], + "score": 1.0, + "content": "International Conference on Learning Representations, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 273, + 506, + 287 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 506, + 287 + ], + "score": 1.0, + "content": "[25] F. Zhou and G. Cong, “On the convergence properties of a k-step averaging stochastic gra-", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 285, + 505, + 296 + ], + "spans": [ + { + "bbox": [ + 127, + 285, + 505, + 296 + ], + "score": 1.0, + "content": "dient descent algorithm for nonconvex optimization,” in Proceedings of the Twenty-Seventh", + "type": "text", + "cross_page": true + } + ], + "index": 15 + }, + { + "bbox": [ + 126, + 294, + 495, + 308 + ], + "spans": [ + { + "bbox": [ + 126, + 294, + 495, + 308 + ], + "score": 1.0, + "content": "International Joint Conference on Artificial Intelligence, IJCAI-18, 7 2018, pp. 3219–3227.", + "type": "text", + "cross_page": true + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 313, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 506, + 326 + ], + "score": 1.0, + "content": "[26] F. Sattler, S. Wiedemann, K.-R. Müller, and W. Samek, “Robust and communication-efficient", + "type": "text", + "cross_page": true + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 323, + 506, + 339 + ], + "spans": [ + { + "bbox": [ + 126, + 323, + 506, + 339 + ], + "score": 1.0, + "content": "federated learning from non-iid data,” IEEE transactions on neural networks and learning", + "type": "text", + "cross_page": true + } + ], + "index": 18 + }, + { + "bbox": [ + 127, + 335, + 312, + 349 + ], + "spans": [ + { + "bbox": [ + 127, + 335, + 312, + 349 + ], + "score": 1.0, + "content": "systems, vol. 31, no. 9, pp. 3400–3413, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 353, + 506, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 506, + 367 + ], + "score": 1.0, + "content": "[27] Y. Zhao, M. Li, L. Lai, N. Suda, D. Civin, and V. Chandra, “Federated learning with non-iid", + "type": "text", + "cross_page": true + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 366, + 316, + 378 + ], + "spans": [ + { + "bbox": [ + 127, + 366, + 316, + 378 + ], + "score": 1.0, + "content": "data,” arXiv preprint arXiv:1806.00582, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 21, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 383, + 506, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 506, + 397 + ], + "score": 1.0, + "content": "[28] J. Wang, V. Tantia, N. Ballas, and M. Rabbat, “Slowmo: Improving communication-efficient", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 394, + 447, + 408 + ], + "spans": [ + { + "bbox": [ + 126, + 394, + 447, + 408 + ], + "score": 1.0, + "content": "distributed sgd with slow momentum,” arXiv preprint arXiv:1910.00643, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 412, + 506, + 427 + ], + "spans": [ + { + "bbox": [ + 104, + 412, + 506, + 427 + ], + "score": 1.0, + "content": "[29] X. Liang, S. Shen, J. Liu, Z. Pan, E. Chen, and Y. Cheng, “Variance reduced local sgd with", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 424, + 433, + 438 + ], + "spans": [ + { + "bbox": [ + 126, + 424, + 433, + 438 + ], + "score": 1.0, + "content": "lower communication complexity,” arXiv preprint arXiv:1912.12844, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 443, + 507, + 457 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 507, + 457 + ], + "score": 1.0, + "content": "[30] P. Sharma, P. Khanduri, S. Bulusu, K. Rajawat, and P. K. Varshney, “Parallel restarted SPI-", + "type": "text", + "cross_page": true + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 455, + 505, + 467 + ], + "spans": [ + { + "bbox": [ + 127, + 455, + 505, + 467 + ], + "score": 1.0, + "content": "DER – communication efficient distributed nonconvex optimization with optimal computation", + "type": "text", + "cross_page": true + } + ], + "index": 27 + }, + { + "bbox": [ + 127, + 465, + 344, + 479 + ], + "spans": [ + { + "bbox": [ + 127, + 465, + 344, + 479 + ], + "score": 1.0, + "content": "complexity,” arXiv preprint arXiv:1912.06036, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 28, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 484, + 506, + 497 + ], + "spans": [ + { + "bbox": [ + 106, + 484, + 506, + 497 + ], + "score": 1.0, + "content": "[31] S. J. Reddi, S. Kale, and S. Kumar, “On the convergence of adam and beyond,” arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 125, + 495, + 232, + 508 + ], + "spans": [ + { + "bbox": [ + 125, + 495, + 232, + 508 + ], + "score": 1.0, + "content": "arXiv:1904.09237, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 511, + 506, + 528 + ], + "spans": [ + { + "bbox": [ + 104, + 511, + 506, + 528 + ], + "score": 1.0, + "content": "[32] A. Koloskova, N. Loizou, S. Boreiri, M. Jaggi, and S. Stich, “A unified theory of decentralized", + "type": "text", + "cross_page": true + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 524, + 506, + 538 + ], + "spans": [ + { + "bbox": [ + 126, + 524, + 506, + 538 + ], + "score": 1.0, + "content": "sgd with changing topology and local updates,” in International Conference on Machine", + "type": "text", + "cross_page": true + } + ], + "index": 32 + }, + { + "bbox": [ + 124, + 534, + 299, + 550 + ], + "spans": [ + { + "bbox": [ + 124, + 534, + 299, + 550 + ], + "score": 1.0, + "content": "Learning. PMLR, 2020, pp. 5381–5393.", + "type": "text", + "cross_page": true + } + ], + "index": 33, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 554, + 506, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 554, + 506, + 568 + ], + "score": 1.0, + "content": "[33] R. Johnson and T. Zhang, “Accelerating stochastic gradient descent using predictive variance", + "type": "text", + "cross_page": true + } + ], + "index": 34, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 564, + 506, + 579 + ], + "spans": [ + { + "bbox": [ + 126, + 564, + 506, + 579 + ], + "score": 1.0, + "content": "reduction,” in Advances in Neural Information Processing Systems 26. Curran Associates,", + "type": "text", + "cross_page": true + } + ], + "index": 35 + }, + { + "bbox": [ + 127, + 576, + 230, + 590 + ], + "spans": [ + { + "bbox": [ + 127, + 576, + 230, + 590 + ], + "score": 1.0, + "content": "Inc., 2013, pp. 315–323.", + "type": "text", + "cross_page": true + } + ], + "index": 36, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 594, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 609 + ], + "score": 1.0, + "content": "[34] L. Bottou, F. E. Curtis, and J. Nocedal, “Optimization methods for large-scale machine learning,”", + "type": "text", + "cross_page": true + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 606, + 324, + 620 + ], + "spans": [ + { + "bbox": [ + 126, + 606, + 324, + 620 + ], + "score": 1.0, + "content": "SIAM Review, vol. 60, no. 2, pp. 223–311, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 38, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 623, + 506, + 638 + ], + "spans": [ + { + "bbox": [ + 104, + 623, + 506, + 638 + ], + "score": 1.0, + "content": "[35] Y. Drori and O. Shamir, “The complexity of finding stationary points with stochastic gradient", + "type": "text", + "cross_page": true + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 636, + 502, + 650 + ], + "spans": [ + { + "bbox": [ + 127, + 636, + 502, + 650 + ], + "score": 1.0, + "content": "descent,” in International Conference on Machine Learning. PMLR, 2020, pp. 2658–2667.", + "type": "text", + "cross_page": true + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 654, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 506, + 667 + ], + "score": 1.0, + "content": "[36] D. A. E. Acar, Y. Zhao, R. Matas, M. Mattina, P. Whatmough, and V. Saligrama, “Federated", + "type": "text", + "cross_page": true + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 665, + 506, + 679 + ], + "spans": [ + { + "bbox": [ + 127, + 665, + 506, + 679 + ], + "score": 1.0, + "content": "learning based on dynamic regularization,” in International Conference on Learning Represen-", + "type": "text", + "cross_page": true + } + ], + "index": 42 + }, + { + "bbox": [ + 127, + 676, + 186, + 690 + ], + "spans": [ + { + "bbox": [ + 127, + 676, + 186, + 690 + ], + "score": 1.0, + "content": "tations, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 43, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 694, + 507, + 708 + ], + "spans": [ + { + "bbox": [ + 106, + 694, + 507, + 708 + ], + "score": 1.0, + "content": "[37] S. Caldas, S. M. K. Duddu, P. Wu, T. Li, J. Konecnˇ ý, H. B. McMahan, V. Smith, and A. Tal-", + "type": "text", + "cross_page": true + } + ], + "index": 44, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 705, + 363, + 719 + ], + "spans": [ + { + "bbox": [ + 126, + 705, + 363, + 719 + ], + "score": 1.0, + "content": "walkar, “Leaf: A benchmark for federated settings,” 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 45, + "is_list_end_line": true + } + ], + "index": 23, + "bbox_fs": [ + 104, + 70, + 508, + 724 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 54, + 507, + 720 + ], + "lines": [ + { + "bbox": [ + 106, + 73, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 73, + 506, + 86 + ], + "score": 1.0, + "content": "[19] B. Woodworth, K. K. Patel, S. U. Stich, Z. Dai, B. Bullins, H. B. McMahan, O. Shamir, and", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 127, + 83, + 500, + 97 + ], + "spans": [ + { + "bbox": [ + 127, + 83, + 500, + 97 + ], + "score": 1.0, + "content": "N. Srebro, “Is local sgd better than minibatch sgd?” arXiv preprint arXiv:2002.07839, 2020.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 104, + 101, + 506, + 117 + ], + "spans": [ + { + "bbox": [ + 104, + 101, + 506, + 117 + ], + "score": 1.0, + "content": "[20] H. Yu and R. Jin, “On the computation and communication complexity of parallel sgd with", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 127, + 113, + 506, + 127 + ], + "spans": [ + { + "bbox": [ + 127, + 113, + 506, + 127 + ], + "score": 1.0, + "content": "dynamic batch sizes for stochastic non-convex optimization,” in International Conference on", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 126, + 124, + 336, + 138 + ], + "spans": [ + { + "bbox": [ + 126, + 124, + 336, + 138 + ], + "score": 1.0, + "content": "Machine Learning. PMLR, 2019, pp. 7174–7183.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 104, + 141, + 508, + 158 + ], + "spans": [ + { + "bbox": [ + 104, + 141, + 508, + 158 + ], + "score": 1.0, + "content": "[21] J. Wang and G. Joshi, “Cooperative sgd: A unified framework for the design and analysis of", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 127, + 154, + 506, + 168 + ], + "spans": [ + { + "bbox": [ + 127, + 154, + 506, + 168 + ], + "score": 1.0, + "content": "local-update sgd algorithms,” Journal of Machine Learning Research, vol. 22, no. 213, pp. 1–50,", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 127, + 162, + 155, + 180 + ], + "spans": [ + { + "bbox": [ + 127, + 162, + 155, + 180 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 184, + 507, + 198 + ], + "spans": [ + { + "bbox": [ + 106, + 184, + 507, + 198 + ], + "score": 1.0, + "content": "[22] A. Khaled, K. Mishchenko, and P. Richtárik, “Better communication complexity for local sgd,”", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 126, + 194, + 181, + 208 + ], + "spans": [ + { + "bbox": [ + 126, + 194, + 181, + 208 + ], + "score": 1.0, + "content": "arXiv, 2019.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 212, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 105, + 212, + 506, + 228 + ], + "score": 1.0, + "content": "[23] S. U. Stich, “Local sgd converges fast and communicates little,” arXiv preprint", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 126, + 223, + 232, + 238 + ], + "spans": [ + { + "bbox": [ + 126, + 223, + 232, + 238 + ], + "score": 1.0, + "content": "arXiv:1805.09767, 2018.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 243, + 506, + 257 + ], + "spans": [ + { + "bbox": [ + 106, + 243, + 506, + 257 + ], + "score": 1.0, + "content": "[24] T. Lin, S. U. Stich, K. K. Patel, and M. Jaggi, “Don’t use large mini-batches, use local sgd,” in", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 126, + 254, + 377, + 268 + ], + "spans": [ + { + "bbox": [ + 126, + 254, + 377, + 268 + ], + "score": 1.0, + "content": "International Conference on Learning Representations, 2020.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 273, + 506, + 287 + ], + "spans": [ + { + "bbox": [ + 106, + 273, + 506, + 287 + ], + "score": 1.0, + "content": "[25] F. Zhou and G. Cong, “On the convergence properties of a k-step averaging stochastic gra-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 127, + 285, + 505, + 296 + ], + "spans": [ + { + "bbox": [ + 127, + 285, + 505, + 296 + ], + "score": 1.0, + "content": "dient descent algorithm for nonconvex optimization,” in Proceedings of the Twenty-Seventh", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 126, + 294, + 495, + 308 + ], + "spans": [ + { + "bbox": [ + 126, + 294, + 495, + 308 + ], + "score": 1.0, + "content": "International Joint Conference on Artificial Intelligence, IJCAI-18, 7 2018, pp. 3219–3227.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 313, + 506, + 326 + ], + "spans": [ + { + "bbox": [ + 105, + 313, + 506, + 326 + ], + "score": 1.0, + "content": "[26] F. Sattler, S. Wiedemann, K.-R. Müller, and W. Samek, “Robust and communication-efficient", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 126, + 323, + 506, + 339 + ], + "spans": [ + { + "bbox": [ + 126, + 323, + 506, + 339 + ], + "score": 1.0, + "content": "federated learning from non-iid data,” IEEE transactions on neural networks and learning", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 127, + 335, + 312, + 349 + ], + "spans": [ + { + "bbox": [ + 127, + 335, + 312, + 349 + ], + "score": 1.0, + "content": "systems, vol. 31, no. 9, pp. 3400–3413, 2019.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 353, + 506, + 367 + ], + "spans": [ + { + "bbox": [ + 106, + 353, + 506, + 367 + ], + "score": 1.0, + "content": "[27] Y. Zhao, M. Li, L. Lai, N. Suda, D. Civin, and V. Chandra, “Federated learning with non-iid", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 127, + 366, + 316, + 378 + ], + "spans": [ + { + "bbox": [ + 127, + 366, + 316, + 378 + ], + "score": 1.0, + "content": "data,” arXiv preprint arXiv:1806.00582, 2018.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 383, + 506, + 397 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 506, + 397 + ], + "score": 1.0, + "content": "[28] J. Wang, V. Tantia, N. Ballas, and M. Rabbat, “Slowmo: Improving communication-efficient", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 126, + 394, + 447, + 408 + ], + "spans": [ + { + "bbox": [ + 126, + 394, + 447, + 408 + ], + "score": 1.0, + "content": "distributed sgd with slow momentum,” arXiv preprint arXiv:1910.00643, 2019.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 412, + 506, + 427 + ], + "spans": [ + { + "bbox": [ + 104, + 412, + 506, + 427 + ], + "score": 1.0, + "content": "[29] X. Liang, S. Shen, J. Liu, Z. Pan, E. Chen, and Y. Cheng, “Variance reduced local sgd with", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 126, + 424, + 433, + 438 + ], + "spans": [ + { + "bbox": [ + 126, + 424, + 433, + 438 + ], + "score": 1.0, + "content": "lower communication complexity,” arXiv preprint arXiv:1912.12844, 2019.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 443, + 507, + 457 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 507, + 457 + ], + "score": 1.0, + "content": "[30] P. Sharma, P. Khanduri, S. Bulusu, K. Rajawat, and P. K. Varshney, “Parallel restarted SPI-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 127, + 455, + 505, + 467 + ], + "spans": [ + { + "bbox": [ + 127, + 455, + 505, + 467 + ], + "score": 1.0, + "content": "DER – communication efficient distributed nonconvex optimization with optimal computation", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 127, + 465, + 344, + 479 + ], + "spans": [ + { + "bbox": [ + 127, + 465, + 344, + 479 + ], + "score": 1.0, + "content": "complexity,” arXiv preprint arXiv:1912.06036, 2019.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 484, + 506, + 497 + ], + "spans": [ + { + "bbox": [ + 106, + 484, + 506, + 497 + ], + "score": 1.0, + "content": "[31] S. J. Reddi, S. Kale, and S. Kumar, “On the convergence of adam and beyond,” arXiv preprint", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 125, + 495, + 232, + 508 + ], + "spans": [ + { + "bbox": [ + 125, + 495, + 232, + 508 + ], + "score": 1.0, + "content": "arXiv:1904.09237, 2019.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 104, + 511, + 506, + 528 + ], + "spans": [ + { + "bbox": [ + 104, + 511, + 506, + 528 + ], + "score": 1.0, + "content": "[32] A. Koloskova, N. Loizou, S. Boreiri, M. Jaggi, and S. Stich, “A unified theory of decentralized", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 126, + 524, + 506, + 538 + ], + "spans": [ + { + "bbox": [ + 126, + 524, + 506, + 538 + ], + "score": 1.0, + "content": "sgd with changing topology and local updates,” in International Conference on Machine", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 124, + 534, + 299, + 550 + ], + "spans": [ + { + "bbox": [ + 124, + 534, + 299, + 550 + ], + "score": 1.0, + "content": "Learning. PMLR, 2020, pp. 5381–5393.", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 554, + 506, + 568 + ], + "spans": [ + { + "bbox": [ + 106, + 554, + 506, + 568 + ], + "score": 1.0, + "content": "[33] R. Johnson and T. Zhang, “Accelerating stochastic gradient descent using predictive variance", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 126, + 564, + 506, + 579 + ], + "spans": [ + { + "bbox": [ + 126, + 564, + 506, + 579 + ], + "score": 1.0, + "content": "reduction,” in Advances in Neural Information Processing Systems 26. Curran Associates,", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 127, + 576, + 230, + 590 + ], + "spans": [ + { + "bbox": [ + 127, + 576, + 230, + 590 + ], + "score": 1.0, + "content": "Inc., 2013, pp. 315–323.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 594, + 505, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 594, + 505, + 609 + ], + "score": 1.0, + "content": "[34] L. Bottou, F. E. Curtis, and J. Nocedal, “Optimization methods for large-scale machine learning,”", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 126, + 606, + 324, + 620 + ], + "spans": [ + { + "bbox": [ + 126, + 606, + 324, + 620 + ], + "score": 1.0, + "content": "SIAM Review, vol. 60, no. 2, pp. 223–311, 2018.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 104, + 623, + 506, + 638 + ], + "spans": [ + { + "bbox": [ + 104, + 623, + 506, + 638 + ], + "score": 1.0, + "content": "[35] Y. Drori and O. Shamir, “The complexity of finding stationary points with stochastic gradient", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 127, + 636, + 502, + 650 + ], + "spans": [ + { + "bbox": [ + 127, + 636, + 502, + 650 + ], + "score": 1.0, + "content": "descent,” in International Conference on Machine Learning. PMLR, 2020, pp. 2658–2667.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 654, + 506, + 667 + ], + "spans": [ + { + "bbox": [ + 105, + 654, + 506, + 667 + ], + "score": 1.0, + "content": "[36] D. A. E. Acar, Y. Zhao, R. Matas, M. Mattina, P. Whatmough, and V. Saligrama, “Federated", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 127, + 665, + 506, + 679 + ], + "spans": [ + { + "bbox": [ + 127, + 665, + 506, + 679 + ], + "score": 1.0, + "content": "learning based on dynamic regularization,” in International Conference on Learning Represen-", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 127, + 676, + 186, + 690 + ], + "spans": [ + { + "bbox": [ + 127, + 676, + 186, + 690 + ], + "score": 1.0, + "content": "tations, 2020.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 694, + 507, + 708 + ], + "spans": [ + { + "bbox": [ + 106, + 694, + 507, + 708 + ], + "score": 1.0, + "content": "[37] S. Caldas, S. M. K. Duddu, P. Wu, T. Li, J. Konecnˇ ý, H. B. McMahan, V. Smith, and A. Tal-", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 126, + 705, + 363, + 719 + ], + "spans": [ + { + "bbox": [ + 126, + 705, + 363, + 719 + ], + "score": 1.0, + "content": "walkar, “Leaf: A benchmark for federated settings,” 2019.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 22.5 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 751 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 312, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 312, + 754 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 105, + 54, + 507, + 720 + ], + "lines": [], + "index": 22.5, + "bbox_fs": [ + 104, + 73, + 508, + 719 + ], + "lines_deleted": true + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/train/J28lNO4p3ki/J28lNO4p3ki_model.json b/parse/train/J28lNO4p3ki/J28lNO4p3ki_model.json new file mode 100644 index 0000000000000000000000000000000000000000..e882c3777a2a08c0ac622b2924bf146977d1074b --- /dev/null +++ b/parse/train/J28lNO4p3ki/J28lNO4p3ki_model.json @@ -0,0 +1,22973 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 397, + 1071, + 1304, + 1071, + 1304, + 1685, + 397, + 1685 + ], + "score": 0.984 + }, + { + "category_id": 1, + "poly": [ + 299, + 1810, + 1402, + 1810, + 1402, + 1932, + 299, + 1932 + ], + "score": 0.969 + }, + { + "category_id": 0, + "poly": [ + 299, + 270, + 1401, + 270, + 1401, + 438, + 299, + 438 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 901, + 554, + 1244, + 554, + 1244, + 647, + 901, + 647 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 318, + 693, + 744, + 693, + 744, + 786, + 318, + 786 + ], + "score": 0.945 + }, + { + "category_id": 1, + "poly": [ + 433, + 832, + 864, + 832, + 864, + 924, + 433, + 924 + ], + "score": 0.944 + }, + { + "category_id": 1, + "poly": [ + 1085, + 693, + 1381, + 693, + 1381, + 785, + 1085, + 785 + ], + "score": 0.942 + }, + { + "category_id": 1, + "poly": [ + 777, + 693, + 1052, + 693, + 1052, + 785, + 777, + 785 + ], + "score": 0.941 + }, + { + "category_id": 1, + "poly": [ + 459, + 555, + 733, + 555, + 733, + 646, + 459, + 646 + ], + "score": 0.932 + }, + { + "category_id": 1, + "poly": [ + 1011, + 832, + 1266, + 832, + 1266, + 925, + 1011, + 925 + ], + "score": 0.923 + }, + { + "category_id": 2, + "poly": [ + 332, + 1954, + 724, + 1954, + 724, + 1982, + 332, + 1982 + ], + "score": 0.907 + }, + { + "category_id": 0, + "poly": [ + 299, + 1739, + 530, + 1739, + 530, + 1778, + 299, + 1778 + ], + "score": 0.898 + }, + { + "category_id": 2, + "poly": [ + 297, + 2033, + 1071, + 2033, + 1071, + 2062, + 297, + 2062 + ], + "score": 0.896 + }, + { + "category_id": 0, + "poly": [ + 788, + 1003, + 913, + 1003, + 913, + 1039, + 788, + 1039 + ], + "score": 0.894 + }, + { + "category_id": 13, + "poly": [ + 655, + 1346, + 742, + 1346, + 742, + 1383, + 655, + 1383 + ], + "score": 0.93, + "latex": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )" + }, + { + "category_id": 13, + "poly": [ + 397, + 1345, + 507, + 1345, + 507, + 1383, + 397, + 1383 + ], + "score": 0.92, + "latex": "\\tilde { \\mathcal O } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1169, + 1357, + 1185, + 1357, + 1185, + 1378, + 1169, + 1378 + ], + "score": 0.6, + "latex": "\\epsilon" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 271.0, + 1301.0, + 271.0, + 1301.0, + 324.0, + 397.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 326.0, + 1345.0, + 326.0, + 1345.0, + 386.0, + 354.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 379.0, + 1406.0, + 379.0, + 1406.0, + 445.0, + 294.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 1947.0, + 730.0, + 1947.0, + 730.0, + 1990.0, + 328.0, + 1990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1736.0, + 536.0, + 1736.0, + 536.0, + 1784.0, + 292.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2028.0, + 1073.0, + 2028.0, + 1073.0, + 2066.0, + 294.0, + 2066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 782.0, + 1000.0, + 920.0, + 1000.0, + 920.0, + 1044.0, + 782.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1074.0, + 1304.0, + 1074.0, + 1304.0, + 1107.0, + 394.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1102.0, + 1306.0, + 1102.0, + 1306.0, + 1138.0, + 394.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1136.0, + 1306.0, + 1136.0, + 1306.0, + 1166.0, + 392.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1166.0, + 1305.0, + 1166.0, + 1305.0, + 1198.0, + 393.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1194.0, + 1305.0, + 1194.0, + 1305.0, + 1228.0, + 393.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1224.0, + 1305.0, + 1224.0, + 1305.0, + 1258.0, + 393.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1258.0, + 1304.0, + 1258.0, + 1304.0, + 1288.0, + 395.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1284.0, + 1305.0, + 1284.0, + 1305.0, + 1320.0, + 393.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1314.0, + 1305.0, + 1314.0, + 1305.0, + 1351.0, + 395.0, + 1351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1343.0, + 396.0, + 1343.0, + 396.0, + 1393.0, + 393.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 1343.0, + 654.0, + 1343.0, + 654.0, + 1393.0, + 508.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 1343.0, + 1168.0, + 1343.0, + 1168.0, + 1393.0, + 743.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1186.0, + 1343.0, + 1307.0, + 1343.0, + 1307.0, + 1393.0, + 1186.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1379.0, + 1306.0, + 1379.0, + 1306.0, + 1415.0, + 393.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1413.0, + 1306.0, + 1413.0, + 1306.0, + 1443.0, + 395.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1441.0, + 1305.0, + 1441.0, + 1305.0, + 1475.0, + 393.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1470.0, + 1309.0, + 1470.0, + 1309.0, + 1507.0, + 393.0, + 1507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1502.0, + 1306.0, + 1502.0, + 1306.0, + 1535.0, + 394.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1532.0, + 1306.0, + 1532.0, + 1306.0, + 1565.0, + 394.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1564.0, + 1306.0, + 1564.0, + 1306.0, + 1597.0, + 394.0, + 1597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1593.0, + 1304.0, + 1593.0, + 1304.0, + 1627.0, + 394.0, + 1627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1622.0, + 1304.0, + 1622.0, + 1304.0, + 1655.0, + 396.0, + 1655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1652.0, + 1110.0, + 1652.0, + 1110.0, + 1688.0, + 394.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1807.0, + 1407.0, + 1807.0, + 1407.0, + 1847.0, + 292.0, + 1847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1842.0, + 1406.0, + 1842.0, + 1406.0, + 1874.0, + 294.0, + 1874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1870.0, + 1407.0, + 1870.0, + 1407.0, + 1905.0, + 293.0, + 1905.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1900.0, + 1407.0, + 1900.0, + 1407.0, + 1936.0, + 294.0, + 1936.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 553.0, + 1172.0, + 553.0, + 1172.0, + 590.0, + 973.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 912.0, + 578.0, + 1233.0, + 578.0, + 1233.0, + 625.0, + 912.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 616.0, + 1244.0, + 616.0, + 1244.0, + 651.0, + 899.0, + 651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 456.0, + 688.0, + 610.0, + 688.0, + 610.0, + 730.0, + 456.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 720.0, + 681.0, + 720.0, + 681.0, + 759.0, + 381.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 754.0, + 749.0, + 754.0, + 749.0, + 791.0, + 313.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 832.0, + 743.0, + 832.0, + 743.0, + 866.0, + 555.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 860.0, + 865.0, + 860.0, + 865.0, + 899.0, + 430.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 528.0, + 896.0, + 769.0, + 896.0, + 769.0, + 924.0, + 528.0, + 924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 690.0, + 1281.0, + 690.0, + 1281.0, + 726.0, + 1183.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1082.0, + 720.0, + 1383.0, + 720.0, + 1383.0, + 759.0, + 1082.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 755.0, + 1347.0, + 755.0, + 1347.0, + 787.0, + 1120.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 831.0, + 691.0, + 1005.0, + 691.0, + 1005.0, + 729.0, + 831.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 776.0, + 722.0, + 1055.0, + 722.0, + 1055.0, + 760.0, + 776.0, + 760.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 814.0, + 754.0, + 1014.0, + 754.0, + 1014.0, + 791.0, + 814.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 557.0, + 713.0, + 557.0, + 713.0, + 586.0, + 475.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 585.0, + 733.0, + 585.0, + 733.0, + 618.0, + 457.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 473.0, + 615.0, + 714.0, + 615.0, + 714.0, + 646.0, + 473.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 826.0, + 1269.0, + 826.0, + 1269.0, + 870.0, + 1009.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1022.0, + 861.0, + 1255.0, + 861.0, + 1255.0, + 899.0, + 1022.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1019.0, + 895.0, + 1260.0, + 895.0, + 1260.0, + 930.0, + 1019.0, + 930.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1504, + 1405, + 1504, + 1405, + 1814, + 297, + 1814 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 296, + 1201, + 1405, + 1201, + 1405, + 1448, + 296, + 1448 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1088, + 1405, + 1088, + 1405, + 1188, + 298, + 1188 + ], + "score": 0.977 + }, + { + "category_id": 4, + "poly": [ + 296, + 596, + 1405, + 596, + 1405, + 825, + 296, + 825 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 299, + 1828, + 1405, + 1828, + 1405, + 1955, + 299, + 1955 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 298, + 878, + 1406, + 878, + 1406, + 972, + 298, + 972 + ], + "score": 0.97 + }, + { + "category_id": 3, + "poly": [ + 351, + 204, + 1316, + 204, + 1316, + 575, + 351, + 575 + ], + "score": 0.969 + }, + { + "category_id": 8, + "poly": [ + 456, + 981, + 1244, + 981, + 1244, + 1074, + 456, + 1074 + ], + "score": 0.955 + }, + { + "category_id": 2, + "poly": [ + 333, + 1976, + 821, + 1976, + 821, + 2007, + 333, + 2007 + ], + "score": 0.921 + }, + { + "category_id": 1, + "poly": [ + 299, + 1460, + 1070, + 1460, + 1070, + 1491, + 299, + 1491 + ], + "score": 0.875 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1012, + 1400, + 1012, + 1400, + 1041, + 1366, + 1041 + ], + "score": 0.865 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.753 + }, + { + "category_id": 14, + "poly": [ + 454, + 978, + 1242, + 978, + 1242, + 1075, + 454, + 1075 + ], + "score": 0.95, + "latex": "\\operatorname* { m i n } _ { x \\in \\mathbb { R } ^ { d } } \\bigg \\{ f ( x ) : = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } f ^ { ( k ) } ( x ) : = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } \\mathbb { E } _ { \\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) } } \\big [ f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) \\big ] \\bigg \\} ." + }, + { + "category_id": 13, + "poly": [ + 768, + 652, + 869, + 652, + 869, + 687, + 768, + 687 + ], + "score": 0.93, + "latex": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1029, + 1596, + 1116, + 1596, + 1116, + 1634, + 1029, + 1634 + ], + "score": 0.93, + "latex": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )" + }, + { + "category_id": 13, + "poly": [ + 371, + 1086, + 540, + 1086, + 540, + 1121, + 371, + 1121 + ], + "score": 0.93, + "latex": "f ^ { ( k ) } : \\mathbb { R } ^ { d } \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 394, + 1920, + 481, + 1920, + 481, + 1956, + 394, + 1956 + ], + "score": 0.93, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 2 } )" + }, + { + "category_id": 13, + "poly": [ + 630, + 1919, + 739, + 1919, + 739, + 1956, + 630, + 1956 + ], + "score": 0.92, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1158, + 1122, + 1262, + 1122, + 1262, + 1157, + 1158, + 1157 + ], + "score": 0.92, + "latex": "k \\in [ K ]" + }, + { + "category_id": 13, + "poly": [ + 1267, + 1085, + 1402, + 1085, + 1402, + 1121, + 1267, + 1121 + ], + "score": 0.92, + "latex": "\\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) }" + }, + { + "category_id": 13, + "poly": [ + 1255, + 710, + 1356, + 710, + 1356, + 738, + 1255, + 738 + ], + "score": 0.91, + "latex": "\\epsilon = 1 0 ^ { - 3 }" + }, + { + "category_id": 13, + "poly": [ + 764, + 1594, + 889, + 1594, + 889, + 1636, + 764, + 1636 + ], + "score": 0.91, + "latex": "\\underline { { \\tilde { \\mathcal { O } } } } ( \\epsilon ^ { - 3 / 2 } ) \\underline { { \\left[ \\frac { 2 } { } \\right] } }" + }, + { + "category_id": 13, + "poly": [ + 519, + 656, + 575, + 656, + 575, + 686, + 519, + 686 + ], + "score": 0.91, + "latex": "( b , I )" + }, + { + "category_id": 13, + "poly": [ + 443, + 1154, + 497, + 1154, + 497, + 1185, + 443, + 1185 + ], + "score": 0.91, + "latex": "\\mathcal { D } ^ { ( k ) }" + }, + { + "category_id": 13, + "poly": [ + 861, + 1120, + 916, + 1120, + 916, + 1151, + 861, + 1151 + ], + "score": 0.9, + "latex": "\\mathcal { D } ^ { ( k ) }" + }, + { + "category_id": 13, + "poly": [ + 713, + 1232, + 765, + 1232, + 765, + 1268, + 713, + 1268 + ], + "score": 0.89, + "latex": "2 . 2 )" + }, + { + "category_id": 13, + "poly": [ + 1064, + 795, + 1119, + 795, + 1119, + 825, + 1064, + 825 + ], + "score": 0.88, + "latex": "\\mathcal { O } ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 476, + 1973, + 526, + 1973, + 526, + 2008, + 476, + 2008 + ], + "score": 0.88, + "latex": "\\tilde { \\mathcal { O } } ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 298, + 795, + 353, + 795, + 353, + 825, + 298, + 825 + ], + "score": 0.85, + "latex": "\\mathcal { O } ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 998, + 1121, + 1035, + 1121, + 1035, + 1151, + 998, + 1151 + ], + "score": 0.85, + "latex": "k ^ { \\mathrm { { t h } } }" + }, + { + "category_id": 13, + "poly": [ + 1151, + 1262, + 1200, + 1262, + 1200, + 1299, + 1151, + 1299 + ], + "score": 0.81, + "latex": "2 . 1 )" + }, + { + "category_id": 13, + "poly": [ + 293, + 1262, + 345, + 1262, + 345, + 1298, + 293, + 1298 + ], + "score": 0.79, + "latex": "2 . 3 )" + }, + { + "category_id": 13, + "poly": [ + 1308, + 913, + 1336, + 913, + 1336, + 939, + 1308, + 939 + ], + "score": 0.79, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 1156, + 1929, + 1171, + 1929, + 1171, + 1950, + 1156, + 1950 + ], + "score": 0.74, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 1070, + 662, + 1083, + 662, + 1083, + 682, + 1070, + 682 + ], + "score": 0.72, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 750, + 1271, + 765, + 1271, + 765, + 1291, + 750, + 1291 + ], + "score": 0.67, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 756, + 1632, + 781, + 1632, + 781, + 1666, + 756, + 1666 + ], + "score": 0.57, + "latex": "\\bigstar" + }, + { + "category_id": 13, + "poly": [ + 397, + 1638, + 412, + 1638, + 412, + 1659, + 397, + 1659 + ], + "score": 0.57, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 704, + 913, + 738, + 913, + 738, + 943, + 704, + 943 + ], + "score": 0.5, + "latex": "\\mathbb { \\lVert \\rVert }" + }, + { + "category_id": 13, + "poly": [ + 747, + 1691, + 772, + 1691, + 772, + 1727, + 747, + 1727 + ], + "score": 0.45, + "latex": "\\checkmark" + }, + { + "category_id": 13, + "poly": [ + 1364, + 1660, + 1395, + 1660, + 1395, + 1698, + 1364, + 1698 + ], + "score": 0.43, + "latex": "\\bigstar \\bigstar" + }, + { + "category_id": 13, + "poly": [ + 340, + 1690, + 403, + 1690, + 403, + 1723, + 340, + 1723 + ], + "score": 0.42, + "latex": "\\pm \\boxed { 5 } \\boxed { 8 } \\parallel" + }, + { + "category_id": 13, + "poly": [ + 454, + 1690, + 530, + 1690, + 530, + 1723, + 454, + 1723 + ], + "score": 0.35, + "latex": "\\mathbb { B } \\mathbb { n o }" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 594.0, + 1407.0, + 594.0, + 1407.0, + 631.0, + 293.0, + 631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 624.0, + 1403.0, + 624.0, + 1403.0, + 657.0, + 295.0, + 657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 649.0, + 518.0, + 649.0, + 518.0, + 692.0, + 289.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 649.0, + 767.0, + 649.0, + 767.0, + 692.0, + 576.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 649.0, + 1069.0, + 649.0, + 1069.0, + 692.0, + 870.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.0, + 649.0, + 1408.0, + 649.0, + 1408.0, + 692.0, + 1084.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 682.0, + 1406.0, + 682.0, + 1406.0, + 717.0, + 293.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 708.0, + 1254.0, + 708.0, + 1254.0, + 746.0, + 293.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1357.0, + 708.0, + 1407.0, + 708.0, + 1407.0, + 746.0, + 1357.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 736.0, + 1408.0, + 736.0, + 1408.0, + 775.0, + 293.0, + 775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 764.0, + 1405.0, + 764.0, + 1405.0, + 800.0, + 294.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 793.0, + 1063.0, + 793.0, + 1063.0, + 829.0, + 354.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1120.0, + 793.0, + 1356.0, + 793.0, + 1356.0, + 829.0, + 1120.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 203.0, + 769.0, + 203.0, + 769.0, + 233.0, + 409.0, + 233.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 204.0, + 1246.0, + 204.0, + 1246.0, + 229.0, + 961.0, + 229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 219.0, + 912.0, + 219.0, + 912.0, + 246.0, + 881.0, + 246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 236.0, + 938.0, + 236.0, + 938.0, + 247.0, + 928.0, + 247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 241.0, + 930.0, + 241.0, + 930.0, + 358.0, + 916.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 244.0, + 769.0, + 244.0, + 769.0, + 276.0, + 549.0, + 276.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 257.0, + 383.0, + 257.0, + 383.0, + 495.0, + 350.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 257.0, + 1050.0, + 257.0, + 1050.0, + 283.0, + 963.0, + 283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 260.0, + 421.0, + 260.0, + 421.0, + 293.0, + 368.0, + 293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 261.0, + 909.0, + 261.0, + 909.0, + 289.0, + 858.0, + 289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 290.0, + 414.0, + 290.0, + 414.0, + 313.0, + 390.0, + 313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 278.0, + 909.0, + 278.0, + 909.0, + 453.0, + 860.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 298.0, + 1281.0, + 298.0, + 1281.0, + 326.0, + 1055.0, + 326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 327.0, + 410.0, + 327.0, + 410.0, + 354.0, + 389.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 323.0, + 1268.0, + 323.0, + 1268.0, + 354.0, + 1055.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 389.0, + 369.0, + 410.0, + 369.0, + 410.0, + 394.0, + 389.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 625.0, + 360.0, + 759.0, + 360.0, + 759.0, + 385.0, + 625.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 887.0, + 351.0, + 907.0, + 351.0, + 907.0, + 370.0, + 887.0, + 370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 923.0, + 356.0, + 942.0, + 356.0, + 942.0, + 408.0, + 923.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 391.0, + 706.0, + 391.0, + 706.0, + 405.0, + 692.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 887.0, + 394.0, + 908.0, + 394.0, + 908.0, + 412.0, + 887.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 411.0, + 574.0, + 411.0, + 574.0, + 440.0, + 482.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 413.0, + 787.0, + 413.0, + 787.0, + 435.0, + 769.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 403.0, + 1264.0, + 403.0, + 1264.0, + 428.0, + 1129.0, + 428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 450.0, + 413.0, + 450.0, + 413.0, + 473.0, + 390.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 501.0, + 443.0, + 710.0, + 443.0, + 710.0, + 474.0, + 501.0, + 474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 728.0, + 445.0, + 758.0, + 445.0, + 758.0, + 472.0, + 728.0, + 472.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 437.0, + 906.0, + 437.0, + 906.0, + 454.0, + 893.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1034.0, + 461.0, + 1048.0, + 461.0, + 1048.0, + 476.0, + 1034.0, + 476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1283.0, + 461.0, + 1296.0, + 461.0, + 1296.0, + 474.0, + 1283.0, + 474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 469.0, + 412.0, + 469.0, + 412.0, + 490.0, + 393.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 471.0, + 577.0, + 471.0, + 577.0, + 532.0, + 451.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 582.0, + 479.0, + 617.0, + 479.0, + 617.0, + 511.0, + 582.0, + 511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 631.0, + 479.0, + 813.0, + 479.0, + 813.0, + 530.0, + 631.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 892.0, + 478.0, + 918.0, + 478.0, + 918.0, + 513.0, + 892.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 967.0, + 491.0, + 983.0, + 491.0, + 983.0, + 515.0, + 967.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1027.0, + 488.0, + 1186.0, + 488.0, + 1186.0, + 538.0, + 1027.0, + 538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1221.0, + 489.0, + 1252.0, + 489.0, + 1252.0, + 516.0, + 1221.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1253.0, + 476.0, + 1283.0, + 476.0, + 1283.0, + 483.0, + 1253.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1286.0, + 490.0, + 1317.0, + 490.0, + 1317.0, + 516.0, + 1286.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 435.0, + 538.0, + 760.0, + 538.0, + 760.0, + 573.0, + 435.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 546.0, + 1293.0, + 546.0, + 1293.0, + 580.0, + 908.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 944.25, + 419.5, + 997.25, + 419.5, + 997.25, + 452.0, + 944.25, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1968.0, + 475.0, + 1968.0, + 475.0, + 2013.0, + 329.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 527.0, + 1968.0, + 828.0, + 1968.0, + 828.0, + 2013.0, + 527.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2093.0, + 838.0, + 2093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1505.0, + 1406.0, + 1505.0, + 1406.0, + 1541.0, + 294.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1537.0, + 1403.0, + 1537.0, + 1403.0, + 1569.0, + 295.0, + 1569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1564.0, + 1405.0, + 1564.0, + 1405.0, + 1601.0, + 294.0, + 1601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1598.0, + 763.0, + 1598.0, + 763.0, + 1635.0, + 292.0, + 1635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 1598.0, + 1028.0, + 1598.0, + 1028.0, + 1635.0, + 890.0, + 1635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1117.0, + 1598.0, + 1405.0, + 1598.0, + 1405.0, + 1635.0, + 1117.0, + 1635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1631.0, + 396.0, + 1631.0, + 396.0, + 1667.0, + 294.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 413.0, + 1631.0, + 755.0, + 1631.0, + 755.0, + 1667.0, + 413.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 782.0, + 1631.0, + 1406.0, + 1631.0, + 1406.0, + 1667.0, + 782.0, + 1667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1659.0, + 1363.0, + 1659.0, + 1363.0, + 1697.0, + 292.0, + 1697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1396.0, + 1659.0, + 1405.0, + 1659.0, + 1405.0, + 1697.0, + 1396.0, + 1697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1688.0, + 339.0, + 1688.0, + 339.0, + 1727.0, + 296.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1688.0, + 453.0, + 1688.0, + 453.0, + 1727.0, + 404.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 1688.0, + 746.0, + 1688.0, + 746.0, + 1727.0, + 531.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 773.0, + 1688.0, + 1405.0, + 1688.0, + 1405.0, + 1727.0, + 773.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1723.0, + 1403.0, + 1723.0, + 1403.0, + 1755.0, + 296.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1754.0, + 1403.0, + 1754.0, + 1403.0, + 1786.0, + 296.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1782.0, + 664.0, + 1782.0, + 664.0, + 1817.0, + 295.0, + 1817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1199.0, + 1407.0, + 1199.0, + 1407.0, + 1238.0, + 293.0, + 1238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1232.0, + 712.0, + 1232.0, + 712.0, + 1267.0, + 295.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1234.0, + 1405.0, + 1234.0, + 1405.0, + 1264.0, + 766.0, + 1264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 1263.0, + 749.0, + 1263.0, + 749.0, + 1298.0, + 346.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 1263.0, + 1150.0, + 1263.0, + 1150.0, + 1298.0, + 766.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1201.0, + 1263.0, + 1405.0, + 1263.0, + 1405.0, + 1298.0, + 1201.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1294.0, + 1405.0, + 1294.0, + 1405.0, + 1328.0, + 294.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1325.0, + 1405.0, + 1325.0, + 1405.0, + 1358.0, + 294.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1351.0, + 1407.0, + 1351.0, + 1407.0, + 1390.0, + 293.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1384.0, + 1406.0, + 1384.0, + 1406.0, + 1421.0, + 295.0, + 1421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1416.0, + 806.0, + 1416.0, + 806.0, + 1450.0, + 295.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1081.0, + 370.0, + 1081.0, + 370.0, + 1125.0, + 291.0, + 1125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 1081.0, + 1266.0, + 1081.0, + 1266.0, + 1125.0, + 541.0, + 1125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1403.0, + 1081.0, + 1407.0, + 1081.0, + 1407.0, + 1125.0, + 1403.0, + 1125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1119.0, + 860.0, + 1119.0, + 860.0, + 1158.0, + 292.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 917.0, + 1119.0, + 997.0, + 1119.0, + 997.0, + 1158.0, + 917.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 1119.0, + 1157.0, + 1119.0, + 1157.0, + 1158.0, + 1036.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 1119.0, + 1405.0, + 1119.0, + 1405.0, + 1158.0, + 1263.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1150.0, + 442.0, + 1150.0, + 442.0, + 1196.0, + 293.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 1150.0, + 1381.0, + 1150.0, + 1381.0, + 1196.0, + 498.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1826.0, + 1406.0, + 1826.0, + 1406.0, + 1863.0, + 293.0, + 1863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1855.0, + 1410.0, + 1855.0, + 1410.0, + 1895.0, + 293.0, + 1895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1890.0, + 1406.0, + 1890.0, + 1406.0, + 1920.0, + 297.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1919.0, + 393.0, + 1919.0, + 393.0, + 1959.0, + 292.0, + 1959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 1919.0, + 629.0, + 1919.0, + 629.0, + 1959.0, + 482.0, + 1959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1919.0, + 1155.0, + 1919.0, + 1155.0, + 1959.0, + 740.0, + 1959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 1919.0, + 1398.0, + 1919.0, + 1398.0, + 1959.0, + 1172.0, + 1959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 879.0, + 1404.0, + 879.0, + 1404.0, + 913.0, + 297.0, + 913.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 912.0, + 703.0, + 912.0, + 703.0, + 943.0, + 293.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 739.0, + 912.0, + 1307.0, + 912.0, + 1307.0, + 943.0, + 739.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1337.0, + 912.0, + 1404.0, + 912.0, + 1404.0, + 943.0, + 1337.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 942.0, + 447.0, + 942.0, + 447.0, + 974.0, + 295.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1458.0, + 1072.0, + 1458.0, + 1072.0, + 1495.0, + 296.0, + 1495.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 296, + 196, + 1413, + 196, + 1413, + 588, + 296, + 588 + ], + "score": 0.984, + "html": "
AlgorithmWorkSampleComm.Minibatch (b)Local Updates (I) /round
FedAvg国园 国国0(€-2)0(c-3/2) 0(c-2)0(1) 0(1) 2(1-v)0(c-1/2) 0(1) 3v
SCAFFOLD*this work 国0(c-2)O(e-3/2) 0(c-2)O(c 4-v) 0(1)O(c−2(4-D)) 0(1)
FedPD/FedProx*四/□O(c-2)0(e-1)0(1)0(e-1)
MIME†/FedGLOMO/80(c-3/2)O(€-3/2)0(1)0(1)
STEM Fed STEM Minibatch STEM* this workO(€-3/2)O(e-1)( 0(1) O(e-1/2)O(∈−(3)) O(∈-1/2) 0(1)
" + }, + { + "category_id": 1, + "poly": [ + 297, + 968, + 1405, + 968, + 1405, + 1637, + 297, + 1637 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 1649, + 1404, + 1649, + 1404, + 1842, + 298, + 1842 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1853, + 1403, + 1853, + 1403, + 2007, + 298, + 2007 + ], + "score": 0.974 + }, + { + "category_id": 7, + "poly": [ + 295, + 603, + 1406, + 603, + 1406, + 802, + 295, + 802 + ], + "score": 0.952 + }, + { + "category_id": 1, + "poly": [ + 300, + 892, + 1401, + 892, + 1401, + 954, + 300, + 954 + ], + "score": 0.946 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 858, + 2061, + 858, + 2085, + 841, + 2085 + ], + "score": 0.643 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.35 + }, + { + "category_id": 13, + "poly": [ + 1004, + 1776, + 1114, + 1776, + 1114, + 1812, + 1004, + 1812 + ], + "score": 0.96, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1007, + 1331, + 1094, + 1331, + 1094, + 1365, + 1007, + 1365 + ], + "score": 0.93, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1316, + 1240, + 1403, + 1240, + 1403, + 1275, + 1316, + 1275 + ], + "score": 0.93, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 1 } )" + }, + { + "category_id": 13, + "poly": [ + 1310, + 1422, + 1396, + 1422, + 1396, + 1456, + 1310, + 1456 + ], + "score": 0.93, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 1 } )" + }, + { + "category_id": 13, + "poly": [ + 628, + 1544, + 714, + 1544, + 714, + 1579, + 628, + 1579 + ], + "score": 0.93, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 1 } )" + }, + { + "category_id": 13, + "poly": [ + 1294, + 1119, + 1403, + 1119, + 1403, + 1154, + 1294, + 1154 + ], + "score": 0.93, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1062, + 1119, + 1149, + 1119, + 1149, + 1154, + 1062, + 1154 + ], + "score": 0.93, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1225, + 405, + 1305, + 405, + 1305, + 435, + 1225, + 435 + ], + "score": 0.93, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 1 } )" + }, + { + "category_id": 13, + "poly": [ + 706, + 294, + 785, + 294, + 785, + 322, + 706, + 322 + ], + "score": 0.92, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 2 } )" + }, + { + "category_id": 13, + "poly": [ + 706, + 370, + 785, + 370, + 785, + 400, + 706, + 400 + ], + "score": 0.92, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 2 } )" + }, + { + "category_id": 13, + "poly": [ + 706, + 404, + 785, + 404, + 785, + 434, + 706, + 434 + ], + "score": 0.92, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1007, + 408, + 1062, + 408, + 1062, + 435, + 1007, + 435 + ], + "score": 0.91, + "latex": "\\mathcal { O } ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 841, + 404, + 921, + 404, + 921, + 434, + 841, + 434 + ], + "score": 0.91, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 1 } )" + }, + { + "category_id": 13, + "poly": [ + 696, + 438, + 796, + 438, + 796, + 468, + 696, + 468 + ], + "score": 0.91, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 832, + 439, + 931, + 439, + 931, + 468, + 832, + 468 + ], + "score": 0.91, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 397, + 1712, + 506, + 1712, + 506, + 1747, + 397, + 1747 + ], + "score": 0.91, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 831, + 336, + 932, + 336, + 932, + 366, + 831, + 366 + ], + "score": 0.91, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1007, + 375, + 1062, + 375, + 1062, + 401, + 1007, + 401 + ], + "score": 0.91, + "latex": "\\mathcal { O } ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 1238, + 374, + 1292, + 374, + 1292, + 401, + 1238, + 401 + ], + "score": 0.91, + "latex": "\\mathcal { O } ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 297, + 661, + 411, + 661, + 411, + 691, + 297, + 691 + ], + "score": 0.9, + "latex": "^ \\circ \\nu \\in [ 0 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 842, + 370, + 921, + 370, + 921, + 400, + 842, + 400 + ], + "score": 0.9, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1007, + 264, + 1061, + 264, + 1061, + 290, + 1007, + 290 + ], + "score": 0.9, + "latex": "\\mathcal { O } ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 841, + 294, + 921, + 294, + 921, + 323, + 841, + 323 + ], + "score": 0.9, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1007, + 442, + 1062, + 442, + 1062, + 469, + 1007, + 469 + ], + "score": 0.89, + "latex": "\\mathcal { O } ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 703, + 773, + 767, + 773, + 767, + 797, + 703, + 797 + ], + "score": 0.89, + "latex": "I = 1" + }, + { + "category_id": 13, + "poly": [ + 832, + 260, + 932, + 260, + 932, + 290, + 832, + 290 + ], + "score": 0.89, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 768, + 1303, + 828, + 1303, + 828, + 1336, + 768, + 1336 + ], + "score": 0.88, + "latex": "\\mathcal { O } ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 606, + 662, + 674, + 662, + 674, + 687, + 606, + 687 + ], + "score": 0.88, + "latex": "\\nu = 1" + }, + { + "category_id": 13, + "poly": [ + 1237, + 442, + 1292, + 442, + 1292, + 469, + 1237, + 469 + ], + "score": 0.88, + "latex": "\\mathcal { O } ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 1194, + 329, + 1333, + 329, + 1333, + 368, + 1194, + 368 + ], + "score": 0.87, + "latex": "\\underline { { \\mathcal { O } \\big ( \\epsilon ^ { - } ^ { 3 \\nu } \\big ) } } \\bigg )" + }, + { + "category_id": 13, + "poly": [ + 1007, + 297, + 1062, + 297, + 1062, + 324, + 1007, + 324 + ], + "score": 0.87, + "latex": "\\mathcal { O } ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 964, + 326, + 1102, + 326, + 1102, + 368, + 964, + 368 + ], + "score": 0.87, + "latex": "\\mathcal { O } \\big ( \\epsilon ^ { - \\frac { 2 ( 1 - \\nu ) } { ( 4 - \\nu ) } } \\big )" + }, + { + "category_id": 13, + "poly": [ + 743, + 662, + 811, + 662, + 811, + 687, + 743, + 687 + ], + "score": 0.87, + "latex": "\\nu = 0" + }, + { + "category_id": 13, + "poly": [ + 1215, + 259, + 1315, + 259, + 1315, + 290, + 1215, + 290 + ], + "score": 0.86, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 1 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 699, + 515, + 799, + 515, + 799, + 547, + 699, + 547 + ], + "score": 0.86, + "latex": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1285, + 660, + 1341, + 660, + 1341, + 691, + 1285, + 691 + ], + "score": 0.86, + "latex": "\\mathcal { O } ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 1235, + 1363, + 1283, + 1363, + 1283, + 1394, + 1235, + 1394 + ], + "score": 0.85, + "latex": "\\pmb { \\mathbb { B 3 } }" + }, + { + "category_id": 13, + "poly": [ + 450, + 1944, + 498, + 1944, + 498, + 1976, + 450, + 1976 + ], + "score": 0.85, + "latex": "\\pmb { \\Vert 2 4 \\Vert }" + }, + { + "category_id": 13, + "poly": [ + 1238, + 297, + 1292, + 297, + 1292, + 324, + 1238, + 324 + ], + "score": 0.83, + "latex": "\\mathcal { O } ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 881, + 1887, + 927, + 1887, + 927, + 1916, + 881, + 1916 + ], + "score": 0.83, + "latex": "\\mathbb { \\lVert 1 9 \\rVert }" + }, + { + "category_id": 13, + "poly": [ + 845, + 516, + 925, + 516, + 925, + 547, + 845, + 547 + ], + "score": 0.82, + "latex": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )" + }, + { + "category_id": 13, + "poly": [ + 1349, + 1180, + 1396, + 1180, + 1396, + 1213, + 1349, + 1213 + ], + "score": 0.81, + "latex": "\\mathbb { \\lVert 1 2 \\rVert }" + }, + { + "category_id": 13, + "poly": [ + 975, + 1150, + 1022, + 1150, + 1022, + 1183, + 975, + 1183 + ], + "score": 0.81, + "latex": "[ \\bar { \\lVert { 4 } } ]" + }, + { + "category_id": 13, + "poly": [ + 327, + 1913, + 375, + 1913, + 375, + 1946, + 327, + 1946 + ], + "score": 0.8, + "latex": "\\mathbb { \\lVert \\rVert 3 \\rVert }" + }, + { + "category_id": 13, + "poly": [ + 979, + 1271, + 1069, + 1271, + 1069, + 1303, + 979, + 1303 + ], + "score": 0.79, + "latex": "\\boxed { 1 5 } , \\boxed { 1 6 }" + }, + { + "category_id": 13, + "poly": [ + 964, + 471, + 1103, + 471, + 1103, + 516, + 964, + 516 + ], + "score": 0.79, + "latex": "\\mathcal { O } \\big ( \\epsilon ^ { - \\frac { 3 ( 1 - \\nu ) } { 2 ( 3 - \\nu ) } } \\big )" + }, + { + "category_id": 13, + "poly": [ + 1031, + 661, + 1089, + 661, + 1089, + 691, + 1031, + 691 + ], + "score": 0.75, + "latex": "\\mathcal { O } ( 1 ) ." + }, + { + "category_id": 13, + "poly": [ + 1189, + 1484, + 1239, + 1484, + 1239, + 1515, + 1189, + 1515 + ], + "score": 0.75, + "latex": "\\mathbb { \\ m }" + }, + { + "category_id": 13, + "poly": [ + 1007, + 519, + 1062, + 519, + 1062, + 546, + 1007, + 546 + ], + "score": 0.73, + "latex": "\\mathcal { O } ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 481, + 1362, + 529, + 1362, + 529, + 1395, + 481, + 1395 + ], + "score": 0.72, + "latex": "\\bar { \\| 1 5 \\| }" + }, + { + "category_id": 13, + "poly": [ + 498, + 1681, + 546, + 1681, + 546, + 1712, + 498, + 1712 + ], + "score": 0.71, + "latex": "\\mathbb { \\lVert 1 7 \\rVert }" + }, + { + "category_id": 13, + "poly": [ + 1073, + 716, + 1105, + 716, + 1105, + 745, + 1073, + 745 + ], + "score": 0.71, + "latex": "\\bigstar \\bigstar" + }, + { + "category_id": 13, + "poly": [ + 518, + 662, + 533, + 662, + 533, + 686, + 518, + 686 + ], + "score": 0.69, + "latex": "^ { b }" + }, + { + "category_id": 13, + "poly": [ + 579, + 662, + 596, + 662, + 596, + 686, + 579, + 686 + ], + "score": 0.64, + "latex": "I" + }, + { + "category_id": 13, + "poly": [ + 1199, + 478, + 1329, + 478, + 1329, + 515, + 1199, + 515 + ], + "score": 0.63, + "latex": "\\mathcal { O } \\bigl ( \\epsilon ^ { - } \\frac { \\nu } { \\left( 3 - \\nu \\right) } \\bigr )" + }, + { + "category_id": 13, + "poly": [ + 1203, + 634, + 1237, + 634, + 1237, + 664, + 1203, + 664 + ], + "score": 0.62, + "latex": "( I )" + }, + { + "category_id": 13, + "poly": [ + 421, + 1815, + 435, + 1815, + 435, + 1836, + 421, + 1836 + ], + "score": 0.61, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 846, + 1808, + 871, + 1808, + 871, + 1845, + 846, + 1845 + ], + "score": 0.56, + "latex": "^ 1" + }, + { + "category_id": 13, + "poly": [ + 805, + 1090, + 853, + 1090, + 853, + 1121, + 805, + 1121 + ], + "score": 0.54, + "latex": "\\bar { \\mathbb { E } 2 } \\mathbb { I }" + }, + { + "category_id": 13, + "poly": [ + 708, + 1158, + 723, + 1158, + 723, + 1179, + 708, + 1179 + ], + "score": 0.51, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 297, + 1423, + 347, + 1423, + 347, + 1455, + 297, + 1455 + ], + "score": 0.5, + "latex": "\\left[ \\left[ 2 9 \\right] \\right]" + }, + { + "category_id": 13, + "poly": [ + 353, + 634, + 384, + 634, + 384, + 662, + 353, + 662 + ], + "score": 0.48, + "latex": "( b )" + }, + { + "category_id": 13, + "poly": [ + 935, + 712, + 960, + 712, + 960, + 748, + 935, + 748 + ], + "score": 0.46, + "latex": "2" + }, + { + "category_id": 13, + "poly": [ + 1206, + 516, + 1316, + 516, + 1316, + 546, + 1206, + 546 + ], + "score": 0.44, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 1 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 553, + 1605, + 587, + 1605, + 587, + 1637, + 553, + 1637 + ], + "score": 0.42, + "latex": "\\pmb { \\mathbb { B } }" + }, + { + "category_id": 13, + "poly": [ + 542, + 1544, + 587, + 1544, + 587, + 1577, + 542, + 1577 + ], + "score": 0.38, + "latex": "\\mathbb { L } 2 \\mathbb { I }" + }, + { + "category_id": 13, + "poly": [ + 983, + 548, + 1083, + 548, + 1083, + 580, + 983, + 580 + ], + "score": 0.36, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 1 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1288, + 211, + 1321, + 211, + 1321, + 241, + 1288, + 241 + ], + "score": 0.34, + "latex": "( I )" + }, + { + "category_id": 13, + "poly": [ + 460, + 1746, + 509, + 1746, + 509, + 1777, + 460, + 1777 + ], + "score": 0.27, + "latex": "[ \\overline { { 1 8 } } ]" + }, + { + "category_id": 13, + "poly": [ + 1200, + 474, + 1328, + 474, + 1328, + 583, + 1200, + 583 + ], + "score": 0.26, + "latex": "\\begin{array} { r l } { { \\mathcal O \\bigl ( \\epsilon ^ { - \\frac { \\nu } { ( 3 - \\nu ) } } \\bigr ) } } \\\\ & { { } \\quad \\mathcal O ( \\epsilon ^ { - 1 / 2 } ) } \\\\ & { { } \\quad \\mathcal O ( 1 ) } \\end{array}" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 601.0, + 1406.0, + 601.0, + 1406.0, + 640.0, + 293.0, + 640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 633.0, + 352.0, + 633.0, + 352.0, + 666.0, + 294.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 633.0, + 1202.0, + 633.0, + 1202.0, + 666.0, + 385.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1238.0, + 633.0, + 1243.0, + 633.0, + 1243.0, + 666.0, + 1238.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 653.0, + 296.0, + 653.0, + 296.0, + 697.0, + 293.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.0, + 653.0, + 517.0, + 653.0, + 517.0, + 697.0, + 412.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 653.0, + 578.0, + 653.0, + 578.0, + 697.0, + 534.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 653.0, + 605.0, + 653.0, + 605.0, + 697.0, + 597.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 675.0, + 653.0, + 742.0, + 653.0, + 742.0, + 697.0, + 675.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 653.0, + 1030.0, + 653.0, + 1030.0, + 697.0, + 812.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 653.0, + 1284.0, + 653.0, + 1284.0, + 697.0, + 1090.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1342.0, + 653.0, + 1411.0, + 653.0, + 1411.0, + 697.0, + 1342.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 686.0, + 1247.0, + 686.0, + 1247.0, + 720.0, + 294.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 710.0, + 934.0, + 710.0, + 934.0, + 752.0, + 293.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 961.0, + 710.0, + 1072.0, + 710.0, + 1072.0, + 752.0, + 961.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1106.0, + 710.0, + 1229.0, + 710.0, + 1229.0, + 752.0, + 1106.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 737.0, + 1129.0, + 737.0, + 1129.0, + 779.0, + 294.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 768.0, + 702.0, + 768.0, + 702.0, + 805.0, + 297.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 768.0, + 1366.0, + 768.0, + 1366.0, + 805.0, + 768.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2058.0, + 860.0, + 2058.0, + 860.0, + 2089.0, + 839.0, + 2089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 967.0, + 1405.0, + 967.0, + 1405.0, + 1003.0, + 292.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1000.0, + 1402.0, + 1000.0, + 1402.0, + 1032.0, + 296.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1028.0, + 1405.0, + 1028.0, + 1405.0, + 1065.0, + 292.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1060.0, + 1405.0, + 1060.0, + 1405.0, + 1095.0, + 295.0, + 1095.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1088.0, + 804.0, + 1088.0, + 804.0, + 1124.0, + 292.0, + 1124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 854.0, + 1088.0, + 1405.0, + 1088.0, + 1405.0, + 1124.0, + 854.0, + 1124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1117.0, + 1061.0, + 1117.0, + 1061.0, + 1157.0, + 291.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1150.0, + 1117.0, + 1293.0, + 1117.0, + 1293.0, + 1157.0, + 1150.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 1117.0, + 1407.0, + 1117.0, + 1407.0, + 1157.0, + 1404.0, + 1157.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1149.0, + 707.0, + 1149.0, + 707.0, + 1186.0, + 292.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 1149.0, + 974.0, + 1149.0, + 974.0, + 1186.0, + 724.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1023.0, + 1149.0, + 1408.0, + 1149.0, + 1408.0, + 1186.0, + 1023.0, + 1186.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1180.0, + 1348.0, + 1180.0, + 1348.0, + 1215.0, + 294.0, + 1215.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1397.0, + 1180.0, + 1407.0, + 1180.0, + 1407.0, + 1215.0, + 1397.0, + 1215.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1210.0, + 1407.0, + 1210.0, + 1407.0, + 1244.0, + 292.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1239.0, + 1315.0, + 1239.0, + 1315.0, + 1278.0, + 294.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1271.0, + 978.0, + 1271.0, + 978.0, + 1306.0, + 296.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1070.0, + 1271.0, + 1407.0, + 1271.0, + 1407.0, + 1306.0, + 1070.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1301.0, + 767.0, + 1301.0, + 767.0, + 1337.0, + 295.0, + 1337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 1301.0, + 1406.0, + 1301.0, + 1406.0, + 1337.0, + 829.0, + 1337.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1332.0, + 1006.0, + 1332.0, + 1006.0, + 1367.0, + 295.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1095.0, + 1332.0, + 1406.0, + 1332.0, + 1406.0, + 1367.0, + 1095.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1361.0, + 480.0, + 1361.0, + 480.0, + 1399.0, + 290.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 530.0, + 1361.0, + 1234.0, + 1361.0, + 1234.0, + 1399.0, + 530.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1284.0, + 1361.0, + 1405.0, + 1361.0, + 1405.0, + 1399.0, + 1284.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1392.0, + 1406.0, + 1392.0, + 1406.0, + 1428.0, + 294.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1420.0, + 1309.0, + 1420.0, + 1309.0, + 1459.0, + 348.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1397.0, + 1420.0, + 1408.0, + 1420.0, + 1408.0, + 1459.0, + 1397.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1452.0, + 1407.0, + 1452.0, + 1407.0, + 1490.0, + 294.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1485.0, + 1188.0, + 1485.0, + 1188.0, + 1519.0, + 292.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1240.0, + 1485.0, + 1405.0, + 1485.0, + 1405.0, + 1519.0, + 1240.0, + 1519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1514.0, + 1405.0, + 1514.0, + 1405.0, + 1549.0, + 294.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1543.0, + 541.0, + 1543.0, + 541.0, + 1581.0, + 292.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 1543.0, + 627.0, + 1543.0, + 627.0, + 1581.0, + 588.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 1543.0, + 1405.0, + 1543.0, + 1405.0, + 1581.0, + 715.0, + 1581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1573.0, + 1407.0, + 1573.0, + 1407.0, + 1612.0, + 292.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1605.0, + 552.0, + 1605.0, + 552.0, + 1639.0, + 293.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 1605.0, + 597.0, + 1605.0, + 597.0, + 1639.0, + 588.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1649.0, + 1406.0, + 1649.0, + 1406.0, + 1685.0, + 294.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1679.0, + 497.0, + 1679.0, + 497.0, + 1718.0, + 293.0, + 1718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1679.0, + 1406.0, + 1679.0, + 1406.0, + 1718.0, + 547.0, + 1718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1705.0, + 396.0, + 1705.0, + 396.0, + 1755.0, + 291.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 507.0, + 1705.0, + 1410.0, + 1705.0, + 1410.0, + 1755.0, + 507.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1743.0, + 459.0, + 1743.0, + 459.0, + 1782.0, + 293.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 1743.0, + 1406.0, + 1743.0, + 1406.0, + 1782.0, + 510.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1773.0, + 1003.0, + 1773.0, + 1003.0, + 1814.0, + 293.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1115.0, + 1773.0, + 1408.0, + 1773.0, + 1408.0, + 1814.0, + 1115.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1810.0, + 420.0, + 1810.0, + 420.0, + 1842.0, + 296.0, + 1842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1810.0, + 845.0, + 1810.0, + 845.0, + 1842.0, + 436.0, + 1842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 1810.0, + 1310.0, + 1810.0, + 1310.0, + 1842.0, + 872.0, + 1842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1855.0, + 1404.0, + 1855.0, + 1404.0, + 1888.0, + 297.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1880.0, + 880.0, + 1880.0, + 880.0, + 1921.0, + 293.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 1880.0, + 1405.0, + 1880.0, + 1405.0, + 1921.0, + 928.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1911.0, + 326.0, + 1911.0, + 326.0, + 1950.0, + 292.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 1911.0, + 1407.0, + 1911.0, + 1407.0, + 1950.0, + 376.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1942.0, + 449.0, + 1942.0, + 449.0, + 1980.0, + 293.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 499.0, + 1942.0, + 1408.0, + 1942.0, + 1408.0, + 1980.0, + 499.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1976.0, + 1404.0, + 1976.0, + 1404.0, + 2009.0, + 296.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 889.0, + 1405.0, + 889.0, + 1405.0, + 931.0, + 294.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 923.0, + 614.0, + 923.0, + 614.0, + 958.0, + 295.0, + 958.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 998, + 1406, + 998, + 1406, + 1254, + 296, + 1254 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 278, + 1405, + 278, + 1405, + 472, + 297, + 472 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 297, + 1604, + 1405, + 1604, + 1405, + 1732, + 297, + 1732 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 297, + 1437, + 1405, + 1437, + 1405, + 1600, + 297, + 1600 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 297, + 1335, + 1406, + 1335, + 1406, + 1430, + 297, + 1430 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 298, + 1839, + 1405, + 1839, + 1405, + 1932, + 298, + 1932 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 298, + 1266, + 1399, + 1266, + 1399, + 1328, + 298, + 1328 + ], + "score": 0.945 + }, + { + "category_id": 1, + "poly": [ + 292, + 201, + 1403, + 201, + 1403, + 266, + 292, + 266 + ], + "score": 0.945 + }, + { + "category_id": 1, + "poly": [ + 294, + 621, + 1401, + 621, + 1401, + 690, + 294, + 690 + ], + "score": 0.941 + }, + { + "category_id": 0, + "poly": [ + 298, + 1769, + 1026, + 1769, + 1026, + 1809, + 298, + 1809 + ], + "score": 0.94 + }, + { + "category_id": 1, + "poly": [ + 301, + 756, + 1402, + 756, + 1402, + 820, + 301, + 820 + ], + "score": 0.938 + }, + { + "category_id": 1, + "poly": [ + 297, + 1942, + 1398, + 1942, + 1398, + 2009, + 297, + 2009 + ], + "score": 0.927 + }, + { + "category_id": 0, + "poly": [ + 298, + 510, + 544, + 510, + 544, + 548, + 298, + 548 + ], + "score": 0.923 + }, + { + "category_id": 1, + "poly": [ + 292, + 578, + 1331, + 578, + 1331, + 612, + 292, + 612 + ], + "score": 0.92 + }, + { + "category_id": 1, + "poly": [ + 292, + 883, + 1088, + 883, + 1088, + 916, + 292, + 916 + ], + "score": 0.888 + }, + { + "category_id": 8, + "poly": [ + 440, + 702, + 1254, + 702, + 1254, + 746, + 440, + 746 + ], + "score": 0.851 + }, + { + "category_id": 8, + "poly": [ + 294, + 928, + 1415, + 928, + 1415, + 973, + 294, + 973 + ], + "score": 0.844 + }, + { + "category_id": 8, + "poly": [ + 511, + 832, + 1187, + 832, + 1187, + 875, + 511, + 875 + ], + "score": 0.84 + }, + { + "category_id": 2, + "poly": [ + 841, + 2062, + 858, + 2062, + 858, + 2084, + 841, + 2084 + ], + "score": 0.774 + }, + { + "category_id": 13, + "poly": [ + 638, + 437, + 762, + 437, + 762, + 471, + 638, + 471 + ], + "score": 0.94, + "latex": "\\{ 1 , \\ldots , T \\}" + }, + { + "category_id": 13, + "poly": [ + 916, + 437, + 1062, + 437, + 1062, + 471, + 916, + 471 + ], + "score": 0.94, + "latex": "{ \\dot { \\mathcal { U } } } \\{ 1 , \\dots , T \\}" + }, + { + "category_id": 13, + "poly": [ + 698, + 1468, + 844, + 1468, + 844, + 1504, + 698, + 1504 + ], + "score": 0.93, + "latex": "\\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) }" + }, + { + "category_id": 13, + "poly": [ + 1210, + 619, + 1344, + 619, + 1344, + 657, + 1210, + 657 + ], + "score": 0.93, + "latex": "f ^ { ( k ) } ( \\cdot , \\xi ^ { ( k ) } )" + }, + { + "category_id": 13, + "poly": [ + 298, + 654, + 433, + 654, + 433, + 690, + 298, + 690 + ], + "score": 0.93, + "latex": "\\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) }" + }, + { + "category_id": 13, + "poly": [ + 716, + 1220, + 832, + 1220, + 832, + 1254, + 716, + 1254 + ], + "score": 0.93, + "latex": "k , \\ell \\in [ K ]" + }, + { + "category_id": 13, + "poly": [ + 513, + 657, + 605, + 657, + 605, + 691, + 513, + 691 + ], + "score": 0.93, + "latex": "k \\in [ K ]" + }, + { + "category_id": 13, + "poly": [ + 495, + 1217, + 637, + 1217, + 637, + 1249, + 495, + 1249 + ], + "score": 0.92, + "latex": "\\mathcal { D } ^ { ( k ) } \\stackrel { = } { = } \\mathcal { D } ^ { ( \\ell ) }" + }, + { + "category_id": 13, + "poly": [ + 962, + 1002, + 1011, + 1002, + 1011, + 1037, + 962, + 1037 + ], + "score": 0.92, + "latex": "f ^ { ( k ) }" + }, + { + "category_id": 13, + "poly": [ + 702, + 374, + 941, + 374, + 941, + 410, + 702, + 410 + ], + "score": 0.92, + "latex": "[ K ] \\triangleq \\{ 1 , 2 , \\dots , K \\}" + }, + { + "category_id": 13, + "poly": [ + 610, + 1157, + 687, + 1157, + 687, + 1189, + 610, + 1189 + ], + "score": 0.92, + "latex": "\\zeta > 0" + }, + { + "category_id": 13, + "poly": [ + 1097, + 1334, + 1269, + 1334, + 1269, + 1369, + 1097, + 1369 + ], + "score": 0.91, + "latex": "\\| \\nabla f ( x ) \\| ^ { 2 } \\leq \\epsilon" + }, + { + "category_id": 13, + "poly": [ + 988, + 1221, + 1056, + 1221, + 1056, + 1253, + 988, + 1253 + ], + "score": 0.91, + "latex": "\\zeta = 0" + }, + { + "category_id": 13, + "poly": [ + 298, + 406, + 349, + 406, + 349, + 440, + 298, + 440 + ], + "score": 0.91, + "latex": "\\langle \\cdot , \\cdot \\rangle" + }, + { + "category_id": 13, + "poly": [ + 1063, + 376, + 1115, + 376, + 1115, + 410, + 1063, + 410 + ], + "score": 0.91, + "latex": "\\| \\cdot \\|" + }, + { + "category_id": 13, + "poly": [ + 819, + 1977, + 847, + 1977, + 847, + 2009, + 819, + 2009 + ], + "score": 0.91, + "latex": "d _ { t } ^ { k }" + }, + { + "category_id": 13, + "poly": [ + 1113, + 279, + 1175, + 279, + 1175, + 312, + 1113, + 312 + ], + "score": 0.91, + "latex": "\\mathbb { E } [ X ]" + }, + { + "category_id": 14, + "poly": [ + 310, + 927, + 1421, + 927, + 1421, + 973, + 310, + 973 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { \\mathbb { \\tilde { z } } \\| \\nabla f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) - \\nabla f ^ { ( k ) } ( x ) \\| ^ { 2 } \\leq \\sigma ^ { 2 } , \\| \\nabla f ^ { ( k ) } ( x ) - \\nabla f ^ { ( \\ell ) } ( x ) \\| ^ { 2 } \\leq \\zeta ^ { 2 } , \\forall \\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) } , \\forall k , \\ell \\in [ K ] . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1183, + 1366, + 1399, + 1366, + 1399, + 1401, + 1183, + 1401 + ], + "score": 0.9, + "latex": "\\begin{array} { r } { \\ddot { \\mathbb { E } } [ \\| \\nabla f ( x _ { t } ) \\| ^ { 2 } ] \\le \\epsilon . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 297, + 1501, + 636, + 1501, + 636, + 1538, + 297, + 1538 + ], + "score": 0.9, + "latex": "( f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) , \\nabla f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) )" + }, + { + "category_id": 14, + "poly": [ + 515, + 831, + 1184, + 831, + 1184, + 874, + 515, + 874 + ], + "score": 0.9, + "latex": "\\mathbb { E } [ \\nabla f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) ] = \\nabla f ^ { ( k ) } ( x ) , \\forall \\xi ^ { ( k ) } \\sim \\mathcal { D } ^ { ( k ) } , \\forall k \\in [ K ] ." + }, + { + "category_id": 13, + "poly": [ + 480, + 376, + 565, + 376, + 565, + 405, + 480, + 405 + ], + "score": 0.9, + "latex": "K \\in \\mathbb N" + }, + { + "category_id": 14, + "poly": [ + 443, + 702, + 1255, + 702, + 1255, + 746, + 443, + 746 + ], + "score": 0.89, + "latex": "\\begin{array} { r } { \\mathbb { E } \\| \\nabla f ^ { ( k ) } ( x ; \\xi ^ { ( k ) } ) - \\nabla f ^ { ( k ) } ( y ; \\xi ^ { ( k ) } ) \\| ^ { 2 } \\leq L ^ { 2 } \\| x - y \\| ^ { 2 } \\mathrm { ~ f o r ~ a l l ~ } x , y \\in \\mathbb { R } ^ { d } . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1255, + 377, + 1282, + 377, + 1282, + 407, + 1255, + 407 + ], + "score": 0.88, + "latex": "\\ell _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 926, + 1469, + 962, + 1469, + 962, + 1500, + 926, + 1500 + ], + "score": 0.87, + "latex": "k ^ { \\mathrm { { t h } } }" + }, + { + "category_id": 13, + "poly": [ + 896, + 408, + 932, + 408, + 932, + 439, + 896, + 439 + ], + "score": 0.85, + "latex": "| B |" + }, + { + "category_id": 13, + "poly": [ + 1209, + 311, + 1246, + 311, + 1246, + 341, + 1209, + 341 + ], + "score": 0.84, + "latex": "\\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 915, + 281, + 944, + 281, + 944, + 307, + 915, + 307 + ], + "score": 0.83, + "latex": "X" + }, + { + "category_id": 13, + "poly": [ + 1122, + 314, + 1146, + 314, + 1146, + 340, + 1122, + 340 + ], + "score": 0.83, + "latex": "\\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 399, + 344, + 417, + 344, + 417, + 370, + 399, + 370 + ], + "score": 0.81, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 1126, + 1126, + 1155, + 1126, + 1155, + 1162, + 1126, + 1162 + ], + "score": 0.79, + "latex": "2 \\cdot" + }, + { + "category_id": 13, + "poly": [ + 364, + 1470, + 413, + 1470, + 413, + 1501, + 364, + 1501 + ], + "score": 0.79, + "latex": "\\textcircled { \\lVert { 3 4 } \\rVert }" + }, + { + "category_id": 13, + "poly": [ + 804, + 1342, + 824, + 1342, + 824, + 1364, + 804, + 1364 + ], + "score": 0.77, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 949, + 1947, + 975, + 1947, + 975, + 1981, + 949, + 1981 + ], + "score": 0.77, + "latex": "^ { 1 . }" + }, + { + "category_id": 13, + "poly": [ + 548, + 999, + 575, + 999, + 575, + 1040, + 548, + 1040 + ], + "score": 0.77, + "latex": "\\boxed { 1 }" + }, + { + "category_id": 13, + "poly": [ + 586, + 314, + 611, + 314, + 611, + 340, + 586, + 340 + ], + "score": 0.76, + "latex": "A" + }, + { + "category_id": 13, + "poly": [ + 1025, + 1370, + 1040, + 1370, + 1040, + 1395, + 1025, + 1395 + ], + "score": 0.75, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 411, + 1126, + 460, + 1126, + 460, + 1158, + 411, + 1158 + ], + "score": 0.74, + "latex": " { \\mathbb { I } } ^ { [ 1 2 ] }" + }, + { + "category_id": 13, + "poly": [ + 1192, + 1402, + 1205, + 1402, + 1205, + 1425, + 1192, + 1425 + ], + "score": 0.74, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 1009, + 1303, + 1024, + 1303, + 1024, + 1324, + 1009, + 1324 + ], + "score": 0.72, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 861, + 409, + 884, + 409, + 884, + 434, + 861, + 434 + ], + "score": 0.72, + "latex": "\\boldsymbol { B }" + }, + { + "category_id": 13, + "poly": [ + 931, + 1343, + 946, + 1343, + 946, + 1364, + 931, + 1364 + ], + "score": 0.72, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 1287, + 578, + 1321, + 578, + 1321, + 614, + 1287, + 614 + ], + "score": 0.72, + "latex": "( 1 )" + }, + { + "category_id": 13, + "poly": [ + 672, + 1126, + 720, + 1126, + 720, + 1157, + 672, + 1157 + ], + "score": 0.71, + "latex": " { \\mathbb { I } } { \\mathrm { 1 8 } } { \\Vert }" + }, + { + "category_id": 13, + "poly": [ + 1163, + 1478, + 1182, + 1478, + 1182, + 1500, + 1163, + 1500 + ], + "score": 0.71, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 1266, + 344, + 1290, + 344, + 1290, + 371, + 1266, + 371 + ], + "score": 0.69, + "latex": "\\mathbb { N }" + }, + { + "category_id": 13, + "poly": [ + 1081, + 1977, + 1096, + 1977, + 1096, + 2003, + 1081, + 2003 + ], + "score": 0.68, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 791, + 1373, + 806, + 1373, + 806, + 1395, + 791, + 1395 + ], + "score": 0.68, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 520, + 1273, + 534, + 1273, + 534, + 1294, + 520, + 1294 + ], + "score": 0.66, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 1289, + 1868, + 1319, + 1868, + 1319, + 1905, + 1289, + 1905 + ], + "score": 0.66, + "latex": "^ { 1 , }" + }, + { + "category_id": 13, + "poly": [ + 582, + 1572, + 597, + 1572, + 597, + 1593, + 582, + 1593 + ], + "score": 0.65, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 421, + 1703, + 435, + 1703, + 435, + 1724, + 421, + 1724 + ], + "score": 0.6, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 490, + 1095, + 524, + 1095, + 524, + 1127, + 490, + 1127 + ], + "score": 0.5, + "latex": " { \\mathbb { I } }" + }, + { + "category_id": 13, + "poly": [ + 774, + 312, + 937, + 312, + 937, + 345, + 774, + 345 + ], + "score": 0.49, + "latex": "\\mathbb { E } [ X | \\mathrm { E v e n t ~ } A ]" + }, + { + "category_id": 13, + "poly": [ + 638, + 1095, + 672, + 1095, + 672, + 1127, + 638, + 1127 + ], + "score": 0.47, + "latex": "\\pmb { \\Vert 6 \\Vert }" + }, + { + "category_id": 13, + "poly": [ + 475, + 1344, + 489, + 1344, + 489, + 1365, + 475, + 1365 + ], + "score": 0.39, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 773, + 312, + 835, + 312, + 835, + 345, + 773, + 345 + ], + "score": 0.38, + "latex": "\\mathbb { E } [ X ]" + }, + { + "category_id": 13, + "poly": [ + 788, + 1096, + 822, + 1096, + 822, + 1128, + 788, + 1128 + ], + "score": 0.35, + "latex": "\\mathbb { [ [ \\big ] ] }" + }, + { + "category_id": 13, + "poly": [ + 799, + 235, + 834, + 235, + 834, + 262, + 799, + 262 + ], + "score": 0.28, + "latex": "\\mathrm { F L }" + }, + { + "category_id": 13, + "poly": [ + 906, + 313, + 934, + 313, + 934, + 344, + 906, + 344 + ], + "score": 0.26, + "latex": "A ]" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1764.0, + 1028.0, + 1764.0, + 1028.0, + 1817.0, + 290.0, + 1817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 506.0, + 546.0, + 506.0, + 546.0, + 554.0, + 291.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2059.0, + 862.0, + 2059.0, + 862.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1001.0, + 547.0, + 1001.0, + 547.0, + 1041.0, + 290.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 1001.0, + 961.0, + 1001.0, + 961.0, + 1041.0, + 576.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 1001.0, + 1408.0, + 1001.0, + 1408.0, + 1041.0, + 1012.0, + 1041.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1035.0, + 1406.0, + 1035.0, + 1406.0, + 1069.0, + 293.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1068.0, + 1403.0, + 1068.0, + 1403.0, + 1099.0, + 296.0, + 1099.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1094.0, + 489.0, + 1094.0, + 489.0, + 1131.0, + 293.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 1094.0, + 637.0, + 1094.0, + 637.0, + 1131.0, + 525.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 1094.0, + 787.0, + 1094.0, + 787.0, + 1131.0, + 673.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 1094.0, + 1406.0, + 1094.0, + 1406.0, + 1131.0, + 823.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1122.0, + 410.0, + 1122.0, + 410.0, + 1163.0, + 293.0, + 1163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 461.0, + 1122.0, + 671.0, + 1122.0, + 671.0, + 1163.0, + 461.0, + 1163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 721.0, + 1122.0, + 1125.0, + 1122.0, + 1125.0, + 1163.0, + 721.0, + 1163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1156.0, + 1122.0, + 1406.0, + 1122.0, + 1406.0, + 1163.0, + 1156.0, + 1163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1157.0, + 609.0, + 1157.0, + 609.0, + 1192.0, + 294.0, + 1192.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 688.0, + 1157.0, + 1404.0, + 1157.0, + 1404.0, + 1192.0, + 688.0, + 1192.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1188.0, + 1406.0, + 1188.0, + 1406.0, + 1220.0, + 294.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1215.0, + 494.0, + 1215.0, + 494.0, + 1256.0, + 293.0, + 1256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 638.0, + 1215.0, + 715.0, + 1215.0, + 715.0, + 1256.0, + 638.0, + 1256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 833.0, + 1215.0, + 987.0, + 1215.0, + 987.0, + 1256.0, + 833.0, + 1256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1057.0, + 1215.0, + 1068.0, + 1215.0, + 1068.0, + 1256.0, + 1057.0, + 1256.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 279.0, + 914.0, + 279.0, + 914.0, + 312.0, + 296.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 945.0, + 279.0, + 1112.0, + 279.0, + 1112.0, + 312.0, + 945.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1176.0, + 279.0, + 1405.0, + 279.0, + 1405.0, + 312.0, + 1176.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 308.0, + 585.0, + 308.0, + 585.0, + 345.0, + 295.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 612.0, + 308.0, + 772.0, + 308.0, + 772.0, + 345.0, + 612.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 308.0, + 1121.0, + 308.0, + 1121.0, + 345.0, + 938.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 308.0, + 1208.0, + 308.0, + 1208.0, + 345.0, + 1147.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1247.0, + 308.0, + 1403.0, + 308.0, + 1403.0, + 345.0, + 1247.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 341.0, + 398.0, + 341.0, + 398.0, + 378.0, + 294.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 418.0, + 341.0, + 1265.0, + 341.0, + 1265.0, + 378.0, + 418.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1291.0, + 341.0, + 1407.0, + 341.0, + 1407.0, + 378.0, + 1291.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 372.0, + 479.0, + 372.0, + 479.0, + 413.0, + 292.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 566.0, + 372.0, + 701.0, + 372.0, + 701.0, + 413.0, + 566.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 372.0, + 1062.0, + 372.0, + 1062.0, + 413.0, + 942.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1116.0, + 372.0, + 1254.0, + 372.0, + 1254.0, + 413.0, + 1116.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1283.0, + 372.0, + 1406.0, + 372.0, + 1406.0, + 413.0, + 1283.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 404.0, + 860.0, + 404.0, + 860.0, + 441.0, + 350.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 885.0, + 404.0, + 895.0, + 404.0, + 895.0, + 441.0, + 885.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 404.0, + 1403.0, + 404.0, + 1403.0, + 441.0, + 933.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 434.0, + 637.0, + 434.0, + 637.0, + 473.0, + 294.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 763.0, + 434.0, + 915.0, + 434.0, + 915.0, + 473.0, + 763.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1063.0, + 434.0, + 1075.0, + 434.0, + 1075.0, + 473.0, + 1063.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1604.0, + 1410.0, + 1604.0, + 1410.0, + 1641.0, + 292.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1635.0, + 1405.0, + 1635.0, + 1405.0, + 1670.0, + 294.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1667.0, + 1405.0, + 1667.0, + 1405.0, + 1700.0, + 295.0, + 1700.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1698.0, + 420.0, + 1698.0, + 420.0, + 1733.0, + 295.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1698.0, + 881.0, + 1698.0, + 881.0, + 1733.0, + 436.0, + 1733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1436.0, + 1408.0, + 1436.0, + 1408.0, + 1473.0, + 294.0, + 1473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1473.0, + 363.0, + 1473.0, + 363.0, + 1500.0, + 293.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 1464.0, + 697.0, + 1464.0, + 697.0, + 1508.0, + 417.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 1464.0, + 925.0, + 1464.0, + 925.0, + 1508.0, + 845.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 963.0, + 1464.0, + 1162.0, + 1464.0, + 1162.0, + 1508.0, + 963.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 1464.0, + 1408.0, + 1464.0, + 1408.0, + 1508.0, + 1183.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1495.0, + 296.0, + 1495.0, + 296.0, + 1541.0, + 287.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 637.0, + 1495.0, + 1410.0, + 1495.0, + 1410.0, + 1541.0, + 637.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1536.0, + 1403.0, + 1536.0, + 1403.0, + 1567.0, + 292.0, + 1567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1567.0, + 581.0, + 1567.0, + 581.0, + 1600.0, + 295.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 598.0, + 1567.0, + 1040.0, + 1567.0, + 1040.0, + 1600.0, + 598.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1333.0, + 474.0, + 1333.0, + 474.0, + 1374.0, + 292.0, + 1374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1333.0, + 803.0, + 1333.0, + 803.0, + 1374.0, + 490.0, + 1374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 1333.0, + 930.0, + 1333.0, + 930.0, + 1374.0, + 825.0, + 1374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 947.0, + 1333.0, + 1096.0, + 1333.0, + 1096.0, + 1374.0, + 947.0, + 1374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1270.0, + 1333.0, + 1411.0, + 1333.0, + 1411.0, + 1374.0, + 1270.0, + 1374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1365.0, + 790.0, + 1365.0, + 790.0, + 1403.0, + 291.0, + 1403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 1365.0, + 1024.0, + 1365.0, + 1024.0, + 1403.0, + 807.0, + 1403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 1365.0, + 1182.0, + 1365.0, + 1182.0, + 1403.0, + 1041.0, + 1403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1400.0, + 1365.0, + 1411.0, + 1365.0, + 1411.0, + 1403.0, + 1400.0, + 1403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1396.0, + 1191.0, + 1396.0, + 1191.0, + 1434.0, + 294.0, + 1434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 1396.0, + 1219.0, + 1396.0, + 1219.0, + 1434.0, + 1206.0, + 1434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1837.0, + 1406.0, + 1837.0, + 1406.0, + 1875.0, + 293.0, + 1875.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1867.0, + 1288.0, + 1867.0, + 1288.0, + 1904.0, + 293.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1320.0, + 1867.0, + 1407.0, + 1867.0, + 1407.0, + 1904.0, + 1320.0, + 1904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1900.0, + 1219.0, + 1900.0, + 1219.0, + 1934.0, + 296.0, + 1934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1263.0, + 519.0, + 1263.0, + 519.0, + 1302.0, + 294.0, + 1302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 1263.0, + 1401.0, + 1263.0, + 1401.0, + 1302.0, + 535.0, + 1302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1294.0, + 1008.0, + 1294.0, + 1008.0, + 1331.0, + 294.0, + 1331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1025.0, + 1294.0, + 1216.0, + 1294.0, + 1216.0, + 1331.0, + 1025.0, + 1331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 200.0, + 1406.0, + 200.0, + 1406.0, + 239.0, + 295.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 231.0, + 798.0, + 231.0, + 798.0, + 268.0, + 294.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 835.0, + 231.0, + 1409.0, + 231.0, + 1409.0, + 268.0, + 835.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 616.0, + 1209.0, + 616.0, + 1209.0, + 661.0, + 293.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1345.0, + 616.0, + 1406.0, + 616.0, + 1406.0, + 661.0, + 1345.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 647.0, + 297.0, + 647.0, + 297.0, + 699.0, + 292.0, + 699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 647.0, + 512.0, + 647.0, + 512.0, + 699.0, + 434.0, + 699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 647.0, + 1279.0, + 647.0, + 1279.0, + 699.0, + 606.0, + 699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 755.0, + 1404.0, + 755.0, + 1404.0, + 791.0, + 296.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 788.0, + 805.0, + 788.0, + 805.0, + 821.0, + 295.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1939.0, + 948.0, + 1939.0, + 948.0, + 1982.0, + 292.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 976.0, + 1939.0, + 1404.0, + 1939.0, + 1404.0, + 1982.0, + 976.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1973.0, + 818.0, + 1973.0, + 818.0, + 2012.0, + 293.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1973.0, + 1080.0, + 1973.0, + 1080.0, + 2012.0, + 848.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 1973.0, + 1404.0, + 1973.0, + 1404.0, + 2012.0, + 1097.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 573.0, + 1286.0, + 573.0, + 1286.0, + 619.0, + 291.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1322.0, + 573.0, + 1329.0, + 573.0, + 1329.0, + 619.0, + 1322.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 882.0, + 1087.0, + 882.0, + 1087.0, + 920.0, + 294.0, + 920.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1145, + 1406, + 1145, + 1406, + 1482, + 297, + 1482 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 296, + 925, + 1405, + 925, + 1405, + 1134, + 296, + 1134 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 298, + 1625, + 1405, + 1625, + 1405, + 1721, + 298, + 1721 + ], + "score": 0.974 + }, + { + "category_id": 8, + "poly": [ + 729, + 1785, + 969, + 1785, + 969, + 1855, + 729, + 1855 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 429, + 1924, + 1259, + 1924, + 1259, + 2003, + 429, + 2003 + ], + "score": 0.942 + }, + { + "category_id": 0, + "poly": [ + 298, + 1568, + 936, + 1568, + 936, + 1601, + 298, + 1601 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 297, + 1873, + 494, + 1873, + 494, + 1904, + 297, + 1904 + ], + "score": 0.922 + }, + { + "category_id": 1, + "poly": [ + 288, + 1729, + 1311, + 1729, + 1311, + 1766, + 288, + 1766 + ], + "score": 0.918 + }, + { + "category_id": 1, + "poly": [ + 298, + 1493, + 1074, + 1493, + 1074, + 1527, + 298, + 1527 + ], + "score": 0.915 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1801, + 1400, + 1801, + 1400, + 1830, + 1366, + 1830 + ], + "score": 0.871 + }, + { + "category_id": 1, + "poly": [ + 304, + 241, + 1407, + 241, + 1407, + 867, + 304, + 867 + ], + "score": 0.777 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.758 + }, + { + "category_id": 2, + "poly": [ + 298, + 201, + 1112, + 201, + 1112, + 236, + 298, + 236 + ], + "score": 0.359 + }, + { + "category_id": 5, + "poly": [ + 304, + 241, + 1407, + 241, + 1407, + 867, + 304, + 867 + ], + "score": 0.109, + "html": "
2: Initialize: Iterate x(k(k) =x1=(k)1: Input: Parameters: c > O,the number of local updates I,batch size b, stepsizes {nt}. ,descent direction d(k)
with d(𝑘)K (k).(k)) ;Si)and |B(𝑘)|= B for k ∈[K].
3: Perform: x2-nd(k), Ak∈[K]
4: fort = 1 to T dox
5:for k=1 to K do#at the WN
6:f(k)1
>)()+(((d(k) M 6
b EBEB
where we choose1B11
7:=b,and at+1 =c·n²;
if t modI=O thend(k)#at the SN
8:=dt+1= k∑k=1dt+1
9:JK (k)
2 := t+1-nt+1dt+1= k∑κ=1 xt+1-nt+1dt+1 #server-side momentum xt+2
10:else xt+2 (k) (k)
#worker-side momentum
11:
12:
13: end for
14:
Return: xa where a~U{1,..,T}.
" + }, + { + "category_id": 14, + "poly": [ + 729, + 1782, + 968, + 1782, + 968, + 1854, + 729, + 1854 + ], + "score": 0.96, + "latex": "\\eta _ { t } = \\frac { \\bar { \\kappa } } { ( w _ { t } + \\sigma ^ { 2 } t ) ^ { 1 / 3 } } ," + }, + { + "category_id": 13, + "poly": [ + 588, + 968, + 711, + 968, + 711, + 1011, + 588, + 1011 + ], + "score": 0.94, + "latex": "\\{ d _ { t + 1 } ^ { ( k ) } \\} _ { k = 1 } ^ { K }" + }, + { + "category_id": 13, + "poly": [ + 298, + 966, + 423, + 966, + 423, + 1012, + 298, + 1012 + ], + "score": 0.93, + "latex": "\\{ x _ { t + 1 } ^ { ( k ) } \\} _ { k = 1 } ^ { K }" + }, + { + "category_id": 13, + "poly": [ + 891, + 320, + 1018, + 320, + 1018, + 362, + 891, + 362 + ], + "score": 0.93, + "latex": "| B _ { 1 } ^ { ( k ) } | = B" + }, + { + "category_id": 14, + "poly": [ + 433, + 1924, + 1258, + 1924, + 1258, + 2004, + 433, + 2004 + ], + "score": 0.93, + "latex": "\\bar { \\kappa } = \\frac { ( b K ) ^ { 2 / 3 } \\sigma ^ { 2 / 3 } } { L } , \\quad w _ { t } = \\operatorname * { m a x } \\bigg \\{ 2 \\sigma ^ { 2 } , 4 0 9 6 L ^ { 3 } I ^ { 3 } \\bar { \\kappa } ^ { 3 } - \\sigma ^ { 2 } t , \\frac { c ^ { 3 } \\bar { \\kappa } ^ { 3 } } { 4 0 9 6 L ^ { 3 } I ^ { 3 } } \\bigg \\} ." + }, + { + "category_id": 13, + "poly": [ + 952, + 1007, + 1008, + 1007, + 1008, + 1043, + 952, + 1043 + ], + "score": 0.92, + "latex": "\\bar { d } _ { t + 1 }" + }, + { + "category_id": 13, + "poly": [ + 515, + 925, + 571, + 925, + 571, + 969, + 515, + 969 + ], + "score": 0.92, + "latex": "x _ { t + 1 } ^ { ( k ) }" + }, + { + "category_id": 13, + "poly": [ + 1086, + 280, + 1399, + 280, + 1399, + 323, + 1086, + 323 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { d _ { 1 } ^ { ( k ) } = \\bar { d } _ { 1 } = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } d _ { 1 } ^ { ( k ) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 729, + 559, + 879, + 559, + 879, + 594, + 729, + 594 + ], + "score": 0.92, + "latex": "a _ { t + 1 } = c \\cdot \\eta _ { t } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 557, + 824, + 741, + 824, + 741, + 859, + 557, + 859 + ], + "score": 0.91, + "latex": "a \\sim \\mathcal { U } \\{ 1 , . . . , T \\}" + }, + { + "category_id": 13, + "poly": [ + 619, + 926, + 669, + 926, + 669, + 967, + 619, + 967 + ], + "score": 0.91, + "latex": "x _ { t } ^ { ( k ) }" + }, + { + "category_id": 13, + "poly": [ + 550, + 279, + 869, + 279, + 869, + 322, + 550, + 322 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { x _ { 1 } ^ { ( k ) } = \\bar { x } _ { 1 } = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } x _ { 1 } ^ { ( k ) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1061, + 326, + 1153, + 326, + 1153, + 361, + 1061, + 361 + ], + "score": 0.91, + "latex": "k \\in [ K ]" + }, + { + "category_id": 13, + "poly": [ + 400, + 320, + 839, + 320, + 839, + 371, + 400, + 371 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { d _ { 1 } ^ { ( k ) } = \\frac { 1 } { B } \\sum _ { \\xi _ { 1 } ^ { ( k ) } \\in \\mathcal { B } _ { 1 } ^ { ( k ) } } \\nabla f ^ { ( k ) } ( x _ { 1 } ^ { ( k ) } ; \\xi _ { 1 } ^ { ( k ) } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1233, + 248, + 1286, + 248, + 1286, + 282, + 1233, + 282 + ], + "score": 0.9, + "latex": "\\{ \\eta _ { t } \\}" + }, + { + "category_id": 13, + "poly": [ + 840, + 1012, + 898, + 1012, + 898, + 1044, + 840, + 1044 + ], + "score": 0.9, + "latex": "\\bar { x } _ { t + 1 }" + }, + { + "category_id": 13, + "poly": [ + 548, + 554, + 671, + 554, + 671, + 595, + 548, + 595 + ], + "score": 0.89, + "latex": "| B _ { t + 1 } ^ { ( k ) } | = b" + }, + { + "category_id": 13, + "poly": [ + 566, + 251, + 631, + 251, + 631, + 277, + 566, + 277 + ], + "score": 0.89, + "latex": "c > 0" + }, + { + "category_id": 13, + "poly": [ + 452, + 370, + 670, + 370, + 670, + 410, + 452, + 410 + ], + "score": 0.88, + "latex": "x _ { 2 } ^ { ( k ) } = x _ { 1 } ^ { k } - \\eta _ { 1 } d _ { 1 } ^ { ( k ) }" + }, + { + "category_id": 13, + "poly": [ + 1152, + 1206, + 1240, + 1206, + 1240, + 1239, + 1152, + 1239 + ], + "score": 0.88, + "latex": "\\overline { { \\mathbb { D } \\mathbb { Z } \\mathbb { G } \\mathbb { S } } }" + }, + { + "category_id": 13, + "poly": [ + 385, + 408, + 450, + 408, + 450, + 435, + 385, + 435 + ], + "score": 0.88, + "latex": "t = 1" + }, + { + "category_id": 13, + "poly": [ + 844, + 1105, + 871, + 1105, + 871, + 1132, + 844, + 1132 + ], + "score": 0.85, + "latex": "a _ { t }" + }, + { + "category_id": 13, + "poly": [ + 446, + 828, + 478, + 828, + 478, + 855, + 446, + 855 + ], + "score": 0.84, + "latex": "\\scriptstyle { \\bar { x } } _ { a }" + }, + { + "category_id": 13, + "poly": [ + 428, + 438, + 497, + 438, + 497, + 465, + 428, + 465 + ], + "score": 0.84, + "latex": "k = 1" + }, + { + "category_id": 13, + "poly": [ + 614, + 1106, + 640, + 1106, + 640, + 1134, + 614, + 1134 + ], + "score": 0.84, + "latex": "\\eta _ { t }" + }, + { + "category_id": 13, + "poly": [ + 524, + 593, + 593, + 593, + 593, + 617, + 524, + 617 + ], + "score": 0.83, + "latex": "I = 0" + }, + { + "category_id": 13, + "poly": [ + 480, + 697, + 760, + 697, + 760, + 739, + 480, + 739 + ], + "score": 0.83, + "latex": "x _ { t + 2 } ^ { ( k ) } = x _ { t + 1 } ^ { ( k ) } - \\eta _ { t + 1 } d _ { t + 1 } ^ { ( k ) }" + }, + { + "category_id": 13, + "poly": [ + 445, + 655, + 1081, + 655, + 1081, + 699, + 445, + 699 + ], + "score": 0.82, + "latex": "\\begin{array} { r } { x _ { t + 2 } ^ { ( k ) } : = \\bar { x } _ { t + 1 } - \\eta _ { t + 1 } \\bar { d } _ { t + 1 } = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } x _ { t + 1 } ^ { ( k ) } - \\eta _ { t + 1 } \\bar { d } _ { t + 1 } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 423, + 455, + 1401, + 455, + 1401, + 554, + 423, + 554 + ], + "score": 0.81, + "latex": "\\mathcal { d } _ { t + 1 } ^ { ( k ) } = \\frac { 1 } { b } \\sum _ { \\xi _ { t + 1 } ^ { ( k ) } \\in \\mathcal { B } _ { t + 1 } ^ { ( k ) } } ^ { \\pi \\Delta } \\nabla f ^ { ( k ) } ( x _ { t + 1 } ^ { ( k ) } ; \\xi _ { t + 1 } ^ { ( k ) } ) + \\left( 1 - a _ { t + 1 } \\right) \\bigg ( d _ { t } ^ { ( k ) } - \\frac { 1 } { b } \\sum _ { \\xi _ { t + 1 } ^ { ( k ) } \\in \\mathcal { B } _ { t + 1 } ^ { ( k ) } } \\nabla f ^ { ( k ) } ( x _ { t } ^ { ( k ) } ; \\xi _ { t + 1 } ^ { ( k ) } ) \\bigg )" + }, + { + "category_id": 13, + "poly": [ + 956, + 251, + 974, + 251, + 974, + 276, + 956, + 276 + ], + "score": 0.74, + "latex": "I" + }, + { + "category_id": 13, + "poly": [ + 1101, + 250, + 1117, + 250, + 1117, + 277, + 1101, + 277 + ], + "score": 0.74, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 817, + 1072, + 833, + 1072, + 833, + 1098, + 817, + 1098 + ], + "score": 0.72, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 529, + 438, + 558, + 438, + 558, + 464, + 529, + 464 + ], + "score": 0.72, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 482, + 409, + 506, + 409, + 506, + 434, + 482, + 434 + ], + "score": 0.71, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 1046, + 1072, + 1065, + 1072, + 1065, + 1099, + 1046, + 1099 + ], + "score": 0.69, + "latex": "I" + }, + { + "category_id": 13, + "poly": [ + 810, + 935, + 828, + 935, + 828, + 961, + 810, + 961 + ], + "score": 0.68, + "latex": "I" + }, + { + "category_id": 13, + "poly": [ + 687, + 373, + 805, + 373, + 805, + 410, + 687, + 410 + ], + "score": 0.67, + "latex": "\\forall k \\in [ K ]" + }, + { + "category_id": 13, + "poly": [ + 452, + 593, + 465, + 593, + 465, + 617, + 452, + 617 + ], + "score": 0.66, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 468, + 1421, + 483, + 1421, + 483, + 1447, + 468, + 1447 + ], + "score": 0.46, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 863, + 1421, + 880, + 1421, + 880, + 1447, + 863, + 1447 + ], + "score": 0.41, + "latex": "I" + }, + { + "category_id": 13, + "poly": [ + 454, + 614, + 800, + 614, + 800, + 656, + 454, + 656 + ], + "score": 0.39, + "latex": "\\begin{array} { r } { d _ { t + 1 } ^ { ( k ) } = \\bar { d } _ { t + 1 } : = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } d _ { t + 1 } ^ { ( k ) } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 453, + 649, + 1081, + 649, + 1081, + 699, + 453, + 699 + ], + "score": 0.28, + "latex": "\\begin{array} { r l } & { x _ { t + 2 } ^ { ( k ) ^ { \\bot } } : = \\bar { x } _ { t + 1 } - \\eta _ { t + 1 } \\bar { d } _ { t + 1 } = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } x _ { t + 1 } ^ { ( k ) } - \\eta _ { t + 1 } \\bar { d } _ { t + 1 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1043, + 1206, + 1104, + 1206, + 1104, + 1239, + 1043, + 1239 + ], + "score": 0.27, + "latex": "\\mathbb { \\left[ \\bigstar \\bigstar \\right] }" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1567.0, + 938.0, + 1567.0, + 938.0, + 1605.0, + 292.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2092.0, + 838.0, + 2092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 192.0, + 1115.0, + 192.0, + 1115.0, + 246.0, + 291.0, + 246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1145.0, + 1404.0, + 1145.0, + 1404.0, + 1182.0, + 295.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1179.0, + 1404.0, + 1179.0, + 1404.0, + 1210.0, + 295.0, + 1210.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1205.0, + 1042.0, + 1205.0, + 1042.0, + 1244.0, + 292.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1105.0, + 1205.0, + 1151.0, + 1205.0, + 1151.0, + 1244.0, + 1105.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 1205.0, + 1407.0, + 1205.0, + 1407.0, + 1244.0, + 1241.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1235.0, + 1407.0, + 1235.0, + 1407.0, + 1275.0, + 291.0, + 1275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1266.0, + 1406.0, + 1266.0, + 1406.0, + 1301.0, + 294.0, + 1301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1299.0, + 1406.0, + 1299.0, + 1406.0, + 1333.0, + 294.0, + 1333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1328.0, + 1408.0, + 1328.0, + 1408.0, + 1363.0, + 294.0, + 1363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1357.0, + 1409.0, + 1357.0, + 1409.0, + 1394.0, + 292.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1389.0, + 1406.0, + 1389.0, + 1406.0, + 1424.0, + 294.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1418.0, + 467.0, + 1418.0, + 467.0, + 1455.0, + 292.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 1418.0, + 862.0, + 1418.0, + 862.0, + 1455.0, + 484.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 881.0, + 1418.0, + 1406.0, + 1418.0, + 1406.0, + 1455.0, + 881.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1450.0, + 1237.0, + 1450.0, + 1237.0, + 1485.0, + 295.0, + 1485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 919.0, + 514.0, + 919.0, + 514.0, + 975.0, + 292.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 919.0, + 618.0, + 919.0, + 618.0, + 975.0, + 572.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 919.0, + 809.0, + 919.0, + 809.0, + 975.0, + 670.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 919.0, + 1411.0, + 919.0, + 1411.0, + 975.0, + 829.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 950.0, + 297.0, + 950.0, + 297.0, + 1029.0, + 284.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 950.0, + 587.0, + 950.0, + 587.0, + 1029.0, + 424.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 712.0, + 950.0, + 1418.0, + 950.0, + 1418.0, + 1029.0, + 712.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1009.0, + 839.0, + 1009.0, + 839.0, + 1046.0, + 291.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 1009.0, + 951.0, + 1009.0, + 951.0, + 1046.0, + 899.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 1009.0, + 1406.0, + 1009.0, + 1406.0, + 1046.0, + 1009.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1038.0, + 1406.0, + 1038.0, + 1406.0, + 1074.0, + 293.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1071.0, + 816.0, + 1071.0, + 816.0, + 1104.0, + 294.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 834.0, + 1071.0, + 1045.0, + 1071.0, + 1045.0, + 1104.0, + 834.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 1071.0, + 1405.0, + 1071.0, + 1405.0, + 1104.0, + 1066.0, + 1104.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1102.0, + 613.0, + 1102.0, + 613.0, + 1136.0, + 294.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 1102.0, + 843.0, + 1102.0, + 843.0, + 1136.0, + 641.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 1102.0, + 1184.0, + 1102.0, + 1184.0, + 1136.0, + 872.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 321.0, + 982.0, + 405.0, + 982.0, + 405.0, + 1001.0, + 321.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1626.0, + 1405.0, + 1626.0, + 1405.0, + 1661.0, + 295.0, + 1661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1658.0, + 1405.0, + 1658.0, + 1405.0, + 1691.0, + 293.0, + 1691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1686.0, + 1122.0, + 1686.0, + 1122.0, + 1722.0, + 293.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1871.0, + 500.0, + 1871.0, + 500.0, + 1907.0, + 295.0, + 1907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1726.0, + 1318.0, + 1726.0, + 1318.0, + 1772.0, + 291.0, + 1772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1491.0, + 1078.0, + 1491.0, + 1078.0, + 1535.0, + 292.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 243.0, + 565.0, + 243.0, + 565.0, + 284.0, + 308.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 632.0, + 243.0, + 955.0, + 243.0, + 955.0, + 284.0, + 632.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 975.0, + 243.0, + 1100.0, + 243.0, + 1100.0, + 284.0, + 975.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1118.0, + 243.0, + 1232.0, + 243.0, + 1232.0, + 284.0, + 1118.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1287.0, + 243.0, + 1298.0, + 243.0, + 1298.0, + 284.0, + 1287.0, + 284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 268.0, + 549.0, + 268.0, + 549.0, + 334.0, + 305.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 268.0, + 1085.0, + 268.0, + 1085.0, + 334.0, + 870.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1400.0, + 268.0, + 1408.0, + 268.0, + 1408.0, + 334.0, + 1400.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 302.0, + 878.0, + 302.0, + 878.0, + 313.0, + 870.0, + 313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 309.0, + 399.0, + 309.0, + 399.0, + 386.0, + 339.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 309.0, + 890.0, + 309.0, + 890.0, + 386.0, + 840.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1019.0, + 309.0, + 1060.0, + 309.0, + 1060.0, + 386.0, + 1019.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1154.0, + 309.0, + 1168.0, + 309.0, + 1168.0, + 386.0, + 1154.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 355.0, + 451.0, + 355.0, + 451.0, + 423.0, + 292.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 671.0, + 355.0, + 686.0, + 355.0, + 686.0, + 423.0, + 671.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 806.0, + 355.0, + 820.0, + 355.0, + 820.0, + 423.0, + 806.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 404.0, + 384.0, + 404.0, + 384.0, + 439.0, + 305.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 404.0, + 481.0, + 404.0, + 481.0, + 439.0, + 451.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 507.0, + 404.0, + 548.0, + 404.0, + 548.0, + 439.0, + 507.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 437.0, + 340.0, + 437.0, + 340.0, + 470.0, + 308.0, + 470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 437.0, + 427.0, + 437.0, + 427.0, + 468.0, + 381.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 437.0, + 528.0, + 437.0, + 528.0, + 468.0, + 498.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 559.0, + 437.0, + 601.0, + 437.0, + 601.0, + 468.0, + 559.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1249.0, + 438.0, + 1404.0, + 438.0, + 1404.0, + 467.0, + 1249.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 477.0, + 339.0, + 477.0, + 339.0, + 510.0, + 308.0, + 510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 456.0, + 422.0, + 456.0, + 422.0, + 526.0, + 417.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 452.0, + 1409.0, + 452.0, + 1409.0, + 531.0, + 1402.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 502.0, + 1201.0, + 502.0, + 1201.0, + 567.0, + 1066.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 540.0, + 547.0, + 540.0, + 547.0, + 607.0, + 342.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 540.0, + 728.0, + 540.0, + 728.0, + 607.0, + 672.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 880.0, + 540.0, + 893.0, + 540.0, + 893.0, + 607.0, + 880.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 588.0, + 342.0, + 588.0, + 342.0, + 655.0, + 305.0, + 655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 584.0, + 451.0, + 584.0, + 451.0, + 667.0, + 421.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 592.0, + 1404.0, + 592.0, + 1404.0, + 621.0, + 1251.0, + 621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 662.0, + 342.0, + 662.0, + 342.0, + 738.0, + 297.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 652.0, + 444.0, + 652.0, + 444.0, + 759.0, + 408.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1082.0, + 649.0, + 1407.0, + 649.0, + 1407.0, + 736.0, + 1082.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 733.0, + 340.0, + 733.0, + 340.0, + 766.0, + 295.0, + 766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 420.0, + 732.0, + 505.0, + 732.0, + 505.0, + 763.0, + 420.0, + 763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 763.0, + 340.0, + 763.0, + 340.0, + 797.0, + 295.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 763.0, + 480.0, + 763.0, + 480.0, + 796.0, + 382.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 792.0, + 439.0, + 792.0, + 439.0, + 829.0, + 293.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 821.0, + 445.0, + 821.0, + 445.0, + 860.0, + 294.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 821.0, + 556.0, + 821.0, + 556.0, + 860.0, + 479.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 821.0, + 749.0, + 821.0, + 749.0, + 860.0, + 742.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 705.25, + 319.5, + 816.25, + 319.5, + 816.25, + 375.0, + 705.25, + 375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 665.0, + 627.0, + 748.0, + 627.0, + 748.0, + 661.0, + 665.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.25, + 626.5, + 802.25, + 626.5, + 802.25, + 662.0, + 742.25, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 791.0, + 648.0, + 883.0, + 648.0, + 883.0, + 699.5, + 791.0, + 699.5 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 1170, + 1405, + 1170, + 1405, + 1661, + 296, + 1661 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 1666, + 1405, + 1666, + 1405, + 2010, + 297, + 2010 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 296, + 821, + 1405, + 821, + 1405, + 1162, + 296, + 1162 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 295, + 196, + 1405, + 196, + 1405, + 297, + 295, + 297 + ], + "score": 0.968 + }, + { + "category_id": 8, + "poly": [ + 378, + 481, + 1332, + 481, + 1332, + 560, + 378, + 560 + ], + "score": 0.938 + }, + { + "category_id": 8, + "poly": [ + 572, + 313, + 1126, + 313, + 1126, + 358, + 572, + 358 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 297, + 371, + 1034, + 371, + 1034, + 411, + 297, + 411 + ], + "score": 0.92 + }, + { + "category_id": 1, + "poly": [ + 302, + 431, + 908, + 431, + 908, + 468, + 302, + 468 + ], + "score": 0.913 + }, + { + "category_id": 1, + "poly": [ + 299, + 784, + 1338, + 784, + 1338, + 818, + 299, + 818 + ], + "score": 0.868 + }, + { + "category_id": 9, + "poly": [ + 1365, + 321, + 1401, + 321, + 1401, + 352, + 1365, + 352 + ], + "score": 0.839 + }, + { + "category_id": 1, + "poly": [ + 346, + 727, + 1267, + 727, + 1267, + 761, + 346, + 761 + ], + "score": 0.828 + }, + { + "category_id": 9, + "poly": [ + 1366, + 507, + 1401, + 507, + 1401, + 537, + 1366, + 537 + ], + "score": 0.828 + }, + { + "category_id": 1, + "poly": [ + 301, + 627, + 1404, + 627, + 1404, + 725, + 301, + 725 + ], + "score": 0.807 + }, + { + "category_id": 2, + "poly": [ + 840, + 2062, + 859, + 2062, + 859, + 2085, + 840, + 2085 + ], + "score": 0.783 + }, + { + "category_id": 1, + "poly": [ + 297, + 591, + 651, + 591, + 651, + 626, + 297, + 626 + ], + "score": 0.67 + }, + { + "category_id": 13, + "poly": [ + 509, + 200, + 1024, + 200, + 1024, + 269, + 509, + 269 + ], + "score": 0.95, + "latex": "\\begin{array} { r } { c = \\frac { 6 4 L ^ { 2 } } { b K } + \\frac { \\sigma ^ { 2 } } { 2 4 \\bar { \\kappa } ^ { 3 } L I } = L ^ { 2 } \\bigg ( \\frac { 6 4 } { b K } + \\frac { 1 } { 2 4 ( b K ) ^ { 2 } I } \\bigg ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 354, + 1851, + 546, + 1851, + 546, + 1888, + 354, + 1888 + ], + "score": 0.93, + "latex": "b = \\mathcal { O } ( T ^ { 1 / 2 } / K )" + }, + { + "category_id": 13, + "poly": [ + 816, + 1440, + 903, + 1440, + 903, + 1475, + 816, + 1475 + ], + "score": 0.93, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 1 } )" + }, + { + "category_id": 13, + "poly": [ + 1196, + 1039, + 1283, + 1039, + 1283, + 1073, + 1196, + 1073 + ], + "score": 0.93, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 1 } )" + }, + { + "category_id": 13, + "poly": [ + 1011, + 852, + 1120, + 852, + 1120, + 889, + 1011, + 889 + ], + "score": 0.93, + "latex": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 983, + 623, + 1091, + 623, + 1091, + 662, + 983, + 662 + ], + "score": 0.93, + "latex": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 694, + 1627, + 803, + 1627, + 803, + 1663, + 694, + 1663 + ], + "score": 0.93, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 1 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 943, + 1350, + 1102, + 1350, + 1102, + 1384, + 943, + 1384 + ], + "score": 0.93, + "latex": "b = I = \\mathcal { O } ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 438, + 1757, + 664, + 1757, + 664, + 1794, + 438, + 1794 + ], + "score": 0.93, + "latex": "I = \\mathcal { O } ( T ^ { 1 / 3 } / K ^ { 2 / 3 } )" + }, + { + "category_id": 13, + "poly": [ + 822, + 947, + 931, + 947, + 931, + 984, + 822, + 984 + ], + "score": 0.93, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 407, + 1009, + 494, + 1009, + 494, + 1043, + 407, + 1043 + ], + "score": 0.93, + "latex": "\\bar { \\mathcal { O } } ( \\epsilon ^ { - 1 } )" + }, + { + "category_id": 13, + "poly": [ + 534, + 659, + 697, + 659, + 697, + 697, + 534, + 697 + ], + "score": 0.93, + "latex": "\\tilde { \\mathcal { O } } ( K ^ { - 1 } \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 991, + 1592, + 1100, + 1592, + 1100, + 1629, + 991, + 1629 + ], + "score": 0.92, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 488, + 375, + 595, + 375, + 595, + 409, + 488, + 409 + ], + "score": 0.92, + "latex": "\\nu \\in [ 0 , 1 ]" + }, + { + "category_id": 14, + "poly": [ + 381, + 483, + 1332, + 483, + 1332, + 560, + 381, + 560 + ], + "score": 0.92, + "latex": "\\mathbb { E } \\Vert \\nabla f ( \\bar { x } _ { a } ) \\Vert ^ { 2 } = \\mathcal { O } \\Bigg ( \\frac { f ( \\bar { x } _ { 1 } ) - f ^ { * } } { K ^ { 2 \\nu / 3 } T ^ { 1 - \\nu / 3 } } \\Bigg ) + \\tilde { \\mathcal { O } } \\Bigg ( \\frac { \\sigma ^ { 2 } } { K ^ { 2 \\nu / 3 } T ^ { 1 - \\nu / 3 } } \\Bigg ) + \\tilde { \\mathcal { O } } \\Bigg ( \\frac { \\zeta ^ { 2 } } { K ^ { 2 \\nu / 3 } T ^ { 1 - \\nu / 3 } } \\Bigg ) ." + }, + { + "category_id": 13, + "poly": [ + 1176, + 853, + 1263, + 853, + 1263, + 889, + 1176, + 889 + ], + "score": 0.92, + "latex": "\\tilde { \\mathcal { O } } ( \\overline { { \\epsilon ^ { - 1 } } } )" + }, + { + "category_id": 13, + "poly": [ + 1173, + 722, + 1259, + 722, + 1259, + 761, + 1173, + 761 + ], + "score": 0.92, + "latex": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )" + }, + { + "category_id": 13, + "poly": [ + 868, + 1595, + 954, + 1595, + 954, + 1628, + 868, + 1628 + ], + "score": 0.91, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 2 } )" + }, + { + "category_id": 13, + "poly": [ + 437, + 592, + 544, + 592, + 544, + 626, + 437, + 626 + ], + "score": 0.91, + "latex": "\\nu \\in [ 0 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 1261, + 1665, + 1373, + 1665, + 1373, + 1698, + 1261, + 1698 + ], + "score": 0.91, + "latex": "\\nu \\in [ 0 , 1 ]" + }, + { + "category_id": 14, + "poly": [ + 573, + 313, + 1124, + 313, + 1124, + 358, + 573, + 358 + ], + "score": 0.91, + "latex": "I = \\mathcal { O } \\big ( ( T / K ^ { 2 } ) ^ { \\nu / 3 } \\big ) , \\quad b = \\mathcal { O } \\big ( ( T / K ^ { 2 } ) ^ { 1 / 2 - \\nu / 2 } \\big )" + }, + { + "category_id": 13, + "poly": [ + 1197, + 1729, + 1269, + 1729, + 1269, + 1756, + 1197, + 1756 + ], + "score": 0.9, + "latex": "\\nu = 1" + }, + { + "category_id": 13, + "poly": [ + 1327, + 1822, + 1397, + 1822, + 1397, + 1849, + 1327, + 1849 + ], + "score": 0.89, + "latex": "\\nu = 0" + }, + { + "category_id": 13, + "poly": [ + 298, + 264, + 386, + 264, + 386, + 292, + 298, + 292 + ], + "score": 0.88, + "latex": "B = b I" + }, + { + "category_id": 13, + "poly": [ + 390, + 437, + 422, + 437, + 422, + 464, + 390, + 464 + ], + "score": 0.85, + "latex": "\\scriptstyle { \\bar { x } } _ { a }" + }, + { + "category_id": 13, + "poly": [ + 1274, + 1697, + 1309, + 1697, + 1309, + 1730, + 1274, + 1730 + ], + "score": 0.85, + "latex": "\\textcircled { 3 }" + }, + { + "category_id": 13, + "poly": [ + 789, + 1561, + 838, + 1561, + 838, + 1594, + 789, + 1594 + ], + "score": 0.83, + "latex": "\\bar { \\mathbb { E } 2 } \\mathbb { I }" + }, + { + "category_id": 13, + "poly": [ + 888, + 1561, + 935, + 1561, + 935, + 1593, + 888, + 1593 + ], + "score": 0.81, + "latex": "\\bar { \\textregistered 4 } \\bar { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1167, + 1470, + 1215, + 1470, + 1215, + 1503, + 1167, + 1503 + ], + "score": 0.78, + "latex": "\\mathbb { I O } ]" + }, + { + "category_id": 13, + "poly": [ + 595, + 1855, + 613, + 1855, + 613, + 1882, + 595, + 1882 + ], + "score": 0.78, + "latex": "I" + }, + { + "category_id": 13, + "poly": [ + 1211, + 979, + 1245, + 979, + 1245, + 1012, + 1211, + 1012 + ], + "score": 0.75, + "latex": " { \\mathbb { I } } ^ { { \\left[ 5 \\right] } }" + }, + { + "category_id": 13, + "poly": [ + 416, + 1733, + 436, + 1733, + 436, + 1755, + 416, + 1755 + ], + "score": 0.75, + "latex": "\\nu" + }, + { + "category_id": 13, + "poly": [ + 438, + 948, + 487, + 948, + 487, + 981, + 438, + 981 + ], + "score": 0.73, + "latex": "[ [ 3 5 ] ]" + }, + { + "category_id": 13, + "poly": [ + 763, + 1698, + 778, + 1698, + 778, + 1724, + 763, + 1724 + ], + "score": 0.72, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 880, + 1729, + 899, + 1729, + 899, + 1755, + 880, + 1755 + ], + "score": 0.72, + "latex": "I" + }, + { + "category_id": 13, + "poly": [ + 1082, + 1102, + 1105, + 1102, + 1105, + 1138, + 1082, + 1138 + ], + "score": 0.72, + "latex": "2 \\cdot" + }, + { + "category_id": 13, + "poly": [ + 1163, + 1352, + 1179, + 1352, + 1179, + 1379, + 1163, + 1379 + ], + "score": 0.71, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 437, + 857, + 453, + 857, + 453, + 885, + 437, + 885 + ], + "score": 0.71, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 1338, + 1729, + 1354, + 1729, + 1354, + 1755, + 1338, + 1755 + ], + "score": 0.7, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 1203, + 819, + 1251, + 819, + 1251, + 853, + 1203, + 853 + ], + "score": 0.69, + "latex": "3 . 1" + }, + { + "category_id": 13, + "poly": [ + 1229, + 1352, + 1247, + 1352, + 1247, + 1378, + 1229, + 1378 + ], + "score": 0.69, + "latex": "I" + }, + { + "category_id": 13, + "poly": [ + 1016, + 1508, + 1031, + 1508, + 1031, + 1529, + 1016, + 1529 + ], + "score": 0.68, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 365, + 857, + 384, + 857, + 384, + 884, + 365, + 884 + ], + "score": 0.67, + "latex": "I" + }, + { + "category_id": 13, + "poly": [ + 1145, + 1010, + 1180, + 1010, + 1180, + 1041, + 1145, + 1041 + ], + "score": 0.67, + "latex": "\\mathbb { P }" + }, + { + "category_id": 13, + "poly": [ + 1312, + 1470, + 1347, + 1470, + 1347, + 1503, + 1312, + 1503 + ], + "score": 0.67, + "latex": "\\bigstar \\bigstar" + }, + { + "category_id": 13, + "poly": [ + 1058, + 1071, + 1093, + 1071, + 1093, + 1102, + 1058, + 1102 + ], + "score": 0.66, + "latex": "\\bigstar \\bigstar" + }, + { + "category_id": 13, + "poly": [ + 395, + 1500, + 445, + 1500, + 445, + 1533, + 395, + 1533 + ], + "score": 0.63, + "latex": "\\pmb { \\Vert 3 6 \\Vert }" + }, + { + "category_id": 13, + "poly": [ + 1124, + 1197, + 1155, + 1197, + 1155, + 1233, + 1124, + 1233 + ], + "score": 0.63, + "latex": "^ { 1 ) }" + }, + { + "category_id": 13, + "poly": [ + 1156, + 1698, + 1175, + 1698, + 1175, + 1724, + 1156, + 1724 + ], + "score": 0.6, + "latex": "I" + }, + { + "category_id": 13, + "poly": [ + 633, + 266, + 650, + 266, + 650, + 290, + 633, + 290 + ], + "score": 0.58, + "latex": "I" + }, + { + "category_id": 13, + "poly": [ + 371, + 382, + 390, + 382, + 390, + 402, + 371, + 402 + ], + "score": 0.56, + "latex": "\\nu" + }, + { + "category_id": 13, + "poly": [ + 695, + 1729, + 712, + 1729, + 712, + 1755, + 695, + 1755 + ], + "score": 0.51, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 772, + 432, + 798, + 432, + 798, + 470, + 772, + 470 + ], + "score": 0.5, + "latex": "\\boldsymbol { l } ," + }, + { + "category_id": 13, + "poly": [ + 1349, + 1538, + 1364, + 1538, + 1364, + 1560, + 1349, + 1560 + ], + "score": 0.42, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 884, + 1478, + 900, + 1478, + 900, + 1499, + 884, + 1499 + ], + "score": 0.4, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 939, + 1626, + 987, + 1626, + 987, + 1665, + 939, + 1665 + ], + "score": 0.33, + "latex": "3 . 2" + }, + { + "category_id": 13, + "poly": [ + 604, + 945, + 631, + 945, + 631, + 985, + 604, + 985 + ], + "score": 0.32, + "latex": "1 ." + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2060.0, + 862.0, + 2060.0, + 862.0, + 2091.0, + 840.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1168.0, + 1406.0, + 1168.0, + 1406.0, + 1202.0, + 295.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1195.0, + 1123.0, + 1195.0, + 1123.0, + 1235.0, + 294.0, + 1235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1156.0, + 1195.0, + 1407.0, + 1195.0, + 1407.0, + 1235.0, + 1156.0, + 1235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1227.0, + 1407.0, + 1227.0, + 1407.0, + 1263.0, + 294.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1259.0, + 1405.0, + 1259.0, + 1405.0, + 1294.0, + 295.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1290.0, + 1405.0, + 1290.0, + 1405.0, + 1324.0, + 294.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1319.0, + 1405.0, + 1319.0, + 1405.0, + 1354.0, + 295.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1350.0, + 942.0, + 1350.0, + 942.0, + 1385.0, + 294.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 1350.0, + 1162.0, + 1350.0, + 1162.0, + 1385.0, + 1103.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 1350.0, + 1228.0, + 1350.0, + 1228.0, + 1385.0, + 1180.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1248.0, + 1350.0, + 1405.0, + 1350.0, + 1405.0, + 1385.0, + 1248.0, + 1385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1380.0, + 1405.0, + 1380.0, + 1405.0, + 1414.0, + 294.0, + 1414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1408.0, + 1405.0, + 1408.0, + 1405.0, + 1445.0, + 293.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1440.0, + 815.0, + 1440.0, + 815.0, + 1475.0, + 294.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 1440.0, + 1407.0, + 1440.0, + 1407.0, + 1475.0, + 904.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1471.0, + 883.0, + 1471.0, + 883.0, + 1505.0, + 295.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 1471.0, + 1166.0, + 1471.0, + 1166.0, + 1505.0, + 901.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1471.0, + 1311.0, + 1471.0, + 1311.0, + 1505.0, + 1216.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 1471.0, + 1405.0, + 1471.0, + 1405.0, + 1505.0, + 1348.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1502.0, + 394.0, + 1502.0, + 394.0, + 1536.0, + 295.0, + 1536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 1502.0, + 1015.0, + 1502.0, + 1015.0, + 1536.0, + 446.0, + 1536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 1502.0, + 1405.0, + 1502.0, + 1405.0, + 1536.0, + 1032.0, + 1536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1529.0, + 1348.0, + 1529.0, + 1348.0, + 1567.0, + 293.0, + 1567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1365.0, + 1529.0, + 1406.0, + 1529.0, + 1406.0, + 1567.0, + 1365.0, + 1567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1558.0, + 788.0, + 1558.0, + 788.0, + 1598.0, + 291.0, + 1598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 1558.0, + 887.0, + 1558.0, + 887.0, + 1598.0, + 839.0, + 1598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 1558.0, + 1405.0, + 1558.0, + 1405.0, + 1598.0, + 936.0, + 1598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1592.0, + 867.0, + 1592.0, + 867.0, + 1630.0, + 293.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 955.0, + 1592.0, + 990.0, + 1592.0, + 990.0, + 1630.0, + 955.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1101.0, + 1592.0, + 1406.0, + 1592.0, + 1406.0, + 1630.0, + 1101.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1626.0, + 693.0, + 1626.0, + 693.0, + 1664.0, + 294.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 1626.0, + 938.0, + 1626.0, + 938.0, + 1664.0, + 804.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 1626.0, + 1400.0, + 1626.0, + 1400.0, + 1664.0, + 988.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1665.0, + 1260.0, + 1665.0, + 1260.0, + 1701.0, + 295.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1374.0, + 1665.0, + 1406.0, + 1665.0, + 1406.0, + 1701.0, + 1374.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1693.0, + 762.0, + 1693.0, + 762.0, + 1735.0, + 292.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 1693.0, + 1155.0, + 1693.0, + 1155.0, + 1735.0, + 779.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1176.0, + 1693.0, + 1273.0, + 1693.0, + 1273.0, + 1735.0, + 1176.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1310.0, + 1693.0, + 1408.0, + 1693.0, + 1408.0, + 1735.0, + 1310.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1726.0, + 415.0, + 1726.0, + 415.0, + 1762.0, + 294.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 1726.0, + 694.0, + 1726.0, + 694.0, + 1762.0, + 437.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 713.0, + 1726.0, + 879.0, + 1726.0, + 879.0, + 1762.0, + 713.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 900.0, + 1726.0, + 1196.0, + 1726.0, + 1196.0, + 1762.0, + 900.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1270.0, + 1726.0, + 1337.0, + 1726.0, + 1337.0, + 1762.0, + 1270.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1355.0, + 1726.0, + 1407.0, + 1726.0, + 1407.0, + 1762.0, + 1355.0, + 1762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1752.0, + 437.0, + 1752.0, + 437.0, + 1800.0, + 291.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 665.0, + 1752.0, + 1408.0, + 1752.0, + 1408.0, + 1800.0, + 665.0, + 1800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1789.0, + 1405.0, + 1789.0, + 1405.0, + 1824.0, + 295.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1819.0, + 1326.0, + 1819.0, + 1326.0, + 1853.0, + 294.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 1819.0, + 1408.0, + 1819.0, + 1408.0, + 1853.0, + 1398.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1849.0, + 353.0, + 1849.0, + 353.0, + 1888.0, + 294.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 547.0, + 1849.0, + 594.0, + 1849.0, + 594.0, + 1888.0, + 547.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 614.0, + 1849.0, + 1406.0, + 1849.0, + 1406.0, + 1888.0, + 614.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1882.0, + 1406.0, + 1882.0, + 1406.0, + 1917.0, + 294.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1912.0, + 1406.0, + 1912.0, + 1406.0, + 1948.0, + 292.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1946.0, + 1403.0, + 1946.0, + 1403.0, + 1979.0, + 295.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1975.0, + 910.0, + 1975.0, + 910.0, + 2011.0, + 295.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1374.0, + 1978.0, + 1403.0, + 1978.0, + 1403.0, + 2004.0, + 1374.0, + 2004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 819.0, + 1202.0, + 819.0, + 1202.0, + 857.0, + 294.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 819.0, + 1407.0, + 819.0, + 1407.0, + 857.0, + 1252.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 850.0, + 364.0, + 850.0, + 364.0, + 893.0, + 291.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 850.0, + 436.0, + 850.0, + 436.0, + 893.0, + 385.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 454.0, + 850.0, + 1010.0, + 850.0, + 1010.0, + 893.0, + 454.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1121.0, + 850.0, + 1175.0, + 850.0, + 1175.0, + 893.0, + 1121.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1264.0, + 850.0, + 1407.0, + 850.0, + 1407.0, + 893.0, + 1264.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 886.0, + 1405.0, + 886.0, + 1405.0, + 921.0, + 294.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 916.0, + 1405.0, + 916.0, + 1405.0, + 950.0, + 294.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 944.0, + 437.0, + 944.0, + 437.0, + 986.0, + 294.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 488.0, + 944.0, + 603.0, + 944.0, + 603.0, + 986.0, + 488.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 632.0, + 944.0, + 821.0, + 944.0, + 821.0, + 986.0, + 632.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 932.0, + 944.0, + 1407.0, + 944.0, + 1407.0, + 986.0, + 932.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 980.0, + 1210.0, + 980.0, + 1210.0, + 1015.0, + 294.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 980.0, + 1407.0, + 980.0, + 1407.0, + 1015.0, + 1246.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1009.0, + 406.0, + 1009.0, + 406.0, + 1044.0, + 294.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 495.0, + 1009.0, + 1144.0, + 1009.0, + 1144.0, + 1044.0, + 495.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 1009.0, + 1406.0, + 1009.0, + 1406.0, + 1044.0, + 1181.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1039.0, + 1195.0, + 1039.0, + 1195.0, + 1074.0, + 295.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1284.0, + 1039.0, + 1406.0, + 1039.0, + 1406.0, + 1074.0, + 1284.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1068.0, + 1057.0, + 1068.0, + 1057.0, + 1107.0, + 293.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1094.0, + 1068.0, + 1405.0, + 1068.0, + 1405.0, + 1107.0, + 1094.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1099.0, + 1081.0, + 1099.0, + 1081.0, + 1138.0, + 294.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1106.0, + 1099.0, + 1407.0, + 1099.0, + 1407.0, + 1138.0, + 1106.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1132.0, + 1330.0, + 1132.0, + 1330.0, + 1163.0, + 296.0, + 1163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1373.0, + 1134.0, + 1402.0, + 1134.0, + 1402.0, + 1159.0, + 1373.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 212.0, + 508.0, + 212.0, + 508.0, + 248.0, + 294.0, + 248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1031.0, + 214.0, + 1406.0, + 214.0, + 1406.0, + 249.0, + 1031.0, + 249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 387.0, + 259.0, + 632.0, + 259.0, + 632.0, + 297.0, + 387.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 651.0, + 259.0, + 1016.0, + 259.0, + 1016.0, + 297.0, + 651.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 558.0, + 203.5, + 629.0, + 203.5, + 629.0, + 237.0, + 558.0, + 237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 206.5, + 830.0, + 206.5, + 830.0, + 256.5, + 735.0, + 256.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 371.0, + 370.0, + 371.0, + 370.0, + 413.0, + 296.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 371.0, + 487.0, + 371.0, + 487.0, + 413.0, + 391.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 596.0, + 371.0, + 1035.0, + 371.0, + 1035.0, + 413.0, + 596.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 429.0, + 389.0, + 429.0, + 389.0, + 473.0, + 301.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 429.0, + 771.0, + 429.0, + 771.0, + 473.0, + 423.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 799.0, + 429.0, + 910.0, + 429.0, + 910.0, + 473.0, + 799.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 782.0, + 1341.0, + 782.0, + 1341.0, + 821.0, + 294.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 724.0, + 1172.0, + 724.0, + 1172.0, + 764.0, + 341.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 724.0, + 1268.0, + 724.0, + 1268.0, + 764.0, + 1260.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 622.0, + 982.0, + 622.0, + 982.0, + 664.0, + 339.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 622.0, + 1406.0, + 622.0, + 1406.0, + 664.0, + 1092.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 656.0, + 533.0, + 656.0, + 533.0, + 702.0, + 339.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 698.0, + 656.0, + 1406.0, + 656.0, + 1406.0, + 702.0, + 698.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 693.0, + 822.0, + 693.0, + 822.0, + 728.0, + 342.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 589.0, + 436.0, + 589.0, + 436.0, + 630.0, + 294.0, + 630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 589.0, + 653.0, + 589.0, + 653.0, + 630.0, + 545.0, + 630.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 296, + 676, + 1407, + 676, + 1407, + 1020, + 296, + 1020 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 298, + 1812, + 1404, + 1812, + 1404, + 2003, + 298, + 2003 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 297, + 1025, + 1406, + 1025, + 1406, + 1290, + 297, + 1290 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 298, + 1298, + 1407, + 1298, + 1407, + 1400, + 298, + 1400 + ], + "score": 0.966 + }, + { + "category_id": 1, + "poly": [ + 297, + 1648, + 1406, + 1648, + 1406, + 1713, + 297, + 1713 + ], + "score": 0.95 + }, + { + "category_id": 8, + "poly": [ + 612, + 1480, + 1129, + 1480, + 1129, + 1552, + 612, + 1552 + ], + "score": 0.943 + }, + { + "category_id": 0, + "poly": [ + 298, + 1753, + 801, + 1753, + 801, + 1789, + 298, + 1789 + ], + "score": 0.924 + }, + { + "category_id": 1, + "poly": [ + 288, + 1585, + 1295, + 1585, + 1295, + 1625, + 288, + 1625 + ], + "score": 0.914 + }, + { + "category_id": 1, + "poly": [ + 302, + 1423, + 900, + 1423, + 900, + 1460, + 302, + 1460 + ], + "score": 0.908 + }, + { + "category_id": 1, + "poly": [ + 305, + 246, + 1355, + 246, + 1355, + 610, + 305, + 610 + ], + "score": 0.768 + }, + { + "category_id": 2, + "poly": [ + 841, + 2060, + 858, + 2060, + 858, + 2084, + 841, + 2084 + ], + "score": 0.714 + }, + { + "category_id": 0, + "poly": [ + 299, + 202, + 711, + 202, + 711, + 235, + 299, + 235 + ], + "score": 0.479 + }, + { + "category_id": 2, + "poly": [ + 841, + 2060, + 859, + 2060, + 859, + 2084, + 841, + 2084 + ], + "score": 0.124 + }, + { + "category_id": 13, + "poly": [ + 999, + 854, + 1163, + 854, + 1163, + 891, + 999, + 891 + ], + "score": 0.95, + "latex": "\\mathcal { O } ( K ^ { - 1 } \\epsilon ^ { - 1 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 784, + 1025, + 980, + 1025, + 980, + 1079, + 784, + 1079 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { \\tilde { \\mathcal { O } } \\biggl ( \\frac { \\zeta ^ { 2 } } { K ^ { 2 \\nu / 3 } T ^ { 1 - \\nu / 3 } } \\biggr ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 899, + 1585, + 986, + 1585, + 986, + 1624, + 899, + 1624 + ], + "score": 0.93, + "latex": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )" + }, + { + "category_id": 13, + "poly": [ + 651, + 1585, + 760, + 1585, + 760, + 1624, + 651, + 1624 + ], + "score": 0.93, + "latex": "\\tilde { \\mathcal O } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1294, + 818, + 1403, + 818, + 1403, + 858, + 1294, + 858 + ], + "score": 0.93, + "latex": "\\tilde { \\mathcal O } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1102, + 1934, + 1189, + 1934, + 1189, + 1969, + 1102, + 1969 + ], + "score": 0.93, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 2 } )" + }, + { + "category_id": 13, + "poly": [ + 712, + 819, + 821, + 819, + 821, + 857, + 712, + 857 + ], + "score": 0.93, + "latex": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 345, + 1966, + 454, + 1966, + 454, + 2003, + 345, + 2003 + ], + "score": 0.93, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 986, + 784, + 1149, + 784, + 1149, + 821, + 986, + 821 + ], + "score": 0.92, + "latex": "\\mathcal { O } ( K ^ { - 1 } \\epsilon ^ { - 1 / 2 } )" + }, + { + "category_id": 14, + "poly": [ + 613, + 1480, + 1131, + 1480, + 1131, + 1550, + 613, + 1550 + ], + "score": 0.92, + "latex": "\\mathbb { E } \\| \\nabla f ( \\bar { x } _ { a } ) \\| ^ { 2 } = \\mathcal { O } \\Big ( \\frac { f ( \\bar { x } _ { 1 } ) - f ^ { * } } { T } \\Big ) + \\tilde { \\mathcal { O } } \\Big ( \\frac { \\sigma ^ { 2 } } { T } \\Big ) ." + }, + { + "category_id": 13, + "poly": [ + 591, + 888, + 677, + 888, + 677, + 924, + 591, + 924 + ], + "score": 0.92, + "latex": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )" + }, + { + "category_id": 13, + "poly": [ + 977, + 893, + 1087, + 893, + 1087, + 926, + 977, + 926 + ], + "score": 0.92, + "latex": "b = \\mathcal { O } ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 1134, + 924, + 1243, + 924, + 1243, + 961, + 1134, + 961 + ], + "score": 0.91, + "latex": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 676, + 745, + 1082, + 745, + 1082, + 787, + 676, + 787 + ], + "score": 0.91, + "latex": "\\tilde { \\mathcal { O } } \\big ( \\operatorname* { m a x } \\big \\{ \\epsilon ^ { - 1 } , ( b \\cdot I ) ^ { - 1 } K ^ { - 1 } \\epsilon ^ { - 3 / 2 } \\big \\} \\big )" + }, + { + "category_id": 13, + "poly": [ + 572, + 1872, + 701, + 1872, + 701, + 1905, + 572, + 1905 + ], + "score": 0.91, + "latex": "\\bar { d } _ { t } = 0 , \\forall t )" + }, + { + "category_id": 13, + "poly": [ + 606, + 925, + 714, + 925, + 714, + 960, + 606, + 960 + ], + "score": 0.91, + "latex": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 558, + 570, + 741, + 570, + 741, + 605, + 558, + 605 + ], + "score": 0.91, + "latex": "a \\sim \\mathcal { U } \\{ 1 , . . . , T \\}" + }, + { + "category_id": 13, + "poly": [ + 1248, + 708, + 1404, + 708, + 1404, + 749, + 1248, + 749 + ], + "score": 0.91, + "latex": "\\bar { \\tilde { O } } ( \\operatorname* { m a x } \\big \\{ ( b \\cdot" + }, + { + "category_id": 13, + "poly": [ + 428, + 246, + 548, + 246, + 548, + 282, + 428, + 282 + ], + "score": 0.9, + "latex": "\\{ \\eta _ { t } \\} _ { t = 0 } ^ { T } ; I" + }, + { + "category_id": 13, + "poly": [ + 1136, + 892, + 1247, + 892, + 1247, + 924, + 1136, + 924 + ], + "score": 0.9, + "latex": "I = { \\mathcal { O } } ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 822, + 1335, + 998, + 1335, + 998, + 1370, + 822, + 1370 + ], + "score": 0.9, + "latex": "b = ( T / K ^ { 2 } ) ^ { 1 / 2 }" + }, + { + "category_id": 13, + "poly": [ + 914, + 1198, + 983, + 1198, + 983, + 1226, + 914, + 1226 + ], + "score": 0.89, + "latex": "I = 1" + }, + { + "category_id": 13, + "poly": [ + 1290, + 1337, + 1398, + 1337, + 1398, + 1364, + 1290, + 1364 + ], + "score": 0.88, + "latex": "B = \\boldsymbol { b } \\cdot \\boldsymbol { I }" + }, + { + "category_id": 13, + "poly": [ + 386, + 282, + 450, + 282, + 450, + 307, + 386, + 307 + ], + "score": 0.88, + "latex": "t = 1" + }, + { + "category_id": 13, + "poly": [ + 297, + 745, + 523, + 745, + 523, + 787, + 297, + 787 + ], + "score": 0.88, + "latex": "I ) \\epsilon ^ { - 1 } , K ^ { - 1 } \\epsilon ^ { - 3 / 2 } \\rbrace )" + }, + { + "category_id": 13, + "poly": [ + 733, + 790, + 783, + 790, + 783, + 815, + 733, + 815 + ], + "score": 0.88, + "latex": "b \\cdot I" + }, + { + "category_id": 13, + "poly": [ + 1325, + 1106, + 1354, + 1106, + 1354, + 1139, + 1325, + 1139 + ], + "score": 0.87, + "latex": "\\zeta ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 794, + 1845, + 948, + 1845, + 948, + 1876, + 794, + 1876 + ], + "score": 0.87, + "latex": "a _ { t } = 1 , ~ \\forall ~ t )" + }, + { + "category_id": 13, + "poly": [ + 742, + 1337, + 809, + 1337, + 809, + 1365, + 742, + 1365 + ], + "score": 0.86, + "latex": "I = 1" + }, + { + "category_id": 13, + "poly": [ + 390, + 1430, + 422, + 1430, + 422, + 1458, + 390, + 1458 + ], + "score": 0.86, + "latex": "\\bar { x } _ { a }" + }, + { + "category_id": 13, + "poly": [ + 728, + 859, + 783, + 859, + 783, + 886, + 728, + 886 + ], + "score": 0.86, + "latex": "b \\cdot I" + }, + { + "category_id": 13, + "poly": [ + 446, + 574, + 478, + 574, + 478, + 602, + 446, + 602 + ], + "score": 0.85, + "latex": "\\scriptstyle { \\bar { x } } _ { a }" + }, + { + "category_id": 13, + "poly": [ + 428, + 311, + 497, + 311, + 497, + 337, + 428, + 337 + ], + "score": 0.84, + "latex": "k = 1" + }, + { + "category_id": 13, + "poly": [ + 1151, + 1138, + 1169, + 1138, + 1169, + 1169, + 1151, + 1169 + ], + "score": 0.83, + "latex": "\\zeta" + }, + { + "category_id": 13, + "poly": [ + 1227, + 1035, + 1262, + 1035, + 1262, + 1071, + 1227, + 1071 + ], + "score": 0.82, + "latex": "\\textcircled{4}" + }, + { + "category_id": 13, + "poly": [ + 626, + 1199, + 644, + 1199, + 644, + 1230, + 626, + 1230 + ], + "score": 0.82, + "latex": "\\zeta" + }, + { + "category_id": 13, + "poly": [ + 711, + 957, + 751, + 957, + 751, + 993, + 711, + 993 + ], + "score": 0.82, + "latex": "\\boxed { 1 \\mathrm { a } }" + }, + { + "category_id": 13, + "poly": [ + 886, + 1080, + 904, + 1080, + 904, + 1108, + 886, + 1108 + ], + "score": 0.8, + "latex": "\\zeta" + }, + { + "category_id": 13, + "poly": [ + 813, + 1103, + 841, + 1103, + 841, + 1141, + 813, + 1141 + ], + "score": 0.8, + "latex": "\\bigstar" + }, + { + "category_id": 13, + "poly": [ + 472, + 990, + 490, + 990, + 490, + 1016, + 472, + 1016 + ], + "score": 0.78, + "latex": "I" + }, + { + "category_id": 13, + "poly": [ + 721, + 1907, + 739, + 1907, + 739, + 1932, + 721, + 1932 + ], + "score": 0.78, + "latex": "I" + }, + { + "category_id": 13, + "poly": [ + 671, + 1937, + 689, + 1937, + 689, + 1963, + 671, + 1963 + ], + "score": 0.77, + "latex": "I" + }, + { + "category_id": 13, + "poly": [ + 458, + 1330, + 507, + 1330, + 507, + 1372, + 458, + 1372 + ], + "score": 0.74, + "latex": "3 . I ." + }, + { + "category_id": 13, + "poly": [ + 529, + 310, + 558, + 310, + 558, + 337, + 529, + 337 + ], + "score": 0.73, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 482, + 283, + 506, + 283, + 506, + 307, + 482, + 307 + ], + "score": 0.73, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 606, + 1937, + 623, + 1937, + 623, + 1963, + 606, + 1963 + ], + "score": 0.71, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 888, + 1299, + 980, + 1299, + 980, + 1335, + 888, + 1335 + ], + "score": 0.7, + "latex": "\\bigstar \\bigstar \\bigstar | \\bigstar |" + }, + { + "category_id": 13, + "poly": [ + 771, + 1425, + 798, + 1425, + 798, + 1462, + 771, + 1462 + ], + "score": 0.69, + "latex": "\\boldsymbol { l } ," + }, + { + "category_id": 13, + "poly": [ + 1103, + 251, + 1118, + 251, + 1118, + 278, + 1103, + 278 + ], + "score": 0.68, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 406, + 990, + 422, + 990, + 422, + 1016, + 406, + 1016 + ], + "score": 0.65, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 655, + 1165, + 773, + 1165, + 773, + 1197, + 655, + 1197 + ], + "score": 0.59, + "latex": "[ \\bigcirc , \\bigcirc , \\bigcirc , \\bigcirc , \\bigcirc ]" + }, + { + "category_id": 13, + "poly": [ + 524, + 421, + 593, + 421, + 593, + 446, + 524, + 446 + ], + "score": 0.59, + "latex": "I = 0" + }, + { + "category_id": 13, + "poly": [ + 1001, + 1906, + 1017, + 1906, + 1017, + 1932, + 1001, + 1932 + ], + "score": 0.53, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 427, + 382, + 672, + 382, + 672, + 422, + 427, + 422 + ], + "score": 0.46, + "latex": "x _ { t + 1 } ^ { ( k ) } = x _ { t } ^ { ( k ) } - \\eta _ { t } d _ { t } ^ { ( k ) }" + }, + { + "category_id": 14, + "poly": [ + 453, + 440, + 800, + 440, + 800, + 504, + 453, + 504 + ], + "score": 0.46, + "latex": "\\begin{array} { r } { \\mathbf { \\Lambda } _ { x _ { t + 1 } ^ { ( k ) } } ^ { ( k ) } = \\bar { x } _ { t + 1 } = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } x _ { t + 1 } ^ { ( k ) } } \\\\ { \\mathbf { u } _ { \\mathrm { ~ i f ~ } } ^ { ( 1 ) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 347, + 707, + 412, + 707, + 412, + 745, + 347, + 745 + ], + "score": 0.41, + "latex": "\\left| \\overline { { \\mathbf { C . 1 0 } } } \\right|" + }, + { + "category_id": 14, + "poly": [ + 426, + 322, + 1041, + 322, + 1041, + 510, + 426, + 510 + ], + "score": 0.39, + "latex": "\\begin{array} { r l } & { \\mathcal { \\kappa } _ { t } ^ { = } \\stackrel { \\mathrm { ~ L ~ U ~ O ~ } \\Lambda } { = } \\mathbf { 0 } } \\\\ & { d _ { t } ^ { ( k ) } = \\frac { 1 } { b } \\sum _ { \\xi _ { t } ^ { ( k ) } \\in \\mathcal { B } _ { t } ^ { ( k ) } } \\nabla f ^ { ( k ) } ( x _ { t } ^ { ( k ) } ; \\xi _ { t } ^ { ( k ) } ) \\mathrm { ~ w i t h ~ } | \\mathcal { B } _ { t } ^ { ( k ) } | = b } \\\\ & { x _ { t + 1 } ^ { ( k ) } = x _ { t } ^ { ( k ) } - \\eta _ { t } d _ { t } ^ { ( k ) } } \\\\ & { \\mathbf { i f } t \\operatorname* { m o d } I = 0 \\mathbf { \\Lambda } \\mathbf { t h e n } } \\\\ & { ~ x _ { t + 1 } ^ { ( k ) } = \\bar { x } _ { t + 1 } = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } x _ { t + 1 } ^ { ( k ) } } \\\\ & { \\mathbf { e n d } \\mathbf { \\Phi } \\mathbf { i f } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 426, + 331, + 867, + 331, + 867, + 383, + 426, + 383 + ], + "score": 0.39, + "latex": "\\begin{array} { r } { \\ddot { d } _ { t } ^ { ( k ) } = \\frac { 1 } { b } \\sum _ { \\xi _ { t } ^ { ( k ) } \\in \\mathcal { B } _ { t } ^ { ( k ) } } \\nabla f ^ { ( k ) } ( x _ { t } ^ { ( k ) } ; \\xi _ { t } ^ { ( k ) } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 453, + 443, + 799, + 443, + 799, + 493, + 453, + 493 + ], + "score": 0.37, + "latex": "\\begin{array} { r } { x _ { t + 1 } ^ { ( k ) } = \\bar { x } _ { t + 1 } = \\frac { 1 } { K } \\sum _ { k = 1 } ^ { K } x _ { t + 1 } ^ { ( k ) } } \\end{array}" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1749.0, + 801.0, + 1749.0, + 801.0, + 1794.0, + 292.0, + 1794.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2059.0, + 860.0, + 2059.0, + 860.0, + 2090.0, + 839.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 197.0, + 712.0, + 197.0, + 712.0, + 239.0, + 294.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2059.0, + 860.0, + 2059.0, + 860.0, + 2091.0, + 839.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 674.0, + 1407.0, + 674.0, + 1407.0, + 714.0, + 294.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 707.0, + 346.0, + 707.0, + 346.0, + 751.0, + 293.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 413.0, + 707.0, + 1247.0, + 707.0, + 1247.0, + 751.0, + 413.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 742.0, + 296.0, + 742.0, + 296.0, + 789.0, + 289.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 524.0, + 742.0, + 675.0, + 742.0, + 675.0, + 789.0, + 524.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1083.0, + 742.0, + 1410.0, + 742.0, + 1410.0, + 789.0, + 1083.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 784.0, + 732.0, + 784.0, + 732.0, + 823.0, + 294.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 784.0, + 985.0, + 784.0, + 985.0, + 823.0, + 784.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1150.0, + 784.0, + 1409.0, + 784.0, + 1409.0, + 823.0, + 1150.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 818.0, + 711.0, + 818.0, + 711.0, + 861.0, + 293.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 818.0, + 1293.0, + 818.0, + 1293.0, + 861.0, + 822.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 818.0, + 1408.0, + 818.0, + 1408.0, + 861.0, + 1404.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 852.0, + 727.0, + 852.0, + 727.0, + 895.0, + 291.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 852.0, + 998.0, + 852.0, + 998.0, + 895.0, + 784.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1164.0, + 852.0, + 1408.0, + 852.0, + 1408.0, + 895.0, + 1164.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 890.0, + 590.0, + 890.0, + 590.0, + 929.0, + 294.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 890.0, + 976.0, + 890.0, + 976.0, + 929.0, + 678.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 890.0, + 1135.0, + 890.0, + 1135.0, + 929.0, + 1088.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1248.0, + 890.0, + 1408.0, + 890.0, + 1408.0, + 929.0, + 1248.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 919.0, + 605.0, + 919.0, + 605.0, + 965.0, + 291.0, + 965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 919.0, + 1133.0, + 919.0, + 1133.0, + 965.0, + 715.0, + 965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1244.0, + 919.0, + 1410.0, + 919.0, + 1410.0, + 965.0, + 1244.0, + 965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 955.0, + 710.0, + 955.0, + 710.0, + 993.0, + 294.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 752.0, + 955.0, + 1406.0, + 955.0, + 1406.0, + 993.0, + 752.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 988.0, + 405.0, + 988.0, + 405.0, + 1023.0, + 296.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 988.0, + 471.0, + 988.0, + 471.0, + 1023.0, + 423.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 988.0, + 859.0, + 988.0, + 859.0, + 1023.0, + 491.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1374.0, + 989.0, + 1405.0, + 989.0, + 1405.0, + 1021.0, + 1374.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1813.0, + 1406.0, + 1813.0, + 1406.0, + 1849.0, + 294.0, + 1849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1844.0, + 793.0, + 1844.0, + 793.0, + 1880.0, + 293.0, + 1880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 949.0, + 1844.0, + 1406.0, + 1844.0, + 1406.0, + 1880.0, + 949.0, + 1880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1874.0, + 571.0, + 1874.0, + 571.0, + 1910.0, + 293.0, + 1910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 702.0, + 1874.0, + 1406.0, + 1874.0, + 1406.0, + 1910.0, + 702.0, + 1910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1901.0, + 720.0, + 1901.0, + 720.0, + 1940.0, + 293.0, + 1940.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1901.0, + 1000.0, + 1901.0, + 1000.0, + 1940.0, + 740.0, + 1940.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1018.0, + 1901.0, + 1405.0, + 1901.0, + 1405.0, + 1940.0, + 1018.0, + 1940.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1930.0, + 605.0, + 1930.0, + 605.0, + 1974.0, + 293.0, + 1974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 624.0, + 1930.0, + 670.0, + 1930.0, + 670.0, + 1974.0, + 624.0, + 1974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 1930.0, + 1101.0, + 1930.0, + 1101.0, + 1974.0, + 690.0, + 1974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1190.0, + 1930.0, + 1404.0, + 1930.0, + 1404.0, + 1974.0, + 1190.0, + 1974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1961.0, + 344.0, + 1961.0, + 344.0, + 2009.0, + 291.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 1961.0, + 776.0, + 1961.0, + 776.0, + 2009.0, + 455.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1023.0, + 783.0, + 1023.0, + 783.0, + 1074.0, + 292.0, + 1074.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 1031.0, + 1226.0, + 1031.0, + 1226.0, + 1075.0, + 981.0, + 1075.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 1031.0, + 1406.0, + 1031.0, + 1406.0, + 1075.0, + 1263.0, + 1075.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1075.0, + 885.0, + 1075.0, + 885.0, + 1111.0, + 296.0, + 1111.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 905.0, + 1075.0, + 1404.0, + 1075.0, + 1404.0, + 1111.0, + 905.0, + 1111.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1106.0, + 812.0, + 1106.0, + 812.0, + 1141.0, + 294.0, + 1141.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 842.0, + 1106.0, + 1324.0, + 1106.0, + 1324.0, + 1141.0, + 842.0, + 1141.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1355.0, + 1106.0, + 1406.0, + 1106.0, + 1406.0, + 1141.0, + 1355.0, + 1141.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1136.0, + 1150.0, + 1136.0, + 1150.0, + 1171.0, + 295.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1170.0, + 1136.0, + 1406.0, + 1136.0, + 1406.0, + 1171.0, + 1170.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1165.0, + 654.0, + 1165.0, + 654.0, + 1199.0, + 294.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 1165.0, + 1406.0, + 1165.0, + 1406.0, + 1199.0, + 774.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1194.0, + 625.0, + 1194.0, + 625.0, + 1230.0, + 294.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 1194.0, + 913.0, + 1194.0, + 913.0, + 1230.0, + 645.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 1194.0, + 1404.0, + 1194.0, + 1404.0, + 1230.0, + 984.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1226.0, + 1407.0, + 1226.0, + 1407.0, + 1261.0, + 295.0, + 1261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1255.0, + 622.0, + 1255.0, + 622.0, + 1294.0, + 293.0, + 1294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1373.0, + 1258.0, + 1406.0, + 1258.0, + 1406.0, + 1289.0, + 1373.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1298.0, + 887.0, + 1298.0, + 887.0, + 1338.0, + 296.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 1298.0, + 1407.0, + 1298.0, + 1407.0, + 1338.0, + 981.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1329.0, + 457.0, + 1329.0, + 457.0, + 1373.0, + 291.0, + 1373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 1329.0, + 741.0, + 1329.0, + 741.0, + 1373.0, + 508.0, + 1373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 810.0, + 1329.0, + 821.0, + 1329.0, + 821.0, + 1373.0, + 810.0, + 1373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 1329.0, + 1289.0, + 1329.0, + 1289.0, + 1373.0, + 999.0, + 1373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 1329.0, + 1412.0, + 1329.0, + 1412.0, + 1373.0, + 1399.0, + 1373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1363.0, + 547.0, + 1363.0, + 547.0, + 1402.0, + 294.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1646.0, + 1404.0, + 1646.0, + 1404.0, + 1687.0, + 292.0, + 1687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1681.0, + 676.0, + 1681.0, + 676.0, + 1713.0, + 295.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1580.0, + 650.0, + 1580.0, + 650.0, + 1633.0, + 290.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 761.0, + 1580.0, + 898.0, + 1580.0, + 898.0, + 1633.0, + 761.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 1580.0, + 1301.0, + 1580.0, + 1301.0, + 1633.0, + 987.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1423.0, + 389.0, + 1423.0, + 389.0, + 1464.0, + 303.0, + 1464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 1423.0, + 770.0, + 1423.0, + 770.0, + 1464.0, + 423.0, + 1464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 799.0, + 1423.0, + 900.0, + 1423.0, + 900.0, + 1464.0, + 799.0, + 1464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 244.0, + 427.0, + 244.0, + 427.0, + 285.0, + 307.0, + 285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 244.0, + 1102.0, + 244.0, + 1102.0, + 285.0, + 549.0, + 285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 244.0, + 1357.0, + 244.0, + 1357.0, + 285.0, + 1119.0, + 285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 277.0, + 385.0, + 277.0, + 385.0, + 312.0, + 307.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 277.0, + 481.0, + 277.0, + 481.0, + 312.0, + 451.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 507.0, + 277.0, + 548.0, + 277.0, + 548.0, + 312.0, + 507.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 310.0, + 340.0, + 310.0, + 340.0, + 376.0, + 307.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 308.0, + 425.0, + 308.0, + 425.0, + 400.0, + 383.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 308.0, + 1060.0, + 308.0, + 1060.0, + 400.0, + 1042.0, + 400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 384.0, + 341.0, + 384.0, + 341.0, + 423.0, + 306.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 419.0, + 374.0, + 425.0, + 374.0, + 425.0, + 432.0, + 419.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 418.0, + 340.0, + 418.0, + 340.0, + 452.0, + 305.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 422.0, + 413.0, + 425.0, + 413.0, + 425.0, + 451.0, + 422.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 447.0, + 340.0, + 447.0, + 340.0, + 484.0, + 307.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 480.0, + 339.0, + 480.0, + 339.0, + 511.0, + 308.0, + 511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 422.0, + 479.0, + 425.0, + 479.0, + 425.0, + 509.0, + 422.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 508.0, + 340.0, + 508.0, + 340.0, + 543.0, + 305.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 380.0, + 504.0, + 482.0, + 504.0, + 482.0, + 543.0, + 380.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 538.0, + 437.0, + 538.0, + 437.0, + 572.0, + 297.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 568.0, + 445.0, + 568.0, + 445.0, + 606.0, + 296.0, + 606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 568.0, + 557.0, + 568.0, + 557.0, + 606.0, + 479.0, + 606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 568.0, + 751.0, + 568.0, + 751.0, + 606.0, + 742.0, + 606.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 375.0, + 691.0, + 375.0, + 691.0, + 435.0, + 576.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 452.0, + 513.0, + 452.0, + 513.0, + 489.5, + 451.0, + 489.5 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1270, + 1405, + 1270, + 1405, + 1579, + 297, + 1579 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1589, + 1404, + 1589, + 1404, + 1804, + 298, + 1804 + ], + "score": 0.98 + }, + { + "category_id": 5, + "poly": [ + 892, + 197, + 1447, + 197, + 1447, + 515, + 892, + 515 + ], + "score": 0.978, + "html": "
AlgorithmTraining Acc.Testing Acc.
FedAvg73.675.4
FedProx80.075.2
FedDyn76.171.3
SCAFFOLD72.573.7
MIME61.558.6
FedGLOMO10.010.0
STEM81.178.5
" + }, + { + "category_id": 5, + "poly": [ + 296, + 197, + 851, + 197, + 851, + 516, + 296, + 516 + ], + "score": 0.977, + "html": "
Algorithm Training Acc.Testing Acc.
FedAvg78.274.1
FedProx79.274.8
FedDyn68.966.0
SCAFFOLD71.974.0
MIME82.676.8
FedGLOMO76.172.8
STEM80.178.8
" + }, + { + "category_id": 1, + "poly": [ + 299, + 1913, + 1402, + 1913, + 1402, + 2007, + 299, + 2007 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 297, + 715, + 1405, + 715, + 1405, + 798, + 297, + 798 + ], + "score": 0.963 + }, + { + "category_id": 8, + "poly": [ + 378, + 983, + 1364, + 983, + 1364, + 1062, + 378, + 1062 + ], + "score": 0.936 + }, + { + "category_id": 8, + "poly": [ + 572, + 817, + 1126, + 817, + 1126, + 863, + 572, + 863 + ], + "score": 0.934 + }, + { + "category_id": 7, + "poly": [ + 328, + 528, + 762, + 528, + 762, + 559, + 328, + 559 + ], + "score": 0.914 + }, + { + "category_id": 1, + "poly": [ + 292, + 876, + 1293, + 876, + 1293, + 913, + 292, + 913 + ], + "score": 0.912 + }, + { + "category_id": 0, + "poly": [ + 300, + 1845, + 602, + 1845, + 602, + 1883, + 300, + 1883 + ], + "score": 0.911 + }, + { + "category_id": 1, + "poly": [ + 301, + 935, + 899, + 935, + 899, + 972, + 301, + 972 + ], + "score": 0.909 + }, + { + "category_id": 9, + "poly": [ + 1365, + 826, + 1401, + 826, + 1401, + 856, + 1365, + 856 + ], + "score": 0.841 + }, + { + "category_id": 7, + "poly": [ + 905, + 528, + 1383, + 528, + 1383, + 559, + 905, + 559 + ], + "score": 0.82 + }, + { + "category_id": 1, + "poly": [ + 296, + 1090, + 761, + 1090, + 761, + 1123, + 296, + 1123 + ], + "score": 0.81 + }, + { + "category_id": 2, + "poly": [ + 840, + 2062, + 858, + 2062, + 858, + 2085, + 840, + 2085 + ], + "score": 0.799 + }, + { + "category_id": 1, + "poly": [ + 336, + 1122, + 1403, + 1122, + 1403, + 1211, + 336, + 1211 + ], + "score": 0.752 + }, + { + "category_id": 1, + "poly": [ + 347, + 1212, + 1296, + 1212, + 1296, + 1246, + 347, + 1246 + ], + "score": 0.68 + }, + { + "category_id": 1, + "poly": [ + 297, + 575, + 1402, + 575, + 1402, + 639, + 297, + 639 + ], + "score": 0.431 + }, + { + "category_id": 7, + "poly": [ + 297, + 575, + 1402, + 575, + 1402, + 639, + 297, + 639 + ], + "score": 0.162 + }, + { + "category_id": 14, + "poly": [ + 381, + 984, + 1365, + 984, + 1365, + 1062, + 381, + 1062 + ], + "score": 0.95, + "latex": "\\mathbb { E } \\Vert \\nabla f ( \\bar { x } _ { a } ) \\Vert ^ { 2 } = \\mathcal { O } \\Bigg ( \\frac { f ( \\bar { x } _ { 1 } ) - f ^ { * } } { K ^ { \\nu / 2 } T ^ { 2 / 3 - \\nu / 6 } } \\Bigg ) + \\mathcal { O } \\Bigg ( \\frac { \\sigma ^ { 2 } } { K ^ { \\nu / 2 } T ^ { 2 / 3 - \\nu / 6 } } \\Bigg ) + \\mathcal { O } \\Bigg ( \\frac { \\zeta ^ { 2 } } { K ^ { \\nu / 2 } T ^ { 2 / 3 - \\nu / 6 } } \\Bigg ) ." + }, + { + "category_id": 13, + "poly": [ + 340, + 748, + 458, + 748, + 458, + 804, + 340, + 804 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { \\eta = \\sqrt { \\frac { b K } { T } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 997, + 1120, + 1084, + 1120, + 1084, + 1155, + 997, + 1155 + ], + "score": 0.92, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1188, + 530, + 1250, + 530, + 1250, + 557, + 1188, + 557 + ], + "score": 0.92, + "latex": "b = 8" + }, + { + "category_id": 13, + "poly": [ + 884, + 877, + 1066, + 877, + 1066, + 910, + 884, + 910 + ], + "score": 0.91, + "latex": "T \\geq 8 1 L ^ { 2 } I ^ { 2 } b K" + }, + { + "category_id": 13, + "poly": [ + 1267, + 1303, + 1373, + 1303, + 1373, + 1336, + 1267, + 1336 + ], + "score": 0.91, + "latex": "\\nu \\in [ 0 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 372, + 879, + 478, + 879, + 478, + 913, + 372, + 913 + ], + "score": 0.91, + "latex": "\\nu \\in [ 0 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 530, + 1151, + 671, + 1151, + 671, + 1185, + 530, + 1185 + ], + "score": 0.91, + "latex": "\\tilde { \\mathcal { O } } ( K ^ { - 1 } \\epsilon ^ { - 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1178, + 1209, + 1286, + 1209, + 1286, + 1246, + 1178, + 1246 + ], + "score": 0.9, + "latex": "\\mathcal { O } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 545, + 1092, + 652, + 1092, + 652, + 1124, + 545, + 1124 + ], + "score": 0.9, + "latex": "\\nu \\in [ 0 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 387, + 1425, + 461, + 1425, + 461, + 1452, + 387, + 1452 + ], + "score": 0.9, + "latex": "\\nu = 0" + }, + { + "category_id": 13, + "poly": [ + 933, + 1454, + 1001, + 1454, + 1001, + 1482, + 933, + 1482 + ], + "score": 0.9, + "latex": "I = 1" + }, + { + "category_id": 14, + "poly": [ + 573, + 816, + 1124, + 816, + 1124, + 862, + 573, + 862 + ], + "score": 0.89, + "latex": "I = \\mathcal { O } \\big ( ( T / K ^ { 3 } ) ^ { \\nu / 4 } \\big ) , \\quad b = \\mathcal { O } \\big ( ( T / K ^ { 3 } ) ^ { 1 / 3 - \\nu / 3 } \\big )" + }, + { + "category_id": 13, + "poly": [ + 543, + 1395, + 618, + 1395, + 618, + 1422, + 543, + 1422 + ], + "score": 0.89, + "latex": "\\nu = 1" + }, + { + "category_id": 13, + "poly": [ + 566, + 530, + 640, + 530, + 640, + 557, + 566, + 557 + ], + "score": 0.88, + "latex": "b = 6 4" + }, + { + "category_id": 13, + "poly": [ + 1300, + 530, + 1377, + 530, + 1377, + 557, + 1300, + 557 + ], + "score": 0.88, + "latex": "I = 6 1" + }, + { + "category_id": 13, + "poly": [ + 391, + 942, + 422, + 942, + 422, + 969, + 391, + 969 + ], + "score": 0.85, + "latex": "\\scriptstyle { \\bar { x } } _ { a }" + }, + { + "category_id": 13, + "poly": [ + 739, + 1301, + 787, + 1301, + 787, + 1333, + 739, + 1333 + ], + "score": 0.84, + "latex": "[ \\mathbb { 1 4 } ]" + }, + { + "category_id": 13, + "poly": [ + 886, + 1393, + 938, + 1393, + 938, + 1428, + 886, + 1428 + ], + "score": 0.83, + "latex": "\\textcircled { 3 . 2 }" + }, + { + "category_id": 13, + "poly": [ + 637, + 1273, + 660, + 1273, + 660, + 1299, + 637, + 1299 + ], + "score": 0.8, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 690, + 1514, + 739, + 1514, + 739, + 1545, + 690, + 1545 + ], + "score": 0.77, + "latex": "[ \\textcircled { 1 4 } ]" + }, + { + "category_id": 13, + "poly": [ + 691, + 530, + 755, + 530, + 755, + 557, + 691, + 557 + ], + "score": 0.77, + "latex": "I = 7" + }, + { + "category_id": 13, + "poly": [ + 792, + 1334, + 809, + 1334, + 809, + 1360, + 792, + 1360 + ], + "score": 0.77, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 580, + 1331, + 626, + 1331, + 626, + 1368, + 580, + 1368 + ], + "score": 0.76, + "latex": "3 . 2" + }, + { + "category_id": 13, + "poly": [ + 959, + 1743, + 979, + 1743, + 979, + 1769, + 959, + 1769 + ], + "score": 0.75, + "latex": "I" + }, + { + "category_id": 13, + "poly": [ + 725, + 1334, + 744, + 1334, + 744, + 1360, + 725, + 1360 + ], + "score": 0.74, + "latex": "I" + }, + { + "category_id": 13, + "poly": [ + 857, + 1361, + 882, + 1361, + 882, + 1396, + 857, + 1396 + ], + "score": 0.72, + "latex": "^ 1" + }, + { + "category_id": 13, + "poly": [ + 905, + 1743, + 921, + 1743, + 921, + 1769, + 905, + 1769 + ], + "score": 0.72, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 1184, + 1713, + 1203, + 1713, + 1203, + 1739, + 1184, + 1739 + ], + "score": 0.72, + "latex": "I" + }, + { + "category_id": 13, + "poly": [ + 1050, + 1945, + 1078, + 1945, + 1078, + 1980, + 1050, + 1980 + ], + "score": 0.71, + "latex": "^ { 1 . }" + }, + { + "category_id": 13, + "poly": [ + 983, + 1652, + 999, + 1652, + 999, + 1679, + 983, + 1679 + ], + "score": 0.69, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 831, + 1484, + 851, + 1484, + 851, + 1515, + 831, + 1515 + ], + "score": 0.68, + "latex": "\\zeta" + }, + { + "category_id": 13, + "poly": [ + 907, + 1976, + 954, + 1976, + 954, + 2007, + 907, + 2007 + ], + "score": 0.64, + "latex": "\\pmb { \\mathbb { B } } 6 \\|" + }, + { + "category_id": 13, + "poly": [ + 737, + 1652, + 755, + 1652, + 755, + 1678, + 737, + 1678 + ], + "score": 0.62, + "latex": "I" + }, + { + "category_id": 13, + "poly": [ + 1112, + 1713, + 1128, + 1713, + 1128, + 1740, + 1112, + 1740 + ], + "score": 0.61, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 1323, + 1482, + 1350, + 1482, + 1350, + 1519, + 1323, + 1519 + ], + "score": 0.56, + "latex": "^ 1" + }, + { + "category_id": 13, + "poly": [ + 772, + 938, + 798, + 938, + 798, + 974, + 772, + 974 + ], + "score": 0.53, + "latex": "\\perp" + }, + { + "category_id": 13, + "poly": [ + 831, + 1482, + 903, + 1482, + 903, + 1515, + 831, + 1515 + ], + "score": 0.38, + "latex": "\\zeta \\equiv \\mathbb { I I } 3 \\mathbb { I }" + }, + { + "category_id": 13, + "poly": [ + 767, + 1543, + 816, + 1543, + 816, + 1580, + 767, + 1580 + ], + "score": 0.33, + "latex": "3 . 2 ." + }, + { + "category_id": 15, + "poly": [ + 328.0, + 527.0, + 565.0, + 527.0, + 565.0, + 563.0, + 328.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 641.0, + 527.0, + 690.0, + 527.0, + 690.0, + 563.0, + 641.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 756.0, + 527.0, + 766.0, + 527.0, + 766.0, + 563.0, + 756.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1839.0, + 606.0, + 1839.0, + 606.0, + 1890.0, + 292.0, + 1890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 526.0, + 1187.0, + 526.0, + 1187.0, + 563.0, + 902.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 526.0, + 1299.0, + 526.0, + 1299.0, + 563.0, + 1251.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 526.0, + 1388.0, + 526.0, + 1388.0, + 563.0, + 1378.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2090.0, + 838.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 574.0, + 1406.0, + 574.0, + 1406.0, + 610.0, + 295.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 604.0, + 1014.0, + 604.0, + 1014.0, + 643.0, + 294.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1267.0, + 636.0, + 1267.0, + 636.0, + 1307.0, + 294.0, + 1307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 661.0, + 1267.0, + 1407.0, + 1267.0, + 1407.0, + 1307.0, + 661.0, + 1307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1299.0, + 738.0, + 1299.0, + 738.0, + 1335.0, + 295.0, + 1335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 788.0, + 1299.0, + 1266.0, + 1299.0, + 1266.0, + 1335.0, + 788.0, + 1335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1374.0, + 1299.0, + 1403.0, + 1299.0, + 1403.0, + 1335.0, + 1374.0, + 1335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1329.0, + 579.0, + 1329.0, + 579.0, + 1367.0, + 294.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 627.0, + 1329.0, + 724.0, + 1329.0, + 724.0, + 1367.0, + 627.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 745.0, + 1329.0, + 791.0, + 1329.0, + 791.0, + 1367.0, + 745.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 810.0, + 1329.0, + 1405.0, + 1329.0, + 1405.0, + 1367.0, + 810.0, + 1367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1362.0, + 856.0, + 1362.0, + 856.0, + 1398.0, + 295.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 1362.0, + 1405.0, + 1362.0, + 1405.0, + 1398.0, + 883.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1390.0, + 542.0, + 1390.0, + 542.0, + 1428.0, + 292.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 619.0, + 1390.0, + 885.0, + 1390.0, + 885.0, + 1428.0, + 619.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 1390.0, + 1403.0, + 1390.0, + 1403.0, + 1428.0, + 939.0, + 1428.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1422.0, + 386.0, + 1422.0, + 386.0, + 1456.0, + 295.0, + 1456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 462.0, + 1422.0, + 1403.0, + 1422.0, + 1403.0, + 1456.0, + 462.0, + 1456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1450.0, + 932.0, + 1450.0, + 932.0, + 1488.0, + 296.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 1450.0, + 1405.0, + 1450.0, + 1405.0, + 1488.0, + 1002.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1482.0, + 830.0, + 1482.0, + 830.0, + 1518.0, + 294.0, + 1518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 1482.0, + 1322.0, + 1482.0, + 1322.0, + 1518.0, + 904.0, + 1518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1351.0, + 1482.0, + 1403.0, + 1482.0, + 1403.0, + 1518.0, + 1351.0, + 1518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1513.0, + 689.0, + 1513.0, + 689.0, + 1548.0, + 295.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 740.0, + 1513.0, + 1407.0, + 1513.0, + 1407.0, + 1548.0, + 740.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1543.0, + 766.0, + 1543.0, + 766.0, + 1580.0, + 295.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1591.0, + 1404.0, + 1591.0, + 1404.0, + 1625.0, + 294.0, + 1625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1618.0, + 1404.0, + 1618.0, + 1404.0, + 1656.0, + 293.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1648.0, + 736.0, + 1648.0, + 736.0, + 1685.0, + 293.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 756.0, + 1648.0, + 982.0, + 1648.0, + 982.0, + 1685.0, + 756.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 1648.0, + 1404.0, + 1648.0, + 1404.0, + 1685.0, + 1000.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1683.0, + 1404.0, + 1683.0, + 1404.0, + 1714.0, + 296.0, + 1714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1710.0, + 1111.0, + 1710.0, + 1111.0, + 1744.0, + 294.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 1710.0, + 1183.0, + 1710.0, + 1183.0, + 1744.0, + 1129.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1204.0, + 1710.0, + 1405.0, + 1710.0, + 1405.0, + 1744.0, + 1204.0, + 1744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1741.0, + 904.0, + 1741.0, + 904.0, + 1779.0, + 293.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 922.0, + 1741.0, + 958.0, + 1741.0, + 958.0, + 1779.0, + 922.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 1741.0, + 1405.0, + 1741.0, + 1405.0, + 1779.0, + 980.0, + 1779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1773.0, + 1029.0, + 1773.0, + 1029.0, + 1807.0, + 296.0, + 1807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1914.0, + 1406.0, + 1914.0, + 1406.0, + 1948.0, + 294.0, + 1948.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1944.0, + 1049.0, + 1944.0, + 1049.0, + 1977.0, + 294.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 1944.0, + 1406.0, + 1944.0, + 1406.0, + 1977.0, + 1079.0, + 1977.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1975.0, + 906.0, + 1975.0, + 906.0, + 2012.0, + 294.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 955.0, + 1975.0, + 1406.0, + 1975.0, + 1406.0, + 2012.0, + 955.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 713.0, + 1407.0, + 713.0, + 1407.0, + 755.0, + 294.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 747.0, + 339.0, + 747.0, + 339.0, + 805.0, + 294.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 459.0, + 747.0, + 599.0, + 747.0, + 599.0, + 805.0, + 459.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 875.0, + 371.0, + 875.0, + 371.0, + 916.0, + 295.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 875.0, + 883.0, + 875.0, + 883.0, + 916.0, + 479.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1067.0, + 875.0, + 1297.0, + 875.0, + 1297.0, + 916.0, + 1067.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 935.0, + 390.0, + 935.0, + 390.0, + 976.0, + 302.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 935.0, + 771.0, + 935.0, + 771.0, + 976.0, + 423.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 799.0, + 935.0, + 900.0, + 935.0, + 900.0, + 976.0, + 799.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1088.0, + 544.0, + 1088.0, + 544.0, + 1128.0, + 294.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 1088.0, + 762.0, + 1088.0, + 762.0, + 1128.0, + 653.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1119.0, + 996.0, + 1119.0, + 996.0, + 1156.0, + 342.0, + 1156.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1085.0, + 1119.0, + 1406.0, + 1119.0, + 1406.0, + 1156.0, + 1085.0, + 1156.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1148.0, + 529.0, + 1148.0, + 529.0, + 1188.0, + 341.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 1148.0, + 1406.0, + 1148.0, + 1406.0, + 1188.0, + 672.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1182.0, + 694.0, + 1182.0, + 694.0, + 1215.0, + 341.0, + 1215.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 1205.0, + 1177.0, + 1205.0, + 1177.0, + 1251.0, + 341.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1287.0, + 1205.0, + 1298.0, + 1205.0, + 1298.0, + 1251.0, + 1287.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 574.0, + 1406.0, + 574.0, + 1406.0, + 610.0, + 295.0, + 610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 604.0, + 1014.0, + 604.0, + 1014.0, + 643.0, + 294.0, + 643.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1425, + 1404, + 1425, + 1404, + 1732, + 297, + 1732 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 298, + 1745, + 1404, + 1745, + 1404, + 1928, + 298, + 1928 + ], + "score": 0.978 + }, + { + "category_id": 5, + "poly": [ + 899, + 197, + 1453, + 197, + 1453, + 516, + 899, + 516 + ], + "score": 0.977, + "html": "
AlgorithmTraining Acc.Testing Acc.
FedAvg40.139.2
FedProx43.543.2
FedDyn43.743.2
SCAFFOLD40.341.3
MIME32.132.1
FedGLOMO40.340.1
STEM44.543.8
" + }, + { + "category_id": 5, + "poly": [ + 296, + 197, + 850, + 197, + 850, + 516, + 296, + 516 + ], + "score": 0.977, + "html": "
AlgorithmTraining Acc.Testing Acc.
FedAvg57.657.1
FedProx59.158.5
FedDyn51.251.3
SCAFFOLD53.154.7
MIME56.155.1
FedGLOMO56.856.1
STEM58.557.4
" + }, + { + "category_id": 1, + "poly": [ + 298, + 1228, + 1404, + 1228, + 1404, + 1413, + 298, + 1413 + ], + "score": 0.976 + }, + { + "category_id": 3, + "poly": [ + 318, + 707, + 1347, + 707, + 1347, + 1082, + 318, + 1082 + ], + "score": 0.967 + }, + { + "category_id": 4, + "poly": [ + 296, + 1099, + 1403, + 1099, + 1403, + 1163, + 296, + 1163 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 298, + 1944, + 1399, + 1944, + 1399, + 2008, + 298, + 2008 + ], + "score": 0.938 + }, + { + "category_id": 6, + "poly": [ + 900, + 538, + 1404, + 538, + 1404, + 601, + 900, + 601 + ], + "score": 0.859 + }, + { + "category_id": 6, + "poly": [ + 295, + 538, + 799, + 538, + 799, + 630, + 295, + 630 + ], + "score": 0.839 + }, + { + "category_id": 2, + "poly": [ + 840, + 2062, + 858, + 2062, + 858, + 2084, + 840, + 2084 + ], + "score": 0.799 + }, + { + "category_id": 13, + "poly": [ + 540, + 1974, + 648, + 1974, + 648, + 2009, + 540, + 2009 + ], + "score": 0.93, + "latex": "a _ { t } = { c } \\eta _ { t } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 967, + 1941, + 1215, + 1941, + 1215, + 1979, + 967, + 1979 + ], + "score": 0.92, + "latex": "\\eta _ { t } = \\bar { \\kappa } / ( w _ { t } + \\sigma ^ { 2 } t ) ^ { 1 / 3 }" + }, + { + "category_id": 13, + "poly": [ + 1199, + 1131, + 1266, + 1131, + 1266, + 1158, + 1199, + 1158 + ], + "score": 0.89, + "latex": "b = 8" + }, + { + "category_id": 13, + "poly": [ + 391, + 600, + 459, + 600, + 459, + 628, + 391, + 628 + ], + "score": 0.89, + "latex": "I = 6" + }, + { + "category_id": 13, + "poly": [ + 1173, + 1837, + 1228, + 1837, + 1228, + 1866, + 1173, + 1866 + ], + "score": 0.87, + "latex": "20 \\%" + }, + { + "category_id": 13, + "poly": [ + 915, + 1837, + 970, + 1837, + 970, + 1866, + 915, + 1866 + ], + "score": 0.87, + "latex": "80 \\%" + }, + { + "category_id": 13, + "poly": [ + 887, + 1775, + 936, + 1775, + 936, + 1806, + 887, + 1806 + ], + "score": 0.86, + "latex": "\\dot { \\left[ \\left| 1 0 \\right| \\right] }" + }, + { + "category_id": 13, + "poly": [ + 751, + 569, + 800, + 569, + 800, + 599, + 751, + 599 + ], + "score": 0.83, + "latex": "b =" + }, + { + "category_id": 13, + "poly": [ + 381, + 1774, + 429, + 1774, + 429, + 1806, + 381, + 1806 + ], + "score": 0.76, + "latex": " { \\mathbb { I } } ^ { \\smash { \\sum } }" + }, + { + "category_id": 13, + "poly": [ + 381, + 1486, + 430, + 1486, + 430, + 1519, + 381, + 1519 + ], + "score": 0.52, + "latex": "\\pmb { \\Vert 3 7 } \\Vert" + }, + { + "category_id": 13, + "poly": [ + 686, + 838, + 755, + 838, + 755, + 861, + 686, + 861 + ], + "score": 0.37, + "latex": "\\mathrm { I } = 5 3 6 " + }, + { + "category_id": 13, + "poly": [ + 1247, + 875, + 1304, + 875, + 1304, + 899, + 1247, + 899 + ], + "score": 0.31, + "latex": "\\mathrm { I } = 6 7 " + }, + { + "category_id": 13, + "poly": [ + 1281, + 1046, + 1338, + 1046, + 1338, + 1080, + 1281, + 1080 + ], + "score": 0.31, + "latex": "\\times { 1 0 } ^ { 4 }" + }, + { + "category_id": 13, + "poly": [ + 687, + 765, + 743, + 765, + 743, + 788, + 687, + 788 + ], + "score": 0.27, + "latex": "\\mathrm { I } = 6 7" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 706.0, + 386.0, + 706.0, + 386.0, + 736.0, + 363.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 718.0, + 928.0, + 718.0, + 928.0, + 728.0, + 920.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1296.0, + 729.0, + 1325.0, + 729.0, + 1325.0, + 737.0, + 1296.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 735.0, + 705.0, + 735.0, + 705.0, + 771.0, + 541.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 735.0, + 961.0, + 735.0, + 961.0, + 906.0, + 838.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 388.0, + 759.0, + 403.0, + 759.0, + 403.0, + 775.0, + 388.0, + 775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 557.0, + 762.0, + 686.0, + 762.0, + 686.0, + 792.0, + 557.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 744.0, + 762.0, + 748.0, + 762.0, + 748.0, + 792.0, + 744.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 551.0, + 769.0, + 568.0, + 769.0, + 568.0, + 784.0, + 551.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 343.0, + 777.0, + 388.0, + 777.0, + 388.0, + 812.0, + 343.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 781.0, + 361.0, + 781.0, + 361.0, + 951.0, + 320.0, + 951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 546.0, + 784.0, + 695.0, + 784.0, + 695.0, + 817.0, + 546.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 551.0, + 807.0, + 717.0, + 807.0, + 717.0, + 843.0, + 551.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 987.0, + 810.0, + 1004.0, + 810.0, + 1004.0, + 821.0, + 987.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 859.0, + 380.0, + 859.0, + 380.0, + 877.0, + 368.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 833.0, + 685.0, + 833.0, + 685.0, + 888.0, + 554.0, + 888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 820.0, + 937.0, + 820.0, + 937.0, + 966.0, + 856.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 949.0, + 840.0, + 983.0, + 840.0, + 983.0, + 895.0, + 949.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1102.0, + 845.0, + 1246.0, + 845.0, + 1246.0, + 904.0, + 1102.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1305.0, + 845.0, + 1310.0, + 845.0, + 1310.0, + 904.0, + 1305.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 893.0, + 416.0, + 893.0, + 416.0, + 907.0, + 407.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1104.0, + 895.0, + 1254.0, + 895.0, + 1254.0, + 928.0, + 1104.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 344.0, + 922.0, + 389.0, + 922.0, + 389.0, + 961.0, + 344.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 934.0, + 940.0, + 934.0, + 940.0, + 973.0, + 889.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1110.0, + 917.0, + 1319.0, + 917.0, + 1319.0, + 974.0, + 1110.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 476.0, + 976.0, + 497.0, + 976.0, + 497.0, + 988.0, + 476.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 973.0, + 952.0, + 973.0, + 952.0, + 1004.0, + 936.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1116.0, + 969.0, + 1263.0, + 969.0, + 1263.0, + 998.0, + 1116.0, + 998.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 366.0, + 1003.0, + 401.0, + 1003.0, + 401.0, + 1051.0, + 366.0, + 1051.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 442.0, + 1019.0, + 489.0, + 1019.0, + 489.0, + 1054.0, + 442.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 1022.0, + 556.0, + 1022.0, + 556.0, + 1051.0, + 535.0, + 1051.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 1018.0, + 647.0, + 1018.0, + 647.0, + 1054.0, + 601.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 688.0, + 1020.0, + 716.0, + 1020.0, + 716.0, + 1052.0, + 688.0, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 1018.0, + 804.0, + 1018.0, + 804.0, + 1053.0, + 757.0, + 1053.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 990.0, + 945.0, + 990.0, + 945.0, + 1044.0, + 878.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1012.0, + 1038.0, + 1012.0, + 1038.0, + 1049.0, + 992.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 1017.0, + 1104.0, + 1017.0, + 1104.0, + 1043.0, + 1087.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1152.0, + 1013.0, + 1199.0, + 1013.0, + 1199.0, + 1049.0, + 1152.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1242.0, + 1014.0, + 1269.0, + 1014.0, + 1269.0, + 1048.0, + 1242.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1310.0, + 1011.0, + 1358.0, + 1011.0, + 1358.0, + 1051.0, + 1310.0, + 1051.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 1049.0, + 683.0, + 1049.0, + 683.0, + 1085.0, + 479.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1014.0, + 1035.0, + 1225.0, + 1035.0, + 1225.0, + 1075.0, + 1014.0, + 1075.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 762.0, + 1095.0, + 762.0, + 1095.0, + 779.0, + 1056.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 779.5, + 1073.0, + 779.5, + 1073.0, + 803.5, + 998.0, + 803.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 813.5, + 1017.0, + 813.5, + 1017.0, + 842.0, + 956.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 928.0, + 838.5, + 977.0, + 838.5, + 977.0, + 864.0, + 928.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 418.0, + 929.0, + 466.0, + 929.0, + 466.0, + 956.0, + 418.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 973.5, + 477.0, + 973.5, + 477.0, + 985.5, + 448.0, + 985.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 712.0, + 989.5, + 802.0, + 989.5, + 802.0, + 1017.5, + 712.0, + 1017.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 731.0, + 1046.0, + 814.0, + 1046.0, + 814.0, + 1090.5, + 731.0, + 1090.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1268.0, + 1041.0, + 1353.0, + 1041.0, + 1353.0, + 1085.0, + 1268.0, + 1085.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1096.0, + 1404.0, + 1096.0, + 1404.0, + 1135.0, + 294.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1130.0, + 1198.0, + 1130.0, + 1198.0, + 1166.0, + 295.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1267.0, + 1130.0, + 1277.0, + 1130.0, + 1277.0, + 1166.0, + 1267.0, + 1166.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 535.0, + 1404.0, + 535.0, + 1404.0, + 575.0, + 899.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 903.0, + 568.0, + 1138.0, + 568.0, + 1138.0, + 601.0, + 903.0, + 601.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 536.0, + 798.0, + 536.0, + 798.0, + 574.0, + 295.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 567.0, + 750.0, + 567.0, + 750.0, + 603.0, + 296.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 599.0, + 390.0, + 599.0, + 390.0, + 631.0, + 296.0, + 631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 599.0, + 471.0, + 599.0, + 471.0, + 631.0, + 460.0, + 631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2060.0, + 860.0, + 2060.0, + 860.0, + 2090.0, + 839.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1427.0, + 1406.0, + 1427.0, + 1406.0, + 1459.0, + 295.0, + 1459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1454.0, + 1404.0, + 1454.0, + 1404.0, + 1493.0, + 292.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1484.0, + 380.0, + 1484.0, + 380.0, + 1523.0, + 296.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.0, + 1484.0, + 1406.0, + 1484.0, + 1406.0, + 1523.0, + 431.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1519.0, + 1405.0, + 1519.0, + 1405.0, + 1551.0, + 296.0, + 1551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1547.0, + 1402.0, + 1547.0, + 1402.0, + 1582.0, + 295.0, + 1582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1579.0, + 1405.0, + 1579.0, + 1405.0, + 1614.0, + 295.0, + 1614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1608.0, + 1405.0, + 1608.0, + 1405.0, + 1643.0, + 294.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1640.0, + 1406.0, + 1640.0, + 1406.0, + 1675.0, + 294.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1670.0, + 1406.0, + 1670.0, + 1406.0, + 1706.0, + 295.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1697.0, + 1409.0, + 1697.0, + 1409.0, + 1736.0, + 292.0, + 1736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1743.0, + 1404.0, + 1743.0, + 1404.0, + 1778.0, + 294.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1772.0, + 380.0, + 1772.0, + 380.0, + 1810.0, + 292.0, + 1810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1772.0, + 886.0, + 1772.0, + 886.0, + 1810.0, + 430.0, + 1810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 937.0, + 1772.0, + 1404.0, + 1772.0, + 1404.0, + 1810.0, + 937.0, + 1810.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1806.0, + 1404.0, + 1806.0, + 1404.0, + 1838.0, + 296.0, + 1838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1837.0, + 914.0, + 1837.0, + 914.0, + 1869.0, + 293.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 971.0, + 1837.0, + 1172.0, + 1837.0, + 1172.0, + 1869.0, + 971.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1229.0, + 1837.0, + 1404.0, + 1837.0, + 1404.0, + 1869.0, + 1229.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1866.0, + 1405.0, + 1866.0, + 1405.0, + 1901.0, + 293.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1895.0, + 1247.0, + 1895.0, + 1247.0, + 1933.0, + 291.0, + 1933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1228.0, + 1405.0, + 1228.0, + 1405.0, + 1265.0, + 292.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1257.0, + 1406.0, + 1257.0, + 1406.0, + 1299.0, + 292.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1290.0, + 1406.0, + 1290.0, + 1406.0, + 1326.0, + 293.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1322.0, + 1405.0, + 1322.0, + 1405.0, + 1354.0, + 296.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1350.0, + 1405.0, + 1350.0, + 1405.0, + 1389.0, + 293.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1380.0, + 1408.0, + 1380.0, + 1408.0, + 1418.0, + 293.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1939.0, + 966.0, + 1939.0, + 966.0, + 1982.0, + 292.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 1939.0, + 1405.0, + 1939.0, + 1405.0, + 1982.0, + 1216.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1975.0, + 539.0, + 1975.0, + 539.0, + 2009.0, + 293.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 649.0, + 1975.0, + 1404.0, + 1975.0, + 1404.0, + 2009.0, + 649.0, + 2009.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1196, + 1406, + 1196, + 1406, + 1510, + 297, + 1510 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 297, + 507, + 1406, + 507, + 1406, + 1086, + 297, + 1086 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 298, + 202, + 1405, + 202, + 1405, + 478, + 298, + 478 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1622, + 1405, + 1622, + 1405, + 1805, + 298, + 1805 + ], + "score": 0.978 + }, + { + "category_id": 0, + "poly": [ + 301, + 1552, + 562, + 1552, + 562, + 1590, + 301, + 1590 + ], + "score": 0.912 + }, + { + "category_id": 0, + "poly": [ + 299, + 1129, + 459, + 1129, + 459, + 1166, + 299, + 1166 + ], + "score": 0.903 + }, + { + "category_id": 2, + "poly": [ + 837, + 2062, + 865, + 2062, + 865, + 2085, + 837, + 2085 + ], + "score": 0.851 + }, + { + "category_id": 13, + "poly": [ + 759, + 232, + 863, + 232, + 863, + 266, + 759, + 266 + ], + "score": 0.93, + "latex": "c \\doteq \\bar { c } / \\bar { \\kappa } ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 839, + 1293, + 925, + 1293, + 925, + 1331, + 839, + 1331 + ], + "score": 0.93, + "latex": "\\tilde { \\mathcal { O } } ( \\epsilon ^ { - 1 } )" + }, + { + "category_id": 13, + "poly": [ + 1101, + 444, + 1362, + 444, + 1362, + 479, + 1101, + 479 + ], + "score": 0.93, + "latex": "\\{ 1 0 ^ { 1 } , 1 0 ^ { 0 } , 1 0 ^ { - 1 } , \\hat { 1 0 } ^ { - 2 } \\}" + }, + { + "category_id": 13, + "poly": [ + 1294, + 1229, + 1403, + 1229, + 1403, + 1266, + 1294, + 1266 + ], + "score": 0.93, + "latex": "\\tilde { \\mathcal O } ( \\epsilon ^ { - 3 / 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1006, + 232, + 1201, + 232, + 1201, + 267, + 1006, + 267 + ], + "score": 0.91, + "latex": "\\bar { \\kappa } \\in [ 1 0 ^ { - 1 } , \\overline { { 1 0 } } ^ { - 2 } ]" + }, + { + "category_id": 13, + "poly": [ + 559, + 232, + 708, + 232, + 708, + 262, + 559, + 262 + ], + "score": 0.91, + "latex": "w _ { t } = \\sigma ^ { 2 } = 1" + }, + { + "category_id": 13, + "poly": [ + 1249, + 355, + 1355, + 355, + 1355, + 385, + 1249, + 385 + ], + "score": 0.91, + "latex": "\\beta _ { k } = 0 . 2" + }, + { + "category_id": 13, + "poly": [ + 329, + 934, + 447, + 934, + 447, + 964, + 329, + 964 + ], + "score": 0.91, + "latex": "9 6 - 9 7 \\%" + }, + { + "category_id": 13, + "poly": [ + 1315, + 813, + 1397, + 813, + 1397, + 841, + 1315, + 841 + ], + "score": 0.9, + "latex": "I = 6 7" + }, + { + "category_id": 13, + "poly": [ + 647, + 813, + 715, + 813, + 715, + 841, + 647, + 841 + ], + "score": 0.89, + "latex": "b = 8" + }, + { + "category_id": 13, + "poly": [ + 510, + 844, + 609, + 844, + 609, + 872, + 510, + 872 + ], + "score": 0.89, + "latex": "I = 5 3 6" + }, + { + "category_id": 13, + "poly": [ + 298, + 966, + 394, + 966, + 394, + 993, + 298, + 993 + ], + "score": 0.89, + "latex": "I = 5 3 6" + }, + { + "category_id": 13, + "poly": [ + 507, + 935, + 593, + 935, + 593, + 963, + 507, + 963 + ], + "score": 0.88, + "latex": "I = 6 7" + }, + { + "category_id": 13, + "poly": [ + 297, + 995, + 366, + 995, + 366, + 1024, + 297, + 1024 + ], + "score": 0.88, + "latex": "I > 1" + }, + { + "category_id": 13, + "poly": [ + 910, + 293, + 958, + 293, + 958, + 325, + 910, + 325 + ], + "score": 0.84, + "latex": "\\lVert \\bar { 3 6 } \\rVert" + }, + { + "category_id": 13, + "poly": [ + 1299, + 202, + 1347, + 202, + 1347, + 237, + 1299, + 237 + ], + "score": 0.84, + "latex": "\\boxed { 3 . 1 }" + }, + { + "category_id": 13, + "poly": [ + 961, + 935, + 1111, + 935, + 1111, + 964, + 961, + 964 + ], + "score": 0.83, + "latex": "5 0 0 0 - 6 0 0 0" + }, + { + "category_id": 13, + "poly": [ + 462, + 262, + 773, + 262, + 773, + 296, + 462, + 296 + ], + "score": 0.82, + "latex": "{ \\bar { \\kappa } } \\in \\{ 1 0 ^ { 1 } , 1 0 ^ { 0 } , 1 0 ^ { - 1 } , 1 0 ^ { - 2 } \\}" + }, + { + "category_id": 13, + "poly": [ + 1047, + 324, + 1095, + 324, + 1095, + 355, + 1047, + 355 + ], + "score": 0.78, + "latex": "\\mathbb { \\lVert 1 7 \\rVert }" + }, + { + "category_id": 13, + "poly": [ + 869, + 325, + 917, + 325, + 917, + 355, + 869, + 355 + ], + "score": 0.75, + "latex": "\\pm \\textcircled { 1 8 } \\textcircled { 1 }" + }, + { + "category_id": 13, + "poly": [ + 717, + 296, + 765, + 296, + 765, + 325, + 717, + 325 + ], + "score": 0.74, + "latex": "[ \\equiv ]" + }, + { + "category_id": 13, + "poly": [ + 688, + 902, + 718, + 902, + 718, + 938, + 688, + 938 + ], + "score": 0.74, + "latex": "2 ." + }, + { + "category_id": 13, + "poly": [ + 961, + 903, + 987, + 903, + 987, + 937, + 961, + 937 + ], + "score": 0.73, + "latex": "\\bigtriangledown" + }, + { + "category_id": 13, + "poly": [ + 1278, + 723, + 1297, + 723, + 1297, + 749, + 1278, + 749 + ], + "score": 0.71, + "latex": "I" + }, + { + "category_id": 13, + "poly": [ + 1029, + 1238, + 1045, + 1238, + 1045, + 1261, + 1029, + 1261 + ], + "score": 0.68, + "latex": "\\epsilon" + }, + { + "category_id": 13, + "poly": [ + 379, + 567, + 405, + 567, + 405, + 605, + 379, + 605 + ], + "score": 0.67, + "latex": "\\bigstar ," + }, + { + "category_id": 13, + "poly": [ + 403, + 749, + 432, + 749, + 432, + 787, + 403, + 787 + ], + "score": 0.63, + "latex": "\\bigstar ," + }, + { + "category_id": 13, + "poly": [ + 777, + 660, + 817, + 660, + 817, + 695, + 777, + 695 + ], + "score": 0.59, + "latex": "\\bigstar" + }, + { + "category_id": 13, + "poly": [ + 336, + 297, + 353, + 297, + 353, + 322, + 336, + 322 + ], + "score": 0.52, + "latex": "\\bar { c }" + }, + { + "category_id": 13, + "poly": [ + 1350, + 384, + 1398, + 384, + 1398, + 416, + 1350, + 416 + ], + "score": 0.5, + "latex": "\\mathbb { \\left[ \\left[ 8 \\right] \\right] }" + }, + { + "category_id": 13, + "poly": [ + 1213, + 723, + 1229, + 723, + 1229, + 750, + 1213, + 750 + ], + "score": 0.49, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 293, + 568, + 333, + 568, + 333, + 605, + 293, + 605 + ], + "score": 0.33, + "latex": "\\boxed { 2 \\mathbf { b } }" + }, + { + "category_id": 13, + "poly": [ + 739, + 689, + 764, + 689, + 764, + 727, + 739, + 727 + ], + "score": 0.29, + "latex": "\\textcircled { 3 }" + }, + { + "category_id": 13, + "poly": [ + 1363, + 538, + 1404, + 538, + 1404, + 575, + 1363, + 575 + ], + "score": 0.27, + "latex": "\\bigstar" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1549.0, + 567.0, + 1549.0, + 567.0, + 1597.0, + 295.0, + 1597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1125.0, + 465.0, + 1125.0, + 465.0, + 1172.0, + 295.0, + 1172.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1196.0, + 1409.0, + 1196.0, + 1409.0, + 1234.0, + 294.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1230.0, + 1028.0, + 1230.0, + 1028.0, + 1268.0, + 292.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1046.0, + 1230.0, + 1293.0, + 1230.0, + 1293.0, + 1268.0, + 1046.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 1230.0, + 1407.0, + 1230.0, + 1407.0, + 1268.0, + 1404.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1261.0, + 1407.0, + 1261.0, + 1407.0, + 1301.0, + 294.0, + 1301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1296.0, + 838.0, + 1296.0, + 838.0, + 1332.0, + 295.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 1296.0, + 1406.0, + 1296.0, + 1406.0, + 1332.0, + 926.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1327.0, + 1406.0, + 1327.0, + 1406.0, + 1363.0, + 295.0, + 1363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1358.0, + 1406.0, + 1358.0, + 1406.0, + 1394.0, + 294.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1387.0, + 1406.0, + 1387.0, + 1406.0, + 1424.0, + 294.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1418.0, + 1407.0, + 1418.0, + 1407.0, + 1455.0, + 292.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1447.0, + 1406.0, + 1447.0, + 1406.0, + 1484.0, + 295.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1479.0, + 1361.0, + 1479.0, + 1361.0, + 1512.0, + 294.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 506.0, + 1406.0, + 506.0, + 1406.0, + 546.0, + 292.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 539.0, + 1362.0, + 539.0, + 1362.0, + 576.0, + 292.0, + 576.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 571.0, + 378.0, + 571.0, + 378.0, + 603.0, + 334.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 571.0, + 1406.0, + 571.0, + 1406.0, + 603.0, + 406.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 599.0, + 1406.0, + 599.0, + 1406.0, + 634.0, + 291.0, + 634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 630.0, + 1404.0, + 630.0, + 1404.0, + 665.0, + 294.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 660.0, + 776.0, + 660.0, + 776.0, + 695.0, + 295.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 818.0, + 660.0, + 1406.0, + 660.0, + 1406.0, + 695.0, + 818.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 689.0, + 738.0, + 689.0, + 738.0, + 728.0, + 294.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 689.0, + 1408.0, + 689.0, + 1408.0, + 728.0, + 765.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 718.0, + 1212.0, + 718.0, + 1212.0, + 759.0, + 292.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1230.0, + 718.0, + 1277.0, + 718.0, + 1277.0, + 759.0, + 1230.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1298.0, + 718.0, + 1409.0, + 718.0, + 1409.0, + 759.0, + 1298.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 750.0, + 402.0, + 750.0, + 402.0, + 787.0, + 294.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 433.0, + 750.0, + 1408.0, + 750.0, + 1408.0, + 787.0, + 433.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 782.0, + 1407.0, + 782.0, + 1407.0, + 816.0, + 292.0, + 816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 813.0, + 646.0, + 813.0, + 646.0, + 848.0, + 294.0, + 848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 813.0, + 1314.0, + 813.0, + 1314.0, + 848.0, + 716.0, + 848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 813.0, + 1407.0, + 813.0, + 1407.0, + 848.0, + 1398.0, + 848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 843.0, + 509.0, + 843.0, + 509.0, + 878.0, + 295.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 843.0, + 1406.0, + 843.0, + 1406.0, + 878.0, + 610.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 873.0, + 1406.0, + 873.0, + 1406.0, + 909.0, + 294.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 903.0, + 687.0, + 903.0, + 687.0, + 941.0, + 294.0, + 941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 903.0, + 960.0, + 903.0, + 960.0, + 941.0, + 719.0, + 941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 903.0, + 1407.0, + 903.0, + 1407.0, + 941.0, + 988.0, + 941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 931.0, + 328.0, + 931.0, + 328.0, + 971.0, + 291.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 931.0, + 506.0, + 931.0, + 506.0, + 971.0, + 448.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.0, + 931.0, + 960.0, + 931.0, + 960.0, + 971.0, + 594.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 931.0, + 1407.0, + 931.0, + 1407.0, + 971.0, + 1112.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 964.0, + 297.0, + 964.0, + 297.0, + 999.0, + 294.0, + 999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 964.0, + 1406.0, + 964.0, + 1406.0, + 999.0, + 395.0, + 999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 992.0, + 296.0, + 992.0, + 296.0, + 1029.0, + 292.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 367.0, + 992.0, + 1406.0, + 992.0, + 1406.0, + 1029.0, + 367.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1025.0, + 1406.0, + 1025.0, + 1406.0, + 1060.0, + 294.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1056.0, + 1312.0, + 1056.0, + 1312.0, + 1091.0, + 295.0, + 1091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 200.0, + 1298.0, + 200.0, + 1298.0, + 238.0, + 293.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1348.0, + 200.0, + 1406.0, + 200.0, + 1406.0, + 238.0, + 1348.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 230.0, + 558.0, + 230.0, + 558.0, + 264.0, + 293.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 230.0, + 758.0, + 230.0, + 758.0, + 264.0, + 709.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 864.0, + 230.0, + 1005.0, + 230.0, + 1005.0, + 264.0, + 864.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1202.0, + 230.0, + 1406.0, + 230.0, + 1406.0, + 264.0, + 1202.0, + 264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 256.0, + 461.0, + 256.0, + 461.0, + 300.0, + 292.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 256.0, + 1408.0, + 256.0, + 1408.0, + 300.0, + 774.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 293.0, + 335.0, + 293.0, + 335.0, + 328.0, + 296.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 293.0, + 716.0, + 293.0, + 716.0, + 328.0, + 354.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 766.0, + 293.0, + 909.0, + 293.0, + 909.0, + 328.0, + 766.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 959.0, + 293.0, + 1405.0, + 293.0, + 1405.0, + 328.0, + 959.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 323.0, + 868.0, + 323.0, + 868.0, + 360.0, + 295.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 918.0, + 323.0, + 1046.0, + 323.0, + 1046.0, + 360.0, + 918.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1096.0, + 323.0, + 1403.0, + 323.0, + 1403.0, + 360.0, + 1096.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 353.0, + 1248.0, + 353.0, + 1248.0, + 388.0, + 293.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1356.0, + 353.0, + 1406.0, + 353.0, + 1406.0, + 388.0, + 1356.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 382.0, + 1349.0, + 382.0, + 1349.0, + 420.0, + 292.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 382.0, + 1407.0, + 382.0, + 1407.0, + 420.0, + 1399.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 414.0, + 1405.0, + 414.0, + 1405.0, + 451.0, + 293.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 442.0, + 1100.0, + 442.0, + 1100.0, + 480.0, + 292.0, + 480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1363.0, + 442.0, + 1373.0, + 442.0, + 1373.0, + 480.0, + 1363.0, + 480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1620.0, + 1407.0, + 1620.0, + 1407.0, + 1659.0, + 293.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1649.0, + 1408.0, + 1649.0, + 1408.0, + 1692.0, + 291.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1678.0, + 1406.0, + 1678.0, + 1406.0, + 1721.0, + 292.0, + 1721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1711.0, + 1405.0, + 1711.0, + 1405.0, + 1749.0, + 293.0, + 1749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1743.0, + 1406.0, + 1743.0, + 1406.0, + 1778.0, + 295.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1773.0, + 670.0, + 1773.0, + 670.0, + 1808.0, + 294.0, + 1808.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 295, + 202, + 1410, + 202, + 1410, + 2025, + 295, + 2025 + ], + "score": 0.893 + }, + { + "category_id": 0, + "poly": [ + 299, + 200, + 456, + 200, + 456, + 235, + 299, + 235 + ], + "score": 0.891 + }, + { + "category_id": 2, + "poly": [ + 835, + 2061, + 863, + 2061, + 863, + 2086, + 835, + 2086 + ], + "score": 0.824 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 196.0, + 460.0, + 196.0, + 460.0, + 241.0, + 295.0, + 241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 868.0, + 2058.0, + 868.0, + 2098.0, + 832.0, + 2098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 196.0, + 463.0, + 196.0, + 463.0, + 243.0, + 294.0, + 243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 248.0, + 1408.0, + 248.0, + 1408.0, + 288.0, + 301.0, + 288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 282.0, + 1314.0, + 282.0, + 1314.0, + 320.0, + 353.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 330.0, + 1408.0, + 330.0, + 1408.0, + 372.0, + 303.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 366.0, + 1406.0, + 366.0, + 1406.0, + 398.0, + 355.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 394.0, + 1406.0, + 394.0, + 1406.0, + 432.0, + 353.0, + 432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 424.0, + 741.0, + 424.0, + 741.0, + 462.0, + 353.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 475.0, + 1408.0, + 475.0, + 1408.0, + 513.0, + 305.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 507.0, + 1406.0, + 507.0, + 1406.0, + 545.0, + 353.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 537.0, + 832.0, + 537.0, + 832.0, + 575.0, + 349.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 585.0, + 1410.0, + 585.0, + 1410.0, + 629.0, + 305.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 619.0, + 1354.0, + 619.0, + 1354.0, + 657.0, + 353.0, + 657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 669.0, + 1406.0, + 669.0, + 1406.0, + 707.0, + 307.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 695.0, + 1408.0, + 695.0, + 1408.0, + 743.0, + 349.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 733.0, + 677.0, + 733.0, + 677.0, + 765.0, + 355.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 781.0, + 1408.0, + 781.0, + 1408.0, + 819.0, + 307.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 815.0, + 807.0, + 815.0, + 807.0, + 847.0, + 355.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 863.0, + 1406.0, + 863.0, + 1406.0, + 901.0, + 307.0, + 901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 889.0, + 1410.0, + 889.0, + 1410.0, + 935.0, + 347.0, + 935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 923.0, + 533.0, + 923.0, + 533.0, + 961.0, + 353.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 975.0, + 1408.0, + 975.0, + 1408.0, + 1013.0, + 307.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1007.0, + 1380.0, + 1007.0, + 1380.0, + 1045.0, + 355.0, + 1045.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 1057.0, + 1406.0, + 1057.0, + 1406.0, + 1095.0, + 307.0, + 1095.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1085.0, + 1348.0, + 1085.0, + 1348.0, + 1130.0, + 351.0, + 1130.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1142.0, + 1404.0, + 1142.0, + 1404.0, + 1174.0, + 297.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1170.0, + 1128.0, + 1170.0, + 1128.0, + 1208.0, + 353.0, + 1208.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1216.0, + 1408.0, + 1216.0, + 1408.0, + 1260.0, + 293.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1252.0, + 1412.0, + 1252.0, + 1412.0, + 1290.0, + 353.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1276.0, + 699.0, + 1276.0, + 699.0, + 1322.0, + 351.0, + 1322.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1332.0, + 1410.0, + 1332.0, + 1410.0, + 1370.0, + 295.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1364.0, + 1406.0, + 1364.0, + 1406.0, + 1402.0, + 353.0, + 1402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1394.0, + 1194.0, + 1394.0, + 1194.0, + 1432.0, + 355.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1442.0, + 1408.0, + 1442.0, + 1408.0, + 1484.0, + 291.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1476.0, + 924.0, + 1476.0, + 924.0, + 1514.0, + 353.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1522.0, + 1412.0, + 1522.0, + 1412.0, + 1568.0, + 291.0, + 1568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1554.0, + 1408.0, + 1554.0, + 1408.0, + 1598.0, + 351.0, + 1598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1584.0, + 832.0, + 1584.0, + 832.0, + 1626.0, + 349.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1636.0, + 1408.0, + 1636.0, + 1408.0, + 1678.0, + 291.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1670.0, + 1410.0, + 1670.0, + 1410.0, + 1708.0, + 353.0, + 1708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1696.0, + 700.0, + 1696.0, + 700.0, + 1738.0, + 351.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1750.0, + 1408.0, + 1750.0, + 1408.0, + 1789.0, + 295.0, + 1789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1779.0, + 1122.0, + 1779.0, + 1122.0, + 1821.0, + 349.0, + 1821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1833.0, + 1406.0, + 1833.0, + 1406.0, + 1871.0, + 295.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1861.0, + 1406.0, + 1861.0, + 1406.0, + 1903.0, + 353.0, + 1903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1893.0, + 645.0, + 1893.0, + 645.0, + 1931.0, + 353.0, + 1931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1945.0, + 1408.0, + 1945.0, + 1408.0, + 1983.0, + 295.0, + 1983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1975.0, + 1234.0, + 1975.0, + 1234.0, + 2013.0, + 353.0, + 2013.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 836, + 2060, + 866, + 2060, + 866, + 2087, + 836, + 2087 + ], + "score": 0.83 + }, + { + "category_id": 1, + "poly": [ + 293, + 151, + 1410, + 151, + 1410, + 2002, + 293, + 2002 + ], + "score": 0.802 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 869.0, + 2058.0, + 869.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 203.0, + 1406.0, + 203.0, + 1406.0, + 241.0, + 295.0, + 241.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 233.0, + 1390.0, + 233.0, + 1390.0, + 272.0, + 353.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 282.0, + 1408.0, + 282.0, + 1408.0, + 325.0, + 291.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 316.0, + 1406.0, + 316.0, + 1406.0, + 355.0, + 353.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 347.0, + 936.0, + 347.0, + 936.0, + 386.0, + 351.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 392.0, + 1412.0, + 392.0, + 1412.0, + 440.0, + 289.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 428.0, + 1408.0, + 428.0, + 1408.0, + 467.0, + 353.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 452.0, + 432.0, + 452.0, + 432.0, + 500.0, + 353.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 512.0, + 1410.0, + 512.0, + 1410.0, + 550.0, + 295.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 540.0, + 505.0, + 540.0, + 505.0, + 579.0, + 351.0, + 579.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 591.0, + 1408.0, + 591.0, + 1408.0, + 635.0, + 293.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 621.0, + 645.0, + 621.0, + 645.0, + 662.0, + 351.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 676.0, + 1406.0, + 676.0, + 1406.0, + 715.0, + 295.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 707.0, + 1048.0, + 707.0, + 1048.0, + 745.0, + 351.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 759.0, + 1408.0, + 759.0, + 1408.0, + 798.0, + 295.0, + 798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 792.0, + 1404.0, + 792.0, + 1404.0, + 824.0, + 355.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 818.0, + 1376.0, + 818.0, + 1376.0, + 857.0, + 351.0, + 857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 871.0, + 1408.0, + 871.0, + 1408.0, + 908.0, + 293.0, + 908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 898.0, + 1408.0, + 898.0, + 1408.0, + 944.0, + 351.0, + 944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 932.0, + 868.0, + 932.0, + 868.0, + 971.0, + 353.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 983.0, + 1406.0, + 983.0, + 1406.0, + 1022.0, + 295.0, + 1022.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1018.0, + 880.0, + 1018.0, + 880.0, + 1050.0, + 355.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1066.0, + 1408.0, + 1066.0, + 1408.0, + 1105.0, + 295.0, + 1105.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1095.0, + 1242.0, + 1095.0, + 1242.0, + 1135.0, + 351.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1146.0, + 1408.0, + 1146.0, + 1408.0, + 1188.0, + 291.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1180.0, + 1204.0, + 1180.0, + 1204.0, + 1217.0, + 351.0, + 1217.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1231.0, + 1410.0, + 1231.0, + 1410.0, + 1270.0, + 295.0, + 1270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1266.0, + 1404.0, + 1266.0, + 1404.0, + 1298.0, + 355.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1294.0, + 956.0, + 1294.0, + 956.0, + 1333.0, + 355.0, + 1333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1345.0, + 1406.0, + 1345.0, + 1406.0, + 1383.0, + 295.0, + 1383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1375.0, + 645.0, + 1375.0, + 645.0, + 1412.0, + 349.0, + 1412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1422.0, + 1408.0, + 1422.0, + 1408.0, + 1469.0, + 291.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1457.0, + 1408.0, + 1457.0, + 1408.0, + 1497.0, + 351.0, + 1497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1485.0, + 833.0, + 1485.0, + 833.0, + 1528.0, + 347.0, + 1528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1540.0, + 1406.0, + 1540.0, + 1406.0, + 1578.0, + 295.0, + 1578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1568.0, + 1408.0, + 1568.0, + 1408.0, + 1611.0, + 351.0, + 1611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1601.0, + 639.0, + 1601.0, + 639.0, + 1639.0, + 353.0, + 1639.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1652.0, + 1404.0, + 1652.0, + 1404.0, + 1692.0, + 293.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1684.0, + 900.0, + 1684.0, + 900.0, + 1723.0, + 351.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1733.0, + 1408.0, + 1733.0, + 1408.0, + 1774.0, + 291.0, + 1774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1767.0, + 1396.0, + 1767.0, + 1396.0, + 1806.0, + 353.0, + 1806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1818.0, + 1406.0, + 1818.0, + 1406.0, + 1853.0, + 293.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1849.0, + 1408.0, + 1849.0, + 1408.0, + 1887.0, + 353.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1879.0, + 517.0, + 1879.0, + 517.0, + 1918.0, + 353.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1930.0, + 1410.0, + 1930.0, + 1410.0, + 1969.0, + 295.0, + 1969.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1961.0, + 1010.0, + 1961.0, + 1010.0, + 1999.0, + 351.0, + 1999.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/KBnXrODoBW/KBnXrODoBW.md b/parse/train/KBnXrODoBW/KBnXrODoBW.md new file mode 100644 index 0000000000000000000000000000000000000000..2935ab5b871b4ccf325292a5f962ad9ef0905eaf --- /dev/null +++ b/parse/train/KBnXrODoBW/KBnXrODoBW.md @@ -0,0 +1,242 @@ +# Pay Attention to MLPs + +Hanxiao Liu, Zihang Dai, David R. So, Quoc V. Le Google Research, Brain Team {hanxiaol,zihangd,davidso,qvl}@google.com + +# Abstract + +Transformers [1] have become one of the most important architectural innovations in deep learning and have enabled many breakthroughs over the past few years. Here we propose a simple network architecture, gMLP, based on MLPs with gating, and show that it can perform as well as Transformers in key language and vision applications. Our comparisons show that self-attention is not critical for Vision Transformers, as $\mathrm { g M L P }$ can achieve the same accuracy. For BERT, our model achieves parity with Transformers on pretraining perplexity and is better on some downstream NLP tasks. On finetuning tasks where gMLP performs worse, making the $\mathrm { g M L P }$ model substantially larger can close the gap with Transformers. In general, our experiments show that gMLP can scale as well as Transformers over increased data and compute. + +# 1 Introduction + +Transformers [1] have enabled many breakthroughs in natural language processing (e.g., [2, 3, 4, 5, 6]) and have been shown to work well for computer vision (e.g., [7, 8, 9, 10]). Thanks to this success, Transformers have largely replaced LSTM-RNN [11] as the default architecture in NLP, and have become an appealing alternative to ConvNets [12, 13, 14, 15, 16, 17] in computer vision. + +The Transformer architecture combines two important concepts: (1) a recurrent-free architecture which computes the representations for each individual token in parallel, and (2) multi-head selfattention blocks which aggregate spatial information across tokens. On one hand, the attention mechanism [18] introduces the inductive bias that the spatial interactions should be dynamically parameterized based on the input representations. On the other hand, it is known that MLPs with static parameterization can represent arbitrary functions [19]. It therefore remains an open question whether the inductive bias in self-attention is essential to the remarkable effectiveness of Transformers. + +Here we study the necessity of self-attention modules in key language and vision applications of Transformers. Specifically, we propose an MLP-based alternative to Transformers without self-attention, which simply consists of channel projections and spatial projections with static parameterization. We experiment with several design choices for this architecture and find spatial projections work well when they are linear and paired with multiplicative gating (Figure 1). We name the model gMLP because it is built out of basic MLP layers with gating. + +We apply $\mathrm { g M L P }$ to image classification and obtain strong results on ImageNet. $\mathrm { g M L P }$ achieves comparable performance with DeiT [8], namely Vision Transformer (ViT) [7] with improved regularization, in a similar training setup. With $66 \%$ less parameters, a gMLP model is $3 \%$ more accurate than MLP-Mixer [20]. Together with Tolstikhin et al. [20], Melas-Kyriazi [21], Touvron et al. [22] and Ding et. al. [23], our results question the necessity of self-attention layers in Vision Transformers. + +We apply gMLP to masked language modeling (MLM) in the BERT [2] setup, one of the most wellestablished applications of Transformers, and find that it is as good as Transformers at minimizing perplexity during pretraining. Our experiments indicate that perplexity is only correlated with model capacity and is insensitive to the presence of self-attention. As capacity increases, we observe that + +def gmlp_block(x, d_model, d_ffn): shortcut $=$ x $\mathrm { ~ x ~ } =$ norm(x, axis="channel") $\mathrm { ~ x ~ } =$ proj(x, d_ffn, axis="channel") $\mathrm { ~ x ~ } =$ gelu(x) $\mathrm { ~ x ~ } =$ spatial_gating_unit(x) $\mathrm { ~ x ~ } =$ proj(x, d_model, axis $= ^ { 1 1 }$ channel") return $\texttt { x + }$ shortcut + +![](images/018d82d7640366b577f21adb6e4e6c60f398a8eb4976b569ff0a2114a750edde.jpg) +Figure 1: Overview of the $\mathrm { g M L P }$ architecture with Spatial Gating Unit (SGU). The model consists of a stack of $L$ blocks with identical structure and size. All projection operations are linear and “ $\odot$ ” refers to element-wise multiplication (linear gating). The input and output protocols follow BERT for NLP and ViT for vision. Unlike Transformers, gMLPs do not require positional encodings, nor is it necessary to mask out the paddings during NLP finetuning. + +def spatial_gating_unit $\mathbf { \Psi } ( \mathbf { x } )$ : u, $\tt { v } =$ split(x, axis="channel") $\tt { v } =$ norm(v, axis="channel") n = get_dim(v, axis $\mathrel { \mathop : } = \mathrel { \mathop : }$ spatial") v = proj(v, n, axis="spatial", init_bias $\mathrel { \mathop : } = 1$ ) return u ∗ v + +both pretraining and finetuning metrics for gMLPs improve as quickly as for Transformers. This is remarkable because it indicates ${ \mathrm { g M L P s } }$ scale just as well as Transformers despite the absence of self-attention, and any performance gap can always be offset by training a larger model with increased data and compute. With a standard 256-batch size $\times \ 1 \mathbf { M }$ -step training setup as in original BERT, a large $\mathrm { g M L P }$ model achieves $8 7 . 7 \%$ accuracy on MNLI and $8 2 . 1 \%$ F1 on SQuAD v2.0. Note, these are better than the $\mathbf { B E R T _ { l a r g e } }$ results reported in Devlin et al. [2] obtained using Transformers. + +For BERT’s finetuning, Transformers can be more practically advantageous over gMLPs on tasks that require cross-sentence alignment (e.g., by $0 . 8 \%$ on MNLI-m in the 300M-param regime), even with similar pretraining perplexity. This problem can be addressed by making gMLPs substantially larger— $3 \times$ as large as Transformers. A more practical solution is to blend in only a tiny bit of selfattention—a single-head self-attention with size up to 128 is sufficient to make $\mathrm { g M L P s }$ outperform Transformers on all NLP tasks we evaluated with even better parameter efficiency. The improvement is sometimes very significant (e.g., $+ 4 . 4 \%$ on SQuAD $\mathrm { v } 2 . 0$ over $\mathbf { B E R T _ { l a r g e } }$ ). + +Overall, the surprising effectiveness of gMLPs in both vision and NLP domains suggests that selfattention is not a necessary ingredient for scaling up machine learning models, although it can be a useful addition depending on the task. With increased data and compute, models with simpler spatial interaction mechanisms such as $\mathrm { g M L P }$ can be as powerful as Transformers and the capacity allocated to self-attention can be either removed or substantially reduced. + +# 2 Model + +Our model, gMLP, consists of a stack of $L$ blocks with identical size and structure. Let $\ b { X } \in \mathbb { R } ^ { n \times d }$ be the token representations with sequence length $n$ and dimension $d$ . Each block is defined as: + +$$ +Z = \sigma ( X U ) , \qquad \tilde { Z } = s ( Z ) , \qquad Y = \tilde { Z } V +$$ + +where $\sigma$ is an activation function such as GeLU [24]. $U$ and $V$ define linear projections along the channel dimension—the same as those in the FFNs of Transformers (e.g., their shapes are $7 6 8 \times 3 0 7 2$ and $3 0 7 2 \times 7 6 8$ for $\mathbf { B E R T _ { b a s e } }$ ). Shortcuts, normalizations and biases are omitted for brevity. + +A key ingredient in the aforementioned formulation is $s ( \cdot )$ , a layer which captures spatial interactions (see below). When $s$ is an identity mapping, the above transformation degenerates to a regular FFN, where individual tokens are processed independently without any cross-token communication. One of our major focuses is therefore to design a good $s$ capable of capturing complex spatial interactions across tokens. The overall block layout is inspired by inverted bottlenecks [25] which define $s ( \cdot )$ as a spatial depthwise convolution. Note, unlike Transformers, our model does not require position embeddings because such information will be captured in $s ( \cdot )$ . + +Our model uses exactly the same input and output protocols as BERT (for NLP) and ViT (for vision). For example, when finetuning on language tasks, we concatenate together multiple text segments followed by paddings, and the predictions are deduced from the last-layer representation of a reserved symbol. Although many of these protocols were introduced for Transformers and hence can be suboptimal for gMLPs, strictly following them helps avoid confounding factors in our experiments and makes our layers more compatible with existing Transformer implementations. + +# 2.1 Spatial Gating Unit + +To enable cross-token interactions, it is necessary for the layer $s ( \cdot )$ to contain a contraction operation over the spatial dimension. The simplistic option would be a linear projection: + +$$ +f _ { W , b } ( Z ) = W Z + b +$$ + +where $W \in \mathbb { R } ^ { n \times n }$ is a matrix for which the size is the same as the sequence length, $n$ , and $b$ refers token-specific biases. For example, if the padded input sequence has 128 tokens, the shape for $W$ will be $1 2 8 \times 1 2 8$ . Unlike self-attention where $W ( Z )$ is dynamically generated from $Z$ , the spatial projection matrix $W$ here in Equation (2) is independent from the input representations. + +In this work, we formulate layer $s ( \cdot )$ as the output of linear gating: + +$$ +s ( Z ) = Z \odot f _ { W , b } ( Z ) +$$ + +where $\odot$ denotes element-wise multiplication. For training stability, we find it critical to initialize $W$ as near-zero values and $b$ as ones, meaning that $f _ { W , b } ( Z ) \approx { \bf 1 }$ and therefore $s ( Z ) \approx Z$ at the beginning of training. This initialization ensures each gMLP block behaves like a regular FFN at the early stage of training, where each token is processed independently, and only gradually injects spatial information across tokens during the course of learning. + +We further find it effective to split $Z$ into two independent parts $( Z _ { 1 } , Z _ { 2 } )$ along the channel dimension for the gating function and for the multiplicative bypass: + +$$ +s ( Z ) = Z _ { 1 } \odot f _ { W , b } ( Z _ { 2 } ) +$$ + +We also normalize the input to $f _ { W , b }$ which empirically improves stability of large NLP models. This gives us the unit illustrated in Figure 1, which we refer to as the Spatial Gating Unit (SGU) in the rest of the paper. In Table 3, we provide ablation studies to compare SGU with several other variants of $s ( \cdot )$ , showing that it works better and narrows the performance gap with self-attention. + +Connections to Existing Layers. The overall formulation of SGU resembles Gated Linear Units (GLUs) [26, 27, 28] as well as earlier works including Highway Networks [29] and LSTM-RNNs [11]. A key distinction is that our gating is computed based on a projection over the spatial (cross-token) dimension rather than the channel (hidden) dimension. SGU is also related to Squeeze-and-Excite (SE) blocks [30] in terms of element-wise multiplication. However, different from SE blocks, SGU does not contain cross-channel projections at all, nor does it enforce permutation invariance (a key feature for content-based attentive modules) due to its static parameterization for the spatial transformation. The spatial projection in SGU could in theory learn to express superficial depthwise convolutions—unlike typical depthwise convolutions with channel-specific filters, SGU learns only a single transformation shared across channels. Finally, we note SGUs offer an alternative mechanism to capture high-order relationships other than self-attention. Specifically, the output for Equation (3) contains up to 2nd-order interactions (e.g., $z _ { i } z _ { j }$ ) whereas output for self-attention (assuming no nonlinearity) contains up to 3rd-order interactions (e.g., $q _ { i } k _ { j } v _ { k } )$ . In terms of computation cost, SGU has $n ^ { 2 } e / 2$ multiply-adds which is comparable to the $2 n ^ { 2 } d$ of dot-product self-attention.1 Both are linear over the input channel size and quadratic over the sequence length $n$ . + +# 3 Image Classification + +Here we examine $\mathrm { g M L P }$ in the vision domain by applying it to the image classification task on ImageNet [31] without using extra data. We compare our MLP-like models with recent attentive models based on vanilla Transformers, including Vision Transformer (ViT) [7], DeiT [8] (ViT with improved regularization), and several other representative convolutional networks. + +Table 1 summarizes the configurations of our gMLP image classification models. The input and output protocols follow ViT/B16 where the raw image is converted into $1 6 \times 1 6$ patches at the stem. The depth and width are chosen so that the models are comparable with ViT/DeiT in capacity. Like Transformers, we find gMLPs tend to drastically overfit the training data. We therefore apply a similar regularization recipe as the one used in DeiT.2 To avoid extensive tuning, we adjust only the strengths of stochastic depth [32] as we move from smaller to larger models in Table 1. All the other hyperparameters remain shared across our three models. See Appendix A.1 for details. + +Table 1: Architecture specifications of $\mathrm { g M L P }$ models for vision. + +
#LdmodeldffnParams (M)FLOPs (B)Survival Prob
gMLP-Ti301287685.92.71.00
gMLP-S30256153619.58.90.95
gMLP-B30512307273.431.60.80
+ +Our ImageNet results are summarized in Table 1 and Figure 2.3 It is interesting to see that gMLPs are comparable with DeiT [8], namely ViT [7] trained using improved regularization. The results suggest that models without self-attention can be as data-efficient as Transformers for image classification. In fact, when the models are properly regularized, their accuracies seem better correlated with capacity instead of the presence of self-attention. Moreover, the accuracy-parameter/FLOPs tradeoff of gMLPs surpasses all concurrently proposed MLP-like architectures [20, 21, 22], which we attribute to the effectiveness of our Spatial Gating Unit (see Table 3 in the next section for an ablation). We also note while ${ \mathrm { g M L P s } }$ are competitive with vanilla Transformers, their performance is behind the best existing ConvNet models (e.g., [33, 34]) or hybrid models (e.g., [35, 36, 37, 38, 10]). + +Table 2: ImageNet-1K results without extra data. + +
ModelImageNet Top-1 (%)*Input ResolutionParams (M)MAdds (B)
ConvNets
ResNet-152 [16]78.32246011.3
RegNetY-8GF[39]81.7224398.0
EfficientNet-B0 [17]77.122450.39
EfficientNet-B3[17]81.6300121.8
EfficientNet-B7 [17]84.36006637.0
NFNet-F0 [33]83.61927212.4
Transformers
ViT-B/16 [7]77.93848655.4
ViT-L/16 [7]76.5384307190.7
DeiT-Ti [8] (ViT+reg)72.222451.3
DeiT-S [8](ViT+reg)79.8224224.6
DeiT-B [8] (ViT+reg)81.82248617.5
MLP-like†
Mixer-B/16 [20]76.42245912.7
Mixer-B/16 (our setup)77.32245912.7
Mixer-L/16 [20]71.822420744.8
ResMLP-12 [22]76.6224153.0
ResMLP-24 [22]79.4224306.0
ResMLP-36[22]79.7224458.9
gMLP-Ti (ours)72.322461.4
gMLP-S (ours)79.6224204.5
gMLP-B (ours)81.62247315.8
+ +\* Standard deviation across multiple independent runs is around 0.1. † Tokenization & embedding process at the stem can be viewed as a convolution. + +Figure 3 visualizes the spatial projection matrices in gMLP-B. Remarkably, the spatial weights after learning exhibit both locality and spatial invariance. In other words, each spatial projection matrix effectively learns to perform convolution with a data-driven, irregular (non-square) kernel shape. + +![](images/4ce114a6aa7138cbd566c0e0b5ee0dbdefe6fbfc6e3bb7bd0ddb7399d3c6d83b.jpg) +Figure 2: ImageNet accuracy vs model capacity. + +![](images/e9c6f27dad6f981df32081593ce223c652c9c19a1ac8f3a7e6353aba3fdced9b.jpg) +Figure 3: Spatial projection weights in gMLPB. Each row shows the filters (reshaped into 2D) for a selected set of tokens in the same layer. + +# 4 Masked Language Modeling with BERT + +Here we conduct empirical studies over the masked language modeling (MLM) task. The input/output protocol for both pretraining and finetuning follows BERT [2]. Different from Transformer-based models, we do not use positional encodings. We also find it unnecessary to mask out tokens in gMLP blocks during finetuning as the model can quickly learn to ignore them. For ablations and case studies, all models are trained with batch size 2048, max length 128 for 125K steps over the RealNews-like subset of C4 [5]. For main results, models are trained with batch size 256, max length 512 for 1M steps over the full English C4 dataset. See Appendix A.2 for details. + +Our preliminary MLM experiments show that gMLPs always learn Toeplitz-like matrices as the spatial weights (Appendix C). This means ${ \mathrm { g M L P s } }$ are able to learn the notion of shift invariance from data, a property naturally implied by the MLM task where any offset of the input sequence does not affect the slot filling outcome. In this case, the learned $f _ { W , b } ( \cdot )$ acts like a 1-d convolution whose kernel size equals the entire sequence length (unlike depthwise convolution with channel-specific filters, here the same $W$ is shared across channels). In the following MLM experiments, we restrict $W$ to be a Toeplitz matrix to avoid redundant model parameterization (since $W$ will be Toeplitz-like regardless after learning). Note this constraint is empirically quality-neutral. + +# 4.1 Ablation: The Importance of Gating in gMLP for BERT’s Pretraining + +In Table 3 below, we establish baselines for our ablation studies. These include: + +1. BERT with a Transformer architecture and learnable absolute position embeddings. +2. BERT with a Transformer architecture and T5-style learnable relative position biases [5]. The biases are both layer- and head-specific as we find this yields the best results. +3. Same as above, but we remove all content-dependent terms inside the softmax and only retain the relative positional biases. This baseline is a straightforward variant of Transformers without self-attention, which can also be viewed as a Random Synthesizer [40]. +4. MLP-Mixer [20] which replaces the multi-head self-attention module in Transformers with a two-layer spatial MLP. This model was developed for image classification and here we investigate it on MLM tasks using the same training setup with BERT and gMLP. + +We compare these baselines against several versions of ${ \mathrm { g M L P s } }$ with similar sizes in Table 3. Note that Multiplicative, Split (last row) is the Spatial Gating Unit we describe in the method section and use in the rest of the paper. First, SGU outperforms other variants in perplexity. Secondly and remarkably, gMLP with SGU also achieves perplexity comparable to Transformer. Note the difference between the strongest baseline (perplexity ${ = } 4 . 2 6 $ ) and ours (perplexity ${ = } 4 . 3 5$ ) is insignificant relative to the perplexity change when the models are scaled (see Table 4 in the next section). Spatial projection weights learned by ${ \mathrm { g M L P s } }$ are visualized in Figure 4. + +Table 3: MLM validation perplexities of Transformer baselines and four versions of ${ \mathrm { g M L P s } }$ . $f$ refers to the spatial linear projection in Equation (2) with input normalization. The MLP-Mixer baseline model has $_ { \mathrm { L } = 2 4 }$ layers with $d _ { \mathrm { m o d e l } } = 7 6 8$ , $d _ { \mathrm { s p a t i a l } } { = } 3 8 4$ and $d _ { \mathrm { f f n } } { = } 3 0 7 2$ . Each gMLP model has $_ { \mathrm { L } = 3 6 }$ layers with $d _ { \mathrm { m o d e l } } = 5 1 2$ and $d _ { \mathrm { f f n } } = 3 0 7 2$ . No positional encodings are used for Mixer or gMLPs. + +
ModelPerplexity*Params (M)
BERTbase4.37110
BERTbase + rel pos4.26110
BERTbase + rel pos - attn5.6496
MLP-Mixer5.34112
Linear gMLP,s(Z)= f(Z)5.1492
Additive gMLP,s(Z)= Z+ f(Z)4.9792
Multiplicative gMLP,s(Z) = Z f(Z)4.5392
Multiplicative,Split gMLP,s(Z) = Z1 f(Z2),Z= Zi||Z24.35102
+ +Standard deviation across multiple independent runs is around 0.01. + +![](images/731fe5bbb1001c5b9f64529e6e9ed073521439ba52c88469a01d1124d0dfaa34.jpg) +Figure 4: Visualization of the spatial filters in $\mathrm { g M L P }$ learned on the MLM task. For each layer in the model we plot the row in $W$ associated with the token in the middle of the sequence. The $\mathbf { X }$ -axis of each subplot has a length of 128 which equal the number of tokens in the sequence. The learned filters appear to be smooth and have several types: forward-looking (e.g., 1st in 2nd row), backward-looking (e.g., 5th in 2nd row) and bi-directional (e.g., 2nd last in the last row). + +# 4.2 Case Study: The Behavior of $\mathbf { g } \mathbf { M L P }$ as Model Size Increases + +In Table 4, we investigate the scaling properties of Transformers and gMLPs in BERT as their model capacity grows. Specifically, we scale the depth of these models by a factor of $\{ 0 . 5 , 1 , 2 , 4 \} \times$ and report the their pretraining MLM perplexities on the validation set as well as finetuning results on the dev sets of two tasks in GLUE [41]. Note each individual Transformer layer is effectively two consecutive blocks: one for self-attention and one for FFN. In the table below we use the notation of $1 2 + 1 2$ to refer to 12 of self-attention blocks plus 12 of FFN blocks in the Transformer baselines. + +Table 4: Pretraining and dev-set finetuning results over increased model capacity. We use the relative positional encoding scheme for Transformers which performs the best in Table 3. + +
Model#LParams (M) PerplexitySST-2 MNLI-m
TransformergMLP6+61867594.915.2590.4 81.591.2 77.7
TransformergMLP12+12361101024.264.3591.3 83.392.3 80.9
TransformergMLP24+24721951873.833.7992.1 85.293.5 82.8
TransformergMLP48+481443653573.473.4392.8 86.395.1 84.6
+ +The results above show that a deep enough gMLP is able to match and even outperform the perplexity of Transformers with comparable capacity.4 In addition, the perplexity-parameter relationships for both architecture families approximately follow a power law (left of Figure 5). This implies the empirical scaling laws originally observed for Transformer-based language models [42] might be broadly applicable across different model families. + +![](images/03ebb35b54fcbc8c3b3a364ab001ed17abb00c96397df6daed10ed232e24238b.jpg) +Figure 5: Scaling properties with respect to perplexity and finetuning accuracies. The figures show that for pretraining, gMLPs are equally good at optimizing perplexity as Transformers. For finetuning, the two model families exhibit comparable scalability despite task-specific offsets. + +Table 4 also leads to an interesting observation that the pretraining perplexities across different model families are not equal in terms of finetuning. While gMLPs outperform Transformers on SST-2, they are worse on MNLI. The results imply that the finetuning performance for NLP tasks is a function of not only the perplexity but also the inductive bias in the architecture. Figure 5 shows that despite the architecture-specific discrepancies between pretraining and finetuning, gMLPs and Transformers exhibit comparable scalability (slope) on both finetuning tasks. This means one can always offset the gap by enlarging the model capacity. In other words, the results indicate that model scalability with respect to downstream metrics can be independent from the presence of self-attention. + +# 4.3 Ablation: The Usefulness of Tiny Attention in BERT’s Finetuning + +So far we have found that self-attention is not a required component to achieve strong MLM perplexity or scalability. At the meantime, we also identified NLP finetuning tasks where gMLPs transfer less well than Transformers (Table 4). The fact that our MLP-like model is advantageous on SST-2 but worse on MNLI is particularly informative—the former is a single-sentence task whereas the latter involves sentence pairs (premise and hypothesis) [43]. We suspect the role of self-attention during finetuning is related to cross-sentence alignment. + +To isolate the effect of self-attention, we experiment with a hybrid model where a tiny self-attention block is attached to the gating function of gMLP (Figure 6). Since gMLP itself is already capable in capturing spatial relationships, we hypothesize that this extra self-attention module does not have to be heavy, and that its presence is more relevant than its capacity. A typical tiny attention module in our experiments has only a single head with size 64, significantly smaller than a typical multi-head self-attention in Transformers with 12 heads and a total size of 768. In the following, we refer to the hybrid model, namely $\mathrm { g M L P }$ with a tiny self-attention, as aMLP (“a” for attention). + +# Pseudo-code for the tiny attention module + +def tiny_attn(x, d_out, d_attn=64): qkv $=$ proj(x, 3 ∗ d_attn, axis="channel") q, k, v $=$ split(qkv, 3, axis="channel") $\kappa =$ einsum("bnd,bmd−>bnm", q, k) a = softmax(w $^ *$ rsqrt(d_attn)) x = einsum("bnm,bmd−>bnd", a, v) return proj(x, d_out, axis="channel") + +![](images/e228c75c14097e90bf3596da75b5d6c0ccbec039247e6155396cb13236ea093c.jpg) +Figure 6: Hybrid spatial gating unit with a tiny self-attention module. We use the normalized input of the $\mathrm { g M L P }$ block (endpoint after the input normalization and right before the channel expansion) as the input to the tiny self-attention. For SGU we have ${ d _ { \mathrm { o u t } } = d _ { \mathrm { f f n } } / 2 }$ due to the channel split. + +In Figure 7, we investigate the transferability of MLM models via the calibration plots between their pretraining perplexities and finetuning metrics. Models evaluated include $\mathbf { B E R T _ { b a s e } }$ , gMLP and its hybrid version aMLP with a 64-d single-head self-attention (Figure 6). The data points were collected by varying the model depth by $\{ 0 . 5 , 1 , 2 \} \times$ or data by $\{ 1 , 2 , 4 , 8 \} \times$ . It can be seen that gMLPs transfer better to SST-2 than Transformers regardless of the presence of self-attention, While gMLP performs worse on MNLI, attaching a tiny bit of self-attention is sufficient to close the gap. In Appendix D we visualize the tiny self-attention modules in aMLP over MNLI examples, showing that they are primarily responsible for the alignment between sentence pairs. + +![](images/263412114ce248c9b9dccbc4dbdcbba36ac2aa648dce820d76c34069690a0cd1.jpg) +Figure 7: Transferability from MLM pretraining perpexity to finetuning accuracies on GLUE. aMLP refers to $\mathrm { g M L P }$ enhanced with a 64-d single-head self-attention, as illustrated in Figure 6. In contrast, each self-attention module in the BERT baseline contains 12 heads with a total size of 768. + +In Figure 8 we put together the scaling properties of the three models, showing that aMLP $\mathrm { ( g M L P + }$ tiny attention) consistently outperforms Transformer on both finetuning tasks. + +![](images/395d749d892cbad8add03840c7c07a237a2c72cef704c6aff4263e86c3de1d09.jpg) +Figure 8: Comparing the scaling properties of Transformers, $\mathrm { g M L P s }$ and aMLPs (with 64-d, singlehead attention). Results were obtained using the same setup in Section 4.2. + +# 4.4 Main Results for MLM in the BERT Setup + +Below we present pretraining and finetuning results in the full BERT setup. Different from ablation and case studies, here we use the full English C4 dataset and adopt a common MLM setup with batch size 256, max length 512 and 1M training steps. For fair comparison, we adjust the depth and width of $\mathrm { g M L P s }$ to ensure comparable model capacity with the Transformer baselines. The model specifications are given in Table 5 and hyperparameters are detailed in Appendix A.2. For finetuning, we report the dev-set performance for SST-2 and MNLI in GLUE [41] and each result entry was obtained by taking the median of five independent runs. In addition, we report finetuning results on SQuAD [44, 45] to test the models’ ability in reasoning over a longer context. + +Results are presented in Table 6. Consistent with our findings earlier in Section 4.1 and Section 4.2, gMLPs are competitive with Transformers in terms of perplexity, especially in the larger scale setup. There are several observations related to the finetuning results: + +First, on finetuning tasks where gMLPs underperform Transformers, the performance gap tends to narrow as the model capacity increases. For example, while $\mathrm { g M L P }$ performs worse by $8 . 5 \%$ on SQuAD-v2.0 in the base scale, the performance gap relative to the baseline decreases to $2 . 7 \%$ at the larger scale. Notably, our ${ \mathrm { g M L P } } _ { \mathrm { l a r g e } }$ achieves $8 9 . 5 \%$ F1 on SQuAD-v1.1 without any self-attention or dynamic spatial parameterization [28], which is well above the $8 8 . 5 \%$ reported for $\mathbf { B E R T _ { b a s e } }$ in Devlin et al. [2] and is only $1 . 4 \%$ away from the original result for $\mathbf { B E R T _ { l a r g e } }$ . We also include one additional data point by scaling up $\mathrm { g M L P }$ even further. The resulting model, $\mathrm { g } \mathrm { \bar { M } L P _ { \mathrm { x l a r g e } } }$ , outperforms $\mathbf { B E R T _ { l a r g e } }$ on SQuAD- $\mathbf { \nabla } \cdot \mathbf { v } 2 . 0 $ —a difficult task involving question-answer pairs—without any self-attention. While this is not a fair comparison due to different model sizes, it is an existence proof that MLP-like models can be competitive with Transformers on challenging NLP tasks. + +Table 5: Model specifications in the full BERT setup. + +
Params (M)FLOPs (B)#Ldmodeldffn
BERTbase gMLPbase aMLPbase110100.812+127683072
130158.0485123072
109128.9365123072
BERTlarge gMLPlarge aMLPlarge336341.224+2410244096
365430.1967683072
316370.3727683072
gMLPxlarge9411091.314410244096
+ +Table 6: Pretraining perplexities and dev-set results for finetuning. “ours” indicates models trained using our setup. We report accuracies for SST-2 and MNLI, and F1 scores for SQuAD v1.1/2.0. + +
PerplexitySST-2MNLI (m/mm)SQuADAttn SizeParams (M)
v1.1v2.0
BERTbase [2]192.784.4/-88.576.3768 (64 × 12)110
BERTbase (ours)4.1793.885.6/85.790.278.6768 (64 × 12)110
gMLPbase4.2894.283.7/84.186.770.11130
aMLPbase3.9593.485.9/85.890.780.964109
BERTlarge [2]193.786.6/-90.981.81024 (64 × 16)336
BERTlarge (ours)3.3594.387.0/87.492.081.01024 (64 × 16)336
gMLPlarge3.3294.886.2/86.589.578.31365
aMLPlarge3.1994.888.4/88.492.285.4128316
gMLPxlarge2.8995.687.7/87.790.982.11941
+ +Furthermore, we show that blending in a tiny single-head self-attention of size either 64 or 128 is sufficient to make gMLPs outperform Transformers of similar capacity, sometimes by a significant margin. For example, our hybrid model $\mathrm { a M L P _ { l a r g e } }$ achieves $4 . 4 \%$ higher F1 than Transformers on SQuAD-v2.0. The results suggest that the capacity in the multi-head self-attention of Transformers can be largely redundant, and that the majority of its functionalities can be captured by the spatial gating unit in gMLPs. The results also imply that the inductive biases in the spatial gating unit of gMLPs and the tiny attention are complementary to each other. While the benefits of architectural inductive bias may vanish over increased compute, tiny attention does improve the practical value of gMLPs in the regime that we investigate in this work. + +# 5 Conclusion + +Since the seminal work of Vaswani et al. [1], Transformers have been widely adopted across NLP and computer vision. This adoption has enabled many impressive results especially in NLP. To date, it is still unclear what empowers such success: is it the feedforward nature of Transformers or is it the multi-head self-attention layers in Transformers? + +Our work suggests a simpler alternative to the multi-head self-attention layers in Transformers. We show that gMLPs, a simple variant of MLPs with gating, can be competitive with Transformers in terms of BERT’s pretraining perplexity and ViT’s accuracy. gMLPs are also comparable with Transformers in terms of the scalability over increased data and compute. As for BERT finetuning, we find gMLPs can achieve appealing results on challenging tasks such as SQuAD without self-attention, and can significantly outperform Transformers in certain cases. We also find the inductive bias in Transformer’s multi-head self-attention useful on downstream tasks that require cross-sentence alignment. However in those cases, making gMLP substantially larger closes the gap with Transformers. More practically, blending a small single-head self-attention into gMLP allows for an even better architecture without the need for increasing model size. + +# Acknowledgements + +We thank Gabriel Bender, Neil Houlsby, Thang Luong, Niki Parmar, Hieu Pham, Jascha SohlDickstein, Noam Shazeer, Ilya Sutskever, Jakob Uszkoreit and Ashish Vaswani for their feedback. + +# References + +[1] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in Neural Information Processing Systems, 2017. +[2] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. In NAACL, 2018. +[3] Zhilin Yang, Zihang Dai, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, and Quoc V Le. Xlnet: Generalized autoregressive pretraining for language understanding. In Advances in Neural Information Processing Systems, 2019. +[4] Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692, 2019. [5] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer. JMLR, 2020. +[6] Tom B Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. In Advances in Neural Information Processing Systems, 2020. +[7] Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale. In ICLR, 2021. [8] Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Hervé Jégou. Training data-efficient image transformers & distillation through attention. arXiv preprint arXiv:2012.12877, 2020. +[9] Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to-end object detection with transformers. In ECCV, 2020. +[10] Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. arXiv preprint arXiv:2103.14030, 2021. +[11] Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory. Neural Computation, 1997. +[12] Y. LeCun, B. Boser, J. S. Denker, D. Henderson, R. E. Howard, W. Hubbard, and L. D. Jackel. Backpropagation applied to handwritten zip code recognition. Neural Computation, 1989. +[13] Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. Advances in Neural Information Processing Systems, 2012. +[14] Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale image recognition. In ICLR, 2015. +[15] Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, and Andrew Rabinovich. Going deeper with convolutions. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1–9, 2015. +[16] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In CVPR, 2016. +[17] Mingxing Tan and Quoc Le. Efficientnet: Rethinking model scaling for convolutional neural networks. In ICML, 2019. +[18] Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. Neural machine translation by jointly learning to align and translate. In ICLR, 2015. +[19] Kurt Hornik, Maxwell Stinchcombe, and Halbert White. Multilayer feedforward networks are universal approximators. Neural networks, 2(5):359–366, 1989. +[20] Ilya Tolstikhin, Neil Houlsby, Alexander Kolesnikov, Lucas Beyer, Xiaohua Zhai, Thomas Unterthiner, Jessica Yung, Daniel Keysers, Jakob Uszkoreit, Mario Lucic, and Alexey Dosovitskiy. Mlp-mixer: An all-mlp architecture for vision. arXiv preprint arXiv:2105.01601, 2021. +[21] Luke Melas-Kyriazi. Do you even need attention? a stack of feed-forward layers does surprisingly well on imagenet. arXiv preprint arXiv:2105.02723, 2021. +[22] Hugo Touvron, Piotr Bojanowski, Mathilde Caron, Matthieu Cord, Alaaeldin El-Nouby, Edouard Grave, Armand Joulin, Gabriel Synnaeve, Jakob Verbeek, and Hervé Jégou. Resmlp: Feedforward networks for image classification with data-efficient training. arXiv preprint arXiv:2105.03404, 2021. +[23] Xiaohan Ding, Xiangyu Zhang, Jungong Han, and Guiguang Ding. Repmlp: Reparameterizing convolutions into fully-connected layers for image recognition. arXiv preprint arXiv:2105.01883, 2021. +[24] Dan Hendrycks and Kevin Gimpel. Gaussian error linear units (gelus). arXiv preprint arXiv:1606.08415, 2016. +[25] Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen. Mobilenetv2: Inverted residuals and linear bottlenecks. In CVPR, 2018. +[26] Yann N Dauphin, Angela Fan, Michael Auli, and David Grangier. Language modeling with gated convolutional networks. In ICML, 2017. +[27] Noam Shazeer. Glu variants improve transformer. arXiv preprint arXiv:2002.05202, 2020. +[28] Felix Wu, Angela Fan, Alexei Baevski, Yann N Dauphin, and Michael Auli. Pay less attention with lightweight and dynamic convolutions. In ICLR, 2019. +[29] Rupesh Kumar Srivastava, Klaus Greff, and Jürgen Schmidhuber. Highway networks. arXiv preprint arXiv:1505.00387, 2015. +[30] Jie Hu, Li Shen, and Gang Sun. Squeeze-and-excitation networks. In CVPR, 2018. +[31] Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A largescale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009. +[32] Gao Huang, Yu Sun, Zhuang Liu, Daniel Sedra, and Kilian Q Weinberger. Deep networks with stochastic depth. In ECCV, 2016. +[33] Andrew Brock, Soham De, Samuel L Smith, and Karen Simonyan. High-performance largescale image recognition without normalization. arXiv preprint arXiv:2102.06171, 2021. +[34] Mingxing Tan and Quoc V Le. Efficientnetv2: Smaller models and faster training. In ICML, 2021. +[35] Irwan Bello. Lambdanetworks: Modeling long-range interactions without attention. In ICLR, 2021. +[36] Ashish Vaswani, Prajit Ramachandran, Aravind Srinivas, Niki Parmar, Blake Hechtman, and Jonathon Shlens. Scaling local self-attention for parameter efficient visual backbones. In CVPR, 2021. +[37] Aravind Srinivas, Tsung-Yi Lin, Niki Parmar, Jonathon Shlens, Pieter Abbeel, and Ashish Vaswani. Bottleneck transformers for visual recognition. arXiv preprint arXiv:2101.11605, 2021. +[38] Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, and Lei Zhang. Cvt: Introducing convolutions to vision transformers. arXiv preprint arXiv:2103.15808, 2021. +[39] Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, and Piotr Dollár. Designing network design spaces. In CVPR, 2020. +[40] Yi Tay, Dara Bahri, Donald Metzler, Da-Cheng Juan, Zhe Zhao, and Che Zheng. Synthesizer: Rethinking self-attention in transformer models. arXiv preprint arXiv:2005.00743, 2020. +[41] Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel R Bowman. Glue: A multi-task benchmark and analysis platform for natural language understanding. In ICLR, 2019. +[42] Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling laws for neural language models. arXiv preprint arXiv:2001.08361, 2020. +[43] Adina Williams, Nikita Nangia, and Samuel R Bowman. A broad-coverage challenge corpus for sentence understanding through inference. In NAACL, 2017. +[44] Pranav Rajpurkar, Jian Zhang, Konstantin Lopyrev, and Percy Liang. Squad: $^ { 1 0 0 , 0 0 0 + }$ questions for machine comprehension of text. In EMNLP, 2016. +[45] Pranav Rajpurkar, Robin Jia, and Percy Liang. Know what you don’t know: Unanswerable questions for squad. In ACL, 2018. +[46] Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In European conference on computer vision, pages 740–755. Springer, 2014. +[47] Mingxing Tan, Ruoming Pang, and Quoc V Le. Efficientdet: Scalable and efficient object detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10781–10790, 2020. +[48] Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, and Ling Shao. Pyramid vision transformer: A versatile backbone for dense prediction without convolutions. arXiv preprint arXiv:2102.12122, 2021. +[49] Josh Beal, Eric Kim, Eric Tzeng, Dong Huk Park, Andrew Zhai, and Dmitry Kislyuk. Toward transformer-based object detection. arXiv preprint arXiv:2012.09958, 2020. \ No newline at end of file diff --git a/parse/train/KBnXrODoBW/KBnXrODoBW_content_list.json b/parse/train/KBnXrODoBW/KBnXrODoBW_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..555ea3106704e7d1dc52ca779eecc832d7001ffb --- /dev/null +++ b/parse/train/KBnXrODoBW/KBnXrODoBW_content_list.json @@ -0,0 +1,1070 @@ +[ + { + "type": "text", + "text": "Pay Attention to MLPs ", + "text_level": 1, + "bbox": [ + 359, + 122, + 640, + 147 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Hanxiao Liu, Zihang Dai, David R. So, Quoc V. Le Google Research, Brain Team {hanxiaol,zihangd,davidso,qvl}@google.com ", + "bbox": [ + 323, + 200, + 676, + 243 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Abstract ", + "text_level": 1, + "bbox": [ + 462, + 279, + 535, + 295 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Transformers [1] have become one of the most important architectural innovations in deep learning and have enabled many breakthroughs over the past few years. Here we propose a simple network architecture, gMLP, based on MLPs with gating, and show that it can perform as well as Transformers in key language and vision applications. Our comparisons show that self-attention is not critical for Vision Transformers, as $\\mathrm { g M L P }$ can achieve the same accuracy. For BERT, our model achieves parity with Transformers on pretraining perplexity and is better on some downstream NLP tasks. On finetuning tasks where gMLP performs worse, making the $\\mathrm { g M L P }$ model substantially larger can close the gap with Transformers. In general, our experiments show that gMLP can scale as well as Transformers over increased data and compute. ", + "bbox": [ + 233, + 309, + 766, + 460 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 Introduction ", + "text_level": 1, + "bbox": [ + 174, + 484, + 310, + 502 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Transformers [1] have enabled many breakthroughs in natural language processing (e.g., [2, 3, 4, 5, 6]) and have been shown to work well for computer vision (e.g., [7, 8, 9, 10]). Thanks to this success, Transformers have largely replaced LSTM-RNN [11] as the default architecture in NLP, and have become an appealing alternative to ConvNets [12, 13, 14, 15, 16, 17] in computer vision. ", + "bbox": [ + 174, + 516, + 825, + 571 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "The Transformer architecture combines two important concepts: (1) a recurrent-free architecture which computes the representations for each individual token in parallel, and (2) multi-head selfattention blocks which aggregate spatial information across tokens. On one hand, the attention mechanism [18] introduces the inductive bias that the spatial interactions should be dynamically parameterized based on the input representations. On the other hand, it is known that MLPs with static parameterization can represent arbitrary functions [19]. It therefore remains an open question whether the inductive bias in self-attention is essential to the remarkable effectiveness of Transformers. ", + "bbox": [ + 174, + 578, + 825, + 675 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Here we study the necessity of self-attention modules in key language and vision applications of Transformers. Specifically, we propose an MLP-based alternative to Transformers without self-attention, which simply consists of channel projections and spatial projections with static parameterization. We experiment with several design choices for this architecture and find spatial projections work well when they are linear and paired with multiplicative gating (Figure 1). We name the model gMLP because it is built out of basic MLP layers with gating. ", + "bbox": [ + 174, + 681, + 825, + 765 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "We apply $\\mathrm { g M L P }$ to image classification and obtain strong results on ImageNet. $\\mathrm { g M L P }$ achieves comparable performance with DeiT [8], namely Vision Transformer (ViT) [7] with improved regularization, in a similar training setup. With $66 \\%$ less parameters, a gMLP model is $3 \\%$ more accurate than MLP-Mixer [20]. Together with Tolstikhin et al. [20], Melas-Kyriazi [21], Touvron et al. [22] and Ding et. al. [23], our results question the necessity of self-attention layers in Vision Transformers. ", + "bbox": [ + 174, + 771, + 825, + 840 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "We apply gMLP to masked language modeling (MLM) in the BERT [2] setup, one of the most wellestablished applications of Transformers, and find that it is as good as Transformers at minimizing perplexity during pretraining. Our experiments indicate that perplexity is only correlated with model capacity and is insensitive to the presence of self-attention. As capacity increases, we observe that ", + "bbox": [ + 174, + 847, + 825, + 902 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "def gmlp_block(x, d_model, d_ffn): shortcut $=$ x $\\mathrm { ~ x ~ } =$ norm(x, axis=\"channel\") $\\mathrm { ~ x ~ } =$ proj(x, d_ffn, axis=\"channel\") $\\mathrm { ~ x ~ } =$ gelu(x) $\\mathrm { ~ x ~ } =$ spatial_gating_unit(x) $\\mathrm { ~ x ~ } =$ proj(x, d_model, axis $= ^ { 1 1 }$ channel\") return $\\texttt { x + }$ shortcut ", + "bbox": [ + 508, + 118, + 753, + 203 + ], + "page_idx": 1 + }, + { + "type": "image", + "img_path": "images/018d82d7640366b577f21adb6e4e6c60f398a8eb4976b569ff0a2114a750edde.jpg", + "image_caption": [ + "Figure 1: Overview of the $\\mathrm { g M L P }$ architecture with Spatial Gating Unit (SGU). The model consists of a stack of $L$ blocks with identical structure and size. All projection operations are linear and “ $\\odot$ ” refers to element-wise multiplication (linear gating). The input and output protocols follow BERT for NLP and ViT for vision. Unlike Transformers, gMLPs do not require positional encodings, nor is it necessary to mask out the paddings during NLP finetuning. " + ], + "image_footnote": [], + "bbox": [ + 184, + 97, + 486, + 284 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "def spatial_gating_unit $\\mathbf { \\Psi } ( \\mathbf { x } )$ : u, $\\tt { v } =$ split(x, axis=\"channel\") $\\tt { v } =$ norm(v, axis=\"channel\") n = get_dim(v, axis $\\mathrel { \\mathop : } = \\mathrel { \\mathop : }$ spatial\") v = proj(v, n, axis=\"spatial\", init_bias $\\mathrel { \\mathop : } = 1$ ) return u ∗ v ", + "bbox": [ + 511, + 214, + 805, + 280 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "both pretraining and finetuning metrics for gMLPs improve as quickly as for Transformers. This is remarkable because it indicates ${ \\mathrm { g M L P s } }$ scale just as well as Transformers despite the absence of self-attention, and any performance gap can always be offset by training a larger model with increased data and compute. With a standard 256-batch size $\\times \\ 1 \\mathbf { M }$ -step training setup as in original BERT, a large $\\mathrm { g M L P }$ model achieves $8 7 . 7 \\%$ accuracy on MNLI and $8 2 . 1 \\%$ F1 on SQuAD v2.0. Note, these are better than the $\\mathbf { B E R T _ { l a r g e } }$ results reported in Devlin et al. [2] obtained using Transformers. ", + "bbox": [ + 173, + 392, + 825, + 478 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "For BERT’s finetuning, Transformers can be more practically advantageous over gMLPs on tasks that require cross-sentence alignment (e.g., by $0 . 8 \\%$ on MNLI-m in the 300M-param regime), even with similar pretraining perplexity. This problem can be addressed by making gMLPs substantially larger— $3 \\times$ as large as Transformers. A more practical solution is to blend in only a tiny bit of selfattention—a single-head self-attention with size up to 128 is sufficient to make $\\mathrm { g M L P s }$ outperform Transformers on all NLP tasks we evaluated with even better parameter efficiency. The improvement is sometimes very significant (e.g., $+ 4 . 4 \\%$ on SQuAD $\\mathrm { v } 2 . 0$ over $\\mathbf { B E R T _ { l a r g e } }$ ). ", + "bbox": [ + 173, + 483, + 825, + 582 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Overall, the surprising effectiveness of gMLPs in both vision and NLP domains suggests that selfattention is not a necessary ingredient for scaling up machine learning models, although it can be a useful addition depending on the task. With increased data and compute, models with simpler spatial interaction mechanisms such as $\\mathrm { g M L P }$ can be as powerful as Transformers and the capacity allocated to self-attention can be either removed or substantially reduced. ", + "bbox": [ + 174, + 585, + 825, + 656 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2 Model ", + "text_level": 1, + "bbox": [ + 174, + 674, + 259, + 691 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Our model, gMLP, consists of a stack of $L$ blocks with identical size and structure. Let $\\ b { X } \\in \\mathbb { R } ^ { n \\times d }$ be the token representations with sequence length $n$ and dimension $d$ . Each block is defined as: ", + "bbox": [ + 173, + 705, + 823, + 734 + ], + "page_idx": 1 + }, + { + "type": "equation", + "img_path": "images/c7f7fd02ef54ad58deab3b63dffb850ea35636c7e3bb6788f9e24746d918a055.jpg", + "text": "$$\nZ = \\sigma ( X U ) , \\qquad \\tilde { Z } = s ( Z ) , \\qquad Y = \\tilde { Z } V\n$$", + "text_format": "latex", + "bbox": [ + 351, + 741, + 647, + 760 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "where $\\sigma$ is an activation function such as GeLU [24]. $U$ and $V$ define linear projections along the channel dimension—the same as those in the FFNs of Transformers (e.g., their shapes are $7 6 8 \\times 3 0 7 2$ and $3 0 7 2 \\times 7 6 8$ for $\\mathbf { B E R T _ { b a s e } }$ ). Shortcuts, normalizations and biases are omitted for brevity. ", + "bbox": [ + 173, + 765, + 825, + 809 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "A key ingredient in the aforementioned formulation is $s ( \\cdot )$ , a layer which captures spatial interactions (see below). When $s$ is an identity mapping, the above transformation degenerates to a regular FFN, where individual tokens are processed independently without any cross-token communication. One of our major focuses is therefore to design a good $s$ capable of capturing complex spatial interactions across tokens. The overall block layout is inspired by inverted bottlenecks [25] which define $s ( \\cdot )$ as a spatial depthwise convolution. Note, unlike Transformers, our model does not require position embeddings because such information will be captured in $s ( \\cdot )$ . ", + "bbox": [ + 173, + 814, + 825, + 912 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Our model uses exactly the same input and output protocols as BERT (for NLP) and ViT (for vision). For example, when finetuning on language tasks, we concatenate together multiple text segments followed by paddings, and the predictions are deduced from the last-layer representation of a reserved symbol. Although many of these protocols were introduced for Transformers and hence can be suboptimal for gMLPs, strictly following them helps avoid confounding factors in our experiments and makes our layers more compatible with existing Transformer implementations. ", + "bbox": [ + 173, + 90, + 825, + 175 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "2.1 Spatial Gating Unit ", + "text_level": 1, + "bbox": [ + 174, + 190, + 349, + 205 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "To enable cross-token interactions, it is necessary for the layer $s ( \\cdot )$ to contain a contraction operation over the spatial dimension. The simplistic option would be a linear projection: ", + "bbox": [ + 173, + 215, + 823, + 244 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/7e5c4dae68c79d1010c2b7fd3237f11b35cf0b787c253c05fa4e0267d371f418.jpg", + "text": "$$\nf _ { W , b } ( Z ) = W Z + b\n$$", + "text_format": "latex", + "bbox": [ + 429, + 251, + 568, + 268 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where $W \\in \\mathbb { R } ^ { n \\times n }$ is a matrix for which the size is the same as the sequence length, $n$ , and $b$ refers token-specific biases. For example, if the padded input sequence has 128 tokens, the shape for $W$ will be $1 2 8 \\times 1 2 8$ . Unlike self-attention where $W ( Z )$ is dynamically generated from $Z$ , the spatial projection matrix $W$ here in Equation (2) is independent from the input representations. ", + "bbox": [ + 173, + 275, + 825, + 332 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "In this work, we formulate layer $s ( \\cdot )$ as the output of linear gating: ", + "bbox": [ + 173, + 337, + 611, + 352 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/b084e72db12a7f0a08d1156d8e0af26ab807cbf9481d25ba50788a491cfa7ca0.jpg", + "text": "$$\ns ( Z ) = Z \\odot f _ { W , b } ( Z )\n$$", + "text_format": "latex", + "bbox": [ + 426, + 358, + 571, + 376 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where $\\odot$ denotes element-wise multiplication. For training stability, we find it critical to initialize $W$ as near-zero values and $b$ as ones, meaning that $f _ { W , b } ( Z ) \\approx { \\bf 1 }$ and therefore $s ( Z ) \\approx Z$ at the beginning of training. This initialization ensures each gMLP block behaves like a regular FFN at the early stage of training, where each token is processed independently, and only gradually injects spatial information across tokens during the course of learning. ", + "bbox": [ + 173, + 381, + 825, + 452 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "We further find it effective to split $Z$ into two independent parts $( Z _ { 1 } , Z _ { 2 } )$ along the channel dimension for the gating function and for the multiplicative bypass: ", + "bbox": [ + 174, + 457, + 820, + 486 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/f6a604c6e6d1a2b5953c15b5a23e51838a39cd7db61dd6c11719cc258562c35a.jpg", + "text": "$$\ns ( Z ) = Z _ { 1 } \\odot f _ { W , b } ( Z _ { 2 } )\n$$", + "text_format": "latex", + "bbox": [ + 419, + 492, + 578, + 511 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "We also normalize the input to $f _ { W , b }$ which empirically improves stability of large NLP models. This gives us the unit illustrated in Figure 1, which we refer to as the Spatial Gating Unit (SGU) in the rest of the paper. In Table 3, we provide ablation studies to compare SGU with several other variants of $s ( \\cdot )$ , showing that it works better and narrows the performance gap with self-attention. ", + "bbox": [ + 173, + 516, + 825, + 571 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Connections to Existing Layers. The overall formulation of SGU resembles Gated Linear Units (GLUs) [26, 27, 28] as well as earlier works including Highway Networks [29] and LSTM-RNNs [11]. A key distinction is that our gating is computed based on a projection over the spatial (cross-token) dimension rather than the channel (hidden) dimension. SGU is also related to Squeeze-and-Excite (SE) blocks [30] in terms of element-wise multiplication. However, different from SE blocks, SGU does not contain cross-channel projections at all, nor does it enforce permutation invariance (a key feature for content-based attentive modules) due to its static parameterization for the spatial transformation. The spatial projection in SGU could in theory learn to express superficial depthwise convolutions—unlike typical depthwise convolutions with channel-specific filters, SGU learns only a single transformation shared across channels. Finally, we note SGUs offer an alternative mechanism to capture high-order relationships other than self-attention. Specifically, the output for Equation (3) contains up to 2nd-order interactions (e.g., $z _ { i } z _ { j }$ ) whereas output for self-attention (assuming no nonlinearity) contains up to 3rd-order interactions (e.g., $q _ { i } k _ { j } v _ { k } )$ . In terms of computation cost, SGU has $n ^ { 2 } e / 2$ multiply-adds which is comparable to the $2 n ^ { 2 } d$ of dot-product self-attention.1 Both are linear over the input channel size and quadratic over the sequence length $n$ . ", + "bbox": [ + 173, + 587, + 825, + 796 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3 Image Classification ", + "text_level": 1, + "bbox": [ + 174, + 814, + 374, + 832 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Here we examine $\\mathrm { g M L P }$ in the vision domain by applying it to the image classification task on ImageNet [31] without using extra data. We compare our MLP-like models with recent attentive models based on vanilla Transformers, including Vision Transformer (ViT) [7], DeiT [8] (ViT with improved regularization), and several other representative convolutional networks. ", + "bbox": [ + 176, + 845, + 823, + 875 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 90, + 823, + 119 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Table 1 summarizes the configurations of our gMLP image classification models. The input and output protocols follow ViT/B16 where the raw image is converted into $1 6 \\times 1 6$ patches at the stem. The depth and width are chosen so that the models are comparable with ViT/DeiT in capacity. Like Transformers, we find gMLPs tend to drastically overfit the training data. We therefore apply a similar regularization recipe as the one used in DeiT.2 To avoid extensive tuning, we adjust only the strengths of stochastic depth [32] as we move from smaller to larger models in Table 1. All the other hyperparameters remain shared across our three models. See Appendix A.1 for details. ", + "bbox": [ + 173, + 126, + 825, + 223 + ], + "page_idx": 3 + }, + { + "type": "table", + "img_path": "images/27f42f69254be8b6e8eeff8fb7a1820daeca371c25a2a48691d8d74cf19caa2f.jpg", + "table_caption": [ + "Table 1: Architecture specifications of $\\mathrm { g M L P }$ models for vision. " + ], + "table_footnote": [], + "table_body": "
#LdmodeldffnParams (M)FLOPs (B)Survival Prob
gMLP-Ti301287685.92.71.00
gMLP-S30256153619.58.90.95
gMLP-B30512307273.431.60.80
", + "bbox": [ + 258, + 256, + 736, + 321 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Our ImageNet results are summarized in Table 1 and Figure 2.3 It is interesting to see that gMLPs are comparable with DeiT [8], namely ViT [7] trained using improved regularization. The results suggest that models without self-attention can be as data-efficient as Transformers for image classification. In fact, when the models are properly regularized, their accuracies seem better correlated with capacity instead of the presence of self-attention. Moreover, the accuracy-parameter/FLOPs tradeoff of gMLPs surpasses all concurrently proposed MLP-like architectures [20, 21, 22], which we attribute to the effectiveness of our Spatial Gating Unit (see Table 3 in the next section for an ablation). We also note while ${ \\mathrm { g M L P s } }$ are competitive with vanilla Transformers, their performance is behind the best existing ConvNet models (e.g., [33, 34]) or hybrid models (e.g., [35, 36, 37, 38, 10]). ", + "bbox": [ + 173, + 330, + 825, + 457 + ], + "page_idx": 3 + }, + { + "type": "table", + "img_path": "images/123f13f92d912e05ece7dd96c75519ea83477124b08ac3939ad58b8b69f2b23d.jpg", + "table_caption": [ + "Table 2: ImageNet-1K results without extra data. " + ], + "table_footnote": [ + "\\* Standard deviation across multiple independent runs is around 0.1. † Tokenization & embedding process at the stem can be viewed as a convolution. " + ], + "table_body": "
ModelImageNet Top-1 (%)*Input ResolutionParams (M)MAdds (B)
ConvNets
ResNet-152 [16]78.32246011.3
RegNetY-8GF[39]81.7224398.0
EfficientNet-B0 [17]77.122450.39
EfficientNet-B3[17]81.6300121.8
EfficientNet-B7 [17]84.36006637.0
NFNet-F0 [33]83.61927212.4
Transformers
ViT-B/16 [7]77.93848655.4
ViT-L/16 [7]76.5384307190.7
DeiT-Ti [8] (ViT+reg)72.222451.3
DeiT-S [8](ViT+reg)79.8224224.6
DeiT-B [8] (ViT+reg)81.82248617.5
MLP-like†
Mixer-B/16 [20]76.42245912.7
Mixer-B/16 (our setup)77.32245912.7
Mixer-L/16 [20]71.822420744.8
ResMLP-12 [22]76.6224153.0
ResMLP-24 [22]79.4224306.0
ResMLP-36[22]79.7224458.9
gMLP-Ti (ours)72.322461.4
gMLP-S (ours)79.6224204.5
gMLP-B (ours)81.62247315.8
", + "bbox": [ + 236, + 488, + 758, + 796 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Figure 3 visualizes the spatial projection matrices in gMLP-B. Remarkably, the spatial weights after learning exhibit both locality and spatial invariance. In other words, each spatial projection matrix effectively learns to perform convolution with a data-driven, irregular (non-square) kernel shape. ", + "bbox": [ + 174, + 833, + 825, + 876 + ], + "page_idx": 3 + }, + { + "type": "image", + "img_path": "images/4ce114a6aa7138cbd566c0e0b5ee0dbdefe6fbfc6e3bb7bd0ddb7399d3c6d83b.jpg", + "image_caption": [ + "Figure 2: ImageNet accuracy vs model capacity. " + ], + "image_footnote": [], + "bbox": [ + 181, + 88, + 496, + 320 + ], + "page_idx": 4 + }, + { + "type": "image", + "img_path": "images/e9c6f27dad6f981df32081593ce223c652c9c19a1ac8f3a7e6353aba3fdced9b.jpg", + "image_caption": [ + "Figure 3: Spatial projection weights in gMLPB. Each row shows the filters (reshaped into 2D) for a selected set of tokens in the same layer. " + ], + "image_footnote": [], + "bbox": [ + 511, + 89, + 821, + 292 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4 Masked Language Modeling with BERT ", + "text_level": 1, + "bbox": [ + 174, + 371, + 542, + 388 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Here we conduct empirical studies over the masked language modeling (MLM) task. The input/output protocol for both pretraining and finetuning follows BERT [2]. Different from Transformer-based models, we do not use positional encodings. We also find it unnecessary to mask out tokens in gMLP blocks during finetuning as the model can quickly learn to ignore them. For ablations and case studies, all models are trained with batch size 2048, max length 128 for 125K steps over the RealNews-like subset of C4 [5]. For main results, models are trained with batch size 256, max length 512 for 1M steps over the full English C4 dataset. See Appendix A.2 for details. ", + "bbox": [ + 173, + 402, + 825, + 501 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Our preliminary MLM experiments show that gMLPs always learn Toeplitz-like matrices as the spatial weights (Appendix C). This means ${ \\mathrm { g M L P s } }$ are able to learn the notion of shift invariance from data, a property naturally implied by the MLM task where any offset of the input sequence does not affect the slot filling outcome. In this case, the learned $f _ { W , b } ( \\cdot )$ acts like a 1-d convolution whose kernel size equals the entire sequence length (unlike depthwise convolution with channel-specific filters, here the same $W$ is shared across channels). In the following MLM experiments, we restrict $W$ to be a Toeplitz matrix to avoid redundant model parameterization (since $W$ will be Toeplitz-like regardless after learning). Note this constraint is empirically quality-neutral. ", + "bbox": [ + 174, + 506, + 825, + 617 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4.1 Ablation: The Importance of Gating in gMLP for BERT’s Pretraining ", + "text_level": 1, + "bbox": [ + 173, + 633, + 699, + 650 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "In Table 3 below, we establish baselines for our ablation studies. These include: ", + "bbox": [ + 173, + 660, + 696, + 675 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "1. BERT with a Transformer architecture and learnable absolute position embeddings. \n2. BERT with a Transformer architecture and T5-style learnable relative position biases [5]. The biases are both layer- and head-specific as we find this yields the best results. \n3. Same as above, but we remove all content-dependent terms inside the softmax and only retain the relative positional biases. This baseline is a straightforward variant of Transformers without self-attention, which can also be viewed as a Random Synthesizer [40]. \n4. MLP-Mixer [20] which replaces the multi-head self-attention module in Transformers with a two-layer spatial MLP. This model was developed for image classification and here we investigate it on MLM tasks using the same training setup with BERT and gMLP. ", + "bbox": [ + 210, + 686, + 826, + 830 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "We compare these baselines against several versions of ${ \\mathrm { g M L P s } }$ with similar sizes in Table 3. Note that Multiplicative, Split (last row) is the Spatial Gating Unit we describe in the method section and use in the rest of the paper. First, SGU outperforms other variants in perplexity. Secondly and remarkably, gMLP with SGU also achieves perplexity comparable to Transformer. Note the difference between the strongest baseline (perplexity ${ = } 4 . 2 6 $ ) and ours (perplexity ${ = } 4 . 3 5$ ) is insignificant relative to the perplexity change when the models are scaled (see Table 4 in the next section). Spatial projection weights learned by ${ \\mathrm { g M L P s } }$ are visualized in Figure 4. ", + "bbox": [ + 174, + 842, + 825, + 911 + ], + "page_idx": 4 + }, + { + "type": "table", + "img_path": "images/230eb41e660952154dd9abe33bc6dbc3d8c04a14f867dc0c2c621db29dfc52f1.jpg", + "table_caption": [ + "Table 3: MLM validation perplexities of Transformer baselines and four versions of ${ \\mathrm { g M L P s } }$ . $f$ refers to the spatial linear projection in Equation (2) with input normalization. The MLP-Mixer baseline model has $_ { \\mathrm { L } = 2 4 }$ layers with $d _ { \\mathrm { m o d e l } } = 7 6 8$ , $d _ { \\mathrm { s p a t i a l } } { = } 3 8 4$ and $d _ { \\mathrm { f f n } } { = } 3 0 7 2$ . Each gMLP model has $_ { \\mathrm { L } = 3 6 }$ layers with $d _ { \\mathrm { m o d e l } } = 5 1 2$ and $d _ { \\mathrm { f f n } } = 3 0 7 2$ . No positional encodings are used for Mixer or gMLPs. " + ], + "table_footnote": [ + "Standard deviation across multiple independent runs is around 0.01. " + ], + "table_body": "
ModelPerplexity*Params (M)
BERTbase4.37110
BERTbase + rel pos4.26110
BERTbase + rel pos - attn5.6496
MLP-Mixer5.34112
Linear gMLP,s(Z)= f(Z)5.1492
Additive gMLP,s(Z)= Z+ f(Z)4.9792
Multiplicative gMLP,s(Z) = Z f(Z)4.5392
Multiplicative,Split gMLP,s(Z) = Z1 f(Z2),Z= Zi||Z24.35102
", + "bbox": [ + 222, + 151, + 772, + 292 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "", + "bbox": [ + 178, + 313, + 823, + 342 + ], + "page_idx": 5 + }, + { + "type": "image", + "img_path": "images/731fe5bbb1001c5b9f64529e6e9ed073521439ba52c88469a01d1124d0dfaa34.jpg", + "image_caption": [ + "Figure 4: Visualization of the spatial filters in $\\mathrm { g M L P }$ learned on the MLM task. For each layer in the model we plot the row in $W$ associated with the token in the middle of the sequence. The $\\mathbf { X }$ -axis of each subplot has a length of 128 which equal the number of tokens in the sequence. The learned filters appear to be smooth and have several types: forward-looking (e.g., 1st in 2nd row), backward-looking (e.g., 5th in 2nd row) and bi-directional (e.g., 2nd last in the last row). " + ], + "image_footnote": [], + "bbox": [ + 264, + 353, + 735, + 443 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "4.2 Case Study: The Behavior of $\\mathbf { g } \\mathbf { M L P }$ as Model Size Increases ", + "text_level": 1, + "bbox": [ + 174, + 537, + 630, + 554 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "In Table 4, we investigate the scaling properties of Transformers and gMLPs in BERT as their model capacity grows. Specifically, we scale the depth of these models by a factor of $\\{ 0 . 5 , 1 , 2 , 4 \\} \\times$ and report the their pretraining MLM perplexities on the validation set as well as finetuning results on the dev sets of two tasks in GLUE [41]. Note each individual Transformer layer is effectively two consecutive blocks: one for self-attention and one for FFN. In the table below we use the notation of $1 2 + 1 2$ to refer to 12 of self-attention blocks plus 12 of FFN blocks in the Transformer baselines. ", + "bbox": [ + 173, + 564, + 825, + 647 + ], + "page_idx": 5 + }, + { + "type": "table", + "img_path": "images/f9b9937d357c9507e5ed75ff7c559cbe2b327ae203d9f1866ee2864e0fbaa0c9.jpg", + "table_caption": [ + "Table 4: Pretraining and dev-set finetuning results over increased model capacity. We use the relative positional encoding scheme for Transformers which performs the best in Table 3. " + ], + "table_footnote": [], + "table_body": "
Model#LParams (M) PerplexitySST-2 MNLI-m
TransformergMLP6+61867594.915.2590.4 81.591.2 77.7
TransformergMLP12+12361101024.264.3591.3 83.392.3 80.9
TransformergMLP24+24721951873.833.7992.1 85.293.5 82.8
TransformergMLP48+481443653573.473.4392.8 86.395.1 84.6
", + "bbox": [ + 279, + 694, + 714, + 842 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "The results above show that a deep enough gMLP is able to match and even outperform the perplexity of Transformers with comparable capacity.4 In addition, the perplexity-parameter relationships for both architecture families approximately follow a power law (left of Figure 5). This implies the empirical scaling laws originally observed for Transformer-based language models [42] might be broadly applicable across different model families. ", + "bbox": [ + 176, + 848, + 825, + 877 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 90, + 825, + 133 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/03ebb35b54fcbc8c3b3a364ab001ed17abb00c96397df6daed10ed232e24238b.jpg", + "image_caption": [ + "Figure 5: Scaling properties with respect to perplexity and finetuning accuracies. The figures show that for pretraining, gMLPs are equally good at optimizing perplexity as Transformers. For finetuning, the two model families exhibit comparable scalability despite task-specific offsets. " + ], + "image_footnote": [], + "bbox": [ + 204, + 143, + 792, + 286 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Table 4 also leads to an interesting observation that the pretraining perplexities across different model families are not equal in terms of finetuning. While gMLPs outperform Transformers on SST-2, they are worse on MNLI. The results imply that the finetuning performance for NLP tasks is a function of not only the perplexity but also the inductive bias in the architecture. Figure 5 shows that despite the architecture-specific discrepancies between pretraining and finetuning, gMLPs and Transformers exhibit comparable scalability (slope) on both finetuning tasks. This means one can always offset the gap by enlarging the model capacity. In other words, the results indicate that model scalability with respect to downstream metrics can be independent from the presence of self-attention. ", + "bbox": [ + 173, + 352, + 825, + 463 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "4.3 Ablation: The Usefulness of Tiny Attention in BERT’s Finetuning ", + "text_level": 1, + "bbox": [ + 174, + 478, + 671, + 494 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "So far we have found that self-attention is not a required component to achieve strong MLM perplexity or scalability. At the meantime, we also identified NLP finetuning tasks where gMLPs transfer less well than Transformers (Table 4). The fact that our MLP-like model is advantageous on SST-2 but worse on MNLI is particularly informative—the former is a single-sentence task whereas the latter involves sentence pairs (premise and hypothesis) [43]. We suspect the role of self-attention during finetuning is related to cross-sentence alignment. ", + "bbox": [ + 174, + 503, + 825, + 588 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "To isolate the effect of self-attention, we experiment with a hybrid model where a tiny self-attention block is attached to the gating function of gMLP (Figure 6). Since gMLP itself is already capable in capturing spatial relationships, we hypothesize that this extra self-attention module does not have to be heavy, and that its presence is more relevant than its capacity. A typical tiny attention module in our experiments has only a single head with size 64, significantly smaller than a typical multi-head self-attention in Transformers with 12 heads and a total size of 768. In the following, we refer to the hybrid model, namely $\\mathrm { g M L P }$ with a tiny self-attention, as aMLP (“a” for attention). ", + "bbox": [ + 173, + 593, + 825, + 691 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Pseudo-code for the tiny attention module ", + "text_level": 1, + "bbox": [ + 491, + 713, + 767, + 727 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "def tiny_attn(x, d_out, d_attn=64): qkv $=$ proj(x, 3 ∗ d_attn, axis=\"channel\") q, k, v $=$ split(qkv, 3, axis=\"channel\") $\\kappa =$ einsum(\"bnd,bmd−>bnm\", q, k) a = softmax(w $^ *$ rsqrt(d_attn)) x = einsum(\"bnm,bmd−>bnd\", a, v) return proj(x, d_out, axis=\"channel\") ", + "bbox": [ + 467, + 732, + 748, + 811 + ], + "page_idx": 6 + }, + { + "type": "image", + "img_path": "images/e228c75c14097e90bf3596da75b5d6c0ccbec039247e6155396cb13236ea093c.jpg", + "image_caption": [ + "Figure 6: Hybrid spatial gating unit with a tiny self-attention module. We use the normalized input of the $\\mathrm { g M L P }$ block (endpoint after the input normalization and right before the channel expansion) as the input to the tiny self-attention. For SGU we have ${ d _ { \\mathrm { o u t } } = d _ { \\mathrm { f f n } } / 2 }$ due to the channel split. " + ], + "image_footnote": [], + "bbox": [ + 205, + 705, + 387, + 818 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "In Figure 7, we investigate the transferability of MLM models via the calibration plots between their pretraining perplexities and finetuning metrics. Models evaluated include $\\mathbf { B E R T _ { b a s e } }$ , gMLP and its hybrid version aMLP with a 64-d single-head self-attention (Figure 6). The data points were collected by varying the model depth by $\\{ 0 . 5 , 1 , 2 \\} \\times$ or data by $\\{ 1 , 2 , 4 , 8 \\} \\times$ . It can be seen that gMLPs transfer better to SST-2 than Transformers regardless of the presence of self-attention, While gMLP performs worse on MNLI, attaching a tiny bit of self-attention is sufficient to close the gap. In Appendix D we visualize the tiny self-attention modules in aMLP over MNLI examples, showing that they are primarily responsible for the alignment between sentence pairs. ", + "bbox": [ + 173, + 882, + 821, + 911 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 90, + 826, + 174 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/263412114ce248c9b9dccbc4dbdcbba36ac2aa648dce820d76c34069690a0cd1.jpg", + "image_caption": [ + "Figure 7: Transferability from MLM pretraining perpexity to finetuning accuracies on GLUE. aMLP refers to $\\mathrm { g M L P }$ enhanced with a 64-d single-head self-attention, as illustrated in Figure 6. In contrast, each self-attention module in the BERT baseline contains 12 heads with a total size of 768. " + ], + "image_footnote": [], + "bbox": [ + 233, + 185, + 735, + 361 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "In Figure 8 we put together the scaling properties of the three models, showing that aMLP $\\mathrm { ( g M L P + }$ tiny attention) consistently outperforms Transformer on both finetuning tasks. ", + "bbox": [ + 173, + 420, + 825, + 449 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/395d749d892cbad8add03840c7c07a237a2c72cef704c6aff4263e86c3de1d09.jpg", + "image_caption": [ + "Figure 8: Comparing the scaling properties of Transformers, $\\mathrm { g M L P s }$ and aMLPs (with 64-d, singlehead attention). Results were obtained using the same setup in Section 4.2. " + ], + "image_footnote": [], + "bbox": [ + 204, + 460, + 792, + 603 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "4.4 Main Results for MLM in the BERT Setup ", + "text_level": 1, + "bbox": [ + 173, + 664, + 511, + 680 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Below we present pretraining and finetuning results in the full BERT setup. Different from ablation and case studies, here we use the full English C4 dataset and adopt a common MLM setup with batch size 256, max length 512 and 1M training steps. For fair comparison, we adjust the depth and width of $\\mathrm { g M L P s }$ to ensure comparable model capacity with the Transformer baselines. The model specifications are given in Table 5 and hyperparameters are detailed in Appendix A.2. For finetuning, we report the dev-set performance for SST-2 and MNLI in GLUE [41] and each result entry was obtained by taking the median of five independent runs. In addition, we report finetuning results on SQuAD [44, 45] to test the models’ ability in reasoning over a longer context. ", + "bbox": [ + 173, + 689, + 825, + 801 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Results are presented in Table 6. Consistent with our findings earlier in Section 4.1 and Section 4.2, gMLPs are competitive with Transformers in terms of perplexity, especially in the larger scale setup. There are several observations related to the finetuning results: ", + "bbox": [ + 174, + 808, + 825, + 849 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "First, on finetuning tasks where gMLPs underperform Transformers, the performance gap tends to narrow as the model capacity increases. For example, while $\\mathrm { g M L P }$ performs worse by $8 . 5 \\%$ on SQuAD-v2.0 in the base scale, the performance gap relative to the baseline decreases to $2 . 7 \\%$ at the larger scale. Notably, our ${ \\mathrm { g M L P } } _ { \\mathrm { l a r g e } }$ achieves $8 9 . 5 \\%$ F1 on SQuAD-v1.1 without any self-attention or dynamic spatial parameterization [28], which is well above the $8 8 . 5 \\%$ reported for $\\mathbf { B E R T _ { b a s e } }$ in Devlin et al. [2] and is only $1 . 4 \\%$ away from the original result for $\\mathbf { B E R T _ { l a r g e } }$ . We also include one additional data point by scaling up $\\mathrm { g M L P }$ even further. The resulting model, $\\mathrm { g } \\mathrm { \\bar { M } L P _ { \\mathrm { x l a r g e } } }$ , outperforms $\\mathbf { B E R T _ { l a r g e } }$ on SQuAD- $\\mathbf { \\nabla } \\cdot \\mathbf { v } 2 . 0 $ —a difficult task involving question-answer pairs—without any self-attention. While this is not a fair comparison due to different model sizes, it is an existence proof that MLP-like models can be competitive with Transformers on challenging NLP tasks. ", + "bbox": [ + 174, + 856, + 825, + 911 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/de04ff846f28de47d3407d02d990efd065f5be28122340787c79d35fd6edcd9d.jpg", + "table_caption": [ + "Table 5: Model specifications in the full BERT setup. " + ], + "table_footnote": [], + "table_body": "
Params (M)FLOPs (B)#Ldmodeldffn
BERTbase gMLPbase aMLPbase110100.812+127683072
130158.0485123072
109128.9365123072
BERTlarge gMLPlarge aMLPlarge336341.224+2410244096
365430.1967683072
316370.3727683072
gMLPxlarge9411091.314410244096
", + "bbox": [ + 290, + 109, + 707, + 239 + ], + "page_idx": 8 + }, + { + "type": "table", + "img_path": "images/5ba707c6c8f0c199b6cc50563f0f4952642180841fe1165c9562b1c6903bcbe3.jpg", + "table_caption": [ + "Table 6: Pretraining perplexities and dev-set results for finetuning. “ours” indicates models trained using our setup. We report accuracies for SST-2 and MNLI, and F1 scores for SQuAD v1.1/2.0. " + ], + "table_footnote": [], + "table_body": "
PerplexitySST-2MNLI (m/mm)SQuADAttn SizeParams (M)
v1.1v2.0
BERTbase [2]192.784.4/-88.576.3768 (64 × 12)110
BERTbase (ours)4.1793.885.6/85.790.278.6768 (64 × 12)110
gMLPbase4.2894.283.7/84.186.770.11130
aMLPbase3.9593.485.9/85.890.780.964109
BERTlarge [2]193.786.6/-90.981.81024 (64 × 16)336
BERTlarge (ours)3.3594.387.0/87.492.081.01024 (64 × 16)336
gMLPlarge3.3294.886.2/86.589.578.31365
aMLPlarge3.1994.888.4/88.492.285.4128316
gMLPxlarge2.8995.687.7/87.790.982.11941
", + "bbox": [ + 184, + 289, + 810, + 497 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "", + "bbox": [ + 173, + 525, + 825, + 608 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Furthermore, we show that blending in a tiny single-head self-attention of size either 64 or 128 is sufficient to make gMLPs outperform Transformers of similar capacity, sometimes by a significant margin. For example, our hybrid model $\\mathrm { a M L P _ { l a r g e } }$ achieves $4 . 4 \\%$ higher F1 than Transformers on SQuAD-v2.0. The results suggest that the capacity in the multi-head self-attention of Transformers can be largely redundant, and that the majority of its functionalities can be captured by the spatial gating unit in gMLPs. The results also imply that the inductive biases in the spatial gating unit of gMLPs and the tiny attention are complementary to each other. While the benefits of architectural inductive bias may vanish over increased compute, tiny attention does improve the practical value of gMLPs in the regime that we investigate in this work. ", + "bbox": [ + 174, + 614, + 825, + 739 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "5 Conclusion ", + "text_level": 1, + "bbox": [ + 174, + 761, + 299, + 779 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Since the seminal work of Vaswani et al. [1], Transformers have been widely adopted across NLP and computer vision. This adoption has enabled many impressive results especially in NLP. To date, it is still unclear what empowers such success: is it the feedforward nature of Transformers or is it the multi-head self-attention layers in Transformers? ", + "bbox": [ + 174, + 794, + 825, + 849 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Our work suggests a simpler alternative to the multi-head self-attention layers in Transformers. We show that gMLPs, a simple variant of MLPs with gating, can be competitive with Transformers in terms of BERT’s pretraining perplexity and ViT’s accuracy. gMLPs are also comparable with Transformers in terms of the scalability over increased data and compute. As for BERT finetuning, we find gMLPs can achieve appealing results on challenging tasks such as SQuAD without self-attention, and can significantly outperform Transformers in certain cases. We also find the inductive bias in Transformer’s multi-head self-attention useful on downstream tasks that require cross-sentence alignment. However in those cases, making gMLP substantially larger closes the gap with Transformers. More practically, blending a small single-head self-attention into gMLP allows for an even better architecture without the need for increasing model size. ", + "bbox": [ + 176, + 856, + 823, + 911 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "", + "bbox": [ + 174, + 92, + 826, + 174 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Acknowledgements ", + "text_level": 1, + "bbox": [ + 176, + 193, + 338, + 210 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "We thank Gabriel Bender, Neil Houlsby, Thang Luong, Niki Parmar, Hieu Pham, Jascha SohlDickstein, Noam Shazeer, Ilya Sutskever, Jakob Uszkoreit and Ashish Vaswani for their feedback. ", + "bbox": [ + 176, + 224, + 825, + 252 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "References ", + "text_level": 1, + "bbox": [ + 174, + 272, + 266, + 287 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "[1] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in Neural Information Processing Systems, 2017. \n[2] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. In NAACL, 2018. \n[3] Zhilin Yang, Zihang Dai, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, and Quoc V Le. Xlnet: Generalized autoregressive pretraining for language understanding. In Advances in Neural Information Processing Systems, 2019. \n[4] Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692, 2019. [5] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified text-to-text transformer. JMLR, 2020. \n[6] Tom B Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners. In Advances in Neural Information Processing Systems, 2020. \n[7] Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al. An image is worth 16x16 words: Transformers for image recognition at scale. In ICLR, 2021. [8] Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and Hervé Jégou. Training data-efficient image transformers & distillation through attention. arXiv preprint arXiv:2012.12877, 2020. \n[9] Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and Sergey Zagoruyko. End-to-end object detection with transformers. In ECCV, 2020. \n[10] Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining Guo. Swin transformer: Hierarchical vision transformer using shifted windows. arXiv preprint arXiv:2103.14030, 2021. \n[11] Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory. Neural Computation, 1997. \n[12] Y. LeCun, B. Boser, J. S. Denker, D. Henderson, R. E. Howard, W. Hubbard, and L. D. Jackel. Backpropagation applied to handwritten zip code recognition. Neural Computation, 1989. \n[13] Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep convolutional neural networks. Advances in Neural Information Processing Systems, 2012. \n[14] Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale image recognition. In ICLR, 2015. \n[15] Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, and Andrew Rabinovich. Going deeper with convolutions. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1–9, 2015. \n[16] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In CVPR, 2016. \n[17] Mingxing Tan and Quoc Le. Efficientnet: Rethinking model scaling for convolutional neural networks. In ICML, 2019. \n[18] Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. Neural machine translation by jointly learning to align and translate. In ICLR, 2015. \n[19] Kurt Hornik, Maxwell Stinchcombe, and Halbert White. Multilayer feedforward networks are universal approximators. Neural networks, 2(5):359–366, 1989. \n[20] Ilya Tolstikhin, Neil Houlsby, Alexander Kolesnikov, Lucas Beyer, Xiaohua Zhai, Thomas Unterthiner, Jessica Yung, Daniel Keysers, Jakob Uszkoreit, Mario Lucic, and Alexey Dosovitskiy. Mlp-mixer: An all-mlp architecture for vision. arXiv preprint arXiv:2105.01601, 2021. \n[21] Luke Melas-Kyriazi. Do you even need attention? a stack of feed-forward layers does surprisingly well on imagenet. arXiv preprint arXiv:2105.02723, 2021. \n[22] Hugo Touvron, Piotr Bojanowski, Mathilde Caron, Matthieu Cord, Alaaeldin El-Nouby, Edouard Grave, Armand Joulin, Gabriel Synnaeve, Jakob Verbeek, and Hervé Jégou. Resmlp: Feedforward networks for image classification with data-efficient training. arXiv preprint arXiv:2105.03404, 2021. \n[23] Xiaohan Ding, Xiangyu Zhang, Jungong Han, and Guiguang Ding. Repmlp: Reparameterizing convolutions into fully-connected layers for image recognition. arXiv preprint arXiv:2105.01883, 2021. \n[24] Dan Hendrycks and Kevin Gimpel. Gaussian error linear units (gelus). arXiv preprint arXiv:1606.08415, 2016. \n[25] Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen. Mobilenetv2: Inverted residuals and linear bottlenecks. In CVPR, 2018. \n[26] Yann N Dauphin, Angela Fan, Michael Auli, and David Grangier. Language modeling with gated convolutional networks. In ICML, 2017. \n[27] Noam Shazeer. Glu variants improve transformer. arXiv preprint arXiv:2002.05202, 2020. \n[28] Felix Wu, Angela Fan, Alexei Baevski, Yann N Dauphin, and Michael Auli. Pay less attention with lightweight and dynamic convolutions. In ICLR, 2019. \n[29] Rupesh Kumar Srivastava, Klaus Greff, and Jürgen Schmidhuber. Highway networks. arXiv preprint arXiv:1505.00387, 2015. \n[30] Jie Hu, Li Shen, and Gang Sun. Squeeze-and-excitation networks. In CVPR, 2018. \n[31] Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A largescale hierarchical image database. In 2009 IEEE conference on computer vision and pattern recognition, pages 248–255. Ieee, 2009. \n[32] Gao Huang, Yu Sun, Zhuang Liu, Daniel Sedra, and Kilian Q Weinberger. Deep networks with stochastic depth. In ECCV, 2016. \n[33] Andrew Brock, Soham De, Samuel L Smith, and Karen Simonyan. High-performance largescale image recognition without normalization. arXiv preprint arXiv:2102.06171, 2021. \n[34] Mingxing Tan and Quoc V Le. Efficientnetv2: Smaller models and faster training. In ICML, 2021. \n[35] Irwan Bello. Lambdanetworks: Modeling long-range interactions without attention. In ICLR, 2021. \n[36] Ashish Vaswani, Prajit Ramachandran, Aravind Srinivas, Niki Parmar, Blake Hechtman, and Jonathon Shlens. Scaling local self-attention for parameter efficient visual backbones. In CVPR, 2021. \n[37] Aravind Srinivas, Tsung-Yi Lin, Niki Parmar, Jonathon Shlens, Pieter Abbeel, and Ashish Vaswani. Bottleneck transformers for visual recognition. arXiv preprint arXiv:2101.11605, 2021. \n[38] Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, and Lei Zhang. Cvt: Introducing convolutions to vision transformers. arXiv preprint arXiv:2103.15808, 2021. \n[39] Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, and Piotr Dollár. Designing network design spaces. In CVPR, 2020. \n[40] Yi Tay, Dara Bahri, Donald Metzler, Da-Cheng Juan, Zhe Zhao, and Che Zheng. Synthesizer: Rethinking self-attention in transformer models. arXiv preprint arXiv:2005.00743, 2020. \n[41] Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel R Bowman. Glue: A multi-task benchmark and analysis platform for natural language understanding. In ICLR, 2019. \n[42] Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling laws for neural language models. arXiv preprint arXiv:2001.08361, 2020. \n[43] Adina Williams, Nikita Nangia, and Samuel R Bowman. A broad-coverage challenge corpus for sentence understanding through inference. In NAACL, 2017. \n[44] Pranav Rajpurkar, Jian Zhang, Konstantin Lopyrev, and Percy Liang. Squad: $^ { 1 0 0 , 0 0 0 + }$ questions for machine comprehension of text. In EMNLP, 2016. \n[45] Pranav Rajpurkar, Robin Jia, and Percy Liang. Know what you don’t know: Unanswerable questions for squad. In ACL, 2018. \n[46] Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr Dollár, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In European conference on computer vision, pages 740–755. Springer, 2014. \n[47] Mingxing Tan, Ruoming Pang, and Quoc V Le. Efficientdet: Scalable and efficient object detection. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 10781–10790, 2020. \n[48] Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping Luo, and Ling Shao. Pyramid vision transformer: A versatile backbone for dense prediction without convolutions. arXiv preprint arXiv:2102.12122, 2021. \n[49] Josh Beal, Eric Kim, Eric Tzeng, Dong Huk Park, Andrew Zhai, and Dmitry Kislyuk. Toward transformer-based object detection. arXiv preprint arXiv:2012.09958, 2020. ", + "bbox": [ + 178, + 291, + 826, + 914 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 56, + 828, + 915 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "", + "bbox": [ + 171, + 89, + 828, + 517 + ], + "page_idx": 11 + } +] \ No newline at end of file diff --git a/parse/train/KBnXrODoBW/KBnXrODoBW_middle.json b/parse/train/KBnXrODoBW/KBnXrODoBW_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..2e67f96c66e44afe116d257d0f910d7d983d2083 --- /dev/null +++ b/parse/train/KBnXrODoBW/KBnXrODoBW_middle.json @@ -0,0 +1,31585 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 220, + 97, + 392, + 117 + ], + "lines": [ + { + "bbox": [ + 217, + 97, + 393, + 117 + ], + "spans": [ + { + "bbox": [ + 217, + 97, + 393, + 117 + ], + "score": 1.0, + "content": "Pay Attention to MLPs", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 198, + 159, + 414, + 193 + ], + "lines": [ + { + "bbox": [ + 199, + 159, + 415, + 172 + ], + "spans": [ + { + "bbox": [ + 199, + 159, + 415, + 172 + ], + "score": 1.0, + "content": "Hanxiao Liu, Zihang Dai, David R. So, Quoc V. Le", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 244, + 171, + 367, + 181 + ], + "spans": [ + { + "bbox": [ + 244, + 171, + 367, + 181 + ], + "score": 1.0, + "content": "Google Research, Brain Team", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 198, + 181, + 414, + 194 + ], + "spans": [ + { + "bbox": [ + 198, + 181, + 414, + 194 + ], + "score": 1.0, + "content": "{hanxiaol,zihangd,davidso,qvl}@google.com", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2 + }, + { + "type": "title", + "bbox": [ + 283, + 221, + 328, + 234 + ], + "lines": [ + { + "bbox": [ + 281, + 220, + 330, + 235 + ], + "spans": [ + { + "bbox": [ + 281, + 220, + 330, + 235 + ], + "score": 1.0, + "content": "Abstract", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 143, + 245, + 469, + 365 + ], + "lines": [ + { + "bbox": [ + 142, + 245, + 469, + 257 + ], + "spans": [ + { + "bbox": [ + 142, + 245, + 469, + 257 + ], + "score": 1.0, + "content": "Transformers [1] have become one of the most important architectural innovations", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 141, + 255, + 471, + 270 + ], + "spans": [ + { + "bbox": [ + 141, + 255, + 471, + 270 + ], + "score": 1.0, + "content": "in deep learning and have enabled many breakthroughs over the past few years.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 266, + 471, + 281 + ], + "spans": [ + { + "bbox": [ + 141, + 266, + 471, + 281 + ], + "score": 1.0, + "content": "Here we propose a simple network architecture, gMLP, based on MLPs with gating,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 278, + 470, + 290 + ], + "spans": [ + { + "bbox": [ + 141, + 278, + 470, + 290 + ], + "score": 1.0, + "content": "and show that it can perform as well as Transformers in key language and vision", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 289, + 469, + 300 + ], + "spans": [ + { + "bbox": [ + 141, + 289, + 469, + 300 + ], + "score": 1.0, + "content": "applications. Our comparisons show that self-attention is not critical for Vision", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 299, + 470, + 312 + ], + "spans": [ + { + "bbox": [ + 141, + 299, + 213, + 312 + ], + "score": 1.0, + "content": "Transformers, as", + "type": "text" + }, + { + "bbox": [ + 214, + 300, + 242, + 311 + ], + "score": 0.75, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 299, + 470, + 312 + ], + "score": 1.0, + "content": "can achieve the same accuracy. For BERT, our model", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 311, + 470, + 323 + ], + "spans": [ + { + "bbox": [ + 141, + 311, + 470, + 323 + ], + "score": 1.0, + "content": "achieves parity with Transformers on pretraining perplexity and is better on some", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 321, + 469, + 334 + ], + "spans": [ + { + "bbox": [ + 141, + 321, + 469, + 334 + ], + "score": 1.0, + "content": "downstream NLP tasks. On finetuning tasks where gMLP performs worse, making", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 332, + 470, + 345 + ], + "spans": [ + { + "bbox": [ + 141, + 332, + 158, + 345 + ], + "score": 1.0, + "content": "the", + "type": "text" + }, + { + "bbox": [ + 158, + 333, + 186, + 344 + ], + "score": 0.3, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 332, + 470, + 345 + ], + "score": 1.0, + "content": "model substantially larger can close the gap with Transformers. In", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 344, + 470, + 356 + ], + "spans": [ + { + "bbox": [ + 141, + 344, + 470, + 356 + ], + "score": 1.0, + "content": "general, our experiments show that gMLP can scale as well as Transformers over", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 355, + 258, + 367 + ], + "spans": [ + { + "bbox": [ + 141, + 355, + 258, + 367 + ], + "score": 1.0, + "content": "increased data and compute.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 10 + }, + { + "type": "title", + "bbox": [ + 107, + 384, + 190, + 398 + ], + "lines": [ + { + "bbox": [ + 105, + 383, + 192, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 383, + 192, + 400 + ], + "score": 1.0, + "content": "1 Introduction", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 107, + 409, + 505, + 453 + ], + "lines": [ + { + "bbox": [ + 105, + 409, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 505, + 421 + ], + "score": 1.0, + "content": "Transformers [1] have enabled many breakthroughs in natural language processing (e.g., [2, 3, 4, 5, 6])", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 420, + 507, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 507, + 433 + ], + "score": 1.0, + "content": "and have been shown to work well for computer vision (e.g., [7, 8, 9, 10]). Thanks to this success,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 431, + 505, + 443 + ], + "spans": [ + { + "bbox": [ + 106, + 431, + 505, + 443 + ], + "score": 1.0, + "content": "Transformers have largely replaced LSTM-RNN [11] as the default architecture in NLP, and have", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 442, + 463, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 463, + 455 + ], + "score": 1.0, + "content": "become an appealing alternative to ConvNets [12, 13, 14, 15, 16, 17] in computer vision.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 107, + 458, + 505, + 535 + ], + "lines": [ + { + "bbox": [ + 105, + 458, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 458, + 505, + 471 + ], + "score": 1.0, + "content": "The Transformer architecture combines two important concepts: (1) a recurrent-free architecture", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 469, + 506, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 506, + 482 + ], + "score": 1.0, + "content": "which computes the representations for each individual token in parallel, and (2) multi-head self-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 480, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 505, + 492 + ], + "score": 1.0, + "content": "attention blocks which aggregate spatial information across tokens. On one hand, the attention", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 491, + 505, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 505, + 504 + ], + "score": 1.0, + "content": "mechanism [18] introduces the inductive bias that the spatial interactions should be dynamically", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 502, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 505, + 514 + ], + "score": 1.0, + "content": "parameterized based on the input representations. On the other hand, it is known that MLPs with static", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 513, + 506, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 506, + 525 + ], + "score": 1.0, + "content": "parameterization can represent arbitrary functions [19]. It therefore remains an open question whether", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 523, + 485, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 485, + 536 + ], + "score": 1.0, + "content": "the inductive bias in self-attention is essential to the remarkable effectiveness of Transformers.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 540, + 505, + 606 + ], + "lines": [ + { + "bbox": [ + 105, + 539, + 507, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 507, + 553 + ], + "score": 1.0, + "content": "Here we study the necessity of self-attention modules in key language and vision applications of Trans-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 551, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 506, + 563 + ], + "score": 1.0, + "content": "formers. Specifically, we propose an MLP-based alternative to Transformers without self-attention,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 563, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 563, + 505, + 574 + ], + "score": 1.0, + "content": "which simply consists of channel projections and spatial projections with static parameterization. We", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 573, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 505, + 585 + ], + "score": 1.0, + "content": "experiment with several design choices for this architecture and find spatial projections work well", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 583, + 506, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 506, + 596 + ], + "score": 1.0, + "content": "when they are linear and paired with multiplicative gating (Figure 1). We name the model gMLP", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 593, + 326, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 593, + 326, + 609 + ], + "score": 1.0, + "content": "because it is built out of basic MLP layers with gating.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 30.5 + }, + { + "type": "text", + "bbox": [ + 107, + 611, + 505, + 666 + ], + "lines": [ + { + "bbox": [ + 106, + 610, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 106, + 610, + 148, + 624 + ], + "score": 1.0, + "content": "We apply", + "type": "text" + }, + { + "bbox": [ + 148, + 611, + 176, + 623 + ], + "score": 0.3, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 610, + 438, + 624 + ], + "score": 1.0, + "content": "to image classification and obtain strong results on ImageNet.", + "type": "text" + }, + { + "bbox": [ + 438, + 611, + 467, + 623 + ], + "score": 0.3, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 610, + 505, + 624 + ], + "score": 1.0, + "content": "achieves", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 622, + 506, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 506, + 635 + ], + "score": 1.0, + "content": "comparable performance with DeiT [8], namely Vision Transformer (ViT) [7] with improved regular-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 633, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 106, + 633, + 269, + 645 + ], + "score": 1.0, + "content": "ization, in a similar training setup. With", + "type": "text" + }, + { + "bbox": [ + 270, + 633, + 290, + 644 + ], + "score": 0.86, + "content": "66 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 633, + 429, + 645 + ], + "score": 1.0, + "content": "less parameters, a gMLP model is", + "type": "text" + }, + { + "bbox": [ + 430, + 633, + 444, + 644 + ], + "score": 0.85, + "content": "3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 633, + 505, + 645 + ], + "score": 1.0, + "content": "more accurate", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 642, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 657 + ], + "score": 1.0, + "content": "than MLP-Mixer [20]. Together with Tolstikhin et al. [20], Melas-Kyriazi [21], Touvron et al. [22]", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 654, + 507, + 668 + ], + "spans": [ + { + "bbox": [ + 106, + 654, + 507, + 668 + ], + "score": 1.0, + "content": "and Ding et. al. [23], our results question the necessity of self-attention layers in Vision Transformers.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 107, + 671, + 505, + 715 + ], + "lines": [ + { + "bbox": [ + 105, + 671, + 506, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 506, + 684 + ], + "score": 1.0, + "content": "We apply gMLP to masked language modeling (MLM) in the BERT [2] setup, one of the most well-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 681, + 505, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 505, + 695 + ], + "score": 1.0, + "content": "established applications of Transformers, and find that it is as good as Transformers at minimizing", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 693, + 506, + 706 + ], + "spans": [ + { + "bbox": [ + 105, + 693, + 506, + 706 + ], + "score": 1.0, + "content": "perplexity during pretraining. Our experiments indicate that perplexity is only correlated with model", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 704, + 506, + 716 + ], + "spans": [ + { + "bbox": [ + 105, + 704, + 506, + 716 + ], + "score": 1.0, + "content": "capacity and is insensitive to the presence of self-attention. As capacity increases, we observe that", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40.5 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 107, + 731, + 385, + 741 + ], + "lines": [ + { + "bbox": [ + 105, + 730, + 387, + 744 + ], + "spans": [ + { + "bbox": [ + 105, + 730, + 387, + 744 + ], + "score": 1.0, + "content": "35th Conference on Neural Information Processing Systems (NeurIPS 2021).", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 220, + 97, + 392, + 117 + ], + "lines": [ + { + "bbox": [ + 217, + 97, + 393, + 117 + ], + "spans": [ + { + "bbox": [ + 217, + 97, + 393, + 117 + ], + "score": 1.0, + "content": "Pay Attention to MLPs", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "text", + "bbox": [ + 198, + 159, + 414, + 193 + ], + "lines": [ + { + "bbox": [ + 199, + 159, + 415, + 172 + ], + "spans": [ + { + "bbox": [ + 199, + 159, + 415, + 172 + ], + "score": 1.0, + "content": "Hanxiao Liu, Zihang Dai, David R. So, Quoc V. Le", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 244, + 171, + 367, + 181 + ], + "spans": [ + { + "bbox": [ + 244, + 171, + 367, + 181 + ], + "score": 1.0, + "content": "Google Research, Brain Team", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 198, + 181, + 414, + 194 + ], + "spans": [ + { + "bbox": [ + 198, + 181, + 414, + 194 + ], + "score": 1.0, + "content": "{hanxiaol,zihangd,davidso,qvl}@google.com", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2, + "bbox_fs": [ + 198, + 159, + 415, + 194 + ] + }, + { + "type": "title", + "bbox": [ + 283, + 221, + 328, + 234 + ], + "lines": [ + { + "bbox": [ + 281, + 220, + 330, + 235 + ], + "spans": [ + { + "bbox": [ + 281, + 220, + 330, + 235 + ], + "score": 1.0, + "content": "Abstract", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 143, + 245, + 469, + 365 + ], + "lines": [ + { + "bbox": [ + 142, + 245, + 469, + 257 + ], + "spans": [ + { + "bbox": [ + 142, + 245, + 469, + 257 + ], + "score": 1.0, + "content": "Transformers [1] have become one of the most important architectural innovations", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 141, + 255, + 471, + 270 + ], + "spans": [ + { + "bbox": [ + 141, + 255, + 471, + 270 + ], + "score": 1.0, + "content": "in deep learning and have enabled many breakthroughs over the past few years.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 141, + 266, + 471, + 281 + ], + "spans": [ + { + "bbox": [ + 141, + 266, + 471, + 281 + ], + "score": 1.0, + "content": "Here we propose a simple network architecture, gMLP, based on MLPs with gating,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 141, + 278, + 470, + 290 + ], + "spans": [ + { + "bbox": [ + 141, + 278, + 470, + 290 + ], + "score": 1.0, + "content": "and show that it can perform as well as Transformers in key language and vision", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 141, + 289, + 469, + 300 + ], + "spans": [ + { + "bbox": [ + 141, + 289, + 469, + 300 + ], + "score": 1.0, + "content": "applications. Our comparisons show that self-attention is not critical for Vision", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 141, + 299, + 470, + 312 + ], + "spans": [ + { + "bbox": [ + 141, + 299, + 213, + 312 + ], + "score": 1.0, + "content": "Transformers, as", + "type": "text" + }, + { + "bbox": [ + 214, + 300, + 242, + 311 + ], + "score": 0.75, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 242, + 299, + 470, + 312 + ], + "score": 1.0, + "content": "can achieve the same accuracy. For BERT, our model", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 141, + 311, + 470, + 323 + ], + "spans": [ + { + "bbox": [ + 141, + 311, + 470, + 323 + ], + "score": 1.0, + "content": "achieves parity with Transformers on pretraining perplexity and is better on some", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 141, + 321, + 469, + 334 + ], + "spans": [ + { + "bbox": [ + 141, + 321, + 469, + 334 + ], + "score": 1.0, + "content": "downstream NLP tasks. On finetuning tasks where gMLP performs worse, making", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 141, + 332, + 470, + 345 + ], + "spans": [ + { + "bbox": [ + 141, + 332, + 158, + 345 + ], + "score": 1.0, + "content": "the", + "type": "text" + }, + { + "bbox": [ + 158, + 333, + 186, + 344 + ], + "score": 0.3, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 186, + 332, + 470, + 345 + ], + "score": 1.0, + "content": "model substantially larger can close the gap with Transformers. In", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 141, + 344, + 470, + 356 + ], + "spans": [ + { + "bbox": [ + 141, + 344, + 470, + 356 + ], + "score": 1.0, + "content": "general, our experiments show that gMLP can scale as well as Transformers over", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 141, + 355, + 258, + 367 + ], + "spans": [ + { + "bbox": [ + 141, + 355, + 258, + 367 + ], + "score": 1.0, + "content": "increased data and compute.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 10, + "bbox_fs": [ + 141, + 245, + 471, + 367 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 384, + 190, + 398 + ], + "lines": [ + { + "bbox": [ + 105, + 383, + 192, + 400 + ], + "spans": [ + { + "bbox": [ + 105, + 383, + 192, + 400 + ], + "score": 1.0, + "content": "1 Introduction", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 107, + 409, + 505, + 453 + ], + "lines": [ + { + "bbox": [ + 105, + 409, + 505, + 421 + ], + "spans": [ + { + "bbox": [ + 105, + 409, + 505, + 421 + ], + "score": 1.0, + "content": "Transformers [1] have enabled many breakthroughs in natural language processing (e.g., [2, 3, 4, 5, 6])", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 420, + 507, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 507, + 433 + ], + "score": 1.0, + "content": "and have been shown to work well for computer vision (e.g., [7, 8, 9, 10]). Thanks to this success,", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 431, + 505, + 443 + ], + "spans": [ + { + "bbox": [ + 106, + 431, + 505, + 443 + ], + "score": 1.0, + "content": "Transformers have largely replaced LSTM-RNN [11] as the default architecture in NLP, and have", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 442, + 463, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 463, + 455 + ], + "score": 1.0, + "content": "become an appealing alternative to ConvNets [12, 13, 14, 15, 16, 17] in computer vision.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5, + "bbox_fs": [ + 105, + 409, + 507, + 455 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 458, + 505, + 535 + ], + "lines": [ + { + "bbox": [ + 105, + 458, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 458, + 505, + 471 + ], + "score": 1.0, + "content": "The Transformer architecture combines two important concepts: (1) a recurrent-free architecture", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 469, + 506, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 506, + 482 + ], + "score": 1.0, + "content": "which computes the representations for each individual token in parallel, and (2) multi-head self-", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 480, + 505, + 492 + ], + "spans": [ + { + "bbox": [ + 105, + 480, + 505, + 492 + ], + "score": 1.0, + "content": "attention blocks which aggregate spatial information across tokens. On one hand, the attention", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 491, + 505, + 504 + ], + "spans": [ + { + "bbox": [ + 105, + 491, + 505, + 504 + ], + "score": 1.0, + "content": "mechanism [18] introduces the inductive bias that the spatial interactions should be dynamically", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 502, + 505, + 514 + ], + "spans": [ + { + "bbox": [ + 105, + 502, + 505, + 514 + ], + "score": 1.0, + "content": "parameterized based on the input representations. On the other hand, it is known that MLPs with static", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 513, + 506, + 525 + ], + "spans": [ + { + "bbox": [ + 105, + 513, + 506, + 525 + ], + "score": 1.0, + "content": "parameterization can represent arbitrary functions [19]. It therefore remains an open question whether", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 523, + 485, + 536 + ], + "spans": [ + { + "bbox": [ + 105, + 523, + 485, + 536 + ], + "score": 1.0, + "content": "the inductive bias in self-attention is essential to the remarkable effectiveness of Transformers.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 24, + "bbox_fs": [ + 105, + 458, + 506, + 536 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 540, + 505, + 606 + ], + "lines": [ + { + "bbox": [ + 105, + 539, + 507, + 553 + ], + "spans": [ + { + "bbox": [ + 105, + 539, + 507, + 553 + ], + "score": 1.0, + "content": "Here we study the necessity of self-attention modules in key language and vision applications of Trans-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 551, + 506, + 563 + ], + "spans": [ + { + "bbox": [ + 105, + 551, + 506, + 563 + ], + "score": 1.0, + "content": "formers. Specifically, we propose an MLP-based alternative to Transformers without self-attention,", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 563, + 505, + 574 + ], + "spans": [ + { + "bbox": [ + 106, + 563, + 505, + 574 + ], + "score": 1.0, + "content": "which simply consists of channel projections and spatial projections with static parameterization. We", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 573, + 505, + 585 + ], + "spans": [ + { + "bbox": [ + 105, + 573, + 505, + 585 + ], + "score": 1.0, + "content": "experiment with several design choices for this architecture and find spatial projections work well", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 583, + 506, + 596 + ], + "spans": [ + { + "bbox": [ + 105, + 583, + 506, + 596 + ], + "score": 1.0, + "content": "when they are linear and paired with multiplicative gating (Figure 1). We name the model gMLP", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 593, + 326, + 609 + ], + "spans": [ + { + "bbox": [ + 105, + 593, + 326, + 609 + ], + "score": 1.0, + "content": "because it is built out of basic MLP layers with gating.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 539, + 507, + 609 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 611, + 505, + 666 + ], + "lines": [ + { + "bbox": [ + 106, + 610, + 505, + 624 + ], + "spans": [ + { + "bbox": [ + 106, + 610, + 148, + 624 + ], + "score": 1.0, + "content": "We apply", + "type": "text" + }, + { + "bbox": [ + 148, + 611, + 176, + 623 + ], + "score": 0.3, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 610, + 438, + 624 + ], + "score": 1.0, + "content": "to image classification and obtain strong results on ImageNet.", + "type": "text" + }, + { + "bbox": [ + 438, + 611, + 467, + 623 + ], + "score": 0.3, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 610, + 505, + 624 + ], + "score": 1.0, + "content": "achieves", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 622, + 506, + 635 + ], + "spans": [ + { + "bbox": [ + 105, + 622, + 506, + 635 + ], + "score": 1.0, + "content": "comparable performance with DeiT [8], namely Vision Transformer (ViT) [7] with improved regular-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 633, + 505, + 645 + ], + "spans": [ + { + "bbox": [ + 106, + 633, + 269, + 645 + ], + "score": 1.0, + "content": "ization, in a similar training setup. With", + "type": "text" + }, + { + "bbox": [ + 270, + 633, + 290, + 644 + ], + "score": 0.86, + "content": "66 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 633, + 429, + 645 + ], + "score": 1.0, + "content": "less parameters, a gMLP model is", + "type": "text" + }, + { + "bbox": [ + 430, + 633, + 444, + 644 + ], + "score": 0.85, + "content": "3 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 445, + 633, + 505, + 645 + ], + "score": 1.0, + "content": "more accurate", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 642, + 505, + 657 + ], + "spans": [ + { + "bbox": [ + 105, + 642, + 505, + 657 + ], + "score": 1.0, + "content": "than MLP-Mixer [20]. Together with Tolstikhin et al. [20], Melas-Kyriazi [21], Touvron et al. [22]", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 654, + 507, + 668 + ], + "spans": [ + { + "bbox": [ + 106, + 654, + 507, + 668 + ], + "score": 1.0, + "content": "and Ding et. al. [23], our results question the necessity of self-attention layers in Vision Transformers.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 36, + "bbox_fs": [ + 105, + 610, + 507, + 668 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 671, + 505, + 715 + ], + "lines": [ + { + "bbox": [ + 105, + 671, + 506, + 684 + ], + "spans": [ + { + "bbox": [ + 105, + 671, + 506, + 684 + ], + "score": 1.0, + "content": "We apply gMLP to masked language modeling (MLM) in the BERT [2] setup, one of the most well-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 681, + 505, + 695 + ], + "spans": [ + { + "bbox": [ + 105, + 681, + 505, + 695 + ], + "score": 1.0, + "content": "established applications of Transformers, and find that it is as good as Transformers at minimizing", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 693, + 506, + 706 + ], + "spans": [ + { + "bbox": [ + 105, + 693, + 506, + 706 + ], + "score": 1.0, + "content": "perplexity during pretraining. Our experiments indicate that perplexity is only correlated with model", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 704, + 506, + 716 + ], + "spans": [ + { + "bbox": [ + 105, + 704, + 506, + 716 + ], + "score": 1.0, + "content": "capacity and is insensitive to the presence of self-attention. As capacity increases, we observe that", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 40.5, + "bbox_fs": [ + 105, + 671, + 506, + 716 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 311, + 94, + 461, + 161 + ], + "lines": [ + { + "bbox": [ + 311, + 93, + 447, + 103 + ], + "spans": [ + { + "bbox": [ + 311, + 93, + 447, + 103 + ], + "score": 1.0, + "content": "def gmlp_block(x, d_model, d_ffn):", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 318, + 102, + 369, + 111 + ], + "spans": [ + { + "bbox": [ + 318, + 102, + 352, + 111 + ], + "score": 1.0, + "content": "shortcut", + "type": "text" + }, + { + "bbox": [ + 352, + 104, + 362, + 110 + ], + "score": 0.33, + "content": "=", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 102, + 369, + 111 + ], + "score": 1.0, + "content": "x", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 318, + 111, + 426, + 119 + ], + "spans": [ + { + "bbox": [ + 318, + 111, + 332, + 119 + ], + "score": 0.58, + "content": "\\mathrm { ~ x ~ } =", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 111, + 426, + 119 + ], + "score": 1.0, + "content": "norm(x, axis=\"channel\")", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 318, + 119, + 454, + 128 + ], + "spans": [ + { + "bbox": [ + 318, + 120, + 332, + 127 + ], + "score": 0.66, + "content": "\\mathrm { ~ x ~ } =", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 119, + 454, + 128 + ], + "score": 1.0, + "content": "proj(x, d_ffn, axis=\"channel\")", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 318, + 128, + 363, + 137 + ], + "spans": [ + { + "bbox": [ + 318, + 129, + 332, + 136 + ], + "score": 0.5, + "content": "\\mathrm { ~ x ~ } =", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 128, + 363, + 137 + ], + "score": 1.0, + "content": "gelu(x)", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 318, + 136, + 422, + 145 + ], + "spans": [ + { + "bbox": [ + 318, + 137, + 333, + 144 + ], + "score": 0.77, + "content": "\\mathrm { ~ x ~ } =", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 136, + 422, + 145 + ], + "score": 1.0, + "content": "spatial_gating_unit(x)", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 318, + 145, + 462, + 153 + ], + "spans": [ + { + "bbox": [ + 318, + 146, + 332, + 153 + ], + "score": 0.67, + "content": "\\mathrm { ~ x ~ } =", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 145, + 417, + 153 + ], + "score": 1.0, + "content": "proj(x, d_model, axis", + "type": "text" + }, + { + "bbox": [ + 417, + 146, + 425, + 152 + ], + "score": 0.44, + "content": "= ^ { 1 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 145, + 462, + 153 + ], + "score": 1.0, + "content": "channel\")", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 317, + 153, + 396, + 162 + ], + "spans": [ + { + "bbox": [ + 317, + 153, + 345, + 162 + ], + "score": 1.0, + "content": "return", + "type": "text" + }, + { + "bbox": [ + 345, + 154, + 360, + 161 + ], + "score": 0.4, + "content": "\\texttt { x + }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 153, + 396, + 162 + ], + "score": 1.0, + "content": "shortcut", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 14.5 + }, + { + "type": "image", + "bbox": [ + 113, + 77, + 298, + 225 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 113, + 77, + 298, + 225 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 77, + 298, + 225 + ], + "spans": [ + { + "bbox": [ + 113, + 77, + 298, + 225 + ], + "score": 0.963, + "type": "image", + "image_path": "018d82d7640366b577f21adb6e4e6c60f398a8eb4976b569ff0a2114a750edde.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 113, + 77, + 298, + 90.45454545454545 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 113, + 90.45454545454545, + 298, + 103.9090909090909 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 113, + 103.9090909090909, + 298, + 117.36363636363636 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 113, + 117.36363636363636, + 298, + 130.8181818181818 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 113, + 130.8181818181818, + 298, + 144.27272727272728 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 113, + 144.27272727272728, + 298, + 157.72727272727275 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 113, + 157.72727272727275, + 298, + 171.18181818181822 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 113, + 171.18181818181822, + 298, + 184.63636363636368 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 113, + 184.63636363636368, + 298, + 198.09090909090915 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 113, + 198.09090909090915, + 298, + 211.54545454545462 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 113, + 211.54545454545462, + 298, + 225.00000000000009 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 236, + 506, + 292 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 236, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 236, + 213, + 249 + ], + "score": 1.0, + "content": "Figure 1: Overview of the", + "type": "text" + }, + { + "bbox": [ + 214, + 237, + 241, + 248 + ], + "score": 0.49, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 236, + 505, + 249 + ], + "score": 1.0, + "content": "architecture with Spatial Gating Unit (SGU). The model consists", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 247, + 506, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 247, + 158, + 261 + ], + "score": 1.0, + "content": "of a stack of", + "type": "text" + }, + { + "bbox": [ + 158, + 248, + 166, + 258 + ], + "score": 0.79, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 247, + 492, + 261 + ], + "score": 1.0, + "content": "blocks with identical structure and size. All projection operations are linear and “", + "type": "text" + }, + { + "bbox": [ + 492, + 248, + 501, + 258 + ], + "score": 0.49, + "content": "\\odot", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 247, + 506, + 261 + ], + "score": 1.0, + "content": "”", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 259, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 506, + 272 + ], + "score": 1.0, + "content": "refers to element-wise multiplication (linear gating). The input and output protocols follow BERT for", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 268, + 506, + 283 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 506, + 283 + ], + "score": 1.0, + "content": "NLP and ViT for vision. Unlike Transformers, gMLPs do not require positional encodings, nor is it", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 280, + 344, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 344, + 294 + ], + "score": 1.0, + "content": "necessary to mask out the paddings during NLP finetuning.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27 + } + ], + "index": 16.0 + }, + { + "type": "text", + "bbox": [ + 313, + 170, + 493, + 222 + ], + "lines": [ + { + "bbox": [ + 313, + 169, + 423, + 179 + ], + "spans": [ + { + "bbox": [ + 313, + 169, + 407, + 179 + ], + "score": 1.0, + "content": "def spatial_gating_unit", + "type": "text" + }, + { + "bbox": [ + 407, + 170, + 418, + 178 + ], + "score": 0.45, + "content": "\\mathbf { \\Psi } ( \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 418, + 169, + 423, + 179 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 320, + 178, + 446, + 187 + ], + "spans": [ + { + "bbox": [ + 320, + 178, + 334, + 187 + ], + "score": 1.0, + "content": "u,", + "type": "text" + }, + { + "bbox": [ + 334, + 180, + 348, + 186 + ], + "score": 0.27, + "content": "\\tt { v } =", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 178, + 446, + 187 + ], + "score": 1.0, + "content": "split(x, axis=\"channel\")", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 321, + 187, + 430, + 195 + ], + "spans": [ + { + "bbox": [ + 321, + 188, + 335, + 194 + ], + "score": 0.47, + "content": "\\tt { v } =", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 187, + 430, + 195 + ], + "score": 1.0, + "content": "norm(v, axis=\"channel\")", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 320, + 196, + 441, + 204 + ], + "spans": [ + { + "bbox": [ + 320, + 196, + 397, + 204 + ], + "score": 1.0, + "content": "n = get_dim(v, axis", + "type": "text" + }, + { + "bbox": [ + 397, + 196, + 404, + 203 + ], + "score": 0.43, + "content": "\\mathrel { \\mathop : } = \\mathrel { \\mathop : }", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 196, + 441, + 204 + ], + "score": 1.0, + "content": "spatial\")", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 320, + 204, + 493, + 213 + ], + "spans": [ + { + "bbox": [ + 320, + 204, + 480, + 213 + ], + "score": 1.0, + "content": "v = proj(v, n, axis=\"spatial\", init_bias", + "type": "text" + }, + { + "bbox": [ + 480, + 204, + 489, + 211 + ], + "score": 0.47, + "content": "\\mathrel { \\mathop : } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 490, + 204, + 493, + 213 + ], + "score": 1.0, + "content": ")", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 321, + 213, + 372, + 221 + ], + "spans": [ + { + "bbox": [ + 321, + 213, + 372, + 221 + ], + "score": 1.0, + "content": "return u ∗ v", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 106, + 311, + 505, + 379 + ], + "lines": [ + { + "bbox": [ + 105, + 312, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 312, + 505, + 325 + ], + "score": 1.0, + "content": "both pretraining and finetuning metrics for gMLPs improve as quickly as for Transformers. This", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 323, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 323, + 244, + 335 + ], + "score": 1.0, + "content": "is remarkable because it indicates", + "type": "text" + }, + { + "bbox": [ + 244, + 323, + 275, + 335 + ], + "score": 0.55, + "content": "{ \\mathrm { g M L P s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 323, + 505, + 335 + ], + "score": 1.0, + "content": "scale just as well as Transformers despite the absence of", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 334, + 505, + 347 + ], + "spans": [ + { + "bbox": [ + 106, + 334, + 505, + 347 + ], + "score": 1.0, + "content": "self-attention, and any performance gap can always be offset by training a larger model with increased", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 344, + 506, + 358 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 309, + 358 + ], + "score": 1.0, + "content": "data and compute. With a standard 256-batch size", + "type": "text" + }, + { + "bbox": [ + 309, + 345, + 335, + 356 + ], + "score": 0.62, + "content": "\\times \\ 1 \\mathbf { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 344, + 506, + 358 + ], + "score": 1.0, + "content": "-step training setup as in original BERT, a", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 356, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 128, + 369 + ], + "score": 1.0, + "content": "large", + "type": "text" + }, + { + "bbox": [ + 129, + 356, + 156, + 367 + ], + "score": 0.25, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 356, + 221, + 369 + ], + "score": 1.0, + "content": "model achieves", + "type": "text" + }, + { + "bbox": [ + 221, + 356, + 249, + 366 + ], + "score": 0.85, + "content": "8 7 . 7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 356, + 345, + 369 + ], + "score": 1.0, + "content": "accuracy on MNLI and", + "type": "text" + }, + { + "bbox": [ + 345, + 356, + 372, + 366 + ], + "score": 0.83, + "content": "8 2 . 1 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 356, + 505, + 369 + ], + "score": 1.0, + "content": "F1 on SQuAD v2.0. Note, these", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 367, + 482, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 367, + 180, + 380 + ], + "score": 1.0, + "content": "are better than the", + "type": "text" + }, + { + "bbox": [ + 181, + 367, + 221, + 379 + ], + "score": 0.46, + "content": "\\mathbf { B E R T _ { l a r g e } }", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 367, + 482, + 380 + ], + "score": 1.0, + "content": "results reported in Devlin et al. [2] obtained using Transformers.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 106, + 383, + 505, + 461 + ], + "lines": [ + { + "bbox": [ + 105, + 383, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 383, + 505, + 396 + ], + "score": 1.0, + "content": "For BERT’s finetuning, Transformers can be more practically advantageous over gMLPs on tasks", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 393, + 506, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 293, + 407 + ], + "score": 1.0, + "content": "that require cross-sentence alignment (e.g., by", + "type": "text" + }, + { + "bbox": [ + 293, + 394, + 316, + 405 + ], + "score": 0.85, + "content": "0 . 8 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 393, + 506, + 407 + ], + "score": 1.0, + "content": "on MNLI-m in the 300M-param regime), even", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 405, + 505, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 505, + 419 + ], + "score": 1.0, + "content": "with similar pretraining perplexity. This problem can be addressed by making gMLPs substantially", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 416, + 506, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 140, + 428 + ], + "score": 1.0, + "content": "larger—", + "type": "text" + }, + { + "bbox": [ + 140, + 416, + 154, + 426 + ], + "score": 0.84, + "content": "3 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 416, + 506, + 428 + ], + "score": 1.0, + "content": "as large as Transformers. A more practical solution is to blend in only a tiny bit of self-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 426, + 505, + 440 + ], + "spans": [ + { + "bbox": [ + 104, + 426, + 425, + 440 + ], + "score": 1.0, + "content": "attention—a single-head self-attention with size up to 128 is sufficient to make", + "type": "text" + }, + { + "bbox": [ + 425, + 427, + 456, + 438 + ], + "score": 0.5, + "content": "\\mathrm { g M L P s }", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 426, + 505, + 440 + ], + "score": 1.0, + "content": "outperform", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 438, + 504, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 504, + 450 + ], + "score": 1.0, + "content": "Transformers on all NLP tasks we evaluated with even better parameter efficiency. The improvement", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 104, + 447, + 411, + 462 + ], + "spans": [ + { + "bbox": [ + 104, + 447, + 247, + 462 + ], + "score": 1.0, + "content": "is sometimes very significant (e.g.,", + "type": "text" + }, + { + "bbox": [ + 247, + 449, + 275, + 459 + ], + "score": 0.89, + "content": "+ 4 . 4 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 447, + 323, + 462 + ], + "score": 1.0, + "content": "on SQuAD", + "type": "text" + }, + { + "bbox": [ + 323, + 449, + 342, + 460 + ], + "score": 0.33, + "content": "\\mathrm { v } 2 . 0", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 447, + 363, + 462 + ], + "score": 1.0, + "content": "over", + "type": "text" + }, + { + "bbox": [ + 363, + 449, + 404, + 461 + ], + "score": 0.7, + "content": "\\mathbf { B E R T _ { l a r g e } }", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 447, + 411, + 462 + ], + "score": 1.0, + "content": ").", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 39 + }, + { + "type": "text", + "bbox": [ + 107, + 464, + 505, + 520 + ], + "lines": [ + { + "bbox": [ + 106, + 465, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 506, + 477 + ], + "score": 1.0, + "content": "Overall, the surprising effectiveness of gMLPs in both vision and NLP domains suggests that self-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 475, + 506, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 506, + 489 + ], + "score": 1.0, + "content": "attention is not a necessary ingredient for scaling up machine learning models, although it can be a", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 487, + 505, + 499 + ], + "spans": [ + { + "bbox": [ + 106, + 487, + 505, + 499 + ], + "score": 1.0, + "content": "useful addition depending on the task. With increased data and compute, models with simpler spatial", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 497, + 506, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 232, + 511 + ], + "score": 1.0, + "content": "interaction mechanisms such as", + "type": "text" + }, + { + "bbox": [ + 232, + 498, + 259, + 509 + ], + "score": 0.5, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 497, + 506, + 511 + ], + "score": 1.0, + "content": "can be as powerful as Transformers and the capacity allocated", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 508, + 362, + 521 + ], + "spans": [ + { + "bbox": [ + 106, + 508, + 362, + 521 + ], + "score": 1.0, + "content": "to self-attention can be either removed or substantially reduced.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 45 + }, + { + "type": "title", + "bbox": [ + 107, + 534, + 159, + 548 + ], + "lines": [ + { + "bbox": [ + 104, + 533, + 160, + 551 + ], + "spans": [ + { + "bbox": [ + 104, + 533, + 160, + 551 + ], + "score": 1.0, + "content": "2 Model", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 48 + }, + { + "type": "text", + "bbox": [ + 106, + 559, + 504, + 582 + ], + "lines": [ + { + "bbox": [ + 105, + 558, + 504, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 558, + 271, + 572 + ], + "score": 1.0, + "content": "Our model, gMLP, consists of a stack of", + "type": "text" + }, + { + "bbox": [ + 271, + 561, + 279, + 570 + ], + "score": 0.83, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 558, + 458, + 572 + ], + "score": 1.0, + "content": "blocks with identical size and structure. Let", + "type": "text" + }, + { + "bbox": [ + 459, + 559, + 504, + 570 + ], + "score": 0.91, + "content": "\\ b { X } \\in \\mathbb { R } ^ { n \\times d }", + "type": "inline_equation" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 570, + 488, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 307, + 583 + ], + "score": 1.0, + "content": "be the token representations with sequence length", + "type": "text" + }, + { + "bbox": [ + 307, + 573, + 314, + 581 + ], + "score": 0.8, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 570, + 376, + 583 + ], + "score": 1.0, + "content": "and dimension", + "type": "text" + }, + { + "bbox": [ + 377, + 572, + 383, + 581 + ], + "score": 0.78, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 570, + 488, + 583 + ], + "score": 1.0, + "content": ". Each block is defined as:", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 49.5 + }, + { + "type": "interline_equation", + "bbox": [ + 215, + 587, + 396, + 602 + ], + "lines": [ + { + "bbox": [ + 215, + 587, + 396, + 602 + ], + "spans": [ + { + "bbox": [ + 215, + 587, + 396, + 602 + ], + "score": 0.92, + "content": "Z = \\sigma ( X U ) , \\qquad \\tilde { Z } = s ( Z ) , \\qquad Y = \\tilde { Z } V", + "type": "interline_equation", + "image_path": "c7f7fd02ef54ad58deab3b63dffb850ea35636c7e3bb6788f9e24746d918a055.jpg" + } + ] + } + ], + "index": 51, + "virtual_lines": [ + { + "bbox": [ + 215, + 587, + 396, + 602 + ], + "spans": [], + "index": 51 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 606, + 505, + 641 + ], + "lines": [ + { + "bbox": [ + 105, + 606, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 133, + 620 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 609, + 141, + 617 + ], + "score": 0.76, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 606, + 325, + 620 + ], + "score": 1.0, + "content": "is an activation function such as GeLU [24].", + "type": "text" + }, + { + "bbox": [ + 325, + 607, + 335, + 617 + ], + "score": 0.78, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 606, + 353, + 620 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 353, + 607, + 363, + 617 + ], + "score": 0.79, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 606, + 505, + 620 + ], + "score": 1.0, + "content": "define linear projections along the", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 617, + 504, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 458, + 630 + ], + "score": 1.0, + "content": "channel dimension—the same as those in the FFNs of Transformers (e.g., their shapes are", + "type": "text" + }, + { + "bbox": [ + 458, + 618, + 504, + 628 + ], + "score": 0.88, + "content": "7 6 8 \\times 3 0 7 2", + "type": "inline_equation" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 628, + 474, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 123, + 641 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 629, + 170, + 640 + ], + "score": 0.86, + "content": "3 0 7 2 \\times 7 6 8", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 628, + 185, + 641 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 185, + 629, + 224, + 640 + ], + "score": 0.7, + "content": "\\mathbf { B E R T _ { b a s e } }", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 628, + 474, + 641 + ], + "score": 1.0, + "content": "). Shortcuts, normalizations and biases are omitted for brevity.", + "type": "text" + } + ], + "index": 54 + } + ], + "index": 53 + }, + { + "type": "text", + "bbox": [ + 106, + 645, + 505, + 723 + ], + "lines": [ + { + "bbox": [ + 104, + 644, + 506, + 659 + ], + "spans": [ + { + "bbox": [ + 104, + 644, + 319, + 659 + ], + "score": 1.0, + "content": "A key ingredient in the aforementioned formulation is", + "type": "text" + }, + { + "bbox": [ + 320, + 645, + 335, + 657 + ], + "score": 0.9, + "content": "s ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 644, + 506, + 659 + ], + "score": 1.0, + "content": ", a layer which captures spatial interactions", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 105, + 655, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 183, + 669 + ], + "score": 1.0, + "content": "(see below). When", + "type": "text" + }, + { + "bbox": [ + 184, + 658, + 190, + 666 + ], + "score": 0.66, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 655, + 506, + 669 + ], + "score": 1.0, + "content": "is an identity mapping, the above transformation degenerates to a regular FFN,", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 104, + 666, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 104, + 666, + 506, + 680 + ], + "score": 1.0, + "content": "where individual tokens are processed independently without any cross-token communication. One", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 106, + 678, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 303, + 691 + ], + "score": 1.0, + "content": "of our major focuses is therefore to design a good", + "type": "text" + }, + { + "bbox": [ + 304, + 681, + 309, + 688 + ], + "score": 0.69, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 678, + 506, + 691 + ], + "score": 1.0, + "content": "capable of capturing complex spatial interactions", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 105, + 688, + 505, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 488, + 702 + ], + "score": 1.0, + "content": "across tokens. The overall block layout is inspired by inverted bottlenecks [25] which define", + "type": "text" + }, + { + "bbox": [ + 489, + 689, + 505, + 701 + ], + "score": 0.88, + "content": "s ( \\cdot )", + "type": "inline_equation" + } + ], + "index": 59 + }, + { + "bbox": [ + 105, + 700, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 505, + 713 + ], + "score": 1.0, + "content": "as a spatial depthwise convolution. Note, unlike Transformers, our model does not require position", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 105, + 711, + 359, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 338, + 724 + ], + "score": 1.0, + "content": "embeddings because such information will be captured in", + "type": "text" + }, + { + "bbox": [ + 338, + 712, + 354, + 723 + ], + "score": 0.92, + "content": "s ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 711, + 359, + 724 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 61 + } + ], + "index": 58 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 339, + 79, + 472, + 89 + ], + "lines": [ + { + "bbox": [ + 338, + 78, + 473, + 91 + ], + "spans": [ + { + "bbox": [ + 338, + 78, + 473, + 91 + ], + "score": 1.0, + "content": "Pseudo-code for the gMLP block", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 753 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 753 + ], + "score": 1.0, + "content": "2", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 311, + 94, + 461, + 161 + ], + "lines": [ + { + "bbox": [ + 311, + 93, + 447, + 103 + ], + "spans": [ + { + "bbox": [ + 311, + 93, + 447, + 103 + ], + "score": 1.0, + "content": "def gmlp_block(x, d_model, d_ffn):", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 318, + 102, + 369, + 111 + ], + "spans": [ + { + "bbox": [ + 318, + 102, + 352, + 111 + ], + "score": 1.0, + "content": "shortcut", + "type": "text" + }, + { + "bbox": [ + 352, + 104, + 362, + 110 + ], + "score": 0.33, + "content": "=", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 102, + 369, + 111 + ], + "score": 1.0, + "content": "x", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 318, + 111, + 426, + 119 + ], + "spans": [ + { + "bbox": [ + 318, + 111, + 332, + 119 + ], + "score": 0.58, + "content": "\\mathrm { ~ x ~ } =", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 111, + 426, + 119 + ], + "score": 1.0, + "content": "norm(x, axis=\"channel\")", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 318, + 119, + 454, + 128 + ], + "spans": [ + { + "bbox": [ + 318, + 120, + 332, + 127 + ], + "score": 0.66, + "content": "\\mathrm { ~ x ~ } =", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 119, + 454, + 128 + ], + "score": 1.0, + "content": "proj(x, d_ffn, axis=\"channel\")", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 318, + 128, + 363, + 137 + ], + "spans": [ + { + "bbox": [ + 318, + 129, + 332, + 136 + ], + "score": 0.5, + "content": "\\mathrm { ~ x ~ } =", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 128, + 363, + 137 + ], + "score": 1.0, + "content": "gelu(x)", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 318, + 136, + 422, + 145 + ], + "spans": [ + { + "bbox": [ + 318, + 137, + 333, + 144 + ], + "score": 0.77, + "content": "\\mathrm { ~ x ~ } =", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 136, + 422, + 145 + ], + "score": 1.0, + "content": "spatial_gating_unit(x)", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 318, + 145, + 462, + 153 + ], + "spans": [ + { + "bbox": [ + 318, + 146, + 332, + 153 + ], + "score": 0.67, + "content": "\\mathrm { ~ x ~ } =", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 145, + 417, + 153 + ], + "score": 1.0, + "content": "proj(x, d_model, axis", + "type": "text" + }, + { + "bbox": [ + 417, + 146, + 425, + 152 + ], + "score": 0.44, + "content": "= ^ { 1 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 145, + 462, + 153 + ], + "score": 1.0, + "content": "channel\")", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 317, + 153, + 396, + 162 + ], + "spans": [ + { + "bbox": [ + 317, + 153, + 345, + 162 + ], + "score": 1.0, + "content": "return", + "type": "text" + }, + { + "bbox": [ + 345, + 154, + 360, + 161 + ], + "score": 0.4, + "content": "\\texttt { x + }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 153, + 396, + 162 + ], + "score": 1.0, + "content": "shortcut", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 14.5, + "bbox_fs": [ + 311, + 93, + 462, + 162 + ] + }, + { + "type": "image", + "bbox": [ + 113, + 77, + 298, + 225 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 113, + 77, + 298, + 225 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 113, + 77, + 298, + 225 + ], + "spans": [ + { + "bbox": [ + 113, + 77, + 298, + 225 + ], + "score": 0.963, + "type": "image", + "image_path": "018d82d7640366b577f21adb6e4e6c60f398a8eb4976b569ff0a2114a750edde.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 113, + 77, + 298, + 90.45454545454545 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 113, + 90.45454545454545, + 298, + 103.9090909090909 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 113, + 103.9090909090909, + 298, + 117.36363636363636 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 113, + 117.36363636363636, + 298, + 130.8181818181818 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 113, + 130.8181818181818, + 298, + 144.27272727272728 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 113, + 144.27272727272728, + 298, + 157.72727272727275 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 113, + 157.72727272727275, + 298, + 171.18181818181822 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 113, + 171.18181818181822, + 298, + 184.63636363636368 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 113, + 184.63636363636368, + 298, + 198.09090909090915 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 113, + 198.09090909090915, + 298, + 211.54545454545462 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 113, + 211.54545454545462, + 298, + 225.00000000000009 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 236, + 506, + 292 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 236, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 106, + 236, + 213, + 249 + ], + "score": 1.0, + "content": "Figure 1: Overview of the", + "type": "text" + }, + { + "bbox": [ + 214, + 237, + 241, + 248 + ], + "score": 0.49, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 236, + 505, + 249 + ], + "score": 1.0, + "content": "architecture with Spatial Gating Unit (SGU). The model consists", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 247, + 506, + 261 + ], + "spans": [ + { + "bbox": [ + 105, + 247, + 158, + 261 + ], + "score": 1.0, + "content": "of a stack of", + "type": "text" + }, + { + "bbox": [ + 158, + 248, + 166, + 258 + ], + "score": 0.79, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 247, + 492, + 261 + ], + "score": 1.0, + "content": "blocks with identical structure and size. All projection operations are linear and “", + "type": "text" + }, + { + "bbox": [ + 492, + 248, + 501, + 258 + ], + "score": 0.49, + "content": "\\odot", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 247, + 506, + 261 + ], + "score": 1.0, + "content": "”", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 259, + 506, + 272 + ], + "spans": [ + { + "bbox": [ + 105, + 259, + 506, + 272 + ], + "score": 1.0, + "content": "refers to element-wise multiplication (linear gating). The input and output protocols follow BERT for", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 268, + 506, + 283 + ], + "spans": [ + { + "bbox": [ + 105, + 268, + 506, + 283 + ], + "score": 1.0, + "content": "NLP and ViT for vision. Unlike Transformers, gMLPs do not require positional encodings, nor is it", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 280, + 344, + 294 + ], + "spans": [ + { + "bbox": [ + 105, + 280, + 344, + 294 + ], + "score": 1.0, + "content": "necessary to mask out the paddings during NLP finetuning.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27 + } + ], + "index": 16.0 + }, + { + "type": "text", + "bbox": [ + 313, + 170, + 493, + 222 + ], + "lines": [ + { + "bbox": [ + 313, + 169, + 423, + 179 + ], + "spans": [ + { + "bbox": [ + 313, + 169, + 407, + 179 + ], + "score": 1.0, + "content": "def spatial_gating_unit", + "type": "text" + }, + { + "bbox": [ + 407, + 170, + 418, + 178 + ], + "score": 0.45, + "content": "\\mathbf { \\Psi } ( \\mathbf { x } )", + "type": "inline_equation" + }, + { + "bbox": [ + 418, + 169, + 423, + 179 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 320, + 178, + 446, + 187 + ], + "spans": [ + { + "bbox": [ + 320, + 178, + 334, + 187 + ], + "score": 1.0, + "content": "u,", + "type": "text" + }, + { + "bbox": [ + 334, + 180, + 348, + 186 + ], + "score": 0.27, + "content": "\\tt { v } =", + "type": "inline_equation" + }, + { + "bbox": [ + 348, + 178, + 446, + 187 + ], + "score": 1.0, + "content": "split(x, axis=\"channel\")", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 321, + 187, + 430, + 195 + ], + "spans": [ + { + "bbox": [ + 321, + 188, + 335, + 194 + ], + "score": 0.47, + "content": "\\tt { v } =", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 187, + 430, + 195 + ], + "score": 1.0, + "content": "norm(v, axis=\"channel\")", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 320, + 196, + 441, + 204 + ], + "spans": [ + { + "bbox": [ + 320, + 196, + 397, + 204 + ], + "score": 1.0, + "content": "n = get_dim(v, axis", + "type": "text" + }, + { + "bbox": [ + 397, + 196, + 404, + 203 + ], + "score": 0.43, + "content": "\\mathrel { \\mathop : } = \\mathrel { \\mathop : }", + "type": "inline_equation" + }, + { + "bbox": [ + 405, + 196, + 441, + 204 + ], + "score": 1.0, + "content": "spatial\")", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 320, + 204, + 493, + 213 + ], + "spans": [ + { + "bbox": [ + 320, + 204, + 480, + 213 + ], + "score": 1.0, + "content": "v = proj(v, n, axis=\"spatial\", init_bias", + "type": "text" + }, + { + "bbox": [ + 480, + 204, + 489, + 211 + ], + "score": 0.47, + "content": "\\mathrel { \\mathop : } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 490, + 204, + 493, + 213 + ], + "score": 1.0, + "content": ")", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 321, + 213, + 372, + 221 + ], + "spans": [ + { + "bbox": [ + 321, + 213, + 372, + 221 + ], + "score": 1.0, + "content": "return u ∗ v", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21.5, + "bbox_fs": [ + 313, + 169, + 493, + 221 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 311, + 505, + 379 + ], + "lines": [ + { + "bbox": [ + 105, + 312, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 105, + 312, + 505, + 325 + ], + "score": 1.0, + "content": "both pretraining and finetuning metrics for gMLPs improve as quickly as for Transformers. This", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 323, + 505, + 335 + ], + "spans": [ + { + "bbox": [ + 106, + 323, + 244, + 335 + ], + "score": 1.0, + "content": "is remarkable because it indicates", + "type": "text" + }, + { + "bbox": [ + 244, + 323, + 275, + 335 + ], + "score": 0.55, + "content": "{ \\mathrm { g M L P s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 323, + 505, + 335 + ], + "score": 1.0, + "content": "scale just as well as Transformers despite the absence of", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 334, + 505, + 347 + ], + "spans": [ + { + "bbox": [ + 106, + 334, + 505, + 347 + ], + "score": 1.0, + "content": "self-attention, and any performance gap can always be offset by training a larger model with increased", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 344, + 506, + 358 + ], + "spans": [ + { + "bbox": [ + 105, + 344, + 309, + 358 + ], + "score": 1.0, + "content": "data and compute. With a standard 256-batch size", + "type": "text" + }, + { + "bbox": [ + 309, + 345, + 335, + 356 + ], + "score": 0.62, + "content": "\\times \\ 1 \\mathbf { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 344, + 506, + 358 + ], + "score": 1.0, + "content": "-step training setup as in original BERT, a", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 356, + 505, + 369 + ], + "spans": [ + { + "bbox": [ + 105, + 356, + 128, + 369 + ], + "score": 1.0, + "content": "large", + "type": "text" + }, + { + "bbox": [ + 129, + 356, + 156, + 367 + ], + "score": 0.25, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 356, + 221, + 369 + ], + "score": 1.0, + "content": "model achieves", + "type": "text" + }, + { + "bbox": [ + 221, + 356, + 249, + 366 + ], + "score": 0.85, + "content": "8 7 . 7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 356, + 345, + 369 + ], + "score": 1.0, + "content": "accuracy on MNLI and", + "type": "text" + }, + { + "bbox": [ + 345, + 356, + 372, + 366 + ], + "score": 0.83, + "content": "8 2 . 1 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 373, + 356, + 505, + 369 + ], + "score": 1.0, + "content": "F1 on SQuAD v2.0. Note, these", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 367, + 482, + 380 + ], + "spans": [ + { + "bbox": [ + 106, + 367, + 180, + 380 + ], + "score": 1.0, + "content": "are better than the", + "type": "text" + }, + { + "bbox": [ + 181, + 367, + 221, + 379 + ], + "score": 0.46, + "content": "\\mathbf { B E R T _ { l a r g e } }", + "type": "inline_equation" + }, + { + "bbox": [ + 221, + 367, + 482, + 380 + ], + "score": 1.0, + "content": "results reported in Devlin et al. [2] obtained using Transformers.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 312, + 506, + 380 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 383, + 505, + 461 + ], + "lines": [ + { + "bbox": [ + 105, + 383, + 505, + 396 + ], + "spans": [ + { + "bbox": [ + 105, + 383, + 505, + 396 + ], + "score": 1.0, + "content": "For BERT’s finetuning, Transformers can be more practically advantageous over gMLPs on tasks", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 105, + 393, + 506, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 393, + 293, + 407 + ], + "score": 1.0, + "content": "that require cross-sentence alignment (e.g., by", + "type": "text" + }, + { + "bbox": [ + 293, + 394, + 316, + 405 + ], + "score": 0.85, + "content": "0 . 8 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 393, + 506, + 407 + ], + "score": 1.0, + "content": "on MNLI-m in the 300M-param regime), even", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 405, + 505, + 419 + ], + "spans": [ + { + "bbox": [ + 105, + 405, + 505, + 419 + ], + "score": 1.0, + "content": "with similar pretraining perplexity. This problem can be addressed by making gMLPs substantially", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 416, + 506, + 428 + ], + "spans": [ + { + "bbox": [ + 105, + 416, + 140, + 428 + ], + "score": 1.0, + "content": "larger—", + "type": "text" + }, + { + "bbox": [ + 140, + 416, + 154, + 426 + ], + "score": 0.84, + "content": "3 \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 416, + 506, + 428 + ], + "score": 1.0, + "content": "as large as Transformers. A more practical solution is to blend in only a tiny bit of self-", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 426, + 505, + 440 + ], + "spans": [ + { + "bbox": [ + 104, + 426, + 425, + 440 + ], + "score": 1.0, + "content": "attention—a single-head self-attention with size up to 128 is sufficient to make", + "type": "text" + }, + { + "bbox": [ + 425, + 427, + 456, + 438 + ], + "score": 0.5, + "content": "\\mathrm { g M L P s }", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 426, + 505, + 440 + ], + "score": 1.0, + "content": "outperform", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 438, + 504, + 450 + ], + "spans": [ + { + "bbox": [ + 105, + 438, + 504, + 450 + ], + "score": 1.0, + "content": "Transformers on all NLP tasks we evaluated with even better parameter efficiency. The improvement", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 104, + 447, + 411, + 462 + ], + "spans": [ + { + "bbox": [ + 104, + 447, + 247, + 462 + ], + "score": 1.0, + "content": "is sometimes very significant (e.g.,", + "type": "text" + }, + { + "bbox": [ + 247, + 449, + 275, + 459 + ], + "score": 0.89, + "content": "+ 4 . 4 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 447, + 323, + 462 + ], + "score": 1.0, + "content": "on SQuAD", + "type": "text" + }, + { + "bbox": [ + 323, + 449, + 342, + 460 + ], + "score": 0.33, + "content": "\\mathrm { v } 2 . 0", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 447, + 363, + 462 + ], + "score": 1.0, + "content": "over", + "type": "text" + }, + { + "bbox": [ + 363, + 449, + 404, + 461 + ], + "score": 0.7, + "content": "\\mathbf { B E R T _ { l a r g e } }", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 447, + 411, + 462 + ], + "score": 1.0, + "content": ").", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 39, + "bbox_fs": [ + 104, + 383, + 506, + 462 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 464, + 505, + 520 + ], + "lines": [ + { + "bbox": [ + 106, + 465, + 506, + 477 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 506, + 477 + ], + "score": 1.0, + "content": "Overall, the surprising effectiveness of gMLPs in both vision and NLP domains suggests that self-", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 105, + 475, + 506, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 506, + 489 + ], + "score": 1.0, + "content": "attention is not a necessary ingredient for scaling up machine learning models, although it can be a", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 487, + 505, + 499 + ], + "spans": [ + { + "bbox": [ + 106, + 487, + 505, + 499 + ], + "score": 1.0, + "content": "useful addition depending on the task. With increased data and compute, models with simpler spatial", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 497, + 506, + 511 + ], + "spans": [ + { + "bbox": [ + 105, + 497, + 232, + 511 + ], + "score": 1.0, + "content": "interaction mechanisms such as", + "type": "text" + }, + { + "bbox": [ + 232, + 498, + 259, + 509 + ], + "score": 0.5, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 259, + 497, + 506, + 511 + ], + "score": 1.0, + "content": "can be as powerful as Transformers and the capacity allocated", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 106, + 508, + 362, + 521 + ], + "spans": [ + { + "bbox": [ + 106, + 508, + 362, + 521 + ], + "score": 1.0, + "content": "to self-attention can be either removed or substantially reduced.", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 45, + "bbox_fs": [ + 105, + 465, + 506, + 521 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 534, + 159, + 548 + ], + "lines": [ + { + "bbox": [ + 104, + 533, + 160, + 551 + ], + "spans": [ + { + "bbox": [ + 104, + 533, + 160, + 551 + ], + "score": 1.0, + "content": "2 Model", + "type": "text" + } + ], + "index": 48 + } + ], + "index": 48 + }, + { + "type": "text", + "bbox": [ + 106, + 559, + 504, + 582 + ], + "lines": [ + { + "bbox": [ + 105, + 558, + 504, + 572 + ], + "spans": [ + { + "bbox": [ + 105, + 558, + 271, + 572 + ], + "score": 1.0, + "content": "Our model, gMLP, consists of a stack of", + "type": "text" + }, + { + "bbox": [ + 271, + 561, + 279, + 570 + ], + "score": 0.83, + "content": "L", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 558, + 458, + 572 + ], + "score": 1.0, + "content": "blocks with identical size and structure. Let", + "type": "text" + }, + { + "bbox": [ + 459, + 559, + 504, + 570 + ], + "score": 0.91, + "content": "\\ b { X } \\in \\mathbb { R } ^ { n \\times d }", + "type": "inline_equation" + } + ], + "index": 49 + }, + { + "bbox": [ + 105, + 570, + 488, + 583 + ], + "spans": [ + { + "bbox": [ + 105, + 570, + 307, + 583 + ], + "score": 1.0, + "content": "be the token representations with sequence length", + "type": "text" + }, + { + "bbox": [ + 307, + 573, + 314, + 581 + ], + "score": 0.8, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 315, + 570, + 376, + 583 + ], + "score": 1.0, + "content": "and dimension", + "type": "text" + }, + { + "bbox": [ + 377, + 572, + 383, + 581 + ], + "score": 0.78, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 570, + 488, + 583 + ], + "score": 1.0, + "content": ". Each block is defined as:", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 49.5, + "bbox_fs": [ + 105, + 558, + 504, + 583 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 215, + 587, + 396, + 602 + ], + "lines": [ + { + "bbox": [ + 215, + 587, + 396, + 602 + ], + "spans": [ + { + "bbox": [ + 215, + 587, + 396, + 602 + ], + "score": 0.92, + "content": "Z = \\sigma ( X U ) , \\qquad \\tilde { Z } = s ( Z ) , \\qquad Y = \\tilde { Z } V", + "type": "interline_equation", + "image_path": "c7f7fd02ef54ad58deab3b63dffb850ea35636c7e3bb6788f9e24746d918a055.jpg" + } + ] + } + ], + "index": 51, + "virtual_lines": [ + { + "bbox": [ + 215, + 587, + 396, + 602 + ], + "spans": [], + "index": 51 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 606, + 505, + 641 + ], + "lines": [ + { + "bbox": [ + 105, + 606, + 505, + 620 + ], + "spans": [ + { + "bbox": [ + 105, + 606, + 133, + 620 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 609, + 141, + 617 + ], + "score": 0.76, + "content": "\\sigma", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 606, + 325, + 620 + ], + "score": 1.0, + "content": "is an activation function such as GeLU [24].", + "type": "text" + }, + { + "bbox": [ + 325, + 607, + 335, + 617 + ], + "score": 0.78, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 335, + 606, + 353, + 620 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 353, + 607, + 363, + 617 + ], + "score": 0.79, + "content": "V", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 606, + 505, + 620 + ], + "score": 1.0, + "content": "define linear projections along the", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 105, + 617, + 504, + 630 + ], + "spans": [ + { + "bbox": [ + 105, + 617, + 458, + 630 + ], + "score": 1.0, + "content": "channel dimension—the same as those in the FFNs of Transformers (e.g., their shapes are", + "type": "text" + }, + { + "bbox": [ + 458, + 618, + 504, + 628 + ], + "score": 0.88, + "content": "7 6 8 \\times 3 0 7 2", + "type": "inline_equation" + } + ], + "index": 53 + }, + { + "bbox": [ + 105, + 628, + 474, + 641 + ], + "spans": [ + { + "bbox": [ + 105, + 628, + 123, + 641 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 123, + 629, + 170, + 640 + ], + "score": 0.86, + "content": "3 0 7 2 \\times 7 6 8", + "type": "inline_equation" + }, + { + "bbox": [ + 171, + 628, + 185, + 641 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 185, + 629, + 224, + 640 + ], + "score": 0.7, + "content": "\\mathbf { B E R T _ { b a s e } }", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 628, + 474, + 641 + ], + "score": 1.0, + "content": "). Shortcuts, normalizations and biases are omitted for brevity.", + "type": "text" + } + ], + "index": 54 + } + ], + "index": 53, + "bbox_fs": [ + 105, + 606, + 505, + 641 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 645, + 505, + 723 + ], + "lines": [ + { + "bbox": [ + 104, + 644, + 506, + 659 + ], + "spans": [ + { + "bbox": [ + 104, + 644, + 319, + 659 + ], + "score": 1.0, + "content": "A key ingredient in the aforementioned formulation is", + "type": "text" + }, + { + "bbox": [ + 320, + 645, + 335, + 657 + ], + "score": 0.9, + "content": "s ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 336, + 644, + 506, + 659 + ], + "score": 1.0, + "content": ", a layer which captures spatial interactions", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 105, + 655, + 506, + 669 + ], + "spans": [ + { + "bbox": [ + 105, + 655, + 183, + 669 + ], + "score": 1.0, + "content": "(see below). When", + "type": "text" + }, + { + "bbox": [ + 184, + 658, + 190, + 666 + ], + "score": 0.66, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 655, + 506, + 669 + ], + "score": 1.0, + "content": "is an identity mapping, the above transformation degenerates to a regular FFN,", + "type": "text" + } + ], + "index": 56 + }, + { + "bbox": [ + 104, + 666, + 506, + 680 + ], + "spans": [ + { + "bbox": [ + 104, + 666, + 506, + 680 + ], + "score": 1.0, + "content": "where individual tokens are processed independently without any cross-token communication. One", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 106, + 678, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 303, + 691 + ], + "score": 1.0, + "content": "of our major focuses is therefore to design a good", + "type": "text" + }, + { + "bbox": [ + 304, + 681, + 309, + 688 + ], + "score": 0.69, + "content": "s", + "type": "inline_equation" + }, + { + "bbox": [ + 309, + 678, + 506, + 691 + ], + "score": 1.0, + "content": "capable of capturing complex spatial interactions", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 105, + 688, + 505, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 488, + 702 + ], + "score": 1.0, + "content": "across tokens. The overall block layout is inspired by inverted bottlenecks [25] which define", + "type": "text" + }, + { + "bbox": [ + 489, + 689, + 505, + 701 + ], + "score": 0.88, + "content": "s ( \\cdot )", + "type": "inline_equation" + } + ], + "index": 59 + }, + { + "bbox": [ + 105, + 700, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 505, + 713 + ], + "score": 1.0, + "content": "as a spatial depthwise convolution. Note, unlike Transformers, our model does not require position", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 105, + 711, + 359, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 711, + 338, + 724 + ], + "score": 1.0, + "content": "embeddings because such information will be captured in", + "type": "text" + }, + { + "bbox": [ + 338, + 712, + 354, + 723 + ], + "score": 0.92, + "content": "s ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 711, + 359, + 724 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 61 + } + ], + "index": 58, + "bbox_fs": [ + 104, + 644, + 506, + 724 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 72, + 505, + 139 + ], + "lines": [ + { + "bbox": [ + 106, + 73, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 73, + 506, + 86 + ], + "score": 1.0, + "content": "Our model uses exactly the same input and output protocols as BERT (for NLP) and ViT (for vision).", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 83, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 506, + 96 + ], + "score": 1.0, + "content": "For example, when finetuning on language tasks, we concatenate together multiple text segments", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 93, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 506, + 107 + ], + "score": 1.0, + "content": "followed by paddings, and the predictions are deduced from the last-layer representation of a reserved", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 106, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 505, + 118 + ], + "score": 1.0, + "content": " symbol. Although many of these protocols were introduced for Transformers and hence can", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 117, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 117, + 505, + 128 + ], + "score": 1.0, + "content": "be suboptimal for gMLPs, strictly following them helps avoid confounding factors in our experiments", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 128, + 441, + 140 + ], + "spans": [ + { + "bbox": [ + 106, + 128, + 441, + 140 + ], + "score": 1.0, + "content": "and makes our layers more compatible with existing Transformer implementations.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 107, + 151, + 214, + 163 + ], + "lines": [ + { + "bbox": [ + 105, + 150, + 215, + 165 + ], + "spans": [ + { + "bbox": [ + 105, + 150, + 215, + 165 + ], + "score": 1.0, + "content": "2.1 Spatial Gating Unit", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 171, + 504, + 194 + ], + "lines": [ + { + "bbox": [ + 106, + 171, + 504, + 184 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 353, + 184 + ], + "score": 1.0, + "content": "To enable cross-token interactions, it is necessary for the layer", + "type": "text" + }, + { + "bbox": [ + 353, + 172, + 369, + 184 + ], + "score": 0.9, + "content": "s ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 171, + 504, + 184 + ], + "score": 1.0, + "content": "to contain a contraction operation", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 182, + 421, + 195 + ], + "spans": [ + { + "bbox": [ + 106, + 182, + 421, + 195 + ], + "score": 1.0, + "content": "over the spatial dimension. The simplistic option would be a linear projection:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "interline_equation", + "bbox": [ + 263, + 199, + 348, + 213 + ], + "lines": [ + { + "bbox": [ + 263, + 199, + 348, + 213 + ], + "spans": [ + { + "bbox": [ + 263, + 199, + 348, + 213 + ], + "score": 0.94, + "content": "f _ { W , b } ( Z ) = W Z + b", + "type": "interline_equation", + "image_path": "7e5c4dae68c79d1010c2b7fd3237f11b35cf0b787c253c05fa4e0267d371f418.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 263, + 199, + 348, + 213 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 218, + 505, + 263 + ], + "lines": [ + { + "bbox": [ + 105, + 217, + 505, + 231 + ], + "spans": [ + { + "bbox": [ + 105, + 217, + 133, + 231 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 218, + 181, + 228 + ], + "score": 0.92, + "content": "W \\in \\mathbb { R } ^ { n \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 217, + 444, + 231 + ], + "score": 1.0, + "content": "is a matrix for which the size is the same as the sequence length,", + "type": "text" + }, + { + "bbox": [ + 445, + 221, + 452, + 228 + ], + "score": 0.76, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 452, + 217, + 473, + 231 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 473, + 219, + 479, + 228 + ], + "score": 0.76, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 217, + 505, + 231 + ], + "score": 1.0, + "content": "refers", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 229, + 504, + 242 + ], + "spans": [ + { + "bbox": [ + 105, + 229, + 492, + 242 + ], + "score": 1.0, + "content": "token-specific biases. For example, if the padded input sequence has 128 tokens, the shape for", + "type": "text" + }, + { + "bbox": [ + 492, + 229, + 504, + 239 + ], + "score": 0.73, + "content": "W", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 240, + 504, + 252 + ], + "spans": [ + { + "bbox": [ + 106, + 241, + 137, + 252 + ], + "score": 1.0, + "content": "will be", + "type": "text" + }, + { + "bbox": [ + 137, + 240, + 176, + 251 + ], + "score": 0.89, + "content": "1 2 8 \\times 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 241, + 294, + 252 + ], + "score": 1.0, + "content": ". Unlike self-attention where", + "type": "text" + }, + { + "bbox": [ + 294, + 240, + 321, + 252 + ], + "score": 0.92, + "content": "W ( Z )", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 241, + 448, + 252 + ], + "score": 1.0, + "content": "is dynamically generated from", + "type": "text" + }, + { + "bbox": [ + 449, + 241, + 457, + 250 + ], + "score": 0.78, + "content": "Z", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 241, + 504, + 252 + ], + "score": 1.0, + "content": ", the spatial", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 251, + 459, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 251, + 177, + 264 + ], + "score": 1.0, + "content": "projection matrix", + "type": "text" + }, + { + "bbox": [ + 178, + 251, + 190, + 261 + ], + "score": 0.46, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 251, + 459, + 264 + ], + "score": 1.0, + "content": "here in Equation (2) is independent from the input representations.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 106, + 267, + 374, + 279 + ], + "lines": [ + { + "bbox": [ + 104, + 264, + 375, + 282 + ], + "spans": [ + { + "bbox": [ + 104, + 264, + 237, + 282 + ], + "score": 1.0, + "content": "In this work, we formulate layer", + "type": "text" + }, + { + "bbox": [ + 237, + 267, + 254, + 279 + ], + "score": 0.91, + "content": "s ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 264, + 375, + 282 + ], + "score": 1.0, + "content": "as the output of linear gating:", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + }, + { + "type": "interline_equation", + "bbox": [ + 261, + 284, + 350, + 298 + ], + "lines": [ + { + "bbox": [ + 261, + 284, + 350, + 298 + ], + "spans": [ + { + "bbox": [ + 261, + 284, + 350, + 298 + ], + "score": 0.93, + "content": "s ( Z ) = Z \\odot f _ { W , b } ( Z )", + "type": "interline_equation", + "image_path": "b084e72db12a7f0a08d1156d8e0af26ab807cbf9481d25ba50788a491cfa7ca0.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 261, + 284, + 350, + 298 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 302, + 505, + 358 + ], + "lines": [ + { + "bbox": [ + 106, + 303, + 505, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 133, + 315 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 304, + 143, + 313 + ], + "score": 0.81, + "content": "\\odot", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 303, + 505, + 315 + ], + "score": 1.0, + "content": "denotes element-wise multiplication. For training stability, we find it critical to initialize", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 313, + 505, + 327 + ], + "spans": [ + { + "bbox": [ + 106, + 314, + 119, + 324 + ], + "score": 0.57, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 313, + 219, + 327 + ], + "score": 1.0, + "content": "as near-zero values and", + "type": "text" + }, + { + "bbox": [ + 219, + 315, + 225, + 324 + ], + "score": 0.76, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 313, + 319, + 327 + ], + "score": 1.0, + "content": "as ones, meaning that", + "type": "text" + }, + { + "bbox": [ + 319, + 314, + 375, + 326 + ], + "score": 0.92, + "content": "f _ { W , b } ( Z ) \\approx { \\bf 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 313, + 434, + 327 + ], + "score": 1.0, + "content": "and therefore", + "type": "text" + }, + { + "bbox": [ + 434, + 314, + 478, + 326 + ], + "score": 0.94, + "content": "s ( Z ) \\approx Z", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 313, + 505, + 327 + ], + "score": 1.0, + "content": "at the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 325, + 505, + 337 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 505, + 337 + ], + "score": 1.0, + "content": "beginning of training. This initialization ensures each gMLP block behaves like a regular FFN at", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 335, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 505, + 349 + ], + "score": 1.0, + "content": "the early stage of training, where each token is processed independently, and only gradually injects", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 345, + 361, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 361, + 361 + ], + "score": 1.0, + "content": "spatial information across tokens during the course of learning.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 107, + 362, + 502, + 385 + ], + "lines": [ + { + "bbox": [ + 105, + 361, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 240, + 376 + ], + "score": 1.0, + "content": "We further find it effective to split", + "type": "text" + }, + { + "bbox": [ + 240, + 363, + 248, + 373 + ], + "score": 0.82, + "content": "Z", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 361, + 355, + 376 + ], + "score": 1.0, + "content": "into two independent parts", + "type": "text" + }, + { + "bbox": [ + 356, + 363, + 390, + 374 + ], + "score": 0.9, + "content": "( Z _ { 1 } , Z _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 361, + 505, + 376 + ], + "score": 1.0, + "content": "along the channel dimension", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 373, + 334, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 334, + 387 + ], + "score": 1.0, + "content": "for the gating function and for the multiplicative bypass:", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + }, + { + "type": "interline_equation", + "bbox": [ + 257, + 390, + 354, + 405 + ], + "lines": [ + { + "bbox": [ + 257, + 390, + 354, + 405 + ], + "spans": [ + { + "bbox": [ + 257, + 390, + 354, + 405 + ], + "score": 0.94, + "content": "s ( Z ) = Z _ { 1 } \\odot f _ { W , b } ( Z _ { 2 } )", + "type": "interline_equation", + "image_path": "f6a604c6e6d1a2b5953c15b5a23e51838a39cd7db61dd6c11719cc258562c35a.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 257, + 390, + 354, + 405 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 409, + 505, + 453 + ], + "lines": [ + { + "bbox": [ + 104, + 408, + 506, + 423 + ], + "spans": [ + { + "bbox": [ + 104, + 408, + 229, + 423 + ], + "score": 1.0, + "content": "We also normalize the input to", + "type": "text" + }, + { + "bbox": [ + 229, + 410, + 249, + 422 + ], + "score": 0.91, + "content": "f _ { W , b }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 408, + 506, + 423 + ], + "score": 1.0, + "content": "which empirically improves stability of large NLP models. This", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 420, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 505, + 433 + ], + "score": 1.0, + "content": "gives us the unit illustrated in Figure 1, which we refer to as the Spatial Gating Unit (SGU) in the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 431, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 505, + 444 + ], + "score": 1.0, + "content": "rest of the paper. In Table 3, we provide ablation studies to compare SGU with several other variants", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 442, + 464, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 117, + 455 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 442, + 133, + 454 + ], + "score": 0.9, + "content": "s ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 442, + 464, + 455 + ], + "score": 1.0, + "content": ", showing that it works better and narrows the performance gap with self-attention.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 106, + 465, + 505, + 631 + ], + "lines": [ + { + "bbox": [ + 106, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 505, + 478 + ], + "score": 1.0, + "content": "Connections to Existing Layers. The overall formulation of SGU resembles Gated Linear Units", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 475, + 506, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 506, + 489 + ], + "score": 1.0, + "content": "(GLUs) [26, 27, 28] as well as earlier works including Highway Networks [29] and LSTM-RNNs [11].", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "score": 1.0, + "content": "A key distinction is that our gating is computed based on a projection over the spatial (cross-token)", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "score": 1.0, + "content": "dimension rather than the channel (hidden) dimension. SGU is also related to Squeeze-and-Excite", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 509, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 106, + 509, + 505, + 522 + ], + "score": 1.0, + "content": "(SE) blocks [30] in terms of element-wise multiplication. However, different from SE blocks, SGU", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 519, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 505, + 533 + ], + "score": 1.0, + "content": "does not contain cross-channel projections at all, nor does it enforce permutation invariance (a", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 531, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 505, + 543 + ], + "score": 1.0, + "content": "key feature for content-based attentive modules) due to its static parameterization for the spatial", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 542, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 542, + 505, + 554 + ], + "score": 1.0, + "content": "transformation. The spatial projection in SGU could in theory learn to express superficial depthwise", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 552, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 505, + 565 + ], + "score": 1.0, + "content": "convolutions—unlike typical depthwise convolutions with channel-specific filters, SGU learns only a", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 564, + 505, + 575 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 505, + 575 + ], + "score": 1.0, + "content": "single transformation shared across channels. Finally, we note SGUs offer an alternative mechanism", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 575, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 575, + 506, + 587 + ], + "score": 1.0, + "content": "to capture high-order relationships other than self-attention. Specifically, the output for Equation (3)", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 585, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 286, + 599 + ], + "score": 1.0, + "content": "contains up to 2nd-order interactions (e.g.,", + "type": "text" + }, + { + "bbox": [ + 286, + 586, + 304, + 598 + ], + "score": 0.84, + "content": "z _ { i } z _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 585, + 506, + 599 + ], + "score": 1.0, + "content": ") whereas output for self-attention (assuming no", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 595, + 506, + 609 + ], + "spans": [ + { + "bbox": [ + 104, + 595, + 328, + 609 + ], + "score": 1.0, + "content": "nonlinearity) contains up to 3rd-order interactions (e.g.,", + "type": "text" + }, + { + "bbox": [ + 329, + 597, + 359, + 608 + ], + "score": 0.9, + "content": "q _ { i } k _ { j } v _ { k } )", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 595, + 506, + 609 + ], + "score": 1.0, + "content": ". In terms of computation cost, SGU", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 608, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 608, + 122, + 621 + ], + "score": 1.0, + "content": "has", + "type": "text" + }, + { + "bbox": [ + 123, + 608, + 149, + 621 + ], + "score": 0.91, + "content": "n ^ { 2 } e / 2", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 608, + 320, + 621 + ], + "score": 1.0, + "content": "multiply-adds which is comparable to the", + "type": "text" + }, + { + "bbox": [ + 321, + 608, + 343, + 619 + ], + "score": 0.9, + "content": "2 n ^ { 2 } d", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 608, + 505, + 621 + ], + "score": 1.0, + "content": "of dot-product self-attention.1 Both are", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 619, + 409, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 398, + 632 + ], + "score": 1.0, + "content": "linear over the input channel size and quadratic over the sequence length", + "type": "text" + }, + { + "bbox": [ + 398, + 621, + 405, + 629 + ], + "score": 0.68, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 619, + 409, + 632 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 35 + }, + { + "type": "title", + "bbox": [ + 107, + 645, + 229, + 659 + ], + "lines": [ + { + "bbox": [ + 104, + 644, + 230, + 663 + ], + "spans": [ + { + "bbox": [ + 104, + 644, + 230, + 663 + ], + "score": 1.0, + "content": "3 Image Classification", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 43 + }, + { + "type": "text", + "bbox": [ + 108, + 670, + 504, + 693 + ], + "lines": [ + { + "bbox": [ + 106, + 670, + 506, + 684 + ], + "spans": [ + { + "bbox": [ + 106, + 670, + 182, + 684 + ], + "score": 1.0, + "content": "Here we examine", + "type": "text" + }, + { + "bbox": [ + 182, + 671, + 210, + 682 + ], + "score": 0.41, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 670, + 506, + 684 + ], + "score": 1.0, + "content": "in the vision domain by applying it to the image classification task on", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 107, + 682, + 505, + 694 + ], + "spans": [ + { + "bbox": [ + 107, + 682, + 505, + 694 + ], + "score": 1.0, + "content": "ImageNet [31] without using extra data. We compare our MLP-like models with recent attentive", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 105, + 701, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 119, + 700, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 119, + 700, + 204, + 713 + ], + "score": 1.0, + "content": "1The input channel size", + "type": "text" + }, + { + "bbox": [ + 205, + 704, + 210, + 711 + ], + "score": 0.39, + "content": "e", + "type": "inline_equation" + }, + { + "bbox": [ + 211, + 700, + 404, + 713 + ], + "score": 1.0, + "content": "for SGU is typically larger than the input channel size", + "type": "text" + }, + { + "bbox": [ + 405, + 702, + 410, + 711 + ], + "score": 0.78, + "content": "d", + "type": "inline_equation" + }, + { + "bbox": [ + 411, + 700, + 505, + 713 + ], + "score": 1.0, + "content": "for self-attention, because", + "type": "text" + } + ] + }, + { + "bbox": [ + 106, + 711, + 374, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 374, + 723 + ], + "score": 1.0, + "content": "the former is applied in the middle of the block after a channel expansion.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 72, + 505, + 139 + ], + "lines": [ + { + "bbox": [ + 106, + 73, + 506, + 86 + ], + "spans": [ + { + "bbox": [ + 106, + 73, + 506, + 86 + ], + "score": 1.0, + "content": "Our model uses exactly the same input and output protocols as BERT (for NLP) and ViT (for vision).", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 83, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 506, + 96 + ], + "score": 1.0, + "content": "For example, when finetuning on language tasks, we concatenate together multiple text segments", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 93, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 93, + 506, + 107 + ], + "score": 1.0, + "content": "followed by paddings, and the predictions are deduced from the last-layer representation of a reserved", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 106, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 106, + 106, + 505, + 118 + ], + "score": 1.0, + "content": " symbol. Although many of these protocols were introduced for Transformers and hence can", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 117, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 117, + 505, + 128 + ], + "score": 1.0, + "content": "be suboptimal for gMLPs, strictly following them helps avoid confounding factors in our experiments", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 128, + 441, + 140 + ], + "spans": [ + { + "bbox": [ + 106, + 128, + 441, + 140 + ], + "score": 1.0, + "content": "and makes our layers more compatible with existing Transformer implementations.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5, + "bbox_fs": [ + 105, + 73, + 506, + 140 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 151, + 214, + 163 + ], + "lines": [ + { + "bbox": [ + 105, + 150, + 215, + 165 + ], + "spans": [ + { + "bbox": [ + 105, + 150, + 215, + 165 + ], + "score": 1.0, + "content": "2.1 Spatial Gating Unit", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 106, + 171, + 504, + 194 + ], + "lines": [ + { + "bbox": [ + 106, + 171, + 504, + 184 + ], + "spans": [ + { + "bbox": [ + 106, + 171, + 353, + 184 + ], + "score": 1.0, + "content": "To enable cross-token interactions, it is necessary for the layer", + "type": "text" + }, + { + "bbox": [ + 353, + 172, + 369, + 184 + ], + "score": 0.9, + "content": "s ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 370, + 171, + 504, + 184 + ], + "score": 1.0, + "content": "to contain a contraction operation", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 182, + 421, + 195 + ], + "spans": [ + { + "bbox": [ + 106, + 182, + 421, + 195 + ], + "score": 1.0, + "content": "over the spatial dimension. The simplistic option would be a linear projection:", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 106, + 171, + 504, + 195 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 263, + 199, + 348, + 213 + ], + "lines": [ + { + "bbox": [ + 263, + 199, + 348, + 213 + ], + "spans": [ + { + "bbox": [ + 263, + 199, + 348, + 213 + ], + "score": 0.94, + "content": "f _ { W , b } ( Z ) = W Z + b", + "type": "interline_equation", + "image_path": "7e5c4dae68c79d1010c2b7fd3237f11b35cf0b787c253c05fa4e0267d371f418.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 263, + 199, + 348, + 213 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 218, + 505, + 263 + ], + "lines": [ + { + "bbox": [ + 105, + 217, + 505, + 231 + ], + "spans": [ + { + "bbox": [ + 105, + 217, + 133, + 231 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 218, + 181, + 228 + ], + "score": 0.92, + "content": "W \\in \\mathbb { R } ^ { n \\times n }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 217, + 444, + 231 + ], + "score": 1.0, + "content": "is a matrix for which the size is the same as the sequence length,", + "type": "text" + }, + { + "bbox": [ + 445, + 221, + 452, + 228 + ], + "score": 0.76, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 452, + 217, + 473, + 231 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 473, + 219, + 479, + 228 + ], + "score": 0.76, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 217, + 505, + 231 + ], + "score": 1.0, + "content": "refers", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 105, + 229, + 504, + 242 + ], + "spans": [ + { + "bbox": [ + 105, + 229, + 492, + 242 + ], + "score": 1.0, + "content": "token-specific biases. For example, if the padded input sequence has 128 tokens, the shape for", + "type": "text" + }, + { + "bbox": [ + 492, + 229, + 504, + 239 + ], + "score": 0.73, + "content": "W", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 240, + 504, + 252 + ], + "spans": [ + { + "bbox": [ + 106, + 241, + 137, + 252 + ], + "score": 1.0, + "content": "will be", + "type": "text" + }, + { + "bbox": [ + 137, + 240, + 176, + 251 + ], + "score": 0.89, + "content": "1 2 8 \\times 1 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 241, + 294, + 252 + ], + "score": 1.0, + "content": ". Unlike self-attention where", + "type": "text" + }, + { + "bbox": [ + 294, + 240, + 321, + 252 + ], + "score": 0.92, + "content": "W ( Z )", + "type": "inline_equation" + }, + { + "bbox": [ + 321, + 241, + 448, + 252 + ], + "score": 1.0, + "content": "is dynamically generated from", + "type": "text" + }, + { + "bbox": [ + 449, + 241, + 457, + 250 + ], + "score": 0.78, + "content": "Z", + "type": "inline_equation" + }, + { + "bbox": [ + 457, + 241, + 504, + 252 + ], + "score": 1.0, + "content": ", the spatial", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 251, + 459, + 264 + ], + "spans": [ + { + "bbox": [ + 105, + 251, + 177, + 264 + ], + "score": 1.0, + "content": "projection matrix", + "type": "text" + }, + { + "bbox": [ + 178, + 251, + 190, + 261 + ], + "score": 0.46, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 251, + 459, + 264 + ], + "score": 1.0, + "content": "here in Equation (2) is independent from the input representations.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 11.5, + "bbox_fs": [ + 105, + 217, + 505, + 264 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 267, + 374, + 279 + ], + "lines": [ + { + "bbox": [ + 104, + 264, + 375, + 282 + ], + "spans": [ + { + "bbox": [ + 104, + 264, + 237, + 282 + ], + "score": 1.0, + "content": "In this work, we formulate layer", + "type": "text" + }, + { + "bbox": [ + 237, + 267, + 254, + 279 + ], + "score": 0.91, + "content": "s ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 264, + 375, + 282 + ], + "score": 1.0, + "content": "as the output of linear gating:", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14, + "bbox_fs": [ + 104, + 264, + 375, + 282 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 261, + 284, + 350, + 298 + ], + "lines": [ + { + "bbox": [ + 261, + 284, + 350, + 298 + ], + "spans": [ + { + "bbox": [ + 261, + 284, + 350, + 298 + ], + "score": 0.93, + "content": "s ( Z ) = Z \\odot f _ { W , b } ( Z )", + "type": "interline_equation", + "image_path": "b084e72db12a7f0a08d1156d8e0af26ab807cbf9481d25ba50788a491cfa7ca0.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 261, + 284, + 350, + 298 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 302, + 505, + 358 + ], + "lines": [ + { + "bbox": [ + 106, + 303, + 505, + 315 + ], + "spans": [ + { + "bbox": [ + 106, + 303, + 133, + 315 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 133, + 304, + 143, + 313 + ], + "score": 0.81, + "content": "\\odot", + "type": "inline_equation" + }, + { + "bbox": [ + 143, + 303, + 505, + 315 + ], + "score": 1.0, + "content": "denotes element-wise multiplication. For training stability, we find it critical to initialize", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 106, + 313, + 505, + 327 + ], + "spans": [ + { + "bbox": [ + 106, + 314, + 119, + 324 + ], + "score": 0.57, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 313, + 219, + 327 + ], + "score": 1.0, + "content": "as near-zero values and", + "type": "text" + }, + { + "bbox": [ + 219, + 315, + 225, + 324 + ], + "score": 0.76, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 313, + 319, + 327 + ], + "score": 1.0, + "content": "as ones, meaning that", + "type": "text" + }, + { + "bbox": [ + 319, + 314, + 375, + 326 + ], + "score": 0.92, + "content": "f _ { W , b } ( Z ) \\approx { \\bf 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 313, + 434, + 327 + ], + "score": 1.0, + "content": "and therefore", + "type": "text" + }, + { + "bbox": [ + 434, + 314, + 478, + 326 + ], + "score": 0.94, + "content": "s ( Z ) \\approx Z", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 313, + 505, + 327 + ], + "score": 1.0, + "content": "at the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 106, + 325, + 505, + 337 + ], + "spans": [ + { + "bbox": [ + 106, + 325, + 505, + 337 + ], + "score": 1.0, + "content": "beginning of training. This initialization ensures each gMLP block behaves like a regular FFN at", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 335, + 505, + 349 + ], + "spans": [ + { + "bbox": [ + 105, + 335, + 505, + 349 + ], + "score": 1.0, + "content": "the early stage of training, where each token is processed independently, and only gradually injects", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 345, + 361, + 361 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 361, + 361 + ], + "score": 1.0, + "content": "spatial information across tokens during the course of learning.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18, + "bbox_fs": [ + 105, + 303, + 505, + 361 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 362, + 502, + 385 + ], + "lines": [ + { + "bbox": [ + 105, + 361, + 505, + 376 + ], + "spans": [ + { + "bbox": [ + 105, + 361, + 240, + 376 + ], + "score": 1.0, + "content": "We further find it effective to split", + "type": "text" + }, + { + "bbox": [ + 240, + 363, + 248, + 373 + ], + "score": 0.82, + "content": "Z", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 361, + 355, + 376 + ], + "score": 1.0, + "content": "into two independent parts", + "type": "text" + }, + { + "bbox": [ + 356, + 363, + 390, + 374 + ], + "score": 0.9, + "content": "( Z _ { 1 } , Z _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 390, + 361, + 505, + 376 + ], + "score": 1.0, + "content": "along the channel dimension", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 373, + 334, + 387 + ], + "spans": [ + { + "bbox": [ + 105, + 373, + 334, + 387 + ], + "score": 1.0, + "content": "for the gating function and for the multiplicative bypass:", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 361, + 505, + 387 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 257, + 390, + 354, + 405 + ], + "lines": [ + { + "bbox": [ + 257, + 390, + 354, + 405 + ], + "spans": [ + { + "bbox": [ + 257, + 390, + 354, + 405 + ], + "score": 0.94, + "content": "s ( Z ) = Z _ { 1 } \\odot f _ { W , b } ( Z _ { 2 } )", + "type": "interline_equation", + "image_path": "f6a604c6e6d1a2b5953c15b5a23e51838a39cd7db61dd6c11719cc258562c35a.jpg" + } + ] + } + ], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 257, + 390, + 354, + 405 + ], + "spans": [], + "index": 23 + } + ] + }, + { + "type": "text", + "bbox": [ + 106, + 409, + 505, + 453 + ], + "lines": [ + { + "bbox": [ + 104, + 408, + 506, + 423 + ], + "spans": [ + { + "bbox": [ + 104, + 408, + 229, + 423 + ], + "score": 1.0, + "content": "We also normalize the input to", + "type": "text" + }, + { + "bbox": [ + 229, + 410, + 249, + 422 + ], + "score": 0.91, + "content": "f _ { W , b }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 408, + 506, + 423 + ], + "score": 1.0, + "content": "which empirically improves stability of large NLP models. This", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 420, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 105, + 420, + 505, + 433 + ], + "score": 1.0, + "content": "gives us the unit illustrated in Figure 1, which we refer to as the Spatial Gating Unit (SGU) in the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 431, + 505, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 431, + 505, + 444 + ], + "score": 1.0, + "content": "rest of the paper. In Table 3, we provide ablation studies to compare SGU with several other variants", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 442, + 464, + 455 + ], + "spans": [ + { + "bbox": [ + 105, + 442, + 117, + 455 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 118, + 442, + 133, + 454 + ], + "score": 0.9, + "content": "s ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 134, + 442, + 464, + 455 + ], + "score": 1.0, + "content": ", showing that it works better and narrows the performance gap with self-attention.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5, + "bbox_fs": [ + 104, + 408, + 506, + 455 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 465, + 505, + 631 + ], + "lines": [ + { + "bbox": [ + 106, + 465, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 106, + 465, + 505, + 478 + ], + "score": 1.0, + "content": "Connections to Existing Layers. The overall formulation of SGU resembles Gated Linear Units", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 475, + 506, + 489 + ], + "spans": [ + { + "bbox": [ + 105, + 475, + 506, + 489 + ], + "score": 1.0, + "content": "(GLUs) [26, 27, 28] as well as earlier works including Highway Networks [29] and LSTM-RNNs [11].", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 105, + 487, + 505, + 500 + ], + "score": 1.0, + "content": "A key distinction is that our gating is computed based on a projection over the spatial (cross-token)", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "spans": [ + { + "bbox": [ + 106, + 498, + 505, + 510 + ], + "score": 1.0, + "content": "dimension rather than the channel (hidden) dimension. SGU is also related to Squeeze-and-Excite", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 106, + 509, + 505, + 522 + ], + "spans": [ + { + "bbox": [ + 106, + 509, + 505, + 522 + ], + "score": 1.0, + "content": "(SE) blocks [30] in terms of element-wise multiplication. However, different from SE blocks, SGU", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 519, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 505, + 533 + ], + "score": 1.0, + "content": "does not contain cross-channel projections at all, nor does it enforce permutation invariance (a", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 106, + 531, + 505, + 543 + ], + "spans": [ + { + "bbox": [ + 106, + 531, + 505, + 543 + ], + "score": 1.0, + "content": "key feature for content-based attentive modules) due to its static parameterization for the spatial", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 542, + 505, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 542, + 505, + 554 + ], + "score": 1.0, + "content": "transformation. The spatial projection in SGU could in theory learn to express superficial depthwise", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 552, + 505, + 565 + ], + "spans": [ + { + "bbox": [ + 106, + 552, + 505, + 565 + ], + "score": 1.0, + "content": "convolutions—unlike typical depthwise convolutions with channel-specific filters, SGU learns only a", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 564, + 505, + 575 + ], + "spans": [ + { + "bbox": [ + 106, + 564, + 505, + 575 + ], + "score": 1.0, + "content": "single transformation shared across channels. Finally, we note SGUs offer an alternative mechanism", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 106, + 575, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 575, + 506, + 587 + ], + "score": 1.0, + "content": "to capture high-order relationships other than self-attention. Specifically, the output for Equation (3)", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 585, + 506, + 599 + ], + "spans": [ + { + "bbox": [ + 105, + 585, + 286, + 599 + ], + "score": 1.0, + "content": "contains up to 2nd-order interactions (e.g.,", + "type": "text" + }, + { + "bbox": [ + 286, + 586, + 304, + 598 + ], + "score": 0.84, + "content": "z _ { i } z _ { j }", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 585, + 506, + 599 + ], + "score": 1.0, + "content": ") whereas output for self-attention (assuming no", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 104, + 595, + 506, + 609 + ], + "spans": [ + { + "bbox": [ + 104, + 595, + 328, + 609 + ], + "score": 1.0, + "content": "nonlinearity) contains up to 3rd-order interactions (e.g.,", + "type": "text" + }, + { + "bbox": [ + 329, + 597, + 359, + 608 + ], + "score": 0.9, + "content": "q _ { i } k _ { j } v _ { k } )", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 595, + 506, + 609 + ], + "score": 1.0, + "content": ". In terms of computation cost, SGU", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 608, + 505, + 621 + ], + "spans": [ + { + "bbox": [ + 106, + 608, + 122, + 621 + ], + "score": 1.0, + "content": "has", + "type": "text" + }, + { + "bbox": [ + 123, + 608, + 149, + 621 + ], + "score": 0.91, + "content": "n ^ { 2 } e / 2", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 608, + 320, + 621 + ], + "score": 1.0, + "content": "multiply-adds which is comparable to the", + "type": "text" + }, + { + "bbox": [ + 321, + 608, + 343, + 619 + ], + "score": 0.9, + "content": "2 n ^ { 2 } d", + "type": "inline_equation" + }, + { + "bbox": [ + 343, + 608, + 505, + 621 + ], + "score": 1.0, + "content": "of dot-product self-attention.1 Both are", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 619, + 409, + 632 + ], + "spans": [ + { + "bbox": [ + 105, + 619, + 398, + 632 + ], + "score": 1.0, + "content": "linear over the input channel size and quadratic over the sequence length", + "type": "text" + }, + { + "bbox": [ + 398, + 621, + 405, + 629 + ], + "score": 0.68, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 406, + 619, + 409, + 632 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 42 + } + ], + "index": 35, + "bbox_fs": [ + 104, + 465, + 506, + 632 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 645, + 229, + 659 + ], + "lines": [ + { + "bbox": [ + 104, + 644, + 230, + 663 + ], + "spans": [ + { + "bbox": [ + 104, + 644, + 230, + 663 + ], + "score": 1.0, + "content": "3 Image Classification", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 43 + }, + { + "type": "text", + "bbox": [ + 108, + 670, + 504, + 693 + ], + "lines": [ + { + "bbox": [ + 106, + 670, + 506, + 684 + ], + "spans": [ + { + "bbox": [ + 106, + 670, + 182, + 684 + ], + "score": 1.0, + "content": "Here we examine", + "type": "text" + }, + { + "bbox": [ + 182, + 671, + 210, + 682 + ], + "score": 0.41, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 670, + 506, + 684 + ], + "score": 1.0, + "content": "in the vision domain by applying it to the image classification task on", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 107, + 682, + 505, + 694 + ], + "spans": [ + { + "bbox": [ + 107, + 682, + 505, + 694 + ], + "score": 1.0, + "content": "ImageNet [31] without using extra data. We compare our MLP-like models with recent attentive", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 106, + 73, + 504, + 84 + ], + "spans": [ + { + "bbox": [ + 106, + 73, + 504, + 84 + ], + "score": 1.0, + "content": "models based on vanilla Transformers, including Vision Transformer (ViT) [7], DeiT [8] (ViT with", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 84, + 437, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 84, + 437, + 96 + ], + "score": 1.0, + "content": "improved regularization), and several other representative convolutional networks.", + "type": "text", + "cross_page": true + } + ], + "index": 1 + } + ], + "index": 44.5, + "bbox_fs": [ + 106, + 670, + 506, + 694 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 72, + 504, + 95 + ], + "lines": [ + { + "bbox": [ + 106, + 73, + 504, + 84 + ], + "spans": [ + { + "bbox": [ + 106, + 73, + 504, + 84 + ], + "score": 1.0, + "content": "models based on vanilla Transformers, including Vision Transformer (ViT) [7], DeiT [8] (ViT with", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 84, + 437, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 84, + 437, + 96 + ], + "score": 1.0, + "content": "improved regularization), and several other representative convolutional networks.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 106, + 100, + 505, + 177 + ], + "lines": [ + { + "bbox": [ + 105, + 99, + 505, + 113 + ], + "spans": [ + { + "bbox": [ + 105, + 99, + 505, + 113 + ], + "score": 1.0, + "content": "Table 1 summarizes the configurations of our gMLP image classification models. The input and", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 110, + 506, + 124 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 396, + 124 + ], + "score": 1.0, + "content": "output protocols follow ViT/B16 where the raw image is converted into", + "type": "text" + }, + { + "bbox": [ + 396, + 111, + 425, + 122 + ], + "score": 0.87, + "content": "1 6 \\times 1 6", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 110, + 506, + 124 + ], + "score": 1.0, + "content": "patches at the stem.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 121, + 505, + 134 + ], + "spans": [ + { + "bbox": [ + 105, + 121, + 505, + 134 + ], + "score": 1.0, + "content": "The depth and width are chosen so that the models are comparable with ViT/DeiT in capacity. Like", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 132, + 506, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 132, + 506, + 145 + ], + "score": 1.0, + "content": "Transformers, we find gMLPs tend to drastically overfit the training data. We therefore apply a", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 142, + 506, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 142, + 506, + 156 + ], + "score": 1.0, + "content": "similar regularization recipe as the one used in DeiT.2 To avoid extensive tuning, we adjust only the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 154, + 506, + 167 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 506, + 167 + ], + "score": 1.0, + "content": "strengths of stochastic depth [32] as we move from smaller to larger models in Table 1. All the other", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 165, + 455, + 177 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 455, + 177 + ], + "score": 1.0, + "content": "hyperparameters remain shared across our three models. See Appendix A.1 for details.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 5 + }, + { + "type": "table", + "bbox": [ + 158, + 203, + 451, + 255 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 176, + 185, + 434, + 197 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 176, + 185, + 434, + 198 + ], + "spans": [ + { + "bbox": [ + 176, + 185, + 331, + 198 + ], + "score": 1.0, + "content": "Table 1: Architecture specifications of", + "type": "text" + }, + { + "bbox": [ + 332, + 185, + 359, + 197 + ], + "score": 0.45, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 185, + 434, + 198 + ], + "score": 1.0, + "content": "models for vision.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "table_body", + "bbox": [ + 158, + 203, + 451, + 255 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 158, + 203, + 451, + 255 + ], + "spans": [ + { + "bbox": [ + 158, + 203, + 451, + 255 + ], + "score": 0.974, + "html": "
#LdmodeldffnParams (M)FLOPs (B)Survival Prob
gMLP-Ti301287685.92.71.00
gMLP-S30256153619.58.90.95
gMLP-B30512307273.431.60.80
", + "type": "table", + "image_path": "27f42f69254be8b6e8eeff8fb7a1820daeca371c25a2a48691d8d74cf19caa2f.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 158, + 203, + 451, + 220.33333333333334 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 158, + 220.33333333333334, + 451, + 237.66666666666669 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 158, + 237.66666666666669, + 451, + 255.00000000000003 + ], + "spans": [], + "index": 12 + } + ] + } + ], + "index": 10.0 + }, + { + "type": "text", + "bbox": [ + 106, + 262, + 505, + 362 + ], + "lines": [ + { + "bbox": [ + 106, + 261, + 506, + 276 + ], + "spans": [ + { + "bbox": [ + 106, + 261, + 506, + 276 + ], + "score": 1.0, + "content": "Our ImageNet results are summarized in Table 1 and Figure 2.3 It is interesting to see that gMLPs are", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 273, + 506, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 506, + 286 + ], + "score": 1.0, + "content": "comparable with DeiT [8], namely ViT [7] trained using improved regularization. The results suggest", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 285, + 505, + 297 + ], + "spans": [ + { + "bbox": [ + 106, + 285, + 505, + 297 + ], + "score": 1.0, + "content": "that models without self-attention can be as data-efficient as Transformers for image classification. In", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 293, + 506, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 506, + 309 + ], + "score": 1.0, + "content": "fact, when the models are properly regularized, their accuracies seem better correlated with capacity", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 306, + 505, + 319 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 505, + 319 + ], + "score": 1.0, + "content": "instead of the presence of self-attention. Moreover, the accuracy-parameter/FLOPs tradeoff of gMLPs", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 317, + 506, + 330 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 506, + 330 + ], + "score": 1.0, + "content": "surpasses all concurrently proposed MLP-like architectures [20, 21, 22], which we attribute to the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 329, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 505, + 340 + ], + "score": 1.0, + "content": "effectiveness of our Spatial Gating Unit (see Table 3 in the next section for an ablation). We also note", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 338, + 506, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 130, + 353 + ], + "score": 1.0, + "content": "while", + "type": "text" + }, + { + "bbox": [ + 130, + 339, + 160, + 351 + ], + "score": 0.57, + "content": "{ \\mathrm { g M L P s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 338, + 506, + 353 + ], + "score": 1.0, + "content": "are competitive with vanilla Transformers, their performance is behind the best existing", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 350, + 413, + 363 + ], + "spans": [ + { + "bbox": [ + 106, + 350, + 413, + 363 + ], + "score": 1.0, + "content": "ConvNet models (e.g., [33, 34]) or hybrid models (e.g., [35, 36, 37, 38, 10]).", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 17 + }, + { + "type": "table", + "bbox": [ + 145, + 387, + 464, + 631 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 208, + 370, + 403, + 382 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 207, + 369, + 404, + 383 + ], + "spans": [ + { + "bbox": [ + 207, + 369, + 404, + 383 + ], + "score": 1.0, + "content": "Table 2: ImageNet-1K results without extra data.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "table_body", + "bbox": [ + 145, + 387, + 464, + 631 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 145, + 387, + 464, + 631 + ], + "spans": [ + { + "bbox": [ + 145, + 387, + 464, + 631 + ], + "score": 0.984, + "html": "
ModelImageNet Top-1 (%)*Input ResolutionParams (M)MAdds (B)
ConvNets
ResNet-152 [16]78.32246011.3
RegNetY-8GF[39]81.7224398.0
EfficientNet-B0 [17]77.122450.39
EfficientNet-B3[17]81.6300121.8
EfficientNet-B7 [17]84.36006637.0
NFNet-F0 [33]83.61927212.4
Transformers
ViT-B/16 [7]77.93848655.4
ViT-L/16 [7]76.5384307190.7
DeiT-Ti [8] (ViT+reg)72.222451.3
DeiT-S [8](ViT+reg)79.8224224.6
DeiT-B [8] (ViT+reg)81.82248617.5
MLP-like†
Mixer-B/16 [20]76.42245912.7
Mixer-B/16 (our setup)77.32245912.7
Mixer-L/16 [20]71.822420744.8
ResMLP-12 [22]76.6224153.0
ResMLP-24 [22]79.4224306.0
ResMLP-36[22]79.7224458.9
gMLP-Ti (ours)72.322461.4
gMLP-S (ours)79.6224204.5
gMLP-B (ours)81.62247315.8
", + "type": "table", + "image_path": "123f13f92d912e05ece7dd96c75519ea83477124b08ac3939ad58b8b69f2b23d.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 145, + 387, + 464, + 468.3333333333333 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 145, + 468.3333333333333, + 464, + 549.6666666666666 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 145, + 549.6666666666666, + 464, + 631.0 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 151, + 633, + 390, + 650 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 152, + 631, + 355, + 642 + ], + "spans": [ + { + "bbox": [ + 152, + 631, + 355, + 642 + ], + "score": 1.0, + "content": "* Standard deviation across multiple independent runs is around 0.1.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 151, + 639, + 392, + 652 + ], + "spans": [ + { + "bbox": [ + 151, + 639, + 392, + 652 + ], + "score": 1.0, + "content": "† Tokenization & embedding process at the stem can be viewed as a convolution.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 660, + 505, + 694 + ], + "lines": [ + { + "bbox": [ + 106, + 660, + 506, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 660, + 506, + 673 + ], + "score": 1.0, + "content": "Figure 3 visualizes the spatial projection matrices in gMLP-B. Remarkably, the spatial weights after", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 671, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 505, + 684 + ], + "score": 1.0, + "content": "learning exhibit both locality and spatial invariance. In other words, each spatial projection matrix", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 682, + 493, + 695 + ], + "spans": [ + { + "bbox": [ + 106, + 682, + 493, + 695 + ], + "score": 1.0, + "content": "effectively learns to perform convolution with a data-driven, irregular (non-square) kernel shape.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 118, + 700, + 402, + 722 + ], + "lines": [ + { + "bbox": [ + 118, + 698, + 376, + 713 + ], + "spans": [ + { + "bbox": [ + 118, + 698, + 376, + 713 + ], + "score": 1.0, + "content": "2Unlike DeiT, we do not use repeated augmentation or random erasing.", + "type": "text" + } + ] + }, + { + "bbox": [ + 118, + 709, + 403, + 724 + ], + "spans": [ + { + "bbox": [ + 118, + 709, + 403, + 724 + ], + "score": 1.0, + "content": "3Additional results on ImageNet-21K and COCO are available in Appendix E.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 741, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 9 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 72, + 504, + 95 + ], + "lines": [], + "index": 0.5, + "bbox_fs": [ + 105, + 73, + 504, + 96 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 106, + 100, + 505, + 177 + ], + "lines": [ + { + "bbox": [ + 105, + 99, + 505, + 113 + ], + "spans": [ + { + "bbox": [ + 105, + 99, + 505, + 113 + ], + "score": 1.0, + "content": "Table 1 summarizes the configurations of our gMLP image classification models. The input and", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 110, + 506, + 124 + ], + "spans": [ + { + "bbox": [ + 105, + 110, + 396, + 124 + ], + "score": 1.0, + "content": "output protocols follow ViT/B16 where the raw image is converted into", + "type": "text" + }, + { + "bbox": [ + 396, + 111, + 425, + 122 + ], + "score": 0.87, + "content": "1 6 \\times 1 6", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 110, + 506, + 124 + ], + "score": 1.0, + "content": "patches at the stem.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 121, + 505, + 134 + ], + "spans": [ + { + "bbox": [ + 105, + 121, + 505, + 134 + ], + "score": 1.0, + "content": "The depth and width are chosen so that the models are comparable with ViT/DeiT in capacity. Like", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 132, + 506, + 145 + ], + "spans": [ + { + "bbox": [ + 105, + 132, + 506, + 145 + ], + "score": 1.0, + "content": "Transformers, we find gMLPs tend to drastically overfit the training data. We therefore apply a", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 142, + 506, + 156 + ], + "spans": [ + { + "bbox": [ + 105, + 142, + 506, + 156 + ], + "score": 1.0, + "content": "similar regularization recipe as the one used in DeiT.2 To avoid extensive tuning, we adjust only the", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 154, + 506, + 167 + ], + "spans": [ + { + "bbox": [ + 106, + 154, + 506, + 167 + ], + "score": 1.0, + "content": "strengths of stochastic depth [32] as we move from smaller to larger models in Table 1. All the other", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 165, + 455, + 177 + ], + "spans": [ + { + "bbox": [ + 105, + 165, + 455, + 177 + ], + "score": 1.0, + "content": "hyperparameters remain shared across our three models. See Appendix A.1 for details.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 5, + "bbox_fs": [ + 105, + 99, + 506, + 177 + ] + }, + { + "type": "table", + "bbox": [ + 158, + 203, + 451, + 255 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 176, + 185, + 434, + 197 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 176, + 185, + 434, + 198 + ], + "spans": [ + { + "bbox": [ + 176, + 185, + 331, + 198 + ], + "score": 1.0, + "content": "Table 1: Architecture specifications of", + "type": "text" + }, + { + "bbox": [ + 332, + 185, + 359, + 197 + ], + "score": 0.45, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 360, + 185, + 434, + 198 + ], + "score": 1.0, + "content": "models for vision.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "table_body", + "bbox": [ + 158, + 203, + 451, + 255 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 158, + 203, + 451, + 255 + ], + "spans": [ + { + "bbox": [ + 158, + 203, + 451, + 255 + ], + "score": 0.974, + "html": "
#LdmodeldffnParams (M)FLOPs (B)Survival Prob
gMLP-Ti301287685.92.71.00
gMLP-S30256153619.58.90.95
gMLP-B30512307273.431.60.80
", + "type": "table", + "image_path": "27f42f69254be8b6e8eeff8fb7a1820daeca371c25a2a48691d8d74cf19caa2f.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 158, + 203, + 451, + 220.33333333333334 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 158, + 220.33333333333334, + 451, + 237.66666666666669 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 158, + 237.66666666666669, + 451, + 255.00000000000003 + ], + "spans": [], + "index": 12 + } + ] + } + ], + "index": 10.0 + }, + { + "type": "text", + "bbox": [ + 106, + 262, + 505, + 362 + ], + "lines": [ + { + "bbox": [ + 106, + 261, + 506, + 276 + ], + "spans": [ + { + "bbox": [ + 106, + 261, + 506, + 276 + ], + "score": 1.0, + "content": "Our ImageNet results are summarized in Table 1 and Figure 2.3 It is interesting to see that gMLPs are", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 273, + 506, + 286 + ], + "spans": [ + { + "bbox": [ + 105, + 273, + 506, + 286 + ], + "score": 1.0, + "content": "comparable with DeiT [8], namely ViT [7] trained using improved regularization. The results suggest", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 285, + 505, + 297 + ], + "spans": [ + { + "bbox": [ + 106, + 285, + 505, + 297 + ], + "score": 1.0, + "content": "that models without self-attention can be as data-efficient as Transformers for image classification. In", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 293, + 506, + 309 + ], + "spans": [ + { + "bbox": [ + 105, + 293, + 506, + 309 + ], + "score": 1.0, + "content": "fact, when the models are properly regularized, their accuracies seem better correlated with capacity", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 306, + 505, + 319 + ], + "spans": [ + { + "bbox": [ + 105, + 306, + 505, + 319 + ], + "score": 1.0, + "content": "instead of the presence of self-attention. Moreover, the accuracy-parameter/FLOPs tradeoff of gMLPs", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 317, + 506, + 330 + ], + "spans": [ + { + "bbox": [ + 105, + 317, + 506, + 330 + ], + "score": 1.0, + "content": "surpasses all concurrently proposed MLP-like architectures [20, 21, 22], which we attribute to the", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 329, + 505, + 340 + ], + "spans": [ + { + "bbox": [ + 106, + 329, + 505, + 340 + ], + "score": 1.0, + "content": "effectiveness of our Spatial Gating Unit (see Table 3 in the next section for an ablation). We also note", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 338, + 506, + 353 + ], + "spans": [ + { + "bbox": [ + 105, + 338, + 130, + 353 + ], + "score": 1.0, + "content": "while", + "type": "text" + }, + { + "bbox": [ + 130, + 339, + 160, + 351 + ], + "score": 0.57, + "content": "{ \\mathrm { g M L P s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 161, + 338, + 506, + 353 + ], + "score": 1.0, + "content": "are competitive with vanilla Transformers, their performance is behind the best existing", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 350, + 413, + 363 + ], + "spans": [ + { + "bbox": [ + 106, + 350, + 413, + 363 + ], + "score": 1.0, + "content": "ConvNet models (e.g., [33, 34]) or hybrid models (e.g., [35, 36, 37, 38, 10]).", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 17, + "bbox_fs": [ + 105, + 261, + 506, + 363 + ] + }, + { + "type": "table", + "bbox": [ + 145, + 387, + 464, + 631 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 208, + 370, + 403, + 382 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 207, + 369, + 404, + 383 + ], + "spans": [ + { + "bbox": [ + 207, + 369, + 404, + 383 + ], + "score": 1.0, + "content": "Table 2: ImageNet-1K results without extra data.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "table_body", + "bbox": [ + 145, + 387, + 464, + 631 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 145, + 387, + 464, + 631 + ], + "spans": [ + { + "bbox": [ + 145, + 387, + 464, + 631 + ], + "score": 0.984, + "html": "
ModelImageNet Top-1 (%)*Input ResolutionParams (M)MAdds (B)
ConvNets
ResNet-152 [16]78.32246011.3
RegNetY-8GF[39]81.7224398.0
EfficientNet-B0 [17]77.122450.39
EfficientNet-B3[17]81.6300121.8
EfficientNet-B7 [17]84.36006637.0
NFNet-F0 [33]83.61927212.4
Transformers
ViT-B/16 [7]77.93848655.4
ViT-L/16 [7]76.5384307190.7
DeiT-Ti [8] (ViT+reg)72.222451.3
DeiT-S [8](ViT+reg)79.8224224.6
DeiT-B [8] (ViT+reg)81.82248617.5
MLP-like†
Mixer-B/16 [20]76.42245912.7
Mixer-B/16 (our setup)77.32245912.7
Mixer-L/16 [20]71.822420744.8
ResMLP-12 [22]76.6224153.0
ResMLP-24 [22]79.4224306.0
ResMLP-36[22]79.7224458.9
gMLP-Ti (ours)72.322461.4
gMLP-S (ours)79.6224204.5
gMLP-B (ours)81.62247315.8
", + "type": "table", + "image_path": "123f13f92d912e05ece7dd96c75519ea83477124b08ac3939ad58b8b69f2b23d.jpg" + } + ] + } + ], + "index": 24, + "virtual_lines": [ + { + "bbox": [ + 145, + 387, + 464, + 468.3333333333333 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 145, + 468.3333333333333, + 464, + 549.6666666666666 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 145, + 549.6666666666666, + 464, + 631.0 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 151, + 633, + 390, + 650 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 152, + 631, + 355, + 642 + ], + "spans": [ + { + "bbox": [ + 152, + 631, + 355, + 642 + ], + "score": 1.0, + "content": "* Standard deviation across multiple independent runs is around 0.1.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 151, + 639, + 392, + 652 + ], + "spans": [ + { + "bbox": [ + 151, + 639, + 392, + 652 + ], + "score": 1.0, + "content": "† Tokenization & embedding process at the stem can be viewed as a convolution.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26.5 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 660, + 505, + 694 + ], + "lines": [ + { + "bbox": [ + 106, + 660, + 506, + 673 + ], + "spans": [ + { + "bbox": [ + 106, + 660, + 506, + 673 + ], + "score": 1.0, + "content": "Figure 3 visualizes the spatial projection matrices in gMLP-B. Remarkably, the spatial weights after", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 671, + 505, + 684 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 505, + 684 + ], + "score": 1.0, + "content": "learning exhibit both locality and spatial invariance. In other words, each spatial projection matrix", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 682, + 493, + 695 + ], + "spans": [ + { + "bbox": [ + 106, + 682, + 493, + 695 + ], + "score": 1.0, + "content": "effectively learns to perform convolution with a data-driven, irregular (non-square) kernel shape.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29, + "bbox_fs": [ + 106, + 660, + 506, + 695 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 111, + 70, + 304, + 254 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 70, + 304, + 254 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 111, + 70, + 304, + 254 + ], + "spans": [ + { + "bbox": [ + 111, + 70, + 304, + 254 + ], + "score": 0.969, + "type": "image", + "image_path": "4ce114a6aa7138cbd566c0e0b5ee0dbdefe6fbfc6e3bb7bd0ddb7399d3c6d83b.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 111, + 70, + 304, + 83.14285714285714 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 111, + 83.14285714285714, + 304, + 96.28571428571428 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 111, + 96.28571428571428, + 304, + 109.42857142857142 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 111, + 109.42857142857142, + 304, + 122.57142857142856 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 111, + 122.57142857142856, + 304, + 135.7142857142857 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 111, + 135.7142857142857, + 304, + 148.85714285714283 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 111, + 148.85714285714283, + 304, + 161.99999999999997 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 111, + 161.99999999999997, + 304, + 175.1428571428571 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 111, + 175.1428571428571, + 304, + 188.28571428571425 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 111, + 188.28571428571425, + 304, + 201.4285714285714 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 111, + 201.4285714285714, + 304, + 214.57142857142853 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 111, + 214.57142857142853, + 304, + 227.71428571428567 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 111, + 227.71428571428567, + 304, + 240.8571428571428 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 111, + 240.8571428571428, + 304, + 253.99999999999994 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 112, + 263, + 304, + 275 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 260, + 306, + 278 + ], + "spans": [ + { + "bbox": [ + 110, + 260, + 306, + 278 + ], + "score": 1.0, + "content": "Figure 2: ImageNet accuracy vs model capacity.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + } + ], + "index": 10.25 + }, + { + "type": "image", + "bbox": [ + 313, + 71, + 503, + 232 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 313, + 71, + 503, + 232 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 313, + 71, + 503, + 232 + ], + "spans": [ + { + "bbox": [ + 313, + 71, + 503, + 232 + ], + "score": 0.972, + "type": "image", + "image_path": "e9c6f27dad6f981df32081593ce223c652c9c19a1ac8f3a7e6353aba3fdced9b.jpg" + } + ] + } + ], + "index": 20.5, + "virtual_lines": [ + { + "bbox": [ + 313, + 71, + 503, + 84.41666666666667 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 313, + 84.41666666666667, + 503, + 97.83333333333334 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 313, + 97.83333333333334, + 503, + 111.25000000000001 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 313, + 111.25000000000001, + 503, + 124.66666666666669 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 313, + 124.66666666666669, + 503, + 138.08333333333334 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 313, + 138.08333333333334, + 503, + 151.5 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 313, + 151.5, + 503, + 164.91666666666666 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 313, + 164.91666666666666, + 503, + 178.33333333333331 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 313, + 178.33333333333331, + 503, + 191.74999999999997 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 313, + 191.74999999999997, + 503, + 205.16666666666663 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 313, + 205.16666666666663, + 503, + 218.5833333333333 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 313, + 218.5833333333333, + 503, + 231.99999999999994 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 314, + 241, + 505, + 275 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 313, + 240, + 507, + 253 + ], + "spans": [ + { + "bbox": [ + 313, + 240, + 507, + 253 + ], + "score": 1.0, + "content": "Figure 3: Spatial projection weights in gMLP-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 313, + 251, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 313, + 251, + 506, + 264 + ], + "score": 1.0, + "content": "B. Each row shows the filters (reshaped into 2D)", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 313, + 262, + 495, + 276 + ], + "spans": [ + { + "bbox": [ + 313, + 262, + 495, + 276 + ], + "score": 1.0, + "content": "for a selected set of tokens in the same layer.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28 + } + ], + "index": 24.25 + }, + { + "type": "title", + "bbox": [ + 107, + 294, + 332, + 308 + ], + "lines": [ + { + "bbox": [ + 104, + 293, + 333, + 311 + ], + "spans": [ + { + "bbox": [ + 104, + 293, + 333, + 311 + ], + "score": 1.0, + "content": "4 Masked Language Modeling with BERT", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 106, + 319, + 505, + 397 + ], + "lines": [ + { + "bbox": [ + 106, + 320, + 505, + 332 + ], + "spans": [ + { + "bbox": [ + 106, + 320, + 505, + 332 + ], + "score": 1.0, + "content": "Here we conduct empirical studies over the masked language modeling (MLM) task. The input/output", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 330, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 330, + 505, + 343 + ], + "score": 1.0, + "content": "protocol for both pretraining and finetuning follows BERT [2]. Different from Transformer-based", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 342, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 505, + 354 + ], + "score": 1.0, + "content": "models, we do not use positional encodings. We also find it unnecessary to mask out tokens", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 353, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 506, + 365 + ], + "score": 1.0, + "content": "in gMLP blocks during finetuning as the model can quickly learn to ignore them. For ablations and", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 363, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 505, + 375 + ], + "score": 1.0, + "content": "case studies, all models are trained with batch size 2048, max length 128 for 125K steps over the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 375, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 106, + 375, + 505, + 386 + ], + "score": 1.0, + "content": "RealNews-like subset of C4 [5]. For main results, models are trained with batch size 256, max length", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 385, + 430, + 398 + ], + "spans": [ + { + "bbox": [ + 106, + 385, + 430, + 398 + ], + "score": 1.0, + "content": "512 for 1M steps over the full English C4 dataset. See Appendix A.2 for details.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 107, + 401, + 505, + 489 + ], + "lines": [ + { + "bbox": [ + 105, + 401, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 505, + 414 + ], + "score": 1.0, + "content": "Our preliminary MLM experiments show that gMLPs always learn Toeplitz-like matrices as the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 412, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 273, + 424 + ], + "score": 1.0, + "content": "spatial weights (Appendix C). This means", + "type": "text" + }, + { + "bbox": [ + 273, + 413, + 303, + 424 + ], + "score": 0.26, + "content": "{ \\mathrm { g M L P s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 412, + 505, + 424 + ], + "score": 1.0, + "content": "are able to learn the notion of shift invariance from", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 424, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 505, + 436 + ], + "score": 1.0, + "content": "data, a property naturally implied by the MLM task where any offset of the input sequence does not", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 434, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 333, + 447 + ], + "score": 1.0, + "content": "affect the slot filling outcome. In this case, the learned", + "type": "text" + }, + { + "bbox": [ + 334, + 434, + 363, + 446 + ], + "score": 0.92, + "content": "f _ { W , b } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 434, + 505, + 447 + ], + "score": 1.0, + "content": "acts like a 1-d convolution whose", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 445, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 505, + 457 + ], + "score": 1.0, + "content": "kernel size equals the entire sequence length (unlike depthwise convolution with channel-specific", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 191, + 469 + ], + "score": 1.0, + "content": "filters, here the same", + "type": "text" + }, + { + "bbox": [ + 192, + 456, + 204, + 466 + ], + "score": 0.36, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 456, + 505, + 469 + ], + "score": 1.0, + "content": "is shared across channels). In the following MLM experiments, we restrict", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 107, + 467, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 107, + 467, + 119, + 477 + ], + "score": 0.42, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 467, + 409, + 479 + ], + "score": 1.0, + "content": "to be a Toeplitz matrix to avoid redundant model parameterization (since", + "type": "text" + }, + { + "bbox": [ + 410, + 467, + 422, + 477 + ], + "score": 0.47, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 467, + 505, + 479 + ], + "score": 1.0, + "content": "will be Toeplitz-like", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 478, + 412, + 490 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 412, + 490 + ], + "score": 1.0, + "content": "regardless after learning). Note this constraint is empirically quality-neutral.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 41.5 + }, + { + "type": "title", + "bbox": [ + 106, + 502, + 428, + 515 + ], + "lines": [ + { + "bbox": [ + 105, + 501, + 430, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 430, + 518 + ], + "score": 1.0, + "content": "4.1 Ablation: The Importance of Gating in gMLP for BERT’s Pretraining", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 46 + }, + { + "type": "text", + "bbox": [ + 106, + 523, + 426, + 535 + ], + "lines": [ + { + "bbox": [ + 106, + 523, + 426, + 536 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 426, + 536 + ], + "score": 1.0, + "content": "In Table 3 below, we establish baselines for our ablation studies. These include:", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 47 + }, + { + "type": "text", + "bbox": [ + 129, + 544, + 506, + 658 + ], + "lines": [ + { + "bbox": [ + 129, + 544, + 477, + 557 + ], + "spans": [ + { + "bbox": [ + 129, + 544, + 477, + 557 + ], + "score": 1.0, + "content": "1. BERT with a Transformer architecture and learnable absolute position embeddings.", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 129, + 560, + 506, + 573 + ], + "spans": [ + { + "bbox": [ + 129, + 560, + 506, + 573 + ], + "score": 1.0, + "content": "2. BERT with a Transformer architecture and T5-style learnable relative position biases [5].", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 142, + 570, + 469, + 583 + ], + "spans": [ + { + "bbox": [ + 142, + 570, + 469, + 583 + ], + "score": 1.0, + "content": "The biases are both layer- and head-specific as we find this yields the best results.", + "type": "text" + } + ], + "index": 50 + }, + { + "bbox": [ + 129, + 586, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 129, + 586, + 505, + 599 + ], + "score": 1.0, + "content": "3. Same as above, but we remove all content-dependent terms inside the softmax and only retain", + "type": "text" + } + ], + "index": 51 + }, + { + "bbox": [ + 142, + 599, + 504, + 609 + ], + "spans": [ + { + "bbox": [ + 142, + 599, + 504, + 609 + ], + "score": 1.0, + "content": "the relative positional biases. This baseline is a straightforward variant of Transformers", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 142, + 609, + 461, + 621 + ], + "spans": [ + { + "bbox": [ + 142, + 609, + 461, + 621 + ], + "score": 1.0, + "content": "without self-attention, which can also be viewed as a Random Synthesizer [40].", + "type": "text" + } + ], + "index": 53 + }, + { + "bbox": [ + 128, + 624, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 128, + 624, + 505, + 636 + ], + "score": 1.0, + "content": "4. MLP-Mixer [20] which replaces the multi-head self-attention module in Transformers with", + "type": "text" + } + ], + "index": 54 + }, + { + "bbox": [ + 141, + 635, + 506, + 648 + ], + "spans": [ + { + "bbox": [ + 141, + 635, + 506, + 648 + ], + "score": 1.0, + "content": "a two-layer spatial MLP. This model was developed for image classification and here we", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 141, + 646, + 469, + 659 + ], + "spans": [ + { + "bbox": [ + 141, + 646, + 469, + 659 + ], + "score": 1.0, + "content": "investigate it on MLM tasks using the same training setup with BERT and gMLP.", + "type": "text" + } + ], + "index": 56 + } + ], + "index": 52 + }, + { + "type": "text", + "bbox": [ + 107, + 667, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 667, + 505, + 679 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 321, + 679 + ], + "score": 1.0, + "content": "We compare these baselines against several versions of", + "type": "text" + }, + { + "bbox": [ + 322, + 667, + 352, + 679 + ], + "score": 0.33, + "content": "{ \\mathrm { g M L P s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 667, + 505, + 679 + ], + "score": 1.0, + "content": "with similar sizes in Table 3. Note that", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "score": 1.0, + "content": "Multiplicative, Split (last row) is the Spatial Gating Unit we describe in the method section and use in", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "the rest of the paper. First, SGU outperforms other variants in perplexity. Secondly and remarkably,", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 106, + 699, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 713 + ], + "score": 1.0, + "content": "gMLP with SGU also achieves perplexity comparable to Transformer. Note the difference between", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 106, + 711, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 242, + 724 + ], + "score": 1.0, + "content": "the strongest baseline (perplexity", + "type": "text" + }, + { + "bbox": [ + 242, + 711, + 268, + 721 + ], + "score": 0.46, + "content": "{ = } 4 . 2 6 ", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 711, + 354, + 724 + ], + "score": 1.0, + "content": ") and ours (perplexity", + "type": "text" + }, + { + "bbox": [ + 354, + 711, + 379, + 721 + ], + "score": 0.42, + "content": "{ = } 4 . 3 5", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 711, + 505, + 724 + ], + "score": 1.0, + "content": ") is insignificant relative to the", + "type": "text" + } + ], + "index": 61 + } + ], + "index": 59 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 740, + 309, + 753 + ], + "spans": [ + { + "bbox": [ + 302, + 740, + 309, + 753 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 111, + 70, + 304, + 254 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 111, + 70, + 304, + 254 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 111, + 70, + 304, + 254 + ], + "spans": [ + { + "bbox": [ + 111, + 70, + 304, + 254 + ], + "score": 0.969, + "type": "image", + "image_path": "4ce114a6aa7138cbd566c0e0b5ee0dbdefe6fbfc6e3bb7bd0ddb7399d3c6d83b.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 111, + 70, + 304, + 83.14285714285714 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 111, + 83.14285714285714, + 304, + 96.28571428571428 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 111, + 96.28571428571428, + 304, + 109.42857142857142 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 111, + 109.42857142857142, + 304, + 122.57142857142856 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 111, + 122.57142857142856, + 304, + 135.7142857142857 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 111, + 135.7142857142857, + 304, + 148.85714285714283 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 111, + 148.85714285714283, + 304, + 161.99999999999997 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 111, + 161.99999999999997, + 304, + 175.1428571428571 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 111, + 175.1428571428571, + 304, + 188.28571428571425 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 111, + 188.28571428571425, + 304, + 201.4285714285714 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 111, + 201.4285714285714, + 304, + 214.57142857142853 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 111, + 214.57142857142853, + 304, + 227.71428571428567 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 111, + 227.71428571428567, + 304, + 240.8571428571428 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 111, + 240.8571428571428, + 304, + 253.99999999999994 + ], + "spans": [], + "index": 13 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 112, + 263, + 304, + 275 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 110, + 260, + 306, + 278 + ], + "spans": [ + { + "bbox": [ + 110, + 260, + 306, + 278 + ], + "score": 1.0, + "content": "Figure 2: ImageNet accuracy vs model capacity.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 14 + } + ], + "index": 10.25 + }, + { + "type": "image", + "bbox": [ + 313, + 71, + 503, + 232 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 313, + 71, + 503, + 232 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 313, + 71, + 503, + 232 + ], + "spans": [ + { + "bbox": [ + 313, + 71, + 503, + 232 + ], + "score": 0.972, + "type": "image", + "image_path": "e9c6f27dad6f981df32081593ce223c652c9c19a1ac8f3a7e6353aba3fdced9b.jpg" + } + ] + } + ], + "index": 20.5, + "virtual_lines": [ + { + "bbox": [ + 313, + 71, + 503, + 84.41666666666667 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 313, + 84.41666666666667, + 503, + 97.83333333333334 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 313, + 97.83333333333334, + 503, + 111.25000000000001 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 313, + 111.25000000000001, + 503, + 124.66666666666669 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 313, + 124.66666666666669, + 503, + 138.08333333333334 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 313, + 138.08333333333334, + 503, + 151.5 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 313, + 151.5, + 503, + 164.91666666666666 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 313, + 164.91666666666666, + 503, + 178.33333333333331 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 313, + 178.33333333333331, + 503, + 191.74999999999997 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 313, + 191.74999999999997, + 503, + 205.16666666666663 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 313, + 205.16666666666663, + 503, + 218.5833333333333 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 313, + 218.5833333333333, + 503, + 231.99999999999994 + ], + "spans": [], + "index": 26 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 314, + 241, + 505, + 275 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 313, + 240, + 507, + 253 + ], + "spans": [ + { + "bbox": [ + 313, + 240, + 507, + 253 + ], + "score": 1.0, + "content": "Figure 3: Spatial projection weights in gMLP-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 313, + 251, + 506, + 264 + ], + "spans": [ + { + "bbox": [ + 313, + 251, + 506, + 264 + ], + "score": 1.0, + "content": "B. Each row shows the filters (reshaped into 2D)", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 313, + 262, + 495, + 276 + ], + "spans": [ + { + "bbox": [ + 313, + 262, + 495, + 276 + ], + "score": 1.0, + "content": "for a selected set of tokens in the same layer.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28 + } + ], + "index": 24.25 + }, + { + "type": "title", + "bbox": [ + 107, + 294, + 332, + 308 + ], + "lines": [ + { + "bbox": [ + 104, + 293, + 333, + 311 + ], + "spans": [ + { + "bbox": [ + 104, + 293, + 333, + 311 + ], + "score": 1.0, + "content": "4 Masked Language Modeling with BERT", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 106, + 319, + 505, + 397 + ], + "lines": [ + { + "bbox": [ + 106, + 320, + 505, + 332 + ], + "spans": [ + { + "bbox": [ + 106, + 320, + 505, + 332 + ], + "score": 1.0, + "content": "Here we conduct empirical studies over the masked language modeling (MLM) task. The input/output", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 330, + 505, + 343 + ], + "spans": [ + { + "bbox": [ + 105, + 330, + 505, + 343 + ], + "score": 1.0, + "content": "protocol for both pretraining and finetuning follows BERT [2]. Different from Transformer-based", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 342, + 505, + 354 + ], + "spans": [ + { + "bbox": [ + 106, + 342, + 505, + 354 + ], + "score": 1.0, + "content": "models, we do not use positional encodings. We also find it unnecessary to mask out tokens", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 353, + 506, + 365 + ], + "spans": [ + { + "bbox": [ + 105, + 353, + 506, + 365 + ], + "score": 1.0, + "content": "in gMLP blocks during finetuning as the model can quickly learn to ignore them. For ablations and", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 105, + 363, + 505, + 375 + ], + "spans": [ + { + "bbox": [ + 105, + 363, + 505, + 375 + ], + "score": 1.0, + "content": "case studies, all models are trained with batch size 2048, max length 128 for 125K steps over the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 106, + 375, + 505, + 386 + ], + "spans": [ + { + "bbox": [ + 106, + 375, + 505, + 386 + ], + "score": 1.0, + "content": "RealNews-like subset of C4 [5]. For main results, models are trained with batch size 256, max length", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 106, + 385, + 430, + 398 + ], + "spans": [ + { + "bbox": [ + 106, + 385, + 430, + 398 + ], + "score": 1.0, + "content": "512 for 1M steps over the full English C4 dataset. See Appendix A.2 for details.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 34, + "bbox_fs": [ + 105, + 320, + 506, + 398 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 401, + 505, + 489 + ], + "lines": [ + { + "bbox": [ + 105, + 401, + 505, + 414 + ], + "spans": [ + { + "bbox": [ + 105, + 401, + 505, + 414 + ], + "score": 1.0, + "content": "Our preliminary MLM experiments show that gMLPs always learn Toeplitz-like matrices as the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 105, + 412, + 505, + 424 + ], + "spans": [ + { + "bbox": [ + 105, + 412, + 273, + 424 + ], + "score": 1.0, + "content": "spatial weights (Appendix C). This means", + "type": "text" + }, + { + "bbox": [ + 273, + 413, + 303, + 424 + ], + "score": 0.26, + "content": "{ \\mathrm { g M L P s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 412, + 505, + 424 + ], + "score": 1.0, + "content": "are able to learn the notion of shift invariance from", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 105, + 424, + 505, + 436 + ], + "spans": [ + { + "bbox": [ + 105, + 424, + 505, + 436 + ], + "score": 1.0, + "content": "data, a property naturally implied by the MLM task where any offset of the input sequence does not", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 105, + 434, + 505, + 447 + ], + "spans": [ + { + "bbox": [ + 105, + 434, + 333, + 447 + ], + "score": 1.0, + "content": "affect the slot filling outcome. In this case, the learned", + "type": "text" + }, + { + "bbox": [ + 334, + 434, + 363, + 446 + ], + "score": 0.92, + "content": "f _ { W , b } ( \\cdot )", + "type": "inline_equation" + }, + { + "bbox": [ + 364, + 434, + 505, + 447 + ], + "score": 1.0, + "content": "acts like a 1-d convolution whose", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 445, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 105, + 445, + 505, + 457 + ], + "score": 1.0, + "content": "kernel size equals the entire sequence length (unlike depthwise convolution with channel-specific", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 456, + 505, + 469 + ], + "spans": [ + { + "bbox": [ + 105, + 456, + 191, + 469 + ], + "score": 1.0, + "content": "filters, here the same", + "type": "text" + }, + { + "bbox": [ + 192, + 456, + 204, + 466 + ], + "score": 0.36, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 456, + 505, + 469 + ], + "score": 1.0, + "content": "is shared across channels). In the following MLM experiments, we restrict", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 107, + 467, + 505, + 479 + ], + "spans": [ + { + "bbox": [ + 107, + 467, + 119, + 477 + ], + "score": 0.42, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 119, + 467, + 409, + 479 + ], + "score": 1.0, + "content": "to be a Toeplitz matrix to avoid redundant model parameterization (since", + "type": "text" + }, + { + "bbox": [ + 410, + 467, + 422, + 477 + ], + "score": 0.47, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 422, + 467, + 505, + 479 + ], + "score": 1.0, + "content": "will be Toeplitz-like", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 105, + 478, + 412, + 490 + ], + "spans": [ + { + "bbox": [ + 105, + 478, + 412, + 490 + ], + "score": 1.0, + "content": "regardless after learning). Note this constraint is empirically quality-neutral.", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 41.5, + "bbox_fs": [ + 105, + 401, + 505, + 490 + ] + }, + { + "type": "title", + "bbox": [ + 106, + 502, + 428, + 515 + ], + "lines": [ + { + "bbox": [ + 105, + 501, + 430, + 518 + ], + "spans": [ + { + "bbox": [ + 105, + 501, + 430, + 518 + ], + "score": 1.0, + "content": "4.1 Ablation: The Importance of Gating in gMLP for BERT’s Pretraining", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 46 + }, + { + "type": "text", + "bbox": [ + 106, + 523, + 426, + 535 + ], + "lines": [ + { + "bbox": [ + 106, + 523, + 426, + 536 + ], + "spans": [ + { + "bbox": [ + 106, + 523, + 426, + 536 + ], + "score": 1.0, + "content": "In Table 3 below, we establish baselines for our ablation studies. These include:", + "type": "text" + } + ], + "index": 47 + } + ], + "index": 47, + "bbox_fs": [ + 106, + 523, + 426, + 536 + ] + }, + { + "type": "list", + "bbox": [ + 129, + 544, + 506, + 658 + ], + "lines": [ + { + "bbox": [ + 129, + 544, + 477, + 557 + ], + "spans": [ + { + "bbox": [ + 129, + 544, + 477, + 557 + ], + "score": 1.0, + "content": "1. BERT with a Transformer architecture and learnable absolute position embeddings.", + "type": "text" + } + ], + "index": 48, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 129, + 560, + 506, + 573 + ], + "spans": [ + { + "bbox": [ + 129, + 560, + 506, + 573 + ], + "score": 1.0, + "content": "2. BERT with a Transformer architecture and T5-style learnable relative position biases [5].", + "type": "text" + } + ], + "index": 49, + "is_list_start_line": true + }, + { + "bbox": [ + 142, + 570, + 469, + 583 + ], + "spans": [ + { + "bbox": [ + 142, + 570, + 469, + 583 + ], + "score": 1.0, + "content": "The biases are both layer- and head-specific as we find this yields the best results.", + "type": "text" + } + ], + "index": 50, + "is_list_end_line": true + }, + { + "bbox": [ + 129, + 586, + 505, + 599 + ], + "spans": [ + { + "bbox": [ + 129, + 586, + 505, + 599 + ], + "score": 1.0, + "content": "3. Same as above, but we remove all content-dependent terms inside the softmax and only retain", + "type": "text" + } + ], + "index": 51, + "is_list_start_line": true + }, + { + "bbox": [ + 142, + 599, + 504, + 609 + ], + "spans": [ + { + "bbox": [ + 142, + 599, + 504, + 609 + ], + "score": 1.0, + "content": "the relative positional biases. This baseline is a straightforward variant of Transformers", + "type": "text" + } + ], + "index": 52 + }, + { + "bbox": [ + 142, + 609, + 461, + 621 + ], + "spans": [ + { + "bbox": [ + 142, + 609, + 461, + 621 + ], + "score": 1.0, + "content": "without self-attention, which can also be viewed as a Random Synthesizer [40].", + "type": "text" + } + ], + "index": 53, + "is_list_end_line": true + }, + { + "bbox": [ + 128, + 624, + 505, + 636 + ], + "spans": [ + { + "bbox": [ + 128, + 624, + 505, + 636 + ], + "score": 1.0, + "content": "4. MLP-Mixer [20] which replaces the multi-head self-attention module in Transformers with", + "type": "text" + } + ], + "index": 54, + "is_list_start_line": true + }, + { + "bbox": [ + 141, + 635, + 506, + 648 + ], + "spans": [ + { + "bbox": [ + 141, + 635, + 506, + 648 + ], + "score": 1.0, + "content": "a two-layer spatial MLP. This model was developed for image classification and here we", + "type": "text" + } + ], + "index": 55 + }, + { + "bbox": [ + 141, + 646, + 469, + 659 + ], + "spans": [ + { + "bbox": [ + 141, + 646, + 469, + 659 + ], + "score": 1.0, + "content": "investigate it on MLM tasks using the same training setup with BERT and gMLP.", + "type": "text" + } + ], + "index": 56, + "is_list_end_line": true + } + ], + "index": 52, + "bbox_fs": [ + 128, + 544, + 506, + 659 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 667, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 667, + 505, + 679 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 321, + 679 + ], + "score": 1.0, + "content": "We compare these baselines against several versions of", + "type": "text" + }, + { + "bbox": [ + 322, + 667, + 352, + 679 + ], + "score": 0.33, + "content": "{ \\mathrm { g M L P s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 352, + 667, + 505, + 679 + ], + "score": 1.0, + "content": "with similar sizes in Table 3. Note that", + "type": "text" + } + ], + "index": 57 + }, + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "score": 1.0, + "content": "Multiplicative, Split (last row) is the Spatial Gating Unit we describe in the method section and use in", + "type": "text" + } + ], + "index": 58 + }, + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 688, + 506, + 702 + ], + "score": 1.0, + "content": "the rest of the paper. First, SGU outperforms other variants in perplexity. Secondly and remarkably,", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 106, + 699, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 713 + ], + "score": 1.0, + "content": "gMLP with SGU also achieves perplexity comparable to Transformer. Note the difference between", + "type": "text" + } + ], + "index": 60 + }, + { + "bbox": [ + 106, + 711, + 505, + 724 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 242, + 724 + ], + "score": 1.0, + "content": "the strongest baseline (perplexity", + "type": "text" + }, + { + "bbox": [ + 242, + 711, + 268, + 721 + ], + "score": 0.46, + "content": "{ = } 4 . 2 6 ", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 711, + 354, + 724 + ], + "score": 1.0, + "content": ") and ours (perplexity", + "type": "text" + }, + { + "bbox": [ + 354, + 711, + 379, + 721 + ], + "score": 0.42, + "content": "{ = } 4 . 3 5", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 711, + 505, + 724 + ], + "score": 1.0, + "content": ") is insignificant relative to the", + "type": "text" + } + ], + "index": 61 + }, + { + "bbox": [ + 106, + 247, + 505, + 262 + ], + "spans": [ + { + "bbox": [ + 106, + 247, + 505, + 262 + ], + "score": 1.0, + "content": "perplexity change when the models are scaled (see Table 4 in the next section). Spatial projection", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 259, + 322, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 259, + 183, + 272 + ], + "score": 1.0, + "content": "weights learned by", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 184, + 259, + 214, + 271 + ], + "score": 0.6, + "content": "{ \\mathrm { g M L P s } }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 215, + 259, + 322, + 272 + ], + "score": 1.0, + "content": "are visualized in Figure 4.", + "type": "text", + "cross_page": true + } + ], + "index": 9 + } + ], + "index": 59, + "bbox_fs": [ + 105, + 667, + 506, + 724 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 136, + 120, + 473, + 232 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 69, + 506, + 114 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 104, + 68, + 505, + 82 + ], + "spans": [ + { + "bbox": [ + 104, + 68, + 437, + 82 + ], + "score": 1.0, + "content": "Table 3: MLM validation perplexities of Transformer baselines and four versions of", + "type": "text" + }, + { + "bbox": [ + 438, + 70, + 467, + 81 + ], + "score": 0.42, + "content": "{ \\mathrm { g M L P s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 68, + 471, + 82 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 472, + 70, + 479, + 81 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 68, + 505, + 82 + ], + "score": 1.0, + "content": "refers", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 81, + 505, + 92 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 92 + ], + "score": 1.0, + "content": "to the spatial linear projection in Equation (2) with input normalization. The MLP-Mixer baseline", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 91, + 505, + 105 + ], + "spans": [ + { + "bbox": [ + 105, + 91, + 149, + 105 + ], + "score": 1.0, + "content": "model has", + "type": "text" + }, + { + "bbox": [ + 149, + 92, + 173, + 102 + ], + "score": 0.81, + "content": "_ { \\mathrm { L } = 2 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 91, + 220, + 105 + ], + "score": 1.0, + "content": "layers with", + "type": "text" + }, + { + "bbox": [ + 221, + 92, + 269, + 102 + ], + "score": 0.89, + "content": "d _ { \\mathrm { m o d e l } } = 7 6 8", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 91, + 272, + 105 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 273, + 92, + 324, + 103 + ], + "score": 0.89, + "content": "d _ { \\mathrm { s p a t i a l } } { = } 3 8 4", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 91, + 341, + 105 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 342, + 92, + 383, + 102 + ], + "score": 0.91, + "content": "d _ { \\mathrm { f f n } } { = } 3 0 7 2", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 91, + 480, + 105 + ], + "score": 1.0, + "content": ". Each gMLP model has", + "type": "text" + }, + { + "bbox": [ + 480, + 92, + 505, + 102 + ], + "score": 0.78, + "content": "_ { \\mathrm { L } = 3 6 }", + "type": "inline_equation" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 102, + 491, + 115 + ], + "spans": [ + { + "bbox": [ + 106, + 103, + 153, + 115 + ], + "score": 1.0, + "content": "layers with", + "type": "text" + }, + { + "bbox": [ + 153, + 102, + 202, + 114 + ], + "score": 0.9, + "content": "d _ { \\mathrm { m o d e l } } = 5 1 2", + "type": "inline_equation" + }, + { + "bbox": [ + 202, + 103, + 219, + 115 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 220, + 103, + 267, + 114 + ], + "score": 0.88, + "content": "d _ { \\mathrm { f f n } } = 3 0 7 2", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 103, + 491, + 115 + ], + "score": 1.0, + "content": ". No positional encodings are used for Mixer or gMLPs.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "table_body", + "bbox": [ + 136, + 120, + 473, + 232 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 136, + 120, + 473, + 232 + ], + "spans": [ + { + "bbox": [ + 136, + 120, + 473, + 232 + ], + "score": 0.979, + "html": "
ModelPerplexity*Params (M)
BERTbase4.37110
BERTbase + rel pos4.26110
BERTbase + rel pos - attn5.6496
MLP-Mixer5.34112
Linear gMLP,s(Z)= f(Z)5.1492
Additive gMLP,s(Z)= Z+ f(Z)4.9792
Multiplicative gMLP,s(Z) = Z f(Z)4.5392
Multiplicative,Split gMLP,s(Z) = Z1 f(Z2),Z= Zi||Z24.35102
", + "type": "table", + "image_path": "230eb41e660952154dd9abe33bc6dbc3d8c04a14f867dc0c2c621db29dfc52f1.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 136, + 120, + 473, + 157.33333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 136, + 157.33333333333334, + 473, + 194.66666666666669 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 136, + 194.66666666666669, + 473, + 232.00000000000003 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 143, + 234, + 396, + 245 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 147, + 233, + 396, + 246 + ], + "spans": [ + { + "bbox": [ + 147, + 233, + 396, + 246 + ], + "score": 1.0, + "content": "Standard deviation across multiple independent runs is around 0.01.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 109, + 248, + 504, + 271 + ], + "lines": [ + { + "bbox": [ + 106, + 247, + 505, + 262 + ], + "spans": [ + { + "bbox": [ + 106, + 247, + 505, + 262 + ], + "score": 1.0, + "content": "perplexity change when the models are scaled (see Table 4 in the next section). Spatial projection", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 106, + 259, + 322, + 272 + ], + "spans": [ + { + "bbox": [ + 106, + 259, + 183, + 272 + ], + "score": 1.0, + "content": "weights learned by", + "type": "text" + }, + { + "bbox": [ + 184, + 259, + 214, + 271 + ], + "score": 0.6, + "content": "{ \\mathrm { g M L P s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 215, + 259, + 322, + 272 + ], + "score": 1.0, + "content": "are visualized in Figure 4.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "image", + "bbox": [ + 162, + 280, + 450, + 351 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 162, + 280, + 450, + 351 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 162, + 280, + 450, + 351 + ], + "spans": [ + { + "bbox": [ + 162, + 280, + 450, + 351 + ], + "score": 0.962, + "type": "image", + "image_path": "731fe5bbb1001c5b9f64529e6e9ed073521439ba52c88469a01d1124d0dfaa34.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 162, + 280, + 450, + 303.6666666666667 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 162, + 303.6666666666667, + 450, + 327.33333333333337 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 162, + 327.33333333333337, + 450, + 351.00000000000006 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 360, + 506, + 416 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 361, + 505, + 372 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 288, + 372 + ], + "score": 1.0, + "content": "Figure 4: Visualization of the spatial filters in", + "type": "text" + }, + { + "bbox": [ + 288, + 361, + 315, + 372 + ], + "score": 0.5, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 361, + 505, + 372 + ], + "score": 1.0, + "content": "learned on the MLM task. For each layer in the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 371, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 371, + 209, + 383 + ], + "score": 1.0, + "content": "model we plot the row in", + "type": "text" + }, + { + "bbox": [ + 209, + 371, + 221, + 381 + ], + "score": 0.5, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 371, + 466, + 383 + ], + "score": 1.0, + "content": "associated with the token in the middle of the sequence. The", + "type": "text" + }, + { + "bbox": [ + 467, + 373, + 473, + 381 + ], + "score": 0.41, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 371, + 506, + 383 + ], + "score": 1.0, + "content": "-axis of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 383, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 505, + 394 + ], + "score": 1.0, + "content": "each subplot has a length of 128 which equal the number of tokens in the sequence. The learned filters", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 392, + 506, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 506, + 407 + ], + "score": 1.0, + "content": "appear to be smooth and have several types: forward-looking (e.g., 1st in 2nd row), backward-looking", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 403, + 387, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 387, + 416 + ], + "score": 1.0, + "content": "(e.g., 5th in 2nd row) and bi-directional (e.g., 2nd last in the last row).", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15 + } + ], + "index": 13.0 + }, + { + "type": "title", + "bbox": [ + 107, + 426, + 386, + 439 + ], + "lines": [ + { + "bbox": [ + 106, + 426, + 387, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 254, + 441 + ], + "score": 1.0, + "content": "4.2 Case Study: The Behavior of", + "type": "text" + }, + { + "bbox": [ + 254, + 427, + 283, + 439 + ], + "score": 0.29, + "content": "\\mathbf { g } \\mathbf { M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 426, + 387, + 441 + ], + "score": 1.0, + "content": "as Model Size Increases", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 447, + 505, + 513 + ], + "lines": [ + { + "bbox": [ + 105, + 447, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 505, + 460 + ], + "score": 1.0, + "content": "In Table 4, we investigate the scaling properties of Transformers and gMLPs in BERT as their model", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 458, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 458, + 427, + 471 + ], + "score": 1.0, + "content": "capacity grows. Specifically, we scale the depth of these models by a factor of", + "type": "text" + }, + { + "bbox": [ + 427, + 458, + 487, + 470 + ], + "score": 0.92, + "content": "\\{ 0 . 5 , 1 , 2 , 4 \\} \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 458, + 505, + 471 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 469, + 505, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 505, + 482 + ], + "score": 1.0, + "content": "report the their pretraining MLM perplexities on the validation set as well as finetuning results on", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 479, + 506, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 506, + 493 + ], + "score": 1.0, + "content": "the dev sets of two tasks in GLUE [41]. Note each individual Transformer layer is effectively two", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 492, + 506, + 503 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 506, + 503 + ], + "score": 1.0, + "content": "consecutive blocks: one for self-attention and one for FFN. In the table below we use the notation of", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 501, + 497, + 514 + ], + "spans": [ + { + "bbox": [ + 106, + 502, + 138, + 513 + ], + "score": 0.87, + "content": "1 2 + 1 2", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 501, + 497, + 514 + ], + "score": 1.0, + "content": "to refer to 12 of self-attention blocks plus 12 of FFN blocks in the Transformer baselines.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21.5 + }, + { + "type": "table", + "bbox": [ + 171, + 550, + 437, + 667 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 521, + 502, + 544 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 521, + 504, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 504, + 534 + ], + "score": 1.0, + "content": "Table 4: Pretraining and dev-set finetuning results over increased model capacity. We use the relative", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 532, + 433, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 433, + 545 + ], + "score": 1.0, + "content": "positional encoding scheme for Transformers which performs the best in Table 3.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "table_body", + "bbox": [ + 171, + 550, + 437, + 667 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 171, + 550, + 437, + 667 + ], + "spans": [ + { + "bbox": [ + 171, + 550, + 437, + 667 + ], + "score": 0.98, + "html": "
Model#LParams (M) PerplexitySST-2 MNLI-m
TransformergMLP6+61867594.915.2590.4 81.591.2 77.7
TransformergMLP12+12361101024.264.3591.3 83.392.3 80.9
TransformergMLP24+24721951873.833.7992.1 85.293.5 82.8
TransformergMLP48+481443653573.473.4392.8 86.395.1 84.6
", + "type": "table", + "image_path": "f9b9937d357c9507e5ed75ff7c559cbe2b327ae203d9f1866ee2864e0fbaa0c9.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 171, + 550, + 437, + 589.0 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 171, + 589.0, + 437, + 628.0 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 171, + 628.0, + 437, + 667.0 + ], + "spans": [], + "index": 29 + } + ] + } + ], + "index": 26.75 + }, + { + "type": "text", + "bbox": [ + 108, + 672, + 505, + 695 + ], + "lines": [ + { + "bbox": [ + 106, + 671, + 505, + 686 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 505, + 686 + ], + "score": 1.0, + "content": "The results above show that a deep enough gMLP is able to match and even outperform the perplexity", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 682, + 506, + 696 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 506, + 696 + ], + "score": 1.0, + "content": "of Transformers with comparable capacity.4 In addition, the perplexity-parameter relationships for", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 30.5 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 106, + 701, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 118, + 699, + 505, + 714 + ], + "spans": [ + { + "bbox": [ + 118, + 699, + 505, + 714 + ], + "score": 1.0, + "content": "4We also experimented with deeper-and-thinner Transformers (with capacity fixed) but found increasing", + "type": "text" + } + ] + }, + { + "bbox": [ + 106, + 711, + 383, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 711, + 383, + 723 + ], + "score": 1.0, + "content": "depth further does not improve perplexity. See Appendix B for more details.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 302, + 741, + 309, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 136, + 120, + 473, + 232 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 106, + 69, + 506, + 114 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 104, + 68, + 505, + 82 + ], + "spans": [ + { + "bbox": [ + 104, + 68, + 437, + 82 + ], + "score": 1.0, + "content": "Table 3: MLM validation perplexities of Transformer baselines and four versions of", + "type": "text" + }, + { + "bbox": [ + 438, + 70, + 467, + 81 + ], + "score": 0.42, + "content": "{ \\mathrm { g M L P s } }", + "type": "inline_equation" + }, + { + "bbox": [ + 468, + 68, + 471, + 82 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 472, + 70, + 479, + 81 + ], + "score": 0.84, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 68, + 505, + 82 + ], + "score": 1.0, + "content": "refers", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 81, + 505, + 92 + ], + "spans": [ + { + "bbox": [ + 105, + 81, + 505, + 92 + ], + "score": 1.0, + "content": "to the spatial linear projection in Equation (2) with input normalization. The MLP-Mixer baseline", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 91, + 505, + 105 + ], + "spans": [ + { + "bbox": [ + 105, + 91, + 149, + 105 + ], + "score": 1.0, + "content": "model has", + "type": "text" + }, + { + "bbox": [ + 149, + 92, + 173, + 102 + ], + "score": 0.81, + "content": "_ { \\mathrm { L } = 2 4 }", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 91, + 220, + 105 + ], + "score": 1.0, + "content": "layers with", + "type": "text" + }, + { + "bbox": [ + 221, + 92, + 269, + 102 + ], + "score": 0.89, + "content": "d _ { \\mathrm { m o d e l } } = 7 6 8", + "type": "inline_equation" + }, + { + "bbox": [ + 269, + 91, + 272, + 105 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 273, + 92, + 324, + 103 + ], + "score": 0.89, + "content": "d _ { \\mathrm { s p a t i a l } } { = } 3 8 4", + "type": "inline_equation" + }, + { + "bbox": [ + 324, + 91, + 341, + 105 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 342, + 92, + 383, + 102 + ], + "score": 0.91, + "content": "d _ { \\mathrm { f f n } } { = } 3 0 7 2", + "type": "inline_equation" + }, + { + "bbox": [ + 383, + 91, + 480, + 105 + ], + "score": 1.0, + "content": ". Each gMLP model has", + "type": "text" + }, + { + "bbox": [ + 480, + 92, + 505, + 102 + ], + "score": 0.78, + "content": "_ { \\mathrm { L } = 3 6 }", + "type": "inline_equation" + } + ], + "index": 2 + }, + { + "bbox": [ + 106, + 102, + 491, + 115 + ], + "spans": [ + { + "bbox": [ + 106, + 103, + 153, + 115 + ], + "score": 1.0, + "content": "layers with", + "type": "text" + }, + { + "bbox": [ + 153, + 102, + 202, + 114 + ], + "score": 0.9, + "content": "d _ { \\mathrm { m o d e l } } = 5 1 2", + "type": "inline_equation" + }, + { + "bbox": [ + 202, + 103, + 219, + 115 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 220, + 103, + 267, + 114 + ], + "score": 0.88, + "content": "d _ { \\mathrm { f f n } } = 3 0 7 2", + "type": "inline_equation" + }, + { + "bbox": [ + 267, + 103, + 491, + 115 + ], + "score": 1.0, + "content": ". No positional encodings are used for Mixer or gMLPs.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "table_body", + "bbox": [ + 136, + 120, + 473, + 232 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 136, + 120, + 473, + 232 + ], + "spans": [ + { + "bbox": [ + 136, + 120, + 473, + 232 + ], + "score": 0.979, + "html": "
ModelPerplexity*Params (M)
BERTbase4.37110
BERTbase + rel pos4.26110
BERTbase + rel pos - attn5.6496
MLP-Mixer5.34112
Linear gMLP,s(Z)= f(Z)5.1492
Additive gMLP,s(Z)= Z+ f(Z)4.9792
Multiplicative gMLP,s(Z) = Z f(Z)4.5392
Multiplicative,Split gMLP,s(Z) = Z1 f(Z2),Z= Zi||Z24.35102
", + "type": "table", + "image_path": "230eb41e660952154dd9abe33bc6dbc3d8c04a14f867dc0c2c621db29dfc52f1.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 136, + 120, + 473, + 157.33333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 136, + 157.33333333333334, + 473, + 194.66666666666669 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 136, + 194.66666666666669, + 473, + 232.00000000000003 + ], + "spans": [], + "index": 6 + } + ] + }, + { + "type": "table_footnote", + "bbox": [ + 143, + 234, + 396, + 245 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 147, + 233, + 396, + 246 + ], + "spans": [ + { + "bbox": [ + 147, + 233, + 396, + 246 + ], + "score": 1.0, + "content": "Standard deviation across multiple independent runs is around 0.01.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 109, + 248, + 504, + 271 + ], + "lines": [], + "index": 8.5, + "bbox_fs": [ + 106, + 247, + 505, + 272 + ], + "lines_deleted": true + }, + { + "type": "image", + "bbox": [ + 162, + 280, + 450, + 351 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 162, + 280, + 450, + 351 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 162, + 280, + 450, + 351 + ], + "spans": [ + { + "bbox": [ + 162, + 280, + 450, + 351 + ], + "score": 0.962, + "type": "image", + "image_path": "731fe5bbb1001c5b9f64529e6e9ed073521439ba52c88469a01d1124d0dfaa34.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 162, + 280, + 450, + 303.6666666666667 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 162, + 303.6666666666667, + 450, + 327.33333333333337 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 162, + 327.33333333333337, + 450, + 351.00000000000006 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 360, + 506, + 416 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 106, + 361, + 505, + 372 + ], + "spans": [ + { + "bbox": [ + 106, + 361, + 288, + 372 + ], + "score": 1.0, + "content": "Figure 4: Visualization of the spatial filters in", + "type": "text" + }, + { + "bbox": [ + 288, + 361, + 315, + 372 + ], + "score": 0.5, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 361, + 505, + 372 + ], + "score": 1.0, + "content": "learned on the MLM task. For each layer in the", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 105, + 371, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 105, + 371, + 209, + 383 + ], + "score": 1.0, + "content": "model we plot the row in", + "type": "text" + }, + { + "bbox": [ + 209, + 371, + 221, + 381 + ], + "score": 0.5, + "content": "W", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 371, + 466, + 383 + ], + "score": 1.0, + "content": "associated with the token in the middle of the sequence. The", + "type": "text" + }, + { + "bbox": [ + 467, + 373, + 473, + 381 + ], + "score": 0.41, + "content": "\\mathbf { X }", + "type": "inline_equation" + }, + { + "bbox": [ + 474, + 371, + 506, + 383 + ], + "score": 1.0, + "content": "-axis of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 106, + 383, + 505, + 394 + ], + "spans": [ + { + "bbox": [ + 106, + 383, + 505, + 394 + ], + "score": 1.0, + "content": "each subplot has a length of 128 which equal the number of tokens in the sequence. The learned filters", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 392, + 506, + 407 + ], + "spans": [ + { + "bbox": [ + 105, + 392, + 506, + 407 + ], + "score": 1.0, + "content": "appear to be smooth and have several types: forward-looking (e.g., 1st in 2nd row), backward-looking", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 403, + 387, + 416 + ], + "spans": [ + { + "bbox": [ + 105, + 403, + 387, + 416 + ], + "score": 1.0, + "content": "(e.g., 5th in 2nd row) and bi-directional (e.g., 2nd last in the last row).", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 15 + } + ], + "index": 13.0 + }, + { + "type": "title", + "bbox": [ + 107, + 426, + 386, + 439 + ], + "lines": [ + { + "bbox": [ + 106, + 426, + 387, + 441 + ], + "spans": [ + { + "bbox": [ + 106, + 426, + 254, + 441 + ], + "score": 1.0, + "content": "4.2 Case Study: The Behavior of", + "type": "text" + }, + { + "bbox": [ + 254, + 427, + 283, + 439 + ], + "score": 0.29, + "content": "\\mathbf { g } \\mathbf { M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 284, + 426, + 387, + 441 + ], + "score": 1.0, + "content": "as Model Size Increases", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 106, + 447, + 505, + 513 + ], + "lines": [ + { + "bbox": [ + 105, + 447, + 505, + 460 + ], + "spans": [ + { + "bbox": [ + 105, + 447, + 505, + 460 + ], + "score": 1.0, + "content": "In Table 4, we investigate the scaling properties of Transformers and gMLPs in BERT as their model", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 458, + 505, + 471 + ], + "spans": [ + { + "bbox": [ + 105, + 458, + 427, + 471 + ], + "score": 1.0, + "content": "capacity grows. Specifically, we scale the depth of these models by a factor of", + "type": "text" + }, + { + "bbox": [ + 427, + 458, + 487, + 470 + ], + "score": 0.92, + "content": "\\{ 0 . 5 , 1 , 2 , 4 \\} \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 487, + 458, + 505, + 471 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 469, + 505, + 482 + ], + "spans": [ + { + "bbox": [ + 105, + 469, + 505, + 482 + ], + "score": 1.0, + "content": "report the their pretraining MLM perplexities on the validation set as well as finetuning results on", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 479, + 506, + 493 + ], + "spans": [ + { + "bbox": [ + 105, + 479, + 506, + 493 + ], + "score": 1.0, + "content": "the dev sets of two tasks in GLUE [41]. Note each individual Transformer layer is effectively two", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 492, + 506, + 503 + ], + "spans": [ + { + "bbox": [ + 106, + 492, + 506, + 503 + ], + "score": 1.0, + "content": "consecutive blocks: one for self-attention and one for FFN. In the table below we use the notation of", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 106, + 501, + 497, + 514 + ], + "spans": [ + { + "bbox": [ + 106, + 502, + 138, + 513 + ], + "score": 0.87, + "content": "1 2 + 1 2", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 501, + 497, + 514 + ], + "score": 1.0, + "content": "to refer to 12 of self-attention blocks plus 12 of FFN blocks in the Transformer baselines.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21.5, + "bbox_fs": [ + 105, + 447, + 506, + 514 + ] + }, + { + "type": "table", + "bbox": [ + 171, + 550, + 437, + 667 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 107, + 521, + 502, + 544 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 521, + 504, + 534 + ], + "spans": [ + { + "bbox": [ + 105, + 521, + 504, + 534 + ], + "score": 1.0, + "content": "Table 4: Pretraining and dev-set finetuning results over increased model capacity. We use the relative", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 532, + 433, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 532, + 433, + 545 + ], + "score": 1.0, + "content": "positional encoding scheme for Transformers which performs the best in Table 3.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "table_body", + "bbox": [ + 171, + 550, + 437, + 667 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 171, + 550, + 437, + 667 + ], + "spans": [ + { + "bbox": [ + 171, + 550, + 437, + 667 + ], + "score": 0.98, + "html": "
Model#LParams (M) PerplexitySST-2 MNLI-m
TransformergMLP6+61867594.915.2590.4 81.591.2 77.7
TransformergMLP12+12361101024.264.3591.3 83.392.3 80.9
TransformergMLP24+24721951873.833.7992.1 85.293.5 82.8
TransformergMLP48+481443653573.473.4392.8 86.395.1 84.6
", + "type": "table", + "image_path": "f9b9937d357c9507e5ed75ff7c559cbe2b327ae203d9f1866ee2864e0fbaa0c9.jpg" + } + ] + } + ], + "index": 28, + "virtual_lines": [ + { + "bbox": [ + 171, + 550, + 437, + 589.0 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 171, + 589.0, + 437, + 628.0 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 171, + 628.0, + 437, + 667.0 + ], + "spans": [], + "index": 29 + } + ] + } + ], + "index": 26.75 + }, + { + "type": "text", + "bbox": [ + 108, + 672, + 505, + 695 + ], + "lines": [ + { + "bbox": [ + 106, + 671, + 505, + 686 + ], + "spans": [ + { + "bbox": [ + 106, + 671, + 505, + 686 + ], + "score": 1.0, + "content": "The results above show that a deep enough gMLP is able to match and even outperform the perplexity", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 682, + 506, + 696 + ], + "spans": [ + { + "bbox": [ + 105, + 682, + 506, + 696 + ], + "score": 1.0, + "content": "of Transformers with comparable capacity.4 In addition, the perplexity-parameter relationships for", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 72, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 505, + 86 + ], + "score": 1.0, + "content": "both architecture families approximately follow a power law (left of Figure 5). This implies the", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 84, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 84, + 505, + 96 + ], + "score": 1.0, + "content": "empirical scaling laws originally observed for Transformer-based language models [42] might be", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 94, + 311, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 311, + 107 + ], + "score": 1.0, + "content": "broadly applicable across different model families.", + "type": "text", + "cross_page": true + } + ], + "index": 2 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 671, + 506, + 696 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 72, + 505, + 106 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 505, + 86 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 505, + 86 + ], + "score": 1.0, + "content": "both architecture families approximately follow a power law (left of Figure 5). This implies the", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 84, + 505, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 84, + 505, + 96 + ], + "score": 1.0, + "content": "empirical scaling laws originally observed for Transformer-based language models [42] might be", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 94, + 311, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 311, + 107 + ], + "score": 1.0, + "content": "broadly applicable across different model families.", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "image", + "bbox": [ + 125, + 114, + 485, + 227 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 125, + 114, + 485, + 227 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 125, + 114, + 485, + 227 + ], + "spans": [ + { + "bbox": [ + 125, + 114, + 485, + 227 + ], + "score": 0.972, + "type": "image", + "image_path": "03ebb35b54fcbc8c3b3a364ab001ed17abb00c96397df6daed10ed232e24238b.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 125, + 114, + 485, + 151.66666666666666 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 125, + 151.66666666666666, + 485, + 189.33333333333331 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 125, + 189.33333333333331, + 485, + 226.99999999999997 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 235, + 506, + 269 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 234, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 505, + 249 + ], + "score": 1.0, + "content": "Figure 5: Scaling properties with respect to perplexity and finetuning accuracies. The figures show", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 245, + 507, + 260 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 507, + 260 + ], + "score": 1.0, + "content": "that for pretraining, gMLPs are equally good at optimizing perplexity as Transformers. For finetuning,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 257, + 437, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 437, + 270 + ], + "score": 1.0, + "content": "the two model families exhibit comparable scalability despite task-specific offsets.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 106, + 279, + 505, + 367 + ], + "lines": [ + { + "bbox": [ + 105, + 278, + 505, + 292 + ], + "spans": [ + { + "bbox": [ + 105, + 278, + 505, + 292 + ], + "score": 1.0, + "content": "Table 4 also leads to an interesting observation that the pretraining perplexities across different model", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 290, + 506, + 303 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 506, + 303 + ], + "score": 1.0, + "content": "families are not equal in terms of finetuning. While gMLPs outperform Transformers on SST-2, they", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 301, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 301, + 506, + 313 + ], + "score": 1.0, + "content": "are worse on MNLI. The results imply that the finetuning performance for NLP tasks is a function of", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 312, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 106, + 312, + 505, + 325 + ], + "score": 1.0, + "content": "not only the perplexity but also the inductive bias in the architecture. Figure 5 shows that despite", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 323, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 323, + 506, + 336 + ], + "score": 1.0, + "content": "the architecture-specific discrepancies between pretraining and finetuning, gMLPs and Transformers", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 334, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 334, + 505, + 346 + ], + "score": 1.0, + "content": "exhibit comparable scalability (slope) on both finetuning tasks. This means one can always offset the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 345, + 506, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 506, + 357 + ], + "score": 1.0, + "content": "gap by enlarging the model capacity. In other words, the results indicate that model scalability with", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 357, + 452, + 368 + ], + "spans": [ + { + "bbox": [ + 105, + 357, + 452, + 368 + ], + "score": 1.0, + "content": "respect to downstream metrics can be independent from the presence of self-attention.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 12.5 + }, + { + "type": "title", + "bbox": [ + 107, + 379, + 411, + 392 + ], + "lines": [ + { + "bbox": [ + 105, + 378, + 411, + 394 + ], + "spans": [ + { + "bbox": [ + 105, + 378, + 411, + 394 + ], + "score": 1.0, + "content": "4.3 Ablation: The Usefulness of Tiny Attention in BERT’s Finetuning", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 107, + 399, + 505, + 466 + ], + "lines": [ + { + "bbox": [ + 105, + 398, + 505, + 413 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 505, + 413 + ], + "score": 1.0, + "content": "So far we have found that self-attention is not a required component to achieve strong MLM perplexity", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 410, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 505, + 423 + ], + "score": 1.0, + "content": "or scalability. At the meantime, we also identified NLP finetuning tasks where gMLPs transfer less", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 422, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 106, + 422, + 505, + 433 + ], + "score": 1.0, + "content": "well than Transformers (Table 4). The fact that our MLP-like model is advantageous on SST-2 but", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 433, + 506, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 506, + 444 + ], + "score": 1.0, + "content": "worse on MNLI is particularly informative—the former is a single-sentence task whereas the latter", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 443, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 505, + 456 + ], + "score": 1.0, + "content": "involves sentence pairs (premise and hypothesis) [43]. We suspect the role of self-attention during", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 454, + 304, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 304, + 467 + ], + "score": 1.0, + "content": "finetuning is related to cross-sentence alignment.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20.5 + }, + { + "type": "text", + "bbox": [ + 106, + 470, + 505, + 548 + ], + "lines": [ + { + "bbox": [ + 105, + 470, + 505, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 505, + 483 + ], + "score": 1.0, + "content": "To isolate the effect of self-attention, we experiment with a hybrid model where a tiny self-attention", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 481, + 505, + 494 + ], + "spans": [ + { + "bbox": [ + 105, + 481, + 505, + 494 + ], + "score": 1.0, + "content": "block is attached to the gating function of gMLP (Figure 6). Since gMLP itself is already capable in", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 493, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 493, + 505, + 505 + ], + "score": 1.0, + "content": "capturing spatial relationships, we hypothesize that this extra self-attention module does not have to", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 504, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 106, + 504, + 505, + 515 + ], + "score": 1.0, + "content": "be heavy, and that its presence is more relevant than its capacity. A typical tiny attention module in", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 514, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 505, + 527 + ], + "score": 1.0, + "content": "our experiments has only a single head with size 64, significantly smaller than a typical multi-head", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 526, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 106, + 526, + 505, + 537 + ], + "score": 1.0, + "content": "self-attention in Transformers with 12 heads and a total size of 768. In the following, we refer to the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 536, + 441, + 549 + ], + "spans": [ + { + "bbox": [ + 106, + 536, + 196, + 549 + ], + "score": 1.0, + "content": "hybrid model, namely", + "type": "text" + }, + { + "bbox": [ + 196, + 536, + 224, + 548 + ], + "score": 0.36, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 536, + 441, + 549 + ], + "score": 1.0, + "content": "with a tiny self-attention, as aMLP (“a” for attention).", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27 + }, + { + "type": "title", + "bbox": [ + 301, + 565, + 470, + 576 + ], + "lines": [ + { + "bbox": [ + 300, + 564, + 471, + 577 + ], + "spans": [ + { + "bbox": [ + 300, + 564, + 471, + 577 + ], + "score": 1.0, + "content": "Pseudo-code for the tiny attention module", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 286, + 580, + 458, + 643 + ], + "lines": [ + { + "bbox": [ + 285, + 580, + 426, + 588 + ], + "spans": [ + { + "bbox": [ + 285, + 580, + 426, + 588 + ], + "score": 1.0, + "content": "def tiny_attn(x, d_out, d_attn=64):", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 292, + 589, + 457, + 597 + ], + "spans": [ + { + "bbox": [ + 292, + 589, + 308, + 597 + ], + "score": 1.0, + "content": "qkv", + "type": "text" + }, + { + "bbox": [ + 308, + 590, + 316, + 596 + ], + "score": 0.68, + "content": "=", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 589, + 457, + 597 + ], + "score": 1.0, + "content": "proj(x, 3 ∗ d_attn, axis=\"channel\")", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 291, + 597, + 449, + 606 + ], + "spans": [ + { + "bbox": [ + 291, + 597, + 323, + 606 + ], + "score": 1.0, + "content": "q, k, v", + "type": "text" + }, + { + "bbox": [ + 324, + 599, + 332, + 605 + ], + "score": 0.35, + "content": "=", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 597, + 449, + 606 + ], + "score": 1.0, + "content": "split(qkv, 3, axis=\"channel\")", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 293, + 606, + 425, + 615 + ], + "spans": [ + { + "bbox": [ + 293, + 607, + 307, + 614 + ], + "score": 0.35, + "content": "\\kappa =", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 606, + 425, + 615 + ], + "score": 1.0, + "content": "einsum(\"bnd,bmd−>bnm\", q, k)", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 291, + 614, + 413, + 623 + ], + "spans": [ + { + "bbox": [ + 291, + 614, + 348, + 623 + ], + "score": 1.0, + "content": "a = softmax(w", + "type": "text" + }, + { + "bbox": [ + 348, + 615, + 355, + 622 + ], + "score": 0.3, + "content": "^ *", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 614, + 413, + 623 + ], + "score": 1.0, + "content": "rsqrt(d_attn))", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 291, + 623, + 424, + 631 + ], + "spans": [ + { + "bbox": [ + 291, + 623, + 424, + 631 + ], + "score": 1.0, + "content": "x = einsum(\"bnm,bmd−>bnd\", a, v)", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 291, + 631, + 441, + 640 + ], + "spans": [ + { + "bbox": [ + 291, + 631, + 441, + 640 + ], + "score": 1.0, + "content": "return proj(x, d_out, axis=\"channel\")", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 37 + }, + { + "type": "image", + "bbox": [ + 126, + 559, + 237, + 648 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 126, + 559, + 237, + 648 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 126, + 559, + 237, + 648 + ], + "spans": [ + { + "bbox": [ + 126, + 559, + 237, + 648 + ], + "score": 0.872, + "type": "image", + "image_path": "e228c75c14097e90bf3596da75b5d6c0ccbec039247e6155396cb13236ea093c.jpg" + } + ] + } + ], + "index": 32.0, + "virtual_lines": [ + { + "bbox": [ + 126, + 559, + 237, + 603.5 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 126, + 603.5, + 237, + 648.0 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 655, + 503, + 690 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 655, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 505, + 669 + ], + "score": 1.0, + "content": "Figure 6: Hybrid spatial gating unit with a tiny self-attention module. We use the normalized input of", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 667, + 505, + 679 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 121, + 679 + ], + "score": 1.0, + "content": "the", + "type": "text" + }, + { + "bbox": [ + 121, + 667, + 149, + 678 + ], + "score": 0.51, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 667, + 505, + 679 + ], + "score": 1.0, + "content": "block (endpoint after the input normalization and right before the channel expansion) as", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 677, + 473, + 691 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 318, + 691 + ], + "score": 1.0, + "content": "the input to the tiny self-attention. For SGU we have", + "type": "text" + }, + { + "bbox": [ + 319, + 678, + 375, + 690 + ], + "score": 0.93, + "content": "{ d _ { \\mathrm { o u t } } = d _ { \\mathrm { f f n } } / 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 677, + 473, + 691 + ], + "score": 1.0, + "content": "due to the channel split.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42 + } + ], + "index": 37.0 + }, + { + "type": "text", + "bbox": [ + 106, + 699, + 503, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "In Figure 7, we investigate the transferability of MLM models via the calibration plots between", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 710, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 432, + 723 + ], + "score": 1.0, + "content": "their pretraining perplexities and finetuning metrics. Models evaluated include", + "type": "text" + }, + { + "bbox": [ + 433, + 711, + 472, + 722 + ], + "score": 0.59, + "content": "\\mathbf { B E R T _ { b a s e } }", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 710, + 505, + 723 + ], + "score": 1.0, + "content": ", gMLP", + "type": "text" + } + ], + "index": 45 + } + ], + "index": 44.5 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 741, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 72, + 505, + 106 + ], + "lines": [], + "index": 1, + "bbox_fs": [ + 105, + 72, + 505, + 107 + ], + "lines_deleted": true + }, + { + "type": "image", + "bbox": [ + 125, + 114, + 485, + 227 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 125, + 114, + 485, + 227 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 125, + 114, + 485, + 227 + ], + "spans": [ + { + "bbox": [ + 125, + 114, + 485, + 227 + ], + "score": 0.972, + "type": "image", + "image_path": "03ebb35b54fcbc8c3b3a364ab001ed17abb00c96397df6daed10ed232e24238b.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 125, + 114, + 485, + 151.66666666666666 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 125, + 151.66666666666666, + 485, + 189.33333333333331 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 125, + 189.33333333333331, + 485, + 226.99999999999997 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 235, + 506, + 269 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 234, + 505, + 249 + ], + "spans": [ + { + "bbox": [ + 105, + 234, + 505, + 249 + ], + "score": 1.0, + "content": "Figure 5: Scaling properties with respect to perplexity and finetuning accuracies. The figures show", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 105, + 245, + 507, + 260 + ], + "spans": [ + { + "bbox": [ + 105, + 245, + 507, + 260 + ], + "score": 1.0, + "content": "that for pretraining, gMLPs are equally good at optimizing perplexity as Transformers. For finetuning,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 105, + 257, + 437, + 270 + ], + "spans": [ + { + "bbox": [ + 105, + 257, + 437, + 270 + ], + "score": 1.0, + "content": "the two model families exhibit comparable scalability despite task-specific offsets.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7 + } + ], + "index": 5.5 + }, + { + "type": "text", + "bbox": [ + 106, + 279, + 505, + 367 + ], + "lines": [ + { + "bbox": [ + 105, + 278, + 505, + 292 + ], + "spans": [ + { + "bbox": [ + 105, + 278, + 505, + 292 + ], + "score": 1.0, + "content": "Table 4 also leads to an interesting observation that the pretraining perplexities across different model", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 290, + 506, + 303 + ], + "spans": [ + { + "bbox": [ + 105, + 290, + 506, + 303 + ], + "score": 1.0, + "content": "families are not equal in terms of finetuning. While gMLPs outperform Transformers on SST-2, they", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 301, + 506, + 313 + ], + "spans": [ + { + "bbox": [ + 106, + 301, + 506, + 313 + ], + "score": 1.0, + "content": "are worse on MNLI. The results imply that the finetuning performance for NLP tasks is a function of", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 312, + 505, + 325 + ], + "spans": [ + { + "bbox": [ + 106, + 312, + 505, + 325 + ], + "score": 1.0, + "content": "not only the perplexity but also the inductive bias in the architecture. Figure 5 shows that despite", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 105, + 323, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 105, + 323, + 506, + 336 + ], + "score": 1.0, + "content": "the architecture-specific discrepancies between pretraining and finetuning, gMLPs and Transformers", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 334, + 505, + 346 + ], + "spans": [ + { + "bbox": [ + 106, + 334, + 505, + 346 + ], + "score": 1.0, + "content": "exhibit comparable scalability (slope) on both finetuning tasks. This means one can always offset the", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 105, + 345, + 506, + 357 + ], + "spans": [ + { + "bbox": [ + 105, + 345, + 506, + 357 + ], + "score": 1.0, + "content": "gap by enlarging the model capacity. In other words, the results indicate that model scalability with", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 105, + 357, + 452, + 368 + ], + "spans": [ + { + "bbox": [ + 105, + 357, + 452, + 368 + ], + "score": 1.0, + "content": "respect to downstream metrics can be independent from the presence of self-attention.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 278, + 506, + 368 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 379, + 411, + 392 + ], + "lines": [ + { + "bbox": [ + 105, + 378, + 411, + 394 + ], + "spans": [ + { + "bbox": [ + 105, + 378, + 411, + 394 + ], + "score": 1.0, + "content": "4.3 Ablation: The Usefulness of Tiny Attention in BERT’s Finetuning", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 107, + 399, + 505, + 466 + ], + "lines": [ + { + "bbox": [ + 105, + 398, + 505, + 413 + ], + "spans": [ + { + "bbox": [ + 105, + 398, + 505, + 413 + ], + "score": 1.0, + "content": "So far we have found that self-attention is not a required component to achieve strong MLM perplexity", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 410, + 505, + 423 + ], + "spans": [ + { + "bbox": [ + 105, + 410, + 505, + 423 + ], + "score": 1.0, + "content": "or scalability. At the meantime, we also identified NLP finetuning tasks where gMLPs transfer less", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 106, + 422, + 505, + 433 + ], + "spans": [ + { + "bbox": [ + 106, + 422, + 505, + 433 + ], + "score": 1.0, + "content": "well than Transformers (Table 4). The fact that our MLP-like model is advantageous on SST-2 but", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 433, + 506, + 444 + ], + "spans": [ + { + "bbox": [ + 105, + 433, + 506, + 444 + ], + "score": 1.0, + "content": "worse on MNLI is particularly informative—the former is a single-sentence task whereas the latter", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 443, + 505, + 456 + ], + "spans": [ + { + "bbox": [ + 105, + 443, + 505, + 456 + ], + "score": 1.0, + "content": "involves sentence pairs (premise and hypothesis) [43]. We suspect the role of self-attention during", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 454, + 304, + 467 + ], + "spans": [ + { + "bbox": [ + 105, + 454, + 304, + 467 + ], + "score": 1.0, + "content": "finetuning is related to cross-sentence alignment.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20.5, + "bbox_fs": [ + 105, + 398, + 506, + 467 + ] + }, + { + "type": "text", + "bbox": [ + 106, + 470, + 505, + 548 + ], + "lines": [ + { + "bbox": [ + 105, + 470, + 505, + 483 + ], + "spans": [ + { + "bbox": [ + 105, + 470, + 505, + 483 + ], + "score": 1.0, + "content": "To isolate the effect of self-attention, we experiment with a hybrid model where a tiny self-attention", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 105, + 481, + 505, + 494 + ], + "spans": [ + { + "bbox": [ + 105, + 481, + 505, + 494 + ], + "score": 1.0, + "content": "block is attached to the gating function of gMLP (Figure 6). Since gMLP itself is already capable in", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 493, + 505, + 505 + ], + "spans": [ + { + "bbox": [ + 106, + 493, + 505, + 505 + ], + "score": 1.0, + "content": "capturing spatial relationships, we hypothesize that this extra self-attention module does not have to", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 504, + 505, + 515 + ], + "spans": [ + { + "bbox": [ + 106, + 504, + 505, + 515 + ], + "score": 1.0, + "content": "be heavy, and that its presence is more relevant than its capacity. A typical tiny attention module in", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 514, + 505, + 527 + ], + "spans": [ + { + "bbox": [ + 105, + 514, + 505, + 527 + ], + "score": 1.0, + "content": "our experiments has only a single head with size 64, significantly smaller than a typical multi-head", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 526, + 505, + 537 + ], + "spans": [ + { + "bbox": [ + 106, + 526, + 505, + 537 + ], + "score": 1.0, + "content": "self-attention in Transformers with 12 heads and a total size of 768. In the following, we refer to the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 536, + 441, + 549 + ], + "spans": [ + { + "bbox": [ + 106, + 536, + 196, + 549 + ], + "score": 1.0, + "content": "hybrid model, namely", + "type": "text" + }, + { + "bbox": [ + 196, + 536, + 224, + 548 + ], + "score": 0.36, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 536, + 441, + 549 + ], + "score": 1.0, + "content": "with a tiny self-attention, as aMLP (“a” for attention).", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 27, + "bbox_fs": [ + 105, + 470, + 505, + 549 + ] + }, + { + "type": "title", + "bbox": [ + 301, + 565, + 470, + 576 + ], + "lines": [ + { + "bbox": [ + 300, + 564, + 471, + 577 + ], + "spans": [ + { + "bbox": [ + 300, + 564, + 471, + 577 + ], + "score": 1.0, + "content": "Pseudo-code for the tiny attention module", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 286, + 580, + 458, + 643 + ], + "lines": [ + { + "bbox": [ + 285, + 580, + 426, + 588 + ], + "spans": [ + { + "bbox": [ + 285, + 580, + 426, + 588 + ], + "score": 1.0, + "content": "def tiny_attn(x, d_out, d_attn=64):", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 292, + 589, + 457, + 597 + ], + "spans": [ + { + "bbox": [ + 292, + 589, + 308, + 597 + ], + "score": 1.0, + "content": "qkv", + "type": "text" + }, + { + "bbox": [ + 308, + 590, + 316, + 596 + ], + "score": 0.68, + "content": "=", + "type": "inline_equation" + }, + { + "bbox": [ + 316, + 589, + 457, + 597 + ], + "score": 1.0, + "content": "proj(x, 3 ∗ d_attn, axis=\"channel\")", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 291, + 597, + 449, + 606 + ], + "spans": [ + { + "bbox": [ + 291, + 597, + 323, + 606 + ], + "score": 1.0, + "content": "q, k, v", + "type": "text" + }, + { + "bbox": [ + 324, + 599, + 332, + 605 + ], + "score": 0.35, + "content": "=", + "type": "inline_equation" + }, + { + "bbox": [ + 332, + 597, + 449, + 606 + ], + "score": 1.0, + "content": "split(qkv, 3, axis=\"channel\")", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 293, + 606, + 425, + 615 + ], + "spans": [ + { + "bbox": [ + 293, + 607, + 307, + 614 + ], + "score": 0.35, + "content": "\\kappa =", + "type": "inline_equation" + }, + { + "bbox": [ + 308, + 606, + 425, + 615 + ], + "score": 1.0, + "content": "einsum(\"bnd,bmd−>bnm\", q, k)", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 291, + 614, + 413, + 623 + ], + "spans": [ + { + "bbox": [ + 291, + 614, + 348, + 623 + ], + "score": 1.0, + "content": "a = softmax(w", + "type": "text" + }, + { + "bbox": [ + 348, + 615, + 355, + 622 + ], + "score": 0.3, + "content": "^ *", + "type": "inline_equation" + }, + { + "bbox": [ + 355, + 614, + 413, + 623 + ], + "score": 1.0, + "content": "rsqrt(d_attn))", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 291, + 623, + 424, + 631 + ], + "spans": [ + { + "bbox": [ + 291, + 623, + 424, + 631 + ], + "score": 1.0, + "content": "x = einsum(\"bnm,bmd−>bnd\", a, v)", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 291, + 631, + 441, + 640 + ], + "spans": [ + { + "bbox": [ + 291, + 631, + 441, + 640 + ], + "score": 1.0, + "content": "return proj(x, d_out, axis=\"channel\")", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 37, + "bbox_fs": [ + 285, + 580, + 457, + 640 + ] + }, + { + "type": "image", + "bbox": [ + 126, + 559, + 237, + 648 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 126, + 559, + 237, + 648 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 126, + 559, + 237, + 648 + ], + "spans": [ + { + "bbox": [ + 126, + 559, + 237, + 648 + ], + "score": 0.872, + "type": "image", + "image_path": "e228c75c14097e90bf3596da75b5d6c0ccbec039247e6155396cb13236ea093c.jpg" + } + ] + } + ], + "index": 32.0, + "virtual_lines": [ + { + "bbox": [ + 126, + 559, + 237, + 603.5 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 126, + 603.5, + 237, + 648.0 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 655, + 503, + 690 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 655, + 505, + 669 + ], + "spans": [ + { + "bbox": [ + 106, + 655, + 505, + 669 + ], + "score": 1.0, + "content": "Figure 6: Hybrid spatial gating unit with a tiny self-attention module. We use the normalized input of", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 106, + 667, + 505, + 679 + ], + "spans": [ + { + "bbox": [ + 106, + 667, + 121, + 679 + ], + "score": 1.0, + "content": "the", + "type": "text" + }, + { + "bbox": [ + 121, + 667, + 149, + 678 + ], + "score": 0.51, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 667, + 505, + 679 + ], + "score": 1.0, + "content": "block (endpoint after the input normalization and right before the channel expansion) as", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 106, + 677, + 473, + 691 + ], + "spans": [ + { + "bbox": [ + 106, + 677, + 318, + 691 + ], + "score": 1.0, + "content": "the input to the tiny self-attention. For SGU we have", + "type": "text" + }, + { + "bbox": [ + 319, + 678, + 375, + 690 + ], + "score": 0.93, + "content": "{ d _ { \\mathrm { o u t } } = d _ { \\mathrm { f f n } } / 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 677, + 473, + 691 + ], + "score": 1.0, + "content": "due to the channel split.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 42 + } + ], + "index": 37.0 + }, + { + "type": "text", + "bbox": [ + 106, + 699, + 503, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 699, + 505, + 712 + ], + "spans": [ + { + "bbox": [ + 106, + 699, + 505, + 712 + ], + "score": 1.0, + "content": "In Figure 7, we investigate the transferability of MLM models via the calibration plots between", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 106, + 710, + 505, + 723 + ], + "spans": [ + { + "bbox": [ + 106, + 710, + 432, + 723 + ], + "score": 1.0, + "content": "their pretraining perplexities and finetuning metrics. Models evaluated include", + "type": "text" + }, + { + "bbox": [ + 433, + 711, + 472, + 722 + ], + "score": 0.59, + "content": "\\mathbf { B E R T _ { b a s e } }", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 710, + 505, + 723 + ], + "score": 1.0, + "content": ", gMLP", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 105, + 72, + 506, + 85 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 506, + 85 + ], + "score": 1.0, + "content": "and its hybrid version aMLP with a 64-d single-head self-attention (Figure 6). The data points were", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 83, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 273, + 96 + ], + "score": 1.0, + "content": "collected by varying the model depth by", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 273, + 84, + 325, + 95 + ], + "score": 0.86, + "content": "\\{ 0 . 5 , 1 , 2 \\} \\times", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 325, + 83, + 370, + 96 + ], + "score": 1.0, + "content": "or data by", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 370, + 84, + 424, + 95 + ], + "score": 0.88, + "content": "\\{ 1 , 2 , 4 , 8 \\} \\times", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 424, + 83, + 506, + 96 + ], + "score": 1.0, + "content": ". It can be seen that", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 94, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 506, + 107 + ], + "score": 1.0, + "content": "gMLPs transfer better to SST-2 than Transformers regardless of the presence of self-attention, While", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 105, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 505, + 118 + ], + "score": 1.0, + "content": "gMLP performs worse on MNLI, attaching a tiny bit of self-attention is sufficient to close the gap. In", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 115, + 506, + 131 + ], + "spans": [ + { + "bbox": [ + 104, + 115, + 506, + 131 + ], + "score": 1.0, + "content": "Appendix D we visualize the tiny self-attention modules in aMLP over MNLI examples, showing", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 127, + 413, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 413, + 140 + ], + "score": 1.0, + "content": "that they are primarily responsible for the alignment between sentence pairs.", + "type": "text", + "cross_page": true + } + ], + "index": 5 + } + ], + "index": 44.5, + "bbox_fs": [ + 106, + 699, + 505, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 72, + 506, + 138 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 506, + 85 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 506, + 85 + ], + "score": 1.0, + "content": "and its hybrid version aMLP with a 64-d single-head self-attention (Figure 6). The data points were", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 105, + 83, + 506, + 96 + ], + "spans": [ + { + "bbox": [ + 105, + 83, + 273, + 96 + ], + "score": 1.0, + "content": "collected by varying the model depth by", + "type": "text" + }, + { + "bbox": [ + 273, + 84, + 325, + 95 + ], + "score": 0.86, + "content": "\\{ 0 . 5 , 1 , 2 \\} \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 83, + 370, + 96 + ], + "score": 1.0, + "content": "or data by", + "type": "text" + }, + { + "bbox": [ + 370, + 84, + 424, + 95 + ], + "score": 0.88, + "content": "\\{ 1 , 2 , 4 , 8 \\} \\times", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 83, + 506, + 96 + ], + "score": 1.0, + "content": ". It can be seen that", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 105, + 94, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 105, + 94, + 506, + 107 + ], + "score": 1.0, + "content": "gMLPs transfer better to SST-2 than Transformers regardless of the presence of self-attention, While", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 105, + 505, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 505, + 118 + ], + "score": 1.0, + "content": "gMLP performs worse on MNLI, attaching a tiny bit of self-attention is sufficient to close the gap. In", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 115, + 506, + 131 + ], + "spans": [ + { + "bbox": [ + 104, + 115, + 506, + 131 + ], + "score": 1.0, + "content": "Appendix D we visualize the tiny self-attention modules in aMLP over MNLI examples, showing", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 127, + 413, + 140 + ], + "spans": [ + { + "bbox": [ + 105, + 127, + 413, + 140 + ], + "score": 1.0, + "content": "that they are primarily responsible for the alignment between sentence pairs.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5 + }, + { + "type": "image", + "bbox": [ + 143, + 147, + 450, + 286 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 143, + 147, + 450, + 286 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 143, + 147, + 450, + 286 + ], + "spans": [ + { + "bbox": [ + 143, + 147, + 450, + 286 + ], + "score": 0.97, + "type": "image", + "image_path": "263412114ce248c9b9dccbc4dbdcbba36ac2aa648dce820d76c34069690a0cd1.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 143, + 147, + 450, + 193.33333333333334 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 143, + 193.33333333333334, + 450, + 239.66666666666669 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 143, + 239.66666666666669, + 450, + 286.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 291, + 505, + 325 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 291, + 506, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 506, + 305 + ], + "score": 1.0, + "content": "Figure 7: Transferability from MLM pretraining perpexity to finetuning accuracies on GLUE. aMLP", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 302, + 507, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 141, + 315 + ], + "score": 1.0, + "content": "refers to", + "type": "text" + }, + { + "bbox": [ + 141, + 303, + 168, + 314 + ], + "score": 0.47, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 302, + 507, + 315 + ], + "score": 1.0, + "content": "enhanced with a 64-d single-head self-attention, as illustrated in Figure 6. In contrast,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 314, + 470, + 325 + ], + "spans": [ + { + "bbox": [ + 106, + 314, + 470, + 325 + ], + "score": 1.0, + "content": "each self-attention module in the BERT baseline contains 12 heads with a total size of 768.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 106, + 333, + 505, + 356 + ], + "lines": [ + { + "bbox": [ + 105, + 333, + 506, + 346 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 469, + 346 + ], + "score": 1.0, + "content": "In Figure 8 we put together the scaling properties of the three models, showing that aMLP", + "type": "text" + }, + { + "bbox": [ + 470, + 334, + 506, + 345 + ], + "score": 0.72, + "content": "\\mathrm { ( g M L P + }", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 344, + 419, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 419, + 357 + ], + "score": 1.0, + "content": "tiny attention) consistently outperforms Transformer on both finetuning tasks.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "image", + "bbox": [ + 125, + 365, + 485, + 478 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 125, + 365, + 485, + 478 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 125, + 365, + 485, + 478 + ], + "spans": [ + { + "bbox": [ + 125, + 365, + 485, + 478 + ], + "score": 0.974, + "type": "image", + "image_path": "395d749d892cbad8add03840c7c07a237a2c72cef704c6aff4263e86c3de1d09.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 125, + 365, + 485, + 402.6666666666667 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 125, + 402.6666666666667, + 485, + 440.33333333333337 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 125, + 440.33333333333337, + 485, + 478.00000000000006 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 486, + 505, + 509 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 486, + 506, + 498 + ], + "spans": [ + { + "bbox": [ + 106, + 486, + 350, + 498 + ], + "score": 1.0, + "content": "Figure 8: Comparing the scaling properties of Transformers,", + "type": "text" + }, + { + "bbox": [ + 350, + 486, + 381, + 497 + ], + "score": 0.27, + "content": "\\mathrm { g M L P s }", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 486, + 506, + 498 + ], + "score": 1.0, + "content": "and aMLPs (with 64-d, single-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 496, + 407, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 407, + 509 + ], + "score": 1.0, + "content": "head attention). Results were obtained using the same setup in Section 4.2.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + } + ], + "index": 16.25 + }, + { + "type": "title", + "bbox": [ + 106, + 526, + 313, + 539 + ], + "lines": [ + { + "bbox": [ + 105, + 524, + 314, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 314, + 542 + ], + "score": 1.0, + "content": "4.4 Main Results for MLM in the BERT Setup", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 546, + 505, + 635 + ], + "lines": [ + { + "bbox": [ + 106, + 547, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 547, + 505, + 559 + ], + "score": 1.0, + "content": "Below we present pretraining and finetuning results in the full BERT setup. Different from ablation", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 558, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 505, + 570 + ], + "score": 1.0, + "content": "and case studies, here we use the full English C4 dataset and adopt a common MLM setup with", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 569, + 505, + 581 + ], + "spans": [ + { + "bbox": [ + 106, + 569, + 505, + 581 + ], + "score": 1.0, + "content": "batch size 256, max length 512 and 1M training steps. For fair comparison, we adjust the depth and", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 580, + 505, + 592 + ], + "spans": [ + { + "bbox": [ + 106, + 580, + 142, + 592 + ], + "score": 1.0, + "content": "width of", + "type": "text" + }, + { + "bbox": [ + 142, + 580, + 174, + 591 + ], + "score": 0.4, + "content": "\\mathrm { g M L P s }", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 580, + 505, + 592 + ], + "score": 1.0, + "content": "to ensure comparable model capacity with the Transformer baselines. The model", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 591, + 506, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 591, + 506, + 604 + ], + "score": 1.0, + "content": "specifications are given in Table 5 and hyperparameters are detailed in Appendix A.2. For finetuning,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 601, + 506, + 615 + ], + "spans": [ + { + "bbox": [ + 104, + 601, + 506, + 615 + ], + "score": 1.0, + "content": "we report the dev-set performance for SST-2 and MNLI in GLUE [41] and each result entry was", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 613, + 505, + 625 + ], + "spans": [ + { + "bbox": [ + 106, + 613, + 505, + 625 + ], + "score": 1.0, + "content": "obtained by taking the median of five independent runs. In addition, we report finetuning results on", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 623, + 420, + 636 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 420, + 636 + ], + "score": 1.0, + "content": "SQuAD [44, 45] to test the models’ ability in reasoning over a longer context.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 107, + 640, + 505, + 673 + ], + "lines": [ + { + "bbox": [ + 105, + 639, + 507, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 507, + 653 + ], + "score": 1.0, + "content": "Results are presented in Table 6. Consistent with our findings earlier in Section 4.1 and Section 4.2,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 649, + 507, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 507, + 664 + ], + "score": 1.0, + "content": "gMLPs are competitive with Transformers in terms of perplexity, especially in the larger scale setup.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 661, + 358, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 358, + 674 + ], + "score": 1.0, + "content": "There are several observations related to the finetuning results:", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 107, + 678, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "score": 1.0, + "content": "First, on finetuning tasks where gMLPs underperform Transformers, the performance gap tends to", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 689, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 689, + 358, + 702 + ], + "score": 1.0, + "content": "narrow as the model capacity increases. For example, while", + "type": "text" + }, + { + "bbox": [ + 358, + 689, + 386, + 700 + ], + "score": 0.36, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 689, + 468, + 702 + ], + "score": 1.0, + "content": "performs worse by", + "type": "text" + }, + { + "bbox": [ + 468, + 689, + 491, + 699 + ], + "score": 0.87, + "content": "8 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 491, + 689, + 506, + 702 + ], + "score": 1.0, + "content": "on", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 699, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 457, + 713 + ], + "score": 1.0, + "content": "SQuAD-v2.0 in the base scale, the performance gap relative to the baseline decreases to", + "type": "text" + }, + { + "bbox": [ + 457, + 700, + 479, + 711 + ], + "score": 0.87, + "content": "2 . 7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 699, + 505, + 713 + ], + "score": 1.0, + "content": "at the", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 710, + 506, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 208, + 724 + ], + "score": 1.0, + "content": "larger scale. Notably, our", + "type": "text" + }, + { + "bbox": [ + 208, + 711, + 248, + 723 + ], + "score": 0.81, + "content": "{ \\mathrm { g M L P } } _ { \\mathrm { l a r g e } }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 710, + 285, + 724 + ], + "score": 1.0, + "content": "achieves", + "type": "text" + }, + { + "bbox": [ + 285, + 711, + 312, + 721 + ], + "score": 0.84, + "content": "8 9 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 710, + 506, + 724 + ], + "score": 1.0, + "content": "F1 on SQuAD-v1.1 without any self-attention or", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32.5 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "spans": [ + { + "bbox": [ + 301, + 740, + 310, + 752 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 12, + "width": 9 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 106, + 72, + 506, + 138 + ], + "lines": [], + "index": 2.5, + "bbox_fs": [ + 104, + 72, + 506, + 140 + ], + "lines_deleted": true + }, + { + "type": "image", + "bbox": [ + 143, + 147, + 450, + 286 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 143, + 147, + 450, + 286 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 143, + 147, + 450, + 286 + ], + "spans": [ + { + "bbox": [ + 143, + 147, + 450, + 286 + ], + "score": 0.97, + "type": "image", + "image_path": "263412114ce248c9b9dccbc4dbdcbba36ac2aa648dce820d76c34069690a0cd1.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 143, + 147, + 450, + 193.33333333333334 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 143, + 193.33333333333334, + 450, + 239.66666666666669 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 143, + 239.66666666666669, + 450, + 286.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 107, + 291, + 505, + 325 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 105, + 291, + 506, + 305 + ], + "spans": [ + { + "bbox": [ + 105, + 291, + 506, + 305 + ], + "score": 1.0, + "content": "Figure 7: Transferability from MLM pretraining perpexity to finetuning accuracies on GLUE. aMLP", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 302, + 507, + 315 + ], + "spans": [ + { + "bbox": [ + 105, + 302, + 141, + 315 + ], + "score": 1.0, + "content": "refers to", + "type": "text" + }, + { + "bbox": [ + 141, + 303, + 168, + 314 + ], + "score": 0.47, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 168, + 302, + 507, + 315 + ], + "score": 1.0, + "content": "enhanced with a 64-d single-head self-attention, as illustrated in Figure 6. In contrast,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 106, + 314, + 470, + 325 + ], + "spans": [ + { + "bbox": [ + 106, + 314, + 470, + 325 + ], + "score": 1.0, + "content": "each self-attention module in the BERT baseline contains 12 heads with a total size of 768.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 106, + 333, + 505, + 356 + ], + "lines": [ + { + "bbox": [ + 105, + 333, + 506, + 346 + ], + "spans": [ + { + "bbox": [ + 105, + 333, + 469, + 346 + ], + "score": 1.0, + "content": "In Figure 8 we put together the scaling properties of the three models, showing that aMLP", + "type": "text" + }, + { + "bbox": [ + 470, + 334, + 506, + 345 + ], + "score": 0.72, + "content": "\\mathrm { ( g M L P + }", + "type": "inline_equation" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 344, + 419, + 357 + ], + "spans": [ + { + "bbox": [ + 106, + 344, + 419, + 357 + ], + "score": 1.0, + "content": "tiny attention) consistently outperforms Transformer on both finetuning tasks.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 105, + 333, + 506, + 357 + ] + }, + { + "type": "image", + "bbox": [ + 125, + 365, + 485, + 478 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 125, + 365, + 485, + 478 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 125, + 365, + 485, + 478 + ], + "spans": [ + { + "bbox": [ + 125, + 365, + 485, + 478 + ], + "score": 0.974, + "type": "image", + "image_path": "395d749d892cbad8add03840c7c07a237a2c72cef704c6aff4263e86c3de1d09.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 125, + 365, + 485, + 402.6666666666667 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 125, + 402.6666666666667, + 485, + 440.33333333333337 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 125, + 440.33333333333337, + 485, + 478.00000000000006 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 106, + 486, + 505, + 509 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 106, + 486, + 506, + 498 + ], + "spans": [ + { + "bbox": [ + 106, + 486, + 350, + 498 + ], + "score": 1.0, + "content": "Figure 8: Comparing the scaling properties of Transformers,", + "type": "text" + }, + { + "bbox": [ + 350, + 486, + 381, + 497 + ], + "score": 0.27, + "content": "\\mathrm { g M L P s }", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 486, + 506, + 498 + ], + "score": 1.0, + "content": "and aMLPs (with 64-d, single-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 496, + 407, + 509 + ], + "spans": [ + { + "bbox": [ + 105, + 496, + 407, + 509 + ], + "score": 1.0, + "content": "head attention). Results were obtained using the same setup in Section 4.2.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + } + ], + "index": 16.25 + }, + { + "type": "title", + "bbox": [ + 106, + 526, + 313, + 539 + ], + "lines": [ + { + "bbox": [ + 105, + 524, + 314, + 542 + ], + "spans": [ + { + "bbox": [ + 105, + 524, + 314, + 542 + ], + "score": 1.0, + "content": "4.4 Main Results for MLM in the BERT Setup", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 19 + }, + { + "type": "text", + "bbox": [ + 106, + 546, + 505, + 635 + ], + "lines": [ + { + "bbox": [ + 106, + 547, + 505, + 559 + ], + "spans": [ + { + "bbox": [ + 106, + 547, + 505, + 559 + ], + "score": 1.0, + "content": "Below we present pretraining and finetuning results in the full BERT setup. Different from ablation", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 106, + 558, + 505, + 570 + ], + "spans": [ + { + "bbox": [ + 106, + 558, + 505, + 570 + ], + "score": 1.0, + "content": "and case studies, here we use the full English C4 dataset and adopt a common MLM setup with", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 569, + 505, + 581 + ], + "spans": [ + { + "bbox": [ + 106, + 569, + 505, + 581 + ], + "score": 1.0, + "content": "batch size 256, max length 512 and 1M training steps. For fair comparison, we adjust the depth and", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 106, + 580, + 505, + 592 + ], + "spans": [ + { + "bbox": [ + 106, + 580, + 142, + 592 + ], + "score": 1.0, + "content": "width of", + "type": "text" + }, + { + "bbox": [ + 142, + 580, + 174, + 591 + ], + "score": 0.4, + "content": "\\mathrm { g M L P s }", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 580, + 505, + 592 + ], + "score": 1.0, + "content": "to ensure comparable model capacity with the Transformer baselines. The model", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 105, + 591, + 506, + 604 + ], + "spans": [ + { + "bbox": [ + 105, + 591, + 506, + 604 + ], + "score": 1.0, + "content": "specifications are given in Table 5 and hyperparameters are detailed in Appendix A.2. For finetuning,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 104, + 601, + 506, + 615 + ], + "spans": [ + { + "bbox": [ + 104, + 601, + 506, + 615 + ], + "score": 1.0, + "content": "we report the dev-set performance for SST-2 and MNLI in GLUE [41] and each result entry was", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 613, + 505, + 625 + ], + "spans": [ + { + "bbox": [ + 106, + 613, + 505, + 625 + ], + "score": 1.0, + "content": "obtained by taking the median of five independent runs. In addition, we report finetuning results on", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 623, + 420, + 636 + ], + "spans": [ + { + "bbox": [ + 106, + 623, + 420, + 636 + ], + "score": 1.0, + "content": "SQuAD [44, 45] to test the models’ ability in reasoning over a longer context.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 23.5, + "bbox_fs": [ + 104, + 547, + 506, + 636 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 640, + 505, + 673 + ], + "lines": [ + { + "bbox": [ + 105, + 639, + 507, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 639, + 507, + 653 + ], + "score": 1.0, + "content": "Results are presented in Table 6. Consistent with our findings earlier in Section 4.1 and Section 4.2,", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 105, + 649, + 507, + 664 + ], + "spans": [ + { + "bbox": [ + 105, + 649, + 507, + 664 + ], + "score": 1.0, + "content": "gMLPs are competitive with Transformers in terms of perplexity, especially in the larger scale setup.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 105, + 661, + 358, + 674 + ], + "spans": [ + { + "bbox": [ + 105, + 661, + 358, + 674 + ], + "score": 1.0, + "content": "There are several observations related to the finetuning results:", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29, + "bbox_fs": [ + 105, + 639, + 507, + 674 + ] + }, + { + "type": "text", + "bbox": [ + 107, + 678, + 505, + 722 + ], + "lines": [ + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "spans": [ + { + "bbox": [ + 105, + 677, + 506, + 691 + ], + "score": 1.0, + "content": "First, on finetuning tasks where gMLPs underperform Transformers, the performance gap tends to", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 689, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 105, + 689, + 358, + 702 + ], + "score": 1.0, + "content": "narrow as the model capacity increases. For example, while", + "type": "text" + }, + { + "bbox": [ + 358, + 689, + 386, + 700 + ], + "score": 0.36, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 387, + 689, + 468, + 702 + ], + "score": 1.0, + "content": "performs worse by", + "type": "text" + }, + { + "bbox": [ + 468, + 689, + 491, + 699 + ], + "score": 0.87, + "content": "8 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 491, + 689, + 506, + 702 + ], + "score": 1.0, + "content": "on", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 105, + 699, + 505, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 699, + 457, + 713 + ], + "score": 1.0, + "content": "SQuAD-v2.0 in the base scale, the performance gap relative to the baseline decreases to", + "type": "text" + }, + { + "bbox": [ + 457, + 700, + 479, + 711 + ], + "score": 0.87, + "content": "2 . 7 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 699, + 505, + 713 + ], + "score": 1.0, + "content": "at the", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 105, + 710, + 506, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 208, + 724 + ], + "score": 1.0, + "content": "larger scale. Notably, our", + "type": "text" + }, + { + "bbox": [ + 208, + 711, + 248, + 723 + ], + "score": 0.81, + "content": "{ \\mathrm { g M L P } } _ { \\mathrm { l a r g e } }", + "type": "inline_equation" + }, + { + "bbox": [ + 249, + 710, + 285, + 724 + ], + "score": 1.0, + "content": "achieves", + "type": "text" + }, + { + "bbox": [ + 285, + 711, + 312, + 721 + ], + "score": 0.84, + "content": "8 9 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 710, + 506, + 724 + ], + "score": 1.0, + "content": "F1 on SQuAD-v1.1 without any self-attention or", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 417, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 417, + 351, + 428 + ], + "score": 1.0, + "content": "dynamic spatial parameterization [28], which is well above the", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 352, + 417, + 378, + 427 + ], + "score": 0.87, + "content": "8 8 . 5 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 379, + 417, + 427, + 428 + ], + "score": 1.0, + "content": "reported for", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 427, + 417, + 465, + 428 + ], + "score": 0.86, + "content": "\\mathbf { B E R T _ { b a s e } }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 466, + 417, + 505, + 428 + ], + "score": 1.0, + "content": "in Devlin", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 427, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 427, + 187, + 441 + ], + "score": 1.0, + "content": "et al. [2] and is only", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 187, + 428, + 208, + 438 + ], + "score": 0.86, + "content": "1 . 4 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 209, + 427, + 338, + 441 + ], + "score": 1.0, + "content": "away from the original result for", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 338, + 428, + 378, + 439 + ], + "score": 0.74, + "content": "\\mathbf { B E R T _ { l a r g e } }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 378, + 427, + 505, + 441 + ], + "score": 1.0, + "content": ". We also include one additional", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 435, + 504, + 453 + ], + "spans": [ + { + "bbox": [ + 104, + 435, + 202, + 453 + ], + "score": 1.0, + "content": "data point by scaling up", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 203, + 439, + 229, + 450 + ], + "score": 0.37, + "content": "\\mathrm { g M L P }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 230, + 435, + 366, + 453 + ], + "score": 1.0, + "content": "even further. The resulting model,", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 366, + 439, + 410, + 450 + ], + "score": 0.86, + "content": "\\mathrm { g } \\mathrm { \\bar { M } L P _ { \\mathrm { x l a r g e } } }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 410, + 435, + 463, + 453 + ], + "score": 1.0, + "content": ", outperforms", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 464, + 438, + 504, + 450 + ], + "score": 0.36, + "content": "\\mathbf { B E R T _ { l a r g e } }", + "type": "inline_equation", + "cross_page": true + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 448, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 153, + 462 + ], + "score": 1.0, + "content": "on SQuAD-", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 154, + 450, + 172, + 460 + ], + "score": 0.35, + "content": "\\mathbf { \\nabla } \\cdot \\mathbf { v } 2 . 0 ", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 172, + 448, + 505, + 462 + ], + "score": 1.0, + "content": "—a difficult task involving question-answer pairs—without any self-attention. While", + "type": "text", + "cross_page": true + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 461, + 505, + 472 + ], + "spans": [ + { + "bbox": [ + 106, + 461, + 505, + 472 + ], + "score": 1.0, + "content": "this is not a fair comparison due to different model sizes, it is an existence proof that MLP-like", + "type": "text", + "cross_page": true + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 471, + 399, + 483 + ], + "spans": [ + { + "bbox": [ + 106, + 471, + 399, + 483 + ], + "score": 1.0, + "content": "models can be competitive with Transformers on challenging NLP tasks.", + "type": "text", + "cross_page": true + } + ], + "index": 14 + } + ], + "index": 32.5, + "bbox_fs": [ + 105, + 677, + 506, + 724 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 178, + 87, + 433, + 190 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 198, + 69, + 412, + 81 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 197, + 67, + 414, + 84 + ], + "spans": [ + { + "bbox": [ + 197, + 67, + 414, + 84 + ], + "score": 1.0, + "content": "Table 5: Model specifications in the full BERT setup.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 178, + 87, + 433, + 190 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 178, + 87, + 433, + 190 + ], + "spans": [ + { + "bbox": [ + 178, + 87, + 433, + 190 + ], + "score": 0.981, + "html": "
Params (M)FLOPs (B)#Ldmodeldffn
BERTbase gMLPbase aMLPbase110100.812+127683072
130158.0485123072
109128.9365123072
BERTlarge gMLPlarge aMLPlarge336341.224+2410244096
365430.1967683072
316370.3727683072
gMLPxlarge9411091.314410244096
", + "type": "table", + "image_path": "de04ff846f28de47d3407d02d990efd065f5be28122340787c79d35fd6edcd9d.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 178, + 87, + 433, + 121.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 178, + 121.33333333333334, + 433, + 155.66666666666669 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 178, + 155.66666666666669, + 433, + 190.00000000000003 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "table", + "bbox": [ + 113, + 229, + 496, + 394 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 104, + 200, + 505, + 223 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 199, + 505, + 213 + ], + "spans": [ + { + "bbox": [ + 105, + 199, + 505, + 213 + ], + "score": 1.0, + "content": "Table 6: Pretraining perplexities and dev-set results for finetuning. “ours” indicates models trained", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 210, + 491, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 491, + 225 + ], + "score": 1.0, + "content": "using our setup. We report accuracies for SST-2 and MNLI, and F1 scores for SQuAD v1.1/2.0.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "table_body", + "bbox": [ + 113, + 229, + 496, + 394 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 113, + 229, + 496, + 394 + ], + "spans": [ + { + "bbox": [ + 113, + 229, + 496, + 394 + ], + "score": 0.986, + "html": "
PerplexitySST-2MNLI (m/mm)SQuADAttn SizeParams (M)
v1.1v2.0
BERTbase [2]192.784.4/-88.576.3768 (64 × 12)110
BERTbase (ours)4.1793.885.6/85.790.278.6768 (64 × 12)110
gMLPbase4.2894.283.7/84.186.770.11130
aMLPbase3.9593.485.9/85.890.780.964109
BERTlarge [2]193.786.6/-90.981.81024 (64 × 16)336
BERTlarge (ours)3.3594.387.0/87.492.081.01024 (64 × 16)336
gMLPlarge3.3294.886.2/86.589.578.31365
aMLPlarge3.1994.888.4/88.492.285.4128316
gMLPxlarge2.8995.687.7/87.790.982.11941
", + "type": "table", + "image_path": "5ba707c6c8f0c199b6cc50563f0f4952642180841fe1165c9562b1c6903bcbe3.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 113, + 229, + 496, + 284.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 113, + 284.0, + 496, + 339.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 113, + 339.0, + 496, + 394.0 + ], + "spans": [], + "index": 8 + } + ] + } + ], + "index": 5.75 + }, + { + "type": "text", + "bbox": [ + 106, + 416, + 505, + 482 + ], + "lines": [ + { + "bbox": [ + 106, + 417, + 505, + 428 + ], + "spans": [ + { + "bbox": [ + 106, + 417, + 351, + 428 + ], + "score": 1.0, + "content": "dynamic spatial parameterization [28], which is well above the", + "type": "text" + }, + { + "bbox": [ + 352, + 417, + 378, + 427 + ], + "score": 0.87, + "content": "8 8 . 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 379, + 417, + 427, + 428 + ], + "score": 1.0, + "content": "reported for", + "type": "text" + }, + { + "bbox": [ + 427, + 417, + 465, + 428 + ], + "score": 0.86, + "content": "\\mathbf { B E R T _ { b a s e } }", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 417, + 505, + 428 + ], + "score": 1.0, + "content": "in Devlin", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 427, + 505, + 441 + ], + "spans": [ + { + "bbox": [ + 105, + 427, + 187, + 441 + ], + "score": 1.0, + "content": "et al. [2] and is only", + "type": "text" + }, + { + "bbox": [ + 187, + 428, + 208, + 438 + ], + "score": 0.86, + "content": "1 . 4 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 209, + 427, + 338, + 441 + ], + "score": 1.0, + "content": "away from the original result for", + "type": "text" + }, + { + "bbox": [ + 338, + 428, + 378, + 439 + ], + "score": 0.74, + "content": "\\mathbf { B E R T _ { l a r g e } }", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 427, + 505, + 441 + ], + "score": 1.0, + "content": ". We also include one additional", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 435, + 504, + 453 + ], + "spans": [ + { + "bbox": [ + 104, + 435, + 202, + 453 + ], + "score": 1.0, + "content": "data point by scaling up", + "type": "text" + }, + { + "bbox": [ + 203, + 439, + 229, + 450 + ], + "score": 0.37, + "content": "\\mathrm { g M L P }", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 435, + 366, + 453 + ], + "score": 1.0, + "content": "even further. The resulting model,", + "type": "text" + }, + { + "bbox": [ + 366, + 439, + 410, + 450 + ], + "score": 0.86, + "content": "\\mathrm { g } \\mathrm { \\bar { M } L P _ { \\mathrm { x l a r g e } } }", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 435, + 463, + 453 + ], + "score": 1.0, + "content": ", outperforms", + "type": "text" + }, + { + "bbox": [ + 464, + 438, + 504, + 450 + ], + "score": 0.36, + "content": "\\mathbf { B E R T _ { l a r g e } }", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 105, + 448, + 505, + 462 + ], + "spans": [ + { + "bbox": [ + 105, + 448, + 153, + 462 + ], + "score": 1.0, + "content": "on SQuAD-", + "type": "text" + }, + { + "bbox": [ + 154, + 450, + 172, + 460 + ], + "score": 0.35, + "content": "\\mathbf { \\nabla } \\cdot \\mathbf { v } 2 . 0 ", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 448, + 505, + 462 + ], + "score": 1.0, + "content": "—a difficult task involving question-answer pairs—without any self-attention. While", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 461, + 505, + 472 + ], + "spans": [ + { + "bbox": [ + 106, + 461, + 505, + 472 + ], + "score": 1.0, + "content": "this is not a fair comparison due to different model sizes, it is an existence proof that MLP-like", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 471, + 399, + 483 + ], + "spans": [ + { + "bbox": [ + 106, + 471, + 399, + 483 + ], + "score": 1.0, + "content": "models can be competitive with Transformers on challenging NLP tasks.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 107, + 487, + 505, + 586 + ], + "lines": [ + { + "bbox": [ + 106, + 488, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 106, + 488, + 505, + 500 + ], + "score": 1.0, + "content": "Furthermore, we show that blending in a tiny single-head self-attention of size either 64 or 128 is", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 499, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 499, + 505, + 511 + ], + "score": 1.0, + "content": "sufficient to make gMLPs outperform Transformers of similar capacity, sometimes by a significant", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 509, + 506, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 270, + 522 + ], + "score": 1.0, + "content": "margin. For example, our hybrid model", + "type": "text" + }, + { + "bbox": [ + 271, + 509, + 311, + 521 + ], + "score": 0.8, + "content": "\\mathrm { a M L P _ { l a r g e } }", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 509, + 349, + 522 + ], + "score": 1.0, + "content": "achieves", + "type": "text" + }, + { + "bbox": [ + 349, + 509, + 372, + 520 + ], + "score": 0.86, + "content": "4 . 4 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 509, + 506, + 522 + ], + "score": 1.0, + "content": "higher F1 than Transformers on", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 520, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 505, + 533 + ], + "score": 1.0, + "content": "SQuAD-v2.0. The results suggest that the capacity in the multi-head self-attention of Transformers", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 531, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 506, + 545 + ], + "score": 1.0, + "content": "can be largely redundant, and that the majority of its functionalities can be captured by the spatial", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 542, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 506, + 555 + ], + "score": 1.0, + "content": "gating unit in gMLPs. The results also imply that the inductive biases in the spatial gating unit of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 553, + 506, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 506, + 565 + ], + "score": 1.0, + "content": "gMLPs and the tiny attention are complementary to each other. While the benefits of architectural", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 564, + 506, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 506, + 577 + ], + "score": 1.0, + "content": "inductive bias may vanish over increased compute, tiny attention does improve the practical value of", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 574, + 322, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 322, + 587 + ], + "score": 1.0, + "content": "gMLPs in the regime that we investigate in this work.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 19 + }, + { + "type": "title", + "bbox": [ + 107, + 603, + 183, + 617 + ], + "lines": [ + { + "bbox": [ + 104, + 600, + 185, + 619 + ], + "spans": [ + { + "bbox": [ + 104, + 600, + 185, + 619 + ], + "score": 1.0, + "content": "5 Conclusion", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 629, + 505, + 673 + ], + "lines": [ + { + "bbox": [ + 106, + 629, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 629, + 505, + 641 + ], + "score": 1.0, + "content": "Since the seminal work of Vaswani et al. [1], Transformers have been widely adopted across NLP", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 640, + 507, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 507, + 653 + ], + "score": 1.0, + "content": "and computer vision. This adoption has enabled many impressive results especially in NLP. To date,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 650, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 505, + 663 + ], + "score": 1.0, + "content": "it is still unclear what empowers such success: is it the feedforward nature of Transformers or is it the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 662, + 304, + 674 + ], + "spans": [ + { + "bbox": [ + 106, + 662, + 304, + 674 + ], + "score": 1.0, + "content": "multi-head self-attention layers in Transformers?", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 108, + 678, + 504, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "score": 1.0, + "content": "Our work suggests a simpler alternative to the multi-head self-attention layers in Transformers. We", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 689, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 106, + 689, + 506, + 702 + ], + "score": 1.0, + "content": "show that gMLPs, a simple variant of MLPs with gating, can be competitive with Transformers", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 700, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 506, + 713 + ], + "score": 1.0, + "content": "in terms of BERT’s pretraining perplexity and ViT’s accuracy. gMLPs are also comparable with", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 710, + 506, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 506, + 724 + ], + "score": 1.0, + "content": "Transformers in terms of the scalability over increased data and compute. As for BERT finetuning, we", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 30.5 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 302, + 742, + 308, + 750 + ], + "lines": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "spans": [ + { + "bbox": [ + 302, + 741, + 309, + 752 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 178, + 87, + 433, + 190 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 198, + 69, + 412, + 81 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 197, + 67, + 414, + 84 + ], + "spans": [ + { + "bbox": [ + 197, + 67, + 414, + 84 + ], + "score": 1.0, + "content": "Table 5: Model specifications in the full BERT setup.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 178, + 87, + 433, + 190 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 178, + 87, + 433, + 190 + ], + "spans": [ + { + "bbox": [ + 178, + 87, + 433, + 190 + ], + "score": 0.981, + "html": "
Params (M)FLOPs (B)#Ldmodeldffn
BERTbase gMLPbase aMLPbase110100.812+127683072
130158.0485123072
109128.9365123072
BERTlarge gMLPlarge aMLPlarge336341.224+2410244096
365430.1967683072
316370.3727683072
gMLPxlarge9411091.314410244096
", + "type": "table", + "image_path": "de04ff846f28de47d3407d02d990efd065f5be28122340787c79d35fd6edcd9d.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 178, + 87, + 433, + 121.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 178, + 121.33333333333334, + 433, + 155.66666666666669 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 178, + 155.66666666666669, + 433, + 190.00000000000003 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "table", + "bbox": [ + 113, + 229, + 496, + 394 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 104, + 200, + 505, + 223 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 105, + 199, + 505, + 213 + ], + "spans": [ + { + "bbox": [ + 105, + 199, + 505, + 213 + ], + "score": 1.0, + "content": "Table 6: Pretraining perplexities and dev-set results for finetuning. “ours” indicates models trained", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 105, + 210, + 491, + 225 + ], + "spans": [ + { + "bbox": [ + 105, + 210, + 491, + 225 + ], + "score": 1.0, + "content": "using our setup. We report accuracies for SST-2 and MNLI, and F1 scores for SQuAD v1.1/2.0.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "table_body", + "bbox": [ + 113, + 229, + 496, + 394 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 113, + 229, + 496, + 394 + ], + "spans": [ + { + "bbox": [ + 113, + 229, + 496, + 394 + ], + "score": 0.986, + "html": "
PerplexitySST-2MNLI (m/mm)SQuADAttn SizeParams (M)
v1.1v2.0
BERTbase [2]192.784.4/-88.576.3768 (64 × 12)110
BERTbase (ours)4.1793.885.6/85.790.278.6768 (64 × 12)110
gMLPbase4.2894.283.7/84.186.770.11130
aMLPbase3.9593.485.9/85.890.780.964109
BERTlarge [2]193.786.6/-90.981.81024 (64 × 16)336
BERTlarge (ours)3.3594.387.0/87.492.081.01024 (64 × 16)336
gMLPlarge3.3294.886.2/86.589.578.31365
aMLPlarge3.1994.888.4/88.492.285.4128316
gMLPxlarge2.8995.687.7/87.790.982.11941
", + "type": "table", + "image_path": "5ba707c6c8f0c199b6cc50563f0f4952642180841fe1165c9562b1c6903bcbe3.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 113, + 229, + 496, + 284.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 113, + 284.0, + 496, + 339.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 113, + 339.0, + 496, + 394.0 + ], + "spans": [], + "index": 8 + } + ] + } + ], + "index": 5.75 + }, + { + "type": "text", + "bbox": [ + 106, + 416, + 505, + 482 + ], + "lines": [], + "index": 11.5, + "bbox_fs": [ + 104, + 417, + 505, + 483 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 107, + 487, + 505, + 586 + ], + "lines": [ + { + "bbox": [ + 106, + 488, + 505, + 500 + ], + "spans": [ + { + "bbox": [ + 106, + 488, + 505, + 500 + ], + "score": 1.0, + "content": "Furthermore, we show that blending in a tiny single-head self-attention of size either 64 or 128 is", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 499, + 505, + 511 + ], + "spans": [ + { + "bbox": [ + 106, + 499, + 505, + 511 + ], + "score": 1.0, + "content": "sufficient to make gMLPs outperform Transformers of similar capacity, sometimes by a significant", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 509, + 506, + 522 + ], + "spans": [ + { + "bbox": [ + 105, + 509, + 270, + 522 + ], + "score": 1.0, + "content": "margin. For example, our hybrid model", + "type": "text" + }, + { + "bbox": [ + 271, + 509, + 311, + 521 + ], + "score": 0.8, + "content": "\\mathrm { a M L P _ { l a r g e } }", + "type": "inline_equation" + }, + { + "bbox": [ + 312, + 509, + 349, + 522 + ], + "score": 1.0, + "content": "achieves", + "type": "text" + }, + { + "bbox": [ + 349, + 509, + 372, + 520 + ], + "score": 0.86, + "content": "4 . 4 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 372, + 509, + 506, + 522 + ], + "score": 1.0, + "content": "higher F1 than Transformers on", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 105, + 520, + 505, + 533 + ], + "spans": [ + { + "bbox": [ + 105, + 520, + 505, + 533 + ], + "score": 1.0, + "content": "SQuAD-v2.0. The results suggest that the capacity in the multi-head self-attention of Transformers", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 105, + 531, + 506, + 545 + ], + "spans": [ + { + "bbox": [ + 105, + 531, + 506, + 545 + ], + "score": 1.0, + "content": "can be largely redundant, and that the majority of its functionalities can be captured by the spatial", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 542, + 506, + 555 + ], + "spans": [ + { + "bbox": [ + 105, + 542, + 506, + 555 + ], + "score": 1.0, + "content": "gating unit in gMLPs. The results also imply that the inductive biases in the spatial gating unit of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 105, + 553, + 506, + 565 + ], + "spans": [ + { + "bbox": [ + 105, + 553, + 506, + 565 + ], + "score": 1.0, + "content": "gMLPs and the tiny attention are complementary to each other. While the benefits of architectural", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 564, + 506, + 577 + ], + "spans": [ + { + "bbox": [ + 105, + 564, + 506, + 577 + ], + "score": 1.0, + "content": "inductive bias may vanish over increased compute, tiny attention does improve the practical value of", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 105, + 574, + 322, + 587 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 322, + 587 + ], + "score": 1.0, + "content": "gMLPs in the regime that we investigate in this work.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 19, + "bbox_fs": [ + 105, + 488, + 506, + 587 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 603, + 183, + 617 + ], + "lines": [ + { + "bbox": [ + 104, + 600, + 185, + 619 + ], + "spans": [ + { + "bbox": [ + 104, + 600, + 185, + 619 + ], + "score": 1.0, + "content": "5 Conclusion", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 107, + 629, + 505, + 673 + ], + "lines": [ + { + "bbox": [ + 106, + 629, + 505, + 641 + ], + "spans": [ + { + "bbox": [ + 106, + 629, + 505, + 641 + ], + "score": 1.0, + "content": "Since the seminal work of Vaswani et al. [1], Transformers have been widely adopted across NLP", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 105, + 640, + 507, + 653 + ], + "spans": [ + { + "bbox": [ + 105, + 640, + 507, + 653 + ], + "score": 1.0, + "content": "and computer vision. This adoption has enabled many impressive results especially in NLP. To date,", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 105, + 650, + 505, + 663 + ], + "spans": [ + { + "bbox": [ + 105, + 650, + 505, + 663 + ], + "score": 1.0, + "content": "it is still unclear what empowers such success: is it the feedforward nature of Transformers or is it the", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 106, + 662, + 304, + 674 + ], + "spans": [ + { + "bbox": [ + 106, + 662, + 304, + 674 + ], + "score": 1.0, + "content": "multi-head self-attention layers in Transformers?", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5, + "bbox_fs": [ + 105, + 629, + 507, + 674 + ] + }, + { + "type": "text", + "bbox": [ + 108, + 678, + 504, + 722 + ], + "lines": [ + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "spans": [ + { + "bbox": [ + 106, + 678, + 505, + 690 + ], + "score": 1.0, + "content": "Our work suggests a simpler alternative to the multi-head self-attention layers in Transformers. We", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 106, + 689, + 506, + 702 + ], + "spans": [ + { + "bbox": [ + 106, + 689, + 506, + 702 + ], + "score": 1.0, + "content": "show that gMLPs, a simple variant of MLPs with gating, can be competitive with Transformers", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 105, + 700, + 506, + 713 + ], + "spans": [ + { + "bbox": [ + 105, + 700, + 506, + 713 + ], + "score": 1.0, + "content": "in terms of BERT’s pretraining perplexity and ViT’s accuracy. gMLPs are also comparable with", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 710, + 506, + 724 + ], + "spans": [ + { + "bbox": [ + 105, + 710, + 506, + 724 + ], + "score": 1.0, + "content": "Transformers in terms of the scalability over increased data and compute. As for BERT finetuning, we", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 106, + 72, + 506, + 85 + ], + "spans": [ + { + "bbox": [ + 106, + 72, + 506, + 85 + ], + "score": 1.0, + "content": "find gMLPs can achieve appealing results on challenging tasks such as SQuAD without self-attention,", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 84, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 84, + 505, + 95 + ], + "score": 1.0, + "content": "and can significantly outperform Transformers in certain cases. We also find the inductive bias in", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 93, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 506, + 107 + ], + "score": 1.0, + "content": "Transformer’s multi-head self-attention useful on downstream tasks that require cross-sentence align-", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 105, + 507, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 507, + 118 + ], + "score": 1.0, + "content": "ment. However in those cases, making gMLP substantially larger closes the gap with Transformers.", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 116, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 116, + 505, + 128 + ], + "score": 1.0, + "content": "More practically, blending a small single-head self-attention into gMLP allows for an even better", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 128, + 330, + 139 + ], + "spans": [ + { + "bbox": [ + 106, + 128, + 330, + 139 + ], + "score": 1.0, + "content": "architecture without the need for increasing model size.", + "type": "text", + "cross_page": true + } + ], + "index": 5 + } + ], + "index": 30.5, + "bbox_fs": [ + 105, + 678, + 506, + 724 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 73, + 506, + 138 + ], + "lines": [ + { + "bbox": [ + 106, + 72, + 506, + 85 + ], + "spans": [ + { + "bbox": [ + 106, + 72, + 506, + 85 + ], + "score": 1.0, + "content": "find gMLPs can achieve appealing results on challenging tasks such as SQuAD without self-attention,", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 106, + 84, + 505, + 95 + ], + "spans": [ + { + "bbox": [ + 106, + 84, + 505, + 95 + ], + "score": 1.0, + "content": "and can significantly outperform Transformers in certain cases. We also find the inductive bias in", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 93, + 506, + 107 + ], + "spans": [ + { + "bbox": [ + 106, + 93, + 506, + 107 + ], + "score": 1.0, + "content": "Transformer’s multi-head self-attention useful on downstream tasks that require cross-sentence align-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 105, + 105, + 507, + 118 + ], + "spans": [ + { + "bbox": [ + 105, + 105, + 507, + 118 + ], + "score": 1.0, + "content": "ment. However in those cases, making gMLP substantially larger closes the gap with Transformers.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 106, + 116, + 505, + 128 + ], + "spans": [ + { + "bbox": [ + 106, + 116, + 505, + 128 + ], + "score": 1.0, + "content": "More practically, blending a small single-head self-attention into gMLP allows for an even better", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 106, + 128, + 330, + 139 + ], + "spans": [ + { + "bbox": [ + 106, + 128, + 330, + 139 + ], + "score": 1.0, + "content": "architecture without the need for increasing model size.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 108, + 153, + 207, + 167 + ], + "lines": [ + { + "bbox": [ + 106, + 152, + 208, + 170 + ], + "spans": [ + { + "bbox": [ + 106, + 152, + 208, + 170 + ], + "score": 1.0, + "content": "Acknowledgements", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 108, + 178, + 505, + 200 + ], + "lines": [ + { + "bbox": [ + 106, + 177, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 106, + 177, + 506, + 190 + ], + "score": 1.0, + "content": "We thank Gabriel Bender, Neil Houlsby, Thang Luong, Niki Parmar, Hieu Pham, Jascha Sohl-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 188, + 500, + 201 + ], + "spans": [ + { + "bbox": [ + 106, + 188, + 500, + 201 + ], + "score": 1.0, + "content": "Dickstein, Noam Shazeer, Ilya Sutskever, Jakob Uszkoreit and Ashish Vaswani for their feedback.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "title", + "bbox": [ + 107, + 216, + 163, + 228 + ], + "lines": [ + { + "bbox": [ + 106, + 214, + 165, + 230 + ], + "spans": [ + { + "bbox": [ + 106, + 214, + 165, + 230 + ], + "score": 1.0, + "content": "References", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 109, + 231, + 506, + 724 + ], + "lines": [ + { + "bbox": [ + 111, + 235, + 507, + 249 + ], + "spans": [ + { + "bbox": [ + 111, + 235, + 507, + 249 + ], + "score": 1.0, + "content": "[1] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez,", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 127, + 245, + 506, + 260 + ], + "spans": [ + { + "bbox": [ + 127, + 245, + 506, + 260 + ], + "score": 1.0, + "content": "Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in Neural Informa-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 127, + 257, + 254, + 270 + ], + "spans": [ + { + "bbox": [ + 127, + 257, + 254, + 270 + ], + "score": 1.0, + "content": "tion Processing Systems, 2017.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 110, + 271, + 507, + 286 + ], + "spans": [ + { + "bbox": [ + 110, + 271, + 507, + 286 + ], + "score": 1.0, + "content": "[2] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 127, + 282, + 441, + 296 + ], + "spans": [ + { + "bbox": [ + 127, + 282, + 441, + 296 + ], + "score": 1.0, + "content": "deep bidirectional transformers for language understanding. In NAACL, 2018.", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 110, + 297, + 506, + 311 + ], + "spans": [ + { + "bbox": [ + 110, + 297, + 506, + 311 + ], + "score": 1.0, + "content": "[3] Zhilin Yang, Zihang Dai, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, and Quoc V", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 127, + 307, + 506, + 321 + ], + "spans": [ + { + "bbox": [ + 127, + 307, + 506, + 321 + ], + "score": 1.0, + "content": "Le. Xlnet: Generalized autoregressive pretraining for language understanding. In Advances in", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 127, + 320, + 315, + 332 + ], + "spans": [ + { + "bbox": [ + 127, + 320, + 315, + 332 + ], + "score": 1.0, + "content": "Neural Information Processing Systems, 2019.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 110, + 333, + 506, + 347 + ], + "spans": [ + { + "bbox": [ + 110, + 333, + 506, + 347 + ], + "score": 1.0, + "content": "[4] Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 126, + 343, + 506, + 360 + ], + "spans": [ + { + "bbox": [ + 126, + 343, + 506, + 360 + ], + "score": 1.0, + "content": "Lewis, Luke Zettlemoyer, and Veselin Stoyanov. Roberta: A robustly optimized bert pretraining", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 127, + 356, + 333, + 369 + ], + "spans": [ + { + "bbox": [ + 127, + 356, + 333, + 369 + ], + "score": 1.0, + "content": "approach. arXiv preprint arXiv:1907.11692, 2019.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 112, + 369, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 112, + 369, + 506, + 383 + ], + "score": 1.0, + "content": "[5] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 128, + 381, + 506, + 394 + ], + "spans": [ + { + "bbox": [ + 128, + 381, + 506, + 394 + ], + "score": 1.0, + "content": "Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 127, + 392, + 280, + 405 + ], + "spans": [ + { + "bbox": [ + 127, + 392, + 280, + 405 + ], + "score": 1.0, + "content": "text-to-text transformer. JMLR, 2020.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 110, + 406, + 506, + 420 + ], + "spans": [ + { + "bbox": [ + 110, + 406, + 506, + 420 + ], + "score": 1.0, + "content": "[6] Tom B Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 128, + 417, + 505, + 430 + ], + "spans": [ + { + "bbox": [ + 128, + 417, + 505, + 430 + ], + "score": 1.0, + "content": "Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 126, + 428, + 452, + 441 + ], + "spans": [ + { + "bbox": [ + 126, + 428, + 452, + 441 + ], + "score": 1.0, + "content": "few-shot learners. In Advances in Neural Information Processing Systems, 2020.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 110, + 442, + 506, + 456 + ], + "spans": [ + { + "bbox": [ + 110, + 442, + 506, + 456 + ], + "score": 1.0, + "content": "[7] Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai,", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 127, + 452, + 507, + 466 + ], + "spans": [ + { + "bbox": [ + 127, + 452, + 507, + 466 + ], + "score": 1.0, + "content": "Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 127, + 464, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 127, + 464, + 505, + 478 + ], + "score": 1.0, + "content": "An image is worth 16x16 words: Transformers for image recognition at scale. In ICLR, 2021.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 112, + 479, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 112, + 479, + 505, + 491 + ], + "score": 1.0, + "content": "[8] Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 127, + 490, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 127, + 490, + 505, + 502 + ], + "score": 1.0, + "content": "Hervé Jégou. Training data-efficient image transformers & distillation through attention. arXiv", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 126, + 501, + 266, + 513 + ], + "spans": [ + { + "bbox": [ + 126, + 501, + 266, + 513 + ], + "score": 1.0, + "content": "preprint arXiv:2012.12877, 2020.", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 110, + 514, + 506, + 528 + ], + "spans": [ + { + "bbox": [ + 110, + 514, + 506, + 528 + ], + "score": 1.0, + "content": "[9] Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 127, + 525, + 464, + 539 + ], + "spans": [ + { + "bbox": [ + 127, + 525, + 464, + 539 + ], + "score": 1.0, + "content": "Sergey Zagoruyko. End-to-end object detection with transformers. In ECCV, 2020.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 539, + 506, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 506, + 554 + ], + "score": 1.0, + "content": "[10] Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 126, + 550, + 506, + 566 + ], + "spans": [ + { + "bbox": [ + 126, + 550, + 506, + 566 + ], + "score": 1.0, + "content": "Guo. Swin transformer: Hierarchical vision transformer using shifted windows. arXiv preprint", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 127, + 562, + 231, + 574 + ], + "spans": [ + { + "bbox": [ + 127, + 562, + 231, + 574 + ], + "score": 1.0, + "content": "arXiv:2103.14030, 2021.", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 105, + 574, + 507, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 507, + 591 + ], + "score": 1.0, + "content": "[11] Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory. Neural Computation,", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 126, + 585, + 154, + 601 + ], + "spans": [ + { + "bbox": [ + 126, + 585, + 154, + 601 + ], + "score": 1.0, + "content": "1997.", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 108, + 602, + 506, + 614 + ], + "spans": [ + { + "bbox": [ + 108, + 602, + 506, + 614 + ], + "score": 1.0, + "content": "[12] Y. LeCun, B. Boser, J. S. Denker, D. Henderson, R. E. Howard, W. Hubbard, and L. D. Jackel.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 127, + 613, + 491, + 626 + ], + "spans": [ + { + "bbox": [ + 127, + 613, + 491, + 626 + ], + "score": 1.0, + "content": "Backpropagation applied to handwritten zip code recognition. Neural Computation, 1989.", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 105, + 626, + 506, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 506, + 642 + ], + "score": 1.0, + "content": "[13] Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 127, + 639, + 493, + 652 + ], + "spans": [ + { + "bbox": [ + 127, + 639, + 493, + 652 + ], + "score": 1.0, + "content": "convolutional neural networks. Advances in Neural Information Processing Systems, 2012.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 652, + 506, + 666 + ], + "spans": [ + { + "bbox": [ + 106, + 652, + 506, + 666 + ], + "score": 1.0, + "content": "[14] Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 127, + 664, + 269, + 675 + ], + "spans": [ + { + "bbox": [ + 127, + 664, + 269, + 675 + ], + "score": 1.0, + "content": "image recognition. In ICLR, 2015.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 108, + 677, + 507, + 691 + ], + "spans": [ + { + "bbox": [ + 108, + 677, + 507, + 691 + ], + "score": 1.0, + "content": "[15] Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov,", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 126, + 688, + 507, + 702 + ], + "spans": [ + { + "bbox": [ + 126, + 688, + 507, + 702 + ], + "score": 1.0, + "content": "Dumitru Erhan, Vincent Vanhoucke, and Andrew Rabinovich. Going deeper with convolutions.", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 126, + 698, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 126, + 698, + 506, + 714 + ], + "score": 1.0, + "content": "In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1–9,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 127, + 711, + 154, + 723 + ], + "spans": [ + { + "bbox": [ + 127, + 711, + 154, + 723 + ], + "score": 1.0, + "content": "2015.", + "type": "text" + } + ], + "index": 49 + } + ], + "index": 29.5 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "10", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 107, + 73, + 506, + 138 + ], + "lines": [], + "index": 2.5, + "bbox_fs": [ + 105, + 72, + 507, + 139 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 108, + 153, + 207, + 167 + ], + "lines": [ + { + "bbox": [ + 106, + 152, + 208, + 170 + ], + "spans": [ + { + "bbox": [ + 106, + 152, + 208, + 170 + ], + "score": 1.0, + "content": "Acknowledgements", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 108, + 178, + 505, + 200 + ], + "lines": [ + { + "bbox": [ + 106, + 177, + 506, + 190 + ], + "spans": [ + { + "bbox": [ + 106, + 177, + 506, + 190 + ], + "score": 1.0, + "content": "We thank Gabriel Bender, Neil Houlsby, Thang Luong, Niki Parmar, Hieu Pham, Jascha Sohl-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 188, + 500, + 201 + ], + "spans": [ + { + "bbox": [ + 106, + 188, + 500, + 201 + ], + "score": 1.0, + "content": "Dickstein, Noam Shazeer, Ilya Sutskever, Jakob Uszkoreit and Ashish Vaswani for their feedback.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 106, + 177, + 506, + 201 + ] + }, + { + "type": "title", + "bbox": [ + 107, + 216, + 163, + 228 + ], + "lines": [ + { + "bbox": [ + 106, + 214, + 165, + 230 + ], + "spans": [ + { + "bbox": [ + 106, + 214, + 165, + 230 + ], + "score": 1.0, + "content": "References", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "list", + "bbox": [ + 109, + 231, + 506, + 724 + ], + "lines": [ + { + "bbox": [ + 111, + 235, + 507, + 249 + ], + "spans": [ + { + "bbox": [ + 111, + 235, + 507, + 249 + ], + "score": 1.0, + "content": "[1] Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez,", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 245, + 506, + 260 + ], + "spans": [ + { + "bbox": [ + 127, + 245, + 506, + 260 + ], + "score": 1.0, + "content": "Lukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in Neural Informa-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 127, + 257, + 254, + 270 + ], + "spans": [ + { + "bbox": [ + 127, + 257, + 254, + 270 + ], + "score": 1.0, + "content": "tion Processing Systems, 2017.", + "type": "text" + } + ], + "index": 12, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 271, + 507, + 286 + ], + "spans": [ + { + "bbox": [ + 110, + 271, + 507, + 286 + ], + "score": 1.0, + "content": "[2] Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of", + "type": "text" + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 282, + 441, + 296 + ], + "spans": [ + { + "bbox": [ + 127, + 282, + 441, + 296 + ], + "score": 1.0, + "content": "deep bidirectional transformers for language understanding. In NAACL, 2018.", + "type": "text" + } + ], + "index": 14, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 297, + 506, + 311 + ], + "spans": [ + { + "bbox": [ + 110, + 297, + 506, + 311 + ], + "score": 1.0, + "content": "[3] Zhilin Yang, Zihang Dai, Yiming Yang, Jaime Carbonell, Ruslan Salakhutdinov, and Quoc V", + "type": "text" + } + ], + "index": 15, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 307, + 506, + 321 + ], + "spans": [ + { + "bbox": [ + 127, + 307, + 506, + 321 + ], + "score": 1.0, + "content": "Le. Xlnet: Generalized autoregressive pretraining for language understanding. In Advances in", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 127, + 320, + 315, + 332 + ], + "spans": [ + { + "bbox": [ + 127, + 320, + 315, + 332 + ], + "score": 1.0, + "content": "Neural Information Processing Systems, 2019.", + "type": "text" + } + ], + "index": 17, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 333, + 506, + 347 + ], + "spans": [ + { + "bbox": [ + 110, + 333, + 506, + 347 + ], + "score": 1.0, + "content": "[4] Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike", + "type": "text" + } + ], + "index": 18, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 343, + 506, + 360 + ], + "spans": [ + { + "bbox": [ + 126, + 343, + 506, + 360 + ], + "score": 1.0, + "content": "Lewis, Luke Zettlemoyer, and Veselin Stoyanov. Roberta: A robustly optimized bert pretraining", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 127, + 356, + 333, + 369 + ], + "spans": [ + { + "bbox": [ + 127, + 356, + 333, + 369 + ], + "score": 1.0, + "content": "approach. arXiv preprint arXiv:1907.11692, 2019.", + "type": "text" + } + ], + "index": 20, + "is_list_end_line": true + }, + { + "bbox": [ + 112, + 369, + 506, + 383 + ], + "spans": [ + { + "bbox": [ + 112, + 369, + 506, + 383 + ], + "score": 1.0, + "content": "[5] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena,", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 128, + 381, + 506, + 394 + ], + "spans": [ + { + "bbox": [ + 128, + 381, + 506, + 394 + ], + "score": 1.0, + "content": "Yanqi Zhou, Wei Li, and Peter J Liu. Exploring the limits of transfer learning with a unified", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 127, + 392, + 280, + 405 + ], + "spans": [ + { + "bbox": [ + 127, + 392, + 280, + 405 + ], + "score": 1.0, + "content": "text-to-text transformer. JMLR, 2020.", + "type": "text" + } + ], + "index": 23, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 406, + 506, + 420 + ], + "spans": [ + { + "bbox": [ + 110, + 406, + 506, + 420 + ], + "score": 1.0, + "content": "[6] Tom B Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal,", + "type": "text" + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 417, + 505, + 430 + ], + "spans": [ + { + "bbox": [ + 128, + 417, + 505, + 430 + ], + "score": 1.0, + "content": "Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 126, + 428, + 452, + 441 + ], + "spans": [ + { + "bbox": [ + 126, + 428, + 452, + 441 + ], + "score": 1.0, + "content": "few-shot learners. In Advances in Neural Information Processing Systems, 2020.", + "type": "text" + } + ], + "index": 26, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 442, + 506, + 456 + ], + "spans": [ + { + "bbox": [ + 110, + 442, + 506, + 456 + ], + "score": 1.0, + "content": "[7] Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai,", + "type": "text" + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 452, + 507, + 466 + ], + "spans": [ + { + "bbox": [ + 127, + 452, + 507, + 466 + ], + "score": 1.0, + "content": "Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, et al.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 127, + 464, + 505, + 478 + ], + "spans": [ + { + "bbox": [ + 127, + 464, + 505, + 478 + ], + "score": 1.0, + "content": "An image is worth 16x16 words: Transformers for image recognition at scale. In ICLR, 2021.", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 112, + 479, + 505, + 491 + ], + "spans": [ + { + "bbox": [ + 112, + 479, + 505, + 491 + ], + "score": 1.0, + "content": "[8] Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, and", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 127, + 490, + 505, + 502 + ], + "spans": [ + { + "bbox": [ + 127, + 490, + 505, + 502 + ], + "score": 1.0, + "content": "Hervé Jégou. Training data-efficient image transformers & distillation through attention. arXiv", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 126, + 501, + 266, + 513 + ], + "spans": [ + { + "bbox": [ + 126, + 501, + 266, + 513 + ], + "score": 1.0, + "content": "preprint arXiv:2012.12877, 2020.", + "type": "text" + } + ], + "index": 32, + "is_list_end_line": true + }, + { + "bbox": [ + 110, + 514, + 506, + 528 + ], + "spans": [ + { + "bbox": [ + 110, + 514, + 506, + 528 + ], + "score": 1.0, + "content": "[9] Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, and", + "type": "text" + } + ], + "index": 33, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 525, + 464, + 539 + ], + "spans": [ + { + "bbox": [ + 127, + 525, + 464, + 539 + ], + "score": 1.0, + "content": "Sergey Zagoruyko. End-to-end object detection with transformers. In ECCV, 2020.", + "type": "text" + } + ], + "index": 34, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 539, + 506, + 554 + ], + "spans": [ + { + "bbox": [ + 106, + 539, + 506, + 554 + ], + "score": 1.0, + "content": "[10] Ze Liu, Yutong Lin, Yue Cao, Han Hu, Yixuan Wei, Zheng Zhang, Stephen Lin, and Baining", + "type": "text" + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 550, + 506, + 566 + ], + "spans": [ + { + "bbox": [ + 126, + 550, + 506, + 566 + ], + "score": 1.0, + "content": "Guo. Swin transformer: Hierarchical vision transformer using shifted windows. arXiv preprint", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 127, + 562, + 231, + 574 + ], + "spans": [ + { + "bbox": [ + 127, + 562, + 231, + 574 + ], + "score": 1.0, + "content": "arXiv:2103.14030, 2021.", + "type": "text" + } + ], + "index": 37, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 574, + 507, + 591 + ], + "spans": [ + { + "bbox": [ + 105, + 574, + 507, + 591 + ], + "score": 1.0, + "content": "[11] Sepp Hochreiter and Jürgen Schmidhuber. Long short-term memory. Neural Computation,", + "type": "text" + } + ], + "index": 38, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 585, + 154, + 601 + ], + "spans": [ + { + "bbox": [ + 126, + 585, + 154, + 601 + ], + "score": 1.0, + "content": "1997.", + "type": "text" + } + ], + "index": 39, + "is_list_end_line": true + }, + { + "bbox": [ + 108, + 602, + 506, + 614 + ], + "spans": [ + { + "bbox": [ + 108, + 602, + 506, + 614 + ], + "score": 1.0, + "content": "[12] Y. LeCun, B. Boser, J. S. Denker, D. Henderson, R. E. Howard, W. Hubbard, and L. D. Jackel.", + "type": "text" + } + ], + "index": 40, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 613, + 491, + 626 + ], + "spans": [ + { + "bbox": [ + 127, + 613, + 491, + 626 + ], + "score": 1.0, + "content": "Backpropagation applied to handwritten zip code recognition. Neural Computation, 1989.", + "type": "text" + } + ], + "index": 41, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 626, + 506, + 642 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 506, + 642 + ], + "score": 1.0, + "content": "[13] Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. Imagenet classification with deep", + "type": "text" + } + ], + "index": 42, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 639, + 493, + 652 + ], + "spans": [ + { + "bbox": [ + 127, + 639, + 493, + 652 + ], + "score": 1.0, + "content": "convolutional neural networks. Advances in Neural Information Processing Systems, 2012.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 106, + 652, + 506, + 666 + ], + "spans": [ + { + "bbox": [ + 106, + 652, + 506, + 666 + ], + "score": 1.0, + "content": "[14] Karen Simonyan and Andrew Zisserman. Very deep convolutional networks for large-scale", + "type": "text" + } + ], + "index": 44, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 664, + 269, + 675 + ], + "spans": [ + { + "bbox": [ + 127, + 664, + 269, + 675 + ], + "score": 1.0, + "content": "image recognition. In ICLR, 2015.", + "type": "text" + } + ], + "index": 45, + "is_list_end_line": true + }, + { + "bbox": [ + 108, + 677, + 507, + 691 + ], + "spans": [ + { + "bbox": [ + 108, + 677, + 507, + 691 + ], + "score": 1.0, + "content": "[15] Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov,", + "type": "text" + } + ], + "index": 46, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 688, + 507, + 702 + ], + "spans": [ + { + "bbox": [ + 126, + 688, + 507, + 702 + ], + "score": 1.0, + "content": "Dumitru Erhan, Vincent Vanhoucke, and Andrew Rabinovich. Going deeper with convolutions.", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 126, + 698, + 506, + 714 + ], + "spans": [ + { + "bbox": [ + 126, + 698, + 506, + 714 + ], + "score": 1.0, + "content": "In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 1–9,", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 127, + 711, + 154, + 723 + ], + "spans": [ + { + "bbox": [ + 127, + 711, + 154, + 723 + ], + "score": 1.0, + "content": "2015.", + "type": "text" + } + ], + "index": 49, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 71, + 505, + 88 + ], + "spans": [ + { + "bbox": [ + 104, + 71, + 505, + 88 + ], + "score": 1.0, + "content": "[16] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 83, + 245, + 98 + ], + "spans": [ + { + "bbox": [ + 126, + 83, + 245, + 98 + ], + "score": 1.0, + "content": "recognition. In CVPR, 2016.", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 99, + 505, + 111 + ], + "spans": [ + { + "bbox": [ + 106, + 99, + 505, + 111 + ], + "score": 1.0, + "content": "[17] Mingxing Tan and Quoc Le. Efficientnet: Rethinking model scaling for convolutional neural", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 110, + 235, + 122 + ], + "spans": [ + { + "bbox": [ + 127, + 110, + 235, + 122 + ], + "score": 1.0, + "content": "networks. In ICML, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 123, + 505, + 140 + ], + "spans": [ + { + "bbox": [ + 104, + 123, + 505, + 140 + ], + "score": 1.0, + "content": "[18] Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. Neural machine translation by jointly", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 136, + 316, + 150 + ], + "spans": [ + { + "bbox": [ + 126, + 136, + 316, + 150 + ], + "score": 1.0, + "content": "learning to align and translate. In ICLR, 2015.", + "type": "text", + "cross_page": true + } + ], + "index": 5, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 151, + 506, + 165 + ], + "spans": [ + { + "bbox": [ + 105, + 151, + 506, + 165 + ], + "score": 1.0, + "content": "[19] Kurt Hornik, Maxwell Stinchcombe, and Halbert White. Multilayer feedforward networks are", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 161, + 385, + 175 + ], + "spans": [ + { + "bbox": [ + 126, + 161, + 385, + 175 + ], + "score": 1.0, + "content": "universal approximators. Neural networks, 2(5):359–366, 1989.", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 177, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 106, + 177, + 506, + 189 + ], + "score": 1.0, + "content": "[20] Ilya Tolstikhin, Neil Houlsby, Alexander Kolesnikov, Lucas Beyer, Xiaohua Zhai, Thomas Un-", + "type": "text", + "cross_page": true + } + ], + "index": 8, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 187, + 507, + 201 + ], + "spans": [ + { + "bbox": [ + 127, + 187, + 507, + 201 + ], + "score": 1.0, + "content": "terthiner, Jessica Yung, Daniel Keysers, Jakob Uszkoreit, Mario Lucic, and Alexey Dosovitskiy.", + "type": "text", + "cross_page": true + } + ], + "index": 9 + }, + { + "bbox": [ + 127, + 198, + 480, + 213 + ], + "spans": [ + { + "bbox": [ + 127, + 198, + 480, + 213 + ], + "score": 1.0, + "content": "Mlp-mixer: An all-mlp architecture for vision. arXiv preprint arXiv:2105.01601, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 212, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 104, + 212, + 506, + 228 + ], + "score": 1.0, + "content": "[21] Luke Melas-Kyriazi. Do you even need attention? a stack of feed-forward layers does surpris-", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 225, + 388, + 239 + ], + "spans": [ + { + "bbox": [ + 127, + 225, + 388, + 239 + ], + "score": 1.0, + "content": "ingly well on imagenet. arXiv preprint arXiv:2105.02723, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 239, + 507, + 254 + ], + "spans": [ + { + "bbox": [ + 106, + 239, + 507, + 254 + ], + "score": 1.0, + "content": "[22] Hugo Touvron, Piotr Bojanowski, Mathilde Caron, Matthieu Cord, Alaaeldin El-Nouby,", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 250, + 506, + 265 + ], + "spans": [ + { + "bbox": [ + 126, + 250, + 506, + 265 + ], + "score": 1.0, + "content": "Edouard Grave, Armand Joulin, Gabriel Synnaeve, Jakob Verbeek, and Hervé Jégou. Resmlp:", + "type": "text", + "cross_page": true + } + ], + "index": 14 + }, + { + "bbox": [ + 126, + 260, + 506, + 276 + ], + "spans": [ + { + "bbox": [ + 126, + 260, + 506, + 276 + ], + "score": 1.0, + "content": "Feedforward networks for image classification with data-efficient training. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 15 + }, + { + "bbox": [ + 125, + 272, + 232, + 285 + ], + "spans": [ + { + "bbox": [ + 125, + 272, + 232, + 285 + ], + "score": 1.0, + "content": "arXiv:2105.03404, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 16, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 286, + 507, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 286, + 507, + 302 + ], + "score": 1.0, + "content": "[23] Xiaohan Ding, Xiangyu Zhang, Jungong Han, and Guiguang Ding. Repmlp: Re-", + "type": "text", + "cross_page": true + } + ], + "index": 17, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 298, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 126, + 298, + 505, + 312 + ], + "score": 1.0, + "content": "parameterizing convolutions into fully-connected layers for image recognition. arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 18 + }, + { + "bbox": [ + 126, + 308, + 231, + 322 + ], + "spans": [ + { + "bbox": [ + 126, + 308, + 231, + 322 + ], + "score": 1.0, + "content": "arXiv:2105.01883, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 324, + 506, + 339 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 506, + 339 + ], + "score": 1.0, + "content": "[24] Dan Hendrycks and Kevin Gimpel. Gaussian error linear units (gelus). arXiv preprint", + "type": "text", + "cross_page": true + } + ], + "index": 20, + "is_list_start_line": true + }, + { + "bbox": [ + 128, + 336, + 231, + 348 + ], + "spans": [ + { + "bbox": [ + 128, + 336, + 231, + 348 + ], + "score": 1.0, + "content": "arXiv:1606.08415, 2016.", + "type": "text", + "cross_page": true + } + ], + "index": 21, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 351, + 507, + 365 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 507, + 365 + ], + "score": 1.0, + "content": "[25] Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen.", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 361, + 417, + 375 + ], + "spans": [ + { + "bbox": [ + 127, + 361, + 417, + 375 + ], + "score": 1.0, + "content": "Mobilenetv2: Inverted residuals and linear bottlenecks. In CVPR, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 23, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 375, + 506, + 392 + ], + "spans": [ + { + "bbox": [ + 104, + 375, + 506, + 392 + ], + "score": 1.0, + "content": "[26] Yann N Dauphin, Angela Fan, Michael Auli, and David Grangier. Language modeling with", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 388, + 316, + 401 + ], + "spans": [ + { + "bbox": [ + 126, + 388, + 316, + 401 + ], + "score": 1.0, + "content": "gated convolutional networks. In ICML, 2017.", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 403, + 493, + 417 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 493, + 417 + ], + "score": 1.0, + "content": "[27] Noam Shazeer. Glu variants improve transformer. arXiv preprint arXiv:2002.05202, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 26, + "is_list_start_line": true + }, + { + "bbox": [ + 106, + 418, + 505, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 418, + 505, + 430 + ], + "score": 1.0, + "content": "[28] Felix Wu, Angela Fan, Alexei Baevski, Yann N Dauphin, and Michael Auli. Pay less attention", + "type": "text", + "cross_page": true + } + ], + "index": 27, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 428, + 369, + 442 + ], + "spans": [ + { + "bbox": [ + 127, + 428, + 369, + 442 + ], + "score": 1.0, + "content": "with lightweight and dynamic convolutions. In ICLR, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 28, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 443, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 505, + 457 + ], + "score": 1.0, + "content": "[29] Rupesh Kumar Srivastava, Klaus Greff, and Jürgen Schmidhuber. Highway networks. arXiv", + "type": "text", + "cross_page": true + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 124, + 454, + 267, + 468 + ], + "spans": [ + { + "bbox": [ + 124, + 454, + 267, + 468 + ], + "score": 1.0, + "content": "preprint arXiv:1505.00387, 2015.", + "type": "text", + "cross_page": true + } + ], + "index": 30, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 470, + 462, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 470, + 462, + 484 + ], + "score": 1.0, + "content": "[30] Jie Hu, Li Shen, and Gang Sun. Squeeze-and-excitation networks. In CVPR, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 31, + "is_list_start_line": true, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 484, + 507, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 507, + 499 + ], + "score": 1.0, + "content": "[31] Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-", + "type": "text", + "cross_page": true + } + ], + "index": 32, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 496, + 506, + 510 + ], + "spans": [ + { + "bbox": [ + 126, + 496, + 506, + 510 + ], + "score": 1.0, + "content": "scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern", + "type": "text", + "cross_page": true + } + ], + "index": 33 + }, + { + "bbox": [ + 125, + 506, + 292, + 520 + ], + "spans": [ + { + "bbox": [ + 125, + 506, + 292, + 520 + ], + "score": 1.0, + "content": "recognition, pages 248–255. Ieee, 2009.", + "type": "text", + "cross_page": true + } + ], + "index": 34, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 521, + 505, + 535 + ], + "spans": [ + { + "bbox": [ + 106, + 521, + 505, + 535 + ], + "score": 1.0, + "content": "[32] Gao Huang, Yu Sun, Zhuang Liu, Daniel Sedra, and Kilian Q Weinberger. Deep networks with", + "type": "text", + "cross_page": true + } + ], + "index": 35, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 532, + 264, + 546 + ], + "spans": [ + { + "bbox": [ + 126, + 532, + 264, + 546 + ], + "score": 1.0, + "content": "stochastic depth. In ECCV, 2016.", + "type": "text", + "cross_page": true + } + ], + "index": 36, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 545, + 507, + 564 + ], + "spans": [ + { + "bbox": [ + 104, + 545, + 507, + 564 + ], + "score": 1.0, + "content": "[33] Andrew Brock, Soham De, Samuel L Smith, and Karen Simonyan. High-performance large-", + "type": "text", + "cross_page": true + } + ], + "index": 37, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 559, + 481, + 572 + ], + "spans": [ + { + "bbox": [ + 126, + 559, + 481, + 572 + ], + "score": 1.0, + "content": "scale image recognition without normalization. arXiv preprint arXiv:2102.06171, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 38, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 573, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 573, + 506, + 587 + ], + "score": 1.0, + "content": "[34] Mingxing Tan and Quoc V Le. Efficientnetv2: Smaller models and faster training. In ICML,", + "type": "text", + "cross_page": true + } + ], + "index": 39, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 583, + 155, + 598 + ], + "spans": [ + { + "bbox": [ + 126, + 583, + 155, + 598 + ], + "score": 1.0, + "content": "2021.", + "type": "text", + "cross_page": true + } + ], + "index": 40, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 599, + 507, + 614 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 507, + 614 + ], + "score": 1.0, + "content": "[35] Irwan Bello. Lambdanetworks: Modeling long-range interactions without attention. In ICLR,", + "type": "text", + "cross_page": true + } + ], + "index": 41, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 610, + 156, + 625 + ], + "spans": [ + { + "bbox": [ + 126, + 610, + 156, + 625 + ], + "score": 1.0, + "content": "2021.", + "type": "text", + "cross_page": true + } + ], + "index": 42, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 626, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 506, + 640 + ], + "score": 1.0, + "content": "[36] Ashish Vaswani, Prajit Ramachandran, Aravind Srinivas, Niki Parmar, Blake Hechtman, and", + "type": "text", + "cross_page": true + } + ], + "index": 43, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 637, + 506, + 651 + ], + "spans": [ + { + "bbox": [ + 126, + 637, + 506, + 651 + ], + "score": 1.0, + "content": "Jonathon Shlens. Scaling local self-attention for parameter efficient visual backbones. In CVPR,", + "type": "text", + "cross_page": true + } + ], + "index": 44 + }, + { + "bbox": [ + 126, + 645, + 155, + 662 + ], + "spans": [ + { + "bbox": [ + 126, + 645, + 155, + 662 + ], + "score": 1.0, + "content": "2021.", + "type": "text", + "cross_page": true + } + ], + "index": 45, + "is_list_end_line": true + }, + { + "bbox": [ + 104, + 661, + 507, + 678 + ], + "spans": [ + { + "bbox": [ + 104, + 661, + 507, + 678 + ], + "score": 1.0, + "content": "[37] Aravind Srinivas, Tsung-Yi Lin, Niki Parmar, Jonathon Shlens, Pieter Abbeel, and Ashish", + "type": "text", + "cross_page": true + } + ], + "index": 46, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 673, + 507, + 687 + ], + "spans": [ + { + "bbox": [ + 127, + 673, + 507, + 687 + ], + "score": 1.0, + "content": "Vaswani. Bottleneck transformers for visual recognition. arXiv preprint arXiv:2101.11605,", + "type": "text", + "cross_page": true + } + ], + "index": 47 + }, + { + "bbox": [ + 126, + 684, + 156, + 699 + ], + "spans": [ + { + "bbox": [ + 126, + 684, + 156, + 699 + ], + "score": 1.0, + "content": "2021.", + "type": "text", + "cross_page": true + } + ], + "index": 48, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 698, + 507, + 715 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 507, + 715 + ], + "score": 1.0, + "content": "[38] Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, and Lei Zhang.", + "type": "text", + "cross_page": true + } + ], + "index": 49, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 711, + 506, + 725 + ], + "spans": [ + { + "bbox": [ + 127, + 711, + 506, + 725 + ], + "score": 1.0, + "content": "Cvt: Introducing convolutions to vision transformers. arXiv preprint arXiv:2103.15808, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 50 + }, + { + "bbox": [ + 105, + 72, + 506, + 87 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 506, + 87 + ], + "score": 1.0, + "content": "[39] Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, and Piotr Dollár. Design-", + "type": "text", + "cross_page": true + } + ], + "index": 0, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 84, + 304, + 96 + ], + "spans": [ + { + "bbox": [ + 127, + 84, + 304, + 96 + ], + "score": 1.0, + "content": "ing network design spaces. In CVPR, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 1, + "is_list_end_line": true + }, + { + "bbox": [ + 107, + 99, + 506, + 111 + ], + "spans": [ + { + "bbox": [ + 107, + 99, + 506, + 111 + ], + "score": 1.0, + "content": "[40] Yi Tay, Dara Bahri, Donald Metzler, Da-Cheng Juan, Zhe Zhao, and Che Zheng. Synthesizer:", + "type": "text", + "cross_page": true + } + ], + "index": 2, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 109, + 486, + 123 + ], + "spans": [ + { + "bbox": [ + 127, + 109, + 486, + 123 + ], + "score": 1.0, + "content": "Rethinking self-attention in transformer models. arXiv preprint arXiv:2005.00743, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 3, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 124, + 507, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 124, + 507, + 138 + ], + "score": 1.0, + "content": "[41] Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel R Bowman.", + "type": "text", + "cross_page": true + } + ], + "index": 4, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 135, + 505, + 148 + ], + "spans": [ + { + "bbox": [ + 127, + 135, + 505, + 148 + ], + "score": 1.0, + "content": "Glue: A multi-task benchmark and analysis platform for natural language understanding. In", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 128, + 147, + 179, + 157 + ], + "spans": [ + { + "bbox": [ + 128, + 147, + 179, + 157 + ], + "score": 1.0, + "content": "ICLR, 2019.", + "type": "text", + "cross_page": true + } + ], + "index": 6, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 161, + 506, + 174 + ], + "spans": [ + { + "bbox": [ + 106, + 161, + 506, + 174 + ], + "score": 1.0, + "content": "[42] Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B Brown, Benjamin Chess, Rewon Child,", + "type": "text", + "cross_page": true + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 171, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 127, + 171, + 505, + 186 + ], + "score": 1.0, + "content": "Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling laws for neural language", + "type": "text", + "cross_page": true + } + ], + "index": 8 + }, + { + "bbox": [ + 127, + 184, + 325, + 194 + ], + "spans": [ + { + "bbox": [ + 127, + 184, + 325, + 194 + ], + "score": 1.0, + "content": "models. arXiv preprint arXiv:2001.08361, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 9, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 197, + 505, + 212 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 505, + 212 + ], + "score": 1.0, + "content": "[43] Adina Williams, Nikita Nangia, and Samuel R Bowman. A broad-coverage challenge corpus", + "type": "text", + "cross_page": true + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 208, + 385, + 221 + ], + "spans": [ + { + "bbox": [ + 127, + 208, + 385, + 221 + ], + "score": 1.0, + "content": "for sentence understanding through inference. In NAACL, 2017.", + "type": "text", + "cross_page": true + } + ], + "index": 11, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 224, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 106, + 224, + 426, + 236 + ], + "score": 1.0, + "content": "[44] Pranav Rajpurkar, Jian Zhang, Konstantin Lopyrev, and Percy Liang. Squad:", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 427, + 224, + 465, + 235 + ], + "score": 0.74, + "content": "^ { 1 0 0 , 0 0 0 + }", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 466, + 224, + 505, + 236 + ], + "score": 1.0, + "content": "questions", + "type": "text", + "cross_page": true + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 235, + 347, + 246 + ], + "spans": [ + { + "bbox": [ + 127, + 235, + 347, + 246 + ], + "score": 1.0, + "content": "for machine comprehension of text. In EMNLP, 2016.", + "type": "text", + "cross_page": true + } + ], + "index": 13, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 249, + 505, + 262 + ], + "spans": [ + { + "bbox": [ + 106, + 249, + 505, + 262 + ], + "score": 1.0, + "content": "[45] Pranav Rajpurkar, Robin Jia, and Percy Liang. Know what you don’t know: Unanswerable", + "type": "text", + "cross_page": true + } + ], + "index": 14, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 260, + 271, + 273 + ], + "spans": [ + { + "bbox": [ + 127, + 260, + 271, + 273 + ], + "score": 1.0, + "content": "questions for squad. In ACL, 2018.", + "type": "text", + "cross_page": true + } + ], + "index": 15, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 275, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 106, + 275, + 505, + 288 + ], + "score": 1.0, + "content": "[46] Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr", + "type": "text", + "cross_page": true + } + ], + "index": 16, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 285, + 506, + 299 + ], + "spans": [ + { + "bbox": [ + 127, + 285, + 506, + 299 + ], + "score": 1.0, + "content": "Dollár, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In European", + "type": "text", + "cross_page": true + } + ], + "index": 17 + }, + { + "bbox": [ + 127, + 298, + 384, + 309 + ], + "spans": [ + { + "bbox": [ + 127, + 298, + 384, + 309 + ], + "score": 1.0, + "content": "conference on computer vision, pages 740–755. Springer, 2014.", + "type": "text", + "cross_page": true + } + ], + "index": 18, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 312, + 506, + 325 + ], + "spans": [ + { + "bbox": [ + 106, + 312, + 506, + 325 + ], + "score": 1.0, + "content": "[47] Mingxing Tan, Ruoming Pang, and Quoc V Le. Efficientdet: Scalable and efficient object detec-", + "type": "text", + "cross_page": true + } + ], + "index": 19, + "is_list_start_line": true + }, + { + "bbox": [ + 126, + 322, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 126, + 322, + 506, + 336 + ], + "score": 1.0, + "content": "tion. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition,", + "type": "text", + "cross_page": true + } + ], + "index": 20 + }, + { + "bbox": [ + 126, + 334, + 238, + 346 + ], + "spans": [ + { + "bbox": [ + 126, + 334, + 238, + 346 + ], + "score": 1.0, + "content": "pages 10781–10790, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 21, + "is_list_end_line": true + }, + { + "bbox": [ + 105, + 347, + 506, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 506, + 362 + ], + "score": 1.0, + "content": "[48] Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping", + "type": "text", + "cross_page": true + } + ], + "index": 22, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 359, + 506, + 372 + ], + "spans": [ + { + "bbox": [ + 127, + 359, + 506, + 372 + ], + "score": 1.0, + "content": "Luo, and Ling Shao. Pyramid vision transformer: A versatile backbone for dense prediction", + "type": "text", + "cross_page": true + } + ], + "index": 23 + }, + { + "bbox": [ + 127, + 371, + 380, + 384 + ], + "spans": [ + { + "bbox": [ + 127, + 371, + 380, + 384 + ], + "score": 1.0, + "content": "without convolutions. arXiv preprint arXiv:2102.12122, 2021.", + "type": "text", + "cross_page": true + } + ], + "index": 24, + "is_list_end_line": true + }, + { + "bbox": [ + 106, + 386, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 106, + 386, + 505, + 398 + ], + "score": 1.0, + "content": "[49] Josh Beal, Eric Kim, Eric Tzeng, Dong Huk Park, Andrew Zhai, and Dmitry Kislyuk. Toward", + "type": "text", + "cross_page": true + } + ], + "index": 25, + "is_list_start_line": true + }, + { + "bbox": [ + 127, + 397, + 434, + 408 + ], + "spans": [ + { + "bbox": [ + 127, + 397, + 434, + 408 + ], + "score": 1.0, + "content": "transformer-based object detection. arXiv preprint arXiv:2012.09958, 2020.", + "type": "text", + "cross_page": true + } + ], + "index": 26, + "is_list_end_line": true + } + ], + "index": 29.5, + "bbox_fs": [ + 105, + 235, + 507, + 723 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 45, + 507, + 725 + ], + "lines": [ + { + "bbox": [ + 104, + 71, + 505, + 88 + ], + "spans": [ + { + "bbox": [ + 104, + 71, + 505, + 88 + ], + "score": 1.0, + "content": "[16] Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 126, + 83, + 245, + 98 + ], + "spans": [ + { + "bbox": [ + 126, + 83, + 245, + 98 + ], + "score": 1.0, + "content": "recognition. In CVPR, 2016.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 106, + 99, + 505, + 111 + ], + "spans": [ + { + "bbox": [ + 106, + 99, + 505, + 111 + ], + "score": 1.0, + "content": "[17] Mingxing Tan and Quoc Le. Efficientnet: Rethinking model scaling for convolutional neural", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 127, + 110, + 235, + 122 + ], + "spans": [ + { + "bbox": [ + 127, + 110, + 235, + 122 + ], + "score": 1.0, + "content": "networks. In ICML, 2019.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 104, + 123, + 505, + 140 + ], + "spans": [ + { + "bbox": [ + 104, + 123, + 505, + 140 + ], + "score": 1.0, + "content": "[18] Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. Neural machine translation by jointly", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 126, + 136, + 316, + 150 + ], + "spans": [ + { + "bbox": [ + 126, + 136, + 316, + 150 + ], + "score": 1.0, + "content": "learning to align and translate. In ICLR, 2015.", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 105, + 151, + 506, + 165 + ], + "spans": [ + { + "bbox": [ + 105, + 151, + 506, + 165 + ], + "score": 1.0, + "content": "[19] Kurt Hornik, Maxwell Stinchcombe, and Halbert White. Multilayer feedforward networks are", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 126, + 161, + 385, + 175 + ], + "spans": [ + { + "bbox": [ + 126, + 161, + 385, + 175 + ], + "score": 1.0, + "content": "universal approximators. Neural networks, 2(5):359–366, 1989.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 106, + 177, + 506, + 189 + ], + "spans": [ + { + "bbox": [ + 106, + 177, + 506, + 189 + ], + "score": 1.0, + "content": "[20] Ilya Tolstikhin, Neil Houlsby, Alexander Kolesnikov, Lucas Beyer, Xiaohua Zhai, Thomas Un-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 127, + 187, + 507, + 201 + ], + "spans": [ + { + "bbox": [ + 127, + 187, + 507, + 201 + ], + "score": 1.0, + "content": "terthiner, Jessica Yung, Daniel Keysers, Jakob Uszkoreit, Mario Lucic, and Alexey Dosovitskiy.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 127, + 198, + 480, + 213 + ], + "spans": [ + { + "bbox": [ + 127, + 198, + 480, + 213 + ], + "score": 1.0, + "content": "Mlp-mixer: An all-mlp architecture for vision. arXiv preprint arXiv:2105.01601, 2021.", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 104, + 212, + 506, + 228 + ], + "spans": [ + { + "bbox": [ + 104, + 212, + 506, + 228 + ], + "score": 1.0, + "content": "[21] Luke Melas-Kyriazi. Do you even need attention? a stack of feed-forward layers does surpris-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 127, + 225, + 388, + 239 + ], + "spans": [ + { + "bbox": [ + 127, + 225, + 388, + 239 + ], + "score": 1.0, + "content": "ingly well on imagenet. arXiv preprint arXiv:2105.02723, 2021.", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 106, + 239, + 507, + 254 + ], + "spans": [ + { + "bbox": [ + 106, + 239, + 507, + 254 + ], + "score": 1.0, + "content": "[22] Hugo Touvron, Piotr Bojanowski, Mathilde Caron, Matthieu Cord, Alaaeldin El-Nouby,", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 126, + 250, + 506, + 265 + ], + "spans": [ + { + "bbox": [ + 126, + 250, + 506, + 265 + ], + "score": 1.0, + "content": "Edouard Grave, Armand Joulin, Gabriel Synnaeve, Jakob Verbeek, and Hervé Jégou. Resmlp:", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 126, + 260, + 506, + 276 + ], + "spans": [ + { + "bbox": [ + 126, + 260, + 506, + 276 + ], + "score": 1.0, + "content": "Feedforward networks for image classification with data-efficient training. arXiv preprint", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 125, + 272, + 232, + 285 + ], + "spans": [ + { + "bbox": [ + 125, + 272, + 232, + 285 + ], + "score": 1.0, + "content": "arXiv:2105.03404, 2021.", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 105, + 286, + 507, + 302 + ], + "spans": [ + { + "bbox": [ + 105, + 286, + 507, + 302 + ], + "score": 1.0, + "content": "[23] Xiaohan Ding, Xiangyu Zhang, Jungong Han, and Guiguang Ding. Repmlp: Re-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 126, + 298, + 505, + 312 + ], + "spans": [ + { + "bbox": [ + 126, + 298, + 505, + 312 + ], + "score": 1.0, + "content": "parameterizing convolutions into fully-connected layers for image recognition. arXiv preprint", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 126, + 308, + 231, + 322 + ], + "spans": [ + { + "bbox": [ + 126, + 308, + 231, + 322 + ], + "score": 1.0, + "content": "arXiv:2105.01883, 2021.", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 105, + 324, + 506, + 339 + ], + "spans": [ + { + "bbox": [ + 105, + 324, + 506, + 339 + ], + "score": 1.0, + "content": "[24] Dan Hendrycks and Kevin Gimpel. Gaussian error linear units (gelus). arXiv preprint", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 128, + 336, + 231, + 348 + ], + "spans": [ + { + "bbox": [ + 128, + 336, + 231, + 348 + ], + "score": 1.0, + "content": "arXiv:1606.08415, 2016.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 351, + 507, + 365 + ], + "spans": [ + { + "bbox": [ + 106, + 351, + 507, + 365 + ], + "score": 1.0, + "content": "[25] Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, and Liang-Chieh Chen.", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 127, + 361, + 417, + 375 + ], + "spans": [ + { + "bbox": [ + 127, + 361, + 417, + 375 + ], + "score": 1.0, + "content": "Mobilenetv2: Inverted residuals and linear bottlenecks. In CVPR, 2018.", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 104, + 375, + 506, + 392 + ], + "spans": [ + { + "bbox": [ + 104, + 375, + 506, + 392 + ], + "score": 1.0, + "content": "[26] Yann N Dauphin, Angela Fan, Michael Auli, and David Grangier. Language modeling with", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 126, + 388, + 316, + 401 + ], + "spans": [ + { + "bbox": [ + 126, + 388, + 316, + 401 + ], + "score": 1.0, + "content": "gated convolutional networks. In ICML, 2017.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 106, + 403, + 493, + 417 + ], + "spans": [ + { + "bbox": [ + 106, + 403, + 493, + 417 + ], + "score": 1.0, + "content": "[27] Noam Shazeer. Glu variants improve transformer. arXiv preprint arXiv:2002.05202, 2020.", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 106, + 418, + 505, + 430 + ], + "spans": [ + { + "bbox": [ + 106, + 418, + 505, + 430 + ], + "score": 1.0, + "content": "[28] Felix Wu, Angela Fan, Alexei Baevski, Yann N Dauphin, and Michael Auli. Pay less attention", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 127, + 428, + 369, + 442 + ], + "spans": [ + { + "bbox": [ + 127, + 428, + 369, + 442 + ], + "score": 1.0, + "content": "with lightweight and dynamic convolutions. In ICLR, 2019.", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 106, + 443, + 505, + 457 + ], + "spans": [ + { + "bbox": [ + 106, + 443, + 505, + 457 + ], + "score": 1.0, + "content": "[29] Rupesh Kumar Srivastava, Klaus Greff, and Jürgen Schmidhuber. Highway networks. arXiv", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 124, + 454, + 267, + 468 + ], + "spans": [ + { + "bbox": [ + 124, + 454, + 267, + 468 + ], + "score": 1.0, + "content": "preprint arXiv:1505.00387, 2015.", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 106, + 470, + 462, + 484 + ], + "spans": [ + { + "bbox": [ + 106, + 470, + 462, + 484 + ], + "score": 1.0, + "content": "[30] Jie Hu, Li Shen, and Gang Sun. Squeeze-and-excitation networks. In CVPR, 2018.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 105, + 484, + 507, + 499 + ], + "spans": [ + { + "bbox": [ + 105, + 484, + 507, + 499 + ], + "score": 1.0, + "content": "[31] Jia Deng, Wei Dong, Richard Socher, Li-Jia Li, Kai Li, and Li Fei-Fei. Imagenet: A large-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 126, + 496, + 506, + 510 + ], + "spans": [ + { + "bbox": [ + 126, + 496, + 506, + 510 + ], + "score": 1.0, + "content": "scale hierarchical image database. In 2009 IEEE conference on computer vision and pattern", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 125, + 506, + 292, + 520 + ], + "spans": [ + { + "bbox": [ + 125, + 506, + 292, + 520 + ], + "score": 1.0, + "content": "recognition, pages 248–255. Ieee, 2009.", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 106, + 521, + 505, + 535 + ], + "spans": [ + { + "bbox": [ + 106, + 521, + 505, + 535 + ], + "score": 1.0, + "content": "[32] Gao Huang, Yu Sun, Zhuang Liu, Daniel Sedra, and Kilian Q Weinberger. Deep networks with", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 126, + 532, + 264, + 546 + ], + "spans": [ + { + "bbox": [ + 126, + 532, + 264, + 546 + ], + "score": 1.0, + "content": "stochastic depth. In ECCV, 2016.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 104, + 545, + 507, + 564 + ], + "spans": [ + { + "bbox": [ + 104, + 545, + 507, + 564 + ], + "score": 1.0, + "content": "[33] Andrew Brock, Soham De, Samuel L Smith, and Karen Simonyan. High-performance large-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 126, + 559, + 481, + 572 + ], + "spans": [ + { + "bbox": [ + 126, + 559, + 481, + 572 + ], + "score": 1.0, + "content": "scale image recognition without normalization. arXiv preprint arXiv:2102.06171, 2021.", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 106, + 573, + 506, + 587 + ], + "spans": [ + { + "bbox": [ + 106, + 573, + 506, + 587 + ], + "score": 1.0, + "content": "[34] Mingxing Tan and Quoc V Le. Efficientnetv2: Smaller models and faster training. In ICML,", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 126, + 583, + 155, + 598 + ], + "spans": [ + { + "bbox": [ + 126, + 583, + 155, + 598 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 106, + 599, + 507, + 614 + ], + "spans": [ + { + "bbox": [ + 106, + 599, + 507, + 614 + ], + "score": 1.0, + "content": "[35] Irwan Bello. Lambdanetworks: Modeling long-range interactions without attention. In ICLR,", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 126, + 610, + 156, + 625 + ], + "spans": [ + { + "bbox": [ + 126, + 610, + 156, + 625 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 105, + 626, + 506, + 640 + ], + "spans": [ + { + "bbox": [ + 105, + 626, + 506, + 640 + ], + "score": 1.0, + "content": "[36] Ashish Vaswani, Prajit Ramachandran, Aravind Srinivas, Niki Parmar, Blake Hechtman, and", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 126, + 637, + 506, + 651 + ], + "spans": [ + { + "bbox": [ + 126, + 637, + 506, + 651 + ], + "score": 1.0, + "content": "Jonathon Shlens. Scaling local self-attention for parameter efficient visual backbones. In CVPR,", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 126, + 645, + 155, + 662 + ], + "spans": [ + { + "bbox": [ + 126, + 645, + 155, + 662 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 104, + 661, + 507, + 678 + ], + "spans": [ + { + "bbox": [ + 104, + 661, + 507, + 678 + ], + "score": 1.0, + "content": "[37] Aravind Srinivas, Tsung-Yi Lin, Niki Parmar, Jonathon Shlens, Pieter Abbeel, and Ashish", + "type": "text" + } + ], + "index": 46 + }, + { + "bbox": [ + 127, + 673, + 507, + 687 + ], + "spans": [ + { + "bbox": [ + 127, + 673, + 507, + 687 + ], + "score": 1.0, + "content": "Vaswani. Bottleneck transformers for visual recognition. arXiv preprint arXiv:2101.11605,", + "type": "text" + } + ], + "index": 47 + }, + { + "bbox": [ + 126, + 684, + 156, + 699 + ], + "spans": [ + { + "bbox": [ + 126, + 684, + 156, + 699 + ], + "score": 1.0, + "content": "2021.", + "type": "text" + } + ], + "index": 48 + }, + { + "bbox": [ + 106, + 698, + 507, + 715 + ], + "spans": [ + { + "bbox": [ + 106, + 698, + 507, + 715 + ], + "score": 1.0, + "content": "[38] Haiping Wu, Bin Xiao, Noel Codella, Mengchen Liu, Xiyang Dai, Lu Yuan, and Lei Zhang.", + "type": "text" + } + ], + "index": 49 + }, + { + "bbox": [ + 127, + 711, + 506, + 725 + ], + "spans": [ + { + "bbox": [ + 127, + 711, + 506, + 725 + ], + "score": 1.0, + "content": "Cvt: Introducing convolutions to vision transformers. arXiv preprint arXiv:2103.15808, 2021.", + "type": "text" + } + ], + "index": 50 + } + ], + "index": 25 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 751 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 312, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 312, + 754 + ], + "score": 1.0, + "content": "11", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 105, + 45, + 507, + 725 + ], + "lines": [], + "index": 25, + "bbox_fs": [ + 104, + 71, + 507, + 725 + ], + "lines_deleted": true + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 105, + 71, + 507, + 410 + ], + "lines": [ + { + "bbox": [ + 105, + 72, + 506, + 87 + ], + "spans": [ + { + "bbox": [ + 105, + 72, + 506, + 87 + ], + "score": 1.0, + "content": "[39] Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, and Piotr Dollár. Design-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 127, + 84, + 304, + 96 + ], + "spans": [ + { + "bbox": [ + 127, + 84, + 304, + 96 + ], + "score": 1.0, + "content": "ing network design spaces. In CVPR, 2020.", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 107, + 99, + 506, + 111 + ], + "spans": [ + { + "bbox": [ + 107, + 99, + 506, + 111 + ], + "score": 1.0, + "content": "[40] Yi Tay, Dara Bahri, Donald Metzler, Da-Cheng Juan, Zhe Zhao, and Che Zheng. Synthesizer:", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 127, + 109, + 486, + 123 + ], + "spans": [ + { + "bbox": [ + 127, + 109, + 486, + 123 + ], + "score": 1.0, + "content": "Rethinking self-attention in transformer models. arXiv preprint arXiv:2005.00743, 2020.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 105, + 124, + 507, + 138 + ], + "spans": [ + { + "bbox": [ + 105, + 124, + 507, + 138 + ], + "score": 1.0, + "content": "[41] Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel R Bowman.", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 127, + 135, + 505, + 148 + ], + "spans": [ + { + "bbox": [ + 127, + 135, + 505, + 148 + ], + "score": 1.0, + "content": "Glue: A multi-task benchmark and analysis platform for natural language understanding. In", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 128, + 147, + 179, + 157 + ], + "spans": [ + { + "bbox": [ + 128, + 147, + 179, + 157 + ], + "score": 1.0, + "content": "ICLR, 2019.", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 106, + 161, + 506, + 174 + ], + "spans": [ + { + "bbox": [ + 106, + 161, + 506, + 174 + ], + "score": 1.0, + "content": "[42] Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B Brown, Benjamin Chess, Rewon Child,", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 127, + 171, + 505, + 186 + ], + "spans": [ + { + "bbox": [ + 127, + 171, + 505, + 186 + ], + "score": 1.0, + "content": "Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. Scaling laws for neural language", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 127, + 184, + 325, + 194 + ], + "spans": [ + { + "bbox": [ + 127, + 184, + 325, + 194 + ], + "score": 1.0, + "content": "models. arXiv preprint arXiv:2001.08361, 2020.", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 105, + 197, + 505, + 212 + ], + "spans": [ + { + "bbox": [ + 105, + 197, + 505, + 212 + ], + "score": 1.0, + "content": "[43] Adina Williams, Nikita Nangia, and Samuel R Bowman. A broad-coverage challenge corpus", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 127, + 208, + 385, + 221 + ], + "spans": [ + { + "bbox": [ + 127, + 208, + 385, + 221 + ], + "score": 1.0, + "content": "for sentence understanding through inference. In NAACL, 2017.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 106, + 224, + 505, + 236 + ], + "spans": [ + { + "bbox": [ + 106, + 224, + 426, + 236 + ], + "score": 1.0, + "content": "[44] Pranav Rajpurkar, Jian Zhang, Konstantin Lopyrev, and Percy Liang. Squad:", + "type": "text" + }, + { + "bbox": [ + 427, + 224, + 465, + 235 + ], + "score": 0.74, + "content": "^ { 1 0 0 , 0 0 0 + }", + "type": "inline_equation" + }, + { + "bbox": [ + 466, + 224, + 505, + 236 + ], + "score": 1.0, + "content": "questions", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 127, + 235, + 347, + 246 + ], + "spans": [ + { + "bbox": [ + 127, + 235, + 347, + 246 + ], + "score": 1.0, + "content": "for machine comprehension of text. In EMNLP, 2016.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 249, + 505, + 262 + ], + "spans": [ + { + "bbox": [ + 106, + 249, + 505, + 262 + ], + "score": 1.0, + "content": "[45] Pranav Rajpurkar, Robin Jia, and Percy Liang. Know what you don’t know: Unanswerable", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 127, + 260, + 271, + 273 + ], + "spans": [ + { + "bbox": [ + 127, + 260, + 271, + 273 + ], + "score": 1.0, + "content": "questions for squad. In ACL, 2018.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 106, + 275, + 505, + 288 + ], + "spans": [ + { + "bbox": [ + 106, + 275, + 505, + 288 + ], + "score": 1.0, + "content": "[46] Tsung-Yi Lin, Michael Maire, Serge Belongie, James Hays, Pietro Perona, Deva Ramanan, Piotr", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 127, + 285, + 506, + 299 + ], + "spans": [ + { + "bbox": [ + 127, + 285, + 506, + 299 + ], + "score": 1.0, + "content": "Dollár, and C Lawrence Zitnick. Microsoft coco: Common objects in context. In European", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 127, + 298, + 384, + 309 + ], + "spans": [ + { + "bbox": [ + 127, + 298, + 384, + 309 + ], + "score": 1.0, + "content": "conference on computer vision, pages 740–755. Springer, 2014.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 106, + 312, + 506, + 325 + ], + "spans": [ + { + "bbox": [ + 106, + 312, + 506, + 325 + ], + "score": 1.0, + "content": "[47] Mingxing Tan, Ruoming Pang, and Quoc V Le. Efficientdet: Scalable and efficient object detec-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 126, + 322, + 506, + 336 + ], + "spans": [ + { + "bbox": [ + 126, + 322, + 506, + 336 + ], + "score": 1.0, + "content": "tion. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition,", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 126, + 334, + 238, + 346 + ], + "spans": [ + { + "bbox": [ + 126, + 334, + 238, + 346 + ], + "score": 1.0, + "content": "pages 10781–10790, 2020.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 105, + 347, + 506, + 362 + ], + "spans": [ + { + "bbox": [ + 105, + 347, + 506, + 362 + ], + "score": 1.0, + "content": "[48] Wenhai Wang, Enze Xie, Xiang Li, Deng-Ping Fan, Kaitao Song, Ding Liang, Tong Lu, Ping", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 127, + 359, + 506, + 372 + ], + "spans": [ + { + "bbox": [ + 127, + 359, + 506, + 372 + ], + "score": 1.0, + "content": "Luo, and Ling Shao. Pyramid vision transformer: A versatile backbone for dense prediction", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 127, + 371, + 380, + 384 + ], + "spans": [ + { + "bbox": [ + 127, + 371, + 380, + 384 + ], + "score": 1.0, + "content": "without convolutions. arXiv preprint arXiv:2102.12122, 2021.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 106, + 386, + 505, + 398 + ], + "spans": [ + { + "bbox": [ + 106, + 386, + 505, + 398 + ], + "score": 1.0, + "content": "[49] Josh Beal, Eric Kim, Eric Tzeng, Dong Huk Park, Andrew Zhai, and Dmitry Kislyuk. Toward", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 127, + 397, + 434, + 408 + ], + "spans": [ + { + "bbox": [ + 127, + 397, + 434, + 408 + ], + "score": 1.0, + "content": "transformer-based object detection. arXiv preprint arXiv:2012.09958, 2020.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 13 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 300, + 741, + 311, + 750 + ], + "lines": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "spans": [ + { + "bbox": [ + 299, + 740, + 313, + 754 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "list", + "bbox": [ + 105, + 71, + 507, + 410 + ], + "lines": [], + "index": 13, + "bbox_fs": [ + 105, + 72, + 507, + 408 + ], + "lines_deleted": true + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/train/KBnXrODoBW/KBnXrODoBW_model.json b/parse/train/KBnXrODoBW/KBnXrODoBW_model.json new file mode 100644 index 0000000000000000000000000000000000000000..d3d2a825fe51c404c7f2d0d13c84f555c1224df8 --- /dev/null +++ b/parse/train/KBnXrODoBW/KBnXrODoBW_model.json @@ -0,0 +1,16796 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 398, + 682, + 1303, + 682, + 1303, + 1016, + 398, + 1016 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1274, + 1404, + 1274, + 1404, + 1487, + 298, + 1487 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 299, + 1501, + 1404, + 1501, + 1404, + 1685, + 299, + 1685 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 298, + 1138, + 1404, + 1138, + 1404, + 1260, + 298, + 1260 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 299, + 1698, + 1403, + 1698, + 1403, + 1851, + 299, + 1851 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 298, + 1864, + 1403, + 1864, + 1403, + 1988, + 298, + 1988 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 551, + 442, + 1152, + 442, + 1152, + 538, + 551, + 538 + ], + "score": 0.931 + }, + { + "category_id": 0, + "poly": [ + 299, + 1068, + 530, + 1068, + 530, + 1106, + 299, + 1106 + ], + "score": 0.906 + }, + { + "category_id": 0, + "poly": [ + 787, + 614, + 912, + 614, + 912, + 651, + 787, + 651 + ], + "score": 0.892 + }, + { + "category_id": 0, + "poly": [ + 613, + 272, + 1089, + 272, + 1089, + 326, + 613, + 326 + ], + "score": 0.885 + }, + { + "category_id": 2, + "poly": [ + 298, + 2033, + 1070, + 2033, + 1070, + 2061, + 298, + 2061 + ], + "score": 0.865 + }, + { + "category_id": 13, + "poly": [ + 750, + 1760, + 806, + 1760, + 806, + 1789, + 750, + 1789 + ], + "score": 0.86, + "latex": "66 \\%" + }, + { + "category_id": 13, + "poly": [ + 1195, + 1760, + 1236, + 1760, + 1236, + 1789, + 1195, + 1789 + ], + "score": 0.85, + "latex": "3 \\%" + }, + { + "category_id": 13, + "poly": [ + 595, + 834, + 673, + 834, + 673, + 866, + 595, + 866 + ], + "score": 0.75, + "latex": "\\mathrm { g M L P }" + }, + { + "category_id": 13, + "poly": [ + 1219, + 1698, + 1298, + 1698, + 1298, + 1731, + 1219, + 1731 + ], + "score": 0.3, + "latex": "\\mathrm { g M L P }" + }, + { + "category_id": 13, + "poly": [ + 413, + 1698, + 490, + 1698, + 490, + 1731, + 413, + 1731 + ], + "score": 0.3, + "latex": "\\mathrm { g M L P }" + }, + { + "category_id": 13, + "poly": [ + 440, + 925, + 518, + 925, + 518, + 957, + 440, + 957 + ], + "score": 0.3, + "latex": "\\mathrm { g M L P }" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1065.0, + 536.0, + 1065.0, + 536.0, + 1113.0, + 292.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 613.0, + 919.0, + 613.0, + 919.0, + 655.0, + 783.0, + 655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 603.0, + 270.0, + 1094.0, + 270.0, + 1094.0, + 327.0, + 603.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 2028.0, + 1075.0, + 2028.0, + 1075.0, + 2067.0, + 293.0, + 2067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 683.0, + 1303.0, + 683.0, + 1303.0, + 715.0, + 395.0, + 715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 709.0, + 1310.0, + 709.0, + 1310.0, + 751.0, + 392.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 740.0, + 1309.0, + 740.0, + 1309.0, + 782.0, + 392.0, + 782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 773.0, + 1306.0, + 773.0, + 1306.0, + 807.0, + 394.0, + 807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 805.0, + 1305.0, + 805.0, + 1305.0, + 835.0, + 394.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 833.0, + 594.0, + 833.0, + 594.0, + 869.0, + 393.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 833.0, + 1306.0, + 833.0, + 1306.0, + 869.0, + 674.0, + 869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 865.0, + 1306.0, + 865.0, + 1306.0, + 899.0, + 394.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 894.0, + 1304.0, + 894.0, + 1304.0, + 929.0, + 393.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 924.0, + 439.0, + 924.0, + 439.0, + 960.0, + 394.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 519.0, + 924.0, + 1306.0, + 924.0, + 1306.0, + 960.0, + 519.0, + 960.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 956.0, + 1307.0, + 956.0, + 1307.0, + 990.0, + 393.0, + 990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 988.0, + 717.0, + 988.0, + 717.0, + 1020.0, + 394.0, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1273.0, + 1404.0, + 1273.0, + 1404.0, + 1310.0, + 293.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1305.0, + 1408.0, + 1305.0, + 1408.0, + 1339.0, + 294.0, + 1339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1335.0, + 1405.0, + 1335.0, + 1405.0, + 1369.0, + 294.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1365.0, + 1404.0, + 1365.0, + 1404.0, + 1400.0, + 293.0, + 1400.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1396.0, + 1405.0, + 1396.0, + 1405.0, + 1429.0, + 293.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1427.0, + 1408.0, + 1427.0, + 1408.0, + 1461.0, + 294.0, + 1461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1455.0, + 1348.0, + 1455.0, + 1348.0, + 1491.0, + 293.0, + 1491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1499.0, + 1409.0, + 1499.0, + 1409.0, + 1537.0, + 293.0, + 1537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1531.0, + 1406.0, + 1531.0, + 1406.0, + 1566.0, + 294.0, + 1566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1564.0, + 1405.0, + 1564.0, + 1405.0, + 1596.0, + 295.0, + 1596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1593.0, + 1405.0, + 1593.0, + 1405.0, + 1626.0, + 294.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1622.0, + 1406.0, + 1622.0, + 1406.0, + 1658.0, + 293.0, + 1658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1649.0, + 908.0, + 1649.0, + 908.0, + 1693.0, + 293.0, + 1693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1137.0, + 1405.0, + 1137.0, + 1405.0, + 1171.0, + 293.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1168.0, + 1410.0, + 1168.0, + 1410.0, + 1204.0, + 293.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1199.0, + 1405.0, + 1199.0, + 1405.0, + 1231.0, + 296.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1228.0, + 1288.0, + 1228.0, + 1288.0, + 1264.0, + 294.0, + 1264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1697.0, + 412.0, + 1697.0, + 412.0, + 1734.0, + 295.0, + 1734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 1697.0, + 1218.0, + 1697.0, + 1218.0, + 1734.0, + 491.0, + 1734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.0, + 1697.0, + 1405.0, + 1697.0, + 1405.0, + 1734.0, + 1299.0, + 1734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1728.0, + 1407.0, + 1728.0, + 1407.0, + 1764.0, + 294.0, + 1764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1760.0, + 749.0, + 1760.0, + 749.0, + 1793.0, + 295.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 1760.0, + 1194.0, + 1760.0, + 1194.0, + 1793.0, + 807.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1237.0, + 1760.0, + 1405.0, + 1760.0, + 1405.0, + 1793.0, + 1237.0, + 1793.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1786.0, + 1403.0, + 1786.0, + 1403.0, + 1825.0, + 294.0, + 1825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1819.0, + 1410.0, + 1819.0, + 1410.0, + 1856.0, + 295.0, + 1856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1864.0, + 1408.0, + 1864.0, + 1408.0, + 1900.0, + 294.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1892.0, + 1404.0, + 1892.0, + 1404.0, + 1932.0, + 292.0, + 1932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1926.0, + 1407.0, + 1926.0, + 1407.0, + 1962.0, + 293.0, + 1962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1956.0, + 1406.0, + 1956.0, + 1406.0, + 1991.0, + 293.0, + 1991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 444.0, + 1154.0, + 444.0, + 1154.0, + 479.0, + 554.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 475.0, + 1021.0, + 475.0, + 1021.0, + 505.0, + 678.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 550.0, + 504.0, + 1151.0, + 504.0, + 1151.0, + 541.0, + 550.0, + 541.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1792, + 1405, + 1792, + 1405, + 2009, + 297, + 2009 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 297, + 1064, + 1404, + 1064, + 1404, + 1281, + 297, + 1281 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 298, + 1291, + 1404, + 1291, + 1404, + 1447, + 298, + 1447 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 866, + 1405, + 866, + 1405, + 1053, + 297, + 1053 + ], + "score": 0.978 + }, + { + "category_id": 3, + "poly": [ + 314, + 215, + 828, + 215, + 828, + 627, + 314, + 627 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 296, + 1686, + 1404, + 1686, + 1404, + 1781, + 296, + 1781 + ], + "score": 0.963 + }, + { + "category_id": 4, + "poly": [ + 296, + 658, + 1408, + 658, + 1408, + 813, + 296, + 813 + ], + "score": 0.959 + }, + { + "category_id": 1, + "poly": [ + 295, + 1555, + 1400, + 1555, + 1400, + 1619, + 295, + 1619 + ], + "score": 0.957 + }, + { + "category_id": 8, + "poly": [ + 601, + 1633, + 1098, + 1633, + 1098, + 1674, + 601, + 1674 + ], + "score": 0.935 + }, + { + "category_id": 0, + "poly": [ + 298, + 1486, + 443, + 1486, + 443, + 1524, + 298, + 1524 + ], + "score": 0.915 + }, + { + "category_id": 9, + "poly": [ + 1366, + 1641, + 1399, + 1641, + 1399, + 1670, + 1366, + 1670 + ], + "score": 0.879 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 858, + 2061, + 858, + 2085, + 841, + 2085 + ], + "score": 0.725 + }, + { + "category_id": 2, + "poly": [ + 942, + 222, + 1312, + 222, + 1312, + 249, + 942, + 249 + ], + "score": 0.718 + }, + { + "category_id": 1, + "poly": [ + 872, + 473, + 1372, + 473, + 1372, + 618, + 872, + 618 + ], + "score": 0.533 + }, + { + "category_id": 1, + "poly": [ + 866, + 263, + 1282, + 263, + 1282, + 448, + 866, + 448 + ], + "score": 0.48 + }, + { + "category_id": 0, + "poly": [ + 942, + 222, + 1312, + 222, + 1312, + 249, + 942, + 249 + ], + "score": 0.229 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.177 + }, + { + "category_id": 14, + "poly": [ + 599, + 1631, + 1101, + 1631, + 1101, + 1674, + 599, + 1674 + ], + "score": 0.92, + "latex": "Z = \\sigma ( X U ) , \\qquad \\tilde { Z } = s ( Z ) , \\qquad Y = \\tilde { Z } V" + }, + { + "category_id": 13, + "poly": [ + 941, + 1980, + 984, + 1980, + 984, + 2009, + 941, + 2009 + ], + "score": 0.92, + "latex": "s ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 1275, + 1554, + 1401, + 1554, + 1401, + 1585, + 1275, + 1585 + ], + "score": 0.91, + "latex": "\\ b { X } \\in \\mathbb { R } ^ { n \\times d }" + }, + { + "category_id": 13, + "poly": [ + 889, + 1793, + 933, + 1793, + 933, + 1827, + 889, + 1827 + ], + "score": 0.9, + "latex": "s ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 688, + 1248, + 766, + 1248, + 766, + 1277, + 688, + 1277 + ], + "score": 0.89, + "latex": "+ 4 . 4 \\%" + }, + { + "category_id": 13, + "poly": [ + 1274, + 1718, + 1401, + 1718, + 1401, + 1747, + 1274, + 1747 + ], + "score": 0.88, + "latex": "7 6 8 \\times 3 0 7 2" + }, + { + "category_id": 13, + "poly": [ + 1359, + 1914, + 1403, + 1914, + 1403, + 1948, + 1359, + 1948 + ], + "score": 0.88, + "latex": "s ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 344, + 1748, + 474, + 1748, + 474, + 1779, + 344, + 1779 + ], + "score": 0.86, + "latex": "3 0 7 2 \\times 7 6 8" + }, + { + "category_id": 13, + "poly": [ + 816, + 1096, + 878, + 1096, + 878, + 1126, + 816, + 1126 + ], + "score": 0.85, + "latex": "0 . 8 \\%" + }, + { + "category_id": 13, + "poly": [ + 616, + 990, + 692, + 990, + 692, + 1019, + 616, + 1019 + ], + "score": 0.85, + "latex": "8 7 . 7 \\%" + }, + { + "category_id": 13, + "poly": [ + 390, + 1158, + 428, + 1158, + 428, + 1186, + 390, + 1186 + ], + "score": 0.84, + "latex": "3 \\times" + }, + { + "category_id": 13, + "poly": [ + 961, + 990, + 1036, + 990, + 1036, + 1019, + 961, + 1019 + ], + "score": 0.83, + "latex": "8 2 . 1 \\%" + }, + { + "category_id": 13, + "poly": [ + 755, + 1559, + 777, + 1559, + 777, + 1584, + 755, + 1584 + ], + "score": 0.83, + "latex": "L" + }, + { + "category_id": 13, + "poly": [ + 854, + 1593, + 874, + 1593, + 874, + 1615, + 854, + 1615 + ], + "score": 0.8, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 983, + 1688, + 1009, + 1688, + 1009, + 1716, + 983, + 1716 + ], + "score": 0.79, + "latex": "V" + }, + { + "category_id": 13, + "poly": [ + 441, + 691, + 463, + 691, + 463, + 717, + 441, + 717 + ], + "score": 0.79, + "latex": "L" + }, + { + "category_id": 13, + "poly": [ + 905, + 1688, + 931, + 1688, + 931, + 1715, + 905, + 1715 + ], + "score": 0.78, + "latex": "U" + }, + { + "category_id": 13, + "poly": [ + 1048, + 1589, + 1064, + 1589, + 1064, + 1614, + 1048, + 1614 + ], + "score": 0.78, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 884, + 383, + 925, + 383, + 925, + 401, + 884, + 401 + ], + "score": 0.77, + "latex": "\\mathrm { ~ x ~ } =" + }, + { + "category_id": 13, + "poly": [ + 372, + 1694, + 393, + 1694, + 393, + 1715, + 372, + 1715 + ], + "score": 0.76, + "latex": "\\sigma" + }, + { + "category_id": 13, + "poly": [ + 1011, + 1248, + 1123, + 1248, + 1123, + 1281, + 1011, + 1281 + ], + "score": 0.7, + "latex": "\\mathbf { B E R T _ { l a r g e } }" + }, + { + "category_id": 13, + "poly": [ + 516, + 1749, + 624, + 1749, + 624, + 1780, + 516, + 1780 + ], + "score": 0.7, + "latex": "\\mathbf { B E R T _ { b a s e } }" + }, + { + "category_id": 13, + "poly": [ + 845, + 1893, + 860, + 1893, + 860, + 1912, + 845, + 1912 + ], + "score": 0.69, + "latex": "s" + }, + { + "category_id": 13, + "poly": [ + 884, + 406, + 924, + 406, + 924, + 426, + 884, + 426 + ], + "score": 0.67, + "latex": "\\mathrm { ~ x ~ } =" + }, + { + "category_id": 13, + "poly": [ + 884, + 336, + 924, + 336, + 924, + 354, + 884, + 354 + ], + "score": 0.66, + "latex": "\\mathrm { ~ x ~ } =" + }, + { + "category_id": 13, + "poly": [ + 512, + 1829, + 529, + 1829, + 529, + 1851, + 512, + 1851 + ], + "score": 0.66, + "latex": "s" + }, + { + "category_id": 13, + "poly": [ + 860, + 960, + 933, + 960, + 933, + 989, + 860, + 989 + ], + "score": 0.62, + "latex": "\\times \\ 1 \\mathbf { M }" + }, + { + "category_id": 13, + "poly": [ + 884, + 311, + 924, + 311, + 924, + 331, + 884, + 331 + ], + "score": 0.58, + "latex": "\\mathrm { ~ x ~ } =" + }, + { + "category_id": 13, + "poly": [ + 680, + 899, + 766, + 899, + 766, + 931, + 680, + 931 + ], + "score": 0.55, + "latex": "{ \\mathrm { g M L P s } }" + }, + { + "category_id": 13, + "poly": [ + 647, + 1384, + 721, + 1384, + 721, + 1416, + 647, + 1416 + ], + "score": 0.5, + "latex": "\\mathrm { g M L P }" + }, + { + "category_id": 13, + "poly": [ + 884, + 359, + 924, + 359, + 924, + 379, + 884, + 379 + ], + "score": 0.5, + "latex": "\\mathrm { ~ x ~ } =" + }, + { + "category_id": 13, + "poly": [ + 1182, + 1188, + 1268, + 1188, + 1268, + 1219, + 1182, + 1219 + ], + "score": 0.5, + "latex": "\\mathrm { g M L P s }" + }, + { + "category_id": 13, + "poly": [ + 1369, + 691, + 1394, + 691, + 1394, + 719, + 1369, + 719 + ], + "score": 0.49, + "latex": "\\odot" + }, + { + "category_id": 13, + "poly": [ + 595, + 660, + 671, + 660, + 671, + 691, + 595, + 691 + ], + "score": 0.49, + "latex": "\\mathrm { g M L P }" + }, + { + "category_id": 13, + "poly": [ + 1335, + 569, + 1361, + 569, + 1361, + 588, + 1335, + 588 + ], + "score": 0.47, + "latex": "\\mathrel { \\mathop : } = 1" + }, + { + "category_id": 13, + "poly": [ + 894, + 523, + 933, + 523, + 933, + 541, + 894, + 541 + ], + "score": 0.47, + "latex": "\\tt { v } =" + }, + { + "category_id": 13, + "poly": [ + 503, + 1021, + 615, + 1021, + 615, + 1055, + 503, + 1055 + ], + "score": 0.46, + "latex": "\\mathbf { B E R T _ { l a r g e } }" + }, + { + "category_id": 13, + "poly": [ + 1132, + 474, + 1162, + 474, + 1162, + 495, + 1132, + 495 + ], + "score": 0.45, + "latex": "\\mathbf { \\Psi } ( \\mathbf { x } )" + }, + { + "category_id": 13, + "poly": [ + 1161, + 406, + 1181, + 406, + 1181, + 423, + 1161, + 423 + ], + "score": 0.44, + "latex": "= ^ { 1 1 }" + }, + { + "category_id": 13, + "poly": [ + 1104, + 547, + 1124, + 547, + 1124, + 564, + 1104, + 564 + ], + "score": 0.43, + "latex": "\\mathrel { \\mathop : } = \\mathrel { \\mathop : }" + }, + { + "category_id": 13, + "poly": [ + 960, + 430, + 1001, + 430, + 1001, + 449, + 960, + 449 + ], + "score": 0.4, + "latex": "\\texttt { x + }" + }, + { + "category_id": 13, + "poly": [ + 899, + 1248, + 952, + 1248, + 952, + 1278, + 899, + 1278 + ], + "score": 0.33, + "latex": "\\mathrm { v } 2 . 0" + }, + { + "category_id": 13, + "poly": [ + 980, + 289, + 1008, + 289, + 1008, + 306, + 980, + 306 + ], + "score": 0.33, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 929, + 500, + 968, + 500, + 968, + 519, + 929, + 519 + ], + "score": 0.27, + "latex": "\\tt { v } =" + }, + { + "category_id": 13, + "poly": [ + 359, + 990, + 436, + 990, + 436, + 1022, + 359, + 1022 + ], + "score": 0.25, + "latex": "\\mathrm { g M L P }" + }, + { + "category_id": 15, + "poly": [ + 495.0, + 253.0, + 535.0, + 253.0, + 535.0, + 283.0, + 495.0, + 283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 294.0, + 492.0, + 294.0, + 492.0, + 330.0, + 374.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 327.0, + 441.0, + 327.0, + 441.0, + 344.0, + 426.0, + 344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 639.0, + 338.0, + 671.0, + 338.0, + 671.0, + 367.0, + 639.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 350.0, + 467.0, + 350.0, + 467.0, + 381.0, + 398.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 371.0, + 484.0, + 371.0, + 484.0, + 402.0, + 381.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 372.0, + 801.0, + 372.0, + 801.0, + 402.0, + 696.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 405.0, + 441.0, + 405.0, + 441.0, + 424.0, + 425.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 403.0, + 755.0, + 403.0, + 755.0, + 415.0, + 743.0, + 415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 387.0, + 424.0, + 477.0, + 424.0, + 477.0, + 453.0, + 387.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 720.0, + 416.0, + 776.0, + 416.0, + 776.0, + 444.0, + 720.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 452.0, + 441.0, + 452.0, + 441.0, + 469.0, + 424.0, + 469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 625.0, + 451.0, + 679.0, + 451.0, + 679.0, + 479.0, + 625.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 446.0, + 755.0, + 446.0, + 755.0, + 459.0, + 742.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 466.0, + 491.0, + 466.0, + 491.0, + 500.0, + 372.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 473.0, + 565.0, + 473.0, + 565.0, + 483.0, + 556.0, + 483.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 484.0, + 572.0, + 484.0, + 572.0, + 494.0, + 563.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 425.0, + 497.0, + 441.0, + 497.0, + 441.0, + 513.0, + 425.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 569.0, + 493.0, + 586.0, + 493.0, + 586.0, + 517.0, + 569.0, + 517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 512.0, + 460.0, + 512.0, + 460.0, + 540.0, + 404.0, + 540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 617.0, + 573.0, + 626.0, + 573.0, + 626.0, + 582.0, + 617.0, + 582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 592.0, + 511.0, + 592.0, + 511.0, + 627.0, + 353.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.25, + 284.5, + 627.25, + 284.5, + 627.25, + 299.5, + 541.25, + 299.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 480.5, + 667.0, + 480.5, + 667.0, + 492.5, + 642.0, + 492.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 538.0, + 611.0, + 538.0, + 611.0, + 548.0, + 590.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 658.0, + 594.0, + 658.0, + 594.0, + 692.0, + 295.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 658.0, + 1404.0, + 658.0, + 1404.0, + 692.0, + 672.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 688.0, + 440.0, + 688.0, + 440.0, + 725.0, + 294.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 688.0, + 1368.0, + 688.0, + 1368.0, + 725.0, + 464.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1395.0, + 688.0, + 1407.0, + 688.0, + 1407.0, + 725.0, + 1395.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 720.0, + 1406.0, + 720.0, + 1406.0, + 757.0, + 294.0, + 757.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 747.0, + 1408.0, + 747.0, + 1408.0, + 787.0, + 293.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 779.0, + 958.0, + 779.0, + 958.0, + 818.0, + 293.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1482.0, + 447.0, + 1482.0, + 447.0, + 1531.0, + 291.0, + 1531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2093.0, + 839.0, + 2093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 941.0, + 219.0, + 1315.0, + 219.0, + 1315.0, + 253.0, + 941.0, + 253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 941.0, + 219.0, + 1315.0, + 219.0, + 1315.0, + 253.0, + 941.0, + 253.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2093.0, + 838.0, + 2093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1790.0, + 888.0, + 1790.0, + 888.0, + 1831.0, + 291.0, + 1831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 1790.0, + 1406.0, + 1790.0, + 1406.0, + 1831.0, + 934.0, + 1831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1822.0, + 511.0, + 1822.0, + 511.0, + 1861.0, + 292.0, + 1861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 530.0, + 1822.0, + 1408.0, + 1822.0, + 1408.0, + 1861.0, + 530.0, + 1861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1851.0, + 1406.0, + 1851.0, + 1406.0, + 1891.0, + 291.0, + 1891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1886.0, + 844.0, + 1886.0, + 844.0, + 1920.0, + 295.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 861.0, + 1886.0, + 1406.0, + 1886.0, + 1406.0, + 1920.0, + 861.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1913.0, + 1358.0, + 1913.0, + 1358.0, + 1951.0, + 292.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 1913.0, + 1407.0, + 1913.0, + 1407.0, + 1951.0, + 1404.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1946.0, + 1405.0, + 1946.0, + 1405.0, + 1981.0, + 294.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1975.0, + 940.0, + 1975.0, + 940.0, + 2013.0, + 294.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 1975.0, + 998.0, + 1975.0, + 998.0, + 2013.0, + 985.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1064.0, + 1405.0, + 1064.0, + 1405.0, + 1101.0, + 294.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1094.0, + 815.0, + 1094.0, + 815.0, + 1133.0, + 294.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 879.0, + 1094.0, + 1406.0, + 1094.0, + 1406.0, + 1133.0, + 879.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1125.0, + 1405.0, + 1125.0, + 1405.0, + 1164.0, + 294.0, + 1164.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1156.0, + 389.0, + 1156.0, + 389.0, + 1191.0, + 292.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 429.0, + 1156.0, + 1406.0, + 1156.0, + 1406.0, + 1191.0, + 429.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1184.0, + 1181.0, + 1184.0, + 1181.0, + 1223.0, + 291.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1269.0, + 1184.0, + 1405.0, + 1184.0, + 1405.0, + 1223.0, + 1269.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1217.0, + 1402.0, + 1217.0, + 1402.0, + 1252.0, + 294.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1244.0, + 687.0, + 1244.0, + 687.0, + 1286.0, + 291.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 1244.0, + 898.0, + 1244.0, + 898.0, + 1286.0, + 767.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 953.0, + 1244.0, + 1010.0, + 1244.0, + 1010.0, + 1286.0, + 953.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1124.0, + 1244.0, + 1142.0, + 1244.0, + 1142.0, + 1286.0, + 1124.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1293.0, + 1408.0, + 1293.0, + 1408.0, + 1326.0, + 297.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1322.0, + 1406.0, + 1322.0, + 1406.0, + 1360.0, + 294.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1354.0, + 1405.0, + 1354.0, + 1405.0, + 1388.0, + 296.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1382.0, + 646.0, + 1382.0, + 646.0, + 1420.0, + 294.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 722.0, + 1382.0, + 1406.0, + 1382.0, + 1406.0, + 1420.0, + 722.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1413.0, + 1006.0, + 1413.0, + 1006.0, + 1449.0, + 295.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 867.0, + 1405.0, + 867.0, + 1405.0, + 903.0, + 294.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 899.0, + 679.0, + 899.0, + 679.0, + 931.0, + 295.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 899.0, + 1405.0, + 899.0, + 1405.0, + 931.0, + 767.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 929.0, + 1405.0, + 929.0, + 1405.0, + 965.0, + 295.0, + 965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 957.0, + 859.0, + 957.0, + 859.0, + 995.0, + 294.0, + 995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 957.0, + 1406.0, + 957.0, + 1406.0, + 995.0, + 934.0, + 995.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 989.0, + 358.0, + 989.0, + 358.0, + 1025.0, + 294.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 989.0, + 615.0, + 989.0, + 615.0, + 1025.0, + 437.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 989.0, + 960.0, + 989.0, + 960.0, + 1025.0, + 693.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1037.0, + 989.0, + 1405.0, + 989.0, + 1405.0, + 1025.0, + 1037.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1021.0, + 502.0, + 1021.0, + 502.0, + 1057.0, + 295.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 616.0, + 1021.0, + 1339.0, + 1021.0, + 1339.0, + 1057.0, + 616.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1684.0, + 371.0, + 1684.0, + 371.0, + 1723.0, + 293.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1684.0, + 904.0, + 1684.0, + 904.0, + 1723.0, + 394.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 932.0, + 1684.0, + 982.0, + 1684.0, + 982.0, + 1723.0, + 932.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1010.0, + 1684.0, + 1405.0, + 1684.0, + 1405.0, + 1723.0, + 1010.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1715.0, + 1273.0, + 1715.0, + 1273.0, + 1752.0, + 294.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1715.0, + 1407.0, + 1715.0, + 1407.0, + 1752.0, + 1402.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1747.0, + 343.0, + 1747.0, + 343.0, + 1783.0, + 294.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 1747.0, + 515.0, + 1747.0, + 515.0, + 1783.0, + 475.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 625.0, + 1747.0, + 1317.0, + 1747.0, + 1317.0, + 1783.0, + 625.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1551.0, + 754.0, + 1551.0, + 754.0, + 1590.0, + 293.0, + 1590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 778.0, + 1551.0, + 1274.0, + 1551.0, + 1274.0, + 1590.0, + 778.0, + 1590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1551.0, + 1407.0, + 1551.0, + 1407.0, + 1590.0, + 1402.0, + 1590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1586.0, + 853.0, + 1586.0, + 853.0, + 1622.0, + 293.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1586.0, + 1047.0, + 1586.0, + 1047.0, + 1622.0, + 875.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1065.0, + 1586.0, + 1358.0, + 1586.0, + 1358.0, + 1622.0, + 1065.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 471.0, + 1131.0, + 471.0, + 1131.0, + 499.0, + 870.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1163.0, + 471.0, + 1177.0, + 471.0, + 1177.0, + 499.0, + 1163.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 889.0, + 497.0, + 928.0, + 497.0, + 928.0, + 521.0, + 889.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 969.0, + 497.0, + 1240.0, + 497.0, + 1240.0, + 521.0, + 969.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 520.0, + 1196.0, + 520.0, + 1196.0, + 544.0, + 934.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 545.0, + 1103.0, + 545.0, + 1103.0, + 568.0, + 890.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1125.0, + 545.0, + 1227.0, + 545.0, + 1227.0, + 568.0, + 1125.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 891.0, + 567.0, + 1334.0, + 567.0, + 1334.0, + 593.0, + 891.0, + 593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1362.0, + 567.0, + 1372.0, + 567.0, + 1372.0, + 593.0, + 1362.0, + 593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 594.0, + 1035.0, + 594.0, + 1035.0, + 615.0, + 893.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 864.0, + 260.0, + 1242.0, + 260.0, + 1242.0, + 288.0, + 864.0, + 288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 884.0, + 285.0, + 979.0, + 285.0, + 979.0, + 310.0, + 884.0, + 310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 285.0, + 1026.0, + 285.0, + 1026.0, + 310.0, + 1009.0, + 310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 879.0, + 309.0, + 883.0, + 309.0, + 883.0, + 332.0, + 879.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 309.0, + 1186.0, + 309.0, + 1186.0, + 332.0, + 925.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 331.0, + 883.0, + 331.0, + 883.0, + 357.0, + 878.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 331.0, + 1262.0, + 331.0, + 1262.0, + 357.0, + 925.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 879.0, + 358.0, + 883.0, + 358.0, + 883.0, + 381.0, + 879.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 358.0, + 1010.0, + 358.0, + 1010.0, + 381.0, + 925.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 380.0, + 883.0, + 380.0, + 883.0, + 403.0, + 878.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 380.0, + 1174.0, + 380.0, + 1174.0, + 403.0, + 926.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 880.0, + 404.0, + 883.0, + 404.0, + 883.0, + 427.0, + 880.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 404.0, + 1160.0, + 404.0, + 1160.0, + 427.0, + 925.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1182.0, + 404.0, + 1285.0, + 404.0, + 1285.0, + 427.0, + 1182.0, + 427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 427.0, + 959.0, + 427.0, + 959.0, + 451.0, + 882.0, + 451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1002.0, + 427.0, + 1100.0, + 427.0, + 1100.0, + 451.0, + 1002.0, + 451.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1292, + 1405, + 1292, + 1405, + 1753, + 297, + 1753 + ], + "score": 0.985 + }, + { + "category_id": 1, + "poly": [ + 297, + 202, + 1405, + 202, + 1405, + 387, + 297, + 387 + ], + "score": 0.982 + }, + { + "category_id": 1, + "poly": [ + 297, + 840, + 1403, + 840, + 1403, + 997, + 297, + 997 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 297, + 606, + 1404, + 606, + 1404, + 731, + 297, + 731 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 297, + 1137, + 1403, + 1137, + 1403, + 1261, + 297, + 1261 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 295, + 477, + 1401, + 477, + 1401, + 541, + 295, + 541 + ], + "score": 0.957 + }, + { + "category_id": 1, + "poly": [ + 298, + 1008, + 1397, + 1008, + 1397, + 1071, + 298, + 1071 + ], + "score": 0.952 + }, + { + "category_id": 1, + "poly": [ + 301, + 1863, + 1401, + 1863, + 1401, + 1927, + 301, + 1927 + ], + "score": 0.943 + }, + { + "category_id": 8, + "poly": [ + 716, + 1087, + 983, + 1087, + 983, + 1125, + 716, + 1125 + ], + "score": 0.939 + }, + { + "category_id": 8, + "poly": [ + 726, + 790, + 973, + 790, + 973, + 829, + 726, + 829 + ], + "score": 0.936 + }, + { + "category_id": 8, + "poly": [ + 733, + 553, + 969, + 553, + 969, + 593, + 733, + 593 + ], + "score": 0.934 + }, + { + "category_id": 1, + "poly": [ + 296, + 742, + 1039, + 742, + 1039, + 776, + 296, + 776 + ], + "score": 0.931 + }, + { + "category_id": 2, + "poly": [ + 294, + 1949, + 1403, + 1949, + 1403, + 2007, + 294, + 2007 + ], + "score": 0.931 + }, + { + "category_id": 0, + "poly": [ + 299, + 421, + 595, + 421, + 595, + 455, + 299, + 455 + ], + "score": 0.928 + }, + { + "category_id": 0, + "poly": [ + 299, + 1794, + 638, + 1794, + 638, + 1833, + 299, + 1833 + ], + "score": 0.918 + }, + { + "category_id": 9, + "poly": [ + 1366, + 795, + 1400, + 795, + 1400, + 824, + 1366, + 824 + ], + "score": 0.903 + }, + { + "category_id": 9, + "poly": [ + 1366, + 558, + 1399, + 558, + 1399, + 588, + 1366, + 588 + ], + "score": 0.883 + }, + { + "category_id": 9, + "poly": [ + 1367, + 1091, + 1399, + 1091, + 1399, + 1120, + 1367, + 1120 + ], + "score": 0.879 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.687 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.342 + }, + { + "category_id": 14, + "poly": [ + 716, + 1086, + 985, + 1086, + 985, + 1125, + 716, + 1125 + ], + "score": 0.94, + "latex": "s ( Z ) = Z _ { 1 } \\odot f _ { W , b } ( Z _ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 1208, + 873, + 1330, + 873, + 1330, + 907, + 1208, + 907 + ], + "score": 0.94, + "latex": "s ( Z ) \\approx Z" + }, + { + "category_id": 14, + "poly": [ + 733, + 554, + 968, + 554, + 968, + 593, + 733, + 593 + ], + "score": 0.94, + "latex": "f _ { W , b } ( Z ) = W Z + b" + }, + { + "category_id": 14, + "poly": [ + 726, + 790, + 974, + 790, + 974, + 828, + 726, + 828 + ], + "score": 0.93, + "latex": "s ( Z ) = Z \\odot f _ { W , b } ( Z )" + }, + { + "category_id": 13, + "poly": [ + 888, + 873, + 1042, + 873, + 1042, + 907, + 888, + 907 + ], + "score": 0.92, + "latex": "f _ { W , b } ( Z ) \\approx { \\bf 1 }" + }, + { + "category_id": 13, + "poly": [ + 372, + 607, + 504, + 607, + 504, + 636, + 372, + 636 + ], + "score": 0.92, + "latex": "W \\in \\mathbb { R } ^ { n \\times n }" + }, + { + "category_id": 13, + "poly": [ + 818, + 668, + 892, + 668, + 892, + 702, + 818, + 702 + ], + "score": 0.92, + "latex": "W ( Z )" + }, + { + "category_id": 13, + "poly": [ + 638, + 1140, + 692, + 1140, + 692, + 1173, + 638, + 1173 + ], + "score": 0.91, + "latex": "f _ { W , b }" + }, + { + "category_id": 13, + "poly": [ + 342, + 1689, + 415, + 1689, + 415, + 1725, + 342, + 1725 + ], + "score": 0.91, + "latex": "n ^ { 2 } e / 2" + }, + { + "category_id": 13, + "poly": [ + 661, + 743, + 706, + 743, + 706, + 777, + 661, + 777 + ], + "score": 0.91, + "latex": "s ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 989, + 1010, + 1084, + 1010, + 1084, + 1041, + 989, + 1041 + ], + "score": 0.9, + "latex": "( Z _ { 1 } , Z _ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 983, + 479, + 1027, + 479, + 1027, + 512, + 983, + 512 + ], + "score": 0.9, + "latex": "s ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 892, + 1691, + 954, + 1691, + 954, + 1720, + 892, + 1720 + ], + "score": 0.9, + "latex": "2 n ^ { 2 } d" + }, + { + "category_id": 13, + "poly": [ + 328, + 1229, + 372, + 1229, + 372, + 1263, + 328, + 1263 + ], + "score": 0.9, + "latex": "s ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 914, + 1659, + 998, + 1659, + 998, + 1691, + 914, + 1691 + ], + "score": 0.9, + "latex": "q _ { i } k _ { j } v _ { k } )" + }, + { + "category_id": 13, + "poly": [ + 382, + 669, + 490, + 669, + 490, + 698, + 382, + 698 + ], + "score": 0.89, + "latex": "1 2 8 \\times 1 2 8" + }, + { + "category_id": 13, + "poly": [ + 796, + 1630, + 847, + 1630, + 847, + 1662, + 796, + 1662 + ], + "score": 0.84, + "latex": "z _ { i } z _ { j }" + }, + { + "category_id": 13, + "poly": [ + 668, + 1011, + 691, + 1011, + 691, + 1037, + 668, + 1037 + ], + "score": 0.82, + "latex": "Z" + }, + { + "category_id": 13, + "poly": [ + 372, + 846, + 398, + 846, + 398, + 872, + 372, + 872 + ], + "score": 0.81, + "latex": "\\odot" + }, + { + "category_id": 13, + "poly": [ + 1125, + 1952, + 1141, + 1952, + 1141, + 1975, + 1125, + 1975 + ], + "score": 0.78, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 1248, + 670, + 1271, + 670, + 1271, + 697, + 1248, + 697 + ], + "score": 0.78, + "latex": "Z" + }, + { + "category_id": 13, + "poly": [ + 611, + 875, + 627, + 875, + 627, + 901, + 611, + 901 + ], + "score": 0.76, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 1315, + 609, + 1331, + 609, + 1331, + 635, + 1315, + 635 + ], + "score": 0.76, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 1237, + 614, + 1256, + 614, + 1256, + 636, + 1237, + 636 + ], + "score": 0.76, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 1368, + 638, + 1401, + 638, + 1401, + 665, + 1368, + 665 + ], + "score": 0.73, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 1107, + 1727, + 1127, + 1727, + 1127, + 1749, + 1107, + 1749 + ], + "score": 0.68, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 297, + 874, + 331, + 874, + 331, + 902, + 297, + 902 + ], + "score": 0.57, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 495, + 699, + 529, + 699, + 529, + 727, + 495, + 727 + ], + "score": 0.46, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 507, + 1865, + 585, + 1865, + 585, + 1897, + 507, + 1897 + ], + "score": 0.41, + "latex": "\\mathrm { g M L P }" + }, + { + "category_id": 13, + "poly": [ + 570, + 1956, + 586, + 1956, + 586, + 1975, + 570, + 1975 + ], + "score": 0.39, + "latex": "e" + }, + { + "category_id": 15, + "poly": [ + 331.0, + 1945.0, + 569.0, + 1945.0, + 569.0, + 1983.0, + 331.0, + 1983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 587.0, + 1945.0, + 1124.0, + 1945.0, + 1124.0, + 1983.0, + 587.0, + 1983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1142.0, + 1945.0, + 1405.0, + 1945.0, + 1405.0, + 1983.0, + 1142.0, + 1983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1976.0, + 1040.0, + 1976.0, + 1040.0, + 2010.0, + 296.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 418.0, + 599.0, + 418.0, + 599.0, + 461.0, + 293.0, + 461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1791.0, + 641.0, + 1791.0, + 641.0, + 1842.0, + 290.0, + 1842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1293.0, + 1405.0, + 1293.0, + 1405.0, + 1328.0, + 296.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1322.0, + 1408.0, + 1322.0, + 1408.0, + 1360.0, + 294.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1355.0, + 1405.0, + 1355.0, + 1405.0, + 1390.0, + 294.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1386.0, + 1403.0, + 1386.0, + 1403.0, + 1417.0, + 295.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1415.0, + 1405.0, + 1415.0, + 1405.0, + 1450.0, + 295.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1444.0, + 1405.0, + 1444.0, + 1405.0, + 1482.0, + 292.0, + 1482.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1477.0, + 1403.0, + 1477.0, + 1403.0, + 1509.0, + 296.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1506.0, + 1403.0, + 1506.0, + 1403.0, + 1541.0, + 295.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1536.0, + 1405.0, + 1536.0, + 1405.0, + 1571.0, + 295.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1568.0, + 1405.0, + 1568.0, + 1405.0, + 1599.0, + 295.0, + 1599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1598.0, + 1406.0, + 1598.0, + 1406.0, + 1633.0, + 295.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1625.0, + 795.0, + 1625.0, + 795.0, + 1665.0, + 292.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1625.0, + 1407.0, + 1625.0, + 1407.0, + 1665.0, + 848.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1655.0, + 913.0, + 1655.0, + 913.0, + 1694.0, + 291.0, + 1694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 1655.0, + 1406.0, + 1655.0, + 1406.0, + 1694.0, + 999.0, + 1694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1690.0, + 341.0, + 1690.0, + 341.0, + 1725.0, + 295.0, + 1725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 416.0, + 1690.0, + 891.0, + 1690.0, + 891.0, + 1725.0, + 416.0, + 1725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 955.0, + 1690.0, + 1405.0, + 1690.0, + 1405.0, + 1725.0, + 955.0, + 1725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1720.0, + 1106.0, + 1720.0, + 1106.0, + 1758.0, + 294.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1128.0, + 1720.0, + 1137.0, + 1720.0, + 1137.0, + 1758.0, + 1128.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 203.0, + 1408.0, + 203.0, + 1408.0, + 239.0, + 295.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 233.0, + 1407.0, + 233.0, + 1407.0, + 269.0, + 294.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 261.0, + 1406.0, + 261.0, + 1406.0, + 298.0, + 294.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 296.0, + 1403.0, + 296.0, + 1403.0, + 328.0, + 295.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 326.0, + 1403.0, + 326.0, + 1403.0, + 358.0, + 295.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 357.0, + 1225.0, + 357.0, + 1225.0, + 389.0, + 296.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 842.0, + 371.0, + 842.0, + 371.0, + 876.0, + 296.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 842.0, + 1405.0, + 842.0, + 1405.0, + 876.0, + 399.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 872.0, + 610.0, + 872.0, + 610.0, + 909.0, + 332.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 872.0, + 887.0, + 872.0, + 887.0, + 909.0, + 628.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 872.0, + 1207.0, + 872.0, + 1207.0, + 909.0, + 1043.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1331.0, + 872.0, + 1405.0, + 872.0, + 1405.0, + 909.0, + 1331.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 904.0, + 1405.0, + 904.0, + 1405.0, + 937.0, + 295.0, + 937.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 931.0, + 1405.0, + 931.0, + 1405.0, + 971.0, + 294.0, + 971.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 961.0, + 1003.0, + 961.0, + 1003.0, + 1004.0, + 292.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 603.0, + 371.0, + 603.0, + 371.0, + 643.0, + 292.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 603.0, + 1236.0, + 603.0, + 1236.0, + 643.0, + 505.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1257.0, + 603.0, + 1314.0, + 603.0, + 1314.0, + 643.0, + 1257.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1332.0, + 603.0, + 1405.0, + 603.0, + 1405.0, + 643.0, + 1332.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 637.0, + 1367.0, + 637.0, + 1367.0, + 673.0, + 294.0, + 673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 670.0, + 381.0, + 670.0, + 381.0, + 702.0, + 295.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 670.0, + 817.0, + 670.0, + 817.0, + 702.0, + 491.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 893.0, + 670.0, + 1247.0, + 670.0, + 1247.0, + 702.0, + 893.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1272.0, + 670.0, + 1402.0, + 670.0, + 1402.0, + 702.0, + 1272.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 699.0, + 494.0, + 699.0, + 494.0, + 734.0, + 292.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 530.0, + 699.0, + 1277.0, + 699.0, + 1277.0, + 734.0, + 530.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1134.0, + 637.0, + 1134.0, + 637.0, + 1176.0, + 291.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 1134.0, + 1408.0, + 1134.0, + 1408.0, + 1176.0, + 693.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1169.0, + 1405.0, + 1169.0, + 1405.0, + 1204.0, + 292.0, + 1204.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1199.0, + 1405.0, + 1199.0, + 1405.0, + 1235.0, + 292.0, + 1235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1229.0, + 327.0, + 1229.0, + 327.0, + 1265.0, + 293.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 373.0, + 1229.0, + 1290.0, + 1229.0, + 1290.0, + 1265.0, + 373.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 477.0, + 982.0, + 477.0, + 982.0, + 512.0, + 295.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1028.0, + 477.0, + 1402.0, + 477.0, + 1402.0, + 512.0, + 1028.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 508.0, + 1171.0, + 508.0, + 1171.0, + 544.0, + 295.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1005.0, + 667.0, + 1005.0, + 667.0, + 1046.0, + 293.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 1005.0, + 988.0, + 1005.0, + 988.0, + 1046.0, + 692.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1085.0, + 1005.0, + 1403.0, + 1005.0, + 1403.0, + 1046.0, + 1085.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1037.0, + 929.0, + 1037.0, + 929.0, + 1075.0, + 294.0, + 1075.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1862.0, + 506.0, + 1862.0, + 506.0, + 1901.0, + 296.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 586.0, + 1862.0, + 1406.0, + 1862.0, + 1406.0, + 1901.0, + 586.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1896.0, + 1404.0, + 1896.0, + 1404.0, + 1928.0, + 298.0, + 1928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 734.0, + 660.0, + 734.0, + 660.0, + 785.0, + 291.0, + 785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 734.0, + 1042.0, + 734.0, + 1042.0, + 785.0, + 707.0, + 785.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 403, + 1077, + 1289, + 1077, + 1289, + 1755, + 403, + 1755 + ], + "score": 0.984, + "html": "
ModelImageNet Top-1 (%)*Input ResolutionParams (M)MAdds (B)
ConvNets
ResNet-152 [16]78.32246011.3
RegNetY-8GF[39]81.7224398.0
EfficientNet-B0 [17]77.122450.39
EfficientNet-B3[17]81.6300121.8
EfficientNet-B7 [17]84.36006637.0
NFNet-F0 [33]83.61927212.4
Transformers
ViT-B/16 [7]77.93848655.4
ViT-L/16 [7]76.5384307190.7
DeiT-Ti [8] (ViT+reg)72.222451.3
DeiT-S [8](ViT+reg)79.8224224.6
DeiT-B [8] (ViT+reg)81.82248617.5
MLP-like†
Mixer-B/16 [20]76.42245912.7
Mixer-B/16 (our setup)77.32245912.7
Mixer-L/16 [20]71.822420744.8
ResMLP-12 [22]76.6224153.0
ResMLP-24 [22]79.4224306.0
ResMLP-36[22]79.7224458.9
gMLP-Ti (ours)72.322461.4
gMLP-S (ours)79.6224204.5
gMLP-B (ours)81.62247315.8
" + }, + { + "category_id": 1, + "poly": [ + 297, + 278, + 1405, + 278, + 1405, + 492, + 297, + 492 + ], + "score": 0.979 + }, + { + "category_id": 5, + "poly": [ + 439, + 564, + 1255, + 564, + 1255, + 710, + 439, + 710 + ], + "score": 0.974, + "html": "
#LdmodeldffnParams (M)FLOPs (B)Survival Prob
gMLP-Ti301287685.92.71.00
gMLP-S30256153619.58.90.95
gMLP-B30512307273.431.60.80
" + }, + { + "category_id": 1, + "poly": [ + 297, + 730, + 1405, + 730, + 1405, + 1008, + 297, + 1008 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 296, + 202, + 1401, + 202, + 1401, + 265, + 296, + 265 + ], + "score": 0.954 + }, + { + "category_id": 7, + "poly": [ + 421, + 1759, + 1086, + 1759, + 1086, + 1808, + 421, + 1808 + ], + "score": 0.934 + }, + { + "category_id": 2, + "poly": [ + 329, + 1947, + 1118, + 1947, + 1118, + 2008, + 329, + 2008 + ], + "score": 0.911 + }, + { + "category_id": 6, + "poly": [ + 580, + 1030, + 1122, + 1030, + 1122, + 1062, + 580, + 1062 + ], + "score": 0.909 + }, + { + "category_id": 6, + "poly": [ + 491, + 515, + 1206, + 515, + 1206, + 548, + 491, + 548 + ], + "score": 0.909 + }, + { + "category_id": 2, + "poly": [ + 840, + 2061, + 858, + 2061, + 858, + 2084, + 840, + 2084 + ], + "score": 0.792 + }, + { + "category_id": 1, + "poly": [ + 298, + 1834, + 1405, + 1834, + 1405, + 1928, + 298, + 1928 + ], + "score": 0.71 + }, + { + "category_id": 6, + "poly": [ + 298, + 1834, + 1405, + 1834, + 1405, + 1928, + 298, + 1928 + ], + "score": 0.206 + }, + { + "category_id": 13, + "poly": [ + 1101, + 310, + 1182, + 310, + 1182, + 339, + 1101, + 339 + ], + "score": 0.87, + "latex": "1 6 \\times 1 6" + }, + { + "category_id": 13, + "poly": [ + 363, + 944, + 447, + 944, + 447, + 976, + 363, + 976 + ], + "score": 0.57, + "latex": "{ \\mathrm { g M L P s } }" + }, + { + "category_id": 13, + "poly": [ + 923, + 516, + 999, + 516, + 999, + 548, + 923, + 548 + ], + "score": 0.45, + "latex": "\\mathrm { g M L P }" + }, + { + "category_id": 15, + "poly": [ + 423.0, + 1755.0, + 987.0, + 1755.0, + 987.0, + 1785.0, + 423.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 422.0, + 1775.0, + 1090.0, + 1775.0, + 1090.0, + 1812.0, + 422.0, + 1812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1941.0, + 1047.0, + 1941.0, + 1047.0, + 1982.0, + 329.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 329.0, + 1970.0, + 1121.0, + 1970.0, + 1121.0, + 2013.0, + 329.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1027.0, + 1124.0, + 1027.0, + 1124.0, + 1065.0, + 575.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 514.0, + 922.0, + 514.0, + 922.0, + 550.0, + 491.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1000.0, + 514.0, + 1207.0, + 514.0, + 1207.0, + 550.0, + 1000.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 863.0, + 2058.0, + 863.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1834.0, + 1406.0, + 1834.0, + 1406.0, + 1871.0, + 295.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1866.0, + 1403.0, + 1866.0, + 1403.0, + 1900.0, + 295.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1897.0, + 1370.0, + 1897.0, + 1370.0, + 1931.0, + 296.0, + 1931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 276.0, + 1405.0, + 276.0, + 1405.0, + 315.0, + 292.0, + 315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 306.0, + 1100.0, + 306.0, + 1100.0, + 345.0, + 294.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 306.0, + 1408.0, + 306.0, + 1408.0, + 345.0, + 1183.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 337.0, + 1405.0, + 337.0, + 1405.0, + 374.0, + 294.0, + 374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 368.0, + 1407.0, + 368.0, + 1407.0, + 405.0, + 294.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 397.0, + 1406.0, + 397.0, + 1406.0, + 435.0, + 294.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 430.0, + 1407.0, + 430.0, + 1407.0, + 465.0, + 295.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 461.0, + 1265.0, + 461.0, + 1265.0, + 494.0, + 294.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 727.0, + 1406.0, + 727.0, + 1406.0, + 767.0, + 295.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 760.0, + 1406.0, + 760.0, + 1406.0, + 797.0, + 294.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 793.0, + 1405.0, + 793.0, + 1405.0, + 826.0, + 295.0, + 826.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 816.0, + 1406.0, + 816.0, + 1406.0, + 861.0, + 292.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 851.0, + 1405.0, + 851.0, + 1405.0, + 887.0, + 294.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 881.0, + 1406.0, + 881.0, + 1406.0, + 918.0, + 292.0, + 918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 914.0, + 1405.0, + 914.0, + 1405.0, + 946.0, + 296.0, + 946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 939.0, + 362.0, + 939.0, + 362.0, + 982.0, + 292.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 939.0, + 1407.0, + 939.0, + 1407.0, + 982.0, + 448.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 973.0, + 1148.0, + 973.0, + 1148.0, + 1009.0, + 296.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 204.0, + 1402.0, + 204.0, + 1402.0, + 236.0, + 297.0, + 236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 234.0, + 1214.0, + 234.0, + 1214.0, + 267.0, + 294.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1834.0, + 1406.0, + 1834.0, + 1406.0, + 1871.0, + 295.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1866.0, + 1403.0, + 1866.0, + 1403.0, + 1900.0, + 295.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1897.0, + 1370.0, + 1897.0, + 1370.0, + 1931.0, + 296.0, + 1931.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 1116, + 1404, + 1116, + 1404, + 1360, + 298, + 1360 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 297, + 888, + 1404, + 888, + 1404, + 1103, + 297, + 1103 + ], + "score": 0.976 + }, + { + "category_id": 3, + "poly": [ + 872, + 199, + 1399, + 199, + 1399, + 646, + 872, + 646 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 299, + 1854, + 1403, + 1854, + 1403, + 2008, + 299, + 2008 + ], + "score": 0.971 + }, + { + "category_id": 3, + "poly": [ + 311, + 196, + 846, + 196, + 846, + 708, + 311, + 708 + ], + "score": 0.969 + }, + { + "category_id": 4, + "poly": [ + 873, + 671, + 1405, + 671, + 1405, + 764, + 873, + 764 + ], + "score": 0.957 + }, + { + "category_id": 1, + "poly": [ + 360, + 1512, + 1406, + 1512, + 1406, + 1828, + 360, + 1828 + ], + "score": 0.955 + }, + { + "category_id": 0, + "poly": [ + 298, + 817, + 923, + 817, + 923, + 857, + 298, + 857 + ], + "score": 0.929 + }, + { + "category_id": 0, + "poly": [ + 297, + 1397, + 1190, + 1397, + 1190, + 1432, + 297, + 1432 + ], + "score": 0.921 + }, + { + "category_id": 4, + "poly": [ + 312, + 732, + 845, + 732, + 845, + 764, + 312, + 764 + ], + "score": 0.916 + }, + { + "category_id": 1, + "poly": [ + 296, + 1454, + 1185, + 1454, + 1185, + 1487, + 296, + 1487 + ], + "score": 0.905 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 858, + 2061, + 858, + 2085, + 841, + 2085 + ], + "score": 0.715 + }, + { + "category_id": 13, + "poly": [ + 928, + 1207, + 1011, + 1207, + 1011, + 1241, + 928, + 1241 + ], + "score": 0.92, + "latex": "f _ { W , b } ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 1139, + 1299, + 1173, + 1299, + 1173, + 1327, + 1139, + 1327 + ], + "score": 0.47, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 674, + 1977, + 745, + 1977, + 745, + 2005, + 674, + 2005 + ], + "score": 0.46, + "latex": "{ = } 4 . 2 6 " + }, + { + "category_id": 13, + "poly": [ + 298, + 1299, + 331, + 1299, + 331, + 1327, + 298, + 1327 + ], + "score": 0.42, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 985, + 1976, + 1055, + 1976, + 1055, + 2004, + 985, + 2004 + ], + "score": 0.42, + "latex": "{ = } 4 . 3 5" + }, + { + "category_id": 13, + "poly": [ + 534, + 1269, + 567, + 1269, + 567, + 1297, + 534, + 1297 + ], + "score": 0.36, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 895, + 1855, + 979, + 1855, + 979, + 1887, + 895, + 1887 + ], + "score": 0.33, + "latex": "{ \\mathrm { g M L P s } }" + }, + { + "category_id": 13, + "poly": [ + 760, + 1148, + 844, + 1148, + 844, + 1178, + 760, + 1178 + ], + "score": 0.26, + "latex": "{ \\mathrm { g M L P s } }" + }, + { + "category_id": 15, + "poly": [ + 948.0, + 263.0, + 1398.0, + 263.0, + 1398.0, + 532.0, + 948.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 283.0, + 932.0, + 283.0, + 932.0, + 303.0, + 894.0, + 303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1159.0, + 294.0, + 1169.0, + 294.0, + 1169.0, + 303.0, + 1159.0, + 303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 873.0, + 343.0, + 942.0, + 343.0, + 942.0, + 369.0, + 873.0, + 369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 345.0, + 1106.0, + 345.0, + 1106.0, + 352.0, + 1090.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 876.0, + 406.0, + 943.0, + 406.0, + 943.0, + 433.0, + 876.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1145.0, + 430.0, + 1155.0, + 430.0, + 1155.0, + 437.0, + 1145.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1228.0, + 429.0, + 1235.0, + 429.0, + 1235.0, + 437.0, + 1228.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 916.0, + 475.0, + 939.0, + 475.0, + 939.0, + 490.0, + 916.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 480.0, + 919.0, + 480.0, + 919.0, + 487.0, + 902.0, + 487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 934.0, + 507.0, + 1405.0, + 507.0, + 1405.0, + 658.0, + 934.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 541.0, + 934.0, + 541.0, + 934.0, + 553.0, + 921.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 748.0, + 197.0, + 824.0, + 197.0, + 824.0, + 228.0, + 748.0, + 228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 225.0, + 372.0, + 225.0, + 372.0, + 257.0, + 334.0, + 257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 623.0, + 228.0, + 708.0, + 228.0, + 708.0, + 259.0, + 623.0, + 259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 663.0, + 269.0, + 678.0, + 269.0, + 678.0, + 280.0, + 663.0, + 280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 515.0, + 277.0, + 587.0, + 277.0, + 587.0, + 307.0, + 515.0, + 307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 296.0, + 615.0, + 296.0, + 615.0, + 330.0, + 513.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 301.0, + 371.0, + 301.0, + 371.0, + 334.0, + 334.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 304.0, + 345.0, + 304.0, + 345.0, + 544.0, + 307.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 428.0, + 306.0, + 511.0, + 306.0, + 511.0, + 335.0, + 428.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 381.0, + 371.0, + 381.0, + 371.0, + 411.0, + 334.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 612.0, + 394.0, + 827.0, + 394.0, + 827.0, + 423.0, + 612.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 609.0, + 427.0, + 723.0, + 427.0, + 723.0, + 459.0, + 609.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 334.0, + 458.0, + 372.0, + 458.0, + 372.0, + 489.0, + 334.0, + 489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 483.0, + 452.0, + 597.0, + 452.0, + 597.0, + 478.0, + 483.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 333.0, + 533.0, + 374.0, + 533.0, + 374.0, + 568.0, + 333.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 635.0, + 529.0, + 833.0, + 529.0, + 833.0, + 567.0, + 635.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 454.0, + 570.0, + 532.0, + 570.0, + 532.0, + 603.0, + 454.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 651.0, + 570.0, + 665.0, + 570.0, + 665.0, + 583.0, + 651.0, + 583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 560.0, + 750.0, + 560.0, + 750.0, + 591.0, + 683.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 590.0, + 554.0, + 590.0, + 554.0, + 628.0, + 455.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 585.0, + 767.0, + 585.0, + 767.0, + 617.0, + 680.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 333.0, + 611.0, + 374.0, + 611.0, + 374.0, + 646.0, + 333.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 683.0, + 613.0, + 805.0, + 613.0, + 805.0, + 646.0, + 683.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 658.0, + 449.0, + 658.0, + 449.0, + 687.0, + 426.0, + 687.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 655.0, + 531.0, + 655.0, + 531.0, + 690.0, + 489.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 655.0, + 605.0, + 655.0, + 605.0, + 690.0, + 563.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 638.0, + 656.0, + 678.0, + 656.0, + 678.0, + 688.0, + 638.0, + 688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 654.0, + 751.0, + 654.0, + 751.0, + 690.0, + 710.0, + 690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 654.0, + 832.0, + 654.0, + 832.0, + 691.0, + 777.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 548.0, + 678.0, + 669.0, + 678.0, + 669.0, + 714.0, + 548.0, + 714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.25, + 208.5, + 860.25, + 208.5, + 860.25, + 263.5, + 716.25, + 263.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.75, + 272.5, + 668.75, + 272.5, + 668.75, + 286.5, + 636.75, + 286.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 530.25, + 329.0, + 711.25, + 329.0, + 711.25, + 371.5, + 530.25, + 371.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 390.5, + 516.0, + 390.5, + 516.0, + 413.5, + 508.0, + 413.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 578.5, + 472.0, + 578.5, + 472.0, + 635.0, + 376.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 668.0, + 1409.0, + 668.0, + 1409.0, + 705.0, + 870.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 698.0, + 1407.0, + 698.0, + 1407.0, + 736.0, + 872.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 872.0, + 729.0, + 1376.0, + 729.0, + 1376.0, + 768.0, + 872.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 816.0, + 927.0, + 816.0, + 927.0, + 864.0, + 291.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1394.0, + 1197.0, + 1394.0, + 1197.0, + 1439.0, + 292.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 723.0, + 852.0, + 723.0, + 852.0, + 773.0, + 306.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2058.0, + 861.0, + 2058.0, + 861.0, + 2093.0, + 839.0, + 2093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1115.0, + 1405.0, + 1115.0, + 1405.0, + 1152.0, + 294.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1147.0, + 759.0, + 1147.0, + 759.0, + 1179.0, + 293.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 1147.0, + 1405.0, + 1147.0, + 1405.0, + 1179.0, + 845.0, + 1179.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1178.0, + 1405.0, + 1178.0, + 1405.0, + 1212.0, + 294.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1207.0, + 927.0, + 1207.0, + 927.0, + 1244.0, + 293.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 1207.0, + 1405.0, + 1207.0, + 1405.0, + 1244.0, + 1012.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1238.0, + 1404.0, + 1238.0, + 1404.0, + 1271.0, + 294.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1267.0, + 533.0, + 1267.0, + 533.0, + 1303.0, + 293.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1267.0, + 1405.0, + 1267.0, + 1405.0, + 1303.0, + 568.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 332.0, + 1299.0, + 1138.0, + 1299.0, + 1138.0, + 1332.0, + 332.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1174.0, + 1299.0, + 1404.0, + 1299.0, + 1404.0, + 1332.0, + 1174.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1330.0, + 1146.0, + 1330.0, + 1146.0, + 1363.0, + 294.0, + 1363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 890.0, + 1405.0, + 890.0, + 1405.0, + 924.0, + 295.0, + 924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 919.0, + 1405.0, + 919.0, + 1405.0, + 954.0, + 294.0, + 954.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 950.0, + 1404.0, + 950.0, + 1404.0, + 984.0, + 295.0, + 984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 981.0, + 1406.0, + 981.0, + 1406.0, + 1015.0, + 294.0, + 1015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1010.0, + 1404.0, + 1010.0, + 1404.0, + 1043.0, + 292.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1042.0, + 1404.0, + 1042.0, + 1404.0, + 1073.0, + 295.0, + 1073.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1072.0, + 1195.0, + 1072.0, + 1195.0, + 1106.0, + 295.0, + 1106.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1854.0, + 894.0, + 1854.0, + 894.0, + 1887.0, + 297.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 980.0, + 1854.0, + 1405.0, + 1854.0, + 1405.0, + 1887.0, + 980.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1885.0, + 1405.0, + 1885.0, + 1405.0, + 1918.0, + 295.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1912.0, + 1407.0, + 1912.0, + 1407.0, + 1951.0, + 293.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1944.0, + 1405.0, + 1944.0, + 1405.0, + 1981.0, + 295.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1975.0, + 673.0, + 1975.0, + 673.0, + 2012.0, + 295.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 746.0, + 1975.0, + 984.0, + 1975.0, + 984.0, + 2012.0, + 746.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 1975.0, + 1405.0, + 1975.0, + 1405.0, + 2012.0, + 1056.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1512.0, + 1326.0, + 1512.0, + 1326.0, + 1548.0, + 361.0, + 1548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1557.0, + 1408.0, + 1557.0, + 1408.0, + 1593.0, + 359.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1586.0, + 1303.0, + 1586.0, + 1303.0, + 1622.0, + 395.0, + 1622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 1630.0, + 1403.0, + 1630.0, + 1403.0, + 1665.0, + 359.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 1664.0, + 1402.0, + 1664.0, + 1402.0, + 1694.0, + 397.0, + 1694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1694.0, + 1283.0, + 1694.0, + 1283.0, + 1727.0, + 396.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1734.0, + 1403.0, + 1734.0, + 1403.0, + 1768.0, + 358.0, + 1768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1766.0, + 1406.0, + 1766.0, + 1406.0, + 1802.0, + 392.0, + 1802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1795.0, + 1304.0, + 1795.0, + 1304.0, + 1832.0, + 392.0, + 1832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1454.0, + 1186.0, + 1454.0, + 1186.0, + 1489.0, + 295.0, + 1489.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1243, + 1405, + 1243, + 1405, + 1427, + 297, + 1427 + ], + "score": 0.982 + }, + { + "category_id": 5, + "poly": [ + 477, + 1530, + 1215, + 1530, + 1215, + 1855, + 477, + 1855 + ], + "score": 0.98, + "html": "
Model#LParams (M) PerplexitySST-2 MNLI-m
TransformergMLP6+61867594.915.2590.4 81.591.2 77.7
TransformergMLP12+12361101024.264.3591.3 83.392.3 80.9
TransformergMLP24+24721951873.833.7992.1 85.293.5 82.8
TransformergMLP48+481443653573.473.4392.8 86.395.1 84.6
" + }, + { + "category_id": 5, + "poly": [ + 379, + 334, + 1315, + 334, + 1315, + 646, + 379, + 646 + ], + "score": 0.979, + "html": "
ModelPerplexity*Params (M)
BERTbase4.37110
BERTbase + rel pos4.26110
BERTbase + rel pos - attn5.6496
MLP-Mixer5.34112
Linear gMLP,s(Z)= f(Z)5.1492
Additive gMLP,s(Z)= Z+ f(Z)4.9792
Multiplicative gMLP,s(Z) = Z f(Z)4.5392
Multiplicative,Split gMLP,s(Z) = Z1 f(Z2),Z= Zi||Z24.35102
" + }, + { + "category_id": 3, + "poly": [ + 450, + 779, + 1251, + 779, + 1251, + 975, + 450, + 975 + ], + "score": 0.962 + }, + { + "category_id": 6, + "poly": [ + 296, + 193, + 1407, + 193, + 1407, + 319, + 296, + 319 + ], + "score": 0.955 + }, + { + "category_id": 0, + "poly": [ + 298, + 1186, + 1073, + 1186, + 1073, + 1221, + 298, + 1221 + ], + "score": 0.932 + }, + { + "category_id": 1, + "poly": [ + 303, + 690, + 1400, + 690, + 1400, + 753, + 303, + 753 + ], + "score": 0.925 + }, + { + "category_id": 2, + "poly": [ + 297, + 1949, + 1403, + 1949, + 1403, + 2008, + 297, + 2008 + ], + "score": 0.892 + }, + { + "category_id": 6, + "poly": [ + 298, + 1449, + 1395, + 1449, + 1395, + 1513, + 298, + 1513 + ], + "score": 0.867 + }, + { + "category_id": 7, + "poly": [ + 398, + 652, + 1102, + 652, + 1102, + 681, + 398, + 681 + ], + "score": 0.851 + }, + { + "category_id": 4, + "poly": [ + 295, + 1001, + 1406, + 1001, + 1406, + 1156, + 295, + 1156 + ], + "score": 0.722 + }, + { + "category_id": 2, + "poly": [ + 839, + 2061, + 859, + 2061, + 859, + 2086, + 839, + 2086 + ], + "score": 0.621 + }, + { + "category_id": 1, + "poly": [ + 300, + 1868, + 1403, + 1868, + 1403, + 1931, + 300, + 1931 + ], + "score": 0.521 + }, + { + "category_id": 7, + "poly": [ + 300, + 1868, + 1403, + 1868, + 1403, + 1931, + 300, + 1931 + ], + "score": 0.491 + }, + { + "category_id": 1, + "poly": [ + 295, + 1001, + 1406, + 1001, + 1406, + 1156, + 295, + 1156 + ], + "score": 0.451 + }, + { + "category_id": 2, + "poly": [ + 839, + 2062, + 859, + 2062, + 859, + 2086, + 839, + 2086 + ], + "score": 0.321 + }, + { + "category_id": 13, + "poly": [ + 1188, + 1274, + 1353, + 1274, + 1353, + 1308, + 1188, + 1308 + ], + "score": 0.92, + "latex": "\\{ 0 . 5 , 1 , 2 , 4 \\} \\times" + }, + { + "category_id": 13, + "poly": [ + 950, + 256, + 1065, + 256, + 1065, + 286, + 950, + 286 + ], + "score": 0.91, + "latex": "d _ { \\mathrm { f f n } } { = } 3 0 7 2" + }, + { + "category_id": 13, + "poly": [ + 427, + 286, + 562, + 286, + 562, + 317, + 427, + 317 + ], + "score": 0.9, + "latex": "d _ { \\mathrm { m o d e l } } = 5 1 2" + }, + { + "category_id": 13, + "poly": [ + 614, + 256, + 748, + 256, + 748, + 286, + 614, + 286 + ], + "score": 0.89, + "latex": "d _ { \\mathrm { m o d e l } } = 7 6 8" + }, + { + "category_id": 13, + "poly": [ + 759, + 256, + 900, + 256, + 900, + 288, + 759, + 288 + ], + "score": 0.89, + "latex": "d _ { \\mathrm { s p a t i a l } } { = } 3 8 4" + }, + { + "category_id": 13, + "poly": [ + 612, + 287, + 742, + 287, + 742, + 317, + 612, + 317 + ], + "score": 0.88, + "latex": "d _ { \\mathrm { f f n } } = 3 0 7 2" + }, + { + "category_id": 13, + "poly": [ + 296, + 1396, + 384, + 1396, + 384, + 1425, + 296, + 1425 + ], + "score": 0.87, + "latex": "1 2 + 1 2" + }, + { + "category_id": 13, + "poly": [ + 529, + 527, + 671, + 527, + 671, + 555, + 529, + 555 + ], + "score": 0.85, + "latex": "s ( Z ) = f ( Z )" + }, + { + "category_id": 13, + "poly": [ + 1312, + 196, + 1332, + 196, + 1332, + 226, + 1312, + 226 + ], + "score": 0.84, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 551, + 554, + 742, + 554, + 742, + 582, + 551, + 582 + ], + "score": 0.83, + "latex": "s ( Z ) = Z + f ( Z )" + }, + { + "category_id": 13, + "poly": [ + 416, + 256, + 482, + 256, + 482, + 284, + 416, + 284 + ], + "score": 0.81, + "latex": "_ { \\mathrm { L } = 2 4 }" + }, + { + "category_id": 13, + "poly": [ + 1336, + 256, + 1403, + 256, + 1403, + 285, + 1336, + 285 + ], + "score": 0.78, + "latex": "_ { \\mathrm { L } = 3 6 }" + }, + { + "category_id": 13, + "poly": [ + 605, + 582, + 797, + 582, + 797, + 610, + 605, + 610 + ], + "score": 0.77, + "latex": "s ( Z ) = Z \\odot f ( Z )" + }, + { + "category_id": 13, + "poly": [ + 512, + 722, + 597, + 722, + 597, + 754, + 512, + 754 + ], + "score": 0.6, + "latex": "{ \\mathrm { g M L P s } }" + }, + { + "category_id": 13, + "poly": [ + 802, + 1003, + 877, + 1003, + 877, + 1035, + 802, + 1035 + ], + "score": 0.5, + "latex": "\\mathrm { g M L P }" + }, + { + "category_id": 13, + "poly": [ + 582, + 1033, + 616, + 1033, + 616, + 1061, + 582, + 1061 + ], + "score": 0.5, + "latex": "W" + }, + { + "category_id": 13, + "poly": [ + 664, + 609, + 1012, + 609, + 1012, + 640, + 664, + 640 + ], + "score": 0.47, + "latex": "s ( Z ) = Z _ { 1 } \\odot f ( Z _ { 2 } ) , Z = Z _ { 1 } \\| Z _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1217, + 195, + 1299, + 195, + 1299, + 226, + 1217, + 226 + ], + "score": 0.42, + "latex": "{ \\mathrm { g M L P s } }" + }, + { + "category_id": 13, + "poly": [ + 1298, + 1037, + 1316, + 1037, + 1316, + 1061, + 1298, + 1061 + ], + "score": 0.41, + "latex": "\\mathbf { X }" + }, + { + "category_id": 13, + "poly": [ + 707, + 1188, + 788, + 1188, + 788, + 1221, + 707, + 1221 + ], + "score": 0.29, + "latex": "\\mathbf { g } \\mathbf { M L P }" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 191.0, + 1216.0, + 191.0, + 1216.0, + 229.0, + 291.0, + 229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 191.0, + 1311.0, + 191.0, + 1311.0, + 229.0, + 1300.0, + 229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1333.0, + 191.0, + 1405.0, + 191.0, + 1405.0, + 229.0, + 1333.0, + 229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 225.0, + 1404.0, + 225.0, + 1404.0, + 258.0, + 294.0, + 258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 254.0, + 415.0, + 254.0, + 415.0, + 294.0, + 293.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 483.0, + 254.0, + 613.0, + 254.0, + 613.0, + 294.0, + 483.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 749.0, + 254.0, + 758.0, + 254.0, + 758.0, + 294.0, + 749.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 254.0, + 949.0, + 254.0, + 949.0, + 294.0, + 901.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1066.0, + 254.0, + 1335.0, + 254.0, + 1335.0, + 294.0, + 1066.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 254.0, + 1408.0, + 254.0, + 1408.0, + 294.0, + 1404.0, + 294.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 287.0, + 426.0, + 287.0, + 426.0, + 320.0, + 295.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 287.0, + 611.0, + 287.0, + 611.0, + 320.0, + 563.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 743.0, + 287.0, + 1365.0, + 287.0, + 1365.0, + 320.0, + 743.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1186.0, + 706.0, + 1186.0, + 706.0, + 1226.0, + 295.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 789.0, + 1186.0, + 1076.0, + 1186.0, + 1076.0, + 1226.0, + 789.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 1942.0, + 1405.0, + 1942.0, + 1405.0, + 1986.0, + 328.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1976.0, + 1065.0, + 1976.0, + 1065.0, + 2011.0, + 296.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1449.0, + 1401.0, + 1449.0, + 1401.0, + 1485.0, + 294.0, + 1485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1480.0, + 1203.0, + 1480.0, + 1203.0, + 1514.0, + 293.0, + 1514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 411.0, + 648.0, + 1101.0, + 648.0, + 1101.0, + 684.0, + 411.0, + 684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1003.0, + 801.0, + 1003.0, + 801.0, + 1036.0, + 295.0, + 1036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 1003.0, + 1403.0, + 1003.0, + 1403.0, + 1036.0, + 878.0, + 1036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1032.0, + 581.0, + 1032.0, + 581.0, + 1065.0, + 294.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 617.0, + 1032.0, + 1297.0, + 1032.0, + 1297.0, + 1065.0, + 617.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 1032.0, + 1406.0, + 1032.0, + 1406.0, + 1065.0, + 1317.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1064.0, + 1404.0, + 1064.0, + 1404.0, + 1097.0, + 295.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1090.0, + 1406.0, + 1090.0, + 1406.0, + 1132.0, + 293.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1122.0, + 1076.0, + 1122.0, + 1076.0, + 1158.0, + 293.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2060.0, + 861.0, + 2060.0, + 861.0, + 2090.0, + 840.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1864.0, + 1403.0, + 1864.0, + 1403.0, + 1906.0, + 295.0, + 1906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1896.0, + 1407.0, + 1896.0, + 1407.0, + 1935.0, + 294.0, + 1935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2059.0, + 861.0, + 2059.0, + 861.0, + 2090.0, + 840.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1244.0, + 1405.0, + 1244.0, + 1405.0, + 1279.0, + 294.0, + 1279.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1273.0, + 1187.0, + 1273.0, + 1187.0, + 1310.0, + 294.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 1273.0, + 1405.0, + 1273.0, + 1405.0, + 1310.0, + 1354.0, + 1310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1305.0, + 1403.0, + 1305.0, + 1403.0, + 1340.0, + 294.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1332.0, + 1406.0, + 1332.0, + 1406.0, + 1371.0, + 294.0, + 1371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1367.0, + 1406.0, + 1367.0, + 1406.0, + 1398.0, + 296.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1392.0, + 295.0, + 1392.0, + 295.0, + 1430.0, + 292.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 1392.0, + 1383.0, + 1392.0, + 1383.0, + 1430.0, + 385.0, + 1430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 687.0, + 1405.0, + 687.0, + 1405.0, + 728.0, + 295.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 720.0, + 511.0, + 720.0, + 511.0, + 756.0, + 297.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 598.0, + 720.0, + 896.0, + 720.0, + 896.0, + 756.0, + 598.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1864.0, + 1403.0, + 1864.0, + 1403.0, + 1906.0, + 295.0, + 1906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1896.0, + 1407.0, + 1896.0, + 1407.0, + 1935.0, + 294.0, + 1935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1003.0, + 801.0, + 1003.0, + 801.0, + 1036.0, + 295.0, + 1036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 1003.0, + 1403.0, + 1003.0, + 1403.0, + 1036.0, + 878.0, + 1036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1032.0, + 581.0, + 1032.0, + 581.0, + 1065.0, + 294.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 617.0, + 1032.0, + 1297.0, + 1032.0, + 1297.0, + 1065.0, + 617.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1317.0, + 1032.0, + 1406.0, + 1032.0, + 1406.0, + 1065.0, + 1317.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1064.0, + 1404.0, + 1064.0, + 1404.0, + 1097.0, + 295.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1090.0, + 1406.0, + 1090.0, + 1406.0, + 1132.0, + 293.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1122.0, + 1076.0, + 1122.0, + 1076.0, + 1158.0, + 293.0, + 1158.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 776, + 1405, + 776, + 1405, + 1022, + 297, + 1022 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 298, + 1110, + 1404, + 1110, + 1404, + 1295, + 298, + 1295 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 297, + 1308, + 1405, + 1308, + 1405, + 1523, + 297, + 1523 + ], + "score": 0.978 + }, + { + "category_id": 3, + "poly": [ + 349, + 319, + 1349, + 319, + 1349, + 633, + 349, + 633 + ], + "score": 0.972 + }, + { + "category_id": 4, + "poly": [ + 297, + 655, + 1407, + 655, + 1407, + 749, + 297, + 749 + ], + "score": 0.959 + }, + { + "category_id": 1, + "poly": [ + 298, + 202, + 1404, + 202, + 1404, + 295, + 298, + 295 + ], + "score": 0.945 + }, + { + "category_id": 1, + "poly": [ + 297, + 1944, + 1399, + 1944, + 1399, + 2008, + 297, + 2008 + ], + "score": 0.94 + }, + { + "category_id": 0, + "poly": [ + 298, + 1055, + 1142, + 1055, + 1142, + 1089, + 298, + 1089 + ], + "score": 0.923 + }, + { + "category_id": 3, + "poly": [ + 350, + 1554, + 661, + 1554, + 661, + 1802, + 350, + 1802 + ], + "score": 0.872 + }, + { + "category_id": 4, + "poly": [ + 296, + 1822, + 1399, + 1822, + 1399, + 1917, + 296, + 1917 + ], + "score": 0.81 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 858, + 2061, + 858, + 2084, + 841, + 2084 + ], + "score": 0.723 + }, + { + "category_id": 0, + "poly": [ + 838, + 1571, + 1308, + 1571, + 1308, + 1602, + 838, + 1602 + ], + "score": 0.428 + }, + { + "category_id": 1, + "poly": [ + 795, + 1612, + 1274, + 1612, + 1274, + 1787, + 795, + 1787 + ], + "score": 0.34 + }, + { + "category_id": 4, + "poly": [ + 295, + 1822, + 1400, + 1822, + 1400, + 1917, + 295, + 1917 + ], + "score": 0.238 + }, + { + "category_id": 13, + "poly": [ + 887, + 1884, + 1044, + 1884, + 1044, + 1917, + 887, + 1917 + ], + "score": 0.93, + "latex": "{ d _ { \\mathrm { o u t } } = d _ { \\mathrm { f f n } } / 2 }" + }, + { + "category_id": 13, + "poly": [ + 857, + 1641, + 878, + 1641, + 878, + 1658, + 857, + 1658 + ], + "score": 0.68, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 1203, + 1976, + 1312, + 1976, + 1312, + 2007, + 1203, + 2007 + ], + "score": 0.59, + "latex": "\\mathbf { B E R T _ { b a s e } }" + }, + { + "category_id": 13, + "poly": [ + 338, + 1853, + 414, + 1853, + 414, + 1886, + 338, + 1886 + ], + "score": 0.51, + "latex": "\\mathrm { g M L P }" + }, + { + "category_id": 13, + "poly": [ + 547, + 1491, + 624, + 1491, + 624, + 1523, + 547, + 1523 + ], + "score": 0.36, + "latex": "\\mathrm { g M L P }" + }, + { + "category_id": 13, + "poly": [ + 814, + 1688, + 855, + 1688, + 855, + 1706, + 814, + 1706 + ], + "score": 0.35, + "latex": "\\kappa =" + }, + { + "category_id": 13, + "poly": [ + 900, + 1664, + 923, + 1664, + 923, + 1682, + 900, + 1682 + ], + "score": 0.35, + "latex": "=" + }, + { + "category_id": 13, + "poly": [ + 968, + 1710, + 987, + 1710, + 987, + 1729, + 968, + 1729 + ], + "score": 0.3, + "latex": "^ *" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 333.0, + 443.0, + 333.0, + 443.0, + 343.0, + 434.0, + 343.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 331.0, + 655.0, + 331.0, + 655.0, + 362.0, + 537.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 324.0, + 744.0, + 324.0, + 744.0, + 358.0, + 707.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 907.0, + 330.0, + 921.0, + 330.0, + 921.0, + 348.0, + 907.0, + 348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 329.0, + 1079.0, + 329.0, + 1079.0, + 362.0, + 1043.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1247.0, + 333.0, + 1257.0, + 333.0, + 1257.0, + 342.0, + 1247.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 346.0, + 341.0, + 411.0, + 341.0, + 411.0, + 566.0, + 346.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 510.0, + 367.0, + 521.0, + 367.0, + 521.0, + 376.0, + 510.0, + 376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 356.0, + 598.0, + 356.0, + 598.0, + 387.0, + 534.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 368.0, + 745.0, + 368.0, + 745.0, + 401.0, + 707.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 378.0, + 1256.0, + 378.0, + 1256.0, + 390.0, + 1245.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 402.0, + 868.0, + 402.0, + 868.0, + 412.0, + 860.0, + 412.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 385.0, + 1080.0, + 385.0, + 1080.0, + 416.0, + 1043.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 673.0, + 407.0, + 756.0, + 407.0, + 756.0, + 502.0, + 673.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 412.0, + 744.0, + 412.0, + 744.0, + 445.0, + 706.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 415.0, + 1052.0, + 415.0, + 1052.0, + 495.0, + 1021.0, + 495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 907.0, + 432.0, + 923.0, + 432.0, + 923.0, + 445.0, + 907.0, + 445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 485.0, + 440.0, + 493.0, + 440.0, + 493.0, + 449.0, + 485.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 440.0, + 1078.0, + 440.0, + 1078.0, + 471.0, + 1043.0, + 471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 456.0, + 410.0, + 456.0, + 410.0, + 487.0, + 370.0, + 487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1116.0, + 464.0, + 1124.0, + 464.0, + 1124.0, + 474.0, + 1116.0, + 474.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 708.0, + 502.0, + 744.0, + 502.0, + 744.0, + 534.0, + 708.0, + 534.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 770.0, + 503.0, + 780.0, + 503.0, + 780.0, + 512.0, + 770.0, + 512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 494.0, + 1080.0, + 494.0, + 1080.0, + 525.0, + 1043.0, + 525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 516.0, + 408.0, + 516.0, + 408.0, + 543.0, + 371.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 778.0, + 536.0, + 793.0, + 536.0, + 793.0, + 550.0, + 778.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 518.0, + 991.0, + 518.0, + 991.0, + 550.0, + 871.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 517.0, + 1326.0, + 517.0, + 1326.0, + 550.0, + 1206.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 708.0, + 545.0, + 745.0, + 545.0, + 745.0, + 577.0, + 708.0, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 554.0, + 856.0, + 554.0, + 856.0, + 564.0, + 847.0, + 564.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 545.0, + 933.0, + 545.0, + 933.0, + 573.0, + 869.0, + 573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 548.0, + 1080.0, + 548.0, + 1080.0, + 581.0, + 1043.0, + 581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1205.0, + 542.0, + 1269.0, + 542.0, + 1269.0, + 575.0, + 1205.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1101.0, + 561.0, + 1118.0, + 561.0, + 1118.0, + 581.0, + 1101.0, + 581.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 581.0, + 504.0, + 581.0, + 504.0, + 615.0, + 457.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 636.0, + 584.0, + 681.0, + 584.0, + 681.0, + 613.0, + 636.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 790.0, + 583.0, + 837.0, + 583.0, + 837.0, + 612.0, + 790.0, + 612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 581.0, + 1019.0, + 581.0, + 1019.0, + 615.0, + 973.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1127.0, + 581.0, + 1173.0, + 581.0, + 1173.0, + 615.0, + 1127.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1304.0, + 583.0, + 1352.0, + 583.0, + 1352.0, + 611.0, + 1304.0, + 611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 603.0, + 634.0, + 603.0, + 634.0, + 636.0, + 436.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 771.0, + 605.0, + 970.0, + 605.0, + 970.0, + 636.0, + 771.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1107.0, + 605.0, + 1305.0, + 605.0, + 1305.0, + 636.0, + 1107.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.25, + 362.0, + 463.25, + 362.0, + 463.25, + 381.0, + 440.25, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 365.0, + 399.0, + 415.0, + 399.0, + 415.0, + 437.5, + 365.0, + 437.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 509.75, + 514.5, + 578.75, + 514.5, + 578.75, + 529.5, + 509.75, + 529.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 651.0, + 1405.0, + 651.0, + 1405.0, + 694.0, + 294.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 683.0, + 1409.0, + 683.0, + 1409.0, + 723.0, + 292.0, + 723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 714.0, + 1215.0, + 714.0, + 1215.0, + 751.0, + 294.0, + 751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1050.0, + 1144.0, + 1050.0, + 1144.0, + 1096.0, + 292.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 382.0, + 1611.0, + 419.0, + 1611.0, + 419.0, + 1642.0, + 382.0, + 1642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 481.0, + 1613.0, + 506.0, + 1613.0, + 506.0, + 1640.0, + 481.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 1636.0, + 403.0, + 1636.0, + 403.0, + 1648.0, + 396.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 439.0, + 1652.0, + 548.0, + 1652.0, + 548.0, + 1685.0, + 439.0, + 1685.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 1687.0, + 499.0, + 1687.0, + 499.0, + 1696.0, + 489.0, + 1696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 579.0, + 1685.0, + 626.0, + 1685.0, + 626.0, + 1717.0, + 579.0, + 1717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 464.0, + 1694.0, + 523.0, + 1694.0, + 523.0, + 1727.0, + 464.0, + 1727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 488.0, + 1726.0, + 500.0, + 1726.0, + 500.0, + 1738.0, + 488.0, + 1738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 598.0, + 1728.0, + 606.0, + 1728.0, + 606.0, + 1736.0, + 598.0, + 1736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 1732.0, + 424.0, + 1732.0, + 424.0, + 1761.0, + 374.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 1760.0, + 409.0, + 1760.0, + 409.0, + 1777.0, + 391.0, + 1777.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 580.75, + 1664.5, + 623.75, + 1664.5, + 623.75, + 1696.0, + 580.75, + 1696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1821.0, + 1405.0, + 1821.0, + 1405.0, + 1859.0, + 295.0, + 1859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1854.0, + 337.0, + 1854.0, + 337.0, + 1888.0, + 296.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 1854.0, + 1405.0, + 1854.0, + 1405.0, + 1888.0, + 415.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1883.0, + 886.0, + 1883.0, + 886.0, + 1921.0, + 295.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1883.0, + 1316.0, + 1883.0, + 1316.0, + 1921.0, + 1045.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2059.0, + 860.0, + 2059.0, + 860.0, + 2091.0, + 839.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 1569.0, + 1309.0, + 1569.0, + 1309.0, + 1605.0, + 836.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1821.0, + 1406.0, + 1821.0, + 1406.0, + 1859.0, + 295.0, + 1859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1852.0, + 337.0, + 1852.0, + 337.0, + 1887.0, + 296.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 1852.0, + 1404.0, + 1852.0, + 1404.0, + 1887.0, + 415.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1883.0, + 886.0, + 1883.0, + 886.0, + 1921.0, + 295.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1883.0, + 1317.0, + 1883.0, + 1317.0, + 1921.0, + 1045.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 774.0, + 1405.0, + 774.0, + 1405.0, + 813.0, + 294.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 806.0, + 1407.0, + 806.0, + 1407.0, + 844.0, + 294.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 838.0, + 1406.0, + 838.0, + 1406.0, + 871.0, + 295.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 869.0, + 1403.0, + 869.0, + 1403.0, + 903.0, + 295.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 898.0, + 1406.0, + 898.0, + 1406.0, + 935.0, + 294.0, + 935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 929.0, + 1403.0, + 929.0, + 1403.0, + 963.0, + 295.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 959.0, + 1406.0, + 959.0, + 1406.0, + 993.0, + 292.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 992.0, + 1256.0, + 992.0, + 1256.0, + 1024.0, + 292.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1107.0, + 1405.0, + 1107.0, + 1405.0, + 1149.0, + 292.0, + 1149.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1140.0, + 1404.0, + 1140.0, + 1404.0, + 1176.0, + 294.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1173.0, + 1405.0, + 1173.0, + 1405.0, + 1205.0, + 296.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1203.0, + 1406.0, + 1203.0, + 1406.0, + 1236.0, + 293.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1233.0, + 1405.0, + 1233.0, + 1405.0, + 1269.0, + 294.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1262.0, + 845.0, + 1262.0, + 845.0, + 1299.0, + 293.0, + 1299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1306.0, + 1405.0, + 1306.0, + 1405.0, + 1342.0, + 294.0, + 1342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1337.0, + 1405.0, + 1337.0, + 1405.0, + 1374.0, + 294.0, + 1374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1370.0, + 1403.0, + 1370.0, + 1403.0, + 1404.0, + 295.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1402.0, + 1403.0, + 1402.0, + 1403.0, + 1433.0, + 296.0, + 1433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1430.0, + 1405.0, + 1430.0, + 1405.0, + 1465.0, + 294.0, + 1465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1462.0, + 1403.0, + 1462.0, + 1403.0, + 1493.0, + 295.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1491.0, + 546.0, + 1491.0, + 546.0, + 1525.0, + 296.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 625.0, + 1491.0, + 1227.0, + 1491.0, + 1227.0, + 1525.0, + 625.0, + 1525.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 201.0, + 1405.0, + 201.0, + 1405.0, + 239.0, + 293.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 234.0, + 1404.0, + 234.0, + 1404.0, + 268.0, + 294.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 262.0, + 866.0, + 262.0, + 866.0, + 299.0, + 294.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1944.0, + 1404.0, + 1944.0, + 1404.0, + 1980.0, + 295.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1973.0, + 1202.0, + 1973.0, + 1202.0, + 2009.0, + 296.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1313.0, + 1973.0, + 1403.0, + 1973.0, + 1403.0, + 2009.0, + 1313.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 793.0, + 1613.0, + 1184.0, + 1613.0, + 1184.0, + 1636.0, + 793.0, + 1636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 813.0, + 1638.0, + 856.0, + 1638.0, + 856.0, + 1660.0, + 813.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 879.0, + 1638.0, + 1271.0, + 1638.0, + 1271.0, + 1660.0, + 879.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 1661.0, + 899.0, + 1661.0, + 899.0, + 1684.0, + 811.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 924.0, + 1661.0, + 1249.0, + 1661.0, + 1249.0, + 1684.0, + 924.0, + 1684.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 1684.0, + 1181.0, + 1684.0, + 1181.0, + 1711.0, + 856.0, + 1711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 1707.0, + 967.0, + 1707.0, + 967.0, + 1731.0, + 811.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 1707.0, + 1149.0, + 1707.0, + 1149.0, + 1731.0, + 988.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 1731.0, + 1180.0, + 1731.0, + 1180.0, + 1755.0, + 811.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 1754.0, + 1227.0, + 1754.0, + 1227.0, + 1779.0, + 811.0, + 1779.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 297, + 1519, + 1405, + 1519, + 1405, + 1765, + 297, + 1765 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 296, + 202, + 1406, + 202, + 1406, + 386, + 296, + 386 + ], + "score": 0.977 + }, + { + "category_id": 3, + "poly": [ + 348, + 1014, + 1349, + 1014, + 1349, + 1329, + 348, + 1329 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 299, + 1884, + 1403, + 1884, + 1403, + 2008, + 299, + 2008 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 299, + 1778, + 1405, + 1778, + 1405, + 1871, + 299, + 1871 + ], + "score": 0.971 + }, + { + "category_id": 3, + "poly": [ + 399, + 410, + 1252, + 410, + 1252, + 796, + 399, + 796 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 297, + 927, + 1403, + 927, + 1403, + 991, + 297, + 991 + ], + "score": 0.943 + }, + { + "category_id": 4, + "poly": [ + 296, + 1351, + 1405, + 1351, + 1405, + 1414, + 296, + 1414 + ], + "score": 0.937 + }, + { + "category_id": 0, + "poly": [ + 297, + 1463, + 870, + 1463, + 870, + 1498, + 297, + 1498 + ], + "score": 0.927 + }, + { + "category_id": 4, + "poly": [ + 298, + 811, + 1404, + 811, + 1404, + 905, + 298, + 905 + ], + "score": 0.916 + }, + { + "category_id": 2, + "poly": [ + 840, + 2062, + 858, + 2062, + 858, + 2085, + 840, + 2085 + ], + "score": 0.798 + }, + { + "category_id": 4, + "poly": [ + 298, + 811, + 1403, + 811, + 1403, + 905, + 298, + 905 + ], + "score": 0.191 + }, + { + "category_id": 13, + "poly": [ + 1030, + 234, + 1178, + 234, + 1178, + 265, + 1030, + 265 + ], + "score": 0.88, + "latex": "\\{ 1 , 2 , 4 , 8 \\} \\times" + }, + { + "category_id": 13, + "poly": [ + 1272, + 1946, + 1333, + 1946, + 1333, + 1975, + 1272, + 1975 + ], + "score": 0.87, + "latex": "2 . 7 \\%" + }, + { + "category_id": 13, + "poly": [ + 1302, + 1915, + 1364, + 1915, + 1364, + 1944, + 1302, + 1944 + ], + "score": 0.87, + "latex": "8 . 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 761, + 234, + 904, + 234, + 904, + 266, + 761, + 266 + ], + "score": 0.86, + "latex": "\\{ 0 . 5 , 1 , 2 \\} \\times" + }, + { + "category_id": 13, + "poly": [ + 794, + 1976, + 867, + 1976, + 867, + 2005, + 794, + 2005 + ], + "score": 0.84, + "latex": "8 9 . 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 579, + 1976, + 691, + 1976, + 691, + 2009, + 579, + 2009 + ], + "score": 0.81, + "latex": "{ \\mathrm { g M L P } } _ { \\mathrm { l a r g e } }" + }, + { + "category_id": 13, + "poly": [ + 1306, + 928, + 1407, + 928, + 1407, + 960, + 1306, + 960 + ], + "score": 0.72, + "latex": "\\mathrm { ( g M L P + }" + }, + { + "category_id": 13, + "poly": [ + 393, + 842, + 468, + 842, + 468, + 874, + 393, + 874 + ], + "score": 0.47, + "latex": "\\mathrm { g M L P }" + }, + { + "category_id": 13, + "poly": [ + 397, + 1612, + 485, + 1612, + 485, + 1644, + 397, + 1644 + ], + "score": 0.4, + "latex": "\\mathrm { g M L P s }" + }, + { + "category_id": 13, + "poly": [ + 997, + 1915, + 1074, + 1915, + 1074, + 1947, + 997, + 1947 + ], + "score": 0.36, + "latex": "\\mathrm { g M L P }" + }, + { + "category_id": 13, + "poly": [ + 974, + 1352, + 1060, + 1352, + 1060, + 1383, + 974, + 1383 + ], + "score": 0.27, + "latex": "\\mathrm { g M L P s }" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1025.0, + 409.0, + 1025.0, + 409.0, + 1057.0, + 370.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1251.0, + 1028.0, + 1260.0, + 1028.0, + 1260.0, + 1038.0, + 1251.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 539.0, + 1029.0, + 655.0, + 1029.0, + 655.0, + 1056.0, + 539.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 1030.0, + 744.0, + 1030.0, + 744.0, + 1063.0, + 706.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1038.0, + 381.0, + 1038.0, + 381.0, + 1263.0, + 347.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 1048.0, + 1079.0, + 1048.0, + 1079.0, + 1080.0, + 1042.0, + 1080.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 1052.0, + 445.0, + 1052.0, + 445.0, + 1063.0, + 436.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 536.0, + 1053.0, + 597.0, + 1053.0, + 597.0, + 1081.0, + 536.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1064.0, + 409.0, + 1064.0, + 409.0, + 1096.0, + 370.0, + 1096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 1070.0, + 745.0, + 1070.0, + 745.0, + 1102.0, + 707.0, + 1102.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 1075.0, + 598.0, + 1075.0, + 598.0, + 1108.0, + 537.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1243.0, + 1089.0, + 1257.0, + 1089.0, + 1257.0, + 1103.0, + 1243.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1045.0, + 1096.0, + 1078.0, + 1096.0, + 1078.0, + 1123.0, + 1045.0, + 1123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 1107.0, + 409.0, + 1107.0, + 409.0, + 1137.0, + 369.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 681.0, + 1108.0, + 744.0, + 1108.0, + 744.0, + 1183.0, + 681.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 910.0, + 1128.0, + 922.0, + 1128.0, + 922.0, + 1138.0, + 910.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 1109.0, + 1052.0, + 1109.0, + 1052.0, + 1190.0, + 1021.0, + 1190.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 1152.0, + 409.0, + 1152.0, + 409.0, + 1183.0, + 371.0, + 1183.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 1147.0, + 744.0, + 1147.0, + 744.0, + 1180.0, + 706.0, + 1180.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1042.0, + 1138.0, + 1078.0, + 1138.0, + 1078.0, + 1170.0, + 1042.0, + 1170.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 1186.0, + 744.0, + 1186.0, + 744.0, + 1219.0, + 707.0, + 1219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 1189.0, + 781.0, + 1189.0, + 781.0, + 1199.0, + 769.0, + 1199.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1184.0, + 1079.0, + 1184.0, + 1079.0, + 1216.0, + 1043.0, + 1216.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1201.0, + 409.0, + 1201.0, + 409.0, + 1232.0, + 370.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 875.0, + 1194.0, + 991.0, + 1194.0, + 991.0, + 1220.0, + 875.0, + 1220.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1210.0, + 1194.0, + 1324.0, + 1194.0, + 1324.0, + 1221.0, + 1210.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 1220.0, + 790.0, + 1220.0, + 790.0, + 1230.0, + 780.0, + 1230.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 1217.0, + 933.0, + 1217.0, + 933.0, + 1244.0, + 870.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1205.0, + 1216.0, + 1269.0, + 1216.0, + 1269.0, + 1246.0, + 1205.0, + 1246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 1226.0, + 744.0, + 1226.0, + 744.0, + 1258.0, + 707.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 871.0, + 1239.0, + 933.0, + 1239.0, + 933.0, + 1271.0, + 871.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1230.0, + 1080.0, + 1230.0, + 1080.0, + 1262.0, + 1043.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 1245.0, + 1117.0, + 1245.0, + 1117.0, + 1260.0, + 1103.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1206.0, + 1239.0, + 1269.0, + 1239.0, + 1269.0, + 1271.0, + 1206.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1252.0, + 410.0, + 1252.0, + 410.0, + 1284.0, + 370.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 456.0, + 1281.0, + 501.0, + 1281.0, + 501.0, + 1306.0, + 456.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 638.0, + 1279.0, + 681.0, + 1279.0, + 681.0, + 1307.0, + 638.0, + 1307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 1265.0, + 745.0, + 1265.0, + 745.0, + 1295.0, + 707.0, + 1295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 792.0, + 1279.0, + 835.0, + 1279.0, + 835.0, + 1306.0, + 792.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 972.0, + 1276.0, + 1018.0, + 1276.0, + 1018.0, + 1311.0, + 972.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1127.0, + 1279.0, + 1170.0, + 1279.0, + 1170.0, + 1308.0, + 1127.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1307.0, + 1281.0, + 1350.0, + 1281.0, + 1350.0, + 1306.0, + 1307.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 1301.0, + 634.0, + 1301.0, + 634.0, + 1331.0, + 437.0, + 1331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 772.0, + 1301.0, + 970.0, + 1301.0, + 970.0, + 1331.0, + 772.0, + 1331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1107.0, + 1301.0, + 1305.0, + 1301.0, + 1305.0, + 1332.0, + 1107.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 914.75, + 1023.5, + 918.75, + 1023.5, + 918.75, + 1052.0, + 914.75, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 483.0, + 1138.0, + 488.0, + 1138.0, + 488.0, + 1147.5, + 483.0, + 1147.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 522.0, + 1211.0, + 578.0, + 1211.0, + 578.0, + 1237.0, + 522.0, + 1237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 415.0, + 444.0, + 415.0, + 444.0, + 436.0, + 421.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 446.0, + 409.0, + 575.0, + 409.0, + 575.0, + 442.0, + 446.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 408.0, + 684.0, + 408.0, + 684.0, + 444.0, + 588.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 695.0, + 408.0, + 789.0, + 408.0, + 789.0, + 441.0, + 695.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 415.0, + 891.0, + 415.0, + 891.0, + 437.0, + 869.0, + 437.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 894.0, + 409.0, + 1020.0, + 409.0, + 1020.0, + 442.0, + 894.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1038.0, + 408.0, + 1131.0, + 408.0, + 1131.0, + 444.0, + 1038.0, + 444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 409.0, + 1236.0, + 409.0, + 1236.0, + 441.0, + 1140.0, + 441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 440.0, + 511.0, + 440.0, + 511.0, + 480.0, + 452.0, + 480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 901.0, + 442.0, + 958.0, + 442.0, + 958.0, + 479.0, + 901.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1223.0, + 472.0, + 1235.0, + 472.0, + 1235.0, + 487.0, + 1223.0, + 487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 776.0, + 479.0, + 789.0, + 479.0, + 789.0, + 493.0, + 776.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1199.0, + 483.0, + 1214.0, + 483.0, + 1214.0, + 500.0, + 1199.0, + 500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 733.0, + 497.0, + 747.0, + 497.0, + 747.0, + 509.0, + 733.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 505.0, + 949.0, + 505.0, + 949.0, + 539.0, + 902.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 507.0, + 1213.0, + 507.0, + 1213.0, + 520.0, + 1200.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 527.0, + 434.0, + 527.0, + 434.0, + 670.0, + 399.0, + 670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 522.0, + 508.0, + 522.0, + 508.0, + 559.0, + 451.0, + 559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 715.0, + 521.0, + 768.0, + 521.0, + 768.0, + 544.0, + 715.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 846.0, + 517.0, + 881.0, + 517.0, + 881.0, + 675.0, + 846.0, + 675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 534.0, + 1193.0, + 534.0, + 1193.0, + 547.0, + 1181.0, + 547.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 707.0, + 551.0, + 723.0, + 551.0, + 723.0, + 572.0, + 707.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1208.0, + 560.0, + 1217.0, + 560.0, + 1217.0, + 571.0, + 1208.0, + 571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 741.0, + 581.0, + 767.0, + 581.0, + 767.0, + 600.0, + 741.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 565.0, + 952.0, + 565.0, + 952.0, + 602.0, + 902.0, + 602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1168.0, + 587.0, + 1179.0, + 587.0, + 1179.0, + 600.0, + 1168.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 456.0, + 606.0, + 498.0, + 606.0, + 498.0, + 641.0, + 456.0, + 641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 633.0, + 609.0, + 642.0, + 609.0, + 642.0, + 618.0, + 633.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 601.0, + 696.0, + 601.0, + 696.0, + 616.0, + 685.0, + 616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 597.0, + 1102.0, + 597.0, + 1102.0, + 614.0, + 1087.0, + 614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 703.0, + 614.0, + 716.0, + 614.0, + 716.0, + 626.0, + 703.0, + 626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 634.0, + 731.0, + 634.0, + 731.0, + 648.0, + 718.0, + 648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 903.0, + 627.0, + 950.0, + 627.0, + 950.0, + 662.0, + 903.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1016.0, + 622.0, + 1036.0, + 622.0, + 1036.0, + 643.0, + 1016.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 506.0, + 649.0, + 522.0, + 649.0, + 522.0, + 665.0, + 506.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 640.0, + 645.0, + 655.0, + 645.0, + 655.0, + 662.0, + 640.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 684.0, + 566.0, + 684.0, + 566.0, + 695.0, + 553.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 904.0, + 687.0, + 946.0, + 687.0, + 946.0, + 721.0, + 904.0, + 721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 713.0, + 779.0, + 713.0, + 779.0, + 752.0, + 523.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 969.0, + 713.0, + 1227.0, + 713.0, + 1227.0, + 752.0, + 969.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 763.0, + 740.0, + 763.0, + 740.0, + 804.0, + 562.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 762.0, + 1188.0, + 762.0, + 1188.0, + 804.0, + 1009.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 499.0, + 1196.0, + 499.0, + 1196.0, + 524.5, + 1188.0, + 524.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1144.0, + 535.0, + 1186.0, + 535.0, + 1186.0, + 556.5, + 1144.0, + 556.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 665.75, + 559.5, + 691.75, + 559.5, + 691.75, + 583.0, + 665.75, + 583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1117.0, + 559.0, + 1148.0, + 559.0, + 1148.0, + 574.5, + 1117.0, + 574.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 685.5, + 505.0, + 685.5, + 505.0, + 721.0, + 453.0, + 721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1351.0, + 973.0, + 1351.0, + 973.0, + 1386.0, + 295.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1061.0, + 1351.0, + 1407.0, + 1351.0, + 1407.0, + 1386.0, + 1061.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1380.0, + 1131.0, + 1380.0, + 1131.0, + 1416.0, + 294.0, + 1416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1457.0, + 873.0, + 1457.0, + 873.0, + 1506.0, + 292.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 812.0, + 1406.0, + 812.0, + 1406.0, + 846.0, + 296.0, + 846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 841.0, + 392.0, + 841.0, + 392.0, + 876.0, + 293.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 841.0, + 1409.0, + 841.0, + 1409.0, + 876.0, + 469.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 874.0, + 1308.0, + 874.0, + 1308.0, + 904.0, + 296.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2090.0, + 838.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 811.0, + 1407.0, + 811.0, + 1407.0, + 848.0, + 294.0, + 848.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 841.0, + 392.0, + 841.0, + 392.0, + 876.0, + 293.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 469.0, + 841.0, + 1408.0, + 841.0, + 1408.0, + 876.0, + 469.0, + 876.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 874.0, + 1309.0, + 874.0, + 1309.0, + 904.0, + 296.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1521.0, + 1405.0, + 1521.0, + 1405.0, + 1554.0, + 295.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1552.0, + 1405.0, + 1552.0, + 1405.0, + 1586.0, + 295.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1582.0, + 1405.0, + 1582.0, + 1405.0, + 1616.0, + 295.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1612.0, + 396.0, + 1612.0, + 396.0, + 1646.0, + 295.0, + 1646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 486.0, + 1612.0, + 1403.0, + 1612.0, + 1403.0, + 1646.0, + 486.0, + 1646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1642.0, + 1408.0, + 1642.0, + 1408.0, + 1679.0, + 292.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1670.0, + 1407.0, + 1670.0, + 1407.0, + 1710.0, + 291.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1703.0, + 1405.0, + 1703.0, + 1405.0, + 1737.0, + 295.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1733.0, + 1167.0, + 1733.0, + 1167.0, + 1767.0, + 295.0, + 1767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 201.0, + 1406.0, + 201.0, + 1406.0, + 238.0, + 293.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 233.0, + 760.0, + 233.0, + 760.0, + 269.0, + 294.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 905.0, + 233.0, + 1029.0, + 233.0, + 1029.0, + 269.0, + 905.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1179.0, + 233.0, + 1406.0, + 233.0, + 1406.0, + 269.0, + 1179.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 262.0, + 1406.0, + 262.0, + 1406.0, + 298.0, + 294.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 294.0, + 1404.0, + 294.0, + 1404.0, + 330.0, + 294.0, + 330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 321.0, + 1407.0, + 321.0, + 1407.0, + 364.0, + 291.0, + 364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 355.0, + 1148.0, + 355.0, + 1148.0, + 390.0, + 294.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1882.0, + 1407.0, + 1882.0, + 1407.0, + 1920.0, + 293.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1914.0, + 996.0, + 1914.0, + 996.0, + 1951.0, + 293.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1075.0, + 1914.0, + 1301.0, + 1914.0, + 1301.0, + 1951.0, + 1075.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1365.0, + 1914.0, + 1407.0, + 1914.0, + 1407.0, + 1951.0, + 1365.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1944.0, + 1271.0, + 1944.0, + 1271.0, + 1983.0, + 293.0, + 1983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1334.0, + 1944.0, + 1405.0, + 1944.0, + 1405.0, + 1983.0, + 1334.0, + 1983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1974.0, + 578.0, + 1974.0, + 578.0, + 2013.0, + 293.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 1974.0, + 793.0, + 1974.0, + 793.0, + 2013.0, + 692.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 868.0, + 1974.0, + 1407.0, + 1974.0, + 1407.0, + 2013.0, + 868.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1775.0, + 1409.0, + 1775.0, + 1409.0, + 1814.0, + 292.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1805.0, + 1409.0, + 1805.0, + 1409.0, + 1846.0, + 293.0, + 1846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1838.0, + 996.0, + 1838.0, + 996.0, + 1873.0, + 294.0, + 1873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 927.0, + 1305.0, + 927.0, + 1305.0, + 963.0, + 293.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 958.0, + 1165.0, + 958.0, + 1165.0, + 994.0, + 295.0, + 994.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 316, + 638, + 1379, + 638, + 1379, + 1097, + 316, + 1097 + ], + "score": 0.986, + "html": "
PerplexitySST-2MNLI (m/mm)SQuADAttn SizeParams (M)
v1.1v2.0
BERTbase [2]192.784.4/-88.576.3768 (64 × 12)110
BERTbase (ours)4.1793.885.6/85.790.278.6768 (64 × 12)110
gMLPbase4.2894.283.7/84.186.770.11130
aMLPbase3.9593.485.9/85.890.780.964109
BERTlarge [2]193.786.6/-90.981.81024 (64 × 16)336
BERTlarge (ours)3.3594.387.0/87.492.081.01024 (64 × 16)336
gMLPlarge3.3294.886.2/86.589.578.31365
aMLPlarge3.1994.888.4/88.492.285.4128316
gMLPxlarge2.8995.687.7/87.790.982.11941
" + }, + { + "category_id": 1, + "poly": [ + 298, + 1354, + 1404, + 1354, + 1404, + 1629, + 298, + 1629 + ], + "score": 0.982 + }, + { + "category_id": 5, + "poly": [ + 496, + 243, + 1205, + 243, + 1205, + 529, + 496, + 529 + ], + "score": 0.981, + "html": "
Params (M)FLOPs (B)#Ldmodeldffn
BERTbase gMLPbase aMLPbase110100.812+127683072
130158.0485123072
109128.9365123072
BERTlarge gMLPlarge aMLPlarge336341.224+2410244096
365430.1967683072
316370.3727683072
gMLPxlarge9411091.314410244096
" + }, + { + "category_id": 1, + "poly": [ + 297, + 1157, + 1403, + 1157, + 1403, + 1341, + 297, + 1341 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 300, + 1884, + 1402, + 1884, + 1402, + 2007, + 300, + 2007 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 299, + 1748, + 1404, + 1748, + 1404, + 1870, + 299, + 1870 + ], + "score": 0.975 + }, + { + "category_id": 6, + "poly": [ + 551, + 193, + 1147, + 193, + 1147, + 227, + 551, + 227 + ], + "score": 0.919 + }, + { + "category_id": 0, + "poly": [ + 298, + 1675, + 509, + 1675, + 509, + 1714, + 298, + 1714 + ], + "score": 0.906 + }, + { + "category_id": 6, + "poly": [ + 290, + 558, + 1404, + 558, + 1404, + 622, + 290, + 622 + ], + "score": 0.887 + }, + { + "category_id": 2, + "poly": [ + 840, + 2062, + 858, + 2062, + 858, + 2084, + 840, + 2084 + ], + "score": 0.798 + }, + { + "category_id": 13, + "poly": [ + 978, + 1159, + 1052, + 1159, + 1052, + 1187, + 978, + 1187 + ], + "score": 0.87, + "latex": "8 8 . 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 1018, + 1220, + 1139, + 1220, + 1139, + 1252, + 1018, + 1252 + ], + "score": 0.86, + "latex": "\\mathrm { g } \\mathrm { \\bar { M } L P _ { \\mathrm { x l a r g e } } }" + }, + { + "category_id": 13, + "poly": [ + 972, + 1416, + 1035, + 1416, + 1035, + 1445, + 972, + 1445 + ], + "score": 0.86, + "latex": "4 . 4 \\%" + }, + { + "category_id": 13, + "poly": [ + 521, + 1189, + 580, + 1189, + 580, + 1218, + 521, + 1218 + ], + "score": 0.86, + "latex": "1 . 4 \\%" + }, + { + "category_id": 13, + "poly": [ + 1188, + 1159, + 1294, + 1159, + 1294, + 1190, + 1188, + 1190 + ], + "score": 0.86, + "latex": "\\mathbf { B E R T _ { b a s e } }" + }, + { + "category_id": 13, + "poly": [ + 753, + 1416, + 866, + 1416, + 866, + 1449, + 753, + 1449 + ], + "score": 0.8, + "latex": "\\mathrm { a M L P _ { l a r g e } }" + }, + { + "category_id": 13, + "poly": [ + 941, + 1190, + 1051, + 1190, + 1051, + 1222, + 941, + 1222 + ], + "score": 0.74, + "latex": "\\mathbf { B E R T _ { l a r g e } }" + }, + { + "category_id": 13, + "poly": [ + 564, + 1220, + 638, + 1220, + 638, + 1251, + 564, + 1251 + ], + "score": 0.37, + "latex": "\\mathrm { g M L P }" + }, + { + "category_id": 13, + "poly": [ + 1289, + 1219, + 1401, + 1219, + 1401, + 1251, + 1289, + 1251 + ], + "score": 0.36, + "latex": "\\mathbf { B E R T _ { l a r g e } }" + }, + { + "category_id": 13, + "poly": [ + 428, + 1250, + 479, + 1250, + 479, + 1278, + 428, + 1278 + ], + "score": 0.35, + "latex": "\\mathbf { \\nabla } \\cdot \\mathbf { v } 2 . 0 " + }, + { + "category_id": 15, + "poly": [ + 549.0, + 187.0, + 1150.0, + 187.0, + 1150.0, + 234.0, + 549.0, + 234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1669.0, + 514.0, + 1669.0, + 514.0, + 1722.0, + 290.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 555.0, + 1405.0, + 555.0, + 1405.0, + 593.0, + 294.0, + 593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 586.0, + 1364.0, + 586.0, + 1364.0, + 625.0, + 293.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2060.0, + 860.0, + 2060.0, + 860.0, + 2090.0, + 839.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1356.0, + 1405.0, + 1356.0, + 1405.0, + 1389.0, + 296.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1387.0, + 1405.0, + 1387.0, + 1405.0, + 1420.0, + 296.0, + 1420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1416.0, + 752.0, + 1416.0, + 752.0, + 1452.0, + 293.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 867.0, + 1416.0, + 971.0, + 1416.0, + 971.0, + 1452.0, + 867.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1036.0, + 1416.0, + 1406.0, + 1416.0, + 1406.0, + 1452.0, + 1036.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1445.0, + 1404.0, + 1445.0, + 1404.0, + 1481.0, + 294.0, + 1481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1475.0, + 1408.0, + 1475.0, + 1408.0, + 1515.0, + 293.0, + 1515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1506.0, + 1406.0, + 1506.0, + 1406.0, + 1542.0, + 294.0, + 1542.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1537.0, + 1406.0, + 1537.0, + 1406.0, + 1571.0, + 294.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1567.0, + 1406.0, + 1567.0, + 1406.0, + 1603.0, + 294.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1596.0, + 897.0, + 1596.0, + 897.0, + 1633.0, + 293.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1159.0, + 977.0, + 1159.0, + 977.0, + 1191.0, + 297.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 1159.0, + 1187.0, + 1159.0, + 1187.0, + 1191.0, + 1053.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1295.0, + 1159.0, + 1403.0, + 1159.0, + 1403.0, + 1191.0, + 1295.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1187.0, + 520.0, + 1187.0, + 520.0, + 1226.0, + 292.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 1187.0, + 940.0, + 1187.0, + 940.0, + 1226.0, + 581.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1187.0, + 1405.0, + 1187.0, + 1405.0, + 1226.0, + 1052.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1211.0, + 563.0, + 1211.0, + 563.0, + 1260.0, + 291.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 639.0, + 1211.0, + 1017.0, + 1211.0, + 1017.0, + 1260.0, + 639.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 1211.0, + 1288.0, + 1211.0, + 1288.0, + 1260.0, + 1140.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 1211.0, + 1405.0, + 1211.0, + 1405.0, + 1260.0, + 1402.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1246.0, + 427.0, + 1246.0, + 427.0, + 1286.0, + 293.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 480.0, + 1246.0, + 1404.0, + 1246.0, + 1404.0, + 1286.0, + 480.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1281.0, + 1404.0, + 1281.0, + 1404.0, + 1312.0, + 296.0, + 1312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1309.0, + 1109.0, + 1309.0, + 1109.0, + 1344.0, + 295.0, + 1344.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1885.0, + 1404.0, + 1885.0, + 1404.0, + 1918.0, + 296.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1914.0, + 1407.0, + 1914.0, + 1407.0, + 1950.0, + 295.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1946.0, + 1406.0, + 1946.0, + 1406.0, + 1982.0, + 294.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1973.0, + 1407.0, + 1973.0, + 1407.0, + 2012.0, + 293.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1749.0, + 1403.0, + 1749.0, + 1403.0, + 1781.0, + 295.0, + 1781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1778.0, + 1409.0, + 1778.0, + 1409.0, + 1814.0, + 294.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1808.0, + 1405.0, + 1808.0, + 1405.0, + 1844.0, + 293.0, + 1844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1840.0, + 846.0, + 1840.0, + 846.0, + 1873.0, + 295.0, + 1873.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 298, + 203, + 1406, + 203, + 1406, + 385, + 298, + 385 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 305, + 644, + 1407, + 644, + 1407, + 2012, + 305, + 2012 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 300, + 495, + 1403, + 495, + 1403, + 556, + 300, + 556 + ], + "score": 0.926 + }, + { + "category_id": 0, + "poly": [ + 301, + 427, + 575, + 427, + 575, + 465, + 301, + 465 + ], + "score": 0.91 + }, + { + "category_id": 0, + "poly": [ + 299, + 600, + 455, + 600, + 455, + 635, + 299, + 635 + ], + "score": 0.898 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2085, + 836, + 2085 + ], + "score": 0.838 + }, + { + "category_id": 15, + "poly": [ + 295.0, + 423.0, + 579.0, + 423.0, + 579.0, + 473.0, + 295.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 596.0, + 460.0, + 596.0, + 460.0, + 641.0, + 295.0, + 641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2057.0, + 871.0, + 2057.0, + 871.0, + 2096.0, + 832.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 202.0, + 1408.0, + 202.0, + 1408.0, + 237.0, + 296.0, + 237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 234.0, + 1403.0, + 234.0, + 1403.0, + 265.0, + 296.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 261.0, + 1407.0, + 261.0, + 1407.0, + 299.0, + 295.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 294.0, + 1409.0, + 294.0, + 1409.0, + 328.0, + 292.0, + 328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 324.0, + 1404.0, + 324.0, + 1404.0, + 356.0, + 296.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 356.0, + 917.0, + 356.0, + 917.0, + 387.0, + 296.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 655.0, + 1409.0, + 655.0, + 1409.0, + 693.0, + 309.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 683.0, + 1407.0, + 683.0, + 1407.0, + 724.0, + 353.0, + 724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 716.0, + 707.0, + 716.0, + 707.0, + 750.0, + 353.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 755.0, + 1409.0, + 755.0, + 1409.0, + 796.0, + 306.0, + 796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 785.0, + 1227.0, + 785.0, + 1227.0, + 823.0, + 354.0, + 823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 826.0, + 1406.0, + 826.0, + 1406.0, + 865.0, + 308.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 855.0, + 1406.0, + 855.0, + 1406.0, + 894.0, + 354.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 889.0, + 876.0, + 889.0, + 876.0, + 923.0, + 354.0, + 923.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 927.0, + 1406.0, + 927.0, + 1406.0, + 966.0, + 308.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 953.0, + 1407.0, + 953.0, + 1407.0, + 1001.0, + 351.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 989.0, + 926.0, + 989.0, + 926.0, + 1025.0, + 353.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 1027.0, + 1407.0, + 1027.0, + 1407.0, + 1065.0, + 312.0, + 1065.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 1059.0, + 1406.0, + 1059.0, + 1406.0, + 1097.0, + 356.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1091.0, + 779.0, + 1091.0, + 779.0, + 1125.0, + 354.0, + 1125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 1129.0, + 1407.0, + 1129.0, + 1407.0, + 1167.0, + 308.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 1161.0, + 1404.0, + 1161.0, + 1404.0, + 1195.0, + 356.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1189.0, + 1257.0, + 1189.0, + 1257.0, + 1226.0, + 351.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 1230.0, + 1407.0, + 1230.0, + 1407.0, + 1268.0, + 308.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1258.0, + 1410.0, + 1258.0, + 1410.0, + 1297.0, + 353.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1290.0, + 1404.0, + 1290.0, + 1404.0, + 1328.0, + 353.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 312.0, + 1333.0, + 1404.0, + 1333.0, + 1404.0, + 1366.0, + 312.0, + 1366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1363.0, + 1404.0, + 1363.0, + 1404.0, + 1397.0, + 354.0, + 1397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1392.0, + 739.0, + 1392.0, + 739.0, + 1427.0, + 351.0, + 1427.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 1430.0, + 1406.0, + 1430.0, + 1406.0, + 1469.0, + 308.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1461.0, + 1289.0, + 1461.0, + 1289.0, + 1498.0, + 353.0, + 1498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1499.0, + 1406.0, + 1499.0, + 1406.0, + 1541.0, + 296.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1528.0, + 1406.0, + 1528.0, + 1406.0, + 1573.0, + 351.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1562.0, + 644.0, + 1562.0, + 644.0, + 1596.0, + 353.0, + 1596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1597.0, + 1409.0, + 1597.0, + 1409.0, + 1643.0, + 294.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1626.0, + 430.0, + 1626.0, + 430.0, + 1671.0, + 351.0, + 1671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 1674.0, + 1407.0, + 1674.0, + 1407.0, + 1707.0, + 300.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1703.0, + 1364.0, + 1703.0, + 1364.0, + 1739.0, + 353.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1739.0, + 1407.0, + 1739.0, + 1407.0, + 1784.0, + 294.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1775.0, + 1370.0, + 1775.0, + 1370.0, + 1813.0, + 353.0, + 1813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1813.0, + 1406.0, + 1813.0, + 1406.0, + 1851.0, + 297.0, + 1851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1846.0, + 748.0, + 1846.0, + 748.0, + 1877.0, + 354.0, + 1877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 300.0, + 1882.0, + 1409.0, + 1882.0, + 1409.0, + 1921.0, + 300.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1912.0, + 1410.0, + 1912.0, + 1410.0, + 1952.0, + 351.0, + 1952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1941.0, + 1407.0, + 1941.0, + 1407.0, + 1986.0, + 350.0, + 1986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1975.0, + 429.0, + 1975.0, + 429.0, + 2010.0, + 353.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 494.0, + 1408.0, + 494.0, + 1408.0, + 529.0, + 296.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 524.0, + 1390.0, + 524.0, + 1390.0, + 559.0, + 295.0, + 559.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 294, + 126, + 1410, + 126, + 1410, + 2014, + 294, + 2014 + ], + "score": 0.841 + }, + { + "category_id": 2, + "poly": [ + 835, + 2061, + 864, + 2061, + 864, + 2087, + 835, + 2087 + ], + "score": 0.84 + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2058.0, + 869.0, + 2058.0, + 869.0, + 2097.0, + 831.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 198.0, + 1405.0, + 198.0, + 1405.0, + 246.0, + 290.0, + 246.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 233.0, + 683.0, + 233.0, + 683.0, + 273.0, + 352.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 277.0, + 1405.0, + 277.0, + 1405.0, + 310.0, + 295.0, + 310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 308.0, + 654.0, + 308.0, + 654.0, + 341.0, + 354.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 343.0, + 1405.0, + 343.0, + 1405.0, + 389.0, + 290.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 378.0, + 879.0, + 378.0, + 879.0, + 418.0, + 352.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 420.0, + 1407.0, + 420.0, + 1407.0, + 459.0, + 293.0, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 449.0, + 1071.0, + 449.0, + 1071.0, + 488.0, + 350.0, + 488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 494.0, + 1407.0, + 494.0, + 1407.0, + 527.0, + 295.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 521.0, + 1409.0, + 521.0, + 1409.0, + 561.0, + 354.0, + 561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 552.0, + 1335.0, + 552.0, + 1335.0, + 592.0, + 354.0, + 592.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 590.0, + 1407.0, + 590.0, + 1407.0, + 635.0, + 290.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 625.0, + 1078.0, + 625.0, + 1078.0, + 664.0, + 354.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 666.0, + 1409.0, + 666.0, + 1409.0, + 706.0, + 295.0, + 706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 695.0, + 1407.0, + 695.0, + 1407.0, + 737.0, + 352.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 724.0, + 1407.0, + 724.0, + 1407.0, + 768.0, + 352.0, + 768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 758.0, + 647.0, + 758.0, + 647.0, + 792.0, + 349.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 797.0, + 1409.0, + 797.0, + 1409.0, + 840.0, + 293.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 828.0, + 1405.0, + 828.0, + 1405.0, + 867.0, + 352.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 857.0, + 643.0, + 857.0, + 643.0, + 896.0, + 352.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 902.0, + 1407.0, + 902.0, + 1407.0, + 942.0, + 293.0, + 942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 935.0, + 643.0, + 935.0, + 643.0, + 969.0, + 356.0, + 969.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 975.0, + 1409.0, + 975.0, + 1409.0, + 1014.0, + 295.0, + 1014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1004.0, + 1160.0, + 1004.0, + 1160.0, + 1043.0, + 354.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1043.0, + 1407.0, + 1043.0, + 1407.0, + 1089.0, + 290.0, + 1089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1078.0, + 879.0, + 1078.0, + 879.0, + 1116.0, + 352.0, + 1116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1120.0, + 1371.0, + 1120.0, + 1371.0, + 1159.0, + 295.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1163.0, + 1405.0, + 1163.0, + 1405.0, + 1196.0, + 297.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1190.0, + 1027.0, + 1190.0, + 1027.0, + 1229.0, + 354.0, + 1229.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1232.0, + 1405.0, + 1232.0, + 1405.0, + 1271.0, + 295.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1263.0, + 744.0, + 1263.0, + 744.0, + 1300.0, + 347.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1306.0, + 1285.0, + 1306.0, + 1285.0, + 1345.0, + 295.0, + 1345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1347.0, + 1409.0, + 1347.0, + 1409.0, + 1387.0, + 293.0, + 1387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1379.0, + 1407.0, + 1379.0, + 1407.0, + 1418.0, + 352.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1408.0, + 812.0, + 1408.0, + 812.0, + 1447.0, + 349.0, + 1447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1449.0, + 1405.0, + 1449.0, + 1405.0, + 1488.0, + 295.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1478.0, + 736.0, + 1478.0, + 736.0, + 1517.0, + 352.0, + 1517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1515.0, + 1409.0, + 1515.0, + 1409.0, + 1567.0, + 291.0, + 1567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1553.0, + 1337.0, + 1553.0, + 1337.0, + 1590.0, + 350.0, + 1590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1594.0, + 1407.0, + 1594.0, + 1407.0, + 1633.0, + 295.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1621.0, + 432.0, + 1621.0, + 432.0, + 1662.0, + 350.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1666.0, + 1409.0, + 1666.0, + 1409.0, + 1706.0, + 295.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1695.0, + 434.0, + 1695.0, + 434.0, + 1737.0, + 352.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1739.0, + 1407.0, + 1739.0, + 1407.0, + 1778.0, + 293.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1770.0, + 1407.0, + 1770.0, + 1407.0, + 1809.0, + 352.0, + 1809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1794.0, + 433.0, + 1794.0, + 433.0, + 1841.0, + 352.0, + 1841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1838.0, + 1409.0, + 1838.0, + 1409.0, + 1884.0, + 290.0, + 1884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1871.0, + 1409.0, + 1871.0, + 1409.0, + 1911.0, + 354.0, + 1911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1900.0, + 434.0, + 1900.0, + 434.0, + 1942.0, + 352.0, + 1942.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1940.0, + 1409.0, + 1940.0, + 1409.0, + 1987.0, + 295.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1975.0, + 1407.0, + 1975.0, + 1407.0, + 2014.0, + 354.0, + 2014.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 293, + 199, + 1410, + 199, + 1410, + 1139, + 293, + 1139 + ], + "score": 0.947 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2086, + 836, + 2086 + ], + "score": 0.712 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2086, + 836, + 2086 + ], + "score": 0.336 + }, + { + "category_id": 13, + "poly": [ + 1187, + 623, + 1294, + 623, + 1294, + 653, + 1187, + 653 + ], + "score": 0.74, + "latex": "^ { 1 0 0 , 0 0 0 + }" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 200.0, + 1408.0, + 200.0, + 1408.0, + 242.0, + 294.0, + 242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 235.0, + 846.0, + 235.0, + 846.0, + 268.0, + 353.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 275.0, + 1407.0, + 275.0, + 1407.0, + 310.0, + 298.0, + 310.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 304.0, + 1350.0, + 304.0, + 1350.0, + 342.0, + 353.0, + 342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 345.0, + 1409.0, + 345.0, + 1409.0, + 384.0, + 294.0, + 384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 375.0, + 1404.0, + 375.0, + 1404.0, + 413.0, + 353.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 410.0, + 498.0, + 410.0, + 498.0, + 438.0, + 356.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 449.0, + 1408.0, + 449.0, + 1408.0, + 484.0, + 295.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 476.0, + 1404.0, + 476.0, + 1404.0, + 517.0, + 353.0, + 517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 513.0, + 903.0, + 513.0, + 903.0, + 541.0, + 355.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 548.0, + 1404.0, + 548.0, + 1404.0, + 589.0, + 292.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 580.0, + 1072.0, + 580.0, + 1072.0, + 615.0, + 355.0, + 615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 623.0, + 1186.0, + 623.0, + 1186.0, + 658.0, + 296.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1295.0, + 623.0, + 1403.0, + 623.0, + 1403.0, + 658.0, + 1295.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 653.0, + 966.0, + 653.0, + 966.0, + 686.0, + 353.0, + 686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 694.0, + 1404.0, + 694.0, + 1404.0, + 729.0, + 295.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 724.0, + 754.0, + 724.0, + 754.0, + 759.0, + 353.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 765.0, + 1404.0, + 765.0, + 1404.0, + 800.0, + 295.0, + 800.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 794.0, + 1406.0, + 794.0, + 1406.0, + 832.0, + 353.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 828.0, + 1068.0, + 828.0, + 1068.0, + 861.0, + 355.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 868.0, + 1407.0, + 868.0, + 1407.0, + 903.0, + 295.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 896.0, + 1408.0, + 896.0, + 1408.0, + 934.0, + 352.0, + 934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 928.0, + 663.0, + 928.0, + 663.0, + 963.0, + 352.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 966.0, + 1406.0, + 966.0, + 1406.0, + 1008.0, + 294.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 999.0, + 1406.0, + 999.0, + 1406.0, + 1035.0, + 353.0, + 1035.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1032.0, + 1057.0, + 1032.0, + 1057.0, + 1067.0, + 355.0, + 1067.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1073.0, + 1404.0, + 1073.0, + 1404.0, + 1108.0, + 296.0, + 1108.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1103.0, + 1208.0, + 1103.0, + 1208.0, + 1135.0, + 353.0, + 1135.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/LT0KSFnQDWF/images/04e2e4e5d0ed8f9e8ae5db2c7cfb7091367193f936bdab3e5fea12c1f6590df0.jpg b/parse/train/LT0KSFnQDWF/images/04e2e4e5d0ed8f9e8ae5db2c7cfb7091367193f936bdab3e5fea12c1f6590df0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..17ce3168d321e38c91a12fd8c95e4b6c2277ecda --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/04e2e4e5d0ed8f9e8ae5db2c7cfb7091367193f936bdab3e5fea12c1f6590df0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:30571817a0d3b3db345adda770e9c98ea7ef30f9d65c8b1a148f720e73f34b08 +size 18444 diff --git a/parse/train/LT0KSFnQDWF/images/1089fb7e093efa53bb55c8ebdd22eb5c686675fec3368a4559f186212eb11900.jpg b/parse/train/LT0KSFnQDWF/images/1089fb7e093efa53bb55c8ebdd22eb5c686675fec3368a4559f186212eb11900.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7f8dc9dd19f68d0686bdb08152031d770c6c38b2 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/1089fb7e093efa53bb55c8ebdd22eb5c686675fec3368a4559f186212eb11900.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:98ce57acc47074784587ee7d50003c518d8f6a7a7d5405dc9379ba5ab9d30afa +size 24937 diff --git a/parse/train/LT0KSFnQDWF/images/10f9a6e2cf42f2c3e77ecbd48e24a0d8f0c85108cdb7656edc995a81caa6375d.jpg b/parse/train/LT0KSFnQDWF/images/10f9a6e2cf42f2c3e77ecbd48e24a0d8f0c85108cdb7656edc995a81caa6375d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8b8ae4394afc71b5cd22803195ec90feac9899cd --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/10f9a6e2cf42f2c3e77ecbd48e24a0d8f0c85108cdb7656edc995a81caa6375d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2869abde7add0ea6c78d5eea42104095e2f2b9ddc69ea4f9c0951006533bf5b0 +size 9803 diff --git a/parse/train/LT0KSFnQDWF/images/124519cea299e5cbf8139dbabbd256084e0fce6502eb5abc9980cb29d643323c.jpg b/parse/train/LT0KSFnQDWF/images/124519cea299e5cbf8139dbabbd256084e0fce6502eb5abc9980cb29d643323c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e78f971a45a4b2a9473398725cb732e0740eda65 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/124519cea299e5cbf8139dbabbd256084e0fce6502eb5abc9980cb29d643323c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7435041f6f842049db77e48630f400bba1401cf917153efdba89c5bf71136f61 +size 16505 diff --git a/parse/train/LT0KSFnQDWF/images/13523f8d2601fcf7bed355d79f46959b877e48ffb34048acea00cef37a906ebd.jpg b/parse/train/LT0KSFnQDWF/images/13523f8d2601fcf7bed355d79f46959b877e48ffb34048acea00cef37a906ebd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c9ddc6570fdbccce6c1e737b54226a3fccb24c07 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/13523f8d2601fcf7bed355d79f46959b877e48ffb34048acea00cef37a906ebd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9331138b26e0658f7adbc1c6fa58eb4f6769da7c1ed2cf9306931a6684f1ff14 +size 8459 diff --git a/parse/train/LT0KSFnQDWF/images/22ef441771a0d819a04c6704db3e41c46f6c58965fab49180e249e5574a07f6c.jpg b/parse/train/LT0KSFnQDWF/images/22ef441771a0d819a04c6704db3e41c46f6c58965fab49180e249e5574a07f6c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..710898a865506cd05af93033df065843ac85536b --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/22ef441771a0d819a04c6704db3e41c46f6c58965fab49180e249e5574a07f6c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:12a670f7988b75f615349c4f33d423267bf1276aa7d3c1f3ca3356daffee189e +size 86893 diff --git a/parse/train/LT0KSFnQDWF/images/23c42dd44f74b7cef2257554b738771fd73d815f870e7378104ed7d87955f789.jpg b/parse/train/LT0KSFnQDWF/images/23c42dd44f74b7cef2257554b738771fd73d815f870e7378104ed7d87955f789.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9bde3232b5d7ddcc06a28ef6b6486038027b654a --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/23c42dd44f74b7cef2257554b738771fd73d815f870e7378104ed7d87955f789.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d6f1cd605cb1038f8a37fe37e6d4a3450132756ff73275e0283b566dffc1a018 +size 22153 diff --git a/parse/train/LT0KSFnQDWF/images/25a1138814be262561ae73e5a9a5a1065a816f83620ad64e2c24f0e15f4d24f9.jpg b/parse/train/LT0KSFnQDWF/images/25a1138814be262561ae73e5a9a5a1065a816f83620ad64e2c24f0e15f4d24f9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3b1b397f46ebe233095eaea64a43c286765b47ca --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/25a1138814be262561ae73e5a9a5a1065a816f83620ad64e2c24f0e15f4d24f9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9ea6d1c6b916945652a82348bad4a9c0fd0f8b2c56162b3a3396975033a227a3 +size 18242 diff --git a/parse/train/LT0KSFnQDWF/images/2681b2ac6450afa3d586c39ac2adceb377dc1e475f9ed5e725b2fd56f968d936.jpg b/parse/train/LT0KSFnQDWF/images/2681b2ac6450afa3d586c39ac2adceb377dc1e475f9ed5e725b2fd56f968d936.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ab602b8b5a6efe32a08eae6be43e095600d008f9 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/2681b2ac6450afa3d586c39ac2adceb377dc1e475f9ed5e725b2fd56f968d936.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2d32d06aaee2fa29d864eddf1e2f5e308bcd3779be964686e76a6176847e65f1 +size 26790 diff --git a/parse/train/LT0KSFnQDWF/images/3e10d41982aa12fdf73c1665514b7f2c699b683937f5596711abbf6b4c812465.jpg b/parse/train/LT0KSFnQDWF/images/3e10d41982aa12fdf73c1665514b7f2c699b683937f5596711abbf6b4c812465.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0b3ce9012e515e63a781a45b05e5317f59caadce --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/3e10d41982aa12fdf73c1665514b7f2c699b683937f5596711abbf6b4c812465.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b1db48aa183e0e0bea8433b6415fb0ab822c974c73c42edb73dfd9dc7ad26406 +size 32083 diff --git a/parse/train/LT0KSFnQDWF/images/41313fddf22f912db9fc2d58f117432a52518237de0fadf5a7723b5cf3c58c9f.jpg b/parse/train/LT0KSFnQDWF/images/41313fddf22f912db9fc2d58f117432a52518237de0fadf5a7723b5cf3c58c9f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..648252e0ee7ac089f9380cececc68a691c519a47 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/41313fddf22f912db9fc2d58f117432a52518237de0fadf5a7723b5cf3c58c9f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cdf490fca9b6d7c28c00ec8981b90205c913cf3081b0082ca89f8ce124de4e7f +size 9927 diff --git a/parse/train/LT0KSFnQDWF/images/46094121df041854d824797b08896e0b137b3312a16418511d451622b135b6b0.jpg b/parse/train/LT0KSFnQDWF/images/46094121df041854d824797b08896e0b137b3312a16418511d451622b135b6b0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7bd8551294e20766f3292d92cf7dc521bc2b2039 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/46094121df041854d824797b08896e0b137b3312a16418511d451622b135b6b0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:09bdf74ebfad4fcfd1f12440c2906e4b69fa3d58ea6074ba434a196bb3528872 +size 53323 diff --git a/parse/train/LT0KSFnQDWF/images/4ce36f4bc031dfa7ff04ec4059cc24056552e8146d79bafdf17e8f117a3073f2.jpg b/parse/train/LT0KSFnQDWF/images/4ce36f4bc031dfa7ff04ec4059cc24056552e8146d79bafdf17e8f117a3073f2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d03cbcb716479308d027ef37621e7a806b23612e --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/4ce36f4bc031dfa7ff04ec4059cc24056552e8146d79bafdf17e8f117a3073f2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e9f0c84c032f0022c24b547fa796b97b79a5ee3931dce00a6e12a3f29ad3b474 +size 10442 diff --git a/parse/train/LT0KSFnQDWF/images/4edcfe48473916fc9aa1042079f50a81efab39d2a1ba4165bb5bc32f559e7e99.jpg b/parse/train/LT0KSFnQDWF/images/4edcfe48473916fc9aa1042079f50a81efab39d2a1ba4165bb5bc32f559e7e99.jpg new file mode 100644 index 0000000000000000000000000000000000000000..deb96f60663ee8d5077244e482ae3ccaf2cc22a3 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/4edcfe48473916fc9aa1042079f50a81efab39d2a1ba4165bb5bc32f559e7e99.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:167adf4328304ea58337a448668fa1b2ef6df98a97fce8669ec50ad589deb7c8 +size 17175 diff --git a/parse/train/LT0KSFnQDWF/images/5787a33e188eb6185f5577b8050fbc11ad9e0892c09154e63ccdb919e90936d3.jpg b/parse/train/LT0KSFnQDWF/images/5787a33e188eb6185f5577b8050fbc11ad9e0892c09154e63ccdb919e90936d3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..07043180cb72a145dc5fa045fba24482bc8b1813 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/5787a33e188eb6185f5577b8050fbc11ad9e0892c09154e63ccdb919e90936d3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:947ba998520f6257a0306409cc0afc94b7d217e9ad829772081802c8f977bb33 +size 8080 diff --git a/parse/train/LT0KSFnQDWF/images/5bc553e98d646dd0b9c601882c543a8b08d59edf1a0d4f23bcaf098c58317902.jpg b/parse/train/LT0KSFnQDWF/images/5bc553e98d646dd0b9c601882c543a8b08d59edf1a0d4f23bcaf098c58317902.jpg new file mode 100644 index 0000000000000000000000000000000000000000..86a08915a452a104b3ebeec5a7b84f71db77395d --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/5bc553e98d646dd0b9c601882c543a8b08d59edf1a0d4f23bcaf098c58317902.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8bd21f2ee93d8c3a58eb3fd80e099f556347d7bcb372d6532971f8e1fd0ad0f0 +size 16760 diff --git a/parse/train/LT0KSFnQDWF/images/5bebaf1856cd286121b79b9d6ea8a4ffc29441daa7b44cb6c9c42bbb7150e4f4.jpg b/parse/train/LT0KSFnQDWF/images/5bebaf1856cd286121b79b9d6ea8a4ffc29441daa7b44cb6c9c42bbb7150e4f4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1087d9c45564017ba6816876b52c40cc37956ea3 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/5bebaf1856cd286121b79b9d6ea8a4ffc29441daa7b44cb6c9c42bbb7150e4f4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1c347c672209578f9bf6a4e4e3b9883c0db977c4187074fe3f5badd04c195db +size 30928 diff --git a/parse/train/LT0KSFnQDWF/images/6f212c13f7a009313cfb9e9a56ca4f705906cf1663bd117c2cf7ed38af91fea4.jpg b/parse/train/LT0KSFnQDWF/images/6f212c13f7a009313cfb9e9a56ca4f705906cf1663bd117c2cf7ed38af91fea4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b16283f5079481c5904b9fff473c09f3de75faf7 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/6f212c13f7a009313cfb9e9a56ca4f705906cf1663bd117c2cf7ed38af91fea4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:42bdae8f817b3c7944ef1a78a37e6d0ef9c5f82e88c4585fcfb5cb85a5babcd6 +size 26923 diff --git a/parse/train/LT0KSFnQDWF/images/6f76980b360de6be1f6d54a267c3e7c0b015c9e56d82c6f65922aa520e86182d.jpg b/parse/train/LT0KSFnQDWF/images/6f76980b360de6be1f6d54a267c3e7c0b015c9e56d82c6f65922aa520e86182d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b068713b34d4d908b687678891bdebfe659e10ac --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/6f76980b360de6be1f6d54a267c3e7c0b015c9e56d82c6f65922aa520e86182d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4fcee7cfbdf6738ddfc0ef0a1b7bfd0b7b7e7de8d3d28216ea674618211db3b1 +size 12243 diff --git a/parse/train/LT0KSFnQDWF/images/7244f94874c2d4e095326b7b6092e6931a8644c7d6e90f79542c8c08101dbfab.jpg b/parse/train/LT0KSFnQDWF/images/7244f94874c2d4e095326b7b6092e6931a8644c7d6e90f79542c8c08101dbfab.jpg new file mode 100644 index 0000000000000000000000000000000000000000..62560887be34658724e4f93d60c7614dcc128602 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/7244f94874c2d4e095326b7b6092e6931a8644c7d6e90f79542c8c08101dbfab.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8badca7d60d4e90063872ec2175f9619ca74e3f5fb7d640149746b00f65e33b8 +size 6409 diff --git a/parse/train/LT0KSFnQDWF/images/7b9dd8e964e8bef74f974866bd2fa8ec9ff6a56b69d3762ba5ce30a3db3f53ed.jpg b/parse/train/LT0KSFnQDWF/images/7b9dd8e964e8bef74f974866bd2fa8ec9ff6a56b69d3762ba5ce30a3db3f53ed.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8eb3ea3e071c6f6a732e9276503de9034e56d9be --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/7b9dd8e964e8bef74f974866bd2fa8ec9ff6a56b69d3762ba5ce30a3db3f53ed.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f237ab48b990fe9d2f27ba16e1cd235dcb23a498d5a10026a7b2d5cf30c61d0c +size 7939 diff --git a/parse/train/LT0KSFnQDWF/images/7bb74554a6945a1d137f4a2baa2be53268cd39a49bec819401c0f352138eb504.jpg b/parse/train/LT0KSFnQDWF/images/7bb74554a6945a1d137f4a2baa2be53268cd39a49bec819401c0f352138eb504.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dbd66ed18622a598b8960a69ef37b69e525f42ef --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/7bb74554a6945a1d137f4a2baa2be53268cd39a49bec819401c0f352138eb504.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:63db5d42ea56ca70054b1d4f8b2ffb0d6efa7bfc5adbc269f5504caa3b27e462 +size 34670 diff --git a/parse/train/LT0KSFnQDWF/images/858225708904ce6221562b5071ec77a84d1dcef00bd6473fe9a59d6c597ab2df.jpg b/parse/train/LT0KSFnQDWF/images/858225708904ce6221562b5071ec77a84d1dcef00bd6473fe9a59d6c597ab2df.jpg new file mode 100644 index 0000000000000000000000000000000000000000..805f47a7d3ad502b3285e5842097bcf36380ee55 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/858225708904ce6221562b5071ec77a84d1dcef00bd6473fe9a59d6c597ab2df.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:764c49223c4c10a14921f54a1f445169f1d5e2a6ac6a51817fd3a5d700491fe2 +size 15177 diff --git a/parse/train/LT0KSFnQDWF/images/8c23bf8f7ddba70663b3d1d33802955e46bd0a3bd42cfc127a10fcfea40e55c1.jpg b/parse/train/LT0KSFnQDWF/images/8c23bf8f7ddba70663b3d1d33802955e46bd0a3bd42cfc127a10fcfea40e55c1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e2c75c9fb87f0b5ca013bdf2129bc2e92eb221a9 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/8c23bf8f7ddba70663b3d1d33802955e46bd0a3bd42cfc127a10fcfea40e55c1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9edfd006f89e96211841f8b69824c579f93c3319f2f29aa781c5f8d62c60c550 +size 17261 diff --git a/parse/train/LT0KSFnQDWF/images/9cb50e991f94017d5824566fdb3b91336f0520088da9bafbb4e04fc9e0ce46bc.jpg b/parse/train/LT0KSFnQDWF/images/9cb50e991f94017d5824566fdb3b91336f0520088da9bafbb4e04fc9e0ce46bc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..935bdebf057c1c62c58a24dae40c32687049b912 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/9cb50e991f94017d5824566fdb3b91336f0520088da9bafbb4e04fc9e0ce46bc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:313eacff252dd070ff28bc61e1d0fcc0351cb9862a93a00f126f0527fb24fcad +size 27681 diff --git a/parse/train/LT0KSFnQDWF/images/9e22c51a198434a1bf8fdc8516b98eb4467dda769bb6bf093d9e74667b2f7a17.jpg b/parse/train/LT0KSFnQDWF/images/9e22c51a198434a1bf8fdc8516b98eb4467dda769bb6bf093d9e74667b2f7a17.jpg new file mode 100644 index 0000000000000000000000000000000000000000..43c92885a9d31ad39b34834a4e15ffa7dc09b822 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/9e22c51a198434a1bf8fdc8516b98eb4467dda769bb6bf093d9e74667b2f7a17.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:22106745dfba8d7a0ca460d9d82eefd226542b97a1147e8c3412ee5ba4f638bb +size 16341 diff --git a/parse/train/LT0KSFnQDWF/images/b3764981e57e06e617e3707d95789d79578c95d116d0a7e241bbcfc404f12c45.jpg b/parse/train/LT0KSFnQDWF/images/b3764981e57e06e617e3707d95789d79578c95d116d0a7e241bbcfc404f12c45.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b48e2edcc078b0fdd1acc61dd495b64b556d7262 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/b3764981e57e06e617e3707d95789d79578c95d116d0a7e241bbcfc404f12c45.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5fd305df32a095f311a0c52bde9e969edfe416efd223f23ab4386930fa58553c +size 9647 diff --git a/parse/train/LT0KSFnQDWF/images/b96c60a684e3813c59c8c43837e24f64019a6016d317aadf9c24b3228b083079.jpg b/parse/train/LT0KSFnQDWF/images/b96c60a684e3813c59c8c43837e24f64019a6016d317aadf9c24b3228b083079.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0f7156b8a9224397ec81f8b591943e435d363668 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/b96c60a684e3813c59c8c43837e24f64019a6016d317aadf9c24b3228b083079.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d8ca5532190f3e0faa99e6ce4c202ec2829c6ccfdb32bf83c8a7f11e83648ee1 +size 8196 diff --git a/parse/train/LT0KSFnQDWF/images/bb7359bd8a26d79ebec4aafa23b72b43a50487e2d12d82687ae893b08ff69ea2.jpg b/parse/train/LT0KSFnQDWF/images/bb7359bd8a26d79ebec4aafa23b72b43a50487e2d12d82687ae893b08ff69ea2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..60e5d43ff88d32a27be77f2f47fc01da8284c9de --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/bb7359bd8a26d79ebec4aafa23b72b43a50487e2d12d82687ae893b08ff69ea2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e765b0c8e8c11d16dfcf9a8efb222ae8a8b96db90fc3f43369be70ed33fe156c +size 28344 diff --git a/parse/train/LT0KSFnQDWF/images/c99b96754b166c38a32a13ca34c38647cffd8cc3ce2227bc73168db62b2545fe.jpg b/parse/train/LT0KSFnQDWF/images/c99b96754b166c38a32a13ca34c38647cffd8cc3ce2227bc73168db62b2545fe.jpg new file mode 100644 index 0000000000000000000000000000000000000000..034d3e262862ebeaefb96b940c499dcfda3468a0 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/c99b96754b166c38a32a13ca34c38647cffd8cc3ce2227bc73168db62b2545fe.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:82547975b69d463d9e28f1c0f4463d8bd9891fe91bd0026b793d04ae1ec7174b +size 151052 diff --git a/parse/train/LT0KSFnQDWF/images/caeffa85d36a7e22dfbfc0fe864aba46ea546b57bd3e94d0fca69ee715dd3ab0.jpg b/parse/train/LT0KSFnQDWF/images/caeffa85d36a7e22dfbfc0fe864aba46ea546b57bd3e94d0fca69ee715dd3ab0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4ea08ba43bd64ce2ce306062b3f2551235ab1abb --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/caeffa85d36a7e22dfbfc0fe864aba46ea546b57bd3e94d0fca69ee715dd3ab0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e8d272da4f84678494fd78fabc877c14fefabc6d9cd15e4cb49358169400c9f9 +size 38522 diff --git a/parse/train/LT0KSFnQDWF/images/cbc128cbb06021b55fcad15e3928c2bb57735857b9109dc218643bf1386102a2.jpg b/parse/train/LT0KSFnQDWF/images/cbc128cbb06021b55fcad15e3928c2bb57735857b9109dc218643bf1386102a2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d73ff10d8e29333bfe72ffc554963d29d6bd1f55 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/cbc128cbb06021b55fcad15e3928c2bb57735857b9109dc218643bf1386102a2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:24a5eaeff27c40daa8aea0d6c7bb9daa346ba50e605831615f5d1ba5d1bbab81 +size 13952 diff --git a/parse/train/LT0KSFnQDWF/images/d4d0b141205432dd2a906648ed75b24a8e82a67667619e3e39ede067c70f9bf7.jpg b/parse/train/LT0KSFnQDWF/images/d4d0b141205432dd2a906648ed75b24a8e82a67667619e3e39ede067c70f9bf7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c05658b66283ad29d35b73a4d6759e54c6e2772d --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/d4d0b141205432dd2a906648ed75b24a8e82a67667619e3e39ede067c70f9bf7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a4268d94ca4cd8d5ba362869d6a210e9d64c43ce994a7c105a1456a46635733d +size 5088 diff --git a/parse/train/LT0KSFnQDWF/images/d86b9a657eb4f2b10f4c661bf2cc136d06735dbd6adc6469c4047b32af9d9ce5.jpg b/parse/train/LT0KSFnQDWF/images/d86b9a657eb4f2b10f4c661bf2cc136d06735dbd6adc6469c4047b32af9d9ce5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5761f7fa914cdc8b3be3a96be26fbddc3b803a52 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/d86b9a657eb4f2b10f4c661bf2cc136d06735dbd6adc6469c4047b32af9d9ce5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:21fc0fbb0bb6801dd9332a47cc14aad45831aa3d6320c9fa90a5c51b717e7493 +size 14405 diff --git a/parse/train/LT0KSFnQDWF/images/e6789c4c424d99a5fcab6814f532faa73ab61e804eadf9f00e662680c2a47950.jpg b/parse/train/LT0KSFnQDWF/images/e6789c4c424d99a5fcab6814f532faa73ab61e804eadf9f00e662680c2a47950.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c49a4dba6269ee2fe2353ad5eb3ca794dab50747 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/e6789c4c424d99a5fcab6814f532faa73ab61e804eadf9f00e662680c2a47950.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dc645031df21386de581fb882ce57b12826a25b423d979ab2acfd3046e3f6bb3 +size 25084 diff --git a/parse/train/LT0KSFnQDWF/images/e7dc0bdfb167876f8297de50e47ce8d7952215916d11990d6abe32e3756fbc1e.jpg b/parse/train/LT0KSFnQDWF/images/e7dc0bdfb167876f8297de50e47ce8d7952215916d11990d6abe32e3756fbc1e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..25bdf853a0cced5fbcff2c524078d99c94f66d82 --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/e7dc0bdfb167876f8297de50e47ce8d7952215916d11990d6abe32e3756fbc1e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:838ce7a17ed65167e8a32c7bf6196630663e0e4a09bc6258a7918d855f6df55f +size 106319 diff --git a/parse/train/LT0KSFnQDWF/images/edd41243c2ca7f5a663142564be6a366550b13fb349a0678e422bdf1d94d5c5f.jpg b/parse/train/LT0KSFnQDWF/images/edd41243c2ca7f5a663142564be6a366550b13fb349a0678e422bdf1d94d5c5f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1df4dfc44606c8ba24446b76cd4872bbe43de32d --- /dev/null +++ b/parse/train/LT0KSFnQDWF/images/edd41243c2ca7f5a663142564be6a366550b13fb349a0678e422bdf1d94d5c5f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f513a09589b5d519a7184a512a6d931276ac1b47042cfb5cb8ee6b04267a378f +size 35888 diff --git a/parse/train/LhY8QdUGSuw/images/01d8bc010e7bc4cba69e6bb815584d87d90580f0eaebed9a2484fb7036781be0.jpg b/parse/train/LhY8QdUGSuw/images/01d8bc010e7bc4cba69e6bb815584d87d90580f0eaebed9a2484fb7036781be0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..174ac8aa6180e13a19c3c349d1d5824862cc1c9b --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/01d8bc010e7bc4cba69e6bb815584d87d90580f0eaebed9a2484fb7036781be0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e02b509c6bd9b58d4e51bc41970a70533eb570a2777343deef654ebe103f327f +size 55179 diff --git a/parse/train/LhY8QdUGSuw/images/0a1d478159be5dbd8574320620780d68cbad18ce3b69b75d88525c34721145d0.jpg b/parse/train/LhY8QdUGSuw/images/0a1d478159be5dbd8574320620780d68cbad18ce3b69b75d88525c34721145d0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bcc57ae644d43d2cdaae94a151fb4cbdcbfa4a8b --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/0a1d478159be5dbd8574320620780d68cbad18ce3b69b75d88525c34721145d0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c9a9beabc59518b5c7a252d35190aaf17ec7aac3257363f7e1cc45d7f3d3e07f +size 37531 diff --git a/parse/train/LhY8QdUGSuw/images/0c1f93a5e253061e71524d4448169be4eb17a45238f9076dd080fa8de7682520.jpg b/parse/train/LhY8QdUGSuw/images/0c1f93a5e253061e71524d4448169be4eb17a45238f9076dd080fa8de7682520.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8779bf9ce8a6536a47da254b45b3a2514739f95c --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/0c1f93a5e253061e71524d4448169be4eb17a45238f9076dd080fa8de7682520.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8858c59cbcbe648d23b9aebfdb2bd2cb320c0689db7e2ecd87a3891227105c9e +size 34206 diff --git a/parse/train/LhY8QdUGSuw/images/10d7acc7c85ddb54b5a1f7a9d2b6d802e716edd510dda6927ec734f59ba4c779.jpg b/parse/train/LhY8QdUGSuw/images/10d7acc7c85ddb54b5a1f7a9d2b6d802e716edd510dda6927ec734f59ba4c779.jpg new file mode 100644 index 0000000000000000000000000000000000000000..12aa752d7416d244171bb2c205c16fd26827904b --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/10d7acc7c85ddb54b5a1f7a9d2b6d802e716edd510dda6927ec734f59ba4c779.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4185d675a5f89aad91b677d25c5faf11fb29dd0f04ab8bb6af7a7ba399814995 +size 46501 diff --git a/parse/train/LhY8QdUGSuw/images/1286bf3918cde9573bebe81d7c0c35537c5f1f5d9a005f6ec519eaa214fa23a7.jpg b/parse/train/LhY8QdUGSuw/images/1286bf3918cde9573bebe81d7c0c35537c5f1f5d9a005f6ec519eaa214fa23a7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ad96838cf12cd0850b9af83e180f7bcb39488bac --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/1286bf3918cde9573bebe81d7c0c35537c5f1f5d9a005f6ec519eaa214fa23a7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:44ff682d6ea37be34318cd6765e12b4f4ae0f4a91d8e3c9c220f658eacea5336 +size 50078 diff --git a/parse/train/LhY8QdUGSuw/images/2b3c2a4076751d50ef22d3907dc455f84464a1aed3850e086b4306ebb52d726e.jpg b/parse/train/LhY8QdUGSuw/images/2b3c2a4076751d50ef22d3907dc455f84464a1aed3850e086b4306ebb52d726e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b001d9b704c174604fd0d0830e15e9ecc048d2e1 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/2b3c2a4076751d50ef22d3907dc455f84464a1aed3850e086b4306ebb52d726e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b8c20136f328a25d0c49848ca122f4fa75a26222a54deb5e88d624f890e6411a +size 50147 diff --git a/parse/train/LhY8QdUGSuw/images/2dfff9c926eeb6d9fa33ca0d0d8bcf0e7abfc1985e3fec48909c31dcd8b984d5.jpg b/parse/train/LhY8QdUGSuw/images/2dfff9c926eeb6d9fa33ca0d0d8bcf0e7abfc1985e3fec48909c31dcd8b984d5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..16b772403d9da83129e53eebb778aa2d34fa137c --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/2dfff9c926eeb6d9fa33ca0d0d8bcf0e7abfc1985e3fec48909c31dcd8b984d5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fa637bdba4d802a5fdd200be2b81219779e66542d6b2b5ab9c9a3a95104e08fb +size 5879 diff --git a/parse/train/LhY8QdUGSuw/images/35dfe190006578fae05569f5bf53d346b76b9a6af954acad93632c9d607345d9.jpg b/parse/train/LhY8QdUGSuw/images/35dfe190006578fae05569f5bf53d346b76b9a6af954acad93632c9d607345d9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b3791069b7fe87752adb350c08bc77119a3973fe --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/35dfe190006578fae05569f5bf53d346b76b9a6af954acad93632c9d607345d9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:265bf169bf8daf4c422618cd7a5aa2fe2b2b2e9432b07271f08dea6cbd9c80e4 +size 59194 diff --git a/parse/train/LhY8QdUGSuw/images/54c39e8d19801972f046935444a440a282fa88813511537bedfe3ab1203c847e.jpg b/parse/train/LhY8QdUGSuw/images/54c39e8d19801972f046935444a440a282fa88813511537bedfe3ab1203c847e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f17fd7d0abde42f7d190bbdcb2fd6da6b2646b92 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/54c39e8d19801972f046935444a440a282fa88813511537bedfe3ab1203c847e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1400979a398d3333d63acd8dbbc0b61f194b05150b2e5f8afa755c682c1480f3 +size 53829 diff --git a/parse/train/LhY8QdUGSuw/images/5507a75b14a33e01fef517018ef0ba97edcb85a091b929d800a7fc761023b095.jpg b/parse/train/LhY8QdUGSuw/images/5507a75b14a33e01fef517018ef0ba97edcb85a091b929d800a7fc761023b095.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9e5af98976850b8dc374115fff2e2d4c3706a783 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/5507a75b14a33e01fef517018ef0ba97edcb85a091b929d800a7fc761023b095.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f43620e9a1ddb6282743d39ef71067989032f8835a5a1fd10c90d11cee45c9fe +size 168925 diff --git a/parse/train/LhY8QdUGSuw/images/5560d915c662208e0d52cb9477642c65c77fbf179501bc79b7d1de8d64727597.jpg b/parse/train/LhY8QdUGSuw/images/5560d915c662208e0d52cb9477642c65c77fbf179501bc79b7d1de8d64727597.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0a26dcd68de90c1510b776b3b14cf5ce717b31b3 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/5560d915c662208e0d52cb9477642c65c77fbf179501bc79b7d1de8d64727597.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4405280781743674f44592233cfa9a1837996ce824406091a15c1624f8909eb1 +size 5899 diff --git a/parse/train/LhY8QdUGSuw/images/5ec6882b50b860c46e6f4b01a17d96b09391c130b52aaf17c44ede2c3d479381.jpg b/parse/train/LhY8QdUGSuw/images/5ec6882b50b860c46e6f4b01a17d96b09391c130b52aaf17c44ede2c3d479381.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e09fc707be9022680c2efc511ff9141f7b7ba37d --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/5ec6882b50b860c46e6f4b01a17d96b09391c130b52aaf17c44ede2c3d479381.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3482b7371e0d4b95641d84331623be39d379f709704eb8ba015d6750979d9aea +size 12050 diff --git a/parse/train/LhY8QdUGSuw/images/64f3ffea5fbcf33063fe5ee2b4b7af43bffb11f768520c54eb86ddd34960eda8.jpg b/parse/train/LhY8QdUGSuw/images/64f3ffea5fbcf33063fe5ee2b4b7af43bffb11f768520c54eb86ddd34960eda8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..24ced34afcf22941a6b9c7be676bc0d07861454c --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/64f3ffea5fbcf33063fe5ee2b4b7af43bffb11f768520c54eb86ddd34960eda8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0486563f93c1a610a2aa56fc66dbdb41c835548a1f496cfd7c7be3286221e19e +size 6559 diff --git a/parse/train/LhY8QdUGSuw/images/6784833ddb4e91dbce15ca23e71963d5878b428c23395b6d834b86c3ad595e7e.jpg b/parse/train/LhY8QdUGSuw/images/6784833ddb4e91dbce15ca23e71963d5878b428c23395b6d834b86c3ad595e7e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..526ad482011a6e09186c5b78ee289bc914afeb80 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/6784833ddb4e91dbce15ca23e71963d5878b428c23395b6d834b86c3ad595e7e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cf3115a020df06faae3699a3330666b4c0190973e0c04373599512ac74a440a1 +size 41767 diff --git a/parse/train/LhY8QdUGSuw/images/6e2fad04e0a002db8542c43cb352a649d5d03328e4c47ff4ee1cafae61635513.jpg b/parse/train/LhY8QdUGSuw/images/6e2fad04e0a002db8542c43cb352a649d5d03328e4c47ff4ee1cafae61635513.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f76be2625bad7809e62650daae9820e95a635f73 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/6e2fad04e0a002db8542c43cb352a649d5d03328e4c47ff4ee1cafae61635513.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:081c9d6b3b40f10a1174ff968506da07afa71aed29709204db7fe3c893b9a4fa +size 111645 diff --git a/parse/train/LhY8QdUGSuw/images/7293a1d894a7f86a197ce3820d943a538ecfac8bc07c8388599461e278723e74.jpg b/parse/train/LhY8QdUGSuw/images/7293a1d894a7f86a197ce3820d943a538ecfac8bc07c8388599461e278723e74.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7c5643a618a48ac15a8dd73f63ebf68c583e61de --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/7293a1d894a7f86a197ce3820d943a538ecfac8bc07c8388599461e278723e74.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2e26430cdf1ea621ca2f97c5d5502f3f278ee24e5bfae84799279c99d44714dd +size 53671 diff --git a/parse/train/LhY8QdUGSuw/images/751d12c2239a6b9cb5e1ce1e312ccd471a884b89fcbcd8afa36f1411e562dd10.jpg b/parse/train/LhY8QdUGSuw/images/751d12c2239a6b9cb5e1ce1e312ccd471a884b89fcbcd8afa36f1411e562dd10.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a0c658dd2e63794bc13b8a4582e4bfcb23fbf818 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/751d12c2239a6b9cb5e1ce1e312ccd471a884b89fcbcd8afa36f1411e562dd10.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:368dd9f03f94c1845c6091138cd7f99563bf169a27acd9d21c64db09b0960b35 +size 39566 diff --git a/parse/train/LhY8QdUGSuw/images/7738d8883272e0945f00bf3a03556f8e5d11b18a8f681f0f28faec91860b6a4a.jpg b/parse/train/LhY8QdUGSuw/images/7738d8883272e0945f00bf3a03556f8e5d11b18a8f681f0f28faec91860b6a4a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7e6056fcf500fd6ddceb4329d1fd4f34a75d86f4 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/7738d8883272e0945f00bf3a03556f8e5d11b18a8f681f0f28faec91860b6a4a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:081e031dbce79c4bcc766dc37471ddee176c9d3e22254c5fcaf00a2cda7e14d9 +size 5876 diff --git a/parse/train/LhY8QdUGSuw/images/7759e854b7587e327c6df0e10615bce5aefeb4f71e514d7eedcbbb714d9f4590.jpg b/parse/train/LhY8QdUGSuw/images/7759e854b7587e327c6df0e10615bce5aefeb4f71e514d7eedcbbb714d9f4590.jpg new file mode 100644 index 0000000000000000000000000000000000000000..803635bde98197db4e135d6784ee71b38a2d1750 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/7759e854b7587e327c6df0e10615bce5aefeb4f71e514d7eedcbbb714d9f4590.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a56a446fc9056c4b87b75b6bcac858a6d3cffa877376e88c3815f08da63fdede +size 45403 diff --git a/parse/train/LhY8QdUGSuw/images/7f40507bbc084480cf35a8735d106497f333c573a088bfb5e52c81ffb580b5fa.jpg b/parse/train/LhY8QdUGSuw/images/7f40507bbc084480cf35a8735d106497f333c573a088bfb5e52c81ffb580b5fa.jpg new file mode 100644 index 0000000000000000000000000000000000000000..35c2a4db31d30cd26cd50eac4cce114b117b19da --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/7f40507bbc084480cf35a8735d106497f333c573a088bfb5e52c81ffb580b5fa.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eac8b6b8bf321ed17ad72ee847eaf87327c2538758f14ec9a6e4154ca7c3ca14 +size 3876 diff --git a/parse/train/LhY8QdUGSuw/images/7fb62347b6c3e11263b8f79f914a23a2b0c512424d1d31faa6d0808e1137d433.jpg b/parse/train/LhY8QdUGSuw/images/7fb62347b6c3e11263b8f79f914a23a2b0c512424d1d31faa6d0808e1137d433.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3ae0108ebfe89ca875fc911d4e70c5ebe3adc562 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/7fb62347b6c3e11263b8f79f914a23a2b0c512424d1d31faa6d0808e1137d433.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:44bfaa49f51f4bda9f20e9d2daeec1a36f99f3594352c5dbc5faa78949949043 +size 9099 diff --git a/parse/train/LhY8QdUGSuw/images/7fba10e508e8f15bcb45cb632a24368a6cd6bfdd6a174508e91812284ace6119.jpg b/parse/train/LhY8QdUGSuw/images/7fba10e508e8f15bcb45cb632a24368a6cd6bfdd6a174508e91812284ace6119.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8646056297265502442a1898f58f5f23721ec12f --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/7fba10e508e8f15bcb45cb632a24368a6cd6bfdd6a174508e91812284ace6119.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9d48707a33a0aad152aeb79e2f7cf71ce0a49ec43ff428ac2067094d3d7c2569 +size 4849 diff --git a/parse/train/LhY8QdUGSuw/images/8553198ed3ce878f4463feb1b92ca6ac388c507393962c63efe02df85ed9a425.jpg b/parse/train/LhY8QdUGSuw/images/8553198ed3ce878f4463feb1b92ca6ac388c507393962c63efe02df85ed9a425.jpg new file mode 100644 index 0000000000000000000000000000000000000000..718f5b83426c0ecd0370c4687545a932acd3b0ad --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/8553198ed3ce878f4463feb1b92ca6ac388c507393962c63efe02df85ed9a425.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:38fd33a187b3d8dbf0630f1c93aa4083dc40f52943b62d601c04ca04107c73af +size 70018 diff --git a/parse/train/LhY8QdUGSuw/images/8b45eecffc780f8e5ffdd981f1c29720590080639d5294339abfdcd35e7e827d.jpg b/parse/train/LhY8QdUGSuw/images/8b45eecffc780f8e5ffdd981f1c29720590080639d5294339abfdcd35e7e827d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..12407ed9cc6d602409b00e27a2bd345ba45e4909 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/8b45eecffc780f8e5ffdd981f1c29720590080639d5294339abfdcd35e7e827d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4ee142750da1214bf2c11133f3a459584b8139aba2b9bf3735fa1ebb87e6023b +size 23022 diff --git a/parse/train/LhY8QdUGSuw/images/93210a561cda29c3ecc977132e1ec315a90f01f1a64db5375460c390077263a9.jpg b/parse/train/LhY8QdUGSuw/images/93210a561cda29c3ecc977132e1ec315a90f01f1a64db5375460c390077263a9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fcc7b5e8a1955db56a5df90a2977179e1625ec1e --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/93210a561cda29c3ecc977132e1ec315a90f01f1a64db5375460c390077263a9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4eb1be87987c5c73cfc678542009aaad8ff47b9dbb956f593320610fe25cd83c +size 5542 diff --git a/parse/train/LhY8QdUGSuw/images/97cddca27ac4c1609e19e4ad01716f32b4b6060f855841e02ef202f7ad83dcf3.jpg b/parse/train/LhY8QdUGSuw/images/97cddca27ac4c1609e19e4ad01716f32b4b6060f855841e02ef202f7ad83dcf3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7d9e09295d28a473843396afd037438317198924 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/97cddca27ac4c1609e19e4ad01716f32b4b6060f855841e02ef202f7ad83dcf3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1be4eae7e1195694d9672c2f40012cc235c449104125688003db0329f3ef6d6c +size 103732 diff --git a/parse/train/LhY8QdUGSuw/images/989d29fbf88957c746c36658e38c694a1fda4765067223fb5123c977c85c3fd4.jpg b/parse/train/LhY8QdUGSuw/images/989d29fbf88957c746c36658e38c694a1fda4765067223fb5123c977c85c3fd4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fd305b748448fe0ba852122357327ac81b0f66a1 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/989d29fbf88957c746c36658e38c694a1fda4765067223fb5123c977c85c3fd4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5b4f9def1570558ea8a5a3ef2698a4687d90e1ee39a1ada07b22a175414f8c3c +size 43913 diff --git a/parse/train/LhY8QdUGSuw/images/9a572e7033cd77924c6de4eba1e53eae57b8a03857e2e5c7c1e96aeb78ed10f9.jpg b/parse/train/LhY8QdUGSuw/images/9a572e7033cd77924c6de4eba1e53eae57b8a03857e2e5c7c1e96aeb78ed10f9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..920d6f217257f2eb07ea7d09556dd6e6b26042b6 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/9a572e7033cd77924c6de4eba1e53eae57b8a03857e2e5c7c1e96aeb78ed10f9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:61200e467884ca9089fc152194554f20f676de118880f8d2a3c4aab0d4797652 +size 46793 diff --git a/parse/train/LhY8QdUGSuw/images/9b5309de8e13131c7c40487d0e4b169bea844843307575030317507dedb50c0e.jpg b/parse/train/LhY8QdUGSuw/images/9b5309de8e13131c7c40487d0e4b169bea844843307575030317507dedb50c0e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..aaf24001abd0d1baae67a7a6ec54fee11b9b32a6 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/9b5309de8e13131c7c40487d0e4b169bea844843307575030317507dedb50c0e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1ef7a3e4ad8ae4d1e12ae80bdafeef2d72d5d6e02a9c9c9b793590ca5273d52f +size 142712 diff --git a/parse/train/LhY8QdUGSuw/images/a220ef5a5b19d8f1139d06f902adf47173caf647598a19fac8a8ab98235b275e.jpg b/parse/train/LhY8QdUGSuw/images/a220ef5a5b19d8f1139d06f902adf47173caf647598a19fac8a8ab98235b275e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6e532b92c1f8ff46a631066d3359fcfea5082771 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/a220ef5a5b19d8f1139d06f902adf47173caf647598a19fac8a8ab98235b275e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:abdf8e0b499142e3e77aa221d75ed615a1dc17844020a0f01fb354d2d0e3ce4a +size 112076 diff --git a/parse/train/LhY8QdUGSuw/images/b1d85f3a5fa54339a3b6bb967c722fbbb4516faa7a86ac3690f9f44842797f8f.jpg b/parse/train/LhY8QdUGSuw/images/b1d85f3a5fa54339a3b6bb967c722fbbb4516faa7a86ac3690f9f44842797f8f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c02bcf9ac42871aeb04e6baba7f9ce9007de1c29 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/b1d85f3a5fa54339a3b6bb967c722fbbb4516faa7a86ac3690f9f44842797f8f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5e9871e99fbe26294849709ad8aa9bfb9122f8e0c64809a0f883105da92fbbea +size 48427 diff --git a/parse/train/LhY8QdUGSuw/images/b3baf907d90078054e843944810b8fa928376820b863fcf18688c2c62f336f5b.jpg b/parse/train/LhY8QdUGSuw/images/b3baf907d90078054e843944810b8fa928376820b863fcf18688c2c62f336f5b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cfe1856881ecbb1a6c08737b81286289754b55c4 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/b3baf907d90078054e843944810b8fa928376820b863fcf18688c2c62f336f5b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4e445370794671d12f42bdc25d7202d5e0d08625dfd928aee4cb73a494483478 +size 23094 diff --git a/parse/train/LhY8QdUGSuw/images/b52d8cd137912b64f9ecf3880dd9f2d238e5a880fcedcca4dfffe3d60785038b.jpg b/parse/train/LhY8QdUGSuw/images/b52d8cd137912b64f9ecf3880dd9f2d238e5a880fcedcca4dfffe3d60785038b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fae8c6bcfcbd1f7074bccdd8896fa8cb98467475 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/b52d8cd137912b64f9ecf3880dd9f2d238e5a880fcedcca4dfffe3d60785038b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d0db568aa44209cc0152b5935d07788012565c83ee2f3230154d249dffb69b5c +size 9427 diff --git a/parse/train/LhY8QdUGSuw/images/b7f4682f0d383ae44ae63e7bb1b04ff9eb952c7c7164efeee34f4d84732f6cac.jpg b/parse/train/LhY8QdUGSuw/images/b7f4682f0d383ae44ae63e7bb1b04ff9eb952c7c7164efeee34f4d84732f6cac.jpg new file mode 100644 index 0000000000000000000000000000000000000000..012f6387e85eecaefb612100b1a3010887a9d2f4 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/b7f4682f0d383ae44ae63e7bb1b04ff9eb952c7c7164efeee34f4d84732f6cac.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d130fde1700b01375bde90c59b851cac2d2ccbb3092d9d43c4670e8d1902541a +size 47576 diff --git a/parse/train/LhY8QdUGSuw/images/bcc2ff7123e38051eeb8a6fd458ec8dee7287267a33219bf8dca1ace03234537.jpg b/parse/train/LhY8QdUGSuw/images/bcc2ff7123e38051eeb8a6fd458ec8dee7287267a33219bf8dca1ace03234537.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d8168e68ef48f1b647c0fa1a26a713c77ac177a9 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/bcc2ff7123e38051eeb8a6fd458ec8dee7287267a33219bf8dca1ace03234537.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:28c0c675d92dc01ae328b1532736cda7ce78896eea332614ee36702ad52d3243 +size 60427 diff --git a/parse/train/LhY8QdUGSuw/images/bef6f0ad266969173dec939f3382fcdf9650004af675922a954db3cdce264098.jpg b/parse/train/LhY8QdUGSuw/images/bef6f0ad266969173dec939f3382fcdf9650004af675922a954db3cdce264098.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b963527cc3e3cdc66c3c5e45c32e54576810982d --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/bef6f0ad266969173dec939f3382fcdf9650004af675922a954db3cdce264098.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:03dad76dfce0074e52162c0ae0de10c13e8fa152fce7fde0c646ad6f02ffdace +size 52655 diff --git a/parse/train/LhY8QdUGSuw/images/c45b34098840087b001418b5c2e872344856165330e725c8dd983b94f41ea99e.jpg b/parse/train/LhY8QdUGSuw/images/c45b34098840087b001418b5c2e872344856165330e725c8dd983b94f41ea99e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e96816c079b496c2453521dc8723b35f7c654fbe --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/c45b34098840087b001418b5c2e872344856165330e725c8dd983b94f41ea99e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dc8b07d72ccc5d77b7fa3be61d8a397bb693968addb38f45b484163f699a5b5f +size 20972 diff --git a/parse/train/LhY8QdUGSuw/images/c7bd96d0456a4214ccd862ce782a9ad0e92ab5b7c2ee612857f364b68e67f3d0.jpg b/parse/train/LhY8QdUGSuw/images/c7bd96d0456a4214ccd862ce782a9ad0e92ab5b7c2ee612857f364b68e67f3d0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5d067070e1df0cfe68121c9110da69eaf8d0522c --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/c7bd96d0456a4214ccd862ce782a9ad0e92ab5b7c2ee612857f364b68e67f3d0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2b4a278f6bfd05cbbdf0b320cda56872f7b31f9b6a848dc3c5e598d3c9639bf8 +size 52247 diff --git a/parse/train/LhY8QdUGSuw/images/caeb1d9636133db42d4e9678e00a9fea90d0bf39678be050d5092c7b521b1d94.jpg b/parse/train/LhY8QdUGSuw/images/caeb1d9636133db42d4e9678e00a9fea90d0bf39678be050d5092c7b521b1d94.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f90125388efadb12fe3e377a5e1888fb5242f5dc --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/caeb1d9636133db42d4e9678e00a9fea90d0bf39678be050d5092c7b521b1d94.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:69cf8248aa6af5811ab6967af1834448a02df07c7b07bc3ac7ae88396f1f960e +size 156201 diff --git a/parse/train/LhY8QdUGSuw/images/cb405a4306c40a91294d386db635da43d561bc8adf93279d58ee48b5de102b43.jpg b/parse/train/LhY8QdUGSuw/images/cb405a4306c40a91294d386db635da43d561bc8adf93279d58ee48b5de102b43.jpg new file mode 100644 index 0000000000000000000000000000000000000000..db605694897dfdcb3f1199aefd3749094b85b3fc --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/cb405a4306c40a91294d386db635da43d561bc8adf93279d58ee48b5de102b43.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:25f0816002204c17b2a558abab077b9e5d5fe4a7c040fc52d7ffdc7190bd503d +size 78818 diff --git a/parse/train/LhY8QdUGSuw/images/ccc557677afea0bb941be42ae608ad511067870b50d6a63b79a1bb7484dc60b5.jpg b/parse/train/LhY8QdUGSuw/images/ccc557677afea0bb941be42ae608ad511067870b50d6a63b79a1bb7484dc60b5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9a40431cb55cd843353214bda4c4a6fc6b080693 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/ccc557677afea0bb941be42ae608ad511067870b50d6a63b79a1bb7484dc60b5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:430e722712773fe8c6a0ff64cc447ee7c938d1b2494e32410db07c08fa730c59 +size 88780 diff --git a/parse/train/LhY8QdUGSuw/images/d611e63cde367ca536e0f6643ab383afee09cc289ec2c502088c61e3cc73273b.jpg b/parse/train/LhY8QdUGSuw/images/d611e63cde367ca536e0f6643ab383afee09cc289ec2c502088c61e3cc73273b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4446ea2fc538be4ca87b935c174705ae94a42d82 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/d611e63cde367ca536e0f6643ab383afee09cc289ec2c502088c61e3cc73273b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:08a23db00c3526aec19edf93febd8bc525918af7af1d02ce1be418489a03c1d4 +size 39495 diff --git a/parse/train/LhY8QdUGSuw/images/d9e738fcd6ce18d35701510190b82266c9c4cfa9974169718abd4a15b8d26993.jpg b/parse/train/LhY8QdUGSuw/images/d9e738fcd6ce18d35701510190b82266c9c4cfa9974169718abd4a15b8d26993.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4263a70f05deda87a23b398b5a8a14f13d2b1e25 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/d9e738fcd6ce18d35701510190b82266c9c4cfa9974169718abd4a15b8d26993.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f574fba2b76074e09aebfcf178e7d3a6262edd5b3827064fbb1c8e2d9ed1bec5 +size 7815 diff --git a/parse/train/LhY8QdUGSuw/images/e0aae519536daa021aaa0f0f7cc8abb09571f71f4be45205b3d066c312214874.jpg b/parse/train/LhY8QdUGSuw/images/e0aae519536daa021aaa0f0f7cc8abb09571f71f4be45205b3d066c312214874.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9ea3a9da00a23f942a3f227387bfcf8aedb8b6f2 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/e0aae519536daa021aaa0f0f7cc8abb09571f71f4be45205b3d066c312214874.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6ddaf60b6fedb637ebd57a48aa683840fff4822530c234b049670e6b11558174 +size 151646 diff --git a/parse/train/LhY8QdUGSuw/images/e18234e09a6d3786a709ac998864e82b0db00b76a82fe85a20bb62a7d2fb3d21.jpg b/parse/train/LhY8QdUGSuw/images/e18234e09a6d3786a709ac998864e82b0db00b76a82fe85a20bb62a7d2fb3d21.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1137fd5aabaced09fa12922ea88901e7664488f4 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/e18234e09a6d3786a709ac998864e82b0db00b76a82fe85a20bb62a7d2fb3d21.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:275ee17abd7601ba45297a42552c2f1ff0e733f684da660c9cff857fa260a44c +size 151940 diff --git a/parse/train/LhY8QdUGSuw/images/e56e761d9357dbd46d926b1676ec05908246c98a27d7e5af13aab5d544b04e4a.jpg b/parse/train/LhY8QdUGSuw/images/e56e761d9357dbd46d926b1676ec05908246c98a27d7e5af13aab5d544b04e4a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7b799653ca287d84a1a76fb28f949e8110a7d00f --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/e56e761d9357dbd46d926b1676ec05908246c98a27d7e5af13aab5d544b04e4a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:17ee51fdc2f80aca546f5161c6883620e3f39defe185d8069ed722b965410ba9 +size 42344 diff --git a/parse/train/LhY8QdUGSuw/images/ec04da581d739ba7b6e3e6a23542ce72fd4e6a3953ad5f4e73573d2c1199f81e.jpg b/parse/train/LhY8QdUGSuw/images/ec04da581d739ba7b6e3e6a23542ce72fd4e6a3953ad5f4e73573d2c1199f81e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..75df4cf92b0899c5e031e4d9e9f1052566e0a16d --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/ec04da581d739ba7b6e3e6a23542ce72fd4e6a3953ad5f4e73573d2c1199f81e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:996856d5b2670d9c89d7601578b485eae433dad044ead7d16451e3415aec2577 +size 53758 diff --git a/parse/train/LhY8QdUGSuw/images/ec90b583322ceabb7cc1c372fbe73932285491b6e11d3b83b592d3267593afe4.jpg b/parse/train/LhY8QdUGSuw/images/ec90b583322ceabb7cc1c372fbe73932285491b6e11d3b83b592d3267593afe4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cedcc65e0905b1ebe7f49b9aefce5c5a83007613 --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/ec90b583322ceabb7cc1c372fbe73932285491b6e11d3b83b592d3267593afe4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f04d662e2e6feeba0329d2068662323a3128507c9b67eba5f31693433ec4435b +size 27093 diff --git a/parse/train/LhY8QdUGSuw/images/f60a3bba076bfdddd01c2acb94dc82366ab2a5daff26a28a2f1a02eaa5331c18.jpg b/parse/train/LhY8QdUGSuw/images/f60a3bba076bfdddd01c2acb94dc82366ab2a5daff26a28a2f1a02eaa5331c18.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a68128b2ba577096c639c1cda2f3bc8a2dd82cea --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/f60a3bba076bfdddd01c2acb94dc82366ab2a5daff26a28a2f1a02eaa5331c18.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7bc81ef1160d959bfd99f879c5a7e1d14a07f09724af7bdaa8564594b81107c1 +size 4794 diff --git a/parse/train/LhY8QdUGSuw/images/f7f09ed8fd905a5739a7af98293034d8a59fa72e4a4b35d72db19658113786f8.jpg b/parse/train/LhY8QdUGSuw/images/f7f09ed8fd905a5739a7af98293034d8a59fa72e4a4b35d72db19658113786f8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7e26b5b13b02cb1d43a67685d382c5e416316ffb --- /dev/null +++ b/parse/train/LhY8QdUGSuw/images/f7f09ed8fd905a5739a7af98293034d8a59fa72e4a4b35d72db19658113786f8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b0fd65f20ae888abaa1c862c30aa91bfb5e8cc170e2ddef102ffe5eb996ccf52 +size 47619 diff --git a/parse/train/Lpfh1Bpqfk/images/1b817452fed13ecd8d39700f88900c771e3a9e3f6dfc8bdc649446f86b908a32.jpg b/parse/train/Lpfh1Bpqfk/images/1b817452fed13ecd8d39700f88900c771e3a9e3f6dfc8bdc649446f86b908a32.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ef2bc452d7ce6e1c8efa85c6110514ae092bc079 --- /dev/null +++ b/parse/train/Lpfh1Bpqfk/images/1b817452fed13ecd8d39700f88900c771e3a9e3f6dfc8bdc649446f86b908a32.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:da1473e6cab1ae0cf986de0047deee35a93446d3f97773dbb922c86226347753 +size 44503 diff --git a/parse/train/Lpfh1Bpqfk/images/294e2fe483ed1695691c2b1470b6210c847eef9d58f017019c1c1cfd8b31695c.jpg b/parse/train/Lpfh1Bpqfk/images/294e2fe483ed1695691c2b1470b6210c847eef9d58f017019c1c1cfd8b31695c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8445aafe9f2f05cb17576f67c2af043ea61386c8 --- /dev/null +++ b/parse/train/Lpfh1Bpqfk/images/294e2fe483ed1695691c2b1470b6210c847eef9d58f017019c1c1cfd8b31695c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fd5b6517f3bc1e68cb7c2cce5f85602ebacc984752971b94e264e5e9e951aacc +size 45291 diff --git a/parse/train/Lpfh1Bpqfk/images/3599e6fb185969911a28ae324a67963ed676830a0d957f351ca943e9dc0aad27.jpg b/parse/train/Lpfh1Bpqfk/images/3599e6fb185969911a28ae324a67963ed676830a0d957f351ca943e9dc0aad27.jpg new file mode 100644 index 0000000000000000000000000000000000000000..85f3deec30fe4ba2caf5a2d464cd9678e26dc079 --- /dev/null +++ b/parse/train/Lpfh1Bpqfk/images/3599e6fb185969911a28ae324a67963ed676830a0d957f351ca943e9dc0aad27.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6264ccd429ac9568efa286acf1b3ada917729716c4cd9d35834c8a7066f07747 +size 22245 diff --git a/parse/train/Lpfh1Bpqfk/images/43debc0650f8f893da5ed8e572faf39672229881bcf0855f86dff3d9306f34f9.jpg b/parse/train/Lpfh1Bpqfk/images/43debc0650f8f893da5ed8e572faf39672229881bcf0855f86dff3d9306f34f9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b37f0facbd709ddc52f6abbffce6f08ee62c36b4 --- /dev/null +++ b/parse/train/Lpfh1Bpqfk/images/43debc0650f8f893da5ed8e572faf39672229881bcf0855f86dff3d9306f34f9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aaddffc95c9e3bdd26309ce36dd71399e1b248321d22366bf5f4f8527001afb7 +size 45627 diff --git a/parse/train/Lpfh1Bpqfk/images/79c844634d5f88ad843d0ccbedac33c38a9980eb6213fb6125e5349400d7d0fd.jpg b/parse/train/Lpfh1Bpqfk/images/79c844634d5f88ad843d0ccbedac33c38a9980eb6213fb6125e5349400d7d0fd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..09073eed802add407408964a0e4aa0aa071139e8 --- /dev/null +++ b/parse/train/Lpfh1Bpqfk/images/79c844634d5f88ad843d0ccbedac33c38a9980eb6213fb6125e5349400d7d0fd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ff5e59e10c615dcaa15f1556a6cb638fd9248e2c4465c63c240fa12deea2aab2 +size 19638 diff --git a/parse/train/Lpfh1Bpqfk/images/aba9b861e8efcb57d826b09c432be1e0382b373a7817ab870f1aaced18950672.jpg b/parse/train/Lpfh1Bpqfk/images/aba9b861e8efcb57d826b09c432be1e0382b373a7817ab870f1aaced18950672.jpg new file mode 100644 index 0000000000000000000000000000000000000000..743a379d321fb1d351a7f356432f792f1d1fc973 --- /dev/null +++ b/parse/train/Lpfh1Bpqfk/images/aba9b861e8efcb57d826b09c432be1e0382b373a7817ab870f1aaced18950672.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:deb01745bc849cb077df0c49ff5d5f73a934103c931a87e7d8109e564c18a73c +size 94580 diff --git a/parse/train/Lpfh1Bpqfk/images/b91aae065046d45a4cf2660ee48625a319b82673ffafb868c9b5ca99235d35b4.jpg b/parse/train/Lpfh1Bpqfk/images/b91aae065046d45a4cf2660ee48625a319b82673ffafb868c9b5ca99235d35b4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8d26f185a5e0f398ccf1126c0b9240155fb221ab --- /dev/null +++ b/parse/train/Lpfh1Bpqfk/images/b91aae065046d45a4cf2660ee48625a319b82673ffafb868c9b5ca99235d35b4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:09ee0af5bc2561a497196747e5b59376a97f4018a19ce8de8c9b5076d551d650 +size 67984 diff --git a/parse/train/Lpfh1Bpqfk/images/be011ecde540fa81014ef4ee685adfdc4ac01205383316fdb5d9823fcc557d92.jpg b/parse/train/Lpfh1Bpqfk/images/be011ecde540fa81014ef4ee685adfdc4ac01205383316fdb5d9823fcc557d92.jpg new file mode 100644 index 0000000000000000000000000000000000000000..eeeb965d4b3e32ed0dd9047164ec73eedbd9e3d9 --- /dev/null +++ b/parse/train/Lpfh1Bpqfk/images/be011ecde540fa81014ef4ee685adfdc4ac01205383316fdb5d9823fcc557d92.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:908cdd25b86f0dee8c84bb0e50e17106e6b18964ebf1af8bf36850035d045dbb +size 41974 diff --git a/parse/train/Lpfh1Bpqfk/images/bfee31f4a850d3ac234c45e7f45341c3c4f9311de9e2162c5b1cbf2aa1a80323.jpg b/parse/train/Lpfh1Bpqfk/images/bfee31f4a850d3ac234c45e7f45341c3c4f9311de9e2162c5b1cbf2aa1a80323.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3a54ae3c1d238363509b04236992506538e2d8a8 --- /dev/null +++ b/parse/train/Lpfh1Bpqfk/images/bfee31f4a850d3ac234c45e7f45341c3c4f9311de9e2162c5b1cbf2aa1a80323.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7bd3bc60e464c6d41d2a04e1366340ee7b2a6fdf0fb1b6a1116dc5b9fca94db7 +size 100339 diff --git a/parse/train/OQ08SN70M1V/images/2ea21def8341a6d5bb6864e6f79d9517f4580e401b096db8d0ee0ac2e3f73f28.jpg b/parse/train/OQ08SN70M1V/images/2ea21def8341a6d5bb6864e6f79d9517f4580e401b096db8d0ee0ac2e3f73f28.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f920677803ba28944e6f6d71998bd60189810876 --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/2ea21def8341a6d5bb6864e6f79d9517f4580e401b096db8d0ee0ac2e3f73f28.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0768d8b123f691725821c2d83d94fba64bd2b26197470f1790345a82f47345cf +size 8787 diff --git a/parse/train/OQ08SN70M1V/images/2fd1f374588616f5a6338fb80df2fa377eeaad249ae87e55e98de34831a2b8d7.jpg b/parse/train/OQ08SN70M1V/images/2fd1f374588616f5a6338fb80df2fa377eeaad249ae87e55e98de34831a2b8d7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c6286039abf201f76b07065de885c6cd71c4fe71 --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/2fd1f374588616f5a6338fb80df2fa377eeaad249ae87e55e98de34831a2b8d7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e76a7474d1265d2c4c6c637cce3f0632715fa03ccc61d9399436d533e85da26b +size 19560 diff --git a/parse/train/OQ08SN70M1V/images/3511fee4606ba2827837421454ce087e956fd928e783321225b60a416864189b.jpg b/parse/train/OQ08SN70M1V/images/3511fee4606ba2827837421454ce087e956fd928e783321225b60a416864189b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9e404650f014bc5d9150d38a0b1f6437257ccc8f --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/3511fee4606ba2827837421454ce087e956fd928e783321225b60a416864189b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:da742034bfe5537273f04c97517ca386aee77e35d783b1a3932ea4ef560769ce +size 26165 diff --git a/parse/train/OQ08SN70M1V/images/37670fd23c9cbecdcd786ba8593afd124fa658bb7fb09cda16df93847b075f2e.jpg b/parse/train/OQ08SN70M1V/images/37670fd23c9cbecdcd786ba8593afd124fa658bb7fb09cda16df93847b075f2e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bab50cf681babd7d954178a64f0d8927557d21af --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/37670fd23c9cbecdcd786ba8593afd124fa658bb7fb09cda16df93847b075f2e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:07f6b3a91403605d7054e836716c4e5e256e9c2c368b7ac1e29971b03c1f8407 +size 36075 diff --git a/parse/train/OQ08SN70M1V/images/3f262e0207a2d613ee5571e007a97bbb41774b850765a7b19e1bd2480ee89cd1.jpg b/parse/train/OQ08SN70M1V/images/3f262e0207a2d613ee5571e007a97bbb41774b850765a7b19e1bd2480ee89cd1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..739017eefba48241a2b05ff57c4891d97db4fb85 --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/3f262e0207a2d613ee5571e007a97bbb41774b850765a7b19e1bd2480ee89cd1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bbc1da61c746bcd118cff8feb4eafcfc04e18be1e2d0c978f3c84941f28eb210 +size 76068 diff --git a/parse/train/OQ08SN70M1V/images/4177d22f8d4f0949cdb246c3364fb3e90bccbc0e329278243ade00fa12ef3fb7.jpg b/parse/train/OQ08SN70M1V/images/4177d22f8d4f0949cdb246c3364fb3e90bccbc0e329278243ade00fa12ef3fb7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c6fbcfb4481a1116cf732b4669d6cdbcefe0f65d --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/4177d22f8d4f0949cdb246c3364fb3e90bccbc0e329278243ade00fa12ef3fb7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f8872762241e1c8196c5347b586522aac343d984da5c901d6c38b92c9d1b5f67 +size 24431 diff --git a/parse/train/OQ08SN70M1V/images/43511ecf9b31accee60f77e94452975f56fba70aedbf22929d63c6f6425738c4.jpg b/parse/train/OQ08SN70M1V/images/43511ecf9b31accee60f77e94452975f56fba70aedbf22929d63c6f6425738c4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..48f5be9ff7f2773e13dfe70aeb9b162243c51843 --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/43511ecf9b31accee60f77e94452975f56fba70aedbf22929d63c6f6425738c4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9baf4dfcb0d1a0edec65cd6d2eede17a25ebf2446a5b5e8492a104b96279f2a2 +size 17434 diff --git a/parse/train/OQ08SN70M1V/images/594e744296588bf3dcf22245b9e01e5557adb6cdef2647b631203f8360026ed7.jpg b/parse/train/OQ08SN70M1V/images/594e744296588bf3dcf22245b9e01e5557adb6cdef2647b631203f8360026ed7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8a2bc7ae2bb9a2a276f8aac545676f0999db038a --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/594e744296588bf3dcf22245b9e01e5557adb6cdef2647b631203f8360026ed7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cf4523eb810f24f1ebc440076822001a2f0b746502f21836f92039d400c43d8c +size 14275 diff --git a/parse/train/OQ08SN70M1V/images/6774e4cff1c4da83d09521055cc23e9dfd7d4cc70106a537169f38f3cf8458b6.jpg b/parse/train/OQ08SN70M1V/images/6774e4cff1c4da83d09521055cc23e9dfd7d4cc70106a537169f38f3cf8458b6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d66d4a3888578fd8101d429023671a4dfc70bf2f --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/6774e4cff1c4da83d09521055cc23e9dfd7d4cc70106a537169f38f3cf8458b6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cd0a60c00134f804b4562ec289fbe4c4cd8aaf827e86a575f4751db03f57e295 +size 12481 diff --git a/parse/train/OQ08SN70M1V/images/69464f3a455946400a8a509121c6b33baa09e829cb304a7916256e3a369b8851.jpg b/parse/train/OQ08SN70M1V/images/69464f3a455946400a8a509121c6b33baa09e829cb304a7916256e3a369b8851.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3175c935225f4aa507b57a5ec2277077b7599b30 --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/69464f3a455946400a8a509121c6b33baa09e829cb304a7916256e3a369b8851.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d0fcc707bd380c95be1d44a12b1b698ee5c6e4239a2ad6c6cd189e90b5715241 +size 23061 diff --git a/parse/train/OQ08SN70M1V/images/893bee63e0019ba2b3dc7b697a1735bf0e6abe18d955dea4bece6a016e905297.jpg b/parse/train/OQ08SN70M1V/images/893bee63e0019ba2b3dc7b697a1735bf0e6abe18d955dea4bece6a016e905297.jpg new file mode 100644 index 0000000000000000000000000000000000000000..55c0cdc104e8dea0728e8b4b46953c18312df9b3 --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/893bee63e0019ba2b3dc7b697a1735bf0e6abe18d955dea4bece6a016e905297.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:61a0f72584b3144612bd709d4d917edb64661e4c34d2c0fe825f079998442181 +size 66877 diff --git a/parse/train/OQ08SN70M1V/images/8fd2cf08c095b3cad513923b03b943d104137ff592ee7b9bdf8cbd7e4d6b60ad.jpg b/parse/train/OQ08SN70M1V/images/8fd2cf08c095b3cad513923b03b943d104137ff592ee7b9bdf8cbd7e4d6b60ad.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c7029de2a3ef6d3a12917561ae3d00fa70026aac --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/8fd2cf08c095b3cad513923b03b943d104137ff592ee7b9bdf8cbd7e4d6b60ad.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:13b60761f7903cda8e324729827c25156306d05ae5da515b968b0970bdeeb800 +size 39037 diff --git a/parse/train/OQ08SN70M1V/images/8fda1870165227e2e77570fa34725795ec5aca7f2f0f543a424ff45bba7dff66.jpg b/parse/train/OQ08SN70M1V/images/8fda1870165227e2e77570fa34725795ec5aca7f2f0f543a424ff45bba7dff66.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2723c06d04abb95717309b9b5b995b246b13be1d --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/8fda1870165227e2e77570fa34725795ec5aca7f2f0f543a424ff45bba7dff66.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a7107af3f49c77d4a87504633f93d052df9b73b9aec31f13d7094d9e3595c5e4 +size 19972 diff --git a/parse/train/OQ08SN70M1V/images/90e87e33e0d474d06695f2af910cc47c043552ae544fc77aa914bbcb413ef411.jpg b/parse/train/OQ08SN70M1V/images/90e87e33e0d474d06695f2af910cc47c043552ae544fc77aa914bbcb413ef411.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a8a95bf991ebac48ab545c39cda4008937dd4d5d --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/90e87e33e0d474d06695f2af910cc47c043552ae544fc77aa914bbcb413ef411.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:31695a14fd9117e82354596bc44627dfd6a63a9e3c13387ae5760d7a30220211 +size 8516 diff --git a/parse/train/OQ08SN70M1V/images/a130ff8e9d49ce3d3cb7db2347ba62e120c691b8d785f39944ac15bca8ffc2d2.jpg b/parse/train/OQ08SN70M1V/images/a130ff8e9d49ce3d3cb7db2347ba62e120c691b8d785f39944ac15bca8ffc2d2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2d75be12b94d7aad2ce5aab0a0ab56a6cf0af5c0 --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/a130ff8e9d49ce3d3cb7db2347ba62e120c691b8d785f39944ac15bca8ffc2d2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e0b13bf8635fdbf871faca7e4e8e0197d772ac8d23a0c8e4689d042f6beb7f03 +size 24485 diff --git a/parse/train/OQ08SN70M1V/images/a7cdb7aac9aa80df4b7f6646bcbb166af4c63be13a3665a5e0b8f969bef9bd0e.jpg b/parse/train/OQ08SN70M1V/images/a7cdb7aac9aa80df4b7f6646bcbb166af4c63be13a3665a5e0b8f969bef9bd0e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..69cab75adef893d3e675deaf0a9d20fda07cae5a --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/a7cdb7aac9aa80df4b7f6646bcbb166af4c63be13a3665a5e0b8f969bef9bd0e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:063f4ea3e4153d5c3e74ab4c2228e28a47819824d01aa3c1b72dfdf8502568e9 +size 36860 diff --git a/parse/train/OQ08SN70M1V/images/b2a74912a6e77c090080c27f318b817820e5551294899fd2ff82b47a1a561520.jpg b/parse/train/OQ08SN70M1V/images/b2a74912a6e77c090080c27f318b817820e5551294899fd2ff82b47a1a561520.jpg new file mode 100644 index 0000000000000000000000000000000000000000..10b585d762f545833d12f48a2542d5948fdef5dd --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/b2a74912a6e77c090080c27f318b817820e5551294899fd2ff82b47a1a561520.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:94692ff4ee1fd0e37b6d332d21c8ce6b02f36dcb7c389799380cc9cc25a16569 +size 12934 diff --git a/parse/train/OQ08SN70M1V/images/b75498f64b027bc49151eb52898c56e85290e1f52ba9a95685a719f55ef1e7e1.jpg b/parse/train/OQ08SN70M1V/images/b75498f64b027bc49151eb52898c56e85290e1f52ba9a95685a719f55ef1e7e1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7081716ed643c080d24325da1ef38b77c40527f3 --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/b75498f64b027bc49151eb52898c56e85290e1f52ba9a95685a719f55ef1e7e1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bccb55ec7e8cdc01725c737c19493d8ab4cd46f607e42aac83112626d47ae81c +size 10119 diff --git a/parse/train/OQ08SN70M1V/images/ba65eba19d22be27fae6120c6a2741dddabcc771d717582ec6be8b09bc5360d5.jpg b/parse/train/OQ08SN70M1V/images/ba65eba19d22be27fae6120c6a2741dddabcc771d717582ec6be8b09bc5360d5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..667d4d92d38602331dd99a63dfd765cbef4acd32 --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/ba65eba19d22be27fae6120c6a2741dddabcc771d717582ec6be8b09bc5360d5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:10dec6fd7d5c12026659a3576fed46de6ec55d2bb0964e4384668057779cb866 +size 18732 diff --git a/parse/train/OQ08SN70M1V/images/c46d8d3c93b763f6fa040c0bd9e72f935f05e0f211ae58d3a6befa44120a77c5.jpg b/parse/train/OQ08SN70M1V/images/c46d8d3c93b763f6fa040c0bd9e72f935f05e0f211ae58d3a6befa44120a77c5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e5c2e8f5d054f9986afc8a6858a16ed8d929c731 --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/c46d8d3c93b763f6fa040c0bd9e72f935f05e0f211ae58d3a6befa44120a77c5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dcde9a116b81ea052be3f8a52207a026b04eea99942546f8542bdfe2d8a4da14 +size 31695 diff --git a/parse/train/OQ08SN70M1V/images/d2acfa692511a819cd233bf0c5096e0ee8dac683de5b4402f1cdb17be9fe880e.jpg b/parse/train/OQ08SN70M1V/images/d2acfa692511a819cd233bf0c5096e0ee8dac683de5b4402f1cdb17be9fe880e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c23a7d4056c0f45f0cdfd7b257ed10c99e9473cc --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/d2acfa692511a819cd233bf0c5096e0ee8dac683de5b4402f1cdb17be9fe880e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:14a2cf96d101a77373901cf2bdba669d7fcb900ac63ecd7ec2c577532d27edc3 +size 31462 diff --git a/parse/train/OQ08SN70M1V/images/e160e05dae6ccc1f2aa277bbe18380ae5f55109c05e73b5889f68a2ead471c27.jpg b/parse/train/OQ08SN70M1V/images/e160e05dae6ccc1f2aa277bbe18380ae5f55109c05e73b5889f68a2ead471c27.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b3cda4ec0eab86cd5f7774069fed34319a74cafc --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/e160e05dae6ccc1f2aa277bbe18380ae5f55109c05e73b5889f68a2ead471c27.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ddb982fdbd034a9c67a0fa26ced5930c750978caafd87a310bfbad15027648e1 +size 60876 diff --git a/parse/train/OQ08SN70M1V/images/f260d02863623e0bba432eeb2881331ba8775b04d069015dde91411d3d2cd440.jpg b/parse/train/OQ08SN70M1V/images/f260d02863623e0bba432eeb2881331ba8775b04d069015dde91411d3d2cd440.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e099203ea8e57059e4be37990e7711d180fb2997 --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/f260d02863623e0bba432eeb2881331ba8775b04d069015dde91411d3d2cd440.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:999d4b8cba4e07dc8a858a0a91b2dee36bb21972845c443f0971113d7691055a +size 31453 diff --git a/parse/train/OQ08SN70M1V/images/ff3de0b9df21ac3bc451dd765ca718f8cd8bc7b4f9d9e5dbf8031f05708b58d5.jpg b/parse/train/OQ08SN70M1V/images/ff3de0b9df21ac3bc451dd765ca718f8cd8bc7b4f9d9e5dbf8031f05708b58d5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..94e4fd1e863f89af930e3a1e93dd681a914b5e76 --- /dev/null +++ b/parse/train/OQ08SN70M1V/images/ff3de0b9df21ac3bc451dd765ca718f8cd8bc7b4f9d9e5dbf8031f05708b58d5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:55f8c61eea20eeeeca08e2e81e4e8ca83192e88679163580c77b573d07cfabf6 +size 26919 diff --git a/parse/train/PS3IMnScugk/images/010da7139c478be872214ea2c500b70f5256aa5879a8b45e8f401cb24f16a4cc.jpg b/parse/train/PS3IMnScugk/images/010da7139c478be872214ea2c500b70f5256aa5879a8b45e8f401cb24f16a4cc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..29f7d87b806f608d43e637c381b504e86a1ea433 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/010da7139c478be872214ea2c500b70f5256aa5879a8b45e8f401cb24f16a4cc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e83a34a729304cc0befab729e5b05f224f0acd6cc9da53c021e9977c4f44db67 +size 97688 diff --git a/parse/train/PS3IMnScugk/images/045a2d63b5f48014d567b2002ee574688de0358f4404cb18ec9bbefb7406348b.jpg b/parse/train/PS3IMnScugk/images/045a2d63b5f48014d567b2002ee574688de0358f4404cb18ec9bbefb7406348b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5b94820234d597ebc00aba713abfbbf665c7476d --- /dev/null +++ b/parse/train/PS3IMnScugk/images/045a2d63b5f48014d567b2002ee574688de0358f4404cb18ec9bbefb7406348b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4fcc688523ddde4d878c15acb5bfc302939b2452af19015a842682113ed5f618 +size 4437 diff --git a/parse/train/PS3IMnScugk/images/08a8f779523e582a10bf277d2ee3960d987e0f7b05be29635adbe1ca40e98be4.jpg b/parse/train/PS3IMnScugk/images/08a8f779523e582a10bf277d2ee3960d987e0f7b05be29635adbe1ca40e98be4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c379b0a137b1636e6530e7d2e04910f62875dee7 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/08a8f779523e582a10bf277d2ee3960d987e0f7b05be29635adbe1ca40e98be4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e79c9ad737b36fa6341aaa3e3b16176862afd55626ff4173e75388dd0b2be667 +size 34721 diff --git a/parse/train/PS3IMnScugk/images/0f768d76e2126cd4611f8a7a352d8ed7b0ac2606a51a8a6919774e6bc3b31728.jpg b/parse/train/PS3IMnScugk/images/0f768d76e2126cd4611f8a7a352d8ed7b0ac2606a51a8a6919774e6bc3b31728.jpg new file mode 100644 index 0000000000000000000000000000000000000000..78083ea70bc4894f5a0d8606bf0120c601088950 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/0f768d76e2126cd4611f8a7a352d8ed7b0ac2606a51a8a6919774e6bc3b31728.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:de28354f3e2b41eda9607d9dcc3cfb38ec92d0af421d831649dd87a346b51d75 +size 4162 diff --git a/parse/train/PS3IMnScugk/images/12c7b9b8e1b1ed79a2d9eca7f3c3fe761f3e59809e6fd1aa7e573361ec72d8f8.jpg b/parse/train/PS3IMnScugk/images/12c7b9b8e1b1ed79a2d9eca7f3c3fe761f3e59809e6fd1aa7e573361ec72d8f8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8f64ce8b92fc242a9d06e38a1ed430d5f5b79e9e --- /dev/null +++ b/parse/train/PS3IMnScugk/images/12c7b9b8e1b1ed79a2d9eca7f3c3fe761f3e59809e6fd1aa7e573361ec72d8f8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8390086cb5d044b0dad87a0648bfd9fe15fa167c78db96519e59db63085a3e6e +size 2043 diff --git a/parse/train/PS3IMnScugk/images/146c7bea2310a1887cf4f82018aa78c620bd682bb4aba6629bc8c9bf22269d34.jpg b/parse/train/PS3IMnScugk/images/146c7bea2310a1887cf4f82018aa78c620bd682bb4aba6629bc8c9bf22269d34.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b70f8aa19f4a741e796eec6a0c09539ff57f4fc0 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/146c7bea2310a1887cf4f82018aa78c620bd682bb4aba6629bc8c9bf22269d34.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ddc8eb9bc8c6d4177f6702bf3b6e4e35ac0c016cd01b021b1162257165cec02a +size 6307 diff --git a/parse/train/PS3IMnScugk/images/162f6f9165d5ea95173d62c2cedb7f002d06baefa87e5ed0349f2843bfa0ab0b.jpg b/parse/train/PS3IMnScugk/images/162f6f9165d5ea95173d62c2cedb7f002d06baefa87e5ed0349f2843bfa0ab0b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..69ebff683e6ae009ba4b56c6c546316246decac5 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/162f6f9165d5ea95173d62c2cedb7f002d06baefa87e5ed0349f2843bfa0ab0b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec66493674e4d5fc275677c6c3ef1f628c191d1ab523f9b9ce9ed62e7929cec3 +size 5824 diff --git a/parse/train/PS3IMnScugk/images/220e079d12b825b1b8005126f87823211b0213a7c002e1946a181257a91f2d1a.jpg b/parse/train/PS3IMnScugk/images/220e079d12b825b1b8005126f87823211b0213a7c002e1946a181257a91f2d1a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ac4081c6d2ecc9f7496f4beeb673bcdf72424e4f --- /dev/null +++ b/parse/train/PS3IMnScugk/images/220e079d12b825b1b8005126f87823211b0213a7c002e1946a181257a91f2d1a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:63dbe59aeed32cd81c7d9f4b29f5b82c7baa147f1a6418bf97043bd552e4f183 +size 5245 diff --git a/parse/train/PS3IMnScugk/images/29a86739ecfb868dbaf60a0ad333443a8cc5fb0698034bc6535601f93bf28546.jpg b/parse/train/PS3IMnScugk/images/29a86739ecfb868dbaf60a0ad333443a8cc5fb0698034bc6535601f93bf28546.jpg new file mode 100644 index 0000000000000000000000000000000000000000..44f63c581d2079781810865d3c4f32a1a9ae419c --- /dev/null +++ b/parse/train/PS3IMnScugk/images/29a86739ecfb868dbaf60a0ad333443a8cc5fb0698034bc6535601f93bf28546.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9f6a1b726768df25432b6b6aa07975e124f78c3e9fa7e2987f514524ef1f1ae7 +size 4549 diff --git a/parse/train/PS3IMnScugk/images/2f51659d4dd0cfc1bf8185b7cca4d9a33ae23af60f0088d1fe507032ae576e38.jpg b/parse/train/PS3IMnScugk/images/2f51659d4dd0cfc1bf8185b7cca4d9a33ae23af60f0088d1fe507032ae576e38.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9181e04c7acf389461e3ab87f621b7bebafd404a --- /dev/null +++ b/parse/train/PS3IMnScugk/images/2f51659d4dd0cfc1bf8185b7cca4d9a33ae23af60f0088d1fe507032ae576e38.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:310c0c57d00f045d98359575eaceb207a51d0b30a3c362c393e201ce916fff19 +size 10522 diff --git a/parse/train/PS3IMnScugk/images/332d40e27af37edf8f661099d1ab669b0e7b1a5d2190d648f4e0d5392a51802d.jpg b/parse/train/PS3IMnScugk/images/332d40e27af37edf8f661099d1ab669b0e7b1a5d2190d648f4e0d5392a51802d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..154b62306f3f9541f9f03af32c8ad035e194f37d --- /dev/null +++ b/parse/train/PS3IMnScugk/images/332d40e27af37edf8f661099d1ab669b0e7b1a5d2190d648f4e0d5392a51802d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:99fb578f25fd7dcb60c40264c9319636f39b5d69ffeb1f652ea0f05f572530f5 +size 18816 diff --git a/parse/train/PS3IMnScugk/images/3ece8a8229f1da33042bdb3d39cbf512e18542f1afc7b1f790a3ff7617dd6b74.jpg b/parse/train/PS3IMnScugk/images/3ece8a8229f1da33042bdb3d39cbf512e18542f1afc7b1f790a3ff7617dd6b74.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e94544e5ad3875f726b6434e616947c4223b822f --- /dev/null +++ b/parse/train/PS3IMnScugk/images/3ece8a8229f1da33042bdb3d39cbf512e18542f1afc7b1f790a3ff7617dd6b74.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2d42957cb5c6683780056c645514622f6c0387341504b1c2dcdad6c27e5c7cac +size 29496 diff --git a/parse/train/PS3IMnScugk/images/474c8db76d93d8262f36055e1d92625372a3048edb47abf1873d302463f1fbf9.jpg b/parse/train/PS3IMnScugk/images/474c8db76d93d8262f36055e1d92625372a3048edb47abf1873d302463f1fbf9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2c3296b76cc5463ae431d96012f03da0be855089 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/474c8db76d93d8262f36055e1d92625372a3048edb47abf1873d302463f1fbf9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cc77a2f0db2393e7e035f0259b53a6bc71400e19348f770c9e96392e38e0c5d1 +size 3467 diff --git a/parse/train/PS3IMnScugk/images/54b175f47d069db5adac0206567c61a1e7135b192849a5b1e87b936efccd862c.jpg b/parse/train/PS3IMnScugk/images/54b175f47d069db5adac0206567c61a1e7135b192849a5b1e87b936efccd862c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0291fa25fc6f8aedd37ced08c59231f4cae302f2 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/54b175f47d069db5adac0206567c61a1e7135b192849a5b1e87b936efccd862c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:71d63f47406fac6d017e4e0ce6ab1f90ba452e6997ed78e266626b8ccb43a203 +size 3142 diff --git a/parse/train/PS3IMnScugk/images/54c509cfb3d5eb93829a5ad063905da3888e74e97f4f73b57b638b8a3340d9d7.jpg b/parse/train/PS3IMnScugk/images/54c509cfb3d5eb93829a5ad063905da3888e74e97f4f73b57b638b8a3340d9d7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ceec1a40029ff2642db5cea91758576dbee48b8d --- /dev/null +++ b/parse/train/PS3IMnScugk/images/54c509cfb3d5eb93829a5ad063905da3888e74e97f4f73b57b638b8a3340d9d7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1f08cc017dd3d6785f6d729de5c8048c3ef6be9cebbef2ad13fda2bbecb0e67 +size 4455 diff --git a/parse/train/PS3IMnScugk/images/626e30a50efc6b441df4beb1037310e57545588dda8080e31e52c6be22d4f8fb.jpg b/parse/train/PS3IMnScugk/images/626e30a50efc6b441df4beb1037310e57545588dda8080e31e52c6be22d4f8fb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..791744db37d6a9398f09e26ca8e6998293a231cc --- /dev/null +++ b/parse/train/PS3IMnScugk/images/626e30a50efc6b441df4beb1037310e57545588dda8080e31e52c6be22d4f8fb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:97df05972d7a0dd50f4768ab5cf36ddaa87657297815bb04618167cdae7884f8 +size 6293 diff --git a/parse/train/PS3IMnScugk/images/662b57c7af50cd3e5b506c123f6802f95eda0db40c7caee885e186eb17cc26e1.jpg b/parse/train/PS3IMnScugk/images/662b57c7af50cd3e5b506c123f6802f95eda0db40c7caee885e186eb17cc26e1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7d07019b48f5aa377408517699c0312925d19321 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/662b57c7af50cd3e5b506c123f6802f95eda0db40c7caee885e186eb17cc26e1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:190f7ab6e4c319a581ed092b9551b954c7fc6ace2e260fd645eb112ddfcc5146 +size 108048 diff --git a/parse/train/PS3IMnScugk/images/6691ccb2f6b51ca608c9ef5f4e59bd85c280588c2f6818bdc7d38544557dec0d.jpg b/parse/train/PS3IMnScugk/images/6691ccb2f6b51ca608c9ef5f4e59bd85c280588c2f6818bdc7d38544557dec0d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f1f245451451faebea8728c9ed06d91c0f87f4be --- /dev/null +++ b/parse/train/PS3IMnScugk/images/6691ccb2f6b51ca608c9ef5f4e59bd85c280588c2f6818bdc7d38544557dec0d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1700697fdfae5abf3ceee5855dedf45ae4d3785162e10175a62cbb09bcce143f +size 6921 diff --git a/parse/train/PS3IMnScugk/images/66ab4b1ffcd80ff7d9593390dfe6135171e04306e5f09a3a5faa8f42c94fc1b3.jpg b/parse/train/PS3IMnScugk/images/66ab4b1ffcd80ff7d9593390dfe6135171e04306e5f09a3a5faa8f42c94fc1b3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4c742c84f24af557bbfd367b22734e7201eaef25 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/66ab4b1ffcd80ff7d9593390dfe6135171e04306e5f09a3a5faa8f42c94fc1b3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ac96d8f0c4e469fd61f68b87f1d20e9173d42b23a36bed2b08afe65a60855fbb +size 9676 diff --git a/parse/train/PS3IMnScugk/images/68b380206b57e3dd8f6ef32c29aeb4a19419a811eecf430abaa404a361ddb255.jpg b/parse/train/PS3IMnScugk/images/68b380206b57e3dd8f6ef32c29aeb4a19419a811eecf430abaa404a361ddb255.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4712421cfec4309d515b9b67f3704802c9949bdf --- /dev/null +++ b/parse/train/PS3IMnScugk/images/68b380206b57e3dd8f6ef32c29aeb4a19419a811eecf430abaa404a361ddb255.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:78ee4bd35fd6d5462f032e4a2fd1bde8f17279e4690636bff1afb1fb129ab879 +size 11056 diff --git a/parse/train/PS3IMnScugk/images/69faebc7cdfbc3bcc0527c2fa58e58071182957a8b79c6d37cc24f2019aaafd2.jpg b/parse/train/PS3IMnScugk/images/69faebc7cdfbc3bcc0527c2fa58e58071182957a8b79c6d37cc24f2019aaafd2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f84d578129aa533e8003a679138e42187425ba63 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/69faebc7cdfbc3bcc0527c2fa58e58071182957a8b79c6d37cc24f2019aaafd2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3c375e984b3791558d57afb4cc6e83d0736110b375975c682c77d3ac4c4a73c7 +size 58375 diff --git a/parse/train/PS3IMnScugk/images/7112765bc2f05c5477f8f447e14c90e628b482c79c1d9875a9274439805becf7.jpg b/parse/train/PS3IMnScugk/images/7112765bc2f05c5477f8f447e14c90e628b482c79c1d9875a9274439805becf7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..366bc4d74859204c65f83de10f21cd3b6cf9bc8a --- /dev/null +++ b/parse/train/PS3IMnScugk/images/7112765bc2f05c5477f8f447e14c90e628b482c79c1d9875a9274439805becf7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cf37e46b86b2782d0193849ced8c7ab3e4605207037387b55e37e9e2c134bca9 +size 13528 diff --git a/parse/train/PS3IMnScugk/images/7e0bcd0cbc227086b9bdb8351c119f1dce20e09a3ae9e530042624784f2db1e5.jpg b/parse/train/PS3IMnScugk/images/7e0bcd0cbc227086b9bdb8351c119f1dce20e09a3ae9e530042624784f2db1e5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2de28fab033074d3a3e029c45f45f3d10a43cec7 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/7e0bcd0cbc227086b9bdb8351c119f1dce20e09a3ae9e530042624784f2db1e5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cb60bd26af236ceccb73914cd2847163fb83ef804f1bb45d09c4ee5dfa875d2d +size 9555 diff --git a/parse/train/PS3IMnScugk/images/7e12847269d0964d0b1efc0f572e0991fd11c3929442b9ffdc76e2d562fa87b5.jpg b/parse/train/PS3IMnScugk/images/7e12847269d0964d0b1efc0f572e0991fd11c3929442b9ffdc76e2d562fa87b5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6de9fa47352078d861d7e874468e3565af1ea3df --- /dev/null +++ b/parse/train/PS3IMnScugk/images/7e12847269d0964d0b1efc0f572e0991fd11c3929442b9ffdc76e2d562fa87b5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c76d6657b12cfa597f5a081a07a9d0f15d6eea74d929b1e07cc72224e3a4ca74 +size 13925 diff --git a/parse/train/PS3IMnScugk/images/82923d66569b93052b9e83149baffab4e1327615cf6521cc2be96d79c0ddead4.jpg b/parse/train/PS3IMnScugk/images/82923d66569b93052b9e83149baffab4e1327615cf6521cc2be96d79c0ddead4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5a73d45e074919d5bdfe0abeea4468fea0ddfad0 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/82923d66569b93052b9e83149baffab4e1327615cf6521cc2be96d79c0ddead4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:364b10ae125cbac17dc6479ea732ad8ea094b29a7ebcc13d8984de02b6f9ccb5 +size 33376 diff --git a/parse/train/PS3IMnScugk/images/83f894b6067d421244ee465c4ac93a50877c0ed48aa5bfadb511af893d868b51.jpg b/parse/train/PS3IMnScugk/images/83f894b6067d421244ee465c4ac93a50877c0ed48aa5bfadb511af893d868b51.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4d8aabfbaf0b31ab2b5496ff1f2657735f9901e0 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/83f894b6067d421244ee465c4ac93a50877c0ed48aa5bfadb511af893d868b51.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e639005ed10d881d1f959fce471fb009d51f4b46eeb82447c4619cde1b2bc323 +size 3745 diff --git a/parse/train/PS3IMnScugk/images/85b81ad796a9d2c0f796ffd77939b8bb84d30fdd5e067d94d56600cbc1b925f3.jpg b/parse/train/PS3IMnScugk/images/85b81ad796a9d2c0f796ffd77939b8bb84d30fdd5e067d94d56600cbc1b925f3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2d0b3f9e1f8ada4565064723f31baf59d5b265e1 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/85b81ad796a9d2c0f796ffd77939b8bb84d30fdd5e067d94d56600cbc1b925f3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0cdebf161eec9e56c1dc61a0fd03893fe3ede5c0f553567c4488379c293861cf +size 14160 diff --git a/parse/train/PS3IMnScugk/images/8ffc15e9a7914505e760b122ffd7973cc0a48dc88008d2b9a316e1c18964391b.jpg b/parse/train/PS3IMnScugk/images/8ffc15e9a7914505e760b122ffd7973cc0a48dc88008d2b9a316e1c18964391b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9343321c0087ff87508c5db7919ae4ce18f246d0 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/8ffc15e9a7914505e760b122ffd7973cc0a48dc88008d2b9a316e1c18964391b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d75753663fe16c20ab22d076886fb59df9e3e7a129be32f1f00ffe0059b26499 +size 4931 diff --git a/parse/train/PS3IMnScugk/images/90e5a910bec2b9f29ac8fa294300ce33c5ee9693859101e6bbd4ea305a8d9aec.jpg b/parse/train/PS3IMnScugk/images/90e5a910bec2b9f29ac8fa294300ce33c5ee9693859101e6bbd4ea305a8d9aec.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4b6ab961e60385e047637f9042af25fc5dc0e449 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/90e5a910bec2b9f29ac8fa294300ce33c5ee9693859101e6bbd4ea305a8d9aec.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4828406832ae53ecc0d1c8a856029abbb66080d4e1aadb6618d1eaeadcde9457 +size 29185 diff --git a/parse/train/PS3IMnScugk/images/94d163da51117ea9445486b736592da4ca481bf14dec3a043d80d293aeb1871b.jpg b/parse/train/PS3IMnScugk/images/94d163da51117ea9445486b736592da4ca481bf14dec3a043d80d293aeb1871b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..db959cf82865190f04dbbd3fc863280bc3243dfb --- /dev/null +++ b/parse/train/PS3IMnScugk/images/94d163da51117ea9445486b736592da4ca481bf14dec3a043d80d293aeb1871b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:46ded14a973cfffa009b7a96d64994f86225bb7949ff7fc0444b8c0307b660ec +size 2718 diff --git a/parse/train/PS3IMnScugk/images/958a8df5af88fc54204e7f733ac11126d6b3d19ff1e80798251cd7fdd910bf59.jpg b/parse/train/PS3IMnScugk/images/958a8df5af88fc54204e7f733ac11126d6b3d19ff1e80798251cd7fdd910bf59.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a351f9466b9c144599cc406be93fbf81e727a003 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/958a8df5af88fc54204e7f733ac11126d6b3d19ff1e80798251cd7fdd910bf59.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b6f307704636d14b18cbfa47b26e10f720e11e13dd246386fdf6f82b0e118e39 +size 3515 diff --git a/parse/train/PS3IMnScugk/images/95fcb92747c5b18d2262db980ea02934e8a4d62660ffe5d2065b7d1af227721a.jpg b/parse/train/PS3IMnScugk/images/95fcb92747c5b18d2262db980ea02934e8a4d62660ffe5d2065b7d1af227721a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8ac0f7cbffb16afe6fb4f1c2ecd36c5d5b703064 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/95fcb92747c5b18d2262db980ea02934e8a4d62660ffe5d2065b7d1af227721a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7c4dd23baca3a6f82c82137ba690b572e5aa3e8c5117567a0352107e22539c23 +size 60595 diff --git a/parse/train/PS3IMnScugk/images/9970f9707fb881457e608de94d36acfe1d09f6938fc2a0386b1a7144bbe2f83c.jpg b/parse/train/PS3IMnScugk/images/9970f9707fb881457e608de94d36acfe1d09f6938fc2a0386b1a7144bbe2f83c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6180759a4db8789cc123ad900ddb30e0a0d7ef60 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/9970f9707fb881457e608de94d36acfe1d09f6938fc2a0386b1a7144bbe2f83c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5b375924b272c623cd286b58d16382602d1abc8ad6b7551ceea80fc427f8a887 +size 34886 diff --git a/parse/train/PS3IMnScugk/images/a40678b4b18b53ff2b22c38f02e4e224b1b9132876421e7aa83e8b56e88df4dd.jpg b/parse/train/PS3IMnScugk/images/a40678b4b18b53ff2b22c38f02e4e224b1b9132876421e7aa83e8b56e88df4dd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1e0cf08a46e35b74a4a88ffc54e8566454f913b2 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/a40678b4b18b53ff2b22c38f02e4e224b1b9132876421e7aa83e8b56e88df4dd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a4b2f269c71bad8d7220df5a70c17ff61f7df1e4021caab6ce5edb8865eea0aa +size 6762 diff --git a/parse/train/PS3IMnScugk/images/a62a53ed1a042ae191e2bb11f194aff3be73c3a693ae2eb264979f3cf1442b17.jpg b/parse/train/PS3IMnScugk/images/a62a53ed1a042ae191e2bb11f194aff3be73c3a693ae2eb264979f3cf1442b17.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6e080a458fbae7103d951d8b7fa6f93ee16ee2e0 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/a62a53ed1a042ae191e2bb11f194aff3be73c3a693ae2eb264979f3cf1442b17.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a4ff78f3fee57f22cd33a9daf058a5b4f923b8b7424145f8e0baef555b50252b +size 57370 diff --git a/parse/train/PS3IMnScugk/images/ad30a87018b066837b8706d2093056cab29a45560ebed7942d7158d508134faf.jpg b/parse/train/PS3IMnScugk/images/ad30a87018b066837b8706d2093056cab29a45560ebed7942d7158d508134faf.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d3966cfad26ad80ad492f40cd859ef01082df8cd --- /dev/null +++ b/parse/train/PS3IMnScugk/images/ad30a87018b066837b8706d2093056cab29a45560ebed7942d7158d508134faf.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b67a800e7ef06a99f36407e819331598b0481de6630d4be2952b085b2d29ab2e +size 39090 diff --git a/parse/train/PS3IMnScugk/images/ad74dbbe710eff2c9778c1ae7ae1efb6309c81c4b5cf526e3242d9ca0e36c198.jpg b/parse/train/PS3IMnScugk/images/ad74dbbe710eff2c9778c1ae7ae1efb6309c81c4b5cf526e3242d9ca0e36c198.jpg new file mode 100644 index 0000000000000000000000000000000000000000..56c94a78e42df9205dfdfaa6b2898ca4b9ceac1b --- /dev/null +++ b/parse/train/PS3IMnScugk/images/ad74dbbe710eff2c9778c1ae7ae1efb6309c81c4b5cf526e3242d9ca0e36c198.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:668cb1a59f31649e8d125b3b6acd13d32f7387220a18a67b07c3a903196696b5 +size 6365 diff --git a/parse/train/PS3IMnScugk/images/afec9d43b3c01eaf8327afb034dc715c70356cb08a599c76be0d10c0fb326089.jpg b/parse/train/PS3IMnScugk/images/afec9d43b3c01eaf8327afb034dc715c70356cb08a599c76be0d10c0fb326089.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7b04f8da348fcef91c303d1cfd99707e96f7b6f2 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/afec9d43b3c01eaf8327afb034dc715c70356cb08a599c76be0d10c0fb326089.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ac40f43bea92c1192c81410a0640febedb68df75aca7107f27d36c39a323cc48 +size 31377 diff --git a/parse/train/PS3IMnScugk/images/b22dd5dbf133998bb856503ea8171f273e6ba4f6752644efa0ca64d41f584545.jpg b/parse/train/PS3IMnScugk/images/b22dd5dbf133998bb856503ea8171f273e6ba4f6752644efa0ca64d41f584545.jpg new file mode 100644 index 0000000000000000000000000000000000000000..386ab1346cd03eaaa89230b4338e470f230f4994 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/b22dd5dbf133998bb856503ea8171f273e6ba4f6752644efa0ca64d41f584545.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5171d28450f1e8c755f65d8285b5cb242bf5a28f344866aa50dba30f28de402f +size 32121 diff --git a/parse/train/PS3IMnScugk/images/bd8e83d50057daa52a6aeb460c6675ba869851fd2cadfa2577e40dcecab57f2c.jpg b/parse/train/PS3IMnScugk/images/bd8e83d50057daa52a6aeb460c6675ba869851fd2cadfa2577e40dcecab57f2c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b5e9bdd9df02a35e82ecbb8af684da5adecccdd1 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/bd8e83d50057daa52a6aeb460c6675ba869851fd2cadfa2577e40dcecab57f2c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:430507521d17e1eedafc5ee89d5284779560ce578238a859a0a28cb1d6fdd889 +size 17413 diff --git a/parse/train/PS3IMnScugk/images/c430672e7e4264a2bae1563dab96acc996feae244f8b3a1d384ca4da7ef221cf.jpg b/parse/train/PS3IMnScugk/images/c430672e7e4264a2bae1563dab96acc996feae244f8b3a1d384ca4da7ef221cf.jpg new file mode 100644 index 0000000000000000000000000000000000000000..624b522db3f79f792a649f747962412e64489d81 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/c430672e7e4264a2bae1563dab96acc996feae244f8b3a1d384ca4da7ef221cf.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6ad29a77219ab2676597797984687006e519fc5d32394d6e0d2cd2ed8dcb07e9 +size 7154 diff --git a/parse/train/PS3IMnScugk/images/c450bd975d7b10ccbfa66a55ac7cca904456a57831df63e362fd15e92225b9f4.jpg b/parse/train/PS3IMnScugk/images/c450bd975d7b10ccbfa66a55ac7cca904456a57831df63e362fd15e92225b9f4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0a0c1955a3f3496b917165a136233537b17dbfe8 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/c450bd975d7b10ccbfa66a55ac7cca904456a57831df63e362fd15e92225b9f4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:41161af79d98bc7775f74d50b089f023399b8672ae569051969b55e8ec4f0137 +size 60650 diff --git a/parse/train/PS3IMnScugk/images/cdc57cf2dd1c625bb3a1b2d5b31f9475c3f13ef7b754aac851e8e84ee25d4892.jpg b/parse/train/PS3IMnScugk/images/cdc57cf2dd1c625bb3a1b2d5b31f9475c3f13ef7b754aac851e8e84ee25d4892.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7676800e5e118331e83d0ec18a50b6471f3f80bf --- /dev/null +++ b/parse/train/PS3IMnScugk/images/cdc57cf2dd1c625bb3a1b2d5b31f9475c3f13ef7b754aac851e8e84ee25d4892.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c18452ac6f031c26c25f2bdafdac2b2f1d8bce24469107b1b81e388b36fcd475 +size 9784 diff --git a/parse/train/PS3IMnScugk/images/e8779c4bfa9aa087e00f348d785e14fe5f63941b703cb1121e013569aa9f0615.jpg b/parse/train/PS3IMnScugk/images/e8779c4bfa9aa087e00f348d785e14fe5f63941b703cb1121e013569aa9f0615.jpg new file mode 100644 index 0000000000000000000000000000000000000000..21b13e453e306dfe8064d45bc07bdcba121b26b3 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/e8779c4bfa9aa087e00f348d785e14fe5f63941b703cb1121e013569aa9f0615.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:529f311cef38a843fe405e4e274b2cb293d812a2a92411319a8dda0571f1d125 +size 57502 diff --git a/parse/train/PS3IMnScugk/images/ecd1b3b02c963b3bd88c4d8c468a03ec1470f302eb3aa5516efdb7577a7c4899.jpg b/parse/train/PS3IMnScugk/images/ecd1b3b02c963b3bd88c4d8c468a03ec1470f302eb3aa5516efdb7577a7c4899.jpg new file mode 100644 index 0000000000000000000000000000000000000000..39f7a0142790be6bd7e9543c7a4197e9bb627829 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/ecd1b3b02c963b3bd88c4d8c468a03ec1470f302eb3aa5516efdb7577a7c4899.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:90916eaac5b70d973408d34742b0009eeeed3f65737c538b06dbb72f9ebc09fb +size 3416 diff --git a/parse/train/PS3IMnScugk/images/f74a6d8544bc93ebb3be6b84b48c7083a92cb42a8377e61247c6217c55800b43.jpg b/parse/train/PS3IMnScugk/images/f74a6d8544bc93ebb3be6b84b48c7083a92cb42a8377e61247c6217c55800b43.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7f26925571dda3c20fa0c535c7cef1e4b184e101 --- /dev/null +++ b/parse/train/PS3IMnScugk/images/f74a6d8544bc93ebb3be6b84b48c7083a92cb42a8377e61247c6217c55800b43.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3612508991789a1d0d482ecdfa06183cb00527e66d556fad051dd70d807261f2 +size 13958 diff --git a/parse/train/S1e2agrFvS/images/2acf8955cf4aaedd8c9da2b31685b68eb67de5353925f9d11efdf9f51934d0c8.jpg b/parse/train/S1e2agrFvS/images/2acf8955cf4aaedd8c9da2b31685b68eb67de5353925f9d11efdf9f51934d0c8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a884622425644dcbef1c795ea0f2937deb5279be --- /dev/null +++ b/parse/train/S1e2agrFvS/images/2acf8955cf4aaedd8c9da2b31685b68eb67de5353925f9d11efdf9f51934d0c8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d750ca0dc219c6b28334beb8d5b7264180916e27b64dbaafdedd522f1cc59a99 +size 14459 diff --git a/parse/train/S1e2agrFvS/images/2d03c4923573c2475e8050a17e113b6465a86f2fc231bc3fafd79f18fac64341.jpg b/parse/train/S1e2agrFvS/images/2d03c4923573c2475e8050a17e113b6465a86f2fc231bc3fafd79f18fac64341.jpg new file mode 100644 index 0000000000000000000000000000000000000000..197ee3a1f4ad1ea7e7cf409dcaae068b6e7e50f0 --- /dev/null +++ b/parse/train/S1e2agrFvS/images/2d03c4923573c2475e8050a17e113b6465a86f2fc231bc3fafd79f18fac64341.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:564581d459f0a92b3ee961443d076b47aa3a42551f067f4e2fffc329520b6e4c +size 56517 diff --git a/parse/train/S1e2agrFvS/images/2f793c5ce0265a3a1d55664b3c393f2998c2c133c17e14f469622e5dd162db64.jpg b/parse/train/S1e2agrFvS/images/2f793c5ce0265a3a1d55664b3c393f2998c2c133c17e14f469622e5dd162db64.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6366305ec7364332d7c291b522e10c67b92be20f --- /dev/null +++ b/parse/train/S1e2agrFvS/images/2f793c5ce0265a3a1d55664b3c393f2998c2c133c17e14f469622e5dd162db64.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:07303c3cc9ec3d2e3eb078493f7efc02666db22733f4f133b522ecf1728bd0ba +size 4981 diff --git a/parse/train/S1e2agrFvS/images/3cceca2eb851cdb224264ef2afcbc172f4cd841f89a4eaae077413f6c6cc266c.jpg b/parse/train/S1e2agrFvS/images/3cceca2eb851cdb224264ef2afcbc172f4cd841f89a4eaae077413f6c6cc266c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f7c23010194379b740b0e77ccca92bb27e2e2a4d --- /dev/null +++ b/parse/train/S1e2agrFvS/images/3cceca2eb851cdb224264ef2afcbc172f4cd841f89a4eaae077413f6c6cc266c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:23a3079fb26b7c213e56161ac484ea16886766f938971596603d3f38af7d01cf +size 113367 diff --git a/parse/train/S1e2agrFvS/images/a4ee85f5eaf30f49afc016da1dd340b01b47deffb64fbccbc51df1f457375d55.jpg b/parse/train/S1e2agrFvS/images/a4ee85f5eaf30f49afc016da1dd340b01b47deffb64fbccbc51df1f457375d55.jpg new file mode 100644 index 0000000000000000000000000000000000000000..daed72ecd9f6e539188b53d4227ff6cc272ef21e --- /dev/null +++ b/parse/train/S1e2agrFvS/images/a4ee85f5eaf30f49afc016da1dd340b01b47deffb64fbccbc51df1f457375d55.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5436fbde5b9668dff9744c4db5d37a7adeedc15007c0bdb0e4180e635ec99057 +size 38578 diff --git a/parse/train/S1e2agrFvS/images/a50da05ef1fa5f5cef731ea6796e109d0b06e3db1f5d492457d846817b3f8c04.jpg b/parse/train/S1e2agrFvS/images/a50da05ef1fa5f5cef731ea6796e109d0b06e3db1f5d492457d846817b3f8c04.jpg new file mode 100644 index 0000000000000000000000000000000000000000..39deefdd27c4646cb41a46d4c9e7cd3cc657393f --- /dev/null +++ b/parse/train/S1e2agrFvS/images/a50da05ef1fa5f5cef731ea6796e109d0b06e3db1f5d492457d846817b3f8c04.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f4d450b44df1ddacd9547b0e31ad8f5b9c1e1408e427e08b88cb8fa9b774e3fb +size 13482 diff --git a/parse/train/S1e2agrFvS/images/b20edf4eefe093900d1db18f675ce18542f188cf1b8bf41a9efd32adc45437f2.jpg b/parse/train/S1e2agrFvS/images/b20edf4eefe093900d1db18f675ce18542f188cf1b8bf41a9efd32adc45437f2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f65e39d8c6f3ac42618f2b405bc2410c90b59b50 --- /dev/null +++ b/parse/train/S1e2agrFvS/images/b20edf4eefe093900d1db18f675ce18542f188cf1b8bf41a9efd32adc45437f2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3457dfe185a1c91870e30f4b3240d3226eccb762dfe0c606dbad6b10db361693 +size 22409 diff --git a/parse/train/S1e2agrFvS/images/bc07b7a4a85402e5c889b2df948fc755829e34632354d767c6b92b0f36270023.jpg b/parse/train/S1e2agrFvS/images/bc07b7a4a85402e5c889b2df948fc755829e34632354d767c6b92b0f36270023.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e2d5fdea28ae2f3b8a79a7ab76d3f4635cf71a1d --- /dev/null +++ b/parse/train/S1e2agrFvS/images/bc07b7a4a85402e5c889b2df948fc755829e34632354d767c6b92b0f36270023.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ff6fa4afb2d46d2ee5d5cd0346022b59ae8d1816e181b39054bf9febdc78386e +size 52281 diff --git a/parse/train/S1e2agrFvS/images/c5147d2535904a02b65bb9ae077644ac3276ce0b15eb1bd098dc340959ab2efc.jpg b/parse/train/S1e2agrFvS/images/c5147d2535904a02b65bb9ae077644ac3276ce0b15eb1bd098dc340959ab2efc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c2d73abc03dd5c1cf86bc57d7965fff56cbcc0ec --- /dev/null +++ b/parse/train/S1e2agrFvS/images/c5147d2535904a02b65bb9ae077644ac3276ce0b15eb1bd098dc340959ab2efc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e876d0bac4d82562fea6b80195e73ca16a8408bb6a5072d616e8330adbdbf346 +size 68654 diff --git a/parse/train/S1e2agrFvS/images/c61d14e7b0c200c366eec12f1591a59da7799ffc6add88a853088690d44a1d4f.jpg b/parse/train/S1e2agrFvS/images/c61d14e7b0c200c366eec12f1591a59da7799ffc6add88a853088690d44a1d4f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..30ae3decf531df94b04a4a8d308db23ce7e0b1f7 --- /dev/null +++ b/parse/train/S1e2agrFvS/images/c61d14e7b0c200c366eec12f1591a59da7799ffc6add88a853088690d44a1d4f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8c34ff44708efde0d03be44fcc786ced9aa32690e975d586782fdbb53bacaa51 +size 2523 diff --git a/parse/train/S1e2agrFvS/images/ce312ac5fbd9f1f38c3d5deec518a9345a11fe4249a02a56d9a14e140a50ebb6.jpg b/parse/train/S1e2agrFvS/images/ce312ac5fbd9f1f38c3d5deec518a9345a11fe4249a02a56d9a14e140a50ebb6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c0e876ca6e68da37c353bc0438b3dd58b1666e68 --- /dev/null +++ b/parse/train/S1e2agrFvS/images/ce312ac5fbd9f1f38c3d5deec518a9345a11fe4249a02a56d9a14e140a50ebb6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:446ba16d86c991c83a825dae18be9047d0fa2ecb228ff5ffb290ed98a4de50ee +size 91683 diff --git a/parse/train/S1e2agrFvS/images/e0e4e28fa4b223da922ee12dca56e9cea4bae54924a5ff1629ae623a4ca5a0c1.jpg b/parse/train/S1e2agrFvS/images/e0e4e28fa4b223da922ee12dca56e9cea4bae54924a5ff1629ae623a4ca5a0c1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3ee30eb93c90fd33546cbaae2647b581bee2b2b3 --- /dev/null +++ b/parse/train/S1e2agrFvS/images/e0e4e28fa4b223da922ee12dca56e9cea4bae54924a5ff1629ae623a4ca5a0c1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7465d12f331e6560dee912c702dcd4be75d6ff0e806f8e518ebcaebaf2f8cb98 +size 9897 diff --git a/parse/train/S1e2agrFvS/images/e9b86a51207f60c4a5b78eb78ca25c4b09289a312ce10480fba9e8816aa1ce54.jpg b/parse/train/S1e2agrFvS/images/e9b86a51207f60c4a5b78eb78ca25c4b09289a312ce10480fba9e8816aa1ce54.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ac000dad7ee9edde3bf85e052a51485987287220 --- /dev/null +++ b/parse/train/S1e2agrFvS/images/e9b86a51207f60c4a5b78eb78ca25c4b09289a312ce10480fba9e8816aa1ce54.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5814cddcac7b99fed7977fa5e0f639bb02f4747a8019918ab5bc3b673140f59b +size 24477 diff --git a/parse/train/S1jBcueAb/images/096f60d487483c7d5b650634c27ed583c0e3aff2aadfd68c5fe49e886efeccc8.jpg b/parse/train/S1jBcueAb/images/096f60d487483c7d5b650634c27ed583c0e3aff2aadfd68c5fe49e886efeccc8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a512402cbb8b905c76a84ef6630a4527c08a4f1f --- /dev/null +++ b/parse/train/S1jBcueAb/images/096f60d487483c7d5b650634c27ed583c0e3aff2aadfd68c5fe49e886efeccc8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9c153cbca87646453111f652f5dbe665c27df070237f7dc65390fc657102f4d2 +size 11545 diff --git a/parse/train/S1jBcueAb/images/14240588660b2b20b035db23332090f4ac5cf8118b18f3c08880e2d11a251819.jpg b/parse/train/S1jBcueAb/images/14240588660b2b20b035db23332090f4ac5cf8118b18f3c08880e2d11a251819.jpg new file mode 100644 index 0000000000000000000000000000000000000000..aa79c31eb0c5a2379680a589b8ae1179072f8b40 --- /dev/null +++ b/parse/train/S1jBcueAb/images/14240588660b2b20b035db23332090f4ac5cf8118b18f3c08880e2d11a251819.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cb5d67a6b44788b9e14d83d7f4d5d0497c9835b2d19c81ebe86e00b2b0692e18 +size 9320 diff --git a/parse/train/S1jBcueAb/images/265630d7e6778d0f67f27e0b381c08a82dc1dcb9a049ca663853a04688083584.jpg b/parse/train/S1jBcueAb/images/265630d7e6778d0f67f27e0b381c08a82dc1dcb9a049ca663853a04688083584.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ebff2ff05ebb3f540527195feb71d8bc4ebe3aad --- /dev/null +++ b/parse/train/S1jBcueAb/images/265630d7e6778d0f67f27e0b381c08a82dc1dcb9a049ca663853a04688083584.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4b7a466a48aedf63e207bc6ddaeb05b0f641285b76504ba2d1fdbe06cdc34f55 +size 48344 diff --git a/parse/train/S1jBcueAb/images/2f5613db18a7755d0700164640879e8ed63126ebbbc45b0a2142ee09e60aa167.jpg b/parse/train/S1jBcueAb/images/2f5613db18a7755d0700164640879e8ed63126ebbbc45b0a2142ee09e60aa167.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4fa2af452fe118a16bb782f523903b9c580c805d --- /dev/null +++ b/parse/train/S1jBcueAb/images/2f5613db18a7755d0700164640879e8ed63126ebbbc45b0a2142ee09e60aa167.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e5d6628a9fca21df35d9b0312a97b8d2adaf513878515f589cc33f50d92b94c8 +size 44782 diff --git a/parse/train/S1jBcueAb/images/5f1ca0052fd45db77a485b46fef4da6ac15831f0543a35fc067a3a5d9893f139.jpg b/parse/train/S1jBcueAb/images/5f1ca0052fd45db77a485b46fef4da6ac15831f0543a35fc067a3a5d9893f139.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ab8c9c4e2728e6c1ad0c4585cdc0721dcc2a1d08 --- /dev/null +++ b/parse/train/S1jBcueAb/images/5f1ca0052fd45db77a485b46fef4da6ac15831f0543a35fc067a3a5d9893f139.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:43efcc661b2b37d6163e06246c4f3497c7347e5be098b95a8d014bb3002ba56b +size 31747 diff --git a/parse/train/S1jBcueAb/images/6903f53201c4f8f2e39a9e26417e3fee749fcfb0da41ae4d857ef67ecb2a12b4.jpg b/parse/train/S1jBcueAb/images/6903f53201c4f8f2e39a9e26417e3fee749fcfb0da41ae4d857ef67ecb2a12b4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..aa541507642ee66788343f88569d21ea031465fe --- /dev/null +++ b/parse/train/S1jBcueAb/images/6903f53201c4f8f2e39a9e26417e3fee749fcfb0da41ae4d857ef67ecb2a12b4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8359bdc02600db3bece761b6c22572703479688f761700c851ae380e96c17485 +size 13254 diff --git a/parse/train/S1jBcueAb/images/6cb9ba4b19cf17c53fe1cfe00369a70fc75b4f550021ce22095947f2664f5762.jpg b/parse/train/S1jBcueAb/images/6cb9ba4b19cf17c53fe1cfe00369a70fc75b4f550021ce22095947f2664f5762.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1d08db5f4f8046ad5ad3f14d5f9929e8bb8a5fc2 --- /dev/null +++ b/parse/train/S1jBcueAb/images/6cb9ba4b19cf17c53fe1cfe00369a70fc75b4f550021ce22095947f2664f5762.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:964508df3eab8b29083185931f123acd0e00937196a2399acca2fe006ede075d +size 10138 diff --git a/parse/train/S1jBcueAb/images/81329cba13885812ffca9626d4c0237a4c9e156fafdbf8ddd326bc504dcb5eb3.jpg b/parse/train/S1jBcueAb/images/81329cba13885812ffca9626d4c0237a4c9e156fafdbf8ddd326bc504dcb5eb3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..55e6defaeeb16d28efbf3396dd912d81d596be66 --- /dev/null +++ b/parse/train/S1jBcueAb/images/81329cba13885812ffca9626d4c0237a4c9e156fafdbf8ddd326bc504dcb5eb3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:be2fefa3a3f61084f881a027e152ae8b66525f803c22609bd88cac4de76baeac +size 7635 diff --git a/parse/train/S1jBcueAb/images/9e5ca465402bf330e97dec896d8a04a1f839ca8ca88436251b93bffcdd2be514.jpg b/parse/train/S1jBcueAb/images/9e5ca465402bf330e97dec896d8a04a1f839ca8ca88436251b93bffcdd2be514.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e52c7f464e597ac49c879288575ea83fe282c445 --- /dev/null +++ b/parse/train/S1jBcueAb/images/9e5ca465402bf330e97dec896d8a04a1f839ca8ca88436251b93bffcdd2be514.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:89275f27bc36f0999d0af218c54071b38de2676bbe9786440965cd7b79209fff +size 72402 diff --git a/parse/train/S1jBcueAb/images/b050caeb6b465b5ea80ddb90338c0904fa8bb8fc6fad2e63215d80cdba234d13.jpg b/parse/train/S1jBcueAb/images/b050caeb6b465b5ea80ddb90338c0904fa8bb8fc6fad2e63215d80cdba234d13.jpg new file mode 100644 index 0000000000000000000000000000000000000000..46436abd6f0525ecdab930a6ac6d37a60e260d3e --- /dev/null +++ b/parse/train/S1jBcueAb/images/b050caeb6b465b5ea80ddb90338c0904fa8bb8fc6fad2e63215d80cdba234d13.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e8e659e138dd59d841ba5d1abc27d5bb77a7ae9ddb4c357be09665abb2fb8302 +size 50144 diff --git a/parse/train/S1jBcueAb/images/b0d76960ab69f9e54f66d9e89f79c9853a314c6993b60fcec05f49031e89e3db.jpg b/parse/train/S1jBcueAb/images/b0d76960ab69f9e54f66d9e89f79c9853a314c6993b60fcec05f49031e89e3db.jpg new file mode 100644 index 0000000000000000000000000000000000000000..37a92543d6d26f32b6f90df413195e72e7ca7691 --- /dev/null +++ b/parse/train/S1jBcueAb/images/b0d76960ab69f9e54f66d9e89f79c9853a314c6993b60fcec05f49031e89e3db.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0bffbb6217b963634552d405d69392b71fbab44726e53a782a5638c12f1cec0b +size 3685 diff --git a/parse/train/S1jBcueAb/images/b495667ab27aaa2b7eb086b94e3371cf8428ea0f1988563991dd3fc71286d5a5.jpg b/parse/train/S1jBcueAb/images/b495667ab27aaa2b7eb086b94e3371cf8428ea0f1988563991dd3fc71286d5a5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b178951f5d959f2adc4e0a6eea6cc4b63c851ec4 --- /dev/null +++ b/parse/train/S1jBcueAb/images/b495667ab27aaa2b7eb086b94e3371cf8428ea0f1988563991dd3fc71286d5a5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dc5b8e636291237ade5cedacab263d395a41c2c60c0b221ae2be0e6701ce147d +size 29307 diff --git a/parse/train/S1jBcueAb/images/c4961a01467cac23e89b7714c7576ac311eef678d3b24728c4a31ddfa360b060.jpg b/parse/train/S1jBcueAb/images/c4961a01467cac23e89b7714c7576ac311eef678d3b24728c4a31ddfa360b060.jpg new file mode 100644 index 0000000000000000000000000000000000000000..884eebe822b79b53ed86ea328f8580861cc4651b --- /dev/null +++ b/parse/train/S1jBcueAb/images/c4961a01467cac23e89b7714c7576ac311eef678d3b24728c4a31ddfa360b060.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e75c63acb8638620e5fb3d0182593dee7b526c63b4855a7013dd044450d0168c +size 8881 diff --git a/parse/train/S1jBcueAb/images/e09e3d824403dffc07fae3fedfa1fb15253a4ce044934e37b56c969748832121.jpg b/parse/train/S1jBcueAb/images/e09e3d824403dffc07fae3fedfa1fb15253a4ce044934e37b56c969748832121.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7d1474048c9a52faea040b4805baf74dc0387611 --- /dev/null +++ b/parse/train/S1jBcueAb/images/e09e3d824403dffc07fae3fedfa1fb15253a4ce044934e37b56c969748832121.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dd6f4ef0beed1b75f61d458f6e8cffd489ddce31a4e94c09328c0b5b7c025abb +size 11777 diff --git a/parse/train/S1lTg3RqYQ/images/22e27578fec5748fd0e43db0b302ce57cc07222ad8a48475f4e83ca6ab4ba63b.jpg b/parse/train/S1lTg3RqYQ/images/22e27578fec5748fd0e43db0b302ce57cc07222ad8a48475f4e83ca6ab4ba63b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..05edb20848285b87983c24b84d19fe36a0787bda --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/22e27578fec5748fd0e43db0b302ce57cc07222ad8a48475f4e83ca6ab4ba63b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ce8c22f06e6ca5f98b2f5ded1d8c6bcce8c69f8610d833bcf487684bb1319bd9 +size 77218 diff --git a/parse/train/S1lTg3RqYQ/images/298dc8ed7f297dc94cd3dee847f5f77dac92f671495dad30365f22be3c49d541.jpg b/parse/train/S1lTg3RqYQ/images/298dc8ed7f297dc94cd3dee847f5f77dac92f671495dad30365f22be3c49d541.jpg new file mode 100644 index 0000000000000000000000000000000000000000..76486e123365a2373f1b24088ae876694ad376cb --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/298dc8ed7f297dc94cd3dee847f5f77dac92f671495dad30365f22be3c49d541.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d52da97992a5de4cbe37644bd80282aa1d4f898d5bb79765230167494df81590 +size 199359 diff --git a/parse/train/S1lTg3RqYQ/images/3b7097e4902d8fde4889de7951e6a83cd042ee8c5f57705a2a731f446d2a9bb6.jpg b/parse/train/S1lTg3RqYQ/images/3b7097e4902d8fde4889de7951e6a83cd042ee8c5f57705a2a731f446d2a9bb6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bd7862436390efbcffff8f346b9880b93a6bcc60 --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/3b7097e4902d8fde4889de7951e6a83cd042ee8c5f57705a2a731f446d2a9bb6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1678b761246b62e5f3aa45b0e9cecfbed71866b275522ad6427e637b14adf10 +size 17393 diff --git a/parse/train/S1lTg3RqYQ/images/3bd5fdbe84bf5f75f81df361a673dbbe153ed663d37da56b3bd1a128c8dfef1b.jpg b/parse/train/S1lTg3RqYQ/images/3bd5fdbe84bf5f75f81df361a673dbbe153ed663d37da56b3bd1a128c8dfef1b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cbf899849fac04b7827b98538e08937ba99051c1 --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/3bd5fdbe84bf5f75f81df361a673dbbe153ed663d37da56b3bd1a128c8dfef1b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:04d08ef66aaf01c48aff27e719919f61e6b2c5220d4de45367eda52b825ae250 +size 20836 diff --git a/parse/train/S1lTg3RqYQ/images/3f8eb7f9c4f123491e4431c1aeba212986b3d0a92d5f79dbc28801f38e5566b0.jpg b/parse/train/S1lTg3RqYQ/images/3f8eb7f9c4f123491e4431c1aeba212986b3d0a92d5f79dbc28801f38e5566b0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..174ed8948cf4ba7b1a5b99c01e45772fea4bdc0d --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/3f8eb7f9c4f123491e4431c1aeba212986b3d0a92d5f79dbc28801f38e5566b0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d7070c8fbdd15ab7255efe0c55052d779680f90a1200814083ddeb1fa9346786 +size 54366 diff --git a/parse/train/S1lTg3RqYQ/images/43bfb59280bb7ed2e5618f2bf82ab4613a793f470b9625e014a812cb83c67edd.jpg b/parse/train/S1lTg3RqYQ/images/43bfb59280bb7ed2e5618f2bf82ab4613a793f470b9625e014a812cb83c67edd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5c85a6a7d50a0066650eb8f5dceed649af7832bc --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/43bfb59280bb7ed2e5618f2bf82ab4613a793f470b9625e014a812cb83c67edd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:70352ee0d1a9aa574fee611a5dd6ecc74d48053b5f08a71afc70371ec4f9f885 +size 37802 diff --git a/parse/train/S1lTg3RqYQ/images/4416f5ffbb793bcfed683a06bdb2bfe20c0b2b4760c708b0c51cf9f9152bb262.jpg b/parse/train/S1lTg3RqYQ/images/4416f5ffbb793bcfed683a06bdb2bfe20c0b2b4760c708b0c51cf9f9152bb262.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7dca893fde82dda153f9d0d147a59f91b81c7d2f --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/4416f5ffbb793bcfed683a06bdb2bfe20c0b2b4760c708b0c51cf9f9152bb262.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:92db278463c346f9c6544681e9cd812cfd6a7c71b6a0ceef1fb692f69f2ef8a1 +size 18774 diff --git a/parse/train/S1lTg3RqYQ/images/4b048788adc62d7d71bf31609d0a41f155d81e68a41f5587698d310d2e823047.jpg b/parse/train/S1lTg3RqYQ/images/4b048788adc62d7d71bf31609d0a41f155d81e68a41f5587698d310d2e823047.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f969874aeb57f33316772454dc6c7a470f33fb89 --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/4b048788adc62d7d71bf31609d0a41f155d81e68a41f5587698d310d2e823047.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5102698bac084356ce3d642caaf1d2c4bf685bcb3feb49e176dfba5130a0eb6a +size 51909 diff --git a/parse/train/S1lTg3RqYQ/images/6baed0ba09c7eaab5124bc350bf55a9b6d3b14dcf7e990a86a2f37b17995c358.jpg b/parse/train/S1lTg3RqYQ/images/6baed0ba09c7eaab5124bc350bf55a9b6d3b14dcf7e990a86a2f37b17995c358.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3ad72a4b580ea462cbe01a520da59f33ed154109 --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/6baed0ba09c7eaab5124bc350bf55a9b6d3b14dcf7e990a86a2f37b17995c358.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2f6173ab5f230715b5249135c658a419ab3d883fe6e6f0868ac80e82d2d922dc +size 64459 diff --git a/parse/train/S1lTg3RqYQ/images/783e3eefb7064478f37ce5c262d14a464de3fca5f0ea402373be8b6a842dc40f.jpg b/parse/train/S1lTg3RqYQ/images/783e3eefb7064478f37ce5c262d14a464de3fca5f0ea402373be8b6a842dc40f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..876e533fab72103a998e7b1add8e294693ef52cc --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/783e3eefb7064478f37ce5c262d14a464de3fca5f0ea402373be8b6a842dc40f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:10da481ed2b5dffe52248060cfc77b5ef67096c6932a6836db311850509a130e +size 108949 diff --git a/parse/train/S1lTg3RqYQ/images/7c062e3e7376003b48e4fa9c0878c43b812dc48ed43fd2721df259eda179734c.jpg b/parse/train/S1lTg3RqYQ/images/7c062e3e7376003b48e4fa9c0878c43b812dc48ed43fd2721df259eda179734c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d2e4784e3343b6c2b6af77ac89c2042dad5230c6 --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/7c062e3e7376003b48e4fa9c0878c43b812dc48ed43fd2721df259eda179734c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:efc0cdbfb9ca7d6d734bf39ae9d49ec5815c2c6dd1babda13c68fe8ef37ffa74 +size 20529 diff --git a/parse/train/S1lTg3RqYQ/images/806a10330128c4642238995a73ece7e808153f3179e9e887b006db61486745d3.jpg b/parse/train/S1lTg3RqYQ/images/806a10330128c4642238995a73ece7e808153f3179e9e887b006db61486745d3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4fa401635276685ce40bb1d377b7fd931c67e849 --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/806a10330128c4642238995a73ece7e808153f3179e9e887b006db61486745d3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2abac488d7f458d8dc00ee8a45ee56329684548dcf1fb948dc5dfd623f1fec38 +size 58740 diff --git a/parse/train/S1lTg3RqYQ/images/8c255f29b6738f5a4e4749333c55cd54176a9e123ad20e18f142c198c3d5337d.jpg b/parse/train/S1lTg3RqYQ/images/8c255f29b6738f5a4e4749333c55cd54176a9e123ad20e18f142c198c3d5337d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3653287530de35325e0bc9f36f9e5ade665e29ff --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/8c255f29b6738f5a4e4749333c55cd54176a9e123ad20e18f142c198c3d5337d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:080311a12a9934e72a727494ec697496630327d2e3d4cce703523ca4f68b4dbd +size 161033 diff --git a/parse/train/S1lTg3RqYQ/images/9a30894d84763197609856cc7cf335f63a1fcd39d00846edbba35e4242d8397e.jpg b/parse/train/S1lTg3RqYQ/images/9a30894d84763197609856cc7cf335f63a1fcd39d00846edbba35e4242d8397e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7d498cadbb9ef440fd1a6b87343a492d5a485265 --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/9a30894d84763197609856cc7cf335f63a1fcd39d00846edbba35e4242d8397e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9f41e18132864207a3de8f3f035690180d86bd25ae9026e59e1e2c48fbb5c218 +size 10540 diff --git a/parse/train/S1lTg3RqYQ/images/aceca2d39713d52e9b56bfd19a3b238dfcf6159335a751237e771650df31abd6.jpg b/parse/train/S1lTg3RqYQ/images/aceca2d39713d52e9b56bfd19a3b238dfcf6159335a751237e771650df31abd6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dd18d2e5d359806d6981b87428246aa321560154 --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/aceca2d39713d52e9b56bfd19a3b238dfcf6159335a751237e771650df31abd6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:212a69f74cf5412e8d81aba2a31befa017d277093ce49b4717fff07dc055e1d9 +size 60441 diff --git a/parse/train/S1lTg3RqYQ/images/b593a0ffab605da9880a50878e261fd54ef6b564791c09fd0426fdfb2e900e53.jpg b/parse/train/S1lTg3RqYQ/images/b593a0ffab605da9880a50878e261fd54ef6b564791c09fd0426fdfb2e900e53.jpg new file mode 100644 index 0000000000000000000000000000000000000000..92f700861d38ed412805260226ac2edbf9f1e6c4 --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/b593a0ffab605da9880a50878e261fd54ef6b564791c09fd0426fdfb2e900e53.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c04fdbd4fbe59f6ec6932850e123548710d501c6ddb8f18fbd92f449d6dbbfbe +size 35433 diff --git a/parse/train/S1lTg3RqYQ/images/c3287d83605afc4690e0245845bf660e98efb9949800f64bc46df3c234194d15.jpg b/parse/train/S1lTg3RqYQ/images/c3287d83605afc4690e0245845bf660e98efb9949800f64bc46df3c234194d15.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ad9e57619b91e97b19f8628b32548a8efd84b4b4 --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/c3287d83605afc4690e0245845bf660e98efb9949800f64bc46df3c234194d15.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cc21b13da246c7463f01b9452cc6307a017581457c58d33dd638348159ac0e46 +size 9418 diff --git a/parse/train/S1lTg3RqYQ/images/c84d7108485946bbfb1b9eefe074542f5a96ce423f09c4d0317f98e3ae4000a8.jpg b/parse/train/S1lTg3RqYQ/images/c84d7108485946bbfb1b9eefe074542f5a96ce423f09c4d0317f98e3ae4000a8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..66017eb0931703acadff7d7a72f4c80b58480317 --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/c84d7108485946bbfb1b9eefe074542f5a96ce423f09c4d0317f98e3ae4000a8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2f175b7945da1ad06c213d7decdeed6543607f598211e7d90ffedab6140902c5 +size 161591 diff --git a/parse/train/S1lTg3RqYQ/images/c9167c0e54d2de08ac3a28f18ad012a737b3ca543404ca5250c0df2a7949a24e.jpg b/parse/train/S1lTg3RqYQ/images/c9167c0e54d2de08ac3a28f18ad012a737b3ca543404ca5250c0df2a7949a24e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..918558beda7f879f0cf88fcd230170c779cf4242 --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/c9167c0e54d2de08ac3a28f18ad012a737b3ca543404ca5250c0df2a7949a24e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c6ef4410defee44afc2d78717300f6de95e05e9f1366ad5b6e9f333cc47ef57c +size 10833 diff --git a/parse/train/S1lTg3RqYQ/images/da01de10be4d71dd8a1a6880327bad58d98ccb5ef0714adc1c9540ea03d965ba.jpg b/parse/train/S1lTg3RqYQ/images/da01de10be4d71dd8a1a6880327bad58d98ccb5ef0714adc1c9540ea03d965ba.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7c5db7049064ee93fcf42df2196561c605c9b319 --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/da01de10be4d71dd8a1a6880327bad58d98ccb5ef0714adc1c9540ea03d965ba.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c0c7eb2ce305eb839f0fb95c4dfe7ae6d576df1f25d8bfeccd4c12fcd363d88c +size 55262 diff --git a/parse/train/S1lTg3RqYQ/images/dc4ba95e385e98b2466cbdccb27591943a02e39eac24b26d9ed785d67b4b07c8.jpg b/parse/train/S1lTg3RqYQ/images/dc4ba95e385e98b2466cbdccb27591943a02e39eac24b26d9ed785d67b4b07c8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f9eee2045b5a6b656699687dfd304956be65f68c --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/dc4ba95e385e98b2466cbdccb27591943a02e39eac24b26d9ed785d67b4b07c8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ee76b2364f0dbac560f1dd11263b9cfc686560fe9cc7a19b48468b39b2c9db73 +size 79166 diff --git a/parse/train/S1lTg3RqYQ/images/e3fe2bb0d9b42bd2d153934fa3de5dd440d74bc5f4f64869c719c30d172ab6a6.jpg b/parse/train/S1lTg3RqYQ/images/e3fe2bb0d9b42bd2d153934fa3de5dd440d74bc5f4f64869c719c30d172ab6a6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3476b969359888a102938cc611aec5006f093c01 --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/e3fe2bb0d9b42bd2d153934fa3de5dd440d74bc5f4f64869c719c30d172ab6a6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3ce027f5a5377da8c14dec16b86579f924807096aeb5b7d2f7b3d6e9eb9dbbd1 +size 31673 diff --git a/parse/train/S1lTg3RqYQ/images/e7a9343c42a683e90599a34376e4054b2a7b1aa7bf53d8005b5e66ff7199a5e3.jpg b/parse/train/S1lTg3RqYQ/images/e7a9343c42a683e90599a34376e4054b2a7b1aa7bf53d8005b5e66ff7199a5e3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..aca5a99ac3b909af4492766aece0c75921830e33 --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/e7a9343c42a683e90599a34376e4054b2a7b1aa7bf53d8005b5e66ff7199a5e3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5fffa198f64401f09f6442ea7ff8b0996b321ed6bdc0f74e4e1d449b18d43759 +size 38902 diff --git a/parse/train/S1lTg3RqYQ/images/f0d470b80b18f67225d4476daa5ad3e0221c9b62a81ad643196881c5389f1585.jpg b/parse/train/S1lTg3RqYQ/images/f0d470b80b18f67225d4476daa5ad3e0221c9b62a81ad643196881c5389f1585.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4a0ea5dcc32f11d3a17259c5454401c2ce77e63c --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/f0d470b80b18f67225d4476daa5ad3e0221c9b62a81ad643196881c5389f1585.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b3b026e3cc90045e6014089d17a91c980679060304d3e0679311633c0eee0e74 +size 41491 diff --git a/parse/train/S1lTg3RqYQ/images/f968503a159bb1daee2dd2edf84da3c5972441cac159a79aec9acaefeb012e6b.jpg b/parse/train/S1lTg3RqYQ/images/f968503a159bb1daee2dd2edf84da3c5972441cac159a79aec9acaefeb012e6b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..49540e06915c076c6c81630cbd5d2afa2e81e233 --- /dev/null +++ b/parse/train/S1lTg3RqYQ/images/f968503a159bb1daee2dd2edf84da3c5972441cac159a79aec9acaefeb012e6b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9483783b616d541fe31c5dc057eb1a78b275dd454e46aa5d822936d97e05d715 +size 197362 diff --git a/parse/train/Sklsm20ctX/images/12721f8289fc6e8305dc6c8ba83c8d9e44f50e66bde71016c6b12f52add3bb27.jpg b/parse/train/Sklsm20ctX/images/12721f8289fc6e8305dc6c8ba83c8d9e44f50e66bde71016c6b12f52add3bb27.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0fec4acfa03c27b5fc2f99824b4dcddf46c0b199 --- /dev/null +++ b/parse/train/Sklsm20ctX/images/12721f8289fc6e8305dc6c8ba83c8d9e44f50e66bde71016c6b12f52add3bb27.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:48ff66d464a6dda3dafe4f6c3b7eef08b3982553609179e3e1766b109130c382 +size 76041 diff --git a/parse/train/Sklsm20ctX/images/18ec610471e417495297d3ada254ce66e758df5b2463d05d5b9be2f597a0bd37.jpg b/parse/train/Sklsm20ctX/images/18ec610471e417495297d3ada254ce66e758df5b2463d05d5b9be2f597a0bd37.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b7776b55fa8897671ba3c894af29a33f021c4fdc --- /dev/null +++ b/parse/train/Sklsm20ctX/images/18ec610471e417495297d3ada254ce66e758df5b2463d05d5b9be2f597a0bd37.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:03e39992e71b947370f1bcd6a86a4c6e39153bbfdd178bf77d5b68927209040b +size 81464 diff --git a/parse/train/Sklsm20ctX/images/27e1273067bcfd690c40306cd806da119675b405423efbb3426709f9cc13b946.jpg b/parse/train/Sklsm20ctX/images/27e1273067bcfd690c40306cd806da119675b405423efbb3426709f9cc13b946.jpg new file mode 100644 index 0000000000000000000000000000000000000000..636e1b7fd66c64e6e901a0e5bd9fe84c613ebc71 --- /dev/null +++ b/parse/train/Sklsm20ctX/images/27e1273067bcfd690c40306cd806da119675b405423efbb3426709f9cc13b946.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a5e489f5a7847769494f313e7005f3b843751f956ec18802fc321fe69fc23054 +size 49761 diff --git a/parse/train/Sklsm20ctX/images/2da34ef7b3aa9d33ff804f1b9ecbb628b3534a4c0a0ed55f49dacca81dfcfb39.jpg b/parse/train/Sklsm20ctX/images/2da34ef7b3aa9d33ff804f1b9ecbb628b3534a4c0a0ed55f49dacca81dfcfb39.jpg new file mode 100644 index 0000000000000000000000000000000000000000..467113f3acde43040dc80f506d4e49a28919327c --- /dev/null +++ b/parse/train/Sklsm20ctX/images/2da34ef7b3aa9d33ff804f1b9ecbb628b3534a4c0a0ed55f49dacca81dfcfb39.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5ec7580e0b966ccc81985be5539a033606f91c7dd15c7e1000ff8c2fbdf094d3 +size 9509 diff --git a/parse/train/Sklsm20ctX/images/2e6850d92597d835ba628f97c0f7711ea1835645b432dfbb185d432047b1c797.jpg b/parse/train/Sklsm20ctX/images/2e6850d92597d835ba628f97c0f7711ea1835645b432dfbb185d432047b1c797.jpg new file mode 100644 index 0000000000000000000000000000000000000000..72b7a1651d881fbae01b031075e10793ad6c0b46 --- /dev/null +++ b/parse/train/Sklsm20ctX/images/2e6850d92597d835ba628f97c0f7711ea1835645b432dfbb185d432047b1c797.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e276d6afedbb43a9152df9da7de57ccb33ca56c0f0820eb2457900097b6bd661 +size 6408 diff --git a/parse/train/Sklsm20ctX/images/3e71c35ab545ef7c2cf7fe70aac3c1b801aeaa9d69edeae00e49c9c58f1ca488.jpg b/parse/train/Sklsm20ctX/images/3e71c35ab545ef7c2cf7fe70aac3c1b801aeaa9d69edeae00e49c9c58f1ca488.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b01c1e20de5a523651d89eace92e1da1506c56e4 --- /dev/null +++ b/parse/train/Sklsm20ctX/images/3e71c35ab545ef7c2cf7fe70aac3c1b801aeaa9d69edeae00e49c9c58f1ca488.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b47452587520463e77b8b88d2277e792ce73ba308fc696a2da4fa888d145a748 +size 117408 diff --git a/parse/train/Sklsm20ctX/images/4135d952788d1437a8eb0c338c4eb28bd17da7bdc1e7d7f6e9980d410c2d2423.jpg b/parse/train/Sklsm20ctX/images/4135d952788d1437a8eb0c338c4eb28bd17da7bdc1e7d7f6e9980d410c2d2423.jpg new file mode 100644 index 0000000000000000000000000000000000000000..167a1dad5ca161213c7a5822d080f204d4212b51 --- /dev/null +++ b/parse/train/Sklsm20ctX/images/4135d952788d1437a8eb0c338c4eb28bd17da7bdc1e7d7f6e9980d410c2d2423.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f69327d80d2f7f76e13d9bc8fe44f4b06fcc2073acc14d23210be88d5de5c253 +size 7219 diff --git a/parse/train/Sklsm20ctX/images/41c030bb667512b63e645f4439dd94c158536b6e1285dc124280366913476cd0.jpg b/parse/train/Sklsm20ctX/images/41c030bb667512b63e645f4439dd94c158536b6e1285dc124280366913476cd0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f0a11dd3c10e7591e542664d966dd83efe64b02e --- /dev/null +++ b/parse/train/Sklsm20ctX/images/41c030bb667512b63e645f4439dd94c158536b6e1285dc124280366913476cd0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c2b9d05f3f73143e046703991affbcc7561acbe95e546adf1acbc5ce0d68699e +size 141026 diff --git a/parse/train/Sklsm20ctX/images/61cbf72e039883c96a15b5be262e8304b5f77fd6ec0e534652890592ab47011c.jpg b/parse/train/Sklsm20ctX/images/61cbf72e039883c96a15b5be262e8304b5f77fd6ec0e534652890592ab47011c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..55a7527413ea1d1eb593d606db8d45bac28c4c43 --- /dev/null +++ b/parse/train/Sklsm20ctX/images/61cbf72e039883c96a15b5be262e8304b5f77fd6ec0e534652890592ab47011c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0205ae3ab85baee42ad8cdccfed7e4246af59f5900bea51944c5c2dc37f25820 +size 5785 diff --git a/parse/train/Sklsm20ctX/images/8899fe9d0a1ef3dcaa5d1a5edf516c162affe6dc3ef72029a5a1bf6c42780dfd.jpg b/parse/train/Sklsm20ctX/images/8899fe9d0a1ef3dcaa5d1a5edf516c162affe6dc3ef72029a5a1bf6c42780dfd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0737b4f8f86d95712961e509cff6aa3f29113b2b --- /dev/null +++ b/parse/train/Sklsm20ctX/images/8899fe9d0a1ef3dcaa5d1a5edf516c162affe6dc3ef72029a5a1bf6c42780dfd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:77dc627fe7c08932bab0eed75febed9db5390921a35285ece60c0f7854db73d6 +size 65766 diff --git a/parse/train/Sklsm20ctX/images/a26e6b0f57bdfdac9fcbc75ddb11695d4d3d55360a134a0e1538445f815fff35.jpg b/parse/train/Sklsm20ctX/images/a26e6b0f57bdfdac9fcbc75ddb11695d4d3d55360a134a0e1538445f815fff35.jpg new file mode 100644 index 0000000000000000000000000000000000000000..78dec5985f8fd461f7144184f80df4d70fcec516 --- /dev/null +++ b/parse/train/Sklsm20ctX/images/a26e6b0f57bdfdac9fcbc75ddb11695d4d3d55360a134a0e1538445f815fff35.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e33f60ed4971fde97a8676e1cc938bd4ce4297fb82b5a7e7295130d215540d0f +size 6010 diff --git a/parse/train/Sklsm20ctX/images/af22a8a4a6d72a51da4964219477015bfefdf2041ad4cc321fcd5ca9d94c28e5.jpg b/parse/train/Sklsm20ctX/images/af22a8a4a6d72a51da4964219477015bfefdf2041ad4cc321fcd5ca9d94c28e5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5e2b2cdc9823314846fc5112ce436e351d330633 --- /dev/null +++ b/parse/train/Sklsm20ctX/images/af22a8a4a6d72a51da4964219477015bfefdf2041ad4cc321fcd5ca9d94c28e5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d6efc577e13bfec1c498e61a0972b2fa1063a96dc558bb3fc2393b0ddc950a77 +size 4911 diff --git a/parse/train/Sklsm20ctX/images/b00a3c03ad9575649264d247c87006c0ad2250785265eaac2510a5062afd0688.jpg b/parse/train/Sklsm20ctX/images/b00a3c03ad9575649264d247c87006c0ad2250785265eaac2510a5062afd0688.jpg new file mode 100644 index 0000000000000000000000000000000000000000..abbad8356346b341cb0e8dc4771bbcbbdc662b31 --- /dev/null +++ b/parse/train/Sklsm20ctX/images/b00a3c03ad9575649264d247c87006c0ad2250785265eaac2510a5062afd0688.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b69243fea7a09359afd40d8e91f736992b9ac15c0acd1b3bfbf1242053c054de +size 31596 diff --git a/parse/train/Sklsm20ctX/images/bc3b0c05124a7d82463e3cb60fe4ad4972449d1f1b688d6d0ee1b750ecbd9381.jpg b/parse/train/Sklsm20ctX/images/bc3b0c05124a7d82463e3cb60fe4ad4972449d1f1b688d6d0ee1b750ecbd9381.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f988df95a01d867c91c14b0684dfeffe921205fc --- /dev/null +++ b/parse/train/Sklsm20ctX/images/bc3b0c05124a7d82463e3cb60fe4ad4972449d1f1b688d6d0ee1b750ecbd9381.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2abdd01f39a627dde91ce524adbab9beecd53efe17c4733ea3e6f7992f934b1f +size 6069 diff --git a/parse/train/Sklsm20ctX/images/d1c163e8a3b024f543bf4bfe9dbdd875881fca86d2b95b586bdff97e6458f9e5.jpg b/parse/train/Sklsm20ctX/images/d1c163e8a3b024f543bf4bfe9dbdd875881fca86d2b95b586bdff97e6458f9e5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..910508ff06125a163eb52f9428664152ceb56627 --- /dev/null +++ b/parse/train/Sklsm20ctX/images/d1c163e8a3b024f543bf4bfe9dbdd875881fca86d2b95b586bdff97e6458f9e5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:08576c331335e49437611d07a92ff150ddb38296f9edc4ff199e81316d79008f +size 37778 diff --git a/parse/train/Sklsm20ctX/images/d2eaaf445d9676d09b0abe1f32ca8f84691878086c2f7fde001b9cc9b9f1b7d8.jpg b/parse/train/Sklsm20ctX/images/d2eaaf445d9676d09b0abe1f32ca8f84691878086c2f7fde001b9cc9b9f1b7d8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..505eb342c55772e910f74f8a43312c889a145aca --- /dev/null +++ b/parse/train/Sklsm20ctX/images/d2eaaf445d9676d09b0abe1f32ca8f84691878086c2f7fde001b9cc9b9f1b7d8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:712751877579dca1b524f8851b4e14f96d843713d32c018f33b7b802ee2a23b3 +size 5690 diff --git a/parse/train/Sklsm20ctX/images/d5cc5dff820ac51df2323423b6c4f33228df28e6347716f00a1bea69f7c54c77.jpg b/parse/train/Sklsm20ctX/images/d5cc5dff820ac51df2323423b6c4f33228df28e6347716f00a1bea69f7c54c77.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b913598affd2329db505ef1d8a6fb1c0378cdf31 --- /dev/null +++ b/parse/train/Sklsm20ctX/images/d5cc5dff820ac51df2323423b6c4f33228df28e6347716f00a1bea69f7c54c77.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:92fd10d77b3da15be9439aa37aebee536d96051f8f2ec566cc5e24596788be47 +size 69018 diff --git a/parse/train/Sklsm20ctX/images/e055e6b40d22d0067714f3e76101f2b2ff58f924492920fd2548f0f2112a06b3.jpg b/parse/train/Sklsm20ctX/images/e055e6b40d22d0067714f3e76101f2b2ff58f924492920fd2548f0f2112a06b3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6caabfc12f1322097e593184e4926ba70612fd2c --- /dev/null +++ b/parse/train/Sklsm20ctX/images/e055e6b40d22d0067714f3e76101f2b2ff58f924492920fd2548f0f2112a06b3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:420b6a2b62042fc241aea79b1808d9650ac81b6d471916360d8375fd79c1e019 +size 39077 diff --git a/parse/train/SkwSJ99ex/images/1dd728f0be161b626ecc836f01f2afb41c510edc591edfc7daf3ba6e8c55bf2c.jpg b/parse/train/SkwSJ99ex/images/1dd728f0be161b626ecc836f01f2afb41c510edc591edfc7daf3ba6e8c55bf2c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b8fa060ed10d1db2177edf5a9a0af7637adecf4a --- /dev/null +++ b/parse/train/SkwSJ99ex/images/1dd728f0be161b626ecc836f01f2afb41c510edc591edfc7daf3ba6e8c55bf2c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e43725b889cad923cf37f7f9ce96d9a3bee13362e5cc4273b1165e073142a646 +size 36344 diff --git a/parse/train/SkwSJ99ex/images/1f1ddf26fa52be669022483cffb83542f1cfa511eca225ece8b3d84a65ede7a6.jpg b/parse/train/SkwSJ99ex/images/1f1ddf26fa52be669022483cffb83542f1cfa511eca225ece8b3d84a65ede7a6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..30529a60cfb50b2aaa479af1d5bc16564836607e --- /dev/null +++ b/parse/train/SkwSJ99ex/images/1f1ddf26fa52be669022483cffb83542f1cfa511eca225ece8b3d84a65ede7a6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3304b66869c5e7775864ebd7f4462b98c5cea003b5844e88971ec8941194b443 +size 73720 diff --git a/parse/train/SkwSJ99ex/images/2232d6c7a9bb95fefe8e126740963b772f960b12f8733e34ed88b42a33cac508.jpg b/parse/train/SkwSJ99ex/images/2232d6c7a9bb95fefe8e126740963b772f960b12f8733e34ed88b42a33cac508.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0ffc07388c0824e0c3e8f70f0c56dc34a0723e49 --- /dev/null +++ b/parse/train/SkwSJ99ex/images/2232d6c7a9bb95fefe8e126740963b772f960b12f8733e34ed88b42a33cac508.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d6aca347fbd6d01b0c676453f122e40f0a3f7a7d8db459068a0c3e1542907f8a +size 29475 diff --git a/parse/train/SkwSJ99ex/images/37406ab441ad031a65f36caeaf38f245fa411d9866dee9bc4cf365c4d7eeb93b.jpg b/parse/train/SkwSJ99ex/images/37406ab441ad031a65f36caeaf38f245fa411d9866dee9bc4cf365c4d7eeb93b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..81dc8fa6a5c4d1d20cbfe7845a3a5a35a3d53413 --- /dev/null +++ b/parse/train/SkwSJ99ex/images/37406ab441ad031a65f36caeaf38f245fa411d9866dee9bc4cf365c4d7eeb93b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0a7bdebbb02b78a03f2280e7bde7753ae8311387fcba5e4574b07f80e858db1b +size 27556 diff --git a/parse/train/SkwSJ99ex/images/6241b61a0e308e4634e1a13f0b72079e965ebee536761369466ad5dfe0c377b3.jpg b/parse/train/SkwSJ99ex/images/6241b61a0e308e4634e1a13f0b72079e965ebee536761369466ad5dfe0c377b3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2b28e1381851888f8ba0ad97cbb55f380c9d2045 --- /dev/null +++ b/parse/train/SkwSJ99ex/images/6241b61a0e308e4634e1a13f0b72079e965ebee536761369466ad5dfe0c377b3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b14d901881c62db529737544f69d97f344c547dec9b974eaa77eecacf83b48ed +size 40038 diff --git a/parse/train/SkwSJ99ex/images/a66c975cba5d4e615e95a4508cc5afd27e476a3892759f9f192fb996c1807126.jpg b/parse/train/SkwSJ99ex/images/a66c975cba5d4e615e95a4508cc5afd27e476a3892759f9f192fb996c1807126.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9c0cca6b0ed70eed4a542ead810a5eb4c0b7ff17 --- /dev/null +++ b/parse/train/SkwSJ99ex/images/a66c975cba5d4e615e95a4508cc5afd27e476a3892759f9f192fb996c1807126.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d072ab3d0c9c1d59e9efef23bff59b0d837f8277998b186e1f3581eb08d31311 +size 88344 diff --git a/parse/train/SkwSJ99ex/images/d8164c6a706959db76993732a4ef66ee08ed4c5647d18d36d0ab6a30db6aa1b1.jpg b/parse/train/SkwSJ99ex/images/d8164c6a706959db76993732a4ef66ee08ed4c5647d18d36d0ab6a30db6aa1b1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..21125f9159b035d8a359dde2192e3939aefc6c85 --- /dev/null +++ b/parse/train/SkwSJ99ex/images/d8164c6a706959db76993732a4ef66ee08ed4c5647d18d36d0ab6a30db6aa1b1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5b2b1b3b55c7eaa601f159619b61b3e4915768753b622c0e7f019a0ed7fe1d44 +size 38036 diff --git a/parse/train/SkwSJ99ex/images/e18bc7b5039b4e44669165106dd20156e587bcdaf3601cec15a21cdb02379189.jpg b/parse/train/SkwSJ99ex/images/e18bc7b5039b4e44669165106dd20156e587bcdaf3601cec15a21cdb02379189.jpg new file mode 100644 index 0000000000000000000000000000000000000000..38083269005177f1ab42647ac5d9e36cce1f628f --- /dev/null +++ b/parse/train/SkwSJ99ex/images/e18bc7b5039b4e44669165106dd20156e587bcdaf3601cec15a21cdb02379189.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f3f6d3687adcda609f3308f534a82c8f1795c1ba059be04ce25c1b2396841f59 +size 43961 diff --git a/parse/train/SkwSJ99ex/images/e349fce92fdfb72a714cb6ee669d2383ba986fde7493f9f0ba586fb6020fdd53.jpg b/parse/train/SkwSJ99ex/images/e349fce92fdfb72a714cb6ee669d2383ba986fde7493f9f0ba586fb6020fdd53.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7e82bf7a7458d88af9dc8f089649c80d4b95e0e5 --- /dev/null +++ b/parse/train/SkwSJ99ex/images/e349fce92fdfb72a714cb6ee669d2383ba986fde7493f9f0ba586fb6020fdd53.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9f1390a2e72b10757887ece3331ae78109af6c9b746b7ec6fbce284fc8141476 +size 47255 diff --git a/parse/train/SkwSJ99ex/images/e4c702046262338a8c54f39d9166419ccab025123de8754da3635703438bcd0e.jpg b/parse/train/SkwSJ99ex/images/e4c702046262338a8c54f39d9166419ccab025123de8754da3635703438bcd0e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..69fd9c1de839c733299ca94f6dbdc5c727053b07 --- /dev/null +++ b/parse/train/SkwSJ99ex/images/e4c702046262338a8c54f39d9166419ccab025123de8754da3635703438bcd0e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1a941066ff9ab5faa0c1d7b8fba72b1797016fab10a98738c09db3813749139a +size 92984 diff --git a/parse/train/SkwSJ99ex/images/e6ce7aabc2125f97135287bd78a1a484db929eb0f676a6318071f19623ca1e94.jpg b/parse/train/SkwSJ99ex/images/e6ce7aabc2125f97135287bd78a1a484db929eb0f676a6318071f19623ca1e94.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9ec31822ab72abe52e623e25285efa1d732eedf5 --- /dev/null +++ b/parse/train/SkwSJ99ex/images/e6ce7aabc2125f97135287bd78a1a484db929eb0f676a6318071f19623ca1e94.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8cb3525c51991772d9e89335da9788da79844e8033452baaf71a8b53dcf77d46 +size 55729 diff --git a/parse/train/SkyQWDcex/images/0b16b7a8c7a2a95fd58d5af3cf1b353d4fdd57544f29bb2089675d4c4d36a5f7.jpg b/parse/train/SkyQWDcex/images/0b16b7a8c7a2a95fd58d5af3cf1b353d4fdd57544f29bb2089675d4c4d36a5f7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2967c889dec31dbfe0c7fe8dd39aaa49a2fa3f0a --- /dev/null +++ b/parse/train/SkyQWDcex/images/0b16b7a8c7a2a95fd58d5af3cf1b353d4fdd57544f29bb2089675d4c4d36a5f7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:810b7a581edb19fe4ac8d3bdb8c996ae23e4fb5096244583b9a5316ec74a51f9 +size 6579 diff --git a/parse/train/SkyQWDcex/images/0cb815ffed145e77c26dbee2b825ecc27c2b0c736f46cc384ba19981d91dc427.jpg b/parse/train/SkyQWDcex/images/0cb815ffed145e77c26dbee2b825ecc27c2b0c736f46cc384ba19981d91dc427.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fae117a103dcd32280aa8b9f0af6819874c4a955 --- /dev/null +++ b/parse/train/SkyQWDcex/images/0cb815ffed145e77c26dbee2b825ecc27c2b0c736f46cc384ba19981d91dc427.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7abfffe282d97936601321c3bb5a02c87ce4909115517db27fc6c9a96e4da4ae +size 7541 diff --git a/parse/train/SkyQWDcex/images/358c8e02d093a79a09b23609572923688357f99af6e65cc671b0e93aaafa535a.jpg b/parse/train/SkyQWDcex/images/358c8e02d093a79a09b23609572923688357f99af6e65cc671b0e93aaafa535a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3584627dc297583180bc0f07afa7ae550e611708 --- /dev/null +++ b/parse/train/SkyQWDcex/images/358c8e02d093a79a09b23609572923688357f99af6e65cc671b0e93aaafa535a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cc7019f583c1f6a72fbc80adbcf9fda9eb75072d5f9dea12e9836617a2f40a5c +size 91599 diff --git a/parse/train/SkyQWDcex/images/3ce11218504d958980c5a0b5c0daaf818bc82911e322d6fa70acb4007d2144a2.jpg b/parse/train/SkyQWDcex/images/3ce11218504d958980c5a0b5c0daaf818bc82911e322d6fa70acb4007d2144a2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e3fb01ae1dfb1f5f3b98490e6d5263db4510daec --- /dev/null +++ b/parse/train/SkyQWDcex/images/3ce11218504d958980c5a0b5c0daaf818bc82911e322d6fa70acb4007d2144a2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:619edb5d4d96f095fd9c8bdb482975a1ec79c9c8350198f5e591962d0b2d2ace +size 3651 diff --git a/parse/train/SkyQWDcex/images/40bd473a6c4d23d5744dfd44e70bdc7ab128d466ec8f02a672a5568c37fa5b06.jpg b/parse/train/SkyQWDcex/images/40bd473a6c4d23d5744dfd44e70bdc7ab128d466ec8f02a672a5568c37fa5b06.jpg new file mode 100644 index 0000000000000000000000000000000000000000..be18efe290971bfc2d72fea75f1147fdf773b698 --- /dev/null +++ b/parse/train/SkyQWDcex/images/40bd473a6c4d23d5744dfd44e70bdc7ab128d466ec8f02a672a5568c37fa5b06.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d94af5f73ec2c5a531463ea93520182150668a1eeb340e78d3b8bfe02325f287 +size 42497 diff --git a/parse/train/SkyQWDcex/images/443e7540b0b549b92448eb235d0ec63fdb4aa2fd3b36ebd0089310bedce1b17b.jpg b/parse/train/SkyQWDcex/images/443e7540b0b549b92448eb235d0ec63fdb4aa2fd3b36ebd0089310bedce1b17b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..915c4f07fee726625d91ed0d08d68124f1d94503 --- /dev/null +++ b/parse/train/SkyQWDcex/images/443e7540b0b549b92448eb235d0ec63fdb4aa2fd3b36ebd0089310bedce1b17b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1290c48ba1f28cc8b8795a7185e89715c36391ddb13fcd69702062a642fca8f9 +size 79426 diff --git a/parse/train/SkyQWDcex/images/6275a64c37ccaf482ebb09b70800a184a280d191381c227afb608db82191bef4.jpg b/parse/train/SkyQWDcex/images/6275a64c37ccaf482ebb09b70800a184a280d191381c227afb608db82191bef4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..139b161b1342bd7660a5f47ce1a2a6b6a1737c67 --- /dev/null +++ b/parse/train/SkyQWDcex/images/6275a64c37ccaf482ebb09b70800a184a280d191381c227afb608db82191bef4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d677e6e894af94276f3a09b31cec234e95d3d56700a9fc598884526a793c893c +size 58680 diff --git a/parse/train/SkyQWDcex/images/64373a95a6b4ff1dc501307c9d860e01add56935983c5e41d26ff00e57549626.jpg b/parse/train/SkyQWDcex/images/64373a95a6b4ff1dc501307c9d860e01add56935983c5e41d26ff00e57549626.jpg new file mode 100644 index 0000000000000000000000000000000000000000..291820d91d26cd1c5d2866bc87eb2b9dd091faf6 --- /dev/null +++ b/parse/train/SkyQWDcex/images/64373a95a6b4ff1dc501307c9d860e01add56935983c5e41d26ff00e57549626.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bfd30d82f2e6d285671b7702d4ed11c50992e4459f21aaafe523711f7d592497 +size 51331 diff --git a/parse/train/SkyQWDcex/images/6ae166f7aefcbea5f5ffc410385ced0339c03d0846c74f328f054a446d0680cf.jpg b/parse/train/SkyQWDcex/images/6ae166f7aefcbea5f5ffc410385ced0339c03d0846c74f328f054a446d0680cf.jpg new file mode 100644 index 0000000000000000000000000000000000000000..670be1bd445cfdb5329e1512414e159f7138f9f4 --- /dev/null +++ b/parse/train/SkyQWDcex/images/6ae166f7aefcbea5f5ffc410385ced0339c03d0846c74f328f054a446d0680cf.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2fec405f399d45871f10620c1e9f97829dc683d1645eef9c338f2e7a0ba5fca0 +size 3459 diff --git a/parse/train/SkyQWDcex/images/8b14046121b47717b9b4018151123e141e1a6284fb44a0172e23952e24580e0f.jpg b/parse/train/SkyQWDcex/images/8b14046121b47717b9b4018151123e141e1a6284fb44a0172e23952e24580e0f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1aefe30da6a2e96836222cbfdc706a80ba347092 --- /dev/null +++ b/parse/train/SkyQWDcex/images/8b14046121b47717b9b4018151123e141e1a6284fb44a0172e23952e24580e0f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b1739ae62bc1edd726676ee7f686e8119210cdd2bc10432c46bfe5905fc9d582 +size 17241 diff --git a/parse/train/SkyQWDcex/images/922424c60c0fe833417fcfbfbb071439d8dda400d17ad7f567682baa2ae62cc6.jpg b/parse/train/SkyQWDcex/images/922424c60c0fe833417fcfbfbb071439d8dda400d17ad7f567682baa2ae62cc6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9c66c787aa6d962ccb369c8b53c7d21be0fb45f3 --- /dev/null +++ b/parse/train/SkyQWDcex/images/922424c60c0fe833417fcfbfbb071439d8dda400d17ad7f567682baa2ae62cc6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5c0df0ad6e15c4c5a7843447705debdce70b6c37197e7a581c3900402abef158 +size 14920 diff --git a/parse/train/SkyQWDcex/images/9686604b9aed49808ce0071485952292d569bc37387b96bfc2d21296fba30651.jpg b/parse/train/SkyQWDcex/images/9686604b9aed49808ce0071485952292d569bc37387b96bfc2d21296fba30651.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e6449e4955883428cf45d7264bfdab8c25331ade --- /dev/null +++ b/parse/train/SkyQWDcex/images/9686604b9aed49808ce0071485952292d569bc37387b96bfc2d21296fba30651.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7db986c2cd7f78d7c892e3ca449d46e1e306b5459fddc632f9c5fe076991f70f +size 3568 diff --git a/parse/train/SkyQWDcex/images/9f6f0cf956ecfb81ce23623065fdbe1e81f6922ac6b507e2fd60d3e235b3fc56.jpg b/parse/train/SkyQWDcex/images/9f6f0cf956ecfb81ce23623065fdbe1e81f6922ac6b507e2fd60d3e235b3fc56.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bc95f4f3281b6710e82e6722a5d0e9c22b91db9d --- /dev/null +++ b/parse/train/SkyQWDcex/images/9f6f0cf956ecfb81ce23623065fdbe1e81f6922ac6b507e2fd60d3e235b3fc56.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8233994c5b10339d1d30d2939d9bd6d31d1a14ca65570fcd78ea1336620086a6 +size 97120 diff --git a/parse/train/SkyQWDcex/images/a0195b5f73bcc63786ad17f5e39faa709bf3440a0ba3e69fbe7e1e76631c3810.jpg b/parse/train/SkyQWDcex/images/a0195b5f73bcc63786ad17f5e39faa709bf3440a0ba3e69fbe7e1e76631c3810.jpg new file mode 100644 index 0000000000000000000000000000000000000000..49a39a19d503dbcb6f8a041e6117aa5bb9d1ab4a --- /dev/null +++ b/parse/train/SkyQWDcex/images/a0195b5f73bcc63786ad17f5e39faa709bf3440a0ba3e69fbe7e1e76631c3810.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eb50981df00d149fcad9b2f9a4ef0c0884ceca94933e514ae4c8d481ab30382c +size 6140 diff --git a/parse/train/SkyQWDcex/images/a327e6af70ecaa851b2ee8dd017a463fa688552a7375b4f522de51f178b11d0e.jpg b/parse/train/SkyQWDcex/images/a327e6af70ecaa851b2ee8dd017a463fa688552a7375b4f522de51f178b11d0e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3fb8ea9b716a82dd880b3bd8ad65f230d7e06758 --- /dev/null +++ b/parse/train/SkyQWDcex/images/a327e6af70ecaa851b2ee8dd017a463fa688552a7375b4f522de51f178b11d0e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6d50eccd79d26833f104b0de44cc80f60293772090af1e7e816b905c91283261 +size 47000 diff --git a/parse/train/SkyQWDcex/images/a44c27c70891d09259aaf0e5293d2e6e4a96214dff1a5db61bcf19d9179e566a.jpg b/parse/train/SkyQWDcex/images/a44c27c70891d09259aaf0e5293d2e6e4a96214dff1a5db61bcf19d9179e566a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b53065667b95754291aa8fe5b2054c55bd89ba86 --- /dev/null +++ b/parse/train/SkyQWDcex/images/a44c27c70891d09259aaf0e5293d2e6e4a96214dff1a5db61bcf19d9179e566a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:eb8644972a951baa0f2f6fb01978e58a7f39850ed33deffa0bcdf9bb550acf8a +size 7893 diff --git a/parse/train/SkyQWDcex/images/ad6ede4c560264dcb9171609f4a5221b67742d04de0f6a5dd1ad63eb3a74dadf.jpg b/parse/train/SkyQWDcex/images/ad6ede4c560264dcb9171609f4a5221b67742d04de0f6a5dd1ad63eb3a74dadf.jpg new file mode 100644 index 0000000000000000000000000000000000000000..03fb0cdbba37a1675a9dc4b076d13c6d61bc91ee --- /dev/null +++ b/parse/train/SkyQWDcex/images/ad6ede4c560264dcb9171609f4a5221b67742d04de0f6a5dd1ad63eb3a74dadf.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e7651415d2e18c644dd8a0c5c785c5eebf66b6af521406f810cbb15d918fd593 +size 41230 diff --git a/parse/train/SkyQWDcex/images/b4da7d441f7fa0fd9474851fc23f000187e261367b2030efcb5b76c398e507db.jpg b/parse/train/SkyQWDcex/images/b4da7d441f7fa0fd9474851fc23f000187e261367b2030efcb5b76c398e507db.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3a1ab48bfe25b90ebcf7d41751298a2ef4d3d9f8 --- /dev/null +++ b/parse/train/SkyQWDcex/images/b4da7d441f7fa0fd9474851fc23f000187e261367b2030efcb5b76c398e507db.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:516c05621f8b1e70a6b62ee8e7d46159a7266c8b5e6a1b015528e1280ca95e7b +size 28596 diff --git a/parse/train/SkyQWDcex/images/cfe138329014b3f7b463a3e582d1d8ef599e89f8412de5bf5a422be2d6512be2.jpg b/parse/train/SkyQWDcex/images/cfe138329014b3f7b463a3e582d1d8ef599e89f8412de5bf5a422be2d6512be2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8ee59b7af6f84d942f04c906ca0a1016d7907973 --- /dev/null +++ b/parse/train/SkyQWDcex/images/cfe138329014b3f7b463a3e582d1d8ef599e89f8412de5bf5a422be2d6512be2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:319148c7483381aec4d5483b5a7096d7cfc7a6e269ac8891e7cec0a475424a36 +size 2969 diff --git a/parse/train/SkyQWDcex/images/ee9293ea751649f654b58f07c93ff0bf2e388a7a07b62395b5755a135fa9ec7c.jpg b/parse/train/SkyQWDcex/images/ee9293ea751649f654b58f07c93ff0bf2e388a7a07b62395b5755a135fa9ec7c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3af53cff6c584904f375a6aff4b70cd030d2cafe --- /dev/null +++ b/parse/train/SkyQWDcex/images/ee9293ea751649f654b58f07c93ff0bf2e388a7a07b62395b5755a135fa9ec7c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:492c6ada1842116d77790c04d3b449348c4a56468c900e23aa862c7fb3375bd6 +size 2907 diff --git a/parse/train/SkyQWDcex/images/f79f6de5ce49e340d423007b9263cab353bd87acab8e16df2db04716cf012cd5.jpg b/parse/train/SkyQWDcex/images/f79f6de5ce49e340d423007b9263cab353bd87acab8e16df2db04716cf012cd5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..083dcb3875352de3483ed0cc8e4e2700b99fdf6f --- /dev/null +++ b/parse/train/SkyQWDcex/images/f79f6de5ce49e340d423007b9263cab353bd87acab8e16df2db04716cf012cd5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5a22de617d361098c1dff7260664b151b6336af61c853adbf22825c79bed1905 +size 3403 diff --git a/parse/train/SkyQWDcex/images/fe2b61c31ee572d9b5361039bd594ece6636a6b9d48463fdd741489f8e216306.jpg b/parse/train/SkyQWDcex/images/fe2b61c31ee572d9b5361039bd594ece6636a6b9d48463fdd741489f8e216306.jpg new file mode 100644 index 0000000000000000000000000000000000000000..995288afe7f36810f7b785c44bf0fef215d07a88 --- /dev/null +++ b/parse/train/SkyQWDcex/images/fe2b61c31ee572d9b5361039bd594ece6636a6b9d48463fdd741489f8e216306.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3a26205e4832e9e86122bbd47ed0c1793506879d3b7d9c44aa80651407ac0408 +size 86299 diff --git a/parse/train/SkyQWDcex/images/ffaeaf501eba6b9fd2998327d2b2287ed17682571e7b437813e1507343346bb5.jpg b/parse/train/SkyQWDcex/images/ffaeaf501eba6b9fd2998327d2b2287ed17682571e7b437813e1507343346bb5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..74bb7dfd6189f9efe98da1138a8c41fa458350f6 --- /dev/null +++ b/parse/train/SkyQWDcex/images/ffaeaf501eba6b9fd2998327d2b2287ed17682571e7b437813e1507343346bb5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b5b29573d49dc99706903438856926025d8e6ba12192ac8aabee242199cd201e +size 3263 diff --git a/parse/train/Syhr6pxCW/images/314099b9c2383b70f21f616c4872ba44803aada1316f1a7031c781ae951d54bd.jpg b/parse/train/Syhr6pxCW/images/314099b9c2383b70f21f616c4872ba44803aada1316f1a7031c781ae951d54bd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b6929248f22851b63e42e22b65668226f487f8c6 --- /dev/null +++ b/parse/train/Syhr6pxCW/images/314099b9c2383b70f21f616c4872ba44803aada1316f1a7031c781ae951d54bd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:be6153fac9f8ab3ebb09ae1d03b3d723bc749c36ed608499813a6097a5ccc00c +size 53929 diff --git a/parse/train/Syhr6pxCW/images/3144b1e738aeb6bf7da809b6afaa1ba121cfb893080b49bf94ddf8a09de676b9.jpg b/parse/train/Syhr6pxCW/images/3144b1e738aeb6bf7da809b6afaa1ba121cfb893080b49bf94ddf8a09de676b9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..83c07ca0fbbf65258716e00a0781dd857d957440 --- /dev/null +++ b/parse/train/Syhr6pxCW/images/3144b1e738aeb6bf7da809b6afaa1ba121cfb893080b49bf94ddf8a09de676b9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c45f7adabd06711101c8df701f7860422d9dbd5463b2a9efd02a5163e51d8fd2 +size 11291 diff --git a/parse/train/Syhr6pxCW/images/32e01ff75a1e6ff1ac336d4b14e3ac4cf819a4220fe18af4b8a6fb853792b996.jpg b/parse/train/Syhr6pxCW/images/32e01ff75a1e6ff1ac336d4b14e3ac4cf819a4220fe18af4b8a6fb853792b996.jpg new file mode 100644 index 0000000000000000000000000000000000000000..89455855e33ab0cd12220a1b69208a126d952ed5 --- /dev/null +++ b/parse/train/Syhr6pxCW/images/32e01ff75a1e6ff1ac336d4b14e3ac4cf819a4220fe18af4b8a6fb853792b996.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2d6ffa04acf2d50e0d3d7ffd69e37d103b8dab5d1b3ed1d7de36fdd5d78ef4d0 +size 34390 diff --git a/parse/train/Syhr6pxCW/images/34433780414fdf8253053f09c62de4cfb93be4e6011215d36265d5ceeb90f64a.jpg b/parse/train/Syhr6pxCW/images/34433780414fdf8253053f09c62de4cfb93be4e6011215d36265d5ceeb90f64a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cb4e578d511fa828d44ac07be99898394f88572d --- /dev/null +++ b/parse/train/Syhr6pxCW/images/34433780414fdf8253053f09c62de4cfb93be4e6011215d36265d5ceeb90f64a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d03fc2a4cfdada7bbae9f5b05594a1052a62d3e897204d2e4710679156e82056 +size 46097 diff --git a/parse/train/Syhr6pxCW/images/5b72414041bcc9c09e54351f07c329d7e9a0f3940b4e99356b12cb6dfb97612e.jpg b/parse/train/Syhr6pxCW/images/5b72414041bcc9c09e54351f07c329d7e9a0f3940b4e99356b12cb6dfb97612e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2375df2303bb28dece60ceb897cf320a589af64a --- /dev/null +++ b/parse/train/Syhr6pxCW/images/5b72414041bcc9c09e54351f07c329d7e9a0f3940b4e99356b12cb6dfb97612e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:25064db85d4f20a5ee57867988661dae7ba0f00c3fb075b384c08950f9710e83 +size 83197 diff --git a/parse/train/Syhr6pxCW/images/5e4b9c11ad2454d91e8a83dfceb66f02de1ec6de487174608c9894f3525df3b7.jpg b/parse/train/Syhr6pxCW/images/5e4b9c11ad2454d91e8a83dfceb66f02de1ec6de487174608c9894f3525df3b7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cbcb8d63213e0995cc694acd187ca13cb6157118 --- /dev/null +++ b/parse/train/Syhr6pxCW/images/5e4b9c11ad2454d91e8a83dfceb66f02de1ec6de487174608c9894f3525df3b7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:af56ffbb3f15327362a82151654d3e47ad4197b27722be37a031f5c55583fe91 +size 62432 diff --git a/parse/train/Syhr6pxCW/images/5e82f8c24138fb7b5f3c36ba1c13ec82f08d69b345016495485b8b3119bcd69e.jpg b/parse/train/Syhr6pxCW/images/5e82f8c24138fb7b5f3c36ba1c13ec82f08d69b345016495485b8b3119bcd69e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..494210e83f8962a08a319adf2f801aaa6c2a6d58 --- /dev/null +++ b/parse/train/Syhr6pxCW/images/5e82f8c24138fb7b5f3c36ba1c13ec82f08d69b345016495485b8b3119bcd69e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6a3b219628344d1687bf2653247f955170c9ef9feb9043297e241e4a1b436f5b +size 37871 diff --git a/parse/train/Syhr6pxCW/images/6cadc4316fc48b462eb4e5fec1d4b1bdc49d53e92ccdc02ae28eb1ea89aaea3b.jpg b/parse/train/Syhr6pxCW/images/6cadc4316fc48b462eb4e5fec1d4b1bdc49d53e92ccdc02ae28eb1ea89aaea3b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..524b43713b1c53ad5afaa04e1ac6df6d9b90fe7f --- /dev/null +++ b/parse/train/Syhr6pxCW/images/6cadc4316fc48b462eb4e5fec1d4b1bdc49d53e92ccdc02ae28eb1ea89aaea3b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:11dec78dd89c7b95713714eb0542fb2a8963b9e5df5b45d843b9902928881fa6 +size 4813 diff --git a/parse/train/Syhr6pxCW/images/7a60c70be503894a96230d770692694874af94ae2f0421246d7855aa77ee1529.jpg b/parse/train/Syhr6pxCW/images/7a60c70be503894a96230d770692694874af94ae2f0421246d7855aa77ee1529.jpg new file mode 100644 index 0000000000000000000000000000000000000000..573a42a08755b2f4ed4c86611e6d4309b5b11c73 --- /dev/null +++ b/parse/train/Syhr6pxCW/images/7a60c70be503894a96230d770692694874af94ae2f0421246d7855aa77ee1529.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3fd6373f3e8be2a57cf926b93669d645c9f28f87e1f2e45bc1f22b146b5cf6e8 +size 60948 diff --git a/parse/train/Syhr6pxCW/images/8297d793c6dcceaaeacf53bfb02de91e2c75d613ca0c3ec26539870b4ec845a0.jpg b/parse/train/Syhr6pxCW/images/8297d793c6dcceaaeacf53bfb02de91e2c75d613ca0c3ec26539870b4ec845a0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..13f31ca0354ffcc786a8b56aff07475fe49258bd --- /dev/null +++ b/parse/train/Syhr6pxCW/images/8297d793c6dcceaaeacf53bfb02de91e2c75d613ca0c3ec26539870b4ec845a0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d365e0bf559126c08b89ec8645c7e6ae765fab5f85a8fcda9af19cac4b66ca71 +size 10190 diff --git a/parse/train/Syhr6pxCW/images/82b6530a39c59797d49f4600bdd196a0e71c5c25b1644e46bf1a210a7e6f8387.jpg b/parse/train/Syhr6pxCW/images/82b6530a39c59797d49f4600bdd196a0e71c5c25b1644e46bf1a210a7e6f8387.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dcc6b711f9790fc9950cd760e2635d794f242d0f --- /dev/null +++ b/parse/train/Syhr6pxCW/images/82b6530a39c59797d49f4600bdd196a0e71c5c25b1644e46bf1a210a7e6f8387.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f83e4cccc71396b914f9af3c944af0ba5240995acc0e7a142bb2a463104a77bd +size 78934 diff --git a/parse/train/Syhr6pxCW/images/89b3472a7e998f2b7a3992225707439ef7cbf865dd2a049fe67462955265e2de.jpg b/parse/train/Syhr6pxCW/images/89b3472a7e998f2b7a3992225707439ef7cbf865dd2a049fe67462955265e2de.jpg new file mode 100644 index 0000000000000000000000000000000000000000..55e3fcc4bf48a969161d98518eeb8bfa232a540a --- /dev/null +++ b/parse/train/Syhr6pxCW/images/89b3472a7e998f2b7a3992225707439ef7cbf865dd2a049fe67462955265e2de.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ff235b04c19aea74a753896f5cfa46a72d743e0308ddda970e71118eabc1209a +size 72598 diff --git a/parse/train/Syhr6pxCW/images/95056bee38f5b7ea90b52c8dd1613acc7f37e0f26d57c6b497b26436b4a73917.jpg b/parse/train/Syhr6pxCW/images/95056bee38f5b7ea90b52c8dd1613acc7f37e0f26d57c6b497b26436b4a73917.jpg new file mode 100644 index 0000000000000000000000000000000000000000..354f214383136c6c30a486419cbdb23d1060ab8d --- /dev/null +++ b/parse/train/Syhr6pxCW/images/95056bee38f5b7ea90b52c8dd1613acc7f37e0f26d57c6b497b26436b4a73917.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:458f6b65db0bc7df0b36024c192b479b383f146b12c2288570b854350003e7a1 +size 35709 diff --git a/parse/train/Syhr6pxCW/images/de3efea17a40c187b7149f3e254e17895474be28324dd201a78ebda0bcc55b5f.jpg b/parse/train/Syhr6pxCW/images/de3efea17a40c187b7149f3e254e17895474be28324dd201a78ebda0bcc55b5f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..025adccb5667fa7b3dab69fed29a8e5cc4b05acb --- /dev/null +++ b/parse/train/Syhr6pxCW/images/de3efea17a40c187b7149f3e254e17895474be28324dd201a78ebda0bcc55b5f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:affc3dfbf7a7121353f81e2013e8757976c6c71c0f36e1592478515586f882ae +size 35957 diff --git a/parse/train/Syhr6pxCW/images/e9178537b132ca5c7eaaba1a7ffa0fd6072239f63e9ab38ffa1eb9e504681538.jpg b/parse/train/Syhr6pxCW/images/e9178537b132ca5c7eaaba1a7ffa0fd6072239f63e9ab38ffa1eb9e504681538.jpg new file mode 100644 index 0000000000000000000000000000000000000000..940b6a337f72c7ab8bcd8157bb2e97292cdda449 --- /dev/null +++ b/parse/train/Syhr6pxCW/images/e9178537b132ca5c7eaaba1a7ffa0fd6072239f63e9ab38ffa1eb9e504681538.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6cd4d90afae16093893d68369d2a6b6792d22437578088202fecbe9477e7b1cd +size 56125 diff --git a/parse/train/Syhr6pxCW/images/ecc8b7ee6bb4092a7a40e66b4e980d1de557f29dd6a223d35ab3c1a23e9b7d9e.jpg b/parse/train/Syhr6pxCW/images/ecc8b7ee6bb4092a7a40e66b4e980d1de557f29dd6a223d35ab3c1a23e9b7d9e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1f569e83bae21e9b80cf488b1f4d8af4e30ddc97 --- /dev/null +++ b/parse/train/Syhr6pxCW/images/ecc8b7ee6bb4092a7a40e66b4e980d1de557f29dd6a223d35ab3c1a23e9b7d9e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:560bb5aa11bb402a32cbf1e5cb9d984aedac9816bd6b18a1f8d030db923393ad +size 33778 diff --git a/parse/train/Syhr6pxCW/images/fa8d7ab945c8757c417464b92f7768ea40b982fae47ff79da4f2038a269f6351.jpg b/parse/train/Syhr6pxCW/images/fa8d7ab945c8757c417464b92f7768ea40b982fae47ff79da4f2038a269f6351.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8fb92cb2701feef849d4ab3b49294609dcad72bf --- /dev/null +++ b/parse/train/Syhr6pxCW/images/fa8d7ab945c8757c417464b92f7768ea40b982fae47ff79da4f2038a269f6351.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b1fb87e0e13e8c5a6d855de509a5097de7a9a6bcd45b63b1decd2462cfc66cbf +size 81391 diff --git a/parse/train/Syhr6pxCW/images/fef432b657129aff7ba768e90d0567485a1736ee78d2d0d40ece24e13adb5fb3.jpg b/parse/train/Syhr6pxCW/images/fef432b657129aff7ba768e90d0567485a1736ee78d2d0d40ece24e13adb5fb3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..29e2dd53a2425dcec4ebd10252d9edb83d7e6d44 --- /dev/null +++ b/parse/train/Syhr6pxCW/images/fef432b657129aff7ba768e90d0567485a1736ee78d2d0d40ece24e13adb5fb3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4d307ecf72a83fe513f0ec700f68dbdb4c4d9dadba0203f55bae40b33971bf28 +size 63809 diff --git a/parse/train/SyxAb30cY7/images/0035a01d3a4b6f29573e9a043351ba7723e0dc16e9bb0c6baa5485c16e2257d1.jpg b/parse/train/SyxAb30cY7/images/0035a01d3a4b6f29573e9a043351ba7723e0dc16e9bb0c6baa5485c16e2257d1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..80575123d770d650dc945480deb0b20aee7c932f --- /dev/null +++ b/parse/train/SyxAb30cY7/images/0035a01d3a4b6f29573e9a043351ba7723e0dc16e9bb0c6baa5485c16e2257d1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d12f8a276afdb3b870e270386d3685f6c6d1de2e746ac7c7231b335c19a4d5be +size 13843 diff --git a/parse/train/SyxAb30cY7/images/0b2b7e56751329b86afd6f10fbc987ae1bfd0803cf303833a4151fe8e5073018.jpg b/parse/train/SyxAb30cY7/images/0b2b7e56751329b86afd6f10fbc987ae1bfd0803cf303833a4151fe8e5073018.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a8c688b56cfeab973eb90f7dccd46d3724038aab --- /dev/null +++ b/parse/train/SyxAb30cY7/images/0b2b7e56751329b86afd6f10fbc987ae1bfd0803cf303833a4151fe8e5073018.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:99719f6d0b2b7863e5c0920e048fe5e9db6ba7cdf7196cc1d6591b88e9c2928c +size 34888 diff --git a/parse/train/SyxAb30cY7/images/14f62442b9b9c598592896760e790d2546187a05e9540a45fd8d59666c990c81.jpg b/parse/train/SyxAb30cY7/images/14f62442b9b9c598592896760e790d2546187a05e9540a45fd8d59666c990c81.jpg new file mode 100644 index 0000000000000000000000000000000000000000..72b4a16c1a113fef855e1139e83fc4809073b08c --- /dev/null +++ b/parse/train/SyxAb30cY7/images/14f62442b9b9c598592896760e790d2546187a05e9540a45fd8d59666c990c81.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:52591191bbcab804305237ec46c82805aac726e61ede88f4b36284e0719aafb0 +size 7006 diff --git a/parse/train/SyxAb30cY7/images/154a1c905478d981d03cc29da33d77c88e340ca489ad327f0274b465ab5950c0.jpg b/parse/train/SyxAb30cY7/images/154a1c905478d981d03cc29da33d77c88e340ca489ad327f0274b465ab5950c0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e0b17be6b0d1de074c267b3e0bd4d06278311fd7 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/154a1c905478d981d03cc29da33d77c88e340ca489ad327f0274b465ab5950c0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6971c17630f61f0ec3ef9cdd17fb8c2cb0d8afbaea04b4e872d4f9b4cb5075ed +size 2877 diff --git a/parse/train/SyxAb30cY7/images/1777f92e260ce321ecce48a10a82075cd93b662cb17b178e5bdea00dab126ac7.jpg b/parse/train/SyxAb30cY7/images/1777f92e260ce321ecce48a10a82075cd93b662cb17b178e5bdea00dab126ac7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d3078c260208da847ad09f031607e07cb0dae556 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/1777f92e260ce321ecce48a10a82075cd93b662cb17b178e5bdea00dab126ac7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:adacd6e940c88ef37d492a504d9f7e5f5c2c5de89e3b8d678c52d83c376be6f5 +size 16435 diff --git a/parse/train/SyxAb30cY7/images/251faad65589865c9f187a423afce9dacaef872ed76ea245b1b2ea2b64ef6872.jpg b/parse/train/SyxAb30cY7/images/251faad65589865c9f187a423afce9dacaef872ed76ea245b1b2ea2b64ef6872.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f1c81ce752f10079c430f808321b4886cd6757c0 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/251faad65589865c9f187a423afce9dacaef872ed76ea245b1b2ea2b64ef6872.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4950daa9afdbd6cf1657a030440417bce9637d91657d1f5dd187964f3ca704ed +size 7465 diff --git a/parse/train/SyxAb30cY7/images/2597fe143c29a132037adead882a1a1016e5b9d9bcb60c2dff641c5d7cb82e9c.jpg b/parse/train/SyxAb30cY7/images/2597fe143c29a132037adead882a1a1016e5b9d9bcb60c2dff641c5d7cb82e9c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5b1f6fabc520283e6a3eaa5e2d784732e4b680d9 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/2597fe143c29a132037adead882a1a1016e5b9d9bcb60c2dff641c5d7cb82e9c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7d68cd68b6a46a424cb4dabe7ed12b5edc5b50c6a316abc72bce69353624b2b8 +size 73407 diff --git a/parse/train/SyxAb30cY7/images/339002471b054e95c8d1749c22777fc38cd56bddaa6d4e22f0dbd0be1c96a5e8.jpg b/parse/train/SyxAb30cY7/images/339002471b054e95c8d1749c22777fc38cd56bddaa6d4e22f0dbd0be1c96a5e8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6203eef3b730680f2ed70a29bdd06516aa723463 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/339002471b054e95c8d1749c22777fc38cd56bddaa6d4e22f0dbd0be1c96a5e8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fea6f84b107d7b6284c713b60e82acdb6115170a3ba95165337a0d06df6cbb95 +size 123999 diff --git a/parse/train/SyxAb30cY7/images/385c205e4c7c2a2a156202997b2cce4414e5f6af278d9c3cdf978ba392979a12.jpg b/parse/train/SyxAb30cY7/images/385c205e4c7c2a2a156202997b2cce4414e5f6af278d9c3cdf978ba392979a12.jpg new file mode 100644 index 0000000000000000000000000000000000000000..391f072ba3419ab78332d50b69daf960692775f3 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/385c205e4c7c2a2a156202997b2cce4414e5f6af278d9c3cdf978ba392979a12.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9e7cabda60563b801ec05fd970f048f4e31b5689e405c8e509b79a40a210b1b +size 2511 diff --git a/parse/train/SyxAb30cY7/images/38facbfd68c823b45564d8be9f0a87e47096838df86c7c18ddeeddf68df09b46.jpg b/parse/train/SyxAb30cY7/images/38facbfd68c823b45564d8be9f0a87e47096838df86c7c18ddeeddf68df09b46.jpg new file mode 100644 index 0000000000000000000000000000000000000000..20380f4263178778bc5983281d090c5bc43c3a72 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/38facbfd68c823b45564d8be9f0a87e47096838df86c7c18ddeeddf68df09b46.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e0db16db75998d6dc4acf573b3b413189edd724a1cb4b54a4137c9dceccc27af +size 29739 diff --git a/parse/train/SyxAb30cY7/images/3d71c0727634c3d8a54dea0100e2bc884c85a5b6478107e48531d53466834ba3.jpg b/parse/train/SyxAb30cY7/images/3d71c0727634c3d8a54dea0100e2bc884c85a5b6478107e48531d53466834ba3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b08104182a00e9c5e9466ff3d82fb6dd8e72d761 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/3d71c0727634c3d8a54dea0100e2bc884c85a5b6478107e48531d53466834ba3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:50b081bc285ac61f9690dfe679d6e857337de72cd2fba5334fc0deb7a2db8339 +size 22071 diff --git a/parse/train/SyxAb30cY7/images/447b946557f267004f860f11d0f13133b4f0fd90be9b562278222ebd5dbbc3af.jpg b/parse/train/SyxAb30cY7/images/447b946557f267004f860f11d0f13133b4f0fd90be9b562278222ebd5dbbc3af.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1b7e83990315a1c3b2cf99a4fe6c1e294dac32a5 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/447b946557f267004f860f11d0f13133b4f0fd90be9b562278222ebd5dbbc3af.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a5bbb115b8dc4f08f18b2126dbf8c52b5b5b07c51a247d419a8f4c213072275c +size 6318 diff --git a/parse/train/SyxAb30cY7/images/5d1251776cd37db4d5efc50795d456f28d98d3b2ed83499291524c707c942d43.jpg b/parse/train/SyxAb30cY7/images/5d1251776cd37db4d5efc50795d456f28d98d3b2ed83499291524c707c942d43.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4206c26b30c14f7d2e677f80c2bcab4c6b64bd39 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/5d1251776cd37db4d5efc50795d456f28d98d3b2ed83499291524c707c942d43.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:47b531dd72701b360b30420b37669c75a329162b20da2946187a273b217321a2 +size 3591 diff --git a/parse/train/SyxAb30cY7/images/603436d4a2274b7027f0e5c64ae190c173359ab6e0f15ff38e9cce182b8aae8c.jpg b/parse/train/SyxAb30cY7/images/603436d4a2274b7027f0e5c64ae190c173359ab6e0f15ff38e9cce182b8aae8c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5ee8ce5c910c195c7b51381430814c6bcc06ab73 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/603436d4a2274b7027f0e5c64ae190c173359ab6e0f15ff38e9cce182b8aae8c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0d57118d014c89de005872f206f0577b2cff2ed09afed89930dc8cb19fa8f092 +size 116634 diff --git a/parse/train/SyxAb30cY7/images/62b5df1c0531806f62241c7a17145fabc1de421e611aed619e37c7b5937ec01f.jpg b/parse/train/SyxAb30cY7/images/62b5df1c0531806f62241c7a17145fabc1de421e611aed619e37c7b5937ec01f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cee85d47d56fb67ccc5b89f8bade14d08b143bde --- /dev/null +++ b/parse/train/SyxAb30cY7/images/62b5df1c0531806f62241c7a17145fabc1de421e611aed619e37c7b5937ec01f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c7c93c3988650bdad99d3c781438bff967afb01f587df68356b70f9db50f2183 +size 351022 diff --git a/parse/train/SyxAb30cY7/images/67c3f79b1f20960edc062608928c488243a0780deb3b65d0a79a2b649cce092d.jpg b/parse/train/SyxAb30cY7/images/67c3f79b1f20960edc062608928c488243a0780deb3b65d0a79a2b649cce092d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..db19b256f615c8287619bc2248ce30e9df2e1846 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/67c3f79b1f20960edc062608928c488243a0780deb3b65d0a79a2b649cce092d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8ff6021d641fc701d78298279c7b303572f963a7f9b262016290be0bffee5b4d +size 4969 diff --git a/parse/train/SyxAb30cY7/images/6a52bcc6ddefb33fc9c7ec8dfd11c2c49d81b2e1aeed6ce9d28acad8db1fb7f6.jpg b/parse/train/SyxAb30cY7/images/6a52bcc6ddefb33fc9c7ec8dfd11c2c49d81b2e1aeed6ce9d28acad8db1fb7f6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7d0b40bb1cf9c4f4c18bc86d17ce228dfa1613e4 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/6a52bcc6ddefb33fc9c7ec8dfd11c2c49d81b2e1aeed6ce9d28acad8db1fb7f6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:acb3180726613309aaf8b2dfcf29878d4b0382118d554dcda1ed62e55e9bec18 +size 30192 diff --git a/parse/train/SyxAb30cY7/images/7175da35414aa36a21a5ac0e91f89e654b99dd73c0f8c81b0c0e5b677ee9c2dc.jpg b/parse/train/SyxAb30cY7/images/7175da35414aa36a21a5ac0e91f89e654b99dd73c0f8c81b0c0e5b677ee9c2dc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4d06610f779c867d9b4df095eb67c1bba5c2e390 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/7175da35414aa36a21a5ac0e91f89e654b99dd73c0f8c81b0c0e5b677ee9c2dc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:de8cfe58f9ebcfe5ec7773b67ef6e72490d028069d5ae6deba34da37d7029b4d +size 2396 diff --git a/parse/train/SyxAb30cY7/images/7d643446e3a6335c69b6352996a9134749c46488027251b6139567d469914f84.jpg b/parse/train/SyxAb30cY7/images/7d643446e3a6335c69b6352996a9134749c46488027251b6139567d469914f84.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c7a7e1286396be242db0dcf5487635f45358b147 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/7d643446e3a6335c69b6352996a9134749c46488027251b6139567d469914f84.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e29a4cd38b326dcf67d648c560720530c56aa63b2416afff92522a3fe4d1b4d9 +size 8749 diff --git a/parse/train/SyxAb30cY7/images/82c35561a11b73c421fe67a23906c4628bbd4989ff5ac0435fed6d38118eb9a9.jpg b/parse/train/SyxAb30cY7/images/82c35561a11b73c421fe67a23906c4628bbd4989ff5ac0435fed6d38118eb9a9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6f633264d819657a9cf1cc44710976be271d68ee --- /dev/null +++ b/parse/train/SyxAb30cY7/images/82c35561a11b73c421fe67a23906c4628bbd4989ff5ac0435fed6d38118eb9a9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:341c2b7b0c2fa08a0d5fc57f7f38de7680e3a240949091ae51a37837d8b6996b +size 72835 diff --git a/parse/train/SyxAb30cY7/images/8a9c400c1bfa1022ac650417e4dbcdce13a0cbfbcbf0c5d3b7c785cd2335f20f.jpg b/parse/train/SyxAb30cY7/images/8a9c400c1bfa1022ac650417e4dbcdce13a0cbfbcbf0c5d3b7c785cd2335f20f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f23de09fd5ab0772044f4d37f030ce9e919cd892 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/8a9c400c1bfa1022ac650417e4dbcdce13a0cbfbcbf0c5d3b7c785cd2335f20f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0d272d6d48b3fc7fa41fc52780e073529bb40f1e54139496f712b128423aeda5 +size 49963 diff --git a/parse/train/SyxAb30cY7/images/8fec7a241f26c123d7c20cb77725a3e92718274370742f21725e377a527f32d2.jpg b/parse/train/SyxAb30cY7/images/8fec7a241f26c123d7c20cb77725a3e92718274370742f21725e377a527f32d2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3fce76b28f8d6903b416059a3dcaaa4e5dbbcd22 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/8fec7a241f26c123d7c20cb77725a3e92718274370742f21725e377a527f32d2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f6ddbcb99f46549fc448622ae19f14f65723373ecb2b913336852ea780f1d1c5 +size 3286 diff --git a/parse/train/SyxAb30cY7/images/957d354e16c984f65b1516ea4b7c51a19fd7f34ade8d87ec8694ca7af3d1a289.jpg b/parse/train/SyxAb30cY7/images/957d354e16c984f65b1516ea4b7c51a19fd7f34ade8d87ec8694ca7af3d1a289.jpg new file mode 100644 index 0000000000000000000000000000000000000000..227567b456b092c99837e521777924795a8d4b38 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/957d354e16c984f65b1516ea4b7c51a19fd7f34ade8d87ec8694ca7af3d1a289.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:138ec44b2a0242786bf5dcf9b1da359aa0bb0ad50b83f0d94d5312bb1811e366 +size 55064 diff --git a/parse/train/SyxAb30cY7/images/a606cea45104d344c4fdc44fcb80fc5e9792a314e798173bdd8bdf396b64bd69.jpg b/parse/train/SyxAb30cY7/images/a606cea45104d344c4fdc44fcb80fc5e9792a314e798173bdd8bdf396b64bd69.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8f1e9237e6895920596a29a381b6ddb4bdbc32b1 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/a606cea45104d344c4fdc44fcb80fc5e9792a314e798173bdd8bdf396b64bd69.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:582e7e79f31787a2b942755a7732ffbcb13c1151cd6193e65c438ce448030b9c +size 102443 diff --git a/parse/train/SyxAb30cY7/images/a9cc3c0789d140d0f53987728544694d1fc04b5cd7985a889c35e94b09b6c523.jpg b/parse/train/SyxAb30cY7/images/a9cc3c0789d140d0f53987728544694d1fc04b5cd7985a889c35e94b09b6c523.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fbc7eb42d4217f49b27c181deb7bbf83717808d4 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/a9cc3c0789d140d0f53987728544694d1fc04b5cd7985a889c35e94b09b6c523.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4723a9869ed42af2c97be33b6668916258412bfd7727a5328efcd9925c2de1d6 +size 20407 diff --git a/parse/train/SyxAb30cY7/images/ba82e0eabc53cb632fccc9b71da16c108a2c986b05d764d2a04eb1c3e2ff408b.jpg b/parse/train/SyxAb30cY7/images/ba82e0eabc53cb632fccc9b71da16c108a2c986b05d764d2a04eb1c3e2ff408b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2938a78b217c7171b1c2fd540c69ec45e3404d64 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/ba82e0eabc53cb632fccc9b71da16c108a2c986b05d764d2a04eb1c3e2ff408b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4e10c9c73ca9c304c5a71d155dd07f4d9818c07b64b1eec94bbf4e7a26b807ce +size 54789 diff --git a/parse/train/SyxAb30cY7/images/be4989108884a0f3034db5ee70d248a57bde34c7ebf36f3d338017440941af49.jpg b/parse/train/SyxAb30cY7/images/be4989108884a0f3034db5ee70d248a57bde34c7ebf36f3d338017440941af49.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b68a4b6fafabb66d4c65a135f465048f94dae2b4 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/be4989108884a0f3034db5ee70d248a57bde34c7ebf36f3d338017440941af49.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d3548b96f8b433291ab124c94c055f507510adcbb216ad1441cfec480fbb84aa +size 6065 diff --git a/parse/train/SyxAb30cY7/images/c0aafe3220e2f2b7b40f27b316738b4b954dddf86caebcebc75449a656a37b0f.jpg b/parse/train/SyxAb30cY7/images/c0aafe3220e2f2b7b40f27b316738b4b954dddf86caebcebc75449a656a37b0f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..272bfb423a7badf421173d9056099a29482e69e7 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/c0aafe3220e2f2b7b40f27b316738b4b954dddf86caebcebc75449a656a37b0f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ebbdf09ff52535bfcec6b21cd02c51eecb01d4b99e353c0a6935fbb9496c4986 +size 3165 diff --git a/parse/train/SyxAb30cY7/images/c66d9f4a25d6dba72f90af5a3cc2a972d1102b444b72850bbb1a0bac04230161.jpg b/parse/train/SyxAb30cY7/images/c66d9f4a25d6dba72f90af5a3cc2a972d1102b444b72850bbb1a0bac04230161.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4ce66ccbb95701b84c396ea421bae1c8d9e01eff --- /dev/null +++ b/parse/train/SyxAb30cY7/images/c66d9f4a25d6dba72f90af5a3cc2a972d1102b444b72850bbb1a0bac04230161.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2a9f1f5d368ecbfad446682a5ed95c733997da9d3e247fb31d18057f91947e09 +size 8722 diff --git a/parse/train/SyxAb30cY7/images/ca212d74ed37061a79acf73540dc079bfb475fcebd6da9f2dbfad17df55d6ffc.jpg b/parse/train/SyxAb30cY7/images/ca212d74ed37061a79acf73540dc079bfb475fcebd6da9f2dbfad17df55d6ffc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3e33721f4fed212c40c01c361ea9c599821951b9 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/ca212d74ed37061a79acf73540dc079bfb475fcebd6da9f2dbfad17df55d6ffc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:743906067b311a2e26be08f5582db214f729adfe55cab89714dafee8989a3757 +size 5607 diff --git a/parse/train/SyxAb30cY7/images/d3d30231252ce370c85f21898bf993fb04fe1ee163c320afe9ac5c08e93cf07e.jpg b/parse/train/SyxAb30cY7/images/d3d30231252ce370c85f21898bf993fb04fe1ee163c320afe9ac5c08e93cf07e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..55851ab715528e6b0d50d39b717421e56d77606a --- /dev/null +++ b/parse/train/SyxAb30cY7/images/d3d30231252ce370c85f21898bf993fb04fe1ee163c320afe9ac5c08e93cf07e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8df81e2370d9c46eda547ea3db45c16fa6dd381edf23229d0989284ec9c22955 +size 5116 diff --git a/parse/train/SyxAb30cY7/images/d576f39e13544b046a929c09d7c0a74343b6ad9d86f37c695b45126fee040e1b.jpg b/parse/train/SyxAb30cY7/images/d576f39e13544b046a929c09d7c0a74343b6ad9d86f37c695b45126fee040e1b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d5b7a2da52b7cafd1879381d91c9d7495f7f21dd --- /dev/null +++ b/parse/train/SyxAb30cY7/images/d576f39e13544b046a929c09d7c0a74343b6ad9d86f37c695b45126fee040e1b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ae1131e9e945e3d4ee692b3fb0e8bcd1bb6d2f299f6e2a9de9b512739ba388c9 +size 17124 diff --git a/parse/train/SyxAb30cY7/images/d72a28db71bafa5f68d5a906a30b14e090a3c1daf9ba84209bf2a528abfc74ce.jpg b/parse/train/SyxAb30cY7/images/d72a28db71bafa5f68d5a906a30b14e090a3c1daf9ba84209bf2a528abfc74ce.jpg new file mode 100644 index 0000000000000000000000000000000000000000..636d94aa2ae3c501f2d59980042c392292e500e2 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/d72a28db71bafa5f68d5a906a30b14e090a3c1daf9ba84209bf2a528abfc74ce.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:866cbd64c054125d42990b5125c069b98cf3b967df0fdd9c4601ece820b0c215 +size 229048 diff --git a/parse/train/SyxAb30cY7/images/d84bab304e5cea8340edcd5dd42746084c3097b15d73548a5bfc3b20b7165e31.jpg b/parse/train/SyxAb30cY7/images/d84bab304e5cea8340edcd5dd42746084c3097b15d73548a5bfc3b20b7165e31.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1ed73ecc4a92fbe64c967b7567c19ae41f3e74cf --- /dev/null +++ b/parse/train/SyxAb30cY7/images/d84bab304e5cea8340edcd5dd42746084c3097b15d73548a5bfc3b20b7165e31.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6cba7a0f6209f91a11e92077642e97113eef536ad3c19d3287ce50a0ee1c2011 +size 12691 diff --git a/parse/train/SyxAb30cY7/images/d93d899be693fdae96a56a4cbb6b5b4ad944a055268d8b29ec27abbed2f30281.jpg b/parse/train/SyxAb30cY7/images/d93d899be693fdae96a56a4cbb6b5b4ad944a055268d8b29ec27abbed2f30281.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1a1feb8642a3fb597a7c65fd46d2a336bbc73d57 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/d93d899be693fdae96a56a4cbb6b5b4ad944a055268d8b29ec27abbed2f30281.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:91291a4f4efc048614d61d748acb2c71a80c8cf2c9cbaac85a748b47f4131652 +size 334673 diff --git a/parse/train/SyxAb30cY7/images/dd0049c26c52721ca9fa5cce191646cbd67a7214764cb6f4acabe058dbdfe908.jpg b/parse/train/SyxAb30cY7/images/dd0049c26c52721ca9fa5cce191646cbd67a7214764cb6f4acabe058dbdfe908.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ada274bb251a22b1f696e464c0e7cc0398a9e0ac --- /dev/null +++ b/parse/train/SyxAb30cY7/images/dd0049c26c52721ca9fa5cce191646cbd67a7214764cb6f4acabe058dbdfe908.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9d43fc6c073207c346cbe74863ce6986a817e01540b63e35d56d36c795625368 +size 12058 diff --git a/parse/train/SyxAb30cY7/images/e9389fe5ce8ee60c31e415d1f9cca55e946577edaf4fccd6a7d90ee268d78c89.jpg b/parse/train/SyxAb30cY7/images/e9389fe5ce8ee60c31e415d1f9cca55e946577edaf4fccd6a7d90ee268d78c89.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fd138d9cebb987a95486f21ae1c0d357f2c667f7 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/e9389fe5ce8ee60c31e415d1f9cca55e946577edaf4fccd6a7d90ee268d78c89.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:56d4427667109e6667f0ca24819dfabc59de1171d69750e3b4d2b1d1f21df6f1 +size 22364 diff --git a/parse/train/SyxAb30cY7/images/f28d01a9563c66a3757b21d088b2f941f2ba0de36f55037507aa52bb0fba8e76.jpg b/parse/train/SyxAb30cY7/images/f28d01a9563c66a3757b21d088b2f941f2ba0de36f55037507aa52bb0fba8e76.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5ed38e46b803cbe315225a3eaa20e241d831a18c --- /dev/null +++ b/parse/train/SyxAb30cY7/images/f28d01a9563c66a3757b21d088b2f941f2ba0de36f55037507aa52bb0fba8e76.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2e9a1fa91b5631f726ae877d2a96c6009e37e4ff5d5a595e0fc255644031edcd +size 7297 diff --git a/parse/train/SyxAb30cY7/images/f33dfb80648a6285e77a72dbf6f3638dd1a117e67d469d447fc7644d1ae4adeb.jpg b/parse/train/SyxAb30cY7/images/f33dfb80648a6285e77a72dbf6f3638dd1a117e67d469d447fc7644d1ae4adeb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..282329e61dc990360b8f4918c4f71bc25ef35a68 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/f33dfb80648a6285e77a72dbf6f3638dd1a117e67d469d447fc7644d1ae4adeb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:281f6e3ef4e18be58233b73d3a5e32b4207849190a1d92ca8d6a173f80185357 +size 1897 diff --git a/parse/train/SyxAb30cY7/images/ff6a598dcd8bedd9e7b1fec6b604de25081ec1424a448e47c2fb1f6e3d517642.jpg b/parse/train/SyxAb30cY7/images/ff6a598dcd8bedd9e7b1fec6b604de25081ec1424a448e47c2fb1f6e3d517642.jpg new file mode 100644 index 0000000000000000000000000000000000000000..203275c57252813daa15b5ee1f6df805fecbfc26 --- /dev/null +++ b/parse/train/SyxAb30cY7/images/ff6a598dcd8bedd9e7b1fec6b604de25081ec1424a448e47c2fb1f6e3d517642.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f8e296a9fa32f82b40ba56e5b4511e6376c961abf0aeb51d9c8d5e7a5170be98 +size 16578 diff --git a/parse/train/SyxKrySYPr/images/197daa0ea022f06a9c301162acd1f2b87383d5171427a9efa2dec05bcb6d8bb5.jpg b/parse/train/SyxKrySYPr/images/197daa0ea022f06a9c301162acd1f2b87383d5171427a9efa2dec05bcb6d8bb5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1ca06cb1acefe83488119b2c9037fa3d31784a7e --- /dev/null +++ b/parse/train/SyxKrySYPr/images/197daa0ea022f06a9c301162acd1f2b87383d5171427a9efa2dec05bcb6d8bb5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b991f6e166f0c7b38f47661cb5bab857288fccf6916e1d6f012bb42a2d70ff6c +size 132950 diff --git a/parse/train/SyxKrySYPr/images/24f71f9da14fbd76fea3e26b5fccfb29c5dcc22f079d1fa3b7883d58778768f3.jpg b/parse/train/SyxKrySYPr/images/24f71f9da14fbd76fea3e26b5fccfb29c5dcc22f079d1fa3b7883d58778768f3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a723158271ebbe48782d45ea0ffdd47277a7de62 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/24f71f9da14fbd76fea3e26b5fccfb29c5dcc22f079d1fa3b7883d58778768f3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:91f48be8d8601d4f65922425957f577ab7ae60878d4982a9a625c5bfe2bbc5bf +size 40477 diff --git a/parse/train/SyxKrySYPr/images/257daa8069812d87b63c712291f85bf13d78624606c701e2c37f2f55acee7055.jpg b/parse/train/SyxKrySYPr/images/257daa8069812d87b63c712291f85bf13d78624606c701e2c37f2f55acee7055.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9525624bb504f0ba1ae6ff5497590b0d94eb2b17 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/257daa8069812d87b63c712291f85bf13d78624606c701e2c37f2f55acee7055.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2a807f2294c5ed57885dfbdd972fb9050e759635fb2bc0b39dd4dc4d450fb256 +size 32411 diff --git a/parse/train/SyxKrySYPr/images/2e121016b1cb7e0d1e7586b617f1b2d947be65c26f8b77a5bc38d70afa177cc3.jpg b/parse/train/SyxKrySYPr/images/2e121016b1cb7e0d1e7586b617f1b2d947be65c26f8b77a5bc38d70afa177cc3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..087bbdd987532bf7ae4be8121aa15e280255cc22 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/2e121016b1cb7e0d1e7586b617f1b2d947be65c26f8b77a5bc38d70afa177cc3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e9d1f0f41982e9938d78a2d8b18f8d804021bde8264fedfda0e442bbd8934ada +size 68738 diff --git a/parse/train/SyxKrySYPr/images/30838ea7d43ed7d927abc8ca5aefe3989b6c1eb607f01824f9f9963d8a07e688.jpg b/parse/train/SyxKrySYPr/images/30838ea7d43ed7d927abc8ca5aefe3989b6c1eb607f01824f9f9963d8a07e688.jpg new file mode 100644 index 0000000000000000000000000000000000000000..beb3c696eec89ccf032ed28a9b6695a18e488748 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/30838ea7d43ed7d927abc8ca5aefe3989b6c1eb607f01824f9f9963d8a07e688.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2ff58b40b5f882803a0b01e8635b60689f1e2a4fe7e9d62a168b6e471bc4f878 +size 33982 diff --git a/parse/train/SyxKrySYPr/images/30a95faf907c738d48f7202db0b5aaaf121b3f2681986a929a1e49b6de29796c.jpg b/parse/train/SyxKrySYPr/images/30a95faf907c738d48f7202db0b5aaaf121b3f2681986a929a1e49b6de29796c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c52eebf1c3b69aa6fbe652d3e556e694e7ebb291 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/30a95faf907c738d48f7202db0b5aaaf121b3f2681986a929a1e49b6de29796c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:006072932ca9d3998be2914fe8b930f510c805ffdf30ee3c13beb945f15b2346 +size 24012 diff --git a/parse/train/SyxKrySYPr/images/32bb564a324859699b7eca89634a55236ea3422a29411827f89b4fb80fc5151e.jpg b/parse/train/SyxKrySYPr/images/32bb564a324859699b7eca89634a55236ea3422a29411827f89b4fb80fc5151e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9d18ab9d15a17e87ac2360a2d350d7ea0a16eb4c --- /dev/null +++ b/parse/train/SyxKrySYPr/images/32bb564a324859699b7eca89634a55236ea3422a29411827f89b4fb80fc5151e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8d1695e8151709744320d92e876fba30b14db43225b216ab9762f7b6db076609 +size 57326 diff --git a/parse/train/SyxKrySYPr/images/3c271cdc012f756b7959308dbfa9e9835346b162065a5b682d958a01e5ecc837.jpg b/parse/train/SyxKrySYPr/images/3c271cdc012f756b7959308dbfa9e9835346b162065a5b682d958a01e5ecc837.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1701fc6f7027980575deaeb95ee710c8356d6e15 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/3c271cdc012f756b7959308dbfa9e9835346b162065a5b682d958a01e5ecc837.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5a74e62b945068b1b09596d534568a561517ff7695e7dabb29c8814fcb6aed0e +size 44938 diff --git a/parse/train/SyxKrySYPr/images/4526bd59eda43dc87adbc6b1c81bb32cf87729033427cdcdd6852b7918412bb3.jpg b/parse/train/SyxKrySYPr/images/4526bd59eda43dc87adbc6b1c81bb32cf87729033427cdcdd6852b7918412bb3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..23e886cc1a5fb4958f28572fbaab8fd39a4534e6 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/4526bd59eda43dc87adbc6b1c81bb32cf87729033427cdcdd6852b7918412bb3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d2800b4be5f44e7ae682cbc529141bec72b882f75786d9f43a136f18569d87e1 +size 33811 diff --git a/parse/train/SyxKrySYPr/images/496fc70ee143b7a9776ec0f677c6b9d43f2167bd8ec27c0208b3deecfc172846.jpg b/parse/train/SyxKrySYPr/images/496fc70ee143b7a9776ec0f677c6b9d43f2167bd8ec27c0208b3deecfc172846.jpg new file mode 100644 index 0000000000000000000000000000000000000000..376ae36518a5d029a98845640cf726ecb20153cb --- /dev/null +++ b/parse/train/SyxKrySYPr/images/496fc70ee143b7a9776ec0f677c6b9d43f2167bd8ec27c0208b3deecfc172846.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:87e9ee2f44de5d2bd183e54087532bfcbb1a51a047c5c525b2e128d90300e4ae +size 43021 diff --git a/parse/train/SyxKrySYPr/images/58c2c94be3a7144c067b1a7b1e712b7fee093fda897725fadd0a13dd16ac8b66.jpg b/parse/train/SyxKrySYPr/images/58c2c94be3a7144c067b1a7b1e712b7fee093fda897725fadd0a13dd16ac8b66.jpg new file mode 100644 index 0000000000000000000000000000000000000000..52547d8960951a435fd4967850e787d8a955a9bb --- /dev/null +++ b/parse/train/SyxKrySYPr/images/58c2c94be3a7144c067b1a7b1e712b7fee093fda897725fadd0a13dd16ac8b66.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e754533dfa69a544f2013363632f4f2e3a32c944b2feade629f9c054938be13f +size 23845 diff --git a/parse/train/SyxKrySYPr/images/58d533e28ba7df1e625d2b01023f1d3355700392e11fce6763e4d05d4607cd65.jpg b/parse/train/SyxKrySYPr/images/58d533e28ba7df1e625d2b01023f1d3355700392e11fce6763e4d05d4607cd65.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f1147b51cf348bbcad1187c73781a3ae4ed2f179 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/58d533e28ba7df1e625d2b01023f1d3355700392e11fce6763e4d05d4607cd65.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0314f3abddb5d4601899eff21b50599509f67938bb19afb43782d4fb2152223a +size 5910 diff --git a/parse/train/SyxKrySYPr/images/6656cb5158812d84c97584e6f9b2c2fa5d242450922c14aa01f7da7e58f6b87b.jpg b/parse/train/SyxKrySYPr/images/6656cb5158812d84c97584e6f9b2c2fa5d242450922c14aa01f7da7e58f6b87b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..63d60c085384bde9d631d632f61ef53690a00000 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/6656cb5158812d84c97584e6f9b2c2fa5d242450922c14aa01f7da7e58f6b87b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9a7bfb32f6c3794a4e6f06b074d4757d493acaf8175fe0af20ca05b769cd1a78 +size 14123 diff --git a/parse/train/SyxKrySYPr/images/6ebc3b88c22a70f561ea1d8adaca940b8b2170d0b3b861cbd61be06aca2ef38d.jpg b/parse/train/SyxKrySYPr/images/6ebc3b88c22a70f561ea1d8adaca940b8b2170d0b3b861cbd61be06aca2ef38d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..057a90914a8d3aaf927248832b385bf0a1b7970a --- /dev/null +++ b/parse/train/SyxKrySYPr/images/6ebc3b88c22a70f561ea1d8adaca940b8b2170d0b3b861cbd61be06aca2ef38d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0461dfc91a83c123d8e5e10bf0cf1a895cc5150b2d1eca497ceaa7c4b7f401be +size 45562 diff --git a/parse/train/SyxKrySYPr/images/6fb6ae6620513c690dab0e51f11b3c89ca9b232165511e3900e6fc82c65855a3.jpg b/parse/train/SyxKrySYPr/images/6fb6ae6620513c690dab0e51f11b3c89ca9b232165511e3900e6fc82c65855a3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..af3ecd090f9ef68b0c94eccf607c44a92ed30a8e --- /dev/null +++ b/parse/train/SyxKrySYPr/images/6fb6ae6620513c690dab0e51f11b3c89ca9b232165511e3900e6fc82c65855a3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8ad341316801fcf470c02347093fa1d99824f07e03dc4f93da44e98efd8d374b +size 23475 diff --git a/parse/train/SyxKrySYPr/images/819f06577ec44d9a36e159282cc7e8c37c22a71d587b02b727e7152b962f1b12.jpg b/parse/train/SyxKrySYPr/images/819f06577ec44d9a36e159282cc7e8c37c22a71d587b02b727e7152b962f1b12.jpg new file mode 100644 index 0000000000000000000000000000000000000000..949c0abc712e184c45c5e8a4af5cad2bfd54d51a --- /dev/null +++ b/parse/train/SyxKrySYPr/images/819f06577ec44d9a36e159282cc7e8c37c22a71d587b02b727e7152b962f1b12.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:76d5c3608107f9da02e9737d8be1975790769120a56bf06415b00146f97e25f4 +size 32905 diff --git a/parse/train/SyxKrySYPr/images/8b9d87fa13e068a7360530a7f2630b072a09f35930098c004bdd5d2b5390bcb2.jpg b/parse/train/SyxKrySYPr/images/8b9d87fa13e068a7360530a7f2630b072a09f35930098c004bdd5d2b5390bcb2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..33ea2b67e804254272c91af19baa764ebbe766db --- /dev/null +++ b/parse/train/SyxKrySYPr/images/8b9d87fa13e068a7360530a7f2630b072a09f35930098c004bdd5d2b5390bcb2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3b9a79273820ce6b174f185b1806ea9f222043d2fb1152e1b74987b668ac3017 +size 28288 diff --git a/parse/train/SyxKrySYPr/images/8fc4a6c86aa3d3bbfe9092b047b2e4f861122c30e7f27e21df89583bc1bfa1d6.jpg b/parse/train/SyxKrySYPr/images/8fc4a6c86aa3d3bbfe9092b047b2e4f861122c30e7f27e21df89583bc1bfa1d6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a8cd254b9368d5da6f156d0478d926e5af062d29 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/8fc4a6c86aa3d3bbfe9092b047b2e4f861122c30e7f27e21df89583bc1bfa1d6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:41d6c35f5090492f9e16bfedf0fa8611294411e1c4da8befbf986c93e066324b +size 7609 diff --git a/parse/train/SyxKrySYPr/images/934c3aca65aec5eaa9ec356178a2503b51f8b5f614064a42547fdfb9398233d6.jpg b/parse/train/SyxKrySYPr/images/934c3aca65aec5eaa9ec356178a2503b51f8b5f614064a42547fdfb9398233d6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3ffa6fa76c5cf2f93cd6936f04a86a405d9894bc --- /dev/null +++ b/parse/train/SyxKrySYPr/images/934c3aca65aec5eaa9ec356178a2503b51f8b5f614064a42547fdfb9398233d6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:09021e56a48377b01e4c6bfb8419f7e1266c3004befafb92e5d103eccafec5c1 +size 43681 diff --git a/parse/train/SyxKrySYPr/images/9640cb6360077a7e9447bb56f43eda27c5cdf1a094b86d0831dc8c9068c46896.jpg b/parse/train/SyxKrySYPr/images/9640cb6360077a7e9447bb56f43eda27c5cdf1a094b86d0831dc8c9068c46896.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9e10612bd24a1a2d85ca6e7ee83cb9227f68d527 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/9640cb6360077a7e9447bb56f43eda27c5cdf1a094b86d0831dc8c9068c46896.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f2bb1d1d38c48d6c95b95e9a29f7a270330ff1beccfbae6ab9e7da8360755df4 +size 13624 diff --git a/parse/train/SyxKrySYPr/images/9699b33dc76b9f9acb9fd78299c693599c409d5ff9b907e13ea043c5daabad93.jpg b/parse/train/SyxKrySYPr/images/9699b33dc76b9f9acb9fd78299c693599c409d5ff9b907e13ea043c5daabad93.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bb7d247b249bee5bacc8901da3df5eab6e836967 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/9699b33dc76b9f9acb9fd78299c693599c409d5ff9b907e13ea043c5daabad93.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:db8055502625bd4a2e7cf511ecb7563261994de87155f01f00b6beba25c2e1e5 +size 7719 diff --git a/parse/train/SyxKrySYPr/images/977370c3f53ef500a9c854cdd933b26384cd5106e2816a6d5ff267f1ede5fe26.jpg b/parse/train/SyxKrySYPr/images/977370c3f53ef500a9c854cdd933b26384cd5106e2816a6d5ff267f1ede5fe26.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2fcf3d2a2ba00d326cc412d50508400809a452cc --- /dev/null +++ b/parse/train/SyxKrySYPr/images/977370c3f53ef500a9c854cdd933b26384cd5106e2816a6d5ff267f1ede5fe26.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1ef62ff84a33dc088e3551676a6a54d6c9f2dbb3300843d4dbc8cc2eea4eb72e +size 22646 diff --git a/parse/train/SyxKrySYPr/images/97c81a43beefd6edb78cd9e4a1c4752ff5fd17095e21bc7df7012545b8dc3242.jpg b/parse/train/SyxKrySYPr/images/97c81a43beefd6edb78cd9e4a1c4752ff5fd17095e21bc7df7012545b8dc3242.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1d690212b6c5e2a37e89d672396e281bb26259b8 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/97c81a43beefd6edb78cd9e4a1c4752ff5fd17095e21bc7df7012545b8dc3242.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3e6ffb910d63af8d42a94c65c74728a03e8ad49c06d5c011fdbb8b1aa2e3526f +size 24473 diff --git a/parse/train/SyxKrySYPr/images/9df923d4f1de7271328f896274725ee57f3708b0653257e8762105b373b0603a.jpg b/parse/train/SyxKrySYPr/images/9df923d4f1de7271328f896274725ee57f3708b0653257e8762105b373b0603a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ed21d1e86bc58c36b4a4706d9f88a9d53bd8bead --- /dev/null +++ b/parse/train/SyxKrySYPr/images/9df923d4f1de7271328f896274725ee57f3708b0653257e8762105b373b0603a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cca9724031d5b920f402a28e88dd0066d9e2a811a63dd4e5a007097b688b4157 +size 33105 diff --git a/parse/train/SyxKrySYPr/images/b4f36c84328e8dadc4527aa5ed297c797760b078dd9d1e0c8fd24e4e46b8749c.jpg b/parse/train/SyxKrySYPr/images/b4f36c84328e8dadc4527aa5ed297c797760b078dd9d1e0c8fd24e4e46b8749c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..345829b44da33ed659edc71dd6dc99456f001067 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/b4f36c84328e8dadc4527aa5ed297c797760b078dd9d1e0c8fd24e4e46b8749c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9371beed65b5a48735f4f295b32380be43464b190045f8d299831492bc16b74a +size 14490 diff --git a/parse/train/SyxKrySYPr/images/b6a1d40e5ab5dc98b878ecf82e90f7ec422bb0cd8ed05d3071675f8d8c4912bb.jpg b/parse/train/SyxKrySYPr/images/b6a1d40e5ab5dc98b878ecf82e90f7ec422bb0cd8ed05d3071675f8d8c4912bb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..81a0815734b994c8fcc2474809f2b6eb84e67325 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/b6a1d40e5ab5dc98b878ecf82e90f7ec422bb0cd8ed05d3071675f8d8c4912bb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a6326148a16aae0bca64edf820034767db7937385fe8cad01836ec076dac1f72 +size 34553 diff --git a/parse/train/SyxKrySYPr/images/b734abe6731fee597aa11f4ed6b490fa8628d7a441406749f59c7a872cd8cd93.jpg b/parse/train/SyxKrySYPr/images/b734abe6731fee597aa11f4ed6b490fa8628d7a441406749f59c7a872cd8cd93.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2f4c2fe552fcfbfd8adc9f43fa8c7444732422cc --- /dev/null +++ b/parse/train/SyxKrySYPr/images/b734abe6731fee597aa11f4ed6b490fa8628d7a441406749f59c7a872cd8cd93.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6fe5b35cf37b0d7e19f3f4864f8e245b64c85b9c7939693ae7a43b0b77d4a566 +size 52582 diff --git a/parse/train/SyxKrySYPr/images/b95c1dda97932b4affb2ac33ca24fd237f0881b8985254a04db746cc06e999df.jpg b/parse/train/SyxKrySYPr/images/b95c1dda97932b4affb2ac33ca24fd237f0881b8985254a04db746cc06e999df.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5209109d84e775330ed16c127f971ee48c9c125a --- /dev/null +++ b/parse/train/SyxKrySYPr/images/b95c1dda97932b4affb2ac33ca24fd237f0881b8985254a04db746cc06e999df.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3b9a6ce377bfebfb0e420f98370d9b8e8c036c58d11dd1b78473cea6a12b0f80 +size 11603 diff --git a/parse/train/SyxKrySYPr/images/c42ffbb16e924e0468e36e9b9f56b2e06f84606fb5a818aab0d9f7fa939a244f.jpg b/parse/train/SyxKrySYPr/images/c42ffbb16e924e0468e36e9b9f56b2e06f84606fb5a818aab0d9f7fa939a244f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e7ab170c25c99ae6c630cff8b3febcb9e4611e95 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/c42ffbb16e924e0468e36e9b9f56b2e06f84606fb5a818aab0d9f7fa939a244f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:af49ec160b33f30bc48a3e294349851f4a5eec5887c7afd83a984b4b49a9794a +size 19219 diff --git a/parse/train/SyxKrySYPr/images/c4fd2e8e004fc98e235575dab9588ef66da14d1a3b1d163131272025031854fc.jpg b/parse/train/SyxKrySYPr/images/c4fd2e8e004fc98e235575dab9588ef66da14d1a3b1d163131272025031854fc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..807bc1edef58546729bc5645f46c791d51e8a0a8 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/c4fd2e8e004fc98e235575dab9588ef66da14d1a3b1d163131272025031854fc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:52d53f864194988b3f3ec88a84c57d22f3708522f994f27a3c9db9314e8d0edb +size 19014 diff --git a/parse/train/SyxKrySYPr/images/c93c85aa1a2d28aacbb978e8fad8f56834ad11844271e51907d7f9c71ed31a7e.jpg b/parse/train/SyxKrySYPr/images/c93c85aa1a2d28aacbb978e8fad8f56834ad11844271e51907d7f9c71ed31a7e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4b40767dfec44ae997230468db7e0cc27aec8d73 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/c93c85aa1a2d28aacbb978e8fad8f56834ad11844271e51907d7f9c71ed31a7e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6f55b5b081fd474a3d9716815fe4acc81d678f0dc8e2cfb68a2c5d568fbddfbb +size 56962 diff --git a/parse/train/SyxKrySYPr/images/c955996cc9fb38c49c6cf553426b490a44fe97ffa8d4215392b25d94d80079a7.jpg b/parse/train/SyxKrySYPr/images/c955996cc9fb38c49c6cf553426b490a44fe97ffa8d4215392b25d94d80079a7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cbec0635b0fb1ec3a8a78140c42e1782436a5e25 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/c955996cc9fb38c49c6cf553426b490a44fe97ffa8d4215392b25d94d80079a7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e51c9a66183181599bf5bebbf72b8c6c3a68259944766c8e6f7f6168e5131025 +size 4811 diff --git a/parse/train/SyxKrySYPr/images/e011eb076bdfdf9235d481830c8195652b3dc62447eb1400ba6952163a568f96.jpg b/parse/train/SyxKrySYPr/images/e011eb076bdfdf9235d481830c8195652b3dc62447eb1400ba6952163a568f96.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e6f797081aa2956946d647f0dd0a7ade47a7ab0a --- /dev/null +++ b/parse/train/SyxKrySYPr/images/e011eb076bdfdf9235d481830c8195652b3dc62447eb1400ba6952163a568f96.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4343ab5ae023786598aad6d0973db4bea949b5b1c8f35e57ae8030c18389bb6a +size 20968 diff --git a/parse/train/SyxKrySYPr/images/e3404dbef0e3d2f4e31a6d5b796425cd37cb031f7065e4b6c24638d0c95f484b.jpg b/parse/train/SyxKrySYPr/images/e3404dbef0e3d2f4e31a6d5b796425cd37cb031f7065e4b6c24638d0c95f484b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bc028786d2d25799277ea7dfaf6306db1dae8d0e --- /dev/null +++ b/parse/train/SyxKrySYPr/images/e3404dbef0e3d2f4e31a6d5b796425cd37cb031f7065e4b6c24638d0c95f484b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:26c1fb816a8d7188f56ad4f95f8e68cd49795e3c25002a9b04970057f71c2816 +size 54362 diff --git a/parse/train/SyxKrySYPr/images/e42a5ac1dffa8ac199cbbe8c0b9dfb0eb55aad66f866f9f219f03630310d0f30.jpg b/parse/train/SyxKrySYPr/images/e42a5ac1dffa8ac199cbbe8c0b9dfb0eb55aad66f866f9f219f03630310d0f30.jpg new file mode 100644 index 0000000000000000000000000000000000000000..02d7b163a0745101904712b408c1d9ee968ad4e5 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/e42a5ac1dffa8ac199cbbe8c0b9dfb0eb55aad66f866f9f219f03630310d0f30.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1dd87f04be54ad37009d8bc2bd6385024d745bd0c598fb9ceaee3441485e6f1e +size 31427 diff --git a/parse/train/SyxKrySYPr/images/e4444a32067774d12165c050fc80da395f468c75bdad10412ac8f7fe3675cc4d.jpg b/parse/train/SyxKrySYPr/images/e4444a32067774d12165c050fc80da395f468c75bdad10412ac8f7fe3675cc4d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d041c989764ea3ef1ec6f84c77299f75f3d3321a --- /dev/null +++ b/parse/train/SyxKrySYPr/images/e4444a32067774d12165c050fc80da395f468c75bdad10412ac8f7fe3675cc4d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5fb03e689b7ac3fe60ee58d7d6a438b11be8b86d70de7674d09a4f75612e9c28 +size 10336 diff --git a/parse/train/SyxKrySYPr/images/ff348e907fc32f4ffc78c25c3bcdf515340bf3bd67691438e31c86d6b21f03c1.jpg b/parse/train/SyxKrySYPr/images/ff348e907fc32f4ffc78c25c3bcdf515340bf3bd67691438e31c86d6b21f03c1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a3b5a6f4d4abc223e9a711be3cfb30c5b3a3e185 --- /dev/null +++ b/parse/train/SyxKrySYPr/images/ff348e907fc32f4ffc78c25c3bcdf515340bf3bd67691438e31c86d6b21f03c1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c13ccd7187623008a4e18fbbf3b9b96d02670bf664f92d55e5a56aaa27eb37f6 +size 4166 diff --git a/parse/train/SyxtJh0qYm/SyxtJh0qYm.md b/parse/train/SyxtJh0qYm/SyxtJh0qYm.md new file mode 100644 index 0000000000000000000000000000000000000000..6255c1620e5ecb260c8d3e04b6b7e93b1a11dafe --- /dev/null +++ b/parse/train/SyxtJh0qYm/SyxtJh0qYm.md @@ -0,0 +1,580 @@ +# VARIATIONAL AUTOENCODER WITH ARBITRARY CONDITIONING + +Oleg Ivanov +Samsung AI Center Moscow Moscow, Russia +tigvarts@gmail.com Michael Figurnov +National Research University Higher School of Economics ∗ Moscow, Russia +michael@figurnov.ru Dmitry Vetrov +Samsung-HSE Laboratory, National Research University Higher School of Economics Samsung AI Center Moscow Moscow, Russia +vetrovd@yandex.ru + +# ABSTRACT + +We propose a single neural probabilistic model based on variational autoencoder that can be conditioned on an arbitrary subset of observed features and then sample the remaining features in “one shot”. The features may be both real-valued and categorical. Training of the model is performed by stochastic variational Bayes. The experimental evaluation on synthetic data, as well as feature imputation and image inpainting problems, shows the effectiveness of the proposed approach and diversity of the generated samples. + +# 1 INTRODUCTION + +In past years, a number of generative probabilistic models based on neural networks have been proposed. The most popular approaches include variational autoencoder (Kingma & Welling, 2013) (VAE) and generative adversarial net (Goodfellow et al., 2014) (GANs). They learn a distribution over objects $p ( x )$ and allow sampling from this distribution. + +In many cases, we are interested in learning a conditional distribution $p ( x | y )$ . For instance, if $x$ is an image of a face, $y$ could be the characteristics describing the face (are glasses present or not; length of hair, etc.) Conditional variational autoencoder (Sohn et al., 2015) and conditional generative adversarial nets (Mirza & Osindero, 2014) are popular methods for this problem. + +In this paper, we consider the problem of learning all conditional distributions of the form $p ( x _ { I } | x _ { U \setminus I } )$ , where $U$ is the set of all features and $I$ is its arbitrary subset. This problem generalizes both learning the joint distribution $p ( x )$ and learning the conditional distribution $p ( x | y )$ . To tackle this problem, we propose a Variational Autoencoder with Arbitrary Conditioning (VAEAC) model. It is a latent variable model similar to VAE, but allows conditioning on an arbitrary subset of the features. The conditioning features affect the prior on the latent Gaussian variables which are used to generate unobserved features. The model is trained using stochastic gradient variational Bayes (Kingma & Welling, 2013). + +We consider two most natural applications of the proposed model. The first one is feature imputation where the goal is to restore the missing features given the observed ones. The imputed values may be valuable by themselves or may improve the performance of other machine learning algorithms which process the dataset. Another application is image inpainting in which the goal is to fill in an unobserved part of an image with an artificial content in a realistic way. This can be used for removing unnecessary objects from the images or, vice versa, for complementing the partially closed or corrupted object. + +The experimental evaluation shows that the proposed model successfully samples from the conditional distributions. The distribution over samples is close to the true conditional distribution. This property is very important when the true distribution has several modes. The model is shown to be effective in feature imputation problem which helps to increase the quality of subsequent discriminative models on different problems from UCI datasets collection (Lichman, 2013). We demonstrate that model can generate diverse and realistic image inpaintings on MNIST (LeCun et al., 1998), Omniglot (Lake et al., 2015) and CelebA (Liu et al., 2015) datasets, and works even better than the current state of the art inpainting techniques in terms of peak signal to noise ratio (PSNR). + +The paper is organized as follows. In section 2 we review the related works. In section 3 we briefly describe variational autoencoders and conditional variational autoencoders. In section 4 we define the problem, describe the VAEAC model and its training procedure. In section 5 we evaluate VAEAC. Section 6 concludes the paper. Appendix contains additional explanations, theoretical analysis, and experiments for VAEAC. + +# 2 RELATED WORK + +Universal Marginalizer (Douglas et al., 2017) is a model based on a feed-forward neural network which approximates marginals of unobserved features conditioned on observable values. A related idea of an autoregressive model of joint probability was previously proposed in Germain et al. (2015) and Uria et al. (2016). The description of the model and comparison with VAEAC are available in section 5.3. + +Yoon et al. (2018) propose a GANs-based model called GAIN which solves the same problem as VAEAC. In contrast to VAEAC, GAIN does not use unobserved data during training, which makes it easier to apply to the missing features imputation problem. Nevertheless, it turns into a disadvantage when the fully-observed training data is available but the missingness rate at the testing stage is high. For example, in inpainting setting GAIN cannot learn the conditional distribution over MNIST digits given one horizontal line of the image while VAEAC can (see appendix D.4). The comparison of VAEAC and GAIN on the missing feature imputation problem is given in section 5.1 and appendix D.2. + +Rezende et al. (2014) [Appendix F], Sohl-Dickstein et al. (2015), Goyal et al. (2017), and Bordes et al. (2017) propose to fill missing data with noise and run Markov chain with a learned transition operator. The stationary distribution of such chains approximates the true conditional distribution of the unobserved features. Bachman & Precup (2015) consider missing feature imputation in terms of Markov decision process and propose LSTM-based sequential decision making model to solve it. Nevertheless, these methods are computationally expensive at the test time and require fully-observed training data. + +Image inpainting is a classic computer vision problem. Most of the earlier methods rely on local and texture information or hand-crafted problem-specific features (Bertalmio et al., 2000). In past years multiple neural network based approaches have been proposed. + +Pathak et al. (2016), Yeh et al. (2016) and Yang et al. (2017) use different kinds and combinations of adversarial, reconstruction, texture and other losses. Li et al. (2017) focuses on face inpainting and uses two adversarial losses and one semantic parsing loss to train the generative model. In Yeh et al. (2017) GANs are first trained on the whole training dataset. The inpainting is an optimization procedure that finds the latent variables that explain the observed features best. Then, the obtained latents are passed through the generative model to restore the unobserved portion of the image. We can say that VAEAC is a similar model which uses prior network to find a proper latents instead of solving the optimization problem. + +All described methods aim to produce a single realistic inpainting, while VAEAC is capable of sampling diverse inpaintings. Additionally, Yeh et al. (2016), Yang et al. (2017) and Yeh et al. (2017) have high testtime computational complexity of inpainting, because they require an optimization problem to be solved. On the other hand, VAEAC is a “single-shot” method with a low computational cost. + +# 3 BACKGROUND + +# 3.1 VARIATIONAL AUTOENCODER + +Variational autoencoder (Kingma & Welling, 2013) (VAE) is a directed generative model with latent variables. The generative process in variational autoencoder is as follows: first, a latent variable $z$ is generated from the prior distribution $p ( z )$ , and then the data $x$ is generated from the generative distribution $p _ { \theta } ( x | z )$ , where $\theta$ are the generative model’s parameters. This process induces the distribution $p _ { \theta } ( x ) = \mathbb { E } _ { p ( z ) } p _ { \theta } ( x | z )$ . The distribution $p _ { \theta } ( x | z )$ is modeled by a neural network with parameters $\theta$ . $p ( z )$ is a standard Gaussian distribution. + +The parameters $\theta$ are tuned by maximizing the likelihood of the training data points $\{ x _ { i } \} _ { i = 1 } ^ { N }$ from the true data distribution $p _ { d } ( x )$ . In general, this optimization problem is challenging due to intractable posterior inference. However, a variational lower bound can be optimized efficiently using backpropagation and stochastic gradient descent: + +$$ +\begin{array} { r l } & { \log p _ { \theta } ( x ) = \mathbb { E } _ { q _ { \phi } ( z | x ) } \log \frac { p _ { \theta } ( x , z ) } { q _ { \phi } ( z | x ) } + D _ { \mathrm { K L } } \big ( q _ { \phi } ( z | x ) \| p ( z | x , \theta ) \big ) } \\ & { \qquad \quad \ge \mathbb { E } _ { q _ { \phi } ( z | x ) } \log p _ { \theta } ( x | z ) - D _ { \mathrm { K L } } \big ( q _ { \phi } ( z | x ) \| p ( z ) \big ) = L _ { V A E } \big ( x ; \theta , \phi \big ) } \end{array} +$$ + +Here $q _ { \phi } ( z | x )$ is a proposal distribution parameterized by neural network with parameters $\phi$ that approximates the posterior $p ( z | x , \theta )$ . Usually this distribution is Gaussian with a diagonal covariance matrix. The closer $q _ { \phi } ( z | x )$ to $p ( z | x , \theta )$ , the tighter variational lower bound $L _ { V A E } ( \theta , \phi )$ . To compute the gradient of the variational lower bound with respect to $\phi$ , reparameterization trick is used: $z = \bar { \mu } _ { \phi } ( x ) + \bar { \varepsilon } \sigma _ { \phi } ( x )$ where $\varepsilon \sim \mathcal { N } ( 0 , I )$ and $\mu _ { \phi }$ and $\sigma _ { \phi }$ are deterministic functions parameterized by neural networks. So the gradient can be estimated using Monte-Carlo method for the first term and computing the second term analytically: + +$$ +\frac { \partial L _ { V A E } ( x ; \theta , \phi ) } { \partial \phi } = \mathbb { E } _ { \varepsilon \sim \mathcal { N } ( 0 , I ) } \frac { \partial } { \partial \phi } \log p _ { \theta } ( x | \mu _ { \phi } ( x ) + \varepsilon \sigma _ { \phi } ( x ) ) - \frac { \partial } { \partial \phi } D _ { \mathrm { K L } } ( q _ { \phi } ( z | x ) \| p ( z ) ) . +$$ + +So $L _ { V A E } ( \theta , \phi )$ can be optimized using stochastic gradient ascent with respect to $\phi$ and $\theta$ + +# 3.2 CONDITIONAL VARIATIONAL AUTOENCODER + +Conditional variational autoencoder (Sohn et al., 2015) (CVAE) approximates the conditional distribution $p _ { d } ( x | y )$ . It outperforms deterministic models when the distribution $p _ { d } ( x | y )$ is multi-modal (diverse $x \mathbf { s }$ are probable for the given $y$ ). For example, assume that $x$ is a real-valued image. Then, a deterministic regression model with mean squared error loss would predict the average blurry value for $x$ . On the other hand, CVAE learns the distribution of $x$ , from which one can sample diverse and realistic objects. + +Variational lower bound for CVAE can be derived similarly to VAE by conditioning all considered distributions on $y$ : + +$$ +L _ { C V A E } ( x , y ; \theta , \psi , \phi ) = \mathbb { E } _ { q _ { \phi } ( z | x , y ) } \log p _ { \theta } ( x | z , y ) - D _ { \mathrm { K L } } ( q _ { \phi } ( z | x , y ) | | p _ { \psi } ( z | y ) ) \leq \log p _ { \theta , \psi } ( x | y ) +$$ + +Similarly to VAE, this objective is optimized using the reparameterization trick. Note that the prior distribution $p _ { \psi } ( z | y )$ is conditioned on $y$ and is modeled by a neural network with parameters $\psi$ . Thus, CVAE uses three trainable neural networks, while VAE only uses two. + +Also authors propose such modifications of CVAE as Gaussian stochastic neural network and hybrid model. These modifications can be applied to our model as well. Nevertheless, we don’t use them, because of their disadvantage which is described in appendix C. + +# 4 VARIATIONAL AUTOENCODER WITH ARBITRARY CONDITIONING + +# 4.1 PROBLEM STATEMENT + +Consider a distribution $p _ { d } ( x )$ over a $D$ -dimensional vector $x$ with real or categorical components. The components of the vector are called features. + +Let binary vector $b \in \{ 0 , 1 \} ^ { D }$ be the binary mask of unobserved features of the object. Then we describe the vector of unobserved features as $x _ { b } = \{ x _ { i : b _ { i } = 1 } \}$ . For example, $x _ { ( 0 , 1 , 1 , 0 , 1 ) } = ( x _ { 2 } , x _ { 3 } , x _ { 5 } )$ . Using this notation we denote $x _ { 1 - b }$ as a vector of observed features. + +Our goal is to build a model of the conditional distribution $p _ { \psi , \theta } ( x _ { b } | x _ { 1 - b } , b ) \approx p _ { d } ( x _ { b } | x _ { 1 - b } , b )$ for an arbitrary $b$ , where $\psi$ and $\theta$ are parameters that are used in our model at the testing stage. + +However, the true distribution $p _ { d } ( x _ { b } | x _ { 1 - b } , b )$ is intractable without strong assumptions about $p _ { d } ( x )$ . Therefore, our model $p _ { \psi , \theta } ( x _ { b } | x _ { 1 - b } , b )$ has to be more precise for some $b$ and less precise for others. To formalize our requirements about the accuracy of our model we introduce the distribution $p ( b )$ over different unobserved feature masks. The distribution $p ( b )$ is arbitrary and may be defined by the user depending on the problem. Generally it should have full support over $\{ 0 , 1 \} ^ { D }$ so that $p _ { \psi , \theta } ( x _ { b } | x _ { 1 - b } , b )$ can evaluate arbitrary conditioning. Nevertheless, it is not necessary if the model is used for specific kinds of conditioning (as we do in section 5.2). + +Using $p ( b )$ we can introduce the following log-likelihood objective function for the model: + +$$ +\operatorname* { m a x } _ { \psi , \theta } \mathbb { E } _ { p _ { d } ( \boldsymbol { x } ) } \mathbb { E } _ { p ( \boldsymbol { b } ) } \log p _ { \psi , \theta } \big ( x _ { b } | \boldsymbol { x } _ { 1 - \boldsymbol { b } } , \boldsymbol { b } \big ) +$$ + +The special cases of the objective (4) are variational autoencoder $( b _ { i } = 1 \forall i \in \{ 1 , \ldots , D \} )$ and conditional variational autoencoder $^ { \textit { b } }$ is constant). + +# 4.2 MODEL DESCRIPTION + +The generative process of our model is similar to the generative process of CVAE: for each object firstly we generate $z \sim p _ { \psi } ( z | x _ { 1 - b } , b )$ using prior network, and then sample unobserved features $x _ { b } ~ \sim ~ p _ { \theta } ( x _ { b } | z , x _ { 1 - b } , b )$ using generative network. This process induces the following model distribution over unobserved features: + +$$ +p _ { \psi , \theta } ( x _ { b } | x _ { 1 - b } , b ) = \mathbb { E } _ { z \sim p _ { \psi } ( z | x _ { 1 - b } , b ) } p _ { \theta } ( x _ { b } | z , x _ { 1 - b } , b ) +$$ + +We use $z ~ \in ~ \mathbb { R } ^ { d }$ , and Gaussian distribution $p _ { \psi }$ over $z$ , with parameters from a neural network with weights $\psi$ : $p _ { \psi } ( z | x _ { 1 - b } , b , \psi ) = \mathcal { N } ( z | \mu _ { \psi } ( x _ { 1 - b } , b ) , \sigma _ { \psi } ^ { 2 } ( x _ { 1 - b } , b ) I )$ . The real-valued components of distribution $p _ { \theta } ( x _ { b } | \boldsymbol { z } , x _ { 1 - b } , b )$ are defined likewise. Each categorical component $i$ of distribution $p _ { \theta } ( x _ { i } | \boldsymbol { z } , x _ { 1 - b } , b )$ is parameterized by a function $w _ { i , \theta } ( z , x _ { 1 - b } , b )$ , whose outputs are logits of probabilities for each category: $x _ { i } \sim \mathrm { C a t } [ \mathrm { S o f t m a x } ( w _ { i , \theta } ( z , x _ { 1 - b } , b ) ) ]$ . Therefore the components of the latent vector $z$ are conditionally independent given $x _ { 1 - b }$ and $b$ , and the components of $x _ { b }$ are conditionally independent given $z , x _ { 1 - b }$ and $b$ . + +The variables $x _ { b }$ and $x _ { 1 - b }$ have variable length that depends on $b$ . So in order to use architectures such as multi-layer perceptron and convolutional neural network we consider $x _ { 1 - b } = x \circ ( 1 - b )$ where $\circ$ is an element-wise product. So in implementation $x _ { 1 - b }$ has fixed length. The output of the generative network also has a fixed length, but we use only unobserved components to compute likelihood. + +The theoretical analysis of the model is available in appendix B.1. + +# 4.3 LEARNING VARIATIONAL AUTOENCODER WITH ARBITRARY CONDITIONING + +# 4.3.1 VARIATIONAL LOWER BOUND + +We can derive a lower bound for $\log p _ { \psi , \theta } ( x _ { b } | x _ { 1 - b } , b )$ as for variational autoencoder: + +$$ +\begin{array} { r l } & { \log p _ { \psi , \theta } ( x _ { b } | x _ { 1 - b } , b ) = \mathbb { E } _ { q _ { \phi } ( z | x , b ) } \log \frac { p _ { \psi , \theta } ( x _ { b } , z | x _ { 1 - b } , b ) } { q _ { \phi } ( z | x , b ) } + D _ { \mathrm { K L } } ( q _ { \phi } ( z | x , b ) | | p _ { \psi , \theta } ( z | x , b ) ) } \\ & { \qquad \geq \mathbb { E } _ { q _ { \phi } ( z | x , b ) } \log p _ { \theta } ( x _ { b } | z , x _ { 1 - b } , b ) - D _ { \mathrm { K L } } ( q _ { \phi } ( z | x , b ) | | p _ { \psi } ( z | x _ { 1 - b } , b ) ) = L _ { V A E A C } ( x , b ; \theta , \psi , \phi ) } \end{array} +$$ + +Therefore we have the following variational lower bound optimization problem: + +$$ +\operatorname* { m a x } _ { \theta , \psi , \phi } \mathbb { E } _ { p _ { d } ( x ) } \mathbb { E } _ { p ( b ) } L _ { V A E A C } ( x , b ; \theta , \psi , \phi ) +$$ + +We use fully-factorized Gaussian proposal distribution $q _ { \phi }$ which allows us to perform reparameterization trick and compute KL divergence analytically in order to optimize (7). + +# 4.3.2 PRIOR IN LATENT SPACE + +During the optimization of objective (7) the parameters $\mu _ { \psi }$ and $\sigma _ { \psi }$ of the prior distribution of $z$ may tend to infinity, since there is no penalty for large values of those parameters. We usually observe the growth of $\left. z \right. _ { 2 }$ during training, though it is slow enough. To prevent potential numerical instabilities, we put a Normal-Gamma prior on the parameters of the prior distribution to prevent the divergence. Formally, we redefine $p _ { \psi } ( z | x _ { 1 - b } , b )$ as follows: + +$$ +p _ { \psi } ( z , \mu _ { \psi } , \sigma _ { \psi } | x _ { 1 - b } , b ) = \mathcal { N } ( z | \mu _ { \psi } , \sigma _ { \psi } ^ { 2 } ) \mathcal { N } ( \mu _ { \psi } | 0 , \sigma _ { \mu } ) \mathrm { G a m m a } ( \sigma _ { \psi } | 2 , \sigma _ { \sigma } ) +$$ + +As a result, the regularizers $- \frac { \mu _ { \psi } ^ { 2 } } { 2 \sigma _ { \mu } ^ { 2 } }$ and $\sigma _ { \sigma } ( \log ( \sigma _ { \psi } ) - \sigma _ { \psi } )$ are added to the model log-likelihood. Hyperparameter $\sigma _ { \mu }$ is chosen to be large $( 1 0 ^ { 4 } )$ and $\sigma _ { \sigma }$ is taken to be a small positive number $( 1 0 ^ { - 4 } )$ . This distribution is close to uniform near zero, so it doesn’t affect the learning process significantly. + +# 4.3.3 MISSING FEATURES + +The optimization objective (7) requires all features of each object at the training stage: some of the features will be observed variables at the input of the model and other will be unobserved features used to evaluate the model. Nevertheless, in some problem settings the training data contains missing features too. We propose the following slight modification of the problem (7) in order to cover such problems as well. + +The missing values cannot be observed so $x _ { i } = \omega \Rightarrow b _ { i } = 1$ , where $\omega$ describes the missing value in the data. In order to meet this requirement, we redefine mask distribution as conditioned on $x$ : $p ( b )$ turns into $p ( b | x )$ in (4) and (7). In the reconstruction loss (5) we simply omit the missing features, i. e. marginalize them out: + +$$ +\log p _ { \theta } ( x _ { b } | z , x _ { 1 - b } , b ) = \sum _ { \substack { i : b _ { i } = 1 , x _ { i } \neq \omega } } \log p _ { \theta } ( x _ { i } | z , x _ { 1 - b } , b ) +$$ + +The proposal network must be able to determine which features came from real object and which are just missing. So we use additional missing features mask which is fed to proposal network together with unobserved features mask $b$ and object $x$ . + +The proposed modifications are evaluated in section 5.1. + +Table 1: NRMSE (for continuous datasets) or PFC (for categorical ones) of imputations. Less is better. + +
Method/DatasetWhiteWineYeastMushroomZooPhishing
MICE0.964± 0.0071.01 ± 0.010.334± 0.0020.19±0.030.422± 0.006
MissForest0.878 ± 0.0091.02 ± 0.060.249 ± 0.0060.16 ±0.020.422 ± 0.009
GAIN0.97 ± 0.020.99 ± 0.030.271 ± 0.0030.20± 0.020.427 ± 0.010
VAEAC0.850 ± 0.0070.94 ± 0.010.244 ± 0.0020.16 ± 0.020.394± 0.006
+ +# 5 EXPERIMENTS + +In this section we validate the performance of VAEAC using several real-world datasets. In the first set of experiments we evaluate VAEAC missing features imputation performance using various UCI datasets (Lichman, 2013). We compare imputations from our model with imputations from such classical methods as MICE (Buuren & Groothuis-Oudshoorn, 2010) and MissForest (Stekhoven & Buhlmann, 2011) and recently ¨ proposed GANs-based method GAIN (Yoon et al., 2018). In the second set of experiments we use VAEAC to solve image inpainting problem. We show inpainitngs generated by VAEAC and compare our model with models from papers Pathak et al. (2016), Yeh et al. (2017) and Li et al. (2017) in terms of peak signal-to-noise ratio (PSNR) of obtained inpaintings on CelebA dataset (Liu et al., 2015) . And finally, we evaluate VAEAC against the competing method called Universal Marginalizer (Douglas et al., 2017). Additional experiments can be found in appendices C and D. The code is available at https://github.com/tigvarts/ vaeac. + +# 5.1 MISSING FEATURES IMPUTATION + +The datasets with missing features are widespread. Consider a dataset with $D$ -dimensional objects $x$ where each feature may be missing (which we denote by $x _ { i } ~ = ~ \omega$ ) and their target values $y$ . The majority of discriminative methods do not support missing values in the objects. The procedure of filling in the missing features values is called missing features imputation. + +In this section we evaluate the quality of imputations produced by VAEAC. For evaluation we use datasets from UCI repository (Lichman, 2013). Before training we drop randomly $50 \%$ of values both in train and test set. After that we impute missing features using MICE (Buuren & Groothuis-Oudshoorn, 2010), MissForest (Stekhoven & Buhlmann, 2011), GAIN (Yoon et al., 2018) and VAEAC trained on the observed data. The ¨ details of GAIN implementation are described in appendix A.4. + +Our model learns the distribution of the imputations, so it is able to sample from this distribution. We replace each object with missing features by $n = 1 0$ objects with sampled imputations, so the size of the dataset increases by $n$ times. This procedure is called missing features multiple imputation. MICE and GAIN are also capable of multiple imputation (we use $n = 1 0$ for them in experiments as well), but MissForest is not. + +For more details about the experimental setup see appendices A.1, A.2, and A.4. + +In table 1 we report NRMSE (i.e. RMSE normalized by the standard deviation of each feature and then averaged over all features) of imputations for continuous datasets and proportion of falsely classified (PFC) for categorical ones. For multiple imputation methods we average imputations of continuous variables and take most frequent imputation for categorical ones for each object. + +We also learn linear or logistic regression and report the regression or classification performance after applying imputations of different methods in table 2. For multiple imputation methods we average predictions for continuous targets and take most frequent prediction for categorical ones for each object in test set. + +Table 2: R2-score (for continuous targets) or accuracy (for categorical ones) of post-imputation regression or classification. Higher is better. + +
Method /DatasetWhiteWineYeastMushroomZ00Phishing
MICE0.13±0.020.41 ±0.020.92± 0.010.78± 0.050.75 ±0.02
MissForest0.17 ± 0.010.42 ± 0.020.972 ± 0.0030.71 ± 0.070.73 ± 0.02
GAIN0.11 ± 0.010.39 ± 0.060.969 ± 0.0050.67 ± 0.060.74 ± 0.03
VAEAC0.17 ± 0.010.43 ± 0.010.983 ± 0.0020.8 ± 0.10.74 ± 0.02
+ +As can be seen from the tables 1 and 2, VAEAC can learn joint data distribution and use it for missing feature imputation. The imputations are competitive with current state of the art imputation methods in terms of RMSE, PFC, post-imputation regression R2-score and classification accuracy. Nevertheless, we don’t claim that our method is state of the art in missing features imputation; for some datasets MICE or MissForest outperform it. The additional experiments can be found in appendix D.2. + +# 5.2 IMAGE INPAINTING + +The image inpainting problem has a number of different formulations. The formulation of our interest is as follows: some of the pixels of an image are unobserved and we want to restore them in a natural way. Unlike the majority of papers, we want to restore not just one most probable inpainting, but the distribution over all possible inpaintings from which we can sample. This distribution is extremely multi-modal because often there is a lot of different possible ways to inpaint the image. + +Unlike the previous subsection, here we have uncorrupted images without missing features in the training set, so $p ( b | x ) = p ( b )$ . + +As we show in section 2, state of the art results use different adversarial losses to achieve more sharp and realistic samples. VAEAC can be adapted to the image inpainting problem by using a combination of those adversarial losses as a part of reconstruction loss $p _ { \theta } ( x _ { b } | \boldsymbol { z } , x _ { 1 - b } , b )$ . Nevertheless, such construction is out of scope for this research, so we leave it for the future work. In the current work we show that the model can generate both diverse and realistic inpaintings. + +In figures 1, 2, 3 and 4 we visualize image inpaintings produced by VAEAC on binarized MNIST (LeCun et al., 1998), Omniglot (Lake et al., 2015) and CelebA (Liu et al., 2015). The details of learning procedure and description of datasets are available in appendixes A.1 and A.3. + +To the best of our knowledge, the most modern inpainting papers don’t consider the diverse inpainting problem, where the goal is to build diverse image inpaintings, so there is no straightforward way to compare with these models. Nevertheless, we compute peak signal-to-noise ratio (PSNR) for one random inpainting from VAEAC and the best PSNR among 10 random inpaintings from VAEAC. One inpainting might not be similar to the original image, so we also measure how good the inpainting which is most similar to the original image reconstructs it. We compare these two metrics computed for certain masks with the PSNRs for the same masks on CelebA from papers Yeh et al. (2017) and Li et al. (2017). The results are available in tables 3 and 4. + +We observe that for the majority of proposed masks our model outperforms the competing methods in terms of PSNR even with one sample, and for the rest (where the inpaintings are significantly diverse) the best PSNR over 10 inpaintings is larger than the same PSNR of the competing models. Even if PSNR does not reflect completely the visual quality of images and tends to encourage blurry VAE samples instead of realistic GANs samples, the results show that VAEAC is able to solve inpainting problem comparably to the state of the art methods. The disadvantage of VAEAC compared to Yeh et al. (2017) and Li et al. (2017) (but not Pathak et al. (2016)) is that it needs the distribution over masks at the training stage to be similar to the distribution over them at the test stage. However, it is not a very strict limitation for the practical usage. + +Table 3: PSNR of inpaintings for different masks for Context Encoder (Pathak et al., 2016), model from “Semantic Image Inpainting with Deep Generative Models” (Yeh et al., 2017) and VAEAC. Higher is better. + +
Method/MasksCenterPatternRandomHalf
Context Encoder 121.319.220.615.5
SIIDGM 119.417.422.813.7
VAEAC, 1 sample22.121.429.314.9
VAEAC,10 samples23.723.329.317.4
+ +Table 4: PSNR of inpaintings for different masks for Context Encoder (Pathak et al., 2016), model from “Generative Face Completion” (Li et al., 2017) and VAEAC. Higher is better. + +
Method/Masks010203040506
Context Encoder218.618.417.919.019.119.3
GFC ²20.019.818.819.719.520.2
VAEAC,1 sample20.821.019.520.320.321.0
VAEAC,10 samples22.022.220.821.721.822.2
+ +# 5.3 UNIVERSAL MARGINALIZER + +Universal Marginalizer (Douglas et al., 2017) (UM) is a model which uses a single neural network to estimate the marginal distributions over the unobserved features. So it optimizes the following objective: + +$$ +\operatorname* { m a x } _ { \theta } \mathbb { E } _ { x \sim p _ { d } ( x ) } \mathbb { E } _ { b \sim p ( b ) } \sum _ { i = 1 } ^ { D } b _ { i } \log p _ { \theta } \big ( x _ { i } | x _ { 1 - b } , b \big ) +$$ + +For given mask $b$ we fix a permutation of its unobserved components: $( i _ { 1 } , i _ { 2 } , \dots , i _ { | b | } )$ , where $| b |$ is a number of unobserved components. Using the learned model and the permutation we can generate objects from joint distribution and estimate their probability using chain rule. + +$$ +\log p _ { \theta } ( x _ { b } | x _ { 1 - b } , b ) = \sum _ { j = 1 } ^ { | b | } \log p _ { \theta } ( x _ { i _ { j } } | x _ { 1 - ( b - \sum _ { k = 1 } ^ { j - 1 } e _ { i _ { k } } ) } , b - \sum _ { k = 1 } ^ { j - 1 } e _ { i _ { k } } ) +$$ + +For example, $p _ { \theta } ( x _ { 1 } , x _ { 4 } , x _ { 5 } | x _ { 2 } , x _ { 3 } ) = p _ { \theta } ( x _ { 4 } | x _ { 2 } , x _ { 3 } ) p _ { \theta } ( x _ { 1 } | x _ { 2 } , x _ { 3 } , x _ { 4 } ) p _ { \theta } ( x _ { 5 } | x _ { 1 } , x _ { 2 } , x _ { 3 } , x _ { 4 } ) .$ + +Conditional sampling or conditional likelihood estimation for one object requires $| b |$ requests to UM to compute $p _ { \theta } ( x _ { i } | x _ { 1 - b } , b )$ . Each request is a forward pass through the neural network. In the case of conditional sampling those requests even cannot be paralleled because the input of the next request contains the output of the previous one. + +We propose a slight modification of the original UM training procedure which allows learning UM efficiently for any kind of masks including those considered in this paper. The details of the modification are described in appendix B.3. + +![](images/d8e3304c44f6641f3fcf46844d8ec076bf559af92021199814e73a3e8e86af57.jpg) +Figure 1: MNIST inpaintings. + +![](images/696a7457eff5fb7c33faf339e1df45fda1cdfba4264c7adccc10fee5a551d327.jpg) +Figure 2: Omniglot inpaintings. + +![](images/22952789d3ea86e1798ed0fcb93d6b5ec6a1f7740b49980530ccffa387686d74.jpg) +Figure 3: CelebA inpaintings. + +![](images/1a5cbe58c9eb61a7543d31d330e9a0fdd262a2099e64f42864b8627f3f719b25.jpg) +Figure 4: CelebA inpaintings with masks from (Yeh et al., 2017). + +Left: input. The gray pixels are unobserved. Middle: samples from VAEAC. Right: ground truth. + +Table 5: VAEAC and UM comparison on MNIST. + +
MethodVAEACUM
Negative log-likelihood6141
Training time (30 epochs)5min 47s3min 14s
Test time (1OO samples generation)0.7ms1s
+ +The results of using this modification of UM are provided in table 5. We can say that the relation between VAEAC and UM is similar to the relation between VAE and PixelCNN. The second one is much slower at the testing stage, but it easily takes into account local dependencies in data while the first one is faster but assumes conditional independence of the outputs. Nevertheless, there are a number of cases where UM cannot learn the distribution well while VAEAC can. For example, when the data is real-valued and marginal distributions have many local optima, there is no straightforward parametrization which allows UM to approximate them, and, therefore also the conditioned joint distribution. An example of such distribution and more illustrations for comparison of VAEAC and UM are available in appendix D.5. + +# 6 CONCLUSION + +In this paper we consider the problem of simultaneous learning of all conditional distributions for a vector. This problem has a number of different special cases with practical applications. We propose neural network based probabilistic model for distribution conditioning learning with Gaussian latent variables. This model is scalable and efficient in inference and learning. We propose several tricks to improve optimization and give recommendations about hyperparameters choice. The model is successfully applied to feature imputation and inpainting tasks. The experimental results show that the model is competitive with state of the art methods for both missing features imputation and image inpainting problems. + +# REFERENCES + +Philip Bachman and Doina Precup. Data generation as sequential decision making. In Advances in Neural Information Processing Systems, pp. 3249–3257, 2015. + +Marcelo Bertalmio, Guillermo Sapiro, Vincent Caselles, and Coloma Ballester. Image inpainting. In Proceedings of the 27th Annual Conference on Computer Graphics and Interactive Techniques, SIGGRAPH ’00, pp. 417–424, New York, NY, USA, 2000. ACM Press/Addison-Wesley Publishing Co. ISBN 1- 58113-208-5. + +Florian Bordes, Sina Honari, and Pascal Vincent. Learning to generate samples from noise through infusion training. In International Conference on Learning Representations, 2017. + +Yuri Burda, Roger Grosse, and Ruslan Salakhutdinov. Importance weighted autoencoders. arXiv preprint arXiv:1509.00519, 2015. + +S van Buuren and Karin Groothuis-Oudshoorn. mice: Multivariate imputation by chained equations in r. Journal of statistical software, pp. 1–68, 2010. + +Laura Douglas, Iliyan Zarov, Konstantinos Gourgoulias, Chris Lucas, Chris Hart, Adam Baker, Maneesh Sahani, Yura Perov, and Saurabh Johri. A universal marginalizer for amortized inference in generative models. arXiv preprint arXiv:1711.00695, 2017. + +Mathieu Germain, Karol Gregor, Iain Murray, and Hugo Larochelle. Made: Masked autoencoder for distribution estimation. In Francis Bach and David Blei (eds.), Proceedings of the 32nd International Conference on Machine Learning, volume 37 of Proceedings of Machine Learning Research, pp. 881–889, Lille, France, 07–09 Jul 2015. PMLR. URL http://proceedings.mlr.press/v37/germain15. html. + +Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in neural information processing systems, pp. 2672–2680, 2014. + +Anirudh Goyal Alias Parth Goyal, Nan Ke, Surya Ganguli, and Yoshua Bengio. Variational walkback: Learning a transition operator as a stochastic recurrent net. In Advances in Neural Information Processing Systems, pp. 4392–4402, 2017. + +Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770–778, 2016. + +Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014. + +Diederik P. Kingma and Max Welling. Auto-encoding variational bayes. CoRR, abs/1312.6114, 2013. + +Brenden M Lake, Ruslan Salakhutdinov, and Joshua B Tenenbaum. Human-level concept learning through probabilistic program induction. Science, 350(6266):1332–1338, 2015. + +Yann LeCun, Corinna Cortes, and Christopher JC Burges. The mnist database of handwritten digits, 1998. + +Yijun Li, Sifei Liu, Jimei Yang, and Ming-Hsuan Yang. Generative face completion. In The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), volume 1, pp. 3, 2017. + +M. Lichman. UCI machine learning repository, 2013. URL http://archive.ics.uci.edu/ml. + +Ziwei Liu, Ping Luo, Xiaogang Wang, and Xiaoou Tang. Deep learning face attributes in the wild. In Proceedings of International Conference on Computer Vision (ICCV), December 2015. + +Alireza Makhzani, Jonathon Shlens, Navdeep Jaitly, and Ian Goodfellow. Adversarial autoencoders. In International Conference on Learning Representations, 2016. URL http://arxiv.org/abs/1511. 05644. + +Xiaojiao Mao, Chunhua Shen, and Yu-Bin Yang. Image restoration using very deep convolutional encoderdecoder networks with symmetric skip connections. In Advances in neural information processing systems, pp. 2802–2810, 2016. + +Mehdi Mirza and Simon Osindero. Conditional generative adversarial nets. CoRR, abs/1411.1784, 2014. + +Deepak Pathak, Philipp Krahenb ¨ uhl, Jeff Donahue, Trevor Darrell, and Alexei Efros. Context encoders: ¨ Feature learning by inpainting. In Computer Vision and Pattern Recognition (CVPR), 2016. + +Danilo Jimenez Rezende, Shakir Mohamed, and Daan Wierstra. Stochastic backpropagation and approximate inference in deep generative models. In International Conference on Machine Learning, pp. 1278– 1286, 2014. + +Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical image computing and computer-assisted intervention, pp. 234–241. Springer, 2015. + +Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In Francis Bach and David Blei (eds.), Proceedings of the 32nd International Conference on Machine Learning, volume 37 of Proceedings of Machine Learning Research, pp. 2256–2265, Lille, France, 07–09 Jul 2015. PMLR. URL http://proceedings.mlr. press/v37/sohl-dickstein15.html. + +Kihyuk Sohn, Honglak Lee, and Xinchen Yan. Learning structured output representation using deep conditional generative models. In C. Cortes, N. D. Lawrence, D. D. Lee, M. Sugiyama, and R. Garnett (eds.), Advances in Neural Information Processing Systems 28, pp. 3483–3491. Curran Associates, Inc., 2015. + +Casper Kaae Sønderby, Tapani Raiko, Lars Maaløe, Søren Kaae Sønderby, and Ole Winther. Ladder variational autoencoders. In Advances in neural information processing systems, pp. 3738–3746, 2016. + +Daniel J Stekhoven and Peter Buhlmann. Missforest - non-parametric missing value imputation for mixed-¨ type data. Bioinformatics, 28(1):112–118, 2011. + +Benigno Uria, Marc-Alexandre Cotˆ e, Karol Gregor, Iain Murray, and Hugo Larochelle. Neural autoregres- ´ sive distribution estimation. The Journal of Machine Learning Research, 17(1):7184–7220, 2016. + +C. Yang, X. Lu, Z. Lin, E. Shechtman, O. Wang, and H. Li. High-resolution image inpainting using multiscale neural patch synthesis. In 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 4076–4084, July 2017. doi: 10.1109/CVPR.2017.434. + +Raymond Yeh, Chen Chen, Teck-Yian Lim, Mark Hasegawa-Johnson, and Minh N. Do. Semantic image inpainting with perceptual and contextual losses. CoRR, abs/1607.07539, 2016. + +Raymond A Yeh, Chen Chen, Teck Yian Lim, Alexander G Schwing, Mark Hasegawa-Johnson, and Minh N Do. Semantic image inpainting with deep generative models. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 5485–5493, 2017. + +Jinsung Yoon, James Jordon, and Mihaela van der Schaar. GAIN: Missing data imputation using generative adversarial nets. In Jennifer Dy and Andreas Krause (eds.), Proceedings of the 35th International Conference on Machine Learning, volume 80 of Proceedings of Machine Learning Research, pp. 5689–5698, Stockholmsmssan, Stockholm Sweden, 10–15 Jul 2018. PMLR. URL http://proceedings.mlr. press/v80/yoon18a.html. + +# APPENDIX + +# A EXPERIMENTAL DETAILS + +A.1 NEURAL NETWORK ARCHITECTURES + +In all experiments we use optimization method Adam (Kingma & Ba, 2014), skip-connections between prior network and generative network inspired by (Mao et al., 2016), (Sønderby et al., 2016) and (Ronneberger et al., 2015), and convolutional neural networks based on ResNet blocks (He et al., 2016). + +Without skip-connections all information for decoder goes through the latent variables. In image inpainting we found skip-connections very useful in both terms of log-likelihood improvement and the image realism, because latent variables are responsible for the global information only while the local information passes through skip-connections. Therefore the border between image and inpainting becomes less conspicuous. + +The main idea of neural networks architecture is reflected in figure 5. + +![](images/5b67d91960752eb18692b5b61ef536a092947e8d387abb82437625c2d41f393a.jpg) +Figure 5: Neural network architecture for inpainting. + +The number of hidden layers, their widths and structure may be different. + +The neural networks we used for image inpainting have He-Uniform initialization of convolutional ResNet blocks, and the skip-connections are implemented using concatenation, not addition. The proposal network structure is exactly the same as the prior network except skip-connections. + +Also one could use much simpler fully-connected networks with one hidden layer as a proposal, prior and generative networks in VAEAC and still obtain nice inpaintings on MNIST. + +# A.2 MISSING FEATURES IMPUTATION + +We split the dataset into train and test set with size ratio 3:1. Before training we drop randomly $50 \%$ of values both in train and test set. We repeat each experiment 5 times with different train-test splits and dropped features and then average results and compute their standard deviation. + +As we show in appendix B.2, the better results can be achieved when the model learns the concatenation of objects features $x$ and targets $y$ . So we treat $y$ as an additional feature that is always unobserved during the testing time. + +To train our model we use distribution $p ( b _ { i } | x )$ in which $p ( b _ { i } | x _ { i } = \omega ) = 1$ and $p ( b _ { i } | x ) = 0 . 2$ otherwise. Also for VAEAC trainig we normalize real-valued features, fix $\sigma _ { \theta } = 1$ in the generative model of VAEAC in order to optimize RMSE, and use $2 5 \%$ of training data as validation set to select the best model among all epochs of training. + +For the test set, the classifier or regressor is applied to each of the $n$ imputed objects and the predictions are combined. For regression problems we report R2-score of combined predictions, so we use averaging as a combination method. For classification problem we report accuracy, and therefore choose the mode. We consider the workflow where the imputed values of $y$ are not fed to the classifier or regressor to make a fair comparison of feature imputation quality. + +Table 6: Generative Face Completion (Li et al., 2017) masks. Image size is 128x128. + +
MaskMeaningX1x2y1y2
01Left half of the face337052115
02Right half of the face577095115
03Two eyes29985273
04Left eye29665273
05Right eye61995273
06Lower half of the face408786123
+ +NRMSE or PFC for dataset is computed as an average of NRMSE or PFC of all features of this dataset. NRMSE of a feature is just RMSE of imputations divided by the standard deviation of this feature. PFC of a feature is a proportion of imputations which are incorrect. + +# A.3 IMAGE INPAINTING DATASETS AND MASKS + +MNIST is a dataset of 60000 train and 10000 test grayscale images of digits from 0 to 9 of size $2 8 \mathbf { x } 2 8$ . We binarize all images in the dataset. For MNIST we consider Bernoulli log-likelihood as the reconstruction loss: $\begin{array} { r } { \log p _ { \theta } ( x _ { b } | z , x _ { 1 - b } , b ) = \sum _ { i : b _ { i } = 1 } \log \mathrm { B e r n o u l l i } ( x _ { i } | p _ { \theta , i } ( z , x _ { 1 - b } , b ) ) } \end{array}$ where $p _ { \theta , i } ( z , x _ { 1 - b } , b )$ is an output of the generative neural network. We use 16 latent variables. In the mask for this dataset the observed pixels form a three pixels wide horizontal line which position is distributed uniformly. + +Omniglot is a dataset of 19280 train and 13180 test black-and-white images of different alphabets symbols of size $1 0 5 \mathrm { x } 1 0 5$ . As in previous section, the brightness of each pixel is treated as a Bernoulli probability of it to be 1. The mask we use is a random rectangular which is described below. We use 64 latent variables. We train model for 50 epochs and choose best model according to IWAE log-likelihood estimation on the validation set after each epoch. + +CelebA is a dataset of 162770 train, 19867 validation and 19962 test color images of faces of celebrities of size $1 7 8 \mathrm { x } 2 1 8$ . Before learning we normalize the channels in dataset. We use logarithm of fully-factorized Gaussian distribution as reconstruction loss. The mask we use is a random rectangular which is describe below. We use 32 latent variables. + +Rectangular mask is the common shape of unobserved region in image inpainting. We use such mask for Omniglot and Celeba. We sample the corner points of rectangles uniprobably on the image, but reject those rectangles which area is less than a quarter of the image area. + +In Li et al. (2017) six different masks O1–O6 are used on the testing stage. We reconstruct the positions of masks from the illustrations in the paper and give their coordinates in table 6. The visualizations of the masks are available in figure 10. + +At the training stage we used a rectangle mask with uniprobable random corners. We reject masks with width or height less than 16pt. We use 64 latent variables and take the best model over 50 epochs based on the validation IWAE log-likelihood estimation. We can obtain slightly higher PSNR values than reported in table 4 if use only masks O1–O6 at the training stage. + +In Yeh et al. (2017) four types of masks are used. Center mask is just an unobserved $3 2 \mathrm { x } 3 2 $ square in the center of 64x64 image. Half mask mean that one of upper, lower, left or right half of the image is unobserved. All these types of a half are equiprobable. Random mask means that we use pixelwise-independent Bernoulli distribution with probability 0.8 to form a mask of unobserved pixels. Pattern mask is proposed in Pathak et al. (2016). As we deduced from the code 3, the generation process is follows: firstly we generate $6 0 0 \times 6 0 0$ one-channel image with uniform distribution over pixels, then bicubically interpolate it to image of size $1 0 0 0 0 \mathrm { x } 1 0 0 0 0$ , and then apply Heaviside step function $H ( x - 0 . 2 5 )$ (i. e. all points with value less than 0.25 are considered as unobserved). To sample a mask we sample a random position in this $1 0 0 0 0 \mathrm { x } 1 0 0 0 0$ binary image and crop $6 4 \mathrm { x } 6 4$ mask. If less than $20 \%$ or more than $30 \%$ of pixel are unobserved, than the mask is rejected and the position is sampled again. In comparison with this paper in section 5.2 we use the same distribution over masks at training and testing stages. We use VAEAC with 64 latent variables and take the best model over 50 epochs based on the validation IWAE log-likelihood estimation. + +# A.4 GAIN IMPLEMENTATION DETAILS + +For missing feature imputation we reimplemented GAIN in PyTorch based on the paper (Yoon et al., 2018) and the available TensorFlow source code for image inpainting 4. + +For categorical features we use one-hot encoding. We observe in experiments that it works better in terms of NRMSE and PFC than processing categorical features in GAIN as continuous ones and then rounding them to the nearest category. + +For categorical features we also use reconstruction loss $\begin{array} { r } { L _ { M } ( x _ { i } , x _ { i } ^ { \prime } ) = - \frac { 1 } { | X _ { i } | } \sum _ { j = 1 } ^ { | X _ { i } | } x _ { i , j } \log ( x _ { i , j } ^ { \prime } ) } \end{array}$ $\left| X _ { i } \right|$ the number of categories of the $i$ -th feature, and $x _ { i , j }$ is the $j$ -th component of one-hot encoding of the feature $x _ { i }$ . Such $L _ { M }$ enforces equal contribution of each categorical feature into the whole reconstruction loss. + +We use one more modification of $L _ { M } ( x , x ^ { \prime } )$ for binary and categorical features. Cross-entropy loss in $L _ { M }$ penalizes incorrect reconstructions of categorical and binary features much more than incorrect reconstructions for continuous ones. To avoid such imbalance we mixed L2 and cross-entropy reconstruction losses for binary and categorical features with weights 0.8 and 0.2 respectively: + +We observe in experiments that this modification also works better in terms of NRMSE and PFC than the original model. + +We use validation set which contains $5 \%$ of the observed features for the best model selection (hyperparameter is the number of iterations). + +In the original GAIN paper authors propose to use cross-validation for hyper-parameter $\alpha \in \{ 0 . 1 , 0 . 5 , 1 , 2 , 1 0 \}$ . We observe that using $\alpha ~ = ~ 1 0$ and a hint $h \ = \ b \circ \ m \ + \ 0 . 5 ( 1 \ - \ b )$ where vector $b$ is sampled from Bernoulli distribution with $p = 0 . 0 1$ provides better results in terms of NRMSE and PFC than the original model with every $\alpha \in \{ 0 . 1 , 0 . 5 , 1 , 2 , 1 0 \}$ . Such hint distribution makes model theoretically inconsistent but works well in practice (see table 7). + +Table 7 shows that our modifications provide consistently not worse or even better imputations than the original GAIN (in terms of NRMSE and PFC, on the considered datasets). So in this paper for the missing feature imputation problem we report the results of our modification of GAIN. + +Table 7: NRMSE (for continuous datasets) or PFC (for categorical ones) of imputations for different GAIN modifications. Less is better. “Our modification” includes the reconstruction loss $L _ { M } ^ { \prime }$ (12), Bernoulli distribution over $b$ in the hint generation procedure, and fixed $\alpha = 1 0$ . Other columns refers original GAIN without these modifications and with different values of $\alpha$ . + +
DatasetOur modificationα=10α=2α=1α= 0.5α=0.1
Boston0.78±0.030.87±0.021.0 ± 0.11.0 ± 0.11.02 ± 0.051.6±0.2
Breast0.67 ± 0.010.80±0.051.00 ± 0.051.10 ± 0.071.19 ± 0.051.52 ± 0.06
Concrete0.96 ± 0.010.98 ± 0.021.02 ± 0.021.13 ± 0.061.17 ± 0.041.3 ± 0.1
Diabetes0.911 ± 0.0090.93 ±0.031.05 ± 0.041.07 ± 0.071.21 ± 0.071.6 ± 0.1
Digits0.79 ± 0.020.88 ± 0.011.05 ± 0.021.13 ± 0.021.24 ± 0.081.4± 0.2
Glass1.06 ± 0.051.04 ± 0.051.19 ± 0.061.4 ± 0.21.6 ± 0.11.81 ± 0.10
Iris0.72 ±0.040.73±0.060.83 ±0.080.97 ± 0.091.2 ± 0.21.3±0.2
Mushroom0.271 ± 0.0030.404 ± 0.0040.52 ± 0.050.55 ± 0.010.56 ± 0.030.64± 0.06
Orthopedic0.91 ± 0.030.91 ±0.081.1 ± 0.11.2 ± 0.11.34 ± 0.081.6 ± 0.2
Phishing0.427 ± 0.0100.52 ±0.020.54±0.020.543 ± 0.0100.56 ± 0.010.57 ± 0.04
WallRobot0.907 ± 0.0050.924± 0.0050.933 ± 0.0080.95 ± 0.011.00 ± 0.021.26 ± 0.04
WhiteWine0.97±0.021.02 ± 0.041.2 ± 0.11.3 ± 0.11.6 ± 0.11.86 ± 0.08
Yeast0.99 ±0.031.3±0.21.6 ± 0.11.83 ± 0.091.9 ±0.12.4± 0.4
Zoo0.20 ±0.020.24± 0.050.35 ± 0.060.36 ± 0.030.43 ± 0.040.433 ± 0.004
+ +# B THEORY + +# B.1 VAEAC UNIVERSALITY + +The theoretical guarantees that VAEAC can model arbitrary distribution are based on the same guarantees for Condtitional Variational Autoencoder (CVAE). We prove below that if CVAE can model each of the conditional distributions $p ( x _ { b } | x _ { 1 - b } )$ , then VAEAC can model all of them. + +We can imagine $2 ^ { D }$ CVAEs learned each for the certain mask. Because neural networks are universal approximators, VAEAC networks could model the union of CVAE networks, so that VAEAC network performs transformation defined by the same network of the corresponding to the given mask CVAE. + +$$ +p _ { \psi , V A E A C } ( z | x _ { 1 - b } , b ) = p _ { \psi , C V A E , 1 - b } ( z | x _ { 1 - b } ) \forall x , b +$$ + +$$ +p _ { \theta , V A E A C } ( x _ { b } | z , x _ { 1 - b } , b ) = p _ { \theta , C V A E , 1 - b } ( x _ { b } | z , x _ { 1 - b } ) \forall z , x , b +$$ + +So if CVAE models any distribution $p ( x | y )$ , VAEAC also do. + +The guarantees for CVAE in the case of continuous variables are based on the point that every smooth distribution can be approximated with a large enough mixture of Gaussians, which is a special case of CVAE’s generative model. These guarantees can be extended on the case of categorical-continuous variables also. Actually, there are distributions over categorical variables which CVAE with Gaussian prior and proposal distributions cannot learn. Nevertheless, this kind of limitation is not fundamental and is caused by poor proposal distribution family. + +# B.2 WHY VAEAC NEEDS TARGET VALUES FOR MISSING FEATURES IMPUTATION? + +Consider a dataset with $D$ -dimensional objects $x$ where each feature may be missing (which we denote by $x _ { i } = \omega$ ) and their target values $y$ . In this section we show that the better results are achieved when our model learns the concatenation of objects features $x$ and targets $y$ . The example that shows the necessity of it is following. Consider a dataset where $x _ { 1 } = 1$ , $x _ { 2 } \sim \mathcal { N } ( \bar { x } _ { 2 } | y , 1 )$ , $p _ { d } ( y = \mathrm { { 0 } ) = { { p } ( y = 5 ) = 0 . 5 } }$ . In this case $p _ { d } ( x _ { 2 } | x _ { 1 } = 1 ) = 0 . 5 \mathcal { N } ( x _ { 2 } | 0 , 1 ) + 0 . 5 \mathcal { N } ( x _ { 2 } | 5 , 1 )$ . We can see that generating data from $p _ { d } ( x _ { 2 } | x _ { 1 } )$ may only confuse the classifier, because with probability 0.5 it generates $x _ { 2 } \sim \bar { \mathcal { N } } ( 0 , 1 )$ for $y = 5$ and $x _ { 2 } \sim \mathcal { N } ( 5 , 1 )$ for $y = 0$ . On the other hand, $p _ { d } ( x _ { 2 } | x _ { 1 } , y ) = \mathcal { N } ( x _ { 2 } | y , 1 )$ . Filling gaps using $p _ { d } ( x _ { 2 } | x _ { 1 } , y )$ may only improve classifier or regressor by giving it some information from the joint distribution $p _ { d } ( x , y )$ and thus simplifying the dependence to be learned at the training time. So we treat $y$ as an additional feature that is always unobserved during the testing time. + +# B.3 UNIVERSAL MARGINALIZER: TRAINING PROCEDURE MODIFICATION + +The problem authors did not address in the original paper is the relation between the distribution of unobserved components $p ( b )$ at the testing stage and the distribution of masks in the requests to UM ${ \hat { p } } ( b )$ . The distribution over masks $p ( b )$ induces the distribution ${ \hat { p } } ( b )$ , and in the most cases $p ( b ) \neq { \hat { p } } ( b )$ . The distribution ${ \hat { p } } ( b )$ also depends on the permutations $( i _ { 1 } , i _ { 2 } , \dots , i _ { | b | } )$ that we use to generate objects. + +We observed in experiments, that UM must be trained using unobserved mask distribution ${ \hat { p } } ( b )$ . For example, if all masks from $p ( b )$ have a fixed number of unobserved components (e. g., $\begin{array} { l } { { \frac { D } { 2 } } } \end{array}$ ), then UM will never see an example of mask with $\begin{array} { r } { { 1 , 2 , \ldots , \frac { D } { 2 } - 1 } } \end{array}$ unobserved components, which is necessary to generate a sample conditioned on $\textstyle { \frac { D } { 2 } }$ components. That leads to drastically low likelihood estimate for the test set and unrealistic samples. + +We developed an easy generative process for ${ \hat { p } } ( b )$ for arbitrary $p ( b )$ if the permutation of unobserved components $( i _ { 1 } , i _ { 2 } , \dots , i _ { | b | } )$ is chosen randomly and equiprobably: firstly we generate $b _ { 0 } \sim p ( b )$ , $u \sim U [ 0 , 1 ]$ , then $b _ { 1 } \sim ( \mathrm { B e r n o u l l i } ( u ) ) ^ { D }$ and $b = b _ { 0 } \circ b _ { 1 }$ . More complicated generative process exists for a sorted permutation where $i _ { j - 1 } < i _ { j } \forall j : 2 \le j \le | b |$ . + +In experiments we use uniform distribution over the permutations. + +# C GAUSSIAN STOCHASTIC NEURAL NETWORK + +Gaussian stochastic neural network (13) and hybrid model (14) are originally proposed in the paper on Conditional VAE (Sohn et al., 2015). The motivation authors mention in the paper is as follows. During training the proposal distribution $q _ { \phi } ( z | x , y )$ is used to generate the latent variables $z$ , while during the testing stage the prior $p _ { \psi } ( z | y )$ is used. KL divergence tries to close the gap between two distributions but, according to authors, it is not enough. To overcome the issue authors propose to use a hybrid model (14), a weighted mixture of variational lower bound (3) and a single-sample Monte-Carlo estimation of log-likelihood (13). The model corresponding to the second term is called Gaussian Stochastic Neural Network (13), because it is a feed-forward neural network with a single Gaussian stochastic layer in the middle. Also GSNN is a special case of CVAE where $q _ { \phi } ( z | x , y ) = p _ { \psi } ( z | y )$ . + +$$ +\begin{array} { r } { L _ { G S N N } ( x , y ; \theta , \psi ) = \mathbb { E } _ { p _ { \psi } ( z | y ) } \log p _ { \theta } ( x | z , y ) \qquad } \\ { L ( x , y ; \theta , \psi , \phi ) = \alpha L _ { C V A E } ( x , y ; \theta , \psi , \phi ) + ( 1 - \alpha ) L _ { G S N N } ( x , y ; \theta , \psi ) , \quad \alpha \in [ 0 , 1 ] } \end{array} +$$ + +Authors report that hybrid model and GSNN outperform CVAE in terms of segmentation accuracy on the majority of datasets. + +We can also add that this technique seems to soften the “holes problem” (Makhzani et al., 2016). In Makhzani et al. (2016) authors observe that vectors $z$ from prior distribution may be different enough from all vectors $z$ from the proposal distribution at the training stage, so the generator network may be confused at the testing stage. Due to this problem CVAE can have good reconstructions of $y$ given $z \sim q _ { \phi } ( z | x , y )$ , while samples of $y$ given $z \sim p _ { \psi } ( z | x )$ are not realistic. + +The same trick is applicable to our model as well: + +$$ +\begin{array} { r l } & { L _ { G S N N } ( x , b ; \theta , \psi ) = \mathbb { E } _ { p _ { \psi } ( z | x _ { 1 - b } , b ) } \log p _ { \theta } ( x _ { b } | z , x _ { 1 - b } , b ) } \\ & { L ( x , b ; \theta , \psi , \phi ) = \alpha L _ { V A E A C } ( x , b ; \theta , \psi , \phi ) + ( 1 - \alpha ) L _ { G S N N } ( x , b ; \theta , \psi ) , \quad \alpha \in [ 0 , 1 ] } \end{array} +$$ + +In order to reflect the difference between sampling $z$ from prior and proposal distributions, authors of CVAE use two methods of log-likelihood estimation: + +$$ +\log p _ { \theta , \psi } ( x | y ) \approx \log \frac { 1 } { S } \sum _ { i = 1 } ^ { S } p _ { \theta } ( x | z _ { i } , y ) , ~ z _ { i } \sim p _ { \psi } ( z | y ) +$$ + +$$ +\log p _ { \theta , \psi } ( x | y ) \approx \log \frac { 1 } { S } \sum _ { i = 1 } ^ { S } \frac { p _ { \theta } ( x | z _ { i } , y ) p _ { \psi } ( z _ { i } | y ) } { q _ { \phi } ( z _ { i } | x , y ) } , ~ z _ { i } \sim q _ { \phi } ( z | x , y ) +$$ + +The first estimator is called Monte-Carlo estimator and the second one is called Importance Sampling estimator (also known as IWAE). They are asymptotically equivalent, but in practice the Monte-Carlo estimator requires much more samples to obtain the same accuracy of estimation. Small $S$ leads to underestimation of the log-likelihood for both Monte-Carlo and Importance Sampling (Burda et al., 2015), but for Monte-Carlo the underestimation is expressed much stronger. + +We perform an additional study of GSNN and hybrid model and show that they have drawbacks when the target distribution $p ( x | y )$ is has multiple different local maximums. + +# C.1 THEORETICAL STUDY + +In this section we show why GSNN cannot learn distributions with several different modes and leads to a blurry image samples. + +For the simplicity of the notation we consider hybrid model for a standard VAE: + +$$ +L ( x ; \phi , \psi , \theta ) = \alpha \mathbb { E } _ { z \sim q _ { \phi } ( z \mid x ) } \log \frac { p _ { \theta } ( x \mid z ) p _ { \psi } ( z ) } { q _ { \phi } ( z \mid x ) } + ( 1 - \alpha ) \mathbb { E } _ { z \sim p _ { \psi } ( z ) } \log p _ { \theta } ( x \mid z ) +$$ + +The hybrid model (16) for VAEAC can be obtained from (19) by replacing $x$ with $x _ { b }$ and conditioning all distributions on $x _ { 1 - b }$ and $b$ . The validity of the further equations and conclusions remains for VAEAC after this replacement. + +Consider now a categorical latent variable $z$ which can take one of $K$ values. Let $x$ be a random variable witfor $p _ { d } ( x )$ to be modele some values following true data distribut. So the true distribution has n: d $\begin{array} { r } { p _ { d } ( x = x _ { i } ) = \frac { 1 } { K } } \end{array}$ $i \in \{ 1 , 2 , \ldots , K \}$ $x _ { 1 } , x _ { 2 } , \dotsc , x _ { K }$ $K$ able modes. Suppose the generator network $N N _ { \theta }$ which models mapping from $z$ to some vector of parameters $\begin{array} { r l r } { v _ { z } } & { { } = } & { \bar { N } N _ { \theta } ( z ) } \end{array}$ . Thus, we define generative distribution as some function of these parameters: $p _ { \theta } ( x | z ) \ : = \ : f ( x , v _ { z } )$ . Therefore, the parameters $\theta$ are just the set of $v _ { 1 } , v _ { 2 } , \dotsc , v _ { K }$ . + +For the simplicity of the model we assume $\begin{array} { r } { p _ { \psi } ( z ) = \frac { 1 } { K } } \end{array}$ . Taking into account $\begin{array} { r } { p _ { \psi } ( z ) = \frac { 1 } { K } } \end{array}$ , we obtain optimal $\begin{array} { r } { q ( z = i | x ) = \frac { f ( x , v _ { i } ) } { \sum _ { j = 1 } ^ { K } f ( x , v _ { j } ) } } \end{array}$ Using (19) and the above formulas for $q _ { \phi } , p _ { \psi }$ and $p _ { \theta }$ we obtain the following optimization problem: + +$$ +\operatorname* { m a x } _ { v _ { 1 } , v _ { 2 } , \ldots , v _ { K } } \frac { 1 } { K } \sum _ { i = 1 } ^ { K } \left[ \alpha \sum _ { j = 1 } ^ { K } \frac { f ( x _ { i } , v _ { j } ) } { \sum _ { k = 1 } ^ { K } f ( x _ { i } , v _ { k } ) } \log \frac { f ( x _ { i } , v _ { j } ) \frac { 1 } { K } } { \frac { f ( x _ { i } , v _ { j } ) } { \sum _ { k = 1 } ^ { K } f ( x _ { i } , v _ { k } ) } } + ( 1 - \alpha ) \sum _ { j = 1 } ^ { K } \frac { 1 } { K } \log f ( x _ { i } , v _ { j } ) \right] +$$ + +Table 8: Negative log-likelihood estimation of a hybrid model on the synthetic data. IS- $S$ refers to Importance Sampling log-likelihood estimation with $S$ samples for each object (18). MC- $S$ refers to Monte-Carlo log-likelihood estimation with $S$ samples for each object (17). + +
VAEAC weightIS-10MC-10
a=10.2285
α = 0.990.3511
α = 0.90.621.7
+ +It is easy to show that (20) is equivalent to the following optimization problem: + +$$ +\operatorname* { m a x } _ { v _ { 1 } , v _ { 2 } , \ldots , v _ { K } } \sum _ { i = 1 } ^ { K } \left[ \alpha \log \frac { \sum _ { j = 1 } ^ { K } f ( x _ { i } , v _ { j } ) } { K } + ( 1 - \alpha ) \sum _ { j = 1 } ^ { K } \frac { 1 } { K } \log f ( x _ { i } , v _ { j } ) \right] +$$ + +It is clear from (21) that when $\alpha = 1$ the log-likelihood of the initial model is optimized. On the other hand, when influe $\alpha = 0$ the optimal point is generative process, a $v _ { 1 } = v _ { 2 } = \cdots = v _ { K } = \operatorname { a r g m a x } _ { v } \sum _ { i = 1 } ^ { K } \log f ( x _ { i } , v ) .$ , i. e. mizes $z$ doesn’telihood $z$ $v$ +estimation of the generative model $f ( x , v )$ for the given dataset of $x$ ’s. For Bernoulli and Gaussian generative distributions $f$ such $v$ is just average of all modes $x _ { 1 } , x _ { 2 } , \dotsc , x _ { K }$ . That explains why further we observe blurry images when using GSNN model. + +The same conclusion holds for for continuous latent variables instead of categorical. Given $K$ different modes in true data distribution, VAE uses proposal network to separate prior distribution into $K$ components (i. e. regions in the latent space), so that each region corresponds to one mode. On the other hand, in GSNN $z$ is sampled independently on the mode which is to be reconstructed from it, so for each $z$ the generator have to produce parameters suitable for all modes. + +From this point of view, there is no difference between VAE and VAEAC. If the true conditional distribution has several different modes, then VAEAC can fit them all, while GSNN learns their average. If true conditional distribution has one mode, GSNN and VAEAC are equal, and GSNN may even learn faster because it has less parameters. + +Hybrid model is a trade-off between VAEAC and GSNN: the closer $\alpha$ to zero, the more blurry and closer to the average is the distribution of the model. The exact dependence of the model distribution on $\alpha$ can be derived analytically for the simple data distributions or evaluated experimentally. We perform such experimental evaluation in the next sections. + +# C.2 SYNTHETIC DATA + +In this section we show that VAEAC is capable of learning a complex multimodal distribution of synthetic +data while GSNN and hybrid model are not. Let $x \in \mathbb { R } ^ { \bar { 2 } }$ and $p ( \bar { b } _ { 1 } = 1 ) = p ( b _ { 2 } = 1 ) = 0 . 5$ . $p _ { d } ( x ) =$ +$\begin{array} { r l } { \frac { 1 } { 8 } \sum _ { i = 1 } ^ { 8 } \mathcal { N } ( x | \mu _ { i } , \frac { 1 } { 1 0 } I ) } & { { } } \end{array}$ where s samp $\mu _ { i } \sim \mathcal N ( \mu _ { i } | 0 , I )$ . The distribution e use multi-layer p $p ( x )$ is plotted in figure 6. The datasetptron with four ReLU layers of size $p _ { d } ( x )$ +400-200-100-50, 25-dimensional Gaussian latent variables. + +For different mixture coefficients $\alpha$ we visualize samples from the learned distributions $p _ { \psi , \theta } ( x _ { 1 } , x _ { 2 } )$ , $p _ { \psi , \theta } ( x _ { 1 } | x _ { 2 } )$ , and $p _ { \psi , \theta } ( x _ { 2 } | x _ { 1 } )$ . The observed features for the conditional distributions are generated from the marginal distributions $p ( x _ { 2 } )$ and $p ( x _ { 1 } )$ respectively. + +We see in table 8 and in figure 7, that even with very small weight GSNN prevents model from learning distributions with several local optimas. GSNN also increases Monte-Carlo log-likelihood estimation with a few samples and decreases much more precise Importance Sampling log-likelihood estimation. When $\alpha = 0 . 9$ the whole distribution structure is lost. + +![](images/7407e71d027399c997ec577ddb914586b76471425f58bff6404a5cd035d567fc.jpg) +Figure 6: Probability density function of synthetic data distribution. + +![](images/a4bd4e815fbae18fb3e5afb3429dcbb43b07bb985c3a0166b18e1666e16f7080.jpg) +Figure 7: VAEAC for synthetic data. + +![](images/54463305da3093916463d301746c0407e7d8d2daa9a2e48291c407dc4d828109.jpg) +Figure 8: MNIST inpaintings. Left: input. The gray pixels are unobserved. Middle: samples from the model. Right: ground truth. + +We see that using $\alpha \neq 1$ ruins multimodality of the restored distribution, so we highly recommend to use $\alpha = 1$ or at least $\alpha \approx 1$ . + +Table 9: Average negative log-likelihood of inpaintings for 1000 objects. IS- $S$ refers to Importance Sampling log-likelihood estimation with $S$ samples for each object (18). MC- $S$ refers to Monte-Carlo log-likelihood estimation with $S$ samples for each object (17). Naive Bayes is a baseline method which assumes pixels and colors independence. + +
MethodMNISTOmniglotCelebA
VAEAC IS-10261±1275±1734035 ± 1609
VAEAC MC-10494±41452 ± 10941513 ± 2163
VAEAC MC-102156 ±12203 ± 15053904 ± 3121
GSNN MC-104141 ±71199 ± 6253427 ± 2208
GSNN MC-10²141 ±11200 ± 6253486 ± 2210
Naive Bayes2052490269480
+ +![](images/3126337f9f6dd982a9196015b6247be9dab5083bb45c2e9eb0945400c4c56f31.jpg) +Figure 9: Convergence of VAE and VAEAC on MNIST dataset. + +C.3 COMPARISON ON THE IMAGE INPAINTING PROBLEM + +In figure 8 we can see that the inpaintings produced by GSNN are smooth, blurry and not diverse compared with VAEAC. + +Table 9 shows that VAEAC learns distribution over inpaintings better than GSNN in terms of test loglikelihood. Nevertheless, Monte-Carlo estimations with a small number of samples sometimes are better for GSNN, which means less local modes in the learned distribution and more blurriness in the samples. + +# D ADDITIONAL EXPERIMENTS + +# D.1 CONVERGENCE SPEED + +In figure 9 one can see that VAEAC has similar convergence speed to VAE in terms of iterations on MNIST dataset. In our experiments we observed the same behaviour for other datasets. Each iteration of VAEAC is about 1.5 times slower than VAE due to usage of three networks instead of two. + +Table 10: NRMSE (for continuous datasets) or PFC (for categorical ones) of imputations. Less is better. + +
DatasetMICEMissForestGAINVAEACGSNNNN
Boston0.69 ± 0.020.58±0.020.78 ± 0.030.71 ± 0.020.70 ± 0.010.69 ± 0.01
Breast0.58 ±0.020.515 ± 0.0080.67 ±0.010.55±0.020.55 ± 0.020.52 ±0.02
Concrete0.850 ± 0.0070.78 ± 0.010.96 ±0.010.84 ±0.020.85 ± 0.012±3
Diabetes0.80 ±0.010.84±0.020.911 ± 0.0090.90 ±0.030.91 ± 0.030.90 ±0.02
Digits0.69 ±0.020.61± 0.020.79±0.020.69 ±0.020.69 ± 0.020.67 ±0.02
Glass0.91 ±0.020.83± 0.041.06 ±0.050.91 ±0.040.91 ± 0.050.87 ± 0.04
Iris0.59 ± 0.020.62 ± 0.040.72 ± 0.040.64± 0.040.62 ± 0.040.61± 0.02
Mushroom0.334 ± 0.0020.249 ±0.0060.271 ±0.0030.241 ± 0.0020.2412 ± 0.00090.239 ± 0.001
Orthopedic0.76 ±0.020.79±0.030.91±0.030.80±0.030.81 ±0.030.81 ±0.02
Phishing0.422 ± 0.0060.422 ±0.0090.427 ± 0.0100.397 ± 0.0100.392 ±0.0090.41 ± 0.01
WallRobot0.885 ± 0.0030.640 ± 0.0030.907 ± 0.0050.78 ±0.010.776 ±0.0070.757± 0.005
WhiteWine0.964 ± 0.0070.878 ±0.0090.97 ± 0.020.850 ± 0.0050.848 ± 0.0070.85 ± 0.01
Yeast0.98±0.021.00 ± 0.020.99 ±0.030.95 ± 0.010.958 ± 0.0070.97 ± 0.03
Zoo0.19 ± 0.030.16 ±0.020.20 ±0.020.16±0.020.17 ±0.020.16 ±0.01
+ +Table 11: R2-score (for continuous targets) or accuracy (for categorical ones) of post-imputation regression or classification. Higher is better. + +
DatasetMICEMissForestGAINVAEACGSNNNN
Boston0.57 ± 0.080.6 ± 0.10.50 ± 0.100.5 ± 0.10.5± 0.10.50±0.09
Breast0.96 ± 0.020.95 ± 0.020.94± 0.010.95 ± 0.020.96 ± 0.020.95 ± 0.02
Concrete0.35 ± 0.050.33 ± 0.040.28±0.060.30 ± 0.080.32 ± 0.050±1
Diabetes0.37 ± 0.060.34±0.060.34±0.030.34± 0.040.33 ±0.040.27±0.06
Digits0.86±0.020.887 ±0.0080.83 ±0.030.892 ± 0.0100.895 ± 0.0100.912 ± 0.010
Glass0.44 ±0.080.53 ±0.050.37±0.050.49 ±0.090.47 ±0.090.48 ±0.09
Iris0.81± 0.020.84±0.020.66 ±0.060.84±0.050.82±0.060.73±0.09
Mushroom0.92 ±0.010.972 ± 0.0030.969 ± 0.0050.987 ± 0.0010.986 ± 0.0020.989 ± 0.003
Orthopedic0.71 ± 0.020.72 ± 0.030.60±0.030.71 ±0.020.70± 0.040.61±0.04
Phishing0.75 ± 0.020.73±0.030.74± 0.030.75 ± 0.010.74±0.040.73±0.02
WallRobot0.55 ±0.010.697 ± 0.0050.56 ±0.010.62±0.020.62 ± 0.010.64±0.02
WhiteWine0.13 ± 0.020.17 ± 0.010.11± 0.010.18 ±0.020.17 ± 0.010.15 ± 0.03
Yeast0.42 ±0.020.41 ±0.020.39 ±0.060.42 ± 0.010.425 ± 0.0100.33±0.03
Zoo0.78 ± 0.060.71 ±0.080.67±0.060.77 ± 0.090.8±0.10.83 ± 0.08
+ +# D.2 MISSING FEATURES IMPUTATION + +We evaluate the quality of imputations on different datasets (mostly from UCI (Lichman, 2013)). The evaluation is performed for VAEAC, GSNN (15) and NN (neural network; can be considered as a special case of GSNN where $p _ { \theta } ( z | x _ { 1 - b } , b )$ is delta-function; produces single imputation). We compare these methods with MICE (Buuren & Groothuis-Oudshoorn, 2010), MissForest (Stekhoven & Buhlmann, 2011), and GAIN ¨ (Yoon et al., 2018). + +We see that for some datasets MICE and MissForest outperform VAEAC, GSNN and NN. The reason is that for some datasets random forest is more natural structure than neural network. + +The results also show that VAEAC, GSNN and NN show similar imputation performance in terms of NRMSE, PFC, post-imputation R2-score and accuracy. Given the result from appendix C we can take this as a weak evidence that the distribution of imputations has only one local maximum for datasets from (Lichman, 2013). + +![](images/21041c7b3c7fc1baa4707c17e3944ec8fc55aadb337ff273f13ec39db52954c7.jpg) +Figure 10: CelebA inpaintings with masks from (Li et al., 2017). ft: input. The gray pixels are unobserved. Middle: samples from VAEAC. Right: ground truth. + +# D.3 FACE INPAINTINGS + +In figure 10 we provide samples of VAEAC on the CelebA dataset for the masks from (Li et al., 2017). + +# D.4 GAIN FOR IMAGE INPAINTING + +GAIN (Yoon et al., 2018) doesnt use unobserved data during training, which makes it easier to apply to the missing features imputation problem. Nevertheless, it turns into a disadvantage when the fully-observed training data is available but the missingness rate at the testing stage is high. + +We consider the horizontal line mask for MNIST which is described in appendix A.3. We use the released GAIN code 5 with a different mask generator. The inpaintings from VAEAC which uses the unobserved pixels during training are available in figure 1. The inpaintings from GAIN which ignores unobserved pixels are provided in figure 11. As can be seen in figure 11, GAIN fails to learn conditional distribution for given mask distribution ${ \dot { p } } ( b )$ . + +Nevertheless, we don’t claim that GAIN is not suitable for image inpainting. As it was shown in the supplementary of (Yoon et al., 2018) and in the corresponding code, GAIN is able to learn conditional distributions when $p ( b )$ is pixel-wise independent Bernoulli distribution with probability 0.5. + +![](images/77978a748fdfece774bbdba6a454e4caa9d264c220caf33b0d01dfee4003f59b.jpg) +Figure 11: MNIST inpaintings from GAIN. + +Left: input. The gray pixels are unobserved. Middle: samples from the model. Right: ground truth. + +![](images/3addba9b5d9e7527dc59c64b67c88b55ee2d758e97d2f9f04dcf482aada3e5b6.jpg) +Figure 12: MNIST inpaintings. Left: input. The gray pixels are unobserved. Middle: samples from the model. Right: ground truth. + +# D.5 UNIVERSAL MARGINALIZER: ILLUSTRATIONS + +In figure 12 we provide samples of Universal Marginalizer (UM) and VAEAC for the same inputs. + +Consider the case when UM marginal distributions are parametrized with Gaussians. The most simple example of a distribution, which UM cannot learn but VAEAC can, is given in figure 13. + +![](images/34b2efedea65ec8dd1cd6662835f9b38b5a648676328c4f707b65cec76fc2de2.jpg) +Figure 13: Distribution learning: VAEAC vs UM. \ No newline at end of file diff --git a/parse/train/SyxtJh0qYm/SyxtJh0qYm_content_list.json b/parse/train/SyxtJh0qYm/SyxtJh0qYm_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..96a50e647b53de90f44b9d1317132b38d772c571 --- /dev/null +++ b/parse/train/SyxtJh0qYm/SyxtJh0qYm_content_list.json @@ -0,0 +1,2990 @@ +[ + { + "type": "text", + "text": "VARIATIONAL AUTOENCODER WITH ARBITRARY CONDITIONING ", + "text_level": 1, + "bbox": [ + 148, + 117, + 558, + 162 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Oleg Ivanov \nSamsung AI Center Moscow Moscow, Russia \ntigvarts@gmail.com Michael Figurnov \nNational Research University Higher School of Economics ∗ Moscow, Russia \nmichael@figurnov.ru Dmitry Vetrov \nSamsung-HSE Laboratory, National Research University Higher School of Economics Samsung AI Center Moscow Moscow, Russia \nvetrovd@yandex.ru ", + "bbox": [ + 156, + 186, + 348, + 243 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "", + "bbox": [ + 387, + 186, + 589, + 257 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "", + "bbox": [ + 622, + 188, + 816, + 284 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "ABSTRACT ", + "text_level": 1, + "bbox": [ + 452, + 320, + 544, + 335 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "We propose a single neural probabilistic model based on variational autoencoder that can be conditioned on an arbitrary subset of observed features and then sample the remaining features in “one shot”. The features may be both real-valued and categorical. Training of the model is performed by stochastic variational Bayes. The experimental evaluation on synthetic data, as well as feature imputation and image inpainting problems, shows the effectiveness of the proposed approach and diversity of the generated samples. ", + "bbox": [ + 207, + 349, + 792, + 434 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 INTRODUCTION ", + "text_level": 1, + "bbox": [ + 150, + 457, + 310, + 473 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "In past years, a number of generative probabilistic models based on neural networks have been proposed. The most popular approaches include variational autoencoder (Kingma & Welling, 2013) (VAE) and generative adversarial net (Goodfellow et al., 2014) (GANs). They learn a distribution over objects $p ( x )$ and allow sampling from this distribution. ", + "bbox": [ + 148, + 488, + 849, + 544 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "In many cases, we are interested in learning a conditional distribution $p ( x | y )$ . For instance, if $x$ is an image of a face, $y$ could be the characteristics describing the face (are glasses present or not; length of hair, etc.) Conditional variational autoencoder (Sohn et al., 2015) and conditional generative adversarial nets (Mirza & Osindero, 2014) are popular methods for this problem. ", + "bbox": [ + 148, + 550, + 849, + 606 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "In this paper, we consider the problem of learning all conditional distributions of the form $p ( x _ { I } | x _ { U \\setminus I } )$ , where $U$ is the set of all features and $I$ is its arbitrary subset. This problem generalizes both learning the joint distribution $p ( x )$ and learning the conditional distribution $p ( x | y )$ . To tackle this problem, we propose a Variational Autoencoder with Arbitrary Conditioning (VAEAC) model. It is a latent variable model similar to VAE, but allows conditioning on an arbitrary subset of the features. The conditioning features affect the prior on the latent Gaussian variables which are used to generate unobserved features. The model is trained using stochastic gradient variational Bayes (Kingma & Welling, 2013). ", + "bbox": [ + 148, + 613, + 851, + 712 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "We consider two most natural applications of the proposed model. The first one is feature imputation where the goal is to restore the missing features given the observed ones. The imputed values may be valuable by themselves or may improve the performance of other machine learning algorithms which process the dataset. Another application is image inpainting in which the goal is to fill in an unobserved part of an image with an artificial content in a realistic way. This can be used for removing unnecessary objects from the images or, vice versa, for complementing the partially closed or corrupted object. ", + "bbox": [ + 148, + 718, + 851, + 801 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "The experimental evaluation shows that the proposed model successfully samples from the conditional distributions. The distribution over samples is close to the true conditional distribution. This property is very important when the true distribution has several modes. The model is shown to be effective in feature imputation problem which helps to increase the quality of subsequent discriminative models on different problems from UCI datasets collection (Lichman, 2013). We demonstrate that model can generate diverse and realistic image inpaintings on MNIST (LeCun et al., 1998), Omniglot (Lake et al., 2015) and CelebA (Liu et al., 2015) datasets, and works even better than the current state of the art inpainting techniques in terms of peak signal to noise ratio (PSNR). ", + "bbox": [ + 148, + 119, + 851, + 232 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "The paper is organized as follows. In section 2 we review the related works. In section 3 we briefly describe variational autoencoders and conditional variational autoencoders. In section 4 we define the problem, describe the VAEAC model and its training procedure. In section 5 we evaluate VAEAC. Section 6 concludes the paper. Appendix contains additional explanations, theoretical analysis, and experiments for VAEAC. ", + "bbox": [ + 148, + 238, + 849, + 295 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "2 RELATED WORK ", + "text_level": 1, + "bbox": [ + 150, + 320, + 318, + 335 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Universal Marginalizer (Douglas et al., 2017) is a model based on a feed-forward neural network which approximates marginals of unobserved features conditioned on observable values. A related idea of an autoregressive model of joint probability was previously proposed in Germain et al. (2015) and Uria et al. (2016). The description of the model and comparison with VAEAC are available in section 5.3. ", + "bbox": [ + 148, + 354, + 851, + 410 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Yoon et al. (2018) propose a GANs-based model called GAIN which solves the same problem as VAEAC. In contrast to VAEAC, GAIN does not use unobserved data during training, which makes it easier to apply to the missing features imputation problem. Nevertheless, it turns into a disadvantage when the fully-observed training data is available but the missingness rate at the testing stage is high. For example, in inpainting setting GAIN cannot learn the conditional distribution over MNIST digits given one horizontal line of the image while VAEAC can (see appendix D.4). The comparison of VAEAC and GAIN on the missing feature imputation problem is given in section 5.1 and appendix D.2. ", + "bbox": [ + 148, + 417, + 851, + 515 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Rezende et al. (2014) [Appendix F], Sohl-Dickstein et al. (2015), Goyal et al. (2017), and Bordes et al. (2017) propose to fill missing data with noise and run Markov chain with a learned transition operator. The stationary distribution of such chains approximates the true conditional distribution of the unobserved features. Bachman & Precup (2015) consider missing feature imputation in terms of Markov decision process and propose LSTM-based sequential decision making model to solve it. Nevertheless, these methods are computationally expensive at the test time and require fully-observed training data. ", + "bbox": [ + 148, + 522, + 849, + 606 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Image inpainting is a classic computer vision problem. Most of the earlier methods rely on local and texture information or hand-crafted problem-specific features (Bertalmio et al., 2000). In past years multiple neural network based approaches have been proposed. ", + "bbox": [ + 148, + 613, + 849, + 655 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "Pathak et al. (2016), Yeh et al. (2016) and Yang et al. (2017) use different kinds and combinations of adversarial, reconstruction, texture and other losses. Li et al. (2017) focuses on face inpainting and uses two adversarial losses and one semantic parsing loss to train the generative model. In Yeh et al. (2017) GANs are first trained on the whole training dataset. The inpainting is an optimization procedure that finds the latent variables that explain the observed features best. Then, the obtained latents are passed through the generative model to restore the unobserved portion of the image. We can say that VAEAC is a similar model which uses prior network to find a proper latents instead of solving the optimization problem. ", + "bbox": [ + 148, + 661, + 849, + 760 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "All described methods aim to produce a single realistic inpainting, while VAEAC is capable of sampling diverse inpaintings. Additionally, Yeh et al. (2016), Yang et al. (2017) and Yeh et al. (2017) have high testtime computational complexity of inpainting, because they require an optimization problem to be solved. On the other hand, VAEAC is a “single-shot” method with a low computational cost. ", + "bbox": [ + 148, + 766, + 849, + 821 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "3 BACKGROUND ", + "text_level": 1, + "bbox": [ + 148, + 118, + 300, + 135 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3.1 VARIATIONAL AUTOENCODER ", + "text_level": 1, + "bbox": [ + 148, + 150, + 397, + 165 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Variational autoencoder (Kingma & Welling, 2013) (VAE) is a directed generative model with latent variables. The generative process in variational autoencoder is as follows: first, a latent variable $z$ is generated from the prior distribution $p ( z )$ , and then the data $x$ is generated from the generative distribution $p _ { \\theta } ( x | z )$ , where $\\theta$ are the generative model’s parameters. This process induces the distribution $p _ { \\theta } ( x ) = \\mathbb { E } _ { p ( z ) } p _ { \\theta } ( x | z )$ . The distribution $p _ { \\theta } ( x | z )$ is modeled by a neural network with parameters $\\theta$ . $p ( z )$ is a standard Gaussian distribution. ", + "bbox": [ + 147, + 176, + 852, + 262 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "The parameters $\\theta$ are tuned by maximizing the likelihood of the training data points $\\{ x _ { i } \\} _ { i = 1 } ^ { N }$ from the true data distribution $p _ { d } ( x )$ . In general, this optimization problem is challenging due to intractable posterior inference. However, a variational lower bound can be optimized efficiently using backpropagation and stochastic gradient descent: ", + "bbox": [ + 148, + 267, + 849, + 325 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/fed675d9452648daab0a6ca45636d1aac44900502c485323a4dc8122dfe3f9fe.jpg", + "text": "$$\n\\begin{array} { r l } & { \\log p _ { \\theta } ( x ) = \\mathbb { E } _ { q _ { \\phi } ( z | x ) } \\log \\frac { p _ { \\theta } ( x , z ) } { q _ { \\phi } ( z | x ) } + D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( z | x ) \\| p ( z | x , \\theta ) \\big ) } \\\\ & { \\qquad \\quad \\ge \\mathbb { E } _ { q _ { \\phi } ( z | x ) } \\log p _ { \\theta } ( x | z ) - D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( z | x ) \\| p ( z ) \\big ) = L _ { V A E } \\big ( x ; \\theta , \\phi \\big ) } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 165, + 335, + 815, + 392 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Here $q _ { \\phi } ( z | x )$ is a proposal distribution parameterized by neural network with parameters $\\phi$ that approximates the posterior $p ( z | x , \\theta )$ . Usually this distribution is Gaussian with a diagonal covariance matrix. The closer $q _ { \\phi } ( z | x )$ to $p ( z | x , \\theta )$ , the tighter variational lower bound $L _ { V A E } ( \\theta , \\phi )$ . To compute the gradient of the variational lower bound with respect to $\\phi$ , reparameterization trick is used: $z = \\bar { \\mu } _ { \\phi } ( x ) + \\bar { \\varepsilon } \\sigma _ { \\phi } ( x )$ where $\\varepsilon \\sim \\mathcal { N } ( 0 , I )$ and $\\mu _ { \\phi }$ and $\\sigma _ { \\phi }$ are deterministic functions parameterized by neural networks. So the gradient can be estimated using Monte-Carlo method for the first term and computing the second term analytically: ", + "bbox": [ + 148, + 396, + 852, + 481 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/392ee19314ad1169b0b4fd7d080fa9dd49328ced339e9d76af78ca28090587f9.jpg", + "text": "$$\n\\frac { \\partial L _ { V A E } ( x ; \\theta , \\phi ) } { \\partial \\phi } = \\mathbb { E } _ { \\varepsilon \\sim \\mathcal { N } ( 0 , I ) } \\frac { \\partial } { \\partial \\phi } \\log p _ { \\theta } ( x | \\mu _ { \\phi } ( x ) + \\varepsilon \\sigma _ { \\phi } ( x ) ) - \\frac { \\partial } { \\partial \\phi } D _ { \\mathrm { K L } } ( q _ { \\phi } ( z | x ) \\| p ( z ) ) .\n$$", + "text_format": "latex", + "bbox": [ + 209, + 488, + 787, + 521 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "So $L _ { V A E } ( \\theta , \\phi )$ can be optimized using stochastic gradient ascent with respect to $\\phi$ and $\\theta$ ", + "bbox": [ + 147, + 527, + 733, + 542 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "3.2 CONDITIONAL VARIATIONAL AUTOENCODER ", + "text_level": 1, + "bbox": [ + 148, + 560, + 503, + 575 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Conditional variational autoencoder (Sohn et al., 2015) (CVAE) approximates the conditional distribution $p _ { d } ( x | y )$ . It outperforms deterministic models when the distribution $p _ { d } ( x | y )$ is multi-modal (diverse $x \\mathbf { s }$ are probable for the given $y$ ). For example, assume that $x$ is a real-valued image. Then, a deterministic regression model with mean squared error loss would predict the average blurry value for $x$ . On the other hand, CVAE learns the distribution of $x$ , from which one can sample diverse and realistic objects. ", + "bbox": [ + 147, + 585, + 851, + 656 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Variational lower bound for CVAE can be derived similarly to VAE by conditioning all considered distributions on $y$ : ", + "bbox": [ + 142, + 662, + 849, + 693 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/1a364d873a354bd42b6ac3c1a473c1e2729b1b87d4dce0f0bfde7ab681819e08.jpg", + "text": "$$\nL _ { C V A E } ( x , y ; \\theta , \\psi , \\phi ) = \\mathbb { E } _ { q _ { \\phi } ( z | x , y ) } \\log p _ { \\theta } ( x | z , y ) - D _ { \\mathrm { K L } } ( q _ { \\phi } ( z | x , y ) | | p _ { \\psi } ( z | y ) ) \\leq \\log p _ { \\theta , \\psi } ( x | y )\n$$", + "text_format": "latex", + "bbox": [ + 186, + 699, + 812, + 718 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Similarly to VAE, this objective is optimized using the reparameterization trick. Note that the prior distribution $p _ { \\psi } ( z | y )$ is conditioned on $y$ and is modeled by a neural network with parameters $\\psi$ . Thus, CVAE uses three trainable neural networks, while VAE only uses two. ", + "bbox": [ + 148, + 731, + 852, + 773 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "Also authors propose such modifications of CVAE as Gaussian stochastic neural network and hybrid model. These modifications can be applied to our model as well. Nevertheless, we don’t use them, because of their disadvantage which is described in appendix C. ", + "bbox": [ + 148, + 779, + 849, + 823 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "4 VARIATIONAL AUTOENCODER WITH ARBITRARY CONDITIONING", + "text_level": 1, + "bbox": [ + 147, + 118, + 720, + 136 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "4.1 PROBLEM STATEMENT ", + "text_level": 1, + "bbox": [ + 148, + 152, + 346, + 167 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Consider a distribution $p _ { d } ( x )$ over a $D$ -dimensional vector $x$ with real or categorical components. The components of the vector are called features. ", + "bbox": [ + 147, + 180, + 849, + 209 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Let binary vector $b \\in \\{ 0 , 1 \\} ^ { D }$ be the binary mask of unobserved features of the object. Then we describe the vector of unobserved features as $x _ { b } = \\{ x _ { i : b _ { i } = 1 } \\}$ . For example, $x _ { ( 0 , 1 , 1 , 0 , 1 ) } = ( x _ { 2 } , x _ { 3 } , x _ { 5 } )$ . Using this notation we denote $x _ { 1 - b }$ as a vector of observed features. ", + "bbox": [ + 147, + 215, + 852, + 258 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Our goal is to build a model of the conditional distribution $p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b ) \\approx p _ { d } ( x _ { b } | x _ { 1 - b } , b )$ for an arbitrary $b$ , where $\\psi$ and $\\theta$ are parameters that are used in our model at the testing stage. ", + "bbox": [ + 148, + 265, + 849, + 294 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "However, the true distribution $p _ { d } ( x _ { b } | x _ { 1 - b } , b )$ is intractable without strong assumptions about $p _ { d } ( x )$ . Therefore, our model $p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b )$ has to be more precise for some $b$ and less precise for others. To formalize our requirements about the accuracy of our model we introduce the distribution $p ( b )$ over different unobserved feature masks. The distribution $p ( b )$ is arbitrary and may be defined by the user depending on the problem. Generally it should have full support over $\\{ 0 , 1 \\} ^ { D }$ so that $p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b )$ can evaluate arbitrary conditioning. Nevertheless, it is not necessary if the model is used for specific kinds of conditioning (as we do in section 5.2). ", + "bbox": [ + 147, + 299, + 851, + 398 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "Using $p ( b )$ we can introduce the following log-likelihood objective function for the model: ", + "bbox": [ + 147, + 404, + 741, + 420 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/6eacc0f28bc2e238ba172a9f915eb5066998bc34e8c21e5ae8d8543613577821.jpg", + "text": "$$\n\\operatorname* { m a x } _ { \\psi , \\theta } \\mathbb { E } _ { p _ { d } ( \\boldsymbol { x } ) } \\mathbb { E } _ { p ( \\boldsymbol { b } ) } \\log p _ { \\psi , \\theta } \\big ( x _ { b } | \\boldsymbol { x } _ { 1 - \\boldsymbol { b } } , \\boldsymbol { b } \\big )\n$$", + "text_format": "latex", + "bbox": [ + 374, + 430, + 625, + 454 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "The special cases of the objective (4) are variational autoencoder $( b _ { i } = 1 \\forall i \\in \\{ 1 , \\ldots , D \\} )$ and conditional variational autoencoder $^ { \\textit { b } }$ is constant). ", + "bbox": [ + 147, + 465, + 851, + 496 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "4.2 MODEL DESCRIPTION ", + "text_level": 1, + "bbox": [ + 148, + 517, + 343, + 531 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "The generative process of our model is similar to the generative process of CVAE: for each object firstly we generate $z \\sim p _ { \\psi } ( z | x _ { 1 - b } , b )$ using prior network, and then sample unobserved features $x _ { b } ~ \\sim ~ p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b )$ using generative network. This process induces the following model distribution over unobserved features: ", + "bbox": [ + 147, + 545, + 851, + 601 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/1010c9b588784d89851c33cee48ac5e4d9b7fc2302b1356bbdce195c382c31ce.jpg", + "text": "$$\np _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b ) = \\mathbb { E } _ { z \\sim p _ { \\psi } ( z | x _ { 1 - b } , b ) } p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b )\n$$", + "text_format": "latex", + "bbox": [ + 325, + 612, + 673, + 631 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "We use $z ~ \\in ~ \\mathbb { R } ^ { d }$ , and Gaussian distribution $p _ { \\psi }$ over $z$ , with parameters from a neural network with weights $\\psi$ : $p _ { \\psi } ( z | x _ { 1 - b } , b , \\psi ) = \\mathcal { N } ( z | \\mu _ { \\psi } ( x _ { 1 - b } , b ) , \\sigma _ { \\psi } ^ { 2 } ( x _ { 1 - b } , b ) I )$ . The real-valued components of distribution $p _ { \\theta } ( x _ { b } | \\boldsymbol { z } , x _ { 1 - b } , b )$ are defined likewise. Each categorical component $i$ of distribution $p _ { \\theta } ( x _ { i } | \\boldsymbol { z } , x _ { 1 - b } , b )$ is parameterized by a function $w _ { i , \\theta } ( z , x _ { 1 - b } , b )$ , whose outputs are logits of probabilities for each category: $x _ { i } \\sim \\mathrm { C a t } [ \\mathrm { S o f t m a x } ( w _ { i , \\theta } ( z , x _ { 1 - b } , b ) ) ]$ . Therefore the components of the latent vector $z$ are conditionally independent given $x _ { 1 - b }$ and $b$ , and the components of $x _ { b }$ are conditionally independent given $z , x _ { 1 - b }$ and $b$ . ", + "bbox": [ + 147, + 650, + 851, + 739 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "The variables $x _ { b }$ and $x _ { 1 - b }$ have variable length that depends on $b$ . So in order to use architectures such as multi-layer perceptron and convolutional neural network we consider $x _ { 1 - b } = x \\circ ( 1 - b )$ where $\\circ$ is an element-wise product. So in implementation $x _ { 1 - b }$ has fixed length. The output of the generative network also has a fixed length, but we use only unobserved components to compute likelihood. ", + "bbox": [ + 148, + 744, + 851, + 801 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "The theoretical analysis of the model is available in appendix B.1. ", + "bbox": [ + 147, + 808, + 578, + 823 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "4.3 LEARNING VARIATIONAL AUTOENCODER WITH ARBITRARY CONDITIONING", + "text_level": 1, + "bbox": [ + 145, + 119, + 717, + 136 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4.3.1 VARIATIONAL LOWER BOUND ", + "text_level": 1, + "bbox": [ + 148, + 146, + 413, + 161 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "We can derive a lower bound for $\\log p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b )$ as for variational autoencoder: ", + "bbox": [ + 147, + 171, + 700, + 188 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/23cab9595cb2df09a6fe520b5c3c4ac610eb31ffc336c62a0f70f0e3846d00aa.jpg", + "text": "$$\n\\begin{array} { r l } & { \\log p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b ) = \\mathbb { E } _ { q _ { \\phi } ( z | x , b ) } \\log \\frac { p _ { \\psi , \\theta } ( x _ { b } , z | x _ { 1 - b } , b ) } { q _ { \\phi } ( z | x , b ) } + D _ { \\mathrm { K L } } ( q _ { \\phi } ( z | x , b ) | | p _ { \\psi , \\theta } ( z | x , b ) ) } \\\\ & { \\qquad \\geq \\mathbb { E } _ { q _ { \\phi } ( z | x , b ) } \\log p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b ) - D _ { \\mathrm { K L } } ( q _ { \\phi } ( z | x , b ) | | p _ { \\psi } ( z | x _ { 1 - b } , b ) ) = L _ { V A E A C } ( x , b ; \\theta , \\psi , \\phi ) } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 163, + 200, + 815, + 256 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Therefore we have the following variational lower bound optimization problem: ", + "bbox": [ + 147, + 261, + 671, + 276 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/0adebae96daeaa3c98aa6ef63b82c1e9fa0b29000c3c6a3af4e0beff92f397c1.jpg", + "text": "$$\n\\operatorname* { m a x } _ { \\theta , \\psi , \\phi } \\mathbb { E } _ { p _ { d } ( x ) } \\mathbb { E } _ { p ( b ) } L _ { V A E A C } ( x , b ; \\theta , \\psi , \\phi )\n$$", + "text_format": "latex", + "bbox": [ + 367, + 285, + 633, + 309 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "We use fully-factorized Gaussian proposal distribution $q _ { \\phi }$ which allows us to perform reparameterization trick and compute KL divergence analytically in order to optimize (7). ", + "bbox": [ + 148, + 318, + 851, + 347 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4.3.2 PRIOR IN LATENT SPACE ", + "text_level": 1, + "bbox": [ + 148, + 363, + 377, + 377 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "During the optimization of objective (7) the parameters $\\mu _ { \\psi }$ and $\\sigma _ { \\psi }$ of the prior distribution of $z$ may tend to infinity, since there is no penalty for large values of those parameters. We usually observe the growth of $\\left. z \\right. _ { 2 }$ during training, though it is slow enough. To prevent potential numerical instabilities, we put a Normal-Gamma prior on the parameters of the prior distribution to prevent the divergence. Formally, we redefine $p _ { \\psi } ( z | x _ { 1 - b } , b )$ as follows: ", + "bbox": [ + 147, + 388, + 851, + 459 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/42bd1f2e7f266115ecbfdf42ffc85525498becdfe4a210dad151bb0713eff63a.jpg", + "text": "$$\np _ { \\psi } ( z , \\mu _ { \\psi } , \\sigma _ { \\psi } | x _ { 1 - b } , b ) = \\mathcal { N } ( z | \\mu _ { \\psi } , \\sigma _ { \\psi } ^ { 2 } ) \\mathcal { N } ( \\mu _ { \\psi } | 0 , \\sigma _ { \\mu } ) \\mathrm { G a m m a } ( \\sigma _ { \\psi } | 2 , \\sigma _ { \\sigma } )\n$$", + "text_format": "latex", + "bbox": [ + 263, + 468, + 735, + 488 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "As a result, the regularizers $- \\frac { \\mu _ { \\psi } ^ { 2 } } { 2 \\sigma _ { \\mu } ^ { 2 } }$ and $\\sigma _ { \\sigma } ( \\log ( \\sigma _ { \\psi } ) - \\sigma _ { \\psi } )$ are added to the model log-likelihood. Hyperparameter $\\sigma _ { \\mu }$ is chosen to be large $( 1 0 ^ { 4 } )$ and $\\sigma _ { \\sigma }$ is taken to be a small positive number $( 1 0 ^ { - 4 } )$ . This distribution is close to uniform near zero, so it doesn’t affect the learning process significantly. ", + "bbox": [ + 147, + 497, + 849, + 551 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "4.3.3 MISSING FEATURES ", + "text_level": 1, + "bbox": [ + 148, + 568, + 341, + 583 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "The optimization objective (7) requires all features of each object at the training stage: some of the features will be observed variables at the input of the model and other will be unobserved features used to evaluate the model. Nevertheless, in some problem settings the training data contains missing features too. We propose the following slight modification of the problem (7) in order to cover such problems as well. ", + "bbox": [ + 148, + 593, + 851, + 650 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "The missing values cannot be observed so $x _ { i } = \\omega \\Rightarrow b _ { i } = 1$ , where $\\omega$ describes the missing value in the data. In order to meet this requirement, we redefine mask distribution as conditioned on $x$ : $p ( b )$ turns into $p ( b | x )$ in (4) and (7). In the reconstruction loss (5) we simply omit the missing features, i. e. marginalize them out: ", + "bbox": [ + 147, + 656, + 851, + 712 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/fd9e44e756fd89b81c9ee7ee19237edf2b2753f3ffd55278ac0c898fa8a3729d.jpg", + "text": "$$\n\\log p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b ) = \\sum _ { \\substack { i : b _ { i } = 1 , x _ { i } \\neq \\omega } } \\log p _ { \\theta } ( x _ { i } | z , x _ { 1 - b } , b )\n$$", + "text_format": "latex", + "bbox": [ + 315, + 709, + 684, + 746 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "The proposal network must be able to determine which features came from real object and which are just missing. So we use additional missing features mask which is fed to proposal network together with unobserved features mask $b$ and object $x$ . ", + "bbox": [ + 148, + 758, + 851, + 801 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "The proposed modifications are evaluated in section 5.1. ", + "bbox": [ + 148, + 808, + 516, + 823 + ], + "page_idx": 4 + }, + { + "type": "table", + "img_path": "images/8a4721c4c0839da51aa1143ab97f04e2031a1758359b3d6ea12919686c3c0741.jpg", + "table_caption": [ + "Table 1: NRMSE (for continuous datasets) or PFC (for categorical ones) of imputations. Less is better. " + ], + "table_footnote": [], + "table_body": "
Method/DatasetWhiteWineYeastMushroomZooPhishing
MICE0.964± 0.0071.01 ± 0.010.334± 0.0020.19±0.030.422± 0.006
MissForest0.878 ± 0.0091.02 ± 0.060.249 ± 0.0060.16 ±0.020.422 ± 0.009
GAIN0.97 ± 0.020.99 ± 0.030.271 ± 0.0030.20± 0.020.427 ± 0.010
VAEAC0.850 ± 0.0070.94 ± 0.010.244 ± 0.0020.16 ± 0.020.394± 0.006
", + "bbox": [ + 148, + 142, + 857, + 215 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "5 EXPERIMENTS ", + "text_level": 1, + "bbox": [ + 148, + 247, + 300, + 263 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "In this section we validate the performance of VAEAC using several real-world datasets. In the first set of experiments we evaluate VAEAC missing features imputation performance using various UCI datasets (Lichman, 2013). We compare imputations from our model with imputations from such classical methods as MICE (Buuren & Groothuis-Oudshoorn, 2010) and MissForest (Stekhoven & Buhlmann, 2011) and recently ¨ proposed GANs-based method GAIN (Yoon et al., 2018). In the second set of experiments we use VAEAC to solve image inpainting problem. We show inpainitngs generated by VAEAC and compare our model with models from papers Pathak et al. (2016), Yeh et al. (2017) and Li et al. (2017) in terms of peak signal-to-noise ratio (PSNR) of obtained inpaintings on CelebA dataset (Liu et al., 2015) . And finally, we evaluate VAEAC against the competing method called Universal Marginalizer (Douglas et al., 2017). Additional experiments can be found in appendices C and D. The code is available at https://github.com/tigvarts/ vaeac. ", + "bbox": [ + 148, + 285, + 851, + 438 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "5.1 MISSING FEATURES IMPUTATION ", + "text_level": 1, + "bbox": [ + 150, + 463, + 418, + 478 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "The datasets with missing features are widespread. Consider a dataset with $D$ -dimensional objects $x$ where each feature may be missing (which we denote by $x _ { i } ~ = ~ \\omega$ ) and their target values $y$ . The majority of discriminative methods do not support missing values in the objects. The procedure of filling in the missing features values is called missing features imputation. ", + "bbox": [ + 148, + 493, + 851, + 549 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "In this section we evaluate the quality of imputations produced by VAEAC. For evaluation we use datasets from UCI repository (Lichman, 2013). Before training we drop randomly $50 \\%$ of values both in train and test set. After that we impute missing features using MICE (Buuren & Groothuis-Oudshoorn, 2010), MissForest (Stekhoven & Buhlmann, 2011), GAIN (Yoon et al., 2018) and VAEAC trained on the observed data. The ¨ details of GAIN implementation are described in appendix A.4. ", + "bbox": [ + 148, + 556, + 849, + 626 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Our model learns the distribution of the imputations, so it is able to sample from this distribution. We replace each object with missing features by $n = 1 0$ objects with sampled imputations, so the size of the dataset increases by $n$ times. This procedure is called missing features multiple imputation. MICE and GAIN are also capable of multiple imputation (we use $n = 1 0$ for them in experiments as well), but MissForest is not. ", + "bbox": [ + 148, + 632, + 849, + 689 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "For more details about the experimental setup see appendices A.1, A.2, and A.4. ", + "bbox": [ + 148, + 695, + 671, + 710 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "In table 1 we report NRMSE (i.e. RMSE normalized by the standard deviation of each feature and then averaged over all features) of imputations for continuous datasets and proportion of falsely classified (PFC) for categorical ones. For multiple imputation methods we average imputations of continuous variables and take most frequent imputation for categorical ones for each object. ", + "bbox": [ + 148, + 717, + 851, + 773 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "We also learn linear or logistic regression and report the regression or classification performance after applying imputations of different methods in table 2. For multiple imputation methods we average predictions for continuous targets and take most frequent prediction for categorical ones for each object in test set. ", + "bbox": [ + 150, + 780, + 849, + 821 + ], + "page_idx": 5 + }, + { + "type": "table", + "img_path": "images/f713f50af0782fe9152224004ad7b2409e899b7731c683b2a4bdd8d12d03d1da.jpg", + "table_caption": [ + "Table 2: R2-score (for continuous targets) or accuracy (for categorical ones) of post-imputation regression or classification. Higher is better. " + ], + "table_footnote": [], + "table_body": "
Method /DatasetWhiteWineYeastMushroomZ00Phishing
MICE0.13±0.020.41 ±0.020.92± 0.010.78± 0.050.75 ±0.02
MissForest0.17 ± 0.010.42 ± 0.020.972 ± 0.0030.71 ± 0.070.73 ± 0.02
GAIN0.11 ± 0.010.39 ± 0.060.969 ± 0.0050.67 ± 0.060.74 ± 0.03
VAEAC0.17 ± 0.010.43 ± 0.010.983 ± 0.0020.8 ± 0.10.74 ± 0.02
", + "bbox": [ + 158, + 156, + 839, + 231 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "As can be seen from the tables 1 and 2, VAEAC can learn joint data distribution and use it for missing feature imputation. The imputations are competitive with current state of the art imputation methods in terms of RMSE, PFC, post-imputation regression R2-score and classification accuracy. Nevertheless, we don’t claim that our method is state of the art in missing features imputation; for some datasets MICE or MissForest outperform it. The additional experiments can be found in appendix D.2. ", + "bbox": [ + 148, + 261, + 851, + 332 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "5.2 IMAGE INPAINTING ", + "text_level": 1, + "bbox": [ + 150, + 354, + 323, + 368 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "The image inpainting problem has a number of different formulations. The formulation of our interest is as follows: some of the pixels of an image are unobserved and we want to restore them in a natural way. Unlike the majority of papers, we want to restore not just one most probable inpainting, but the distribution over all possible inpaintings from which we can sample. This distribution is extremely multi-modal because often there is a lot of different possible ways to inpaint the image. ", + "bbox": [ + 148, + 382, + 849, + 452 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "Unlike the previous subsection, here we have uncorrupted images without missing features in the training set, so $p ( b | x ) = p ( b )$ . ", + "bbox": [ + 147, + 459, + 848, + 488 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "As we show in section 2, state of the art results use different adversarial losses to achieve more sharp and realistic samples. VAEAC can be adapted to the image inpainting problem by using a combination of those adversarial losses as a part of reconstruction loss $p _ { \\theta } ( x _ { b } | \\boldsymbol { z } , x _ { 1 - b } , b )$ . Nevertheless, such construction is out of scope for this research, so we leave it for the future work. In the current work we show that the model can generate both diverse and realistic inpaintings. ", + "bbox": [ + 148, + 494, + 851, + 564 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "In figures 1, 2, 3 and 4 we visualize image inpaintings produced by VAEAC on binarized MNIST (LeCun et al., 1998), Omniglot (Lake et al., 2015) and CelebA (Liu et al., 2015). The details of learning procedure and description of datasets are available in appendixes A.1 and A.3. ", + "bbox": [ + 148, + 571, + 851, + 613 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "To the best of our knowledge, the most modern inpainting papers don’t consider the diverse inpainting problem, where the goal is to build diverse image inpaintings, so there is no straightforward way to compare with these models. Nevertheless, we compute peak signal-to-noise ratio (PSNR) for one random inpainting from VAEAC and the best PSNR among 10 random inpaintings from VAEAC. One inpainting might not be similar to the original image, so we also measure how good the inpainting which is most similar to the original image reconstructs it. We compare these two metrics computed for certain masks with the PSNRs for the same masks on CelebA from papers Yeh et al. (2017) and Li et al. (2017). The results are available in tables 3 and 4. ", + "bbox": [ + 147, + 619, + 851, + 732 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "We observe that for the majority of proposed masks our model outperforms the competing methods in terms of PSNR even with one sample, and for the rest (where the inpaintings are significantly diverse) the best PSNR over 10 inpaintings is larger than the same PSNR of the competing models. Even if PSNR does not reflect completely the visual quality of images and tends to encourage blurry VAE samples instead of realistic GANs samples, the results show that VAEAC is able to solve inpainting problem comparably to the state of the art methods. The disadvantage of VAEAC compared to Yeh et al. (2017) and Li et al. (2017) (but not Pathak et al. (2016)) is that it needs the distribution over masks at the training stage to be similar to the distribution over them at the test stage. However, it is not a very strict limitation for the practical usage. ", + "bbox": [ + 148, + 738, + 851, + 821 + ], + "page_idx": 6 + }, + { + "type": "table", + "img_path": "images/5026e89f974837461ecf99590c5cc0d8aa5c4f5fb91a1c67e40c826d645b2e9c.jpg", + "table_caption": [ + "Table 3: PSNR of inpaintings for different masks for Context Encoder (Pathak et al., 2016), model from “Semantic Image Inpainting with Deep Generative Models” (Yeh et al., 2017) and VAEAC. Higher is better. " + ], + "table_footnote": [], + "table_body": "
Method/MasksCenterPatternRandomHalf
Context Encoder 121.319.220.615.5
SIIDGM 119.417.422.813.7
VAEAC, 1 sample22.121.429.314.9
VAEAC,10 samples23.723.329.317.4
", + "bbox": [ + 294, + 162, + 702, + 237 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/651996553ca128a2271022b3f6a402952fe4db1b266c3fc41a688b8cbaf6c735.jpg", + "table_caption": [ + "Table 4: PSNR of inpaintings for different masks for Context Encoder (Pathak et al., 2016), model from “Generative Face Completion” (Li et al., 2017) and VAEAC. Higher is better. " + ], + "table_footnote": [], + "table_body": "
Method/Masks010203040506
Context Encoder218.618.417.919.019.119.3
GFC ²20.019.818.819.719.520.2
VAEAC,1 sample20.821.019.520.320.321.0
VAEAC,10 samples22.022.220.821.721.822.2
", + "bbox": [ + 276, + 301, + 722, + 376 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "", + "bbox": [ + 150, + 405, + 851, + 434 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "5.3 UNIVERSAL MARGINALIZER ", + "text_level": 1, + "bbox": [ + 148, + 450, + 387, + 465 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Universal Marginalizer (Douglas et al., 2017) (UM) is a model which uses a single neural network to estimate the marginal distributions over the unobserved features. So it optimizes the following objective: ", + "bbox": [ + 148, + 476, + 849, + 505 + ], + "page_idx": 7 + }, + { + "type": "equation", + "img_path": "images/acbf51b909c72b7e6a56e480067a9402f1912d3b957be1fdb0f7435430250ce2.jpg", + "text": "$$\n\\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { x \\sim p _ { d } ( x ) } \\mathbb { E } _ { b \\sim p ( b ) } \\sum _ { i = 1 } ^ { D } b _ { i } \\log p _ { \\theta } \\big ( x _ { i } | x _ { 1 - b } , b \\big )\n$$", + "text_format": "latex", + "bbox": [ + 344, + 508, + 655, + 553 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "For given mask $b$ we fix a permutation of its unobserved components: $( i _ { 1 } , i _ { 2 } , \\dots , i _ { | b | } )$ , where $| b |$ is a number of unobserved components. Using the learned model and the permutation we can generate objects from joint distribution and estimate their probability using chain rule. ", + "bbox": [ + 147, + 563, + 852, + 606 + ], + "page_idx": 7 + }, + { + "type": "equation", + "img_path": "images/29a588d63f7f9b56a9a4a873e8bb8c472a9609e06bfe1de20bf3ead78d6521b2.jpg", + "text": "$$\n\\log p _ { \\theta } ( x _ { b } | x _ { 1 - b } , b ) = \\sum _ { j = 1 } ^ { | b | } \\log p _ { \\theta } ( x _ { i _ { j } } | x _ { 1 - ( b - \\sum _ { k = 1 } ^ { j - 1 } e _ { i _ { k } } ) } , b - \\sum _ { k = 1 } ^ { j - 1 } e _ { i _ { k } } )\n$$", + "text_format": "latex", + "bbox": [ + 281, + 609, + 720, + 655 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "For example, $p _ { \\theta } ( x _ { 1 } , x _ { 4 } , x _ { 5 } | x _ { 2 } , x _ { 3 } ) = p _ { \\theta } ( x _ { 4 } | x _ { 2 } , x _ { 3 } ) p _ { \\theta } ( x _ { 1 } | x _ { 2 } , x _ { 3 } , x _ { 4 } ) p _ { \\theta } ( x _ { 5 } | x _ { 1 } , x _ { 2 } , x _ { 3 } , x _ { 4 } ) .$ ", + "bbox": [ + 148, + 659, + 745, + 676 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Conditional sampling or conditional likelihood estimation for one object requires $| b |$ requests to UM to compute $p _ { \\theta } ( x _ { i } | x _ { 1 - b } , b )$ . Each request is a forward pass through the neural network. In the case of conditional sampling those requests even cannot be paralleled because the input of the next request contains the output of the previous one. ", + "bbox": [ + 147, + 680, + 851, + 737 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "We propose a slight modification of the original UM training procedure which allows learning UM efficiently for any kind of masks including those considered in this paper. The details of the modification are described in appendix B.3. ", + "bbox": [ + 148, + 743, + 851, + 785 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/d8e3304c44f6641f3fcf46844d8ec076bf559af92021199814e73a3e8e86af57.jpg", + "image_caption": [ + "Figure 1: MNIST inpaintings. " + ], + "image_footnote": [], + "bbox": [ + 176, + 137, + 482, + 333 + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/696a7457eff5fb7c33faf339e1df45fda1cdfba4264c7adccc10fee5a551d327.jpg", + "image_caption": [ + "Figure 2: Omniglot inpaintings. " + ], + "image_footnote": [], + "bbox": [ + 529, + 137, + 831, + 333 + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/22952789d3ea86e1798ed0fcb93d6b5ec6a1f7740b49980530ccffa387686d74.jpg", + "image_caption": [ + "Figure 3: CelebA inpaintings. " + ], + "image_footnote": [], + "bbox": [ + 161, + 387, + 491, + 738 + ], + "page_idx": 8 + }, + { + "type": "image", + "img_path": "images/1a5cbe58c9eb61a7543d31d330e9a0fdd262a2099e64f42864b8627f3f719b25.jpg", + "image_caption": [ + "Figure 4: CelebA inpaintings with masks from (Yeh et al., 2017). " + ], + "image_footnote": [], + "bbox": [ + 519, + 385, + 839, + 737 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Left: input. The gray pixels are unobserved. Middle: samples from VAEAC. Right: ground truth. ", + "bbox": [ + 174, + 789, + 815, + 804 + ], + "page_idx": 8 + }, + { + "type": "table", + "img_path": "images/cbb12bac85d7349bfd6184e4aa519378b2ab3130d2d4cea064e8fd8f042f3272.jpg", + "table_caption": [ + "Table 5: VAEAC and UM comparison on MNIST. " + ], + "table_footnote": [], + "table_body": "
MethodVAEACUM
Negative log-likelihood6141
Training time (30 epochs)5min 47s3min 14s
Test time (1OO samples generation)0.7ms1s
", + "bbox": [ + 297, + 141, + 704, + 202 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "The results of using this modification of UM are provided in table 5. We can say that the relation between VAEAC and UM is similar to the relation between VAE and PixelCNN. The second one is much slower at the testing stage, but it easily takes into account local dependencies in data while the first one is faster but assumes conditional independence of the outputs. Nevertheless, there are a number of cases where UM cannot learn the distribution well while VAEAC can. For example, when the data is real-valued and marginal distributions have many local optima, there is no straightforward parametrization which allows UM to approximate them, and, therefore also the conditioned joint distribution. An example of such distribution and more illustrations for comparison of VAEAC and UM are available in appendix D.5. ", + "bbox": [ + 147, + 229, + 851, + 342 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "6 CONCLUSION ", + "text_level": 1, + "bbox": [ + 148, + 363, + 292, + 380 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "In this paper we consider the problem of simultaneous learning of all conditional distributions for a vector. This problem has a number of different special cases with practical applications. We propose neural network based probabilistic model for distribution conditioning learning with Gaussian latent variables. This model is scalable and efficient in inference and learning. We propose several tricks to improve optimization and give recommendations about hyperparameters choice. The model is successfully applied to feature imputation and inpainting tasks. The experimental results show that the model is competitive with state of the art methods for both missing features imputation and image inpainting problems. ", + "bbox": [ + 148, + 396, + 851, + 494 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "REFERENCES ", + "text_level": 1, + "bbox": [ + 148, + 517, + 259, + 532 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Philip Bachman and Doina Precup. Data generation as sequential decision making. In Advances in Neural Information Processing Systems, pp. 3249–3257, 2015. ", + "bbox": [ + 145, + 542, + 849, + 570 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Marcelo Bertalmio, Guillermo Sapiro, Vincent Caselles, and Coloma Ballester. Image inpainting. In Proceedings of the 27th Annual Conference on Computer Graphics and Interactive Techniques, SIGGRAPH ’00, pp. 417–424, New York, NY, USA, 2000. ACM Press/Addison-Wesley Publishing Co. ISBN 1- 58113-208-5. ", + "bbox": [ + 147, + 583, + 851, + 640 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Florian Bordes, Sina Honari, and Pascal Vincent. Learning to generate samples from noise through infusion training. In International Conference on Learning Representations, 2017. ", + "bbox": [ + 143, + 654, + 849, + 683 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Yuri Burda, Roger Grosse, and Ruslan Salakhutdinov. Importance weighted autoencoders. arXiv preprint arXiv:1509.00519, 2015. ", + "bbox": [ + 145, + 695, + 848, + 724 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "S van Buuren and Karin Groothuis-Oudshoorn. mice: Multivariate imputation by chained equations in r. Journal of statistical software, pp. 1–68, 2010. ", + "bbox": [ + 143, + 738, + 848, + 767 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Laura Douglas, Iliyan Zarov, Konstantinos Gourgoulias, Chris Lucas, Chris Hart, Adam Baker, Maneesh Sahani, Yura Perov, and Saurabh Johri. A universal marginalizer for amortized inference in generative models. arXiv preprint arXiv:1711.00695, 2017. ", + "bbox": [ + 148, + 780, + 851, + 821 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "Mathieu Germain, Karol Gregor, Iain Murray, and Hugo Larochelle. Made: Masked autoencoder for distribution estimation. In Francis Bach and David Blei (eds.), Proceedings of the 32nd International Conference on Machine Learning, volume 37 of Proceedings of Machine Learning Research, pp. 881–889, Lille, France, 07–09 Jul 2015. PMLR. URL http://proceedings.mlr.press/v37/germain15. html. ", + "bbox": [ + 147, + 119, + 851, + 189 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in neural information processing systems, pp. 2672–2680, 2014. ", + "bbox": [ + 147, + 199, + 852, + 242 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Anirudh Goyal Alias Parth Goyal, Nan Ke, Surya Ganguli, and Yoshua Bengio. Variational walkback: Learning a transition operator as a stochastic recurrent net. In Advances in Neural Information Processing Systems, pp. 4392–4402, 2017. ", + "bbox": [ + 148, + 251, + 848, + 294 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770–778, 2016. ", + "bbox": [ + 143, + 301, + 848, + 332 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014. ", + "bbox": [ + 145, + 340, + 849, + 369 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Diederik P. Kingma and Max Welling. Auto-encoding variational bayes. CoRR, abs/1312.6114, 2013. ", + "bbox": [ + 150, + 377, + 815, + 393 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Brenden M Lake, Ruslan Salakhutdinov, and Joshua B Tenenbaum. Human-level concept learning through probabilistic program induction. Science, 350(6266):1332–1338, 2015. ", + "bbox": [ + 147, + 401, + 849, + 431 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Yann LeCun, Corinna Cortes, and Christopher JC Burges. The mnist database of handwritten digits, 1998. ", + "bbox": [ + 148, + 439, + 846, + 455 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Yijun Li, Sifei Liu, Jimei Yang, and Ming-Hsuan Yang. Generative face completion. In The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), volume 1, pp. 3, 2017. ", + "bbox": [ + 147, + 463, + 849, + 493 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "M. Lichman. UCI machine learning repository, 2013. URL http://archive.ics.uci.edu/ml. ", + "bbox": [ + 148, + 501, + 830, + 517 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Ziwei Liu, Ping Luo, Xiaogang Wang, and Xiaoou Tang. Deep learning face attributes in the wild. In Proceedings of International Conference on Computer Vision (ICCV), December 2015. ", + "bbox": [ + 147, + 525, + 848, + 555 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Alireza Makhzani, Jonathon Shlens, Navdeep Jaitly, and Ian Goodfellow. Adversarial autoencoders. In International Conference on Learning Representations, 2016. URL http://arxiv.org/abs/1511. 05644. ", + "bbox": [ + 148, + 563, + 848, + 606 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Xiaojiao Mao, Chunhua Shen, and Yu-Bin Yang. Image restoration using very deep convolutional encoderdecoder networks with symmetric skip connections. In Advances in neural information processing systems, pp. 2802–2810, 2016. ", + "bbox": [ + 147, + 614, + 848, + 657 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Mehdi Mirza and Simon Osindero. Conditional generative adversarial nets. CoRR, abs/1411.1784, 2014. ", + "bbox": [ + 145, + 665, + 841, + 683 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Deepak Pathak, Philipp Krahenb ¨ uhl, Jeff Donahue, Trevor Darrell, and Alexei Efros. Context encoders: ¨ Feature learning by inpainting. In Computer Vision and Pattern Recognition (CVPR), 2016. ", + "bbox": [ + 147, + 689, + 848, + 719 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Danilo Jimenez Rezende, Shakir Mohamed, and Daan Wierstra. Stochastic backpropagation and approximate inference in deep generative models. In International Conference on Machine Learning, pp. 1278– 1286, 2014. ", + "bbox": [ + 148, + 728, + 848, + 771 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical image computing and computer-assisted intervention, pp. 234–241. Springer, 2015. ", + "bbox": [ + 148, + 780, + 851, + 823 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning using nonequilibrium thermodynamics. In Francis Bach and David Blei (eds.), Proceedings of the 32nd International Conference on Machine Learning, volume 37 of Proceedings of Machine Learning Research, pp. 2256–2265, Lille, France, 07–09 Jul 2015. PMLR. URL http://proceedings.mlr. press/v37/sohl-dickstein15.html. ", + "bbox": [ + 148, + 121, + 851, + 189 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Kihyuk Sohn, Honglak Lee, and Xinchen Yan. Learning structured output representation using deep conditional generative models. In C. Cortes, N. D. Lawrence, D. D. Lee, M. Sugiyama, and R. Garnett (eds.), Advances in Neural Information Processing Systems 28, pp. 3483–3491. Curran Associates, Inc., 2015. ", + "bbox": [ + 147, + 200, + 851, + 242 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Casper Kaae Sønderby, Tapani Raiko, Lars Maaløe, Søren Kaae Sønderby, and Ole Winther. Ladder variational autoencoders. In Advances in neural information processing systems, pp. 3738–3746, 2016. ", + "bbox": [ + 142, + 251, + 849, + 280 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Daniel J Stekhoven and Peter Buhlmann. Missforest - non-parametric missing value imputation for mixed-¨ type data. Bioinformatics, 28(1):112–118, 2011. ", + "bbox": [ + 143, + 290, + 849, + 319 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Benigno Uria, Marc-Alexandre Cotˆ e, Karol Gregor, Iain Murray, and Hugo Larochelle. Neural autoregres- ´ sive distribution estimation. The Journal of Machine Learning Research, 17(1):7184–7220, 2016. ", + "bbox": [ + 145, + 327, + 849, + 357 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "C. Yang, X. Lu, Z. Lin, E. Shechtman, O. Wang, and H. Li. High-resolution image inpainting using multiscale neural patch synthesis. In 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 4076–4084, July 2017. doi: 10.1109/CVPR.2017.434. ", + "bbox": [ + 148, + 364, + 849, + 409 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Raymond Yeh, Chen Chen, Teck-Yian Lim, Mark Hasegawa-Johnson, and Minh N. Do. Semantic image inpainting with perceptual and contextual losses. CoRR, abs/1607.07539, 2016. ", + "bbox": [ + 143, + 417, + 849, + 445 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Raymond A Yeh, Chen Chen, Teck Yian Lim, Alexander G Schwing, Mark Hasegawa-Johnson, and Minh N Do. Semantic image inpainting with deep generative models. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pp. 5485–5493, 2017. ", + "bbox": [ + 148, + 455, + 849, + 498 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Jinsung Yoon, James Jordon, and Mihaela van der Schaar. GAIN: Missing data imputation using generative adversarial nets. In Jennifer Dy and Andreas Krause (eds.), Proceedings of the 35th International Conference on Machine Learning, volume 80 of Proceedings of Machine Learning Research, pp. 5689–5698, Stockholmsmssan, Stockholm Sweden, 10–15 Jul 2018. PMLR. URL http://proceedings.mlr. press/v80/yoon18a.html. ", + "bbox": [ + 148, + 507, + 851, + 578 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "APPENDIX ", + "text_level": 1, + "bbox": [ + 150, + 604, + 236, + 619 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "A EXPERIMENTAL DETAILS ", + "text_level": 1, + "bbox": [ + 150, + 638, + 393, + 655 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "A.1 NEURAL NETWORK ARCHITECTURES ", + "bbox": [ + 150, + 670, + 452, + 684 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "In all experiments we use optimization method Adam (Kingma & Ba, 2014), skip-connections between prior network and generative network inspired by (Mao et al., 2016), (Sønderby et al., 2016) and (Ronneberger et al., 2015), and convolutional neural networks based on ResNet blocks (He et al., 2016). ", + "bbox": [ + 148, + 695, + 851, + 738 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "Without skip-connections all information for decoder goes through the latent variables. In image inpainting we found skip-connections very useful in both terms of log-likelihood improvement and the image realism, because latent variables are responsible for the global information only while the local information passes through skip-connections. Therefore the border between image and inpainting becomes less conspicuous. ", + "bbox": [ + 148, + 744, + 851, + 801 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "The main idea of neural networks architecture is reflected in figure 5. ", + "bbox": [ + 147, + 808, + 598, + 821 + ], + "page_idx": 11 + }, + { + "type": "image", + "img_path": "images/5b67d91960752eb18692b5b61ef536a092947e8d387abb82437625c2d41f393a.jpg", + "image_caption": [ + "Figure 5: Neural network architecture for inpainting. " + ], + "image_footnote": [], + "bbox": [ + 161, + 137, + 833, + 352 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "The number of hidden layers, their widths and structure may be different. ", + "bbox": [ + 148, + 426, + 627, + 441 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "The neural networks we used for image inpainting have He-Uniform initialization of convolutional ResNet blocks, and the skip-connections are implemented using concatenation, not addition. The proposal network structure is exactly the same as the prior network except skip-connections. ", + "bbox": [ + 148, + 448, + 852, + 491 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "Also one could use much simpler fully-connected networks with one hidden layer as a proposal, prior and generative networks in VAEAC and still obtain nice inpaintings on MNIST. ", + "bbox": [ + 147, + 497, + 849, + 526 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "A.2 MISSING FEATURES IMPUTATION ", + "text_level": 1, + "bbox": [ + 150, + 559, + 421, + 573 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "We split the dataset into train and test set with size ratio 3:1. Before training we drop randomly $50 \\%$ of values both in train and test set. We repeat each experiment 5 times with different train-test splits and dropped features and then average results and compute their standard deviation. ", + "bbox": [ + 147, + 590, + 856, + 633 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "As we show in appendix B.2, the better results can be achieved when the model learns the concatenation of objects features $x$ and targets $y$ . So we treat $y$ as an additional feature that is always unobserved during the testing time. ", + "bbox": [ + 147, + 640, + 852, + 683 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "To train our model we use distribution $p ( b _ { i } | x )$ in which $p ( b _ { i } | x _ { i } = \\omega ) = 1$ and $p ( b _ { i } | x ) = 0 . 2$ otherwise. Also for VAEAC trainig we normalize real-valued features, fix $\\sigma _ { \\theta } = 1$ in the generative model of VAEAC in order to optimize RMSE, and use $2 5 \\%$ of training data as validation set to select the best model among all epochs of training. ", + "bbox": [ + 148, + 689, + 849, + 746 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "For the test set, the classifier or regressor is applied to each of the $n$ imputed objects and the predictions are combined. For regression problems we report R2-score of combined predictions, so we use averaging as a combination method. For classification problem we report accuracy, and therefore choose the mode. We consider the workflow where the imputed values of $y$ are not fed to the classifier or regressor to make a fair comparison of feature imputation quality. ", + "bbox": [ + 148, + 752, + 851, + 823 + ], + "page_idx": 12 + }, + { + "type": "table", + "img_path": "images/04d4687dcf9699d512eac4a44124a37465506a687d19b17f4d1f42a5bbb655e0.jpg", + "table_caption": [ + "Table 6: Generative Face Completion (Li et al., 2017) masks. Image size is 128x128. " + ], + "table_footnote": [], + "table_body": "
MaskMeaningX1x2y1y2
01Left half of the face337052115
02Right half of the face577095115
03Two eyes29985273
04Left eye29665273
05Right eye61995273
06Lower half of the face408786123
", + "bbox": [ + 312, + 142, + 686, + 244 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "NRMSE or PFC for dataset is computed as an average of NRMSE or PFC of all features of this dataset. NRMSE of a feature is just RMSE of imputations divided by the standard deviation of this feature. PFC of a feature is a proportion of imputations which are incorrect. ", + "bbox": [ + 148, + 270, + 849, + 313 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "A.3 IMAGE INPAINTING DATASETS AND MASKS ", + "text_level": 1, + "bbox": [ + 150, + 329, + 493, + 344 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "MNIST is a dataset of 60000 train and 10000 test grayscale images of digits from 0 to 9 of size $2 8 \\mathbf { x } 2 8$ . We binarize all images in the dataset. For MNIST we consider Bernoulli log-likelihood as the reconstruction loss: $\\begin{array} { r } { \\log p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b ) = \\sum _ { i : b _ { i } = 1 } \\log \\mathrm { B e r n o u l l i } ( x _ { i } | p _ { \\theta , i } ( z , x _ { 1 - b } , b ) ) } \\end{array}$ where $p _ { \\theta , i } ( z , x _ { 1 - b } , b )$ is an output of the generative neural network. We use 16 latent variables. In the mask for this dataset the observed pixels form a three pixels wide horizontal line which position is distributed uniformly. ", + "bbox": [ + 147, + 356, + 851, + 426 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Omniglot is a dataset of 19280 train and 13180 test black-and-white images of different alphabets symbols of size $1 0 5 \\mathrm { x } 1 0 5$ . As in previous section, the brightness of each pixel is treated as a Bernoulli probability of it to be 1. The mask we use is a random rectangular which is described below. We use 64 latent variables. We train model for 50 epochs and choose best model according to IWAE log-likelihood estimation on the validation set after each epoch. ", + "bbox": [ + 147, + 441, + 849, + 512 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "CelebA is a dataset of 162770 train, 19867 validation and 19962 test color images of faces of celebrities of size $1 7 8 \\mathrm { x } 2 1 8$ . Before learning we normalize the channels in dataset. We use logarithm of fully-factorized Gaussian distribution as reconstruction loss. The mask we use is a random rectangular which is describe below. We use 32 latent variables. ", + "bbox": [ + 148, + 527, + 849, + 584 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "Rectangular mask is the common shape of unobserved region in image inpainting. We use such mask for Omniglot and Celeba. We sample the corner points of rectangles uniprobably on the image, but reject those rectangles which area is less than a quarter of the image area. ", + "bbox": [ + 148, + 599, + 851, + 642 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "In Li et al. (2017) six different masks O1–O6 are used on the testing stage. We reconstruct the positions of masks from the illustrations in the paper and give their coordinates in table 6. The visualizations of the masks are available in figure 10. ", + "bbox": [ + 148, + 659, + 849, + 700 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "At the training stage we used a rectangle mask with uniprobable random corners. We reject masks with width or height less than 16pt. We use 64 latent variables and take the best model over 50 epochs based on the validation IWAE log-likelihood estimation. We can obtain slightly higher PSNR values than reported in table 4 if use only masks O1–O6 at the training stage. ", + "bbox": [ + 148, + 707, + 849, + 763 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "In Yeh et al. (2017) four types of masks are used. Center mask is just an unobserved $3 2 \\mathrm { x } 3 2 $ square in the center of 64x64 image. Half mask mean that one of upper, lower, left or right half of the image is unobserved. All these types of a half are equiprobable. Random mask means that we use pixelwise-independent Bernoulli distribution with probability 0.8 to form a mask of unobserved pixels. Pattern mask is proposed in Pathak et al. (2016). As we deduced from the code 3, the generation process is follows: firstly we generate $6 0 0 \\times 6 0 0$ one-channel image with uniform distribution over pixels, then bicubically interpolate it to image of size $1 0 0 0 0 \\mathrm { x } 1 0 0 0 0$ , and then apply Heaviside step function $H ( x - 0 . 2 5 )$ (i. e. all points with value less than 0.25 are considered as unobserved). To sample a mask we sample a random position in this $1 0 0 0 0 \\mathrm { x } 1 0 0 0 0$ binary image and crop $6 4 \\mathrm { x } 6 4$ mask. If less than $20 \\%$ or more than $30 \\%$ of pixel are unobserved, than the mask is rejected and the position is sampled again. In comparison with this paper in section 5.2 we use the same distribution over masks at training and testing stages. We use VAEAC with 64 latent variables and take the best model over 50 epochs based on the validation IWAE log-likelihood estimation. ", + "bbox": [ + 148, + 780, + 851, + 823 + ], + "page_idx": 13 + }, + { + "type": "text", + "text": "", + "bbox": [ + 148, + 119, + 851, + 246 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "A.4 GAIN IMPLEMENTATION DETAILS ", + "text_level": 1, + "bbox": [ + 150, + 271, + 431, + 285 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "For missing feature imputation we reimplemented GAIN in PyTorch based on the paper (Yoon et al., 2018) and the available TensorFlow source code for image inpainting 4. ", + "bbox": [ + 147, + 300, + 844, + 330 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "For categorical features we use one-hot encoding. We observe in experiments that it works better in terms of NRMSE and PFC than processing categorical features in GAIN as continuous ones and then rounding them to the nearest category. ", + "bbox": [ + 148, + 335, + 851, + 378 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "For categorical features we also use reconstruction loss $\\begin{array} { r } { L _ { M } ( x _ { i } , x _ { i } ^ { \\prime } ) = - \\frac { 1 } { | X _ { i } | } \\sum _ { j = 1 } ^ { | X _ { i } | } x _ { i , j } \\log ( x _ { i , j } ^ { \\prime } ) } \\end{array}$ $\\left| X _ { i } \\right|$ the number of categories of the $i$ -th feature, and $x _ { i , j }$ is the $j$ -th component of one-hot encoding of the feature $x _ { i }$ . Such $L _ { M }$ enforces equal contribution of each categorical feature into the whole reconstruction loss. ", + "bbox": [ + 148, + 386, + 849, + 434 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "We use one more modification of $L _ { M } ( x , x ^ { \\prime } )$ for binary and categorical features. Cross-entropy loss in $L _ { M }$ penalizes incorrect reconstructions of categorical and binary features much more than incorrect reconstructions for continuous ones. To avoid such imbalance we mixed L2 and cross-entropy reconstruction losses for binary and categorical features with weights 0.8 and 0.2 respectively: ", + "bbox": [ + 148, + 439, + 848, + 496 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "We observe in experiments that this modification also works better in terms of NRMSE and PFC than the original model. ", + "bbox": [ + 151, + 566, + 851, + 595 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "We use validation set which contains $5 \\%$ of the observed features for the best model selection (hyperparameter is the number of iterations). ", + "bbox": [ + 147, + 602, + 849, + 630 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "In the original GAIN paper authors propose to use cross-validation for hyper-parameter $\\alpha \\in \\{ 0 . 1 , 0 . 5 , 1 , 2 , 1 0 \\}$ . We observe that using $\\alpha ~ = ~ 1 0$ and a hint $h \\ = \\ b \\circ \\ m \\ + \\ 0 . 5 ( 1 \\ - \\ b )$ where vector $b$ is sampled from Bernoulli distribution with $p = 0 . 0 1$ provides better results in terms of NRMSE and PFC than the original model with every $\\alpha \\in \\{ 0 . 1 , 0 . 5 , 1 , 2 , 1 0 \\}$ . Such hint distribution makes model theoretically inconsistent but works well in practice (see table 7). ", + "bbox": [ + 148, + 637, + 851, + 707 + ], + "page_idx": 14 + }, + { + "type": "text", + "text": "Table 7 shows that our modifications provide consistently not worse or even better imputations than the original GAIN (in terms of NRMSE and PFC, on the considered datasets). So in this paper for the missing feature imputation problem we report the results of our modification of GAIN. ", + "bbox": [ + 148, + 713, + 849, + 756 + ], + "page_idx": 14 + }, + { + "type": "table", + "img_path": "images/bd6b6b98454a1eb53c64484cc4462e2c757911704352b0336a4bd11fa6d9a4c0.jpg", + "table_caption": [ + "Table 7: NRMSE (for continuous datasets) or PFC (for categorical ones) of imputations for different GAIN modifications. Less is better. “Our modification” includes the reconstruction loss $L _ { M } ^ { \\prime }$ (12), Bernoulli distribution over $b$ in the hint generation procedure, and fixed $\\alpha = 1 0$ . Other columns refers original GAIN without these modifications and with different values of $\\alpha$ . " + ], + "table_footnote": [], + "table_body": "
DatasetOur modificationα=10α=2α=1α= 0.5α=0.1
Boston0.78±0.030.87±0.021.0 ± 0.11.0 ± 0.11.02 ± 0.051.6±0.2
Breast0.67 ± 0.010.80±0.051.00 ± 0.051.10 ± 0.071.19 ± 0.051.52 ± 0.06
Concrete0.96 ± 0.010.98 ± 0.021.02 ± 0.021.13 ± 0.061.17 ± 0.041.3 ± 0.1
Diabetes0.911 ± 0.0090.93 ±0.031.05 ± 0.041.07 ± 0.071.21 ± 0.071.6 ± 0.1
Digits0.79 ± 0.020.88 ± 0.011.05 ± 0.021.13 ± 0.021.24 ± 0.081.4± 0.2
Glass1.06 ± 0.051.04 ± 0.051.19 ± 0.061.4 ± 0.21.6 ± 0.11.81 ± 0.10
Iris0.72 ±0.040.73±0.060.83 ±0.080.97 ± 0.091.2 ± 0.21.3±0.2
Mushroom0.271 ± 0.0030.404 ± 0.0040.52 ± 0.050.55 ± 0.010.56 ± 0.030.64± 0.06
Orthopedic0.91 ± 0.030.91 ±0.081.1 ± 0.11.2 ± 0.11.34 ± 0.081.6 ± 0.2
Phishing0.427 ± 0.0100.52 ±0.020.54±0.020.543 ± 0.0100.56 ± 0.010.57 ± 0.04
WallRobot0.907 ± 0.0050.924± 0.0050.933 ± 0.0080.95 ± 0.011.00 ± 0.021.26 ± 0.04
WhiteWine0.97±0.021.02 ± 0.041.2 ± 0.11.3 ± 0.11.6 ± 0.11.86 ± 0.08
Yeast0.99 ±0.031.3±0.21.6 ± 0.11.83 ± 0.091.9 ±0.12.4± 0.4
Zoo0.20 ±0.020.24± 0.050.35 ± 0.060.36 ± 0.030.43 ± 0.040.433 ± 0.004
", + "bbox": [ + 150, + 184, + 848, + 391 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "B THEORY ", + "text_level": 1, + "bbox": [ + 148, + 415, + 253, + 431 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "B.1 VAEAC UNIVERSALITY ", + "text_level": 1, + "bbox": [ + 148, + 446, + 359, + 462 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "The theoretical guarantees that VAEAC can model arbitrary distribution are based on the same guarantees for Condtitional Variational Autoencoder (CVAE). We prove below that if CVAE can model each of the conditional distributions $p ( x _ { b } | x _ { 1 - b } )$ , then VAEAC can model all of them. ", + "bbox": [ + 148, + 473, + 851, + 516 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "We can imagine $2 ^ { D }$ CVAEs learned each for the certain mask. Because neural networks are universal approximators, VAEAC networks could model the union of CVAE networks, so that VAEAC network performs transformation defined by the same network of the corresponding to the given mask CVAE. ", + "bbox": [ + 150, + 522, + 849, + 565 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/ebd07ceb7235754b921ee91da23d856f5d6aa8b4c7359835c4a8bda147c1fb4c.jpg", + "text": "$$\np _ { \\psi , V A E A C } ( z | x _ { 1 - b } , b ) = p _ { \\psi , C V A E , 1 - b } ( z | x _ { 1 - b } ) \\forall x , b\n$$", + "text_format": "latex", + "bbox": [ + 315, + 573, + 678, + 590 + ], + "page_idx": 15 + }, + { + "type": "equation", + "img_path": "images/6dbaafed2c4133461a5e6db047a62653e52e4a0973ef592faa2ba8d413f8f487.jpg", + "text": "$$\np _ { \\theta , V A E A C } ( x _ { b } | z , x _ { 1 - b } , b ) = p _ { \\theta , C V A E , 1 - b } ( x _ { b } | z , x _ { 1 - b } ) \\forall z , x , b\n$$", + "text_format": "latex", + "bbox": [ + 294, + 597, + 705, + 614 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "So if CVAE models any distribution $p ( x | y )$ , VAEAC also do. ", + "bbox": [ + 148, + 617, + 549, + 632 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "The guarantees for CVAE in the case of continuous variables are based on the point that every smooth distribution can be approximated with a large enough mixture of Gaussians, which is a special case of CVAE’s generative model. These guarantees can be extended on the case of categorical-continuous variables also. Actually, there are distributions over categorical variables which CVAE with Gaussian prior and proposal distributions cannot learn. Nevertheless, this kind of limitation is not fundamental and is caused by poor proposal distribution family. ", + "bbox": [ + 148, + 638, + 852, + 723 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "B.2 WHY VAEAC NEEDS TARGET VALUES FOR MISSING FEATURES IMPUTATION? ", + "text_level": 1, + "bbox": [ + 148, + 739, + 738, + 755 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "Consider a dataset with $D$ -dimensional objects $x$ where each feature may be missing (which we denote by $x _ { i } = \\omega$ ) and their target values $y$ . In this section we show that the better results are achieved when our model learns the concatenation of objects features $x$ and targets $y$ . The example that shows the necessity of it is following. Consider a dataset where $x _ { 1 } = 1$ , $x _ { 2 } \\sim \\mathcal { N } ( \\bar { x } _ { 2 } | y , 1 )$ , $p _ { d } ( y = \\mathrm { { 0 } ) = { { p } ( y = 5 ) = 0 . 5 } }$ . In this case $p _ { d } ( x _ { 2 } | x _ { 1 } = 1 ) = 0 . 5 \\mathcal { N } ( x _ { 2 } | 0 , 1 ) + 0 . 5 \\mathcal { N } ( x _ { 2 } | 5 , 1 )$ . We can see that generating data from $p _ { d } ( x _ { 2 } | x _ { 1 } )$ may only confuse the classifier, because with probability 0.5 it generates $x _ { 2 } \\sim \\bar { \\mathcal { N } } ( 0 , 1 )$ for $y = 5$ and $x _ { 2 } \\sim \\mathcal { N } ( 5 , 1 )$ for $y = 0$ . On the other hand, $p _ { d } ( x _ { 2 } | x _ { 1 } , y ) = \\mathcal { N } ( x _ { 2 } | y , 1 )$ . Filling gaps using $p _ { d } ( x _ { 2 } | x _ { 1 } , y )$ may only improve classifier or regressor by giving it some information from the joint distribution $p _ { d } ( x , y )$ and thus simplifying the dependence to be learned at the training time. So we treat $y$ as an additional feature that is always unobserved during the testing time. ", + "bbox": [ + 148, + 766, + 851, + 823 + ], + "page_idx": 15 + }, + { + "type": "text", + "text": "", + "bbox": [ + 148, + 119, + 851, + 204 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "B.3 UNIVERSAL MARGINALIZER: TRAINING PROCEDURE MODIFICATION ", + "text_level": 1, + "bbox": [ + 147, + 220, + 671, + 236 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "The problem authors did not address in the original paper is the relation between the distribution of unobserved components $p ( b )$ at the testing stage and the distribution of masks in the requests to UM ${ \\hat { p } } ( b )$ . The distribution over masks $p ( b )$ induces the distribution ${ \\hat { p } } ( b )$ , and in the most cases $p ( b ) \\neq { \\hat { p } } ( b )$ . The distribution ${ \\hat { p } } ( b )$ also depends on the permutations $( i _ { 1 } , i _ { 2 } , \\dots , i _ { | b | } )$ that we use to generate objects. ", + "bbox": [ + 148, + 246, + 851, + 304 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "We observed in experiments, that UM must be trained using unobserved mask distribution ${ \\hat { p } } ( b )$ . For example, if all masks from $p ( b )$ have a fixed number of unobserved components (e. g., $\\begin{array} { l } { { \\frac { D } { 2 } } } \\end{array}$ ), then UM will never see an example of mask with $\\begin{array} { r } { { 1 , 2 , \\ldots , \\frac { D } { 2 } - 1 } } \\end{array}$ unobserved components, which is necessary to generate a sample conditioned on $\\textstyle { \\frac { D } { 2 } }$ components. That leads to drastically low likelihood estimate for the test set and unrealistic samples. ", + "bbox": [ + 147, + 310, + 851, + 388 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "We developed an easy generative process for ${ \\hat { p } } ( b )$ for arbitrary $p ( b )$ if the permutation of unobserved components $( i _ { 1 } , i _ { 2 } , \\dots , i _ { | b | } )$ is chosen randomly and equiprobably: firstly we generate $b _ { 0 } \\sim p ( b )$ , $u \\sim U [ 0 , 1 ]$ , then $b _ { 1 } \\sim ( \\mathrm { B e r n o u l l i } ( u ) ) ^ { D }$ and $b = b _ { 0 } \\circ b _ { 1 }$ . More complicated generative process exists for a sorted permutation where $i _ { j - 1 } < i _ { j } \\forall j : 2 \\le j \\le | b |$ . ", + "bbox": [ + 148, + 393, + 851, + 455 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "In experiments we use uniform distribution over the permutations. ", + "bbox": [ + 148, + 459, + 580, + 474 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "C GAUSSIAN STOCHASTIC NEURAL NETWORK ", + "text_level": 1, + "bbox": [ + 148, + 494, + 553, + 511 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Gaussian stochastic neural network (13) and hybrid model (14) are originally proposed in the paper on Conditional VAE (Sohn et al., 2015). The motivation authors mention in the paper is as follows. During training the proposal distribution $q _ { \\phi } ( z | x , y )$ is used to generate the latent variables $z$ , while during the testing stage the prior $p _ { \\psi } ( z | y )$ is used. KL divergence tries to close the gap between two distributions but, according to authors, it is not enough. To overcome the issue authors propose to use a hybrid model (14), a weighted mixture of variational lower bound (3) and a single-sample Monte-Carlo estimation of log-likelihood (13). The model corresponding to the second term is called Gaussian Stochastic Neural Network (13), because it is a feed-forward neural network with a single Gaussian stochastic layer in the middle. Also GSNN is a special case of CVAE where $q _ { \\phi } ( z | x , y ) = p _ { \\psi } ( z | y )$ . ", + "bbox": [ + 147, + 525, + 851, + 654 + ], + "page_idx": 16 + }, + { + "type": "equation", + "img_path": "images/6eda0f59e7f92ff41e13fc03a857ebaa2e07dbc1dc3cfeb2358c4de8661097dd.jpg", + "text": "$$\n\\begin{array} { r } { L _ { G S N N } ( x , y ; \\theta , \\psi ) = \\mathbb { E } _ { p _ { \\psi } ( z | y ) } \\log p _ { \\theta } ( x | z , y ) \\qquad } \\\\ { L ( x , y ; \\theta , \\psi , \\phi ) = \\alpha L _ { C V A E } ( x , y ; \\theta , \\psi , \\phi ) + ( 1 - \\alpha ) L _ { G S N N } ( x , y ; \\theta , \\psi ) , \\quad \\alpha \\in [ 0 , 1 ] } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 215, + 670, + 782, + 710 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "Authors report that hybrid model and GSNN outperform CVAE in terms of segmentation accuracy on the majority of datasets. ", + "bbox": [ + 147, + 717, + 852, + 746 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "We can also add that this technique seems to soften the “holes problem” (Makhzani et al., 2016). In Makhzani et al. (2016) authors observe that vectors $z$ from prior distribution may be different enough from all vectors $z$ from the proposal distribution at the training stage, so the generator network may be confused at the testing stage. Due to this problem CVAE can have good reconstructions of $y$ given $z \\sim q _ { \\phi } ( z | x , y )$ , while samples of $y$ given $z \\sim p _ { \\psi } ( z | x )$ are not realistic. ", + "bbox": [ + 148, + 752, + 851, + 823 + ], + "page_idx": 16 + }, + { + "type": "text", + "text": "The same trick is applicable to our model as well: ", + "bbox": [ + 147, + 119, + 475, + 136 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/c33502d1bb26f985e83664b8c446d3988a4e59347df95247f5e7bef837ca1711.jpg", + "text": "$$\n\\begin{array} { r l } & { L _ { G S N N } ( x , b ; \\theta , \\psi ) = \\mathbb { E } _ { p _ { \\psi } ( z | x _ { 1 - b } , b ) } \\log p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b ) } \\\\ & { L ( x , b ; \\theta , \\psi , \\phi ) = \\alpha L _ { V A E A C } ( x , b ; \\theta , \\psi , \\phi ) + ( 1 - \\alpha ) L _ { G S N N } ( x , b ; \\theta , \\psi ) , \\quad \\alpha \\in [ 0 , 1 ] } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 214, + 142, + 784, + 181 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "In order to reflect the difference between sampling $z$ from prior and proposal distributions, authors of CVAE use two methods of log-likelihood estimation: ", + "bbox": [ + 143, + 194, + 856, + 223 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/583a13fe422be7a69bc3cb86f05f54ddd22119c286cd1b5e0c02498f08abc599.jpg", + "text": "$$\n\\log p _ { \\theta , \\psi } ( x | y ) \\approx \\log \\frac { 1 } { S } \\sum _ { i = 1 } ^ { S } p _ { \\theta } ( x | z _ { i } , y ) , ~ z _ { i } \\sim p _ { \\psi } ( z | y )\n$$", + "text_format": "latex", + "bbox": [ + 316, + 231, + 684, + 275 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/6ccb2cb6db773a0184ddddcd940622b1174f4ffcf1db0c6a0ee7b8d59e0ea61b.jpg", + "text": "$$\n\\log p _ { \\theta , \\psi } ( x | y ) \\approx \\log \\frac { 1 } { S } \\sum _ { i = 1 } ^ { S } \\frac { p _ { \\theta } ( x | z _ { i } , y ) p _ { \\psi } ( z _ { i } | y ) } { q _ { \\phi } ( z _ { i } | x , y ) } , ~ z _ { i } \\sim q _ { \\phi } ( z | x , y )\n$$", + "text_format": "latex", + "bbox": [ + 279, + 281, + 722, + 325 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "The first estimator is called Monte-Carlo estimator and the second one is called Importance Sampling estimator (also known as IWAE). They are asymptotically equivalent, but in practice the Monte-Carlo estimator requires much more samples to obtain the same accuracy of estimation. Small $S$ leads to underestimation of the log-likelihood for both Monte-Carlo and Importance Sampling (Burda et al., 2015), but for Monte-Carlo the underestimation is expressed much stronger. ", + "bbox": [ + 147, + 329, + 852, + 400 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "We perform an additional study of GSNN and hybrid model and show that they have drawbacks when the target distribution $p ( x | y )$ is has multiple different local maximums. ", + "bbox": [ + 147, + 405, + 851, + 434 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "C.1 THEORETICAL STUDY ", + "text_level": 1, + "bbox": [ + 148, + 452, + 344, + 467 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "In this section we show why GSNN cannot learn distributions with several different modes and leads to a blurry image samples. ", + "bbox": [ + 145, + 478, + 851, + 508 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "For the simplicity of the notation we consider hybrid model for a standard VAE: ", + "bbox": [ + 150, + 513, + 674, + 529 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/c1559493b96d28a0f47430d3d517775906770a2762939c39f8bd304ea066eecd.jpg", + "text": "$$\nL ( x ; \\phi , \\psi , \\theta ) = \\alpha \\mathbb { E } _ { z \\sim q _ { \\phi } ( z \\mid x ) } \\log \\frac { p _ { \\theta } ( x \\mid z ) p _ { \\psi } ( z ) } { q _ { \\phi } ( z \\mid x ) } + ( 1 - \\alpha ) \\mathbb { E } _ { z \\sim p _ { \\psi } ( z ) } \\log p _ { \\theta } ( x \\mid z )\n$$", + "text_format": "latex", + "bbox": [ + 238, + 536, + 759, + 570 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "The hybrid model (16) for VAEAC can be obtained from (19) by replacing $x$ with $x _ { b }$ and conditioning all distributions on $x _ { 1 - b }$ and $b$ . The validity of the further equations and conclusions remains for VAEAC after this replacement. ", + "bbox": [ + 147, + 577, + 854, + 619 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Consider now a categorical latent variable $z$ which can take one of $K$ values. Let $x$ be a random variable witfor $p _ { d } ( x )$ to be modele some values following true data distribut. So the true distribution has n: d $\\begin{array} { r } { p _ { d } ( x = x _ { i } ) = \\frac { 1 } { K } } \\end{array}$ $i \\in \\{ 1 , 2 , \\ldots , K \\}$ $x _ { 1 } , x _ { 2 } , \\dotsc , x _ { K }$ $K$ able modes. Suppose the generator network $N N _ { \\theta }$ which models mapping from $z$ to some vector of parameters $\\begin{array} { r l r } { v _ { z } } & { { } = } & { \\bar { N } N _ { \\theta } ( z ) } \\end{array}$ . Thus, we define generative distribution as some function of these parameters: $p _ { \\theta } ( x | z ) \\ : = \\ : f ( x , v _ { z } )$ . Therefore, the parameters $\\theta$ are just the set of $v _ { 1 } , v _ { 2 } , \\dotsc , v _ { K }$ . ", + "bbox": [ + 147, + 626, + 851, + 712 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "For the simplicity of the model we assume $\\begin{array} { r } { p _ { \\psi } ( z ) = \\frac { 1 } { K } } \\end{array}$ . Taking into account $\\begin{array} { r } { p _ { \\psi } ( z ) = \\frac { 1 } { K } } \\end{array}$ , we obtain optimal $\\begin{array} { r } { q ( z = i | x ) = \\frac { f ( x , v _ { i } ) } { \\sum _ { j = 1 } ^ { K } f ( x , v _ { j } ) } } \\end{array}$ Using (19) and the above formulas for $q _ { \\phi } , p _ { \\psi }$ and $p _ { \\theta }$ we obtain the following optimization problem: ", + "bbox": [ + 147, + 715, + 851, + 768 + ], + "page_idx": 17 + }, + { + "type": "equation", + "img_path": "images/07b4e4827ba4de27da9506a427292dfd27100f30c89238d1b1a7b66b439dfe73.jpg", + "text": "$$\n\\operatorname* { m a x } _ { v _ { 1 } , v _ { 2 } , \\ldots , v _ { K } } \\frac { 1 } { K } \\sum _ { i = 1 } ^ { K } \\left[ \\alpha \\sum _ { j = 1 } ^ { K } \\frac { f ( x _ { i } , v _ { j } ) } { \\sum _ { k = 1 } ^ { K } f ( x _ { i } , v _ { k } ) } \\log \\frac { f ( x _ { i } , v _ { j } ) \\frac { 1 } { K } } { \\frac { f ( x _ { i } , v _ { j } ) } { \\sum _ { k = 1 } ^ { K } f ( x _ { i } , v _ { k } ) } } + ( 1 - \\alpha ) \\sum _ { j = 1 } ^ { K } \\frac { 1 } { K } \\log f ( x _ { i } , v _ { j } ) \\right]\n$$", + "text_format": "latex", + "bbox": [ + 178, + 776, + 792, + 827 + ], + "page_idx": 17 + }, + { + "type": "text", + "text": "Table 8: Negative log-likelihood estimation of a hybrid model on the synthetic data. IS- $S$ refers to Importance Sampling log-likelihood estimation with $S$ samples for each object (18). MC- $S$ refers to Monte-Carlo log-likelihood estimation with $S$ samples for each object (17). ", + "bbox": [ + 147, + 117, + 851, + 160 + ], + "page_idx": 18 + }, + { + "type": "table", + "img_path": "images/65cc93c875903136313a61e9ac8afdb60ebc425608483cce53104e3064921176.jpg", + "table_caption": [], + "table_footnote": [], + "table_body": "
VAEAC weightIS-10MC-10
a=10.2285
α = 0.990.3511
α = 0.90.621.7
", + "bbox": [ + 375, + 170, + 622, + 229 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "It is easy to show that (20) is equivalent to the following optimization problem: ", + "bbox": [ + 145, + 252, + 665, + 268 + ], + "page_idx": 18 + }, + { + "type": "equation", + "img_path": "images/fba50a965f88da607ef696227097c6d0af74dd0f5711b3ed679ed890002f6b06.jpg", + "text": "$$\n\\operatorname* { m a x } _ { v _ { 1 } , v _ { 2 } , \\ldots , v _ { K } } \\sum _ { i = 1 } ^ { K } \\left[ \\alpha \\log \\frac { \\sum _ { j = 1 } ^ { K } f ( x _ { i } , v _ { j } ) } { K } + ( 1 - \\alpha ) \\sum _ { j = 1 } ^ { K } \\frac { 1 } { K } \\log f ( x _ { i } , v _ { j } ) \\right]\n$$", + "text_format": "latex", + "bbox": [ + 264, + 272, + 735, + 323 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "It is clear from (21) that when $\\alpha = 1$ the log-likelihood of the initial model is optimized. On the other hand, when influe $\\alpha = 0$ the optimal point is generative process, a $v _ { 1 } = v _ { 2 } = \\cdots = v _ { K } = \\operatorname { a r g m a x } _ { v } \\sum _ { i = 1 } ^ { K } \\log f ( x _ { i } , v ) .$ , i. e. mizes $z$ doesn’telihood $z$ $v$ \nestimation of the generative model $f ( x , v )$ for the given dataset of $x$ ’s. For Bernoulli and Gaussian generative distributions $f$ such $v$ is just average of all modes $x _ { 1 } , x _ { 2 } , \\dotsc , x _ { K }$ . That explains why further we observe blurry images when using GSNN model. ", + "bbox": [ + 147, + 333, + 851, + 420 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "The same conclusion holds for for continuous latent variables instead of categorical. Given $K$ different modes in true data distribution, VAE uses proposal network to separate prior distribution into $K$ components (i. e. regions in the latent space), so that each region corresponds to one mode. On the other hand, in GSNN $z$ is sampled independently on the mode which is to be reconstructed from it, so for each $z$ the generator have to produce parameters suitable for all modes. ", + "bbox": [ + 148, + 426, + 851, + 497 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "From this point of view, there is no difference between VAE and VAEAC. If the true conditional distribution has several different modes, then VAEAC can fit them all, while GSNN learns their average. If true conditional distribution has one mode, GSNN and VAEAC are equal, and GSNN may even learn faster because it has less parameters. ", + "bbox": [ + 148, + 503, + 849, + 560 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "Hybrid model is a trade-off between VAEAC and GSNN: the closer $\\alpha$ to zero, the more blurry and closer to the average is the distribution of the model. The exact dependence of the model distribution on $\\alpha$ can be derived analytically for the simple data distributions or evaluated experimentally. We perform such experimental evaluation in the next sections. ", + "bbox": [ + 148, + 565, + 849, + 622 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "C.2 SYNTHETIC DATA ", + "text_level": 1, + "bbox": [ + 148, + 638, + 315, + 654 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "In this section we show that VAEAC is capable of learning a complex multimodal distribution of synthetic \ndata while GSNN and hybrid model are not. Let $x \\in \\mathbb { R } ^ { \\bar { 2 } }$ and $p ( \\bar { b } _ { 1 } = 1 ) = p ( b _ { 2 } = 1 ) = 0 . 5$ . $p _ { d } ( x ) =$ \n$\\begin{array} { r l } { \\frac { 1 } { 8 } \\sum _ { i = 1 } ^ { 8 } \\mathcal { N } ( x | \\mu _ { i } , \\frac { 1 } { 1 0 } I ) } & { { } } \\end{array}$ where s samp $\\mu _ { i } \\sim \\mathcal N ( \\mu _ { i } | 0 , I )$ . The distribution e use multi-layer p $p ( x )$ is plotted in figure 6. The datasetptron with four ReLU layers of size $p _ { d } ( x )$ \n400-200-100-50, 25-dimensional Gaussian latent variables. ", + "bbox": [ + 148, + 665, + 851, + 738 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "For different mixture coefficients $\\alpha$ we visualize samples from the learned distributions $p _ { \\psi , \\theta } ( x _ { 1 } , x _ { 2 } )$ , $p _ { \\psi , \\theta } ( x _ { 1 } | x _ { 2 } )$ , and $p _ { \\psi , \\theta } ( x _ { 2 } | x _ { 1 } )$ . The observed features for the conditional distributions are generated from the marginal distributions $p ( x _ { 2 } )$ and $p ( x _ { 1 } )$ respectively. ", + "bbox": [ + 148, + 744, + 851, + 787 + ], + "page_idx": 18 + }, + { + "type": "text", + "text": "We see in table 8 and in figure 7, that even with very small weight GSNN prevents model from learning distributions with several local optimas. GSNN also increases Monte-Carlo log-likelihood estimation with a few samples and decreases much more precise Importance Sampling log-likelihood estimation. When $\\alpha = 0 . 9$ the whole distribution structure is lost. ", + "bbox": [ + 147, + 794, + 849, + 823 + ], + "page_idx": 18 + }, + { + "type": "image", + "img_path": "images/7407e71d027399c997ec577ddb914586b76471425f58bff6404a5cd035d567fc.jpg", + "image_caption": [ + "Figure 6: Probability density function of synthetic data distribution. " + ], + "image_footnote": [], + "bbox": [ + 217, + 185, + 348, + 267 + ], + "page_idx": 19 + }, + { + "type": "image", + "img_path": "images/a4bd4e815fbae18fb3e5afb3429dcbb43b07bb985c3a0166b18e1666e16f7080.jpg", + "image_caption": [ + "Figure 7: VAEAC for synthetic data. " + ], + "image_footnote": [], + "bbox": [ + 539, + 119, + 838, + 347 + ], + "page_idx": 19 + }, + { + "type": "image", + "img_path": "images/54463305da3093916463d301746c0407e7d8d2daa9a2e48291c407dc4d828109.jpg", + "image_caption": [ + "Figure 8: MNIST inpaintings. Left: input. The gray pixels are unobserved. Middle: samples from the model. Right: ground truth. " + ], + "image_footnote": [], + "bbox": [ + 173, + 429, + 834, + 659 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "", + "bbox": [ + 148, + 758, + 849, + 787 + ], + "page_idx": 19 + }, + { + "type": "text", + "text": "We see that using $\\alpha \\neq 1$ ruins multimodality of the restored distribution, so we highly recommend to use $\\alpha = 1$ or at least $\\alpha \\approx 1$ . ", + "bbox": [ + 147, + 794, + 846, + 823 + ], + "page_idx": 19 + }, + { + "type": "table", + "img_path": "images/a38ab45727edbf8bab4aa92d10957fbb17674c382278cea78615c2779533612a.jpg", + "table_caption": [ + "Table 9: Average negative log-likelihood of inpaintings for 1000 objects. IS- $S$ refers to Importance Sampling log-likelihood estimation with $S$ samples for each object (18). MC- $S$ refers to Monte-Carlo log-likelihood estimation with $S$ samples for each object (17). Naive Bayes is a baseline method which assumes pixels and colors independence. " + ], + "table_footnote": [], + "table_body": "
MethodMNISTOmniglotCelebA
VAEAC IS-10261±1275±1734035 ± 1609
VAEAC MC-10494±41452 ± 10941513 ± 2163
VAEAC MC-102156 ±12203 ± 15053904 ± 3121
GSNN MC-104141 ±71199 ± 6253427 ± 2208
GSNN MC-10²141 ±11200 ± 6253486 ± 2210
Naive Bayes2052490269480
", + "bbox": [ + 289, + 184, + 709, + 287 + ], + "page_idx": 20 + }, + { + "type": "image", + "img_path": "images/3126337f9f6dd982a9196015b6247be9dab5083bb45c2e9eb0945400c4c56f31.jpg", + "image_caption": [ + "Figure 9: Convergence of VAE and VAEAC on MNIST dataset. " + ], + "image_footnote": [], + "bbox": [ + 339, + 329, + 632, + 479 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "C.3 COMPARISON ON THE IMAGE INPAINTING PROBLEM ", + "bbox": [ + 147, + 563, + 553, + 578 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "In figure 8 we can see that the inpaintings produced by GSNN are smooth, blurry and not diverse compared with VAEAC. ", + "bbox": [ + 147, + 594, + 849, + 623 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "Table 9 shows that VAEAC learns distribution over inpaintings better than GSNN in terms of test loglikelihood. Nevertheless, Monte-Carlo estimations with a small number of samples sometimes are better for GSNN, which means less local modes in the learned distribution and more blurriness in the samples. ", + "bbox": [ + 148, + 630, + 851, + 672 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "D ADDITIONAL EXPERIMENTS ", + "text_level": 1, + "bbox": [ + 148, + 708, + 418, + 724 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "D.1 CONVERGENCE SPEED ", + "text_level": 1, + "bbox": [ + 148, + 747, + 349, + 762 + ], + "page_idx": 20 + }, + { + "type": "text", + "text": "In figure 9 one can see that VAEAC has similar convergence speed to VAE in terms of iterations on MNIST dataset. In our experiments we observed the same behaviour for other datasets. Each iteration of VAEAC is about 1.5 times slower than VAE due to usage of three networks instead of two. ", + "bbox": [ + 148, + 780, + 851, + 821 + ], + "page_idx": 20 + }, + { + "type": "table", + "img_path": "images/a28b110ac71efa360491f823a261c429125aef3652390412ba5272709177d2d9.jpg", + "table_caption": [ + "Table 10: NRMSE (for continuous datasets) or PFC (for categorical ones) of imputations. Less is better. " + ], + "table_footnote": [], + "table_body": "
DatasetMICEMissForestGAINVAEACGSNNNN
Boston0.69 ± 0.020.58±0.020.78 ± 0.030.71 ± 0.020.70 ± 0.010.69 ± 0.01
Breast0.58 ±0.020.515 ± 0.0080.67 ±0.010.55±0.020.55 ± 0.020.52 ±0.02
Concrete0.850 ± 0.0070.78 ± 0.010.96 ±0.010.84 ±0.020.85 ± 0.012±3
Diabetes0.80 ±0.010.84±0.020.911 ± 0.0090.90 ±0.030.91 ± 0.030.90 ±0.02
Digits0.69 ±0.020.61± 0.020.79±0.020.69 ±0.020.69 ± 0.020.67 ±0.02
Glass0.91 ±0.020.83± 0.041.06 ±0.050.91 ±0.040.91 ± 0.050.87 ± 0.04
Iris0.59 ± 0.020.62 ± 0.040.72 ± 0.040.64± 0.040.62 ± 0.040.61± 0.02
Mushroom0.334 ± 0.0020.249 ±0.0060.271 ±0.0030.241 ± 0.0020.2412 ± 0.00090.239 ± 0.001
Orthopedic0.76 ±0.020.79±0.030.91±0.030.80±0.030.81 ±0.030.81 ±0.02
Phishing0.422 ± 0.0060.422 ±0.0090.427 ± 0.0100.397 ± 0.0100.392 ±0.0090.41 ± 0.01
WallRobot0.885 ± 0.0030.640 ± 0.0030.907 ± 0.0050.78 ±0.010.776 ±0.0070.757± 0.005
WhiteWine0.964 ± 0.0070.878 ±0.0090.97 ± 0.020.850 ± 0.0050.848 ± 0.0070.85 ± 0.01
Yeast0.98±0.021.00 ± 0.020.99 ±0.030.95 ± 0.010.958 ± 0.0070.97 ± 0.03
Zoo0.19 ± 0.030.16 ±0.020.20 ±0.020.16±0.020.17 ±0.020.16 ±0.01
", + "bbox": [ + 155, + 141, + 848, + 343 + ], + "page_idx": 21 + }, + { + "type": "table", + "img_path": "images/7d3a3cbee5bc4c33e77c2d6020bcaa263208c0156dec7db727d7d8e3202cc1d5.jpg", + "table_caption": [ + "Table 11: R2-score (for continuous targets) or accuracy (for categorical ones) of post-imputation regression or classification. Higher is better. " + ], + "table_footnote": [], + "table_body": "
DatasetMICEMissForestGAINVAEACGSNNNN
Boston0.57 ± 0.080.6 ± 0.10.50 ± 0.100.5 ± 0.10.5± 0.10.50±0.09
Breast0.96 ± 0.020.95 ± 0.020.94± 0.010.95 ± 0.020.96 ± 0.020.95 ± 0.02
Concrete0.35 ± 0.050.33 ± 0.040.28±0.060.30 ± 0.080.32 ± 0.050±1
Diabetes0.37 ± 0.060.34±0.060.34±0.030.34± 0.040.33 ±0.040.27±0.06
Digits0.86±0.020.887 ±0.0080.83 ±0.030.892 ± 0.0100.895 ± 0.0100.912 ± 0.010
Glass0.44 ±0.080.53 ±0.050.37±0.050.49 ±0.090.47 ±0.090.48 ±0.09
Iris0.81± 0.020.84±0.020.66 ±0.060.84±0.050.82±0.060.73±0.09
Mushroom0.92 ±0.010.972 ± 0.0030.969 ± 0.0050.987 ± 0.0010.986 ± 0.0020.989 ± 0.003
Orthopedic0.71 ± 0.020.72 ± 0.030.60±0.030.71 ±0.020.70± 0.040.61±0.04
Phishing0.75 ± 0.020.73±0.030.74± 0.030.75 ± 0.010.74±0.040.73±0.02
WallRobot0.55 ±0.010.697 ± 0.0050.56 ±0.010.62±0.020.62 ± 0.010.64±0.02
WhiteWine0.13 ± 0.020.17 ± 0.010.11± 0.010.18 ±0.020.17 ± 0.010.15 ± 0.03
Yeast0.42 ±0.020.41 ±0.020.39 ±0.060.42 ± 0.010.425 ± 0.0100.33±0.03
Zoo0.78 ± 0.060.71 ±0.080.67±0.060.77 ± 0.090.8±0.10.83 ± 0.08
", + "bbox": [ + 151, + 396, + 848, + 601 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "D.2 MISSING FEATURES IMPUTATION ", + "text_level": 1, + "bbox": [ + 148, + 627, + 423, + 642 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "We evaluate the quality of imputations on different datasets (mostly from UCI (Lichman, 2013)). The evaluation is performed for VAEAC, GSNN (15) and NN (neural network; can be considered as a special case of GSNN where $p _ { \\theta } ( z | x _ { 1 - b } , b )$ is delta-function; produces single imputation). We compare these methods with MICE (Buuren & Groothuis-Oudshoorn, 2010), MissForest (Stekhoven & Buhlmann, 2011), and GAIN ¨ (Yoon et al., 2018). ", + "bbox": [ + 148, + 654, + 851, + 724 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "We see that for some datasets MICE and MissForest outperform VAEAC, GSNN and NN. The reason is that for some datasets random forest is more natural structure than neural network. ", + "bbox": [ + 145, + 731, + 851, + 760 + ], + "page_idx": 21 + }, + { + "type": "text", + "text": "The results also show that VAEAC, GSNN and NN show similar imputation performance in terms of NRMSE, PFC, post-imputation R2-score and accuracy. Given the result from appendix C we can take this as a weak evidence that the distribution of imputations has only one local maximum for datasets from (Lichman, 2013). ", + "bbox": [ + 148, + 766, + 851, + 823 + ], + "page_idx": 21 + }, + { + "type": "image", + "img_path": "images/21041c7b3c7fc1baa4707c17e3944ec8fc55aadb337ff273f13ec39db52954c7.jpg", + "image_caption": [ + "Figure 10: CelebA inpaintings with masks from (Li et al., 2017). ft: input. The gray pixels are unobserved. Middle: samples from VAEAC. Right: ground truth. " + ], + "image_footnote": [], + "bbox": [ + 245, + 119, + 759, + 458 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "D.3 FACE INPAINTINGS ", + "text_level": 1, + "bbox": [ + 148, + 537, + 323, + 553 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "In figure 10 we provide samples of VAEAC on the CelebA dataset for the masks from (Li et al., 2017). ", + "bbox": [ + 150, + 564, + 816, + 579 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "D.4 GAIN FOR IMAGE INPAINTING ", + "text_level": 1, + "bbox": [ + 148, + 598, + 406, + 613 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "GAIN (Yoon et al., 2018) doesnt use unobserved data during training, which makes it easier to apply to the missing features imputation problem. Nevertheless, it turns into a disadvantage when the fully-observed training data is available but the missingness rate at the testing stage is high. ", + "bbox": [ + 148, + 626, + 849, + 667 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "We consider the horizontal line mask for MNIST which is described in appendix A.3. We use the released GAIN code 5 with a different mask generator. The inpaintings from VAEAC which uses the unobserved pixels during training are available in figure 1. The inpaintings from GAIN which ignores unobserved pixels are provided in figure 11. As can be seen in figure 11, GAIN fails to learn conditional distribution for given mask distribution ${ \\dot { p } } ( b )$ . ", + "bbox": [ + 148, + 674, + 849, + 744 + ], + "page_idx": 22 + }, + { + "type": "text", + "text": "Nevertheless, we don’t claim that GAIN is not suitable for image inpainting. As it was shown in the supplementary of (Yoon et al., 2018) and in the corresponding code, GAIN is able to learn conditional distributions when $p ( b )$ is pixel-wise independent Bernoulli distribution with probability 0.5. ", + "bbox": [ + 148, + 751, + 849, + 794 + ], + "page_idx": 22 + }, + { + "type": "image", + "img_path": "images/77978a748fdfece774bbdba6a454e4caa9d264c220caf33b0d01dfee4003f59b.jpg", + "image_caption": [ + "Figure 11: MNIST inpaintings from GAIN. " + ], + "image_footnote": [], + "bbox": [ + 348, + 122, + 650, + 290 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Left: input. The gray pixels are unobserved. Middle: samples from the model. Right: ground truth. ", + "bbox": [ + 166, + 320, + 823, + 335 + ], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/3addba9b5d9e7527dc59c64b67c88b55ee2d758e97d2f9f04dcf482aada3e5b6.jpg", + "image_caption": [ + "Figure 12: MNIST inpaintings. Left: input. The gray pixels are unobserved. Middle: samples from the model. Right: ground truth. " + ], + "image_footnote": [], + "bbox": [ + 174, + 353, + 834, + 580 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "D.5 UNIVERSAL MARGINALIZER: ILLUSTRATIONS ", + "text_level": 1, + "bbox": [ + 147, + 654, + 511, + 669 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "In figure 12 we provide samples of Universal Marginalizer (UM) and VAEAC for the same inputs. ", + "bbox": [ + 148, + 679, + 792, + 695 + ], + "page_idx": 23 + }, + { + "type": "text", + "text": "Consider the case when UM marginal distributions are parametrized with Gaussians. The most simple example of a distribution, which UM cannot learn but VAEAC can, is given in figure 13. ", + "bbox": [ + 143, + 700, + 852, + 731 + ], + "page_idx": 23 + }, + { + "type": "image", + "img_path": "images/34b2efedea65ec8dd1cd6662835f9b38b5a648676328c4f707b65cec76fc2de2.jpg", + "image_caption": [ + "Figure 13: Distribution learning: VAEAC vs UM. " + ], + "image_footnote": [], + "bbox": [ + 158, + 375, + 839, + 530 + ], + "page_idx": 24 + } +] \ No newline at end of file diff --git a/parse/train/SyxtJh0qYm/SyxtJh0qYm_middle.json b/parse/train/SyxtJh0qYm/SyxtJh0qYm_middle.json new file mode 100644 index 0000000000000000000000000000000000000000..678da957d3fcc10d21511ab39dba5c7bc86ef903 --- /dev/null +++ b/parse/train/SyxtJh0qYm/SyxtJh0qYm_middle.json @@ -0,0 +1,57926 @@ +{ + "pdf_info": [ + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 91, + 93, + 342, + 129 + ], + "lines": [ + { + "bbox": [ + 90, + 91, + 313, + 110 + ], + "spans": [ + { + "bbox": [ + 90, + 91, + 313, + 110 + ], + "score": 1.0, + "content": "VARIATIONAL AUTOENCODER", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 90, + 111, + 343, + 131 + ], + "spans": [ + { + "bbox": [ + 90, + 111, + 343, + 131 + ], + "score": 1.0, + "content": "WITH ARBITRARY CONDITIONING", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "text", + "bbox": [ + 96, + 148, + 213, + 193 + ], + "lines": [ + { + "bbox": [ + 95, + 148, + 150, + 160 + ], + "spans": [ + { + "bbox": [ + 95, + 148, + 150, + 160 + ], + "score": 1.0, + "content": "Oleg Ivanov", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 96, + 160, + 213, + 171 + ], + "spans": [ + { + "bbox": [ + 96, + 160, + 213, + 171 + ], + "score": 1.0, + "content": "Samsung AI Center Moscow", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 95, + 170, + 164, + 182 + ], + "spans": [ + { + "bbox": [ + 95, + 170, + 164, + 182 + ], + "score": 1.0, + "content": "Moscow, Russia", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 96, + 182, + 207, + 194 + ], + "spans": [ + { + "bbox": [ + 96, + 182, + 207, + 194 + ], + "score": 1.0, + "content": "tigvarts@gmail.com", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 237, + 148, + 361, + 204 + ], + "lines": [ + { + "bbox": [ + 236, + 147, + 316, + 162 + ], + "spans": [ + { + "bbox": [ + 236, + 147, + 316, + 162 + ], + "score": 1.0, + "content": "Michael Figurnov", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 236, + 158, + 358, + 172 + ], + "spans": [ + { + "bbox": [ + 236, + 158, + 358, + 172 + ], + "score": 1.0, + "content": "National Research University", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 236, + 170, + 362, + 181 + ], + "spans": [ + { + "bbox": [ + 236, + 170, + 362, + 181 + ], + "score": 1.0, + "content": "Higher School of Economics ∗", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 236, + 181, + 305, + 193 + ], + "spans": [ + { + "bbox": [ + 236, + 181, + 305, + 193 + ], + "score": 1.0, + "content": "Moscow, Russia", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 236, + 192, + 353, + 204 + ], + "spans": [ + { + "bbox": [ + 236, + 192, + 353, + 204 + ], + "score": 1.0, + "content": "michael@figurnov.ru", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 381, + 149, + 500, + 225 + ], + "lines": [ + { + "bbox": [ + 381, + 148, + 445, + 161 + ], + "spans": [ + { + "bbox": [ + 381, + 148, + 445, + 161 + ], + "score": 1.0, + "content": "Dmitry Vetrov", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 381, + 158, + 491, + 172 + ], + "spans": [ + { + "bbox": [ + 381, + 158, + 491, + 172 + ], + "score": 1.0, + "content": "Samsung-HSE Laboratory,", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 380, + 169, + 501, + 183 + ], + "spans": [ + { + "bbox": [ + 380, + 169, + 501, + 183 + ], + "score": 1.0, + "content": "National Research University", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 380, + 181, + 499, + 193 + ], + "spans": [ + { + "bbox": [ + 380, + 181, + 499, + 193 + ], + "score": 1.0, + "content": "Higher School of Economics", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 381, + 192, + 498, + 204 + ], + "spans": [ + { + "bbox": [ + 381, + 192, + 498, + 204 + ], + "score": 1.0, + "content": "Samsung AI Center Moscow", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 380, + 203, + 449, + 215 + ], + "spans": [ + { + "bbox": [ + 380, + 203, + 449, + 215 + ], + "score": 1.0, + "content": "Moscow, Russia", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 381, + 214, + 486, + 227 + ], + "spans": [ + { + "bbox": [ + 381, + 214, + 486, + 227 + ], + "score": 1.0, + "content": "vetrovd@yandex.ru", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 14 + }, + { + "type": "title", + "bbox": [ + 277, + 254, + 333, + 266 + ], + "lines": [ + { + "bbox": [ + 276, + 254, + 335, + 267 + ], + "spans": [ + { + "bbox": [ + 276, + 254, + 335, + 267 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 127, + 277, + 485, + 344 + ], + "lines": [ + { + "bbox": [ + 126, + 278, + 486, + 290 + ], + "spans": [ + { + "bbox": [ + 126, + 278, + 486, + 290 + ], + "score": 1.0, + "content": "We propose a single neural probabilistic model based on variational autoencoder that can", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 125, + 288, + 486, + 302 + ], + "spans": [ + { + "bbox": [ + 125, + 288, + 486, + 302 + ], + "score": 1.0, + "content": "be conditioned on an arbitrary subset of observed features and then sample the remaining", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 125, + 298, + 486, + 313 + ], + "spans": [ + { + "bbox": [ + 125, + 298, + 486, + 313 + ], + "score": 1.0, + "content": "features in “one shot”. The features may be both real-valued and categorical. Training", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 125, + 310, + 486, + 324 + ], + "spans": [ + { + "bbox": [ + 125, + 310, + 486, + 324 + ], + "score": 1.0, + "content": "of the model is performed by stochastic variational Bayes. The experimental evaluation", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 125, + 322, + 486, + 334 + ], + "spans": [ + { + "bbox": [ + 125, + 322, + 486, + 334 + ], + "score": 1.0, + "content": "on synthetic data, as well as feature imputation and image inpainting problems, shows the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 125, + 332, + 441, + 345 + ], + "spans": [ + { + "bbox": [ + 125, + 332, + 441, + 345 + ], + "score": 1.0, + "content": "effectiveness of the proposed approach and diversity of the generated samples.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21.5 + }, + { + "type": "title", + "bbox": [ + 92, + 362, + 190, + 375 + ], + "lines": [ + { + "bbox": [ + 89, + 361, + 192, + 378 + ], + "spans": [ + { + "bbox": [ + 89, + 361, + 192, + 378 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 91, + 387, + 520, + 431 + ], + "lines": [ + { + "bbox": [ + 90, + 386, + 521, + 399 + ], + "spans": [ + { + "bbox": [ + 90, + 386, + 521, + 399 + ], + "score": 1.0, + "content": "In past years, a number of generative probabilistic models based on neural networks have been proposed. The", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 90, + 397, + 521, + 410 + ], + "spans": [ + { + "bbox": [ + 90, + 397, + 521, + 410 + ], + "score": 1.0, + "content": "most popular approaches include variational autoencoder (Kingma & Welling, 2013) (VAE) and generative", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 90, + 409, + 521, + 421 + ], + "spans": [ + { + "bbox": [ + 90, + 409, + 457, + 421 + ], + "score": 1.0, + "content": "adversarial net (Goodfellow et al., 2014) (GANs). They learn a distribution over objects", + "type": "text" + }, + { + "bbox": [ + 457, + 409, + 477, + 421 + ], + "score": 0.92, + "content": "p ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 409, + 521, + 421 + ], + "score": 1.0, + "content": "and allow", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 89, + 419, + 219, + 433 + ], + "spans": [ + { + "bbox": [ + 89, + 419, + 219, + 433 + ], + "score": 1.0, + "content": "sampling from this distribution.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 91, + 436, + 520, + 480 + ], + "lines": [ + { + "bbox": [ + 89, + 436, + 521, + 449 + ], + "spans": [ + { + "bbox": [ + 89, + 436, + 370, + 449 + ], + "score": 1.0, + "content": "In many cases, we are interested in learning a conditional distribution", + "type": "text" + }, + { + "bbox": [ + 370, + 437, + 397, + 449 + ], + "score": 0.92, + "content": "p ( x | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 436, + 465, + 449 + ], + "score": 1.0, + "content": ". For instance, if", + "type": "text" + }, + { + "bbox": [ + 465, + 439, + 472, + 447 + ], + "score": 0.77, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 436, + 521, + 449 + ], + "score": 1.0, + "content": "is an image", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 89, + 447, + 522, + 460 + ], + "spans": [ + { + "bbox": [ + 89, + 447, + 131, + 460 + ], + "score": 1.0, + "content": "of a face,", + "type": "text" + }, + { + "bbox": [ + 131, + 450, + 138, + 459 + ], + "score": 0.76, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 447, + 522, + 460 + ], + "score": 1.0, + "content": "could be the characteristics describing the face (are glasses present or not; length of hair, etc.)", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 89, + 458, + 522, + 471 + ], + "spans": [ + { + "bbox": [ + 89, + 458, + 522, + 471 + ], + "score": 1.0, + "content": "Conditional variational autoencoder (Sohn et al., 2015) and conditional generative adversarial nets (Mirza", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 89, + 469, + 321, + 482 + ], + "spans": [ + { + "bbox": [ + 89, + 469, + 321, + 482 + ], + "score": 1.0, + "content": "& Osindero, 2014) are popular methods for this problem.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 91, + 486, + 521, + 564 + ], + "lines": [ + { + "bbox": [ + 88, + 484, + 522, + 501 + ], + "spans": [ + { + "bbox": [ + 88, + 484, + 470, + 501 + ], + "score": 1.0, + "content": "In this paper, we consider the problem of learning all conditional distributions of the form", + "type": "text" + }, + { + "bbox": [ + 470, + 486, + 517, + 499 + ], + "score": 0.91, + "content": "p ( x _ { I } | x _ { U \\setminus I } )", + "type": "inline_equation" + }, + { + "bbox": [ + 518, + 484, + 522, + 501 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 89, + 496, + 522, + 510 + ], + "spans": [ + { + "bbox": [ + 89, + 496, + 117, + 510 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 118, + 498, + 127, + 507 + ], + "score": 0.79, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 496, + 244, + 510 + ], + "score": 1.0, + "content": "is the set of all features and", + "type": "text" + }, + { + "bbox": [ + 245, + 498, + 251, + 507 + ], + "score": 0.68, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 496, + 522, + 510 + ], + "score": 1.0, + "content": "is its arbitrary subset. This problem generalizes both learning the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 89, + 507, + 522, + 522 + ], + "spans": [ + { + "bbox": [ + 89, + 507, + 159, + 522 + ], + "score": 1.0, + "content": "joint distribution", + "type": "text" + }, + { + "bbox": [ + 159, + 508, + 178, + 520 + ], + "score": 0.91, + "content": "p ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 507, + 341, + 522 + ], + "score": 1.0, + "content": "and learning the conditional distribution", + "type": "text" + }, + { + "bbox": [ + 341, + 509, + 369, + 520 + ], + "score": 0.94, + "content": "p ( x | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 507, + 522, + 522 + ], + "score": 1.0, + "content": ". To tackle this problem, we propose a", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 89, + 518, + 522, + 532 + ], + "spans": [ + { + "bbox": [ + 89, + 518, + 522, + 532 + ], + "score": 1.0, + "content": "Variational Autoencoder with Arbitrary Conditioning (VAEAC) model. It is a latent variable model similar", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 88, + 529, + 522, + 543 + ], + "spans": [ + { + "bbox": [ + 88, + 529, + 522, + 543 + ], + "score": 1.0, + "content": "to VAE, but allows conditioning on an arbitrary subset of the features. The conditioning features affect the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 88, + 541, + 522, + 554 + ], + "spans": [ + { + "bbox": [ + 88, + 541, + 522, + 554 + ], + "score": 1.0, + "content": "prior on the latent Gaussian variables which are used to generate unobserved features. The model is trained", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 89, + 551, + 376, + 565 + ], + "spans": [ + { + "bbox": [ + 89, + 551, + 376, + 565 + ], + "score": 1.0, + "content": "using stochastic gradient variational Bayes (Kingma & Welling, 2013).", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 37 + }, + { + "type": "text", + "bbox": [ + 91, + 569, + 521, + 635 + ], + "lines": [ + { + "bbox": [ + 89, + 568, + 522, + 582 + ], + "spans": [ + { + "bbox": [ + 89, + 568, + 522, + 582 + ], + "score": 1.0, + "content": "We consider two most natural applications of the proposed model. The first one is feature imputation where", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 89, + 579, + 522, + 592 + ], + "spans": [ + { + "bbox": [ + 89, + 579, + 522, + 592 + ], + "score": 1.0, + "content": "the goal is to restore the missing features given the observed ones. The imputed values may be valuable by", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 90, + 592, + 521, + 603 + ], + "spans": [ + { + "bbox": [ + 90, + 592, + 521, + 603 + ], + "score": 1.0, + "content": "themselves or may improve the performance of other machine learning algorithms which process the dataset.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 89, + 602, + 522, + 615 + ], + "spans": [ + { + "bbox": [ + 89, + 602, + 522, + 615 + ], + "score": 1.0, + "content": "Another application is image inpainting in which the goal is to fill in an unobserved part of an image with", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 89, + 612, + 521, + 625 + ], + "spans": [ + { + "bbox": [ + 89, + 612, + 521, + 625 + ], + "score": 1.0, + "content": "an artificial content in a realistic way. This can be used for removing unnecessary objects from the images", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 89, + 624, + 386, + 637 + ], + "spans": [ + { + "bbox": [ + 89, + 624, + 386, + 637 + ], + "score": 1.0, + "content": "or, vice versa, for complementing the partially closed or corrupted object.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 43.5 + } + ], + "page_idx": 0, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 103, + 641, + 209, + 651 + ], + "lines": [ + { + "bbox": [ + 102, + 639, + 211, + 653 + ], + "spans": [ + { + "bbox": [ + 102, + 639, + 211, + 653 + ], + "score": 1.0, + "content": "∗Author is in DeepMind now.", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 91, + 40, + 277, + 50 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 286, + 671, + 292, + 680 + ], + "lines": [ + { + "bbox": [ + 286, + 671, + 293, + 682 + ], + "spans": [ + { + "bbox": [ + 286, + 671, + 293, + 682 + ], + "score": 1.0, + "content": "1", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 91, + 93, + 342, + 129 + ], + "lines": [ + { + "bbox": [ + 90, + 91, + 313, + 110 + ], + "spans": [ + { + "bbox": [ + 90, + 91, + 313, + 110 + ], + "score": 1.0, + "content": "VARIATIONAL AUTOENCODER", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 90, + 111, + 343, + 131 + ], + "spans": [ + { + "bbox": [ + 90, + 111, + 343, + 131 + ], + "score": 1.0, + "content": "WITH ARBITRARY CONDITIONING", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "list", + "bbox": [ + 96, + 148, + 213, + 193 + ], + "lines": [ + { + "bbox": [ + 95, + 148, + 150, + 160 + ], + "spans": [ + { + "bbox": [ + 95, + 148, + 150, + 160 + ], + "score": 1.0, + "content": "Oleg Ivanov", + "type": "text" + } + ], + "index": 2, + "is_list_end_line": true + }, + { + "bbox": [ + 96, + 160, + 213, + 171 + ], + "spans": [ + { + "bbox": [ + 96, + 160, + 213, + 171 + ], + "score": 1.0, + "content": "Samsung AI Center Moscow", + "type": "text" + } + ], + "index": 3, + "is_list_start_line": true + }, + { + "bbox": [ + 95, + 170, + 164, + 182 + ], + "spans": [ + { + "bbox": [ + 95, + 170, + 164, + 182 + ], + "score": 1.0, + "content": "Moscow, Russia", + "type": "text" + } + ], + "index": 4, + "is_list_end_line": true + }, + { + "bbox": [ + 96, + 182, + 207, + 194 + ], + "spans": [ + { + "bbox": [ + 96, + 182, + 207, + 194 + ], + "score": 1.0, + "content": "tigvarts@gmail.com", + "type": "text" + } + ], + "index": 5, + "is_list_start_line": true + }, + { + "bbox": [ + 236, + 147, + 316, + 162 + ], + "spans": [ + { + "bbox": [ + 236, + 147, + 316, + 162 + ], + "score": 1.0, + "content": "Michael Figurnov", + "type": "text" + } + ], + "index": 6, + "is_list_end_line": true + }, + { + "bbox": [ + 236, + 158, + 358, + 172 + ], + "spans": [ + { + "bbox": [ + 236, + 158, + 358, + 172 + ], + "score": 1.0, + "content": "National Research University", + "type": "text" + } + ], + "index": 7, + "is_list_start_line": true + }, + { + "bbox": [ + 236, + 170, + 362, + 181 + ], + "spans": [ + { + "bbox": [ + 236, + 170, + 362, + 181 + ], + "score": 1.0, + "content": "Higher School of Economics ∗", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 236, + 181, + 305, + 193 + ], + "spans": [ + { + "bbox": [ + 236, + 181, + 305, + 193 + ], + "score": 1.0, + "content": "Moscow, Russia", + "type": "text" + } + ], + "index": 9, + "is_list_end_line": true + }, + { + "bbox": [ + 236, + 192, + 353, + 204 + ], + "spans": [ + { + "bbox": [ + 236, + 192, + 353, + 204 + ], + "score": 1.0, + "content": "michael@figurnov.ru", + "type": "text" + } + ], + "index": 10, + "is_list_start_line": true + }, + { + "bbox": [ + 381, + 148, + 445, + 161 + ], + "spans": [ + { + "bbox": [ + 381, + 148, + 445, + 161 + ], + "score": 1.0, + "content": "Dmitry Vetrov", + "type": "text" + } + ], + "index": 11, + "is_list_end_line": true + }, + { + "bbox": [ + 381, + 158, + 491, + 172 + ], + "spans": [ + { + "bbox": [ + 381, + 158, + 491, + 172 + ], + "score": 1.0, + "content": "Samsung-HSE Laboratory,", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 380, + 169, + 501, + 183 + ], + "spans": [ + { + "bbox": [ + 380, + 169, + 501, + 183 + ], + "score": 1.0, + "content": "National Research University", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 380, + 181, + 499, + 193 + ], + "spans": [ + { + "bbox": [ + 380, + 181, + 499, + 193 + ], + "score": 1.0, + "content": "Higher School of Economics", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 381, + 192, + 498, + 204 + ], + "spans": [ + { + "bbox": [ + 381, + 192, + 498, + 204 + ], + "score": 1.0, + "content": "Samsung AI Center Moscow", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 380, + 203, + 449, + 215 + ], + "spans": [ + { + "bbox": [ + 380, + 203, + 449, + 215 + ], + "score": 1.0, + "content": "Moscow, Russia", + "type": "text" + } + ], + "index": 16, + "is_list_end_line": true + }, + { + "bbox": [ + 381, + 214, + 486, + 227 + ], + "spans": [ + { + "bbox": [ + 381, + 214, + 486, + 227 + ], + "score": 1.0, + "content": "vetrovd@yandex.ru", + "type": "text" + } + ], + "index": 17, + "is_list_start_line": true, + "is_list_end_line": true + } + ], + "index": 3.5, + "bbox_fs": [ + 95, + 148, + 213, + 194 + ] + }, + { + "type": "list", + "bbox": [ + 237, + 148, + 361, + 204 + ], + "lines": [], + "index": 8, + "bbox_fs": [ + 236, + 147, + 362, + 204 + ], + "lines_deleted": true + }, + { + "type": "list", + "bbox": [ + 381, + 149, + 500, + 225 + ], + "lines": [], + "index": 14, + "bbox_fs": [ + 380, + 148, + 501, + 227 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 277, + 254, + 333, + 266 + ], + "lines": [ + { + "bbox": [ + 276, + 254, + 335, + 267 + ], + "spans": [ + { + "bbox": [ + 276, + 254, + 335, + 267 + ], + "score": 1.0, + "content": "ABSTRACT", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 127, + 277, + 485, + 344 + ], + "lines": [ + { + "bbox": [ + 126, + 278, + 486, + 290 + ], + "spans": [ + { + "bbox": [ + 126, + 278, + 486, + 290 + ], + "score": 1.0, + "content": "We propose a single neural probabilistic model based on variational autoencoder that can", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 125, + 288, + 486, + 302 + ], + "spans": [ + { + "bbox": [ + 125, + 288, + 486, + 302 + ], + "score": 1.0, + "content": "be conditioned on an arbitrary subset of observed features and then sample the remaining", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 125, + 298, + 486, + 313 + ], + "spans": [ + { + "bbox": [ + 125, + 298, + 486, + 313 + ], + "score": 1.0, + "content": "features in “one shot”. The features may be both real-valued and categorical. Training", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 125, + 310, + 486, + 324 + ], + "spans": [ + { + "bbox": [ + 125, + 310, + 486, + 324 + ], + "score": 1.0, + "content": "of the model is performed by stochastic variational Bayes. The experimental evaluation", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 125, + 322, + 486, + 334 + ], + "spans": [ + { + "bbox": [ + 125, + 322, + 486, + 334 + ], + "score": 1.0, + "content": "on synthetic data, as well as feature imputation and image inpainting problems, shows the", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 125, + 332, + 441, + 345 + ], + "spans": [ + { + "bbox": [ + 125, + 332, + 441, + 345 + ], + "score": 1.0, + "content": "effectiveness of the proposed approach and diversity of the generated samples.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 21.5, + "bbox_fs": [ + 125, + 278, + 486, + 345 + ] + }, + { + "type": "title", + "bbox": [ + 92, + 362, + 190, + 375 + ], + "lines": [ + { + "bbox": [ + 89, + 361, + 192, + 378 + ], + "spans": [ + { + "bbox": [ + 89, + 361, + 192, + 378 + ], + "score": 1.0, + "content": "1 INTRODUCTION", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 91, + 387, + 520, + 431 + ], + "lines": [ + { + "bbox": [ + 90, + 386, + 521, + 399 + ], + "spans": [ + { + "bbox": [ + 90, + 386, + 521, + 399 + ], + "score": 1.0, + "content": "In past years, a number of generative probabilistic models based on neural networks have been proposed. The", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 90, + 397, + 521, + 410 + ], + "spans": [ + { + "bbox": [ + 90, + 397, + 521, + 410 + ], + "score": 1.0, + "content": "most popular approaches include variational autoencoder (Kingma & Welling, 2013) (VAE) and generative", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 90, + 409, + 521, + 421 + ], + "spans": [ + { + "bbox": [ + 90, + 409, + 457, + 421 + ], + "score": 1.0, + "content": "adversarial net (Goodfellow et al., 2014) (GANs). They learn a distribution over objects", + "type": "text" + }, + { + "bbox": [ + 457, + 409, + 477, + 421 + ], + "score": 0.92, + "content": "p ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 477, + 409, + 521, + 421 + ], + "score": 1.0, + "content": "and allow", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 89, + 419, + 219, + 433 + ], + "spans": [ + { + "bbox": [ + 89, + 419, + 219, + 433 + ], + "score": 1.0, + "content": "sampling from this distribution.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5, + "bbox_fs": [ + 89, + 386, + 521, + 433 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 436, + 520, + 480 + ], + "lines": [ + { + "bbox": [ + 89, + 436, + 521, + 449 + ], + "spans": [ + { + "bbox": [ + 89, + 436, + 370, + 449 + ], + "score": 1.0, + "content": "In many cases, we are interested in learning a conditional distribution", + "type": "text" + }, + { + "bbox": [ + 370, + 437, + 397, + 449 + ], + "score": 0.92, + "content": "p ( x | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 436, + 465, + 449 + ], + "score": 1.0, + "content": ". For instance, if", + "type": "text" + }, + { + "bbox": [ + 465, + 439, + 472, + 447 + ], + "score": 0.77, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 436, + 521, + 449 + ], + "score": 1.0, + "content": "is an image", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 89, + 447, + 522, + 460 + ], + "spans": [ + { + "bbox": [ + 89, + 447, + 131, + 460 + ], + "score": 1.0, + "content": "of a face,", + "type": "text" + }, + { + "bbox": [ + 131, + 450, + 138, + 459 + ], + "score": 0.76, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 138, + 447, + 522, + 460 + ], + "score": 1.0, + "content": "could be the characteristics describing the face (are glasses present or not; length of hair, etc.)", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 89, + 458, + 522, + 471 + ], + "spans": [ + { + "bbox": [ + 89, + 458, + 522, + 471 + ], + "score": 1.0, + "content": "Conditional variational autoencoder (Sohn et al., 2015) and conditional generative adversarial nets (Mirza", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 89, + 469, + 321, + 482 + ], + "spans": [ + { + "bbox": [ + 89, + 469, + 321, + 482 + ], + "score": 1.0, + "content": "& Osindero, 2014) are popular methods for this problem.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31.5, + "bbox_fs": [ + 89, + 436, + 522, + 482 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 486, + 521, + 564 + ], + "lines": [ + { + "bbox": [ + 88, + 484, + 522, + 501 + ], + "spans": [ + { + "bbox": [ + 88, + 484, + 470, + 501 + ], + "score": 1.0, + "content": "In this paper, we consider the problem of learning all conditional distributions of the form", + "type": "text" + }, + { + "bbox": [ + 470, + 486, + 517, + 499 + ], + "score": 0.91, + "content": "p ( x _ { I } | x _ { U \\setminus I } )", + "type": "inline_equation" + }, + { + "bbox": [ + 518, + 484, + 522, + 501 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 89, + 496, + 522, + 510 + ], + "spans": [ + { + "bbox": [ + 89, + 496, + 117, + 510 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 118, + 498, + 127, + 507 + ], + "score": 0.79, + "content": "U", + "type": "inline_equation" + }, + { + "bbox": [ + 127, + 496, + 244, + 510 + ], + "score": 1.0, + "content": "is the set of all features and", + "type": "text" + }, + { + "bbox": [ + 245, + 498, + 251, + 507 + ], + "score": 0.68, + "content": "I", + "type": "inline_equation" + }, + { + "bbox": [ + 252, + 496, + 522, + 510 + ], + "score": 1.0, + "content": "is its arbitrary subset. This problem generalizes both learning the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 89, + 507, + 522, + 522 + ], + "spans": [ + { + "bbox": [ + 89, + 507, + 159, + 522 + ], + "score": 1.0, + "content": "joint distribution", + "type": "text" + }, + { + "bbox": [ + 159, + 508, + 178, + 520 + ], + "score": 0.91, + "content": "p ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 179, + 507, + 341, + 522 + ], + "score": 1.0, + "content": "and learning the conditional distribution", + "type": "text" + }, + { + "bbox": [ + 341, + 509, + 369, + 520 + ], + "score": 0.94, + "content": "p ( x | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 369, + 507, + 522, + 522 + ], + "score": 1.0, + "content": ". To tackle this problem, we propose a", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 89, + 518, + 522, + 532 + ], + "spans": [ + { + "bbox": [ + 89, + 518, + 522, + 532 + ], + "score": 1.0, + "content": "Variational Autoencoder with Arbitrary Conditioning (VAEAC) model. It is a latent variable model similar", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 88, + 529, + 522, + 543 + ], + "spans": [ + { + "bbox": [ + 88, + 529, + 522, + 543 + ], + "score": 1.0, + "content": "to VAE, but allows conditioning on an arbitrary subset of the features. The conditioning features affect the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 88, + 541, + 522, + 554 + ], + "spans": [ + { + "bbox": [ + 88, + 541, + 522, + 554 + ], + "score": 1.0, + "content": "prior on the latent Gaussian variables which are used to generate unobserved features. The model is trained", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 89, + 551, + 376, + 565 + ], + "spans": [ + { + "bbox": [ + 89, + 551, + 376, + 565 + ], + "score": 1.0, + "content": "using stochastic gradient variational Bayes (Kingma & Welling, 2013).", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 37, + "bbox_fs": [ + 88, + 484, + 522, + 565 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 569, + 521, + 635 + ], + "lines": [ + { + "bbox": [ + 89, + 568, + 522, + 582 + ], + "spans": [ + { + "bbox": [ + 89, + 568, + 522, + 582 + ], + "score": 1.0, + "content": "We consider two most natural applications of the proposed model. The first one is feature imputation where", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 89, + 579, + 522, + 592 + ], + "spans": [ + { + "bbox": [ + 89, + 579, + 522, + 592 + ], + "score": 1.0, + "content": "the goal is to restore the missing features given the observed ones. The imputed values may be valuable by", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 90, + 592, + 521, + 603 + ], + "spans": [ + { + "bbox": [ + 90, + 592, + 521, + 603 + ], + "score": 1.0, + "content": "themselves or may improve the performance of other machine learning algorithms which process the dataset.", + "type": "text" + } + ], + "index": 43 + }, + { + "bbox": [ + 89, + 602, + 522, + 615 + ], + "spans": [ + { + "bbox": [ + 89, + 602, + 522, + 615 + ], + "score": 1.0, + "content": "Another application is image inpainting in which the goal is to fill in an unobserved part of an image with", + "type": "text" + } + ], + "index": 44 + }, + { + "bbox": [ + 89, + 612, + 521, + 625 + ], + "spans": [ + { + "bbox": [ + 89, + 612, + 521, + 625 + ], + "score": 1.0, + "content": "an artificial content in a realistic way. This can be used for removing unnecessary objects from the images", + "type": "text" + } + ], + "index": 45 + }, + { + "bbox": [ + 89, + 624, + 386, + 637 + ], + "spans": [ + { + "bbox": [ + 89, + 624, + 386, + 637 + ], + "score": 1.0, + "content": "or, vice versa, for complementing the partially closed or corrupted object.", + "type": "text" + } + ], + "index": 46 + } + ], + "index": 43.5, + "bbox_fs": [ + 89, + 568, + 522, + 637 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 91, + 95, + 521, + 184 + ], + "lines": [ + { + "bbox": [ + 90, + 96, + 521, + 108 + ], + "spans": [ + { + "bbox": [ + 90, + 96, + 521, + 108 + ], + "score": 1.0, + "content": "The experimental evaluation shows that the proposed model successfully samples from the conditional dis-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 89, + 105, + 521, + 120 + ], + "spans": [ + { + "bbox": [ + 89, + 105, + 521, + 120 + ], + "score": 1.0, + "content": "tributions. The distribution over samples is close to the true conditional distribution. This property is very", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 89, + 118, + 522, + 130 + ], + "spans": [ + { + "bbox": [ + 89, + 118, + 522, + 130 + ], + "score": 1.0, + "content": "important when the true distribution has several modes. The model is shown to be effective in feature", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 89, + 129, + 522, + 141 + ], + "spans": [ + { + "bbox": [ + 89, + 129, + 522, + 141 + ], + "score": 1.0, + "content": "imputation problem which helps to increase the quality of subsequent discriminative models on different", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 90, + 139, + 521, + 151 + ], + "spans": [ + { + "bbox": [ + 90, + 139, + 521, + 151 + ], + "score": 1.0, + "content": "problems from UCI datasets collection (Lichman, 2013). We demonstrate that model can generate diverse", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 89, + 151, + 521, + 163 + ], + "spans": [ + { + "bbox": [ + 89, + 151, + 521, + 163 + ], + "score": 1.0, + "content": "and realistic image inpaintings on MNIST (LeCun et al., 1998), Omniglot (Lake et al., 2015) and CelebA", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 89, + 160, + 522, + 174 + ], + "spans": [ + { + "bbox": [ + 89, + 160, + 522, + 174 + ], + "score": 1.0, + "content": "(Liu et al., 2015) datasets, and works even better than the current state of the art inpainting techniques in", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 89, + 172, + 264, + 186 + ], + "spans": [ + { + "bbox": [ + 89, + 172, + 264, + 186 + ], + "score": 1.0, + "content": "terms of peak signal to noise ratio (PSNR).", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 91, + 189, + 520, + 234 + ], + "lines": [ + { + "bbox": [ + 90, + 189, + 522, + 202 + ], + "spans": [ + { + "bbox": [ + 90, + 189, + 522, + 202 + ], + "score": 1.0, + "content": "The paper is organized as follows. In section 2 we review the related works. In section 3 we briefly describe", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 89, + 199, + 522, + 213 + ], + "spans": [ + { + "bbox": [ + 89, + 199, + 522, + 213 + ], + "score": 1.0, + "content": "variational autoencoders and conditional variational autoencoders. In section 4 we define the problem, de-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 89, + 210, + 522, + 225 + ], + "spans": [ + { + "bbox": [ + 89, + 210, + 522, + 225 + ], + "score": 1.0, + "content": "scribe the VAEAC model and its training procedure. In section 5 we evaluate VAEAC. Section 6 concludes", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 89, + 221, + 509, + 235 + ], + "spans": [ + { + "bbox": [ + 89, + 221, + 509, + 235 + ], + "score": 1.0, + "content": "the paper. Appendix contains additional explanations, theoretical analysis, and experiments for VAEAC.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5 + }, + { + "type": "title", + "bbox": [ + 92, + 254, + 195, + 266 + ], + "lines": [ + { + "bbox": [ + 88, + 253, + 197, + 268 + ], + "spans": [ + { + "bbox": [ + 88, + 253, + 197, + 268 + ], + "score": 1.0, + "content": "2 RELATED WORK", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 91, + 281, + 521, + 325 + ], + "lines": [ + { + "bbox": [ + 91, + 282, + 521, + 294 + ], + "spans": [ + { + "bbox": [ + 91, + 282, + 521, + 294 + ], + "score": 1.0, + "content": "Universal Marginalizer (Douglas et al., 2017) is a model based on a feed-forward neural network which", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 88, + 291, + 522, + 305 + ], + "spans": [ + { + "bbox": [ + 88, + 291, + 522, + 305 + ], + "score": 1.0, + "content": "approximates marginals of unobserved features conditioned on observable values. A related idea of an", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 88, + 303, + 522, + 316 + ], + "spans": [ + { + "bbox": [ + 88, + 303, + 522, + 316 + ], + "score": 1.0, + "content": "autoregressive model of joint probability was previously proposed in Germain et al. (2015) and Uria et al.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 90, + 314, + 472, + 327 + ], + "spans": [ + { + "bbox": [ + 90, + 314, + 472, + 327 + ], + "score": 1.0, + "content": "(2016). The description of the model and comparison with VAEAC are available in section 5.3.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14.5 + }, + { + "type": "text", + "bbox": [ + 91, + 331, + 521, + 408 + ], + "lines": [ + { + "bbox": [ + 90, + 331, + 520, + 344 + ], + "spans": [ + { + "bbox": [ + 90, + 331, + 520, + 344 + ], + "score": 1.0, + "content": "Yoon et al. (2018) propose a GANs-based model called GAIN which solves the same problem as VAEAC.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 89, + 342, + 522, + 356 + ], + "spans": [ + { + "bbox": [ + 89, + 342, + 522, + 356 + ], + "score": 1.0, + "content": "In contrast to VAEAC, GAIN does not use unobserved data during training, which makes it easier to apply to", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 90, + 353, + 521, + 366 + ], + "spans": [ + { + "bbox": [ + 90, + 353, + 521, + 366 + ], + "score": 1.0, + "content": "the missing features imputation problem. Nevertheless, it turns into a disadvantage when the fully-observed", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 89, + 363, + 522, + 378 + ], + "spans": [ + { + "bbox": [ + 89, + 363, + 522, + 378 + ], + "score": 1.0, + "content": "training data is available but the missingness rate at the testing stage is high. For example, in inpainting", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 89, + 375, + 521, + 388 + ], + "spans": [ + { + "bbox": [ + 89, + 375, + 521, + 388 + ], + "score": 1.0, + "content": "setting GAIN cannot learn the conditional distribution over MNIST digits given one horizontal line of the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 89, + 385, + 521, + 399 + ], + "spans": [ + { + "bbox": [ + 89, + 385, + 521, + 399 + ], + "score": 1.0, + "content": "image while VAEAC can (see appendix D.4). The comparison of VAEAC and GAIN on the missing feature", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 89, + 397, + 337, + 410 + ], + "spans": [ + { + "bbox": [ + 89, + 397, + 337, + 410 + ], + "score": 1.0, + "content": "imputation problem is given in section 5.1 and appendix D.2.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 91, + 414, + 520, + 480 + ], + "lines": [ + { + "bbox": [ + 90, + 414, + 522, + 426 + ], + "spans": [ + { + "bbox": [ + 90, + 414, + 522, + 426 + ], + "score": 1.0, + "content": "Rezende et al. (2014) [Appendix F], Sohl-Dickstein et al. (2015), Goyal et al. (2017), and Bordes et al.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 89, + 424, + 522, + 438 + ], + "spans": [ + { + "bbox": [ + 89, + 424, + 522, + 438 + ], + "score": 1.0, + "content": "(2017) propose to fill missing data with noise and run Markov chain with a learned transition operator. The", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 89, + 436, + 522, + 449 + ], + "spans": [ + { + "bbox": [ + 89, + 436, + 522, + 449 + ], + "score": 1.0, + "content": "stationary distribution of such chains approximates the true conditional distribution of the unobserved fea-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 89, + 447, + 522, + 460 + ], + "spans": [ + { + "bbox": [ + 89, + 447, + 522, + 460 + ], + "score": 1.0, + "content": "tures. Bachman & Precup (2015) consider missing feature imputation in terms of Markov decision process", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 89, + 458, + 522, + 471 + ], + "spans": [ + { + "bbox": [ + 89, + 458, + 522, + 471 + ], + "score": 1.0, + "content": "and propose LSTM-based sequential decision making model to solve it. Nevertheless, these methods are", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 88, + 469, + 423, + 482 + ], + "spans": [ + { + "bbox": [ + 88, + 469, + 423, + 482 + ], + "score": 1.0, + "content": "computationally expensive at the test time and require fully-observed training data.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26.5 + }, + { + "type": "text", + "bbox": [ + 91, + 486, + 520, + 519 + ], + "lines": [ + { + "bbox": [ + 90, + 484, + 522, + 499 + ], + "spans": [ + { + "bbox": [ + 90, + 484, + 522, + 499 + ], + "score": 1.0, + "content": "Image inpainting is a classic computer vision problem. Most of the earlier methods rely on local and texture", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 90, + 497, + 521, + 509 + ], + "spans": [ + { + "bbox": [ + 90, + 497, + 521, + 509 + ], + "score": 1.0, + "content": "information or hand-crafted problem-specific features (Bertalmio et al., 2000). In past years multiple neural", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 89, + 507, + 281, + 521 + ], + "spans": [ + { + "bbox": [ + 89, + 507, + 281, + 521 + ], + "score": 1.0, + "content": "network based approaches have been proposed.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 91, + 524, + 520, + 602 + ], + "lines": [ + { + "bbox": [ + 89, + 524, + 522, + 537 + ], + "spans": [ + { + "bbox": [ + 89, + 524, + 522, + 537 + ], + "score": 1.0, + "content": "Pathak et al. (2016), Yeh et al. (2016) and Yang et al. (2017) use different kinds and combinations of ad-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 88, + 535, + 522, + 548 + ], + "spans": [ + { + "bbox": [ + 88, + 535, + 522, + 548 + ], + "score": 1.0, + "content": "versarial, reconstruction, texture and other losses. Li et al. (2017) focuses on face inpainting and uses two", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 89, + 546, + 523, + 560 + ], + "spans": [ + { + "bbox": [ + 89, + 546, + 523, + 560 + ], + "score": 1.0, + "content": "adversarial losses and one semantic parsing loss to train the generative model. In Yeh et al. (2017) GANs", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 88, + 557, + 522, + 571 + ], + "spans": [ + { + "bbox": [ + 88, + 557, + 522, + 571 + ], + "score": 1.0, + "content": "are first trained on the whole training dataset. The inpainting is an optimization procedure that finds the", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 89, + 569, + 522, + 581 + ], + "spans": [ + { + "bbox": [ + 89, + 569, + 522, + 581 + ], + "score": 1.0, + "content": "latent variables that explain the observed features best. Then, the obtained latents are passed through the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 88, + 579, + 522, + 592 + ], + "spans": [ + { + "bbox": [ + 88, + 579, + 522, + 592 + ], + "score": 1.0, + "content": "generative model to restore the unobserved portion of the image. We can say that VAEAC is a similar model", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 89, + 590, + 464, + 604 + ], + "spans": [ + { + "bbox": [ + 89, + 590, + 464, + 604 + ], + "score": 1.0, + "content": "which uses prior network to find a proper latents instead of solving the optimization problem.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 36 + }, + { + "type": "text", + "bbox": [ + 91, + 607, + 520, + 651 + ], + "lines": [ + { + "bbox": [ + 88, + 605, + 522, + 622 + ], + "spans": [ + { + "bbox": [ + 88, + 605, + 522, + 622 + ], + "score": 1.0, + "content": "All described methods aim to produce a single realistic inpainting, while VAEAC is capable of sampling", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 90, + 618, + 521, + 631 + ], + "spans": [ + { + "bbox": [ + 90, + 618, + 521, + 631 + ], + "score": 1.0, + "content": "diverse inpaintings. Additionally, Yeh et al. (2016), Yang et al. (2017) and Yeh et al. (2017) have high test-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 89, + 629, + 522, + 642 + ], + "spans": [ + { + "bbox": [ + 89, + 629, + 522, + 642 + ], + "score": 1.0, + "content": "time computational complexity of inpainting, because they require an optimization problem to be solved.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 90, + 640, + 431, + 653 + ], + "spans": [ + { + "bbox": [ + 90, + 640, + 431, + 653 + ], + "score": 1.0, + "content": "On the other hand, VAEAC is a “single-shot” method with a low computational cost.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 41.5 + } + ], + "page_idx": 1, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 40, + 277, + 50 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 286, + 671, + 292, + 680 + ], + "lines": [ + { + "bbox": [ + 285, + 670, + 294, + 683 + ], + "spans": [ + { + "bbox": [ + 285, + 670, + 294, + 683 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 13, + "width": 9 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 91, + 95, + 521, + 184 + ], + "lines": [ + { + "bbox": [ + 90, + 96, + 521, + 108 + ], + "spans": [ + { + "bbox": [ + 90, + 96, + 521, + 108 + ], + "score": 1.0, + "content": "The experimental evaluation shows that the proposed model successfully samples from the conditional dis-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 89, + 105, + 521, + 120 + ], + "spans": [ + { + "bbox": [ + 89, + 105, + 521, + 120 + ], + "score": 1.0, + "content": "tributions. The distribution over samples is close to the true conditional distribution. This property is very", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 89, + 118, + 522, + 130 + ], + "spans": [ + { + "bbox": [ + 89, + 118, + 522, + 130 + ], + "score": 1.0, + "content": "important when the true distribution has several modes. The model is shown to be effective in feature", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 89, + 129, + 522, + 141 + ], + "spans": [ + { + "bbox": [ + 89, + 129, + 522, + 141 + ], + "score": 1.0, + "content": "imputation problem which helps to increase the quality of subsequent discriminative models on different", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 90, + 139, + 521, + 151 + ], + "spans": [ + { + "bbox": [ + 90, + 139, + 521, + 151 + ], + "score": 1.0, + "content": "problems from UCI datasets collection (Lichman, 2013). We demonstrate that model can generate diverse", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 89, + 151, + 521, + 163 + ], + "spans": [ + { + "bbox": [ + 89, + 151, + 521, + 163 + ], + "score": 1.0, + "content": "and realistic image inpaintings on MNIST (LeCun et al., 1998), Omniglot (Lake et al., 2015) and CelebA", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 89, + 160, + 522, + 174 + ], + "spans": [ + { + "bbox": [ + 89, + 160, + 522, + 174 + ], + "score": 1.0, + "content": "(Liu et al., 2015) datasets, and works even better than the current state of the art inpainting techniques in", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 89, + 172, + 264, + 186 + ], + "spans": [ + { + "bbox": [ + 89, + 172, + 264, + 186 + ], + "score": 1.0, + "content": "terms of peak signal to noise ratio (PSNR).", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 3.5, + "bbox_fs": [ + 89, + 96, + 522, + 186 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 189, + 520, + 234 + ], + "lines": [ + { + "bbox": [ + 90, + 189, + 522, + 202 + ], + "spans": [ + { + "bbox": [ + 90, + 189, + 522, + 202 + ], + "score": 1.0, + "content": "The paper is organized as follows. In section 2 we review the related works. In section 3 we briefly describe", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 89, + 199, + 522, + 213 + ], + "spans": [ + { + "bbox": [ + 89, + 199, + 522, + 213 + ], + "score": 1.0, + "content": "variational autoencoders and conditional variational autoencoders. In section 4 we define the problem, de-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 89, + 210, + 522, + 225 + ], + "spans": [ + { + "bbox": [ + 89, + 210, + 522, + 225 + ], + "score": 1.0, + "content": "scribe the VAEAC model and its training procedure. In section 5 we evaluate VAEAC. Section 6 concludes", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 89, + 221, + 509, + 235 + ], + "spans": [ + { + "bbox": [ + 89, + 221, + 509, + 235 + ], + "score": 1.0, + "content": "the paper. Appendix contains additional explanations, theoretical analysis, and experiments for VAEAC.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5, + "bbox_fs": [ + 89, + 189, + 522, + 235 + ] + }, + { + "type": "title", + "bbox": [ + 92, + 254, + 195, + 266 + ], + "lines": [ + { + "bbox": [ + 88, + 253, + 197, + 268 + ], + "spans": [ + { + "bbox": [ + 88, + 253, + 197, + 268 + ], + "score": 1.0, + "content": "2 RELATED WORK", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 91, + 281, + 521, + 325 + ], + "lines": [ + { + "bbox": [ + 91, + 282, + 521, + 294 + ], + "spans": [ + { + "bbox": [ + 91, + 282, + 521, + 294 + ], + "score": 1.0, + "content": "Universal Marginalizer (Douglas et al., 2017) is a model based on a feed-forward neural network which", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 88, + 291, + 522, + 305 + ], + "spans": [ + { + "bbox": [ + 88, + 291, + 522, + 305 + ], + "score": 1.0, + "content": "approximates marginals of unobserved features conditioned on observable values. A related idea of an", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 88, + 303, + 522, + 316 + ], + "spans": [ + { + "bbox": [ + 88, + 303, + 522, + 316 + ], + "score": 1.0, + "content": "autoregressive model of joint probability was previously proposed in Germain et al. (2015) and Uria et al.", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 90, + 314, + 472, + 327 + ], + "spans": [ + { + "bbox": [ + 90, + 314, + 472, + 327 + ], + "score": 1.0, + "content": "(2016). The description of the model and comparison with VAEAC are available in section 5.3.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 14.5, + "bbox_fs": [ + 88, + 282, + 522, + 327 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 331, + 521, + 408 + ], + "lines": [ + { + "bbox": [ + 90, + 331, + 520, + 344 + ], + "spans": [ + { + "bbox": [ + 90, + 331, + 520, + 344 + ], + "score": 1.0, + "content": "Yoon et al. (2018) propose a GANs-based model called GAIN which solves the same problem as VAEAC.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 89, + 342, + 522, + 356 + ], + "spans": [ + { + "bbox": [ + 89, + 342, + 522, + 356 + ], + "score": 1.0, + "content": "In contrast to VAEAC, GAIN does not use unobserved data during training, which makes it easier to apply to", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 90, + 353, + 521, + 366 + ], + "spans": [ + { + "bbox": [ + 90, + 353, + 521, + 366 + ], + "score": 1.0, + "content": "the missing features imputation problem. Nevertheless, it turns into a disadvantage when the fully-observed", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 89, + 363, + 522, + 378 + ], + "spans": [ + { + "bbox": [ + 89, + 363, + 522, + 378 + ], + "score": 1.0, + "content": "training data is available but the missingness rate at the testing stage is high. For example, in inpainting", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 89, + 375, + 521, + 388 + ], + "spans": [ + { + "bbox": [ + 89, + 375, + 521, + 388 + ], + "score": 1.0, + "content": "setting GAIN cannot learn the conditional distribution over MNIST digits given one horizontal line of the", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 89, + 385, + 521, + 399 + ], + "spans": [ + { + "bbox": [ + 89, + 385, + 521, + 399 + ], + "score": 1.0, + "content": "image while VAEAC can (see appendix D.4). The comparison of VAEAC and GAIN on the missing feature", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 89, + 397, + 337, + 410 + ], + "spans": [ + { + "bbox": [ + 89, + 397, + 337, + 410 + ], + "score": 1.0, + "content": "imputation problem is given in section 5.1 and appendix D.2.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20, + "bbox_fs": [ + 89, + 331, + 522, + 410 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 414, + 520, + 480 + ], + "lines": [ + { + "bbox": [ + 90, + 414, + 522, + 426 + ], + "spans": [ + { + "bbox": [ + 90, + 414, + 522, + 426 + ], + "score": 1.0, + "content": "Rezende et al. (2014) [Appendix F], Sohl-Dickstein et al. (2015), Goyal et al. (2017), and Bordes et al.", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 89, + 424, + 522, + 438 + ], + "spans": [ + { + "bbox": [ + 89, + 424, + 522, + 438 + ], + "score": 1.0, + "content": "(2017) propose to fill missing data with noise and run Markov chain with a learned transition operator. The", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 89, + 436, + 522, + 449 + ], + "spans": [ + { + "bbox": [ + 89, + 436, + 522, + 449 + ], + "score": 1.0, + "content": "stationary distribution of such chains approximates the true conditional distribution of the unobserved fea-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 89, + 447, + 522, + 460 + ], + "spans": [ + { + "bbox": [ + 89, + 447, + 522, + 460 + ], + "score": 1.0, + "content": "tures. Bachman & Precup (2015) consider missing feature imputation in terms of Markov decision process", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 89, + 458, + 522, + 471 + ], + "spans": [ + { + "bbox": [ + 89, + 458, + 522, + 471 + ], + "score": 1.0, + "content": "and propose LSTM-based sequential decision making model to solve it. Nevertheless, these methods are", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 88, + 469, + 423, + 482 + ], + "spans": [ + { + "bbox": [ + 88, + 469, + 423, + 482 + ], + "score": 1.0, + "content": "computationally expensive at the test time and require fully-observed training data.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 26.5, + "bbox_fs": [ + 88, + 414, + 522, + 482 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 486, + 520, + 519 + ], + "lines": [ + { + "bbox": [ + 90, + 484, + 522, + 499 + ], + "spans": [ + { + "bbox": [ + 90, + 484, + 522, + 499 + ], + "score": 1.0, + "content": "Image inpainting is a classic computer vision problem. Most of the earlier methods rely on local and texture", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 90, + 497, + 521, + 509 + ], + "spans": [ + { + "bbox": [ + 90, + 497, + 521, + 509 + ], + "score": 1.0, + "content": "information or hand-crafted problem-specific features (Bertalmio et al., 2000). In past years multiple neural", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 89, + 507, + 281, + 521 + ], + "spans": [ + { + "bbox": [ + 89, + 507, + 281, + 521 + ], + "score": 1.0, + "content": "network based approaches have been proposed.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 89, + 484, + 522, + 521 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 524, + 520, + 602 + ], + "lines": [ + { + "bbox": [ + 89, + 524, + 522, + 537 + ], + "spans": [ + { + "bbox": [ + 89, + 524, + 522, + 537 + ], + "score": 1.0, + "content": "Pathak et al. (2016), Yeh et al. (2016) and Yang et al. (2017) use different kinds and combinations of ad-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 88, + 535, + 522, + 548 + ], + "spans": [ + { + "bbox": [ + 88, + 535, + 522, + 548 + ], + "score": 1.0, + "content": "versarial, reconstruction, texture and other losses. Li et al. (2017) focuses on face inpainting and uses two", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 89, + 546, + 523, + 560 + ], + "spans": [ + { + "bbox": [ + 89, + 546, + 523, + 560 + ], + "score": 1.0, + "content": "adversarial losses and one semantic parsing loss to train the generative model. In Yeh et al. (2017) GANs", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 88, + 557, + 522, + 571 + ], + "spans": [ + { + "bbox": [ + 88, + 557, + 522, + 571 + ], + "score": 1.0, + "content": "are first trained on the whole training dataset. The inpainting is an optimization procedure that finds the", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 89, + 569, + 522, + 581 + ], + "spans": [ + { + "bbox": [ + 89, + 569, + 522, + 581 + ], + "score": 1.0, + "content": "latent variables that explain the observed features best. Then, the obtained latents are passed through the", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 88, + 579, + 522, + 592 + ], + "spans": [ + { + "bbox": [ + 88, + 579, + 522, + 592 + ], + "score": 1.0, + "content": "generative model to restore the unobserved portion of the image. We can say that VAEAC is a similar model", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 89, + 590, + 464, + 604 + ], + "spans": [ + { + "bbox": [ + 89, + 590, + 464, + 604 + ], + "score": 1.0, + "content": "which uses prior network to find a proper latents instead of solving the optimization problem.", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 36, + "bbox_fs": [ + 88, + 524, + 523, + 604 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 607, + 520, + 651 + ], + "lines": [ + { + "bbox": [ + 88, + 605, + 522, + 622 + ], + "spans": [ + { + "bbox": [ + 88, + 605, + 522, + 622 + ], + "score": 1.0, + "content": "All described methods aim to produce a single realistic inpainting, while VAEAC is capable of sampling", + "type": "text" + } + ], + "index": 40 + }, + { + "bbox": [ + 90, + 618, + 521, + 631 + ], + "spans": [ + { + "bbox": [ + 90, + 618, + 521, + 631 + ], + "score": 1.0, + "content": "diverse inpaintings. Additionally, Yeh et al. (2016), Yang et al. (2017) and Yeh et al. (2017) have high test-", + "type": "text" + } + ], + "index": 41 + }, + { + "bbox": [ + 89, + 629, + 522, + 642 + ], + "spans": [ + { + "bbox": [ + 89, + 629, + 522, + 642 + ], + "score": 1.0, + "content": "time computational complexity of inpainting, because they require an optimization problem to be solved.", + "type": "text" + } + ], + "index": 42 + }, + { + "bbox": [ + 90, + 640, + 431, + 653 + ], + "spans": [ + { + "bbox": [ + 90, + 640, + 431, + 653 + ], + "score": 1.0, + "content": "On the other hand, VAEAC is a “single-shot” method with a low computational cost.", + "type": "text" + } + ], + "index": 43 + } + ], + "index": 41.5, + "bbox_fs": [ + 88, + 605, + 522, + 653 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 91, + 94, + 184, + 107 + ], + "lines": [ + { + "bbox": [ + 88, + 93, + 185, + 110 + ], + "spans": [ + { + "bbox": [ + 88, + 93, + 185, + 110 + ], + "score": 1.0, + "content": "3 BACKGROUND", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 91, + 119, + 243, + 131 + ], + "lines": [ + { + "bbox": [ + 90, + 119, + 244, + 131 + ], + "spans": [ + { + "bbox": [ + 90, + 119, + 244, + 131 + ], + "score": 1.0, + "content": "3.1 VARIATIONAL AUTOENCODER", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 90, + 140, + 522, + 208 + ], + "lines": [ + { + "bbox": [ + 90, + 139, + 522, + 153 + ], + "spans": [ + { + "bbox": [ + 90, + 139, + 522, + 153 + ], + "score": 1.0, + "content": "Variational autoencoder (Kingma & Welling, 2013) (VAE) is a directed generative model with latent vari-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 90, + 151, + 522, + 163 + ], + "spans": [ + { + "bbox": [ + 90, + 151, + 462, + 163 + ], + "score": 1.0, + "content": "ables. The generative process in variational autoencoder is as follows: first, a latent variable", + "type": "text" + }, + { + "bbox": [ + 462, + 153, + 469, + 161 + ], + "score": 0.76, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 151, + 522, + 163 + ], + "score": 1.0, + "content": "is generated", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 90, + 162, + 522, + 175 + ], + "spans": [ + { + "bbox": [ + 90, + 162, + 199, + 175 + ], + "score": 1.0, + "content": "from the prior distribution", + "type": "text" + }, + { + "bbox": [ + 200, + 162, + 218, + 174 + ], + "score": 0.92, + "content": "p ( z )", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 162, + 295, + 175 + ], + "score": 1.0, + "content": ", and then the data", + "type": "text" + }, + { + "bbox": [ + 295, + 164, + 302, + 172 + ], + "score": 0.77, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 162, + 485, + 175 + ], + "score": 1.0, + "content": "is generated from the generative distribution", + "type": "text" + }, + { + "bbox": [ + 486, + 162, + 517, + 174 + ], + "score": 0.9, + "content": "p _ { \\theta } ( x | z )", + "type": "inline_equation" + }, + { + "bbox": [ + 517, + 162, + 522, + 175 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 89, + 172, + 522, + 186 + ], + "spans": [ + { + "bbox": [ + 89, + 172, + 117, + 186 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 117, + 174, + 123, + 183 + ], + "score": 0.79, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 172, + 428, + 186 + ], + "score": 1.0, + "content": "are the generative model’s parameters. This process induces the distribution", + "type": "text" + }, + { + "bbox": [ + 428, + 173, + 518, + 186 + ], + "score": 0.92, + "content": "p _ { \\theta } ( x ) = \\mathbb { E } _ { p ( z ) } p _ { \\theta } ( x | z )", + "type": "inline_equation" + }, + { + "bbox": [ + 518, + 172, + 522, + 186 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 89, + 184, + 522, + 198 + ], + "spans": [ + { + "bbox": [ + 89, + 184, + 158, + 198 + ], + "score": 1.0, + "content": "The distribution", + "type": "text" + }, + { + "bbox": [ + 158, + 185, + 190, + 197 + ], + "score": 0.93, + "content": "p _ { \\theta } ( x | z )", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 184, + 393, + 198 + ], + "score": 1.0, + "content": "is modeled by a neural network with parameters", + "type": "text" + }, + { + "bbox": [ + 393, + 185, + 399, + 195 + ], + "score": 0.53, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 184, + 406, + 198 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 406, + 185, + 425, + 197 + ], + "score": 0.86, + "content": "p ( z )", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 184, + 522, + 198 + ], + "score": 1.0, + "content": "is a standard Gaussian", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 89, + 195, + 141, + 208 + ], + "spans": [ + { + "bbox": [ + 89, + 195, + 141, + 208 + ], + "score": 1.0, + "content": "distribution.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 4.5 + }, + { + "type": "text", + "bbox": [ + 91, + 212, + 520, + 258 + ], + "lines": [ + { + "bbox": [ + 88, + 210, + 523, + 229 + ], + "spans": [ + { + "bbox": [ + 88, + 210, + 155, + 229 + ], + "score": 1.0, + "content": "The parameters", + "type": "text" + }, + { + "bbox": [ + 155, + 214, + 162, + 223 + ], + "score": 0.79, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 210, + 431, + 229 + ], + "score": 1.0, + "content": "are tuned by maximizing the likelihood of the training data points", + "type": "text" + }, + { + "bbox": [ + 431, + 212, + 464, + 225 + ], + "score": 0.93, + "content": "\\{ x _ { i } \\} _ { i = 1 } ^ { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 210, + 523, + 229 + ], + "score": 1.0, + "content": "from the true", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 89, + 222, + 522, + 237 + ], + "spans": [ + { + "bbox": [ + 89, + 222, + 159, + 237 + ], + "score": 1.0, + "content": "data distribution", + "type": "text" + }, + { + "bbox": [ + 160, + 224, + 184, + 236 + ], + "score": 0.92, + "content": "p _ { d } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 222, + 522, + 237 + ], + "score": 1.0, + "content": ". In general, this optimization problem is challenging due to intractable posterior", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 89, + 235, + 522, + 249 + ], + "spans": [ + { + "bbox": [ + 89, + 235, + 522, + 249 + ], + "score": 1.0, + "content": "inference. However, a variational lower bound can be optimized efficiently using backpropagation and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 90, + 246, + 202, + 258 + ], + "spans": [ + { + "bbox": [ + 90, + 246, + 202, + 258 + ], + "score": 1.0, + "content": "stochastic gradient descent:", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5 + }, + { + "type": "interline_equation", + "bbox": [ + 101, + 266, + 499, + 311 + ], + "lines": [ + { + "bbox": [ + 101, + 266, + 499, + 311 + ], + "spans": [ + { + "bbox": [ + 101, + 266, + 499, + 311 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\log p _ { \\theta } ( x ) = \\mathbb { E } _ { q _ { \\phi } ( z | x ) } \\log \\frac { p _ { \\theta } ( x , z ) } { q _ { \\phi } ( z | x ) } + D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( z | x ) \\| p ( z | x , \\theta ) \\big ) } \\\\ & { \\qquad \\quad \\ge \\mathbb { E } _ { q _ { \\phi } ( z | x ) } \\log p _ { \\theta } ( x | z ) - D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( z | x ) \\| p ( z ) \\big ) = L _ { V A E } \\big ( x ; \\theta , \\phi \\big ) } \\end{array}", + "type": "interline_equation", + "image_path": "fed675d9452648daab0a6ca45636d1aac44900502c485323a4dc8122dfe3f9fe.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 101, + 266, + 499, + 281.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 101, + 281.0, + 499, + 296.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 101, + 296.0, + 499, + 311.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 91, + 314, + 522, + 381 + ], + "lines": [ + { + "bbox": [ + 88, + 313, + 522, + 328 + ], + "spans": [ + { + "bbox": [ + 88, + 313, + 113, + 328 + ], + "score": 1.0, + "content": "Here", + "type": "text" + }, + { + "bbox": [ + 113, + 315, + 145, + 327 + ], + "score": 0.93, + "content": "q _ { \\phi } ( z | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 313, + 457, + 328 + ], + "score": 1.0, + "content": "is a proposal distribution parameterized by neural network with parameters", + "type": "text" + }, + { + "bbox": [ + 457, + 315, + 465, + 326 + ], + "score": 0.86, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 313, + 522, + 328 + ], + "score": 1.0, + "content": "that approxi-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 89, + 325, + 522, + 338 + ], + "spans": [ + { + "bbox": [ + 89, + 325, + 170, + 338 + ], + "score": 1.0, + "content": "mates the posterior", + "type": "text" + }, + { + "bbox": [ + 171, + 327, + 206, + 338 + ], + "score": 0.91, + "content": "p ( z | x , \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 325, + 522, + 338 + ], + "score": 1.0, + "content": ". Usually this distribution is Gaussian with a diagonal covariance matrix. The", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 90, + 336, + 521, + 349 + ], + "spans": [ + { + "bbox": [ + 90, + 337, + 117, + 349 + ], + "score": 1.0, + "content": "closer", + "type": "text" + }, + { + "bbox": [ + 117, + 337, + 149, + 349 + ], + "score": 0.92, + "content": "q _ { \\phi } ( z | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 337, + 160, + 349 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 161, + 338, + 197, + 349 + ], + "score": 0.92, + "content": "p ( z | x , \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 337, + 342, + 349 + ], + "score": 1.0, + "content": ", the tighter variational lower bound", + "type": "text" + }, + { + "bbox": [ + 342, + 336, + 392, + 349 + ], + "score": 0.92, + "content": "L _ { V A E } ( \\theta , \\phi )", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 337, + 521, + 349 + ], + "score": 1.0, + "content": ". To compute the gradient of the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 89, + 347, + 522, + 360 + ], + "spans": [ + { + "bbox": [ + 89, + 347, + 254, + 360 + ], + "score": 1.0, + "content": "variational lower bound with respect to", + "type": "text" + }, + { + "bbox": [ + 254, + 348, + 261, + 359 + ], + "score": 0.83, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 347, + 402, + 360 + ], + "score": 1.0, + "content": ", reparameterization trick is used:", + "type": "text" + }, + { + "bbox": [ + 402, + 347, + 492, + 360 + ], + "score": 0.92, + "content": "z = \\bar { \\mu } _ { \\phi } ( x ) + \\bar { \\varepsilon } \\sigma _ { \\phi } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 347, + 522, + 360 + ], + "score": 1.0, + "content": "where", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 91, + 358, + 522, + 371 + ], + "spans": [ + { + "bbox": [ + 91, + 358, + 142, + 371 + ], + "score": 0.92, + "content": "\\varepsilon \\sim \\mathcal { N } ( 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 358, + 161, + 371 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 161, + 360, + 173, + 371 + ], + "score": 0.84, + "content": "\\mu _ { \\phi }", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 358, + 192, + 371 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 192, + 360, + 204, + 371 + ], + "score": 0.84, + "content": "\\sigma _ { \\phi }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 358, + 522, + 371 + ], + "score": 1.0, + "content": "are deterministic functions parameterized by neural networks. So the gradient", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 89, + 369, + 517, + 382 + ], + "spans": [ + { + "bbox": [ + 89, + 369, + 517, + 382 + ], + "score": 1.0, + "content": "can be estimated using Monte-Carlo method for the first term and computing the second term analytically:", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 17.5 + }, + { + "type": "interline_equation", + "bbox": [ + 128, + 387, + 482, + 413 + ], + "lines": [ + { + "bbox": [ + 128, + 387, + 482, + 413 + ], + "spans": [ + { + "bbox": [ + 128, + 387, + 482, + 413 + ], + "score": 0.92, + "content": "\\frac { \\partial L _ { V A E } ( x ; \\theta , \\phi ) } { \\partial \\phi } = \\mathbb { E } _ { \\varepsilon \\sim \\mathcal { N } ( 0 , I ) } \\frac { \\partial } { \\partial \\phi } \\log p _ { \\theta } ( x | \\mu _ { \\phi } ( x ) + \\varepsilon \\sigma _ { \\phi } ( x ) ) - \\frac { \\partial } { \\partial \\phi } D _ { \\mathrm { K L } } ( q _ { \\phi } ( z | x ) \\| p ( z ) ) .", + "type": "interline_equation", + "image_path": "392ee19314ad1169b0b4fd7d080fa9dd49328ced339e9d76af78ca28090587f9.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 128, + 387, + 482, + 413 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 90, + 418, + 449, + 430 + ], + "lines": [ + { + "bbox": [ + 89, + 418, + 446, + 432 + ], + "spans": [ + { + "bbox": [ + 89, + 418, + 103, + 432 + ], + "score": 1.0, + "content": "So", + "type": "text" + }, + { + "bbox": [ + 104, + 419, + 153, + 431 + ], + "score": 0.92, + "content": "L _ { V A E } ( \\theta , \\phi )", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 418, + 415, + 432 + ], + "score": 1.0, + "content": "can be optimized using stochastic gradient ascent with respect to", + "type": "text" + }, + { + "bbox": [ + 415, + 419, + 422, + 430 + ], + "score": 0.86, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 418, + 440, + 432 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 441, + 419, + 446, + 429 + ], + "score": 0.79, + "content": "\\theta", + "type": "inline_equation" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "title", + "bbox": [ + 91, + 444, + 308, + 456 + ], + "lines": [ + { + "bbox": [ + 89, + 443, + 309, + 457 + ], + "spans": [ + { + "bbox": [ + 89, + 443, + 309, + 457 + ], + "score": 1.0, + "content": "3.2 CONDITIONAL VARIATIONAL AUTOENCODER", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 90, + 464, + 521, + 520 + ], + "lines": [ + { + "bbox": [ + 90, + 465, + 522, + 478 + ], + "spans": [ + { + "bbox": [ + 90, + 465, + 522, + 478 + ], + "score": 1.0, + "content": "Conditional variational autoencoder (Sohn et al., 2015) (CVAE) approximates the conditional distribution", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 91, + 475, + 522, + 489 + ], + "spans": [ + { + "bbox": [ + 91, + 476, + 122, + 488 + ], + "score": 0.92, + "content": "p _ { d } ( x | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 123, + 475, + 363, + 489 + ], + "score": 1.0, + "content": ". It outperforms deterministic models when the distribution", + "type": "text" + }, + { + "bbox": [ + 364, + 476, + 396, + 488 + ], + "score": 0.93, + "content": "p _ { d } ( x | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 475, + 494, + 489 + ], + "score": 1.0, + "content": "is multi-modal (diverse", + "type": "text" + }, + { + "bbox": [ + 494, + 478, + 505, + 486 + ], + "score": 0.64, + "content": "x \\mathbf { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 506, + 475, + 522, + 489 + ], + "score": 1.0, + "content": "are", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 88, + 486, + 522, + 500 + ], + "spans": [ + { + "bbox": [ + 88, + 486, + 178, + 500 + ], + "score": 1.0, + "content": "probable for the given", + "type": "text" + }, + { + "bbox": [ + 178, + 489, + 185, + 498 + ], + "score": 0.64, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 486, + 294, + 500 + ], + "score": 1.0, + "content": "). For example, assume that", + "type": "text" + }, + { + "bbox": [ + 294, + 488, + 301, + 497 + ], + "score": 0.8, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 301, + 486, + 522, + 500 + ], + "score": 1.0, + "content": "is a real-valued image. Then, a deterministic regression", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 88, + 497, + 522, + 510 + ], + "spans": [ + { + "bbox": [ + 88, + 497, + 405, + 510 + ], + "score": 1.0, + "content": "model with mean squared error loss would predict the average blurry value for", + "type": "text" + }, + { + "bbox": [ + 405, + 500, + 412, + 508 + ], + "score": 0.78, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 497, + 522, + 510 + ], + "score": 1.0, + "content": ". On the other hand, CVAE", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 88, + 508, + 428, + 522 + ], + "spans": [ + { + "bbox": [ + 88, + 508, + 190, + 522 + ], + "score": 1.0, + "content": "learns the distribution of", + "type": "text" + }, + { + "bbox": [ + 191, + 511, + 197, + 519 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 508, + 428, + 522 + ], + "score": 1.0, + "content": ", from which one can sample diverse and realistic objects.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 87, + 525, + 520, + 549 + ], + "lines": [ + { + "bbox": [ + 90, + 524, + 520, + 538 + ], + "spans": [ + { + "bbox": [ + 90, + 524, + 520, + 538 + ], + "score": 1.0, + "content": "Variational lower bound for CVAE can be derived similarly to VAE by conditioning all considered distribu-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 89, + 535, + 136, + 551 + ], + "spans": [ + { + "bbox": [ + 89, + 535, + 125, + 551 + ], + "score": 1.0, + "content": "tions on", + "type": "text" + }, + { + "bbox": [ + 125, + 540, + 131, + 549 + ], + "score": 0.72, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 535, + 136, + 551 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + }, + { + "type": "interline_equation", + "bbox": [ + 114, + 554, + 497, + 569 + ], + "lines": [ + { + "bbox": [ + 114, + 554, + 497, + 569 + ], + "spans": [ + { + "bbox": [ + 114, + 554, + 497, + 569 + ], + "score": 0.91, + "content": "L _ { C V A E } ( x , y ; \\theta , \\psi , \\phi ) = \\mathbb { E } _ { q _ { \\phi } ( z | x , y ) } \\log p _ { \\theta } ( x | z , y ) - D _ { \\mathrm { K L } } ( q _ { \\phi } ( z | x , y ) | | p _ { \\psi } ( z | y ) ) \\leq \\log p _ { \\theta , \\psi } ( x | y )", + "type": "interline_equation", + "image_path": "1a364d873a354bd42b6ac3c1a473c1e2729b1b87d4dce0f0bfde7ab681819e08.jpg" + } + ] + } + ], + "index": 31, + "virtual_lines": [ + { + "bbox": [ + 114, + 554, + 497, + 569 + ], + "spans": [], + "index": 31 + } + ] + }, + { + "type": "text", + "bbox": [ + 91, + 579, + 522, + 613 + ], + "lines": [ + { + "bbox": [ + 89, + 578, + 521, + 592 + ], + "spans": [ + { + "bbox": [ + 89, + 578, + 521, + 592 + ], + "score": 1.0, + "content": "Similarly to VAE, this objective is optimized using the reparameterization trick. Note that the prior distribu-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 90, + 590, + 522, + 604 + ], + "spans": [ + { + "bbox": [ + 90, + 590, + 108, + 604 + ], + "score": 1.0, + "content": "tion", + "type": "text" + }, + { + "bbox": [ + 109, + 591, + 141, + 603 + ], + "score": 0.93, + "content": "p _ { \\psi } ( z | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 590, + 214, + 604 + ], + "score": 1.0, + "content": "is conditioned on", + "type": "text" + }, + { + "bbox": [ + 215, + 592, + 221, + 602 + ], + "score": 0.77, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 590, + 435, + 604 + ], + "score": 1.0, + "content": "and is modeled by a neural network with parameters", + "type": "text" + }, + { + "bbox": [ + 435, + 591, + 443, + 602 + ], + "score": 0.86, + "content": "\\psi", + "type": "inline_equation" + }, + { + "bbox": [ + 443, + 590, + 522, + 604 + ], + "score": 1.0, + "content": ". Thus, CVAE uses", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 89, + 600, + 325, + 614 + ], + "spans": [ + { + "bbox": [ + 89, + 600, + 325, + 614 + ], + "score": 1.0, + "content": "three trainable neural networks, while VAE only uses two.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33 + }, + { + "type": "text", + "bbox": [ + 91, + 617, + 520, + 652 + ], + "lines": [ + { + "bbox": [ + 90, + 618, + 521, + 630 + ], + "spans": [ + { + "bbox": [ + 90, + 618, + 521, + 630 + ], + "score": 1.0, + "content": "Also authors propose such modifications of CVAE as Gaussian stochastic neural network and hybrid model.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 89, + 628, + 522, + 641 + ], + "spans": [ + { + "bbox": [ + 89, + 628, + 522, + 641 + ], + "score": 1.0, + "content": "These modifications can be applied to our model as well. Nevertheless, we don’t use them, because of their", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 90, + 640, + 282, + 653 + ], + "spans": [ + { + "bbox": [ + 90, + 640, + 282, + 653 + ], + "score": 1.0, + "content": "disadvantage which is described in appendix C.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36 + } + ], + "page_idx": 2, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 39, + 277, + 51 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 286, + 671, + 293, + 680 + ], + "lines": [ + { + "bbox": [ + 285, + 670, + 293, + 681 + ], + "spans": [ + { + "bbox": [ + 285, + 670, + 293, + 681 + ], + "score": 1.0, + "content": "3", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 91, + 94, + 184, + 107 + ], + "lines": [ + { + "bbox": [ + 88, + 93, + 185, + 110 + ], + "spans": [ + { + "bbox": [ + 88, + 93, + 185, + 110 + ], + "score": 1.0, + "content": "3 BACKGROUND", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 91, + 119, + 243, + 131 + ], + "lines": [ + { + "bbox": [ + 90, + 119, + 244, + 131 + ], + "spans": [ + { + "bbox": [ + 90, + 119, + 244, + 131 + ], + "score": 1.0, + "content": "3.1 VARIATIONAL AUTOENCODER", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 90, + 140, + 522, + 208 + ], + "lines": [ + { + "bbox": [ + 90, + 139, + 522, + 153 + ], + "spans": [ + { + "bbox": [ + 90, + 139, + 522, + 153 + ], + "score": 1.0, + "content": "Variational autoencoder (Kingma & Welling, 2013) (VAE) is a directed generative model with latent vari-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 90, + 151, + 522, + 163 + ], + "spans": [ + { + "bbox": [ + 90, + 151, + 462, + 163 + ], + "score": 1.0, + "content": "ables. The generative process in variational autoencoder is as follows: first, a latent variable", + "type": "text" + }, + { + "bbox": [ + 462, + 153, + 469, + 161 + ], + "score": 0.76, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 151, + 522, + 163 + ], + "score": 1.0, + "content": "is generated", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 90, + 162, + 522, + 175 + ], + "spans": [ + { + "bbox": [ + 90, + 162, + 199, + 175 + ], + "score": 1.0, + "content": "from the prior distribution", + "type": "text" + }, + { + "bbox": [ + 200, + 162, + 218, + 174 + ], + "score": 0.92, + "content": "p ( z )", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 162, + 295, + 175 + ], + "score": 1.0, + "content": ", and then the data", + "type": "text" + }, + { + "bbox": [ + 295, + 164, + 302, + 172 + ], + "score": 0.77, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 303, + 162, + 485, + 175 + ], + "score": 1.0, + "content": "is generated from the generative distribution", + "type": "text" + }, + { + "bbox": [ + 486, + 162, + 517, + 174 + ], + "score": 0.9, + "content": "p _ { \\theta } ( x | z )", + "type": "inline_equation" + }, + { + "bbox": [ + 517, + 162, + 522, + 175 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 89, + 172, + 522, + 186 + ], + "spans": [ + { + "bbox": [ + 89, + 172, + 117, + 186 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 117, + 174, + 123, + 183 + ], + "score": 0.79, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 172, + 428, + 186 + ], + "score": 1.0, + "content": "are the generative model’s parameters. This process induces the distribution", + "type": "text" + }, + { + "bbox": [ + 428, + 173, + 518, + 186 + ], + "score": 0.92, + "content": "p _ { \\theta } ( x ) = \\mathbb { E } _ { p ( z ) } p _ { \\theta } ( x | z )", + "type": "inline_equation" + }, + { + "bbox": [ + 518, + 172, + 522, + 186 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 89, + 184, + 522, + 198 + ], + "spans": [ + { + "bbox": [ + 89, + 184, + 158, + 198 + ], + "score": 1.0, + "content": "The distribution", + "type": "text" + }, + { + "bbox": [ + 158, + 185, + 190, + 197 + ], + "score": 0.93, + "content": "p _ { \\theta } ( x | z )", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 184, + 393, + 198 + ], + "score": 1.0, + "content": "is modeled by a neural network with parameters", + "type": "text" + }, + { + "bbox": [ + 393, + 185, + 399, + 195 + ], + "score": 0.53, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 400, + 184, + 406, + 198 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 406, + 185, + 425, + 197 + ], + "score": 0.86, + "content": "p ( z )", + "type": "inline_equation" + }, + { + "bbox": [ + 425, + 184, + 522, + 198 + ], + "score": 1.0, + "content": "is a standard Gaussian", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 89, + 195, + 141, + 208 + ], + "spans": [ + { + "bbox": [ + 89, + 195, + 141, + 208 + ], + "score": 1.0, + "content": "distribution.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 4.5, + "bbox_fs": [ + 89, + 139, + 522, + 208 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 212, + 520, + 258 + ], + "lines": [ + { + "bbox": [ + 88, + 210, + 523, + 229 + ], + "spans": [ + { + "bbox": [ + 88, + 210, + 155, + 229 + ], + "score": 1.0, + "content": "The parameters", + "type": "text" + }, + { + "bbox": [ + 155, + 214, + 162, + 223 + ], + "score": 0.79, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 162, + 210, + 431, + 229 + ], + "score": 1.0, + "content": "are tuned by maximizing the likelihood of the training data points", + "type": "text" + }, + { + "bbox": [ + 431, + 212, + 464, + 225 + ], + "score": 0.93, + "content": "\\{ x _ { i } \\} _ { i = 1 } ^ { N }", + "type": "inline_equation" + }, + { + "bbox": [ + 464, + 210, + 523, + 229 + ], + "score": 1.0, + "content": "from the true", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 89, + 222, + 522, + 237 + ], + "spans": [ + { + "bbox": [ + 89, + 222, + 159, + 237 + ], + "score": 1.0, + "content": "data distribution", + "type": "text" + }, + { + "bbox": [ + 160, + 224, + 184, + 236 + ], + "score": 0.92, + "content": "p _ { d } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 222, + 522, + 237 + ], + "score": 1.0, + "content": ". In general, this optimization problem is challenging due to intractable posterior", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 89, + 235, + 522, + 249 + ], + "spans": [ + { + "bbox": [ + 89, + 235, + 522, + 249 + ], + "score": 1.0, + "content": "inference. However, a variational lower bound can be optimized efficiently using backpropagation and", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 90, + 246, + 202, + 258 + ], + "spans": [ + { + "bbox": [ + 90, + 246, + 202, + 258 + ], + "score": 1.0, + "content": "stochastic gradient descent:", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 9.5, + "bbox_fs": [ + 88, + 210, + 523, + 258 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 101, + 266, + 499, + 311 + ], + "lines": [ + { + "bbox": [ + 101, + 266, + 499, + 311 + ], + "spans": [ + { + "bbox": [ + 101, + 266, + 499, + 311 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\log p _ { \\theta } ( x ) = \\mathbb { E } _ { q _ { \\phi } ( z | x ) } \\log \\frac { p _ { \\theta } ( x , z ) } { q _ { \\phi } ( z | x ) } + D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( z | x ) \\| p ( z | x , \\theta ) \\big ) } \\\\ & { \\qquad \\quad \\ge \\mathbb { E } _ { q _ { \\phi } ( z | x ) } \\log p _ { \\theta } ( x | z ) - D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( z | x ) \\| p ( z ) \\big ) = L _ { V A E } \\big ( x ; \\theta , \\phi \\big ) } \\end{array}", + "type": "interline_equation", + "image_path": "fed675d9452648daab0a6ca45636d1aac44900502c485323a4dc8122dfe3f9fe.jpg" + } + ] + } + ], + "index": 13, + "virtual_lines": [ + { + "bbox": [ + 101, + 266, + 499, + 281.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 101, + 281.0, + 499, + 296.0 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 101, + 296.0, + 499, + 311.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "text", + "bbox": [ + 91, + 314, + 522, + 381 + ], + "lines": [ + { + "bbox": [ + 88, + 313, + 522, + 328 + ], + "spans": [ + { + "bbox": [ + 88, + 313, + 113, + 328 + ], + "score": 1.0, + "content": "Here", + "type": "text" + }, + { + "bbox": [ + 113, + 315, + 145, + 327 + ], + "score": 0.93, + "content": "q _ { \\phi } ( z | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 313, + 457, + 328 + ], + "score": 1.0, + "content": "is a proposal distribution parameterized by neural network with parameters", + "type": "text" + }, + { + "bbox": [ + 457, + 315, + 465, + 326 + ], + "score": 0.86, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 313, + 522, + 328 + ], + "score": 1.0, + "content": "that approxi-", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 89, + 325, + 522, + 338 + ], + "spans": [ + { + "bbox": [ + 89, + 325, + 170, + 338 + ], + "score": 1.0, + "content": "mates the posterior", + "type": "text" + }, + { + "bbox": [ + 171, + 327, + 206, + 338 + ], + "score": 0.91, + "content": "p ( z | x , \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 325, + 522, + 338 + ], + "score": 1.0, + "content": ". Usually this distribution is Gaussian with a diagonal covariance matrix. The", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 90, + 336, + 521, + 349 + ], + "spans": [ + { + "bbox": [ + 90, + 337, + 117, + 349 + ], + "score": 1.0, + "content": "closer", + "type": "text" + }, + { + "bbox": [ + 117, + 337, + 149, + 349 + ], + "score": 0.92, + "content": "q _ { \\phi } ( z | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 337, + 160, + 349 + ], + "score": 1.0, + "content": "to", + "type": "text" + }, + { + "bbox": [ + 161, + 338, + 197, + 349 + ], + "score": 0.92, + "content": "p ( z | x , \\theta )", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 337, + 342, + 349 + ], + "score": 1.0, + "content": ", the tighter variational lower bound", + "type": "text" + }, + { + "bbox": [ + 342, + 336, + 392, + 349 + ], + "score": 0.92, + "content": "L _ { V A E } ( \\theta , \\phi )", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 337, + 521, + 349 + ], + "score": 1.0, + "content": ". To compute the gradient of the", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 89, + 347, + 522, + 360 + ], + "spans": [ + { + "bbox": [ + 89, + 347, + 254, + 360 + ], + "score": 1.0, + "content": "variational lower bound with respect to", + "type": "text" + }, + { + "bbox": [ + 254, + 348, + 261, + 359 + ], + "score": 0.83, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 262, + 347, + 402, + 360 + ], + "score": 1.0, + "content": ", reparameterization trick is used:", + "type": "text" + }, + { + "bbox": [ + 402, + 347, + 492, + 360 + ], + "score": 0.92, + "content": "z = \\bar { \\mu } _ { \\phi } ( x ) + \\bar { \\varepsilon } \\sigma _ { \\phi } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 347, + 522, + 360 + ], + "score": 1.0, + "content": "where", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 91, + 358, + 522, + 371 + ], + "spans": [ + { + "bbox": [ + 91, + 358, + 142, + 371 + ], + "score": 0.92, + "content": "\\varepsilon \\sim \\mathcal { N } ( 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 142, + 358, + 161, + 371 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 161, + 360, + 173, + 371 + ], + "score": 0.84, + "content": "\\mu _ { \\phi }", + "type": "inline_equation" + }, + { + "bbox": [ + 173, + 358, + 192, + 371 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 192, + 360, + 204, + 371 + ], + "score": 0.84, + "content": "\\sigma _ { \\phi }", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 358, + 522, + 371 + ], + "score": 1.0, + "content": "are deterministic functions parameterized by neural networks. So the gradient", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 89, + 369, + 517, + 382 + ], + "spans": [ + { + "bbox": [ + 89, + 369, + 517, + 382 + ], + "score": 1.0, + "content": "can be estimated using Monte-Carlo method for the first term and computing the second term analytically:", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 17.5, + "bbox_fs": [ + 88, + 313, + 522, + 382 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 128, + 387, + 482, + 413 + ], + "lines": [ + { + "bbox": [ + 128, + 387, + 482, + 413 + ], + "spans": [ + { + "bbox": [ + 128, + 387, + 482, + 413 + ], + "score": 0.92, + "content": "\\frac { \\partial L _ { V A E } ( x ; \\theta , \\phi ) } { \\partial \\phi } = \\mathbb { E } _ { \\varepsilon \\sim \\mathcal { N } ( 0 , I ) } \\frac { \\partial } { \\partial \\phi } \\log p _ { \\theta } ( x | \\mu _ { \\phi } ( x ) + \\varepsilon \\sigma _ { \\phi } ( x ) ) - \\frac { \\partial } { \\partial \\phi } D _ { \\mathrm { K L } } ( q _ { \\phi } ( z | x ) \\| p ( z ) ) .", + "type": "interline_equation", + "image_path": "392ee19314ad1169b0b4fd7d080fa9dd49328ced339e9d76af78ca28090587f9.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 128, + 387, + 482, + 413 + ], + "spans": [], + "index": 21 + } + ] + }, + { + "type": "text", + "bbox": [ + 90, + 418, + 449, + 430 + ], + "lines": [ + { + "bbox": [ + 89, + 418, + 446, + 432 + ], + "spans": [ + { + "bbox": [ + 89, + 418, + 103, + 432 + ], + "score": 1.0, + "content": "So", + "type": "text" + }, + { + "bbox": [ + 104, + 419, + 153, + 431 + ], + "score": 0.92, + "content": "L _ { V A E } ( \\theta , \\phi )", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 418, + 415, + 432 + ], + "score": 1.0, + "content": "can be optimized using stochastic gradient ascent with respect to", + "type": "text" + }, + { + "bbox": [ + 415, + 419, + 422, + 430 + ], + "score": 0.86, + "content": "\\phi", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 418, + 440, + 432 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 441, + 419, + 446, + 429 + ], + "score": 0.79, + "content": "\\theta", + "type": "inline_equation" + } + ], + "index": 22 + } + ], + "index": 22, + "bbox_fs": [ + 89, + 418, + 446, + 432 + ] + }, + { + "type": "title", + "bbox": [ + 91, + 444, + 308, + 456 + ], + "lines": [ + { + "bbox": [ + 89, + 443, + 309, + 457 + ], + "spans": [ + { + "bbox": [ + 89, + 443, + 309, + 457 + ], + "score": 1.0, + "content": "3.2 CONDITIONAL VARIATIONAL AUTOENCODER", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 90, + 464, + 521, + 520 + ], + "lines": [ + { + "bbox": [ + 90, + 465, + 522, + 478 + ], + "spans": [ + { + "bbox": [ + 90, + 465, + 522, + 478 + ], + "score": 1.0, + "content": "Conditional variational autoencoder (Sohn et al., 2015) (CVAE) approximates the conditional distribution", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 91, + 475, + 522, + 489 + ], + "spans": [ + { + "bbox": [ + 91, + 476, + 122, + 488 + ], + "score": 0.92, + "content": "p _ { d } ( x | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 123, + 475, + 363, + 489 + ], + "score": 1.0, + "content": ". It outperforms deterministic models when the distribution", + "type": "text" + }, + { + "bbox": [ + 364, + 476, + 396, + 488 + ], + "score": 0.93, + "content": "p _ { d } ( x | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 396, + 475, + 494, + 489 + ], + "score": 1.0, + "content": "is multi-modal (diverse", + "type": "text" + }, + { + "bbox": [ + 494, + 478, + 505, + 486 + ], + "score": 0.64, + "content": "x \\mathbf { s }", + "type": "inline_equation" + }, + { + "bbox": [ + 506, + 475, + 522, + 489 + ], + "score": 1.0, + "content": "are", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 88, + 486, + 522, + 500 + ], + "spans": [ + { + "bbox": [ + 88, + 486, + 178, + 500 + ], + "score": 1.0, + "content": "probable for the given", + "type": "text" + }, + { + "bbox": [ + 178, + 489, + 185, + 498 + ], + "score": 0.64, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 486, + 294, + 500 + ], + "score": 1.0, + "content": "). For example, assume that", + "type": "text" + }, + { + "bbox": [ + 294, + 488, + 301, + 497 + ], + "score": 0.8, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 301, + 486, + 522, + 500 + ], + "score": 1.0, + "content": "is a real-valued image. Then, a deterministic regression", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 88, + 497, + 522, + 510 + ], + "spans": [ + { + "bbox": [ + 88, + 497, + 405, + 510 + ], + "score": 1.0, + "content": "model with mean squared error loss would predict the average blurry value for", + "type": "text" + }, + { + "bbox": [ + 405, + 500, + 412, + 508 + ], + "score": 0.78, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 412, + 497, + 522, + 510 + ], + "score": 1.0, + "content": ". On the other hand, CVAE", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 88, + 508, + 428, + 522 + ], + "spans": [ + { + "bbox": [ + 88, + 508, + 190, + 522 + ], + "score": 1.0, + "content": "learns the distribution of", + "type": "text" + }, + { + "bbox": [ + 191, + 511, + 197, + 519 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 508, + 428, + 522 + ], + "score": 1.0, + "content": ", from which one can sample diverse and realistic objects.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26, + "bbox_fs": [ + 88, + 465, + 522, + 522 + ] + }, + { + "type": "text", + "bbox": [ + 87, + 525, + 520, + 549 + ], + "lines": [ + { + "bbox": [ + 90, + 524, + 520, + 538 + ], + "spans": [ + { + "bbox": [ + 90, + 524, + 520, + 538 + ], + "score": 1.0, + "content": "Variational lower bound for CVAE can be derived similarly to VAE by conditioning all considered distribu-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 89, + 535, + 136, + 551 + ], + "spans": [ + { + "bbox": [ + 89, + 535, + 125, + 551 + ], + "score": 1.0, + "content": "tions on", + "type": "text" + }, + { + "bbox": [ + 125, + 540, + 131, + 549 + ], + "score": 0.72, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 131, + 535, + 136, + 551 + ], + "score": 1.0, + "content": ":", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5, + "bbox_fs": [ + 89, + 524, + 520, + 551 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 114, + 554, + 497, + 569 + ], + "lines": [ + { + "bbox": [ + 114, + 554, + 497, + 569 + ], + "spans": [ + { + "bbox": [ + 114, + 554, + 497, + 569 + ], + "score": 0.91, + "content": "L _ { C V A E } ( x , y ; \\theta , \\psi , \\phi ) = \\mathbb { E } _ { q _ { \\phi } ( z | x , y ) } \\log p _ { \\theta } ( x | z , y ) - D _ { \\mathrm { K L } } ( q _ { \\phi } ( z | x , y ) | | p _ { \\psi } ( z | y ) ) \\leq \\log p _ { \\theta , \\psi } ( x | y )", + "type": "interline_equation", + "image_path": "1a364d873a354bd42b6ac3c1a473c1e2729b1b87d4dce0f0bfde7ab681819e08.jpg" + } + ] + } + ], + "index": 31, + "virtual_lines": [ + { + "bbox": [ + 114, + 554, + 497, + 569 + ], + "spans": [], + "index": 31 + } + ] + }, + { + "type": "text", + "bbox": [ + 91, + 579, + 522, + 613 + ], + "lines": [ + { + "bbox": [ + 89, + 578, + 521, + 592 + ], + "spans": [ + { + "bbox": [ + 89, + 578, + 521, + 592 + ], + "score": 1.0, + "content": "Similarly to VAE, this objective is optimized using the reparameterization trick. Note that the prior distribu-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 90, + 590, + 522, + 604 + ], + "spans": [ + { + "bbox": [ + 90, + 590, + 108, + 604 + ], + "score": 1.0, + "content": "tion", + "type": "text" + }, + { + "bbox": [ + 109, + 591, + 141, + 603 + ], + "score": 0.93, + "content": "p _ { \\psi } ( z | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 590, + 214, + 604 + ], + "score": 1.0, + "content": "is conditioned on", + "type": "text" + }, + { + "bbox": [ + 215, + 592, + 221, + 602 + ], + "score": 0.77, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 590, + 435, + 604 + ], + "score": 1.0, + "content": "and is modeled by a neural network with parameters", + "type": "text" + }, + { + "bbox": [ + 435, + 591, + 443, + 602 + ], + "score": 0.86, + "content": "\\psi", + "type": "inline_equation" + }, + { + "bbox": [ + 443, + 590, + 522, + 604 + ], + "score": 1.0, + "content": ". Thus, CVAE uses", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 89, + 600, + 325, + 614 + ], + "spans": [ + { + "bbox": [ + 89, + 600, + 325, + 614 + ], + "score": 1.0, + "content": "three trainable neural networks, while VAE only uses two.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33, + "bbox_fs": [ + 89, + 578, + 522, + 614 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 617, + 520, + 652 + ], + "lines": [ + { + "bbox": [ + 90, + 618, + 521, + 630 + ], + "spans": [ + { + "bbox": [ + 90, + 618, + 521, + 630 + ], + "score": 1.0, + "content": "Also authors propose such modifications of CVAE as Gaussian stochastic neural network and hybrid model.", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 89, + 628, + 522, + 641 + ], + "spans": [ + { + "bbox": [ + 89, + 628, + 522, + 641 + ], + "score": 1.0, + "content": "These modifications can be applied to our model as well. Nevertheless, we don’t use them, because of their", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 90, + 640, + 282, + 653 + ], + "spans": [ + { + "bbox": [ + 90, + 640, + 282, + 653 + ], + "score": 1.0, + "content": "disadvantage which is described in appendix C.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36, + "bbox_fs": [ + 89, + 618, + 522, + 653 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 90, + 94, + 441, + 108 + ], + "lines": [ + { + "bbox": [ + 88, + 93, + 442, + 109 + ], + "spans": [ + { + "bbox": [ + 88, + 93, + 442, + 109 + ], + "score": 1.0, + "content": "4 VARIATIONAL AUTOENCODER WITH ARBITRARY CONDITIONING", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 91, + 121, + 212, + 133 + ], + "lines": [ + { + "bbox": [ + 90, + 122, + 212, + 133 + ], + "spans": [ + { + "bbox": [ + 90, + 122, + 212, + 133 + ], + "score": 1.0, + "content": "4.1 PROBLEM STATEMENT", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 90, + 143, + 520, + 166 + ], + "lines": [ + { + "bbox": [ + 89, + 142, + 521, + 156 + ], + "spans": [ + { + "bbox": [ + 89, + 142, + 187, + 156 + ], + "score": 1.0, + "content": "Consider a distribution", + "type": "text" + }, + { + "bbox": [ + 188, + 144, + 212, + 156 + ], + "score": 0.93, + "content": "p _ { d } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 142, + 244, + 156 + ], + "score": 1.0, + "content": "over a", + "type": "text" + }, + { + "bbox": [ + 244, + 144, + 253, + 154 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 142, + 336, + 156 + ], + "score": 1.0, + "content": "-dimensional vector", + "type": "text" + }, + { + "bbox": [ + 337, + 146, + 344, + 154 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 142, + 521, + 156 + ], + "score": 1.0, + "content": "with real or categorical components. The", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 90, + 155, + 271, + 167 + ], + "spans": [ + { + "bbox": [ + 90, + 155, + 271, + 167 + ], + "score": 1.0, + "content": "components of the vector are called features.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + }, + { + "type": "text", + "bbox": [ + 90, + 171, + 522, + 205 + ], + "lines": [ + { + "bbox": [ + 90, + 171, + 521, + 184 + ], + "spans": [ + { + "bbox": [ + 90, + 171, + 163, + 184 + ], + "score": 1.0, + "content": "Let binary vector", + "type": "text" + }, + { + "bbox": [ + 163, + 171, + 213, + 184 + ], + "score": 0.93, + "content": "b \\in \\{ 0 , 1 \\} ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 171, + 521, + 184 + ], + "score": 1.0, + "content": "be the binary mask of unobserved features of the object. Then we describe", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 88, + 181, + 523, + 198 + ], + "spans": [ + { + "bbox": [ + 88, + 181, + 239, + 198 + ], + "score": 1.0, + "content": "the vector of unobserved features as", + "type": "text" + }, + { + "bbox": [ + 240, + 183, + 303, + 195 + ], + "score": 0.93, + "content": "x _ { b } = \\{ x _ { i : b _ { i } = 1 } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 181, + 366, + 198 + ], + "score": 1.0, + "content": ". For example,", + "type": "text" + }, + { + "bbox": [ + 366, + 183, + 471, + 195 + ], + "score": 0.91, + "content": "x _ { ( 0 , 1 , 1 , 0 , 1 ) } = ( x _ { 2 } , x _ { 3 } , x _ { 5 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 181, + 523, + 198 + ], + "score": 1.0, + "content": ". Using this", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 89, + 192, + 322, + 206 + ], + "spans": [ + { + "bbox": [ + 89, + 192, + 168, + 206 + ], + "score": 1.0, + "content": "notation we denote", + "type": "text" + }, + { + "bbox": [ + 169, + 195, + 190, + 205 + ], + "score": 0.89, + "content": "x _ { 1 - b }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 192, + 322, + 206 + ], + "score": 1.0, + "content": "as a vector of observed features.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 91, + 210, + 520, + 233 + ], + "lines": [ + { + "bbox": [ + 89, + 209, + 521, + 224 + ], + "spans": [ + { + "bbox": [ + 89, + 209, + 320, + 224 + ], + "score": 1.0, + "content": "Our goal is to build a model of the conditional distribution", + "type": "text" + }, + { + "bbox": [ + 320, + 210, + 459, + 223 + ], + "score": 0.92, + "content": "p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b ) \\approx p _ { d } ( x _ { b } | x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 209, + 521, + 224 + ], + "score": 1.0, + "content": "for an arbitrary", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 91, + 220, + 406, + 235 + ], + "spans": [ + { + "bbox": [ + 91, + 222, + 96, + 231 + ], + "score": 0.74, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 96, + 220, + 126, + 235 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 126, + 222, + 135, + 233 + ], + "score": 0.85, + "content": "\\psi", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 220, + 153, + 235 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 153, + 222, + 159, + 231 + ], + "score": 0.79, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 220, + 406, + 235 + ], + "score": 1.0, + "content": "are parameters that are used in our model at the testing stage.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5 + }, + { + "type": "text", + "bbox": [ + 90, + 237, + 521, + 316 + ], + "lines": [ + { + "bbox": [ + 89, + 237, + 522, + 252 + ], + "spans": [ + { + "bbox": [ + 89, + 237, + 212, + 252 + ], + "score": 1.0, + "content": "However, the true distribution", + "type": "text" + }, + { + "bbox": [ + 212, + 238, + 271, + 250 + ], + "score": 0.93, + "content": "p _ { d } ( x _ { b } | x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 237, + 463, + 252 + ], + "score": 1.0, + "content": "is intractable without strong assumptions about", + "type": "text" + }, + { + "bbox": [ + 463, + 238, + 487, + 250 + ], + "score": 0.91, + "content": "p _ { d } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 488, + 237, + 522, + 252 + ], + "score": 1.0, + "content": ". There-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 89, + 249, + 522, + 263 + ], + "spans": [ + { + "bbox": [ + 89, + 249, + 154, + 263 + ], + "score": 1.0, + "content": "fore, our model", + "type": "text" + }, + { + "bbox": [ + 155, + 249, + 221, + 262 + ], + "score": 0.92, + "content": "p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 249, + 352, + 263 + ], + "score": 1.0, + "content": "has to be more precise for some", + "type": "text" + }, + { + "bbox": [ + 352, + 250, + 358, + 259 + ], + "score": 0.69, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 249, + 522, + 263 + ], + "score": 1.0, + "content": "and less precise for others. To formalize", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 89, + 260, + 521, + 273 + ], + "spans": [ + { + "bbox": [ + 89, + 260, + 417, + 273 + ], + "score": 1.0, + "content": "our requirements about the accuracy of our model we introduce the distribution", + "type": "text" + }, + { + "bbox": [ + 417, + 260, + 435, + 272 + ], + "score": 0.91, + "content": "p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 436, + 260, + 521, + 273 + ], + "score": 1.0, + "content": "over different unob-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 89, + 271, + 521, + 284 + ], + "spans": [ + { + "bbox": [ + 89, + 271, + 250, + 284 + ], + "score": 1.0, + "content": "served feature masks. The distribution", + "type": "text" + }, + { + "bbox": [ + 250, + 271, + 268, + 283 + ], + "score": 0.91, + "content": "p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 271, + 521, + 284 + ], + "score": 1.0, + "content": "is arbitrary and may be defined by the user depending on the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 88, + 280, + 522, + 297 + ], + "spans": [ + { + "bbox": [ + 88, + 280, + 300, + 297 + ], + "score": 1.0, + "content": "problem. Generally it should have full support over", + "type": "text" + }, + { + "bbox": [ + 300, + 282, + 333, + 294 + ], + "score": 0.94, + "content": "\\{ 0 , 1 \\} ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 280, + 363, + 297 + ], + "score": 1.0, + "content": "so that", + "type": "text" + }, + { + "bbox": [ + 364, + 282, + 430, + 294 + ], + "score": 0.92, + "content": "p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 280, + 522, + 297 + ], + "score": 1.0, + "content": "can evaluate arbitrary", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 89, + 292, + 522, + 306 + ], + "spans": [ + { + "bbox": [ + 89, + 292, + 522, + 306 + ], + "score": 1.0, + "content": "conditioning. Nevertheless, it is not necessary if the model is used for specific kinds of conditioning (as we", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 89, + 304, + 164, + 317 + ], + "spans": [ + { + "bbox": [ + 89, + 304, + 164, + 317 + ], + "score": 1.0, + "content": "do in section 5.2).", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 90, + 320, + 454, + 333 + ], + "lines": [ + { + "bbox": [ + 90, + 320, + 455, + 334 + ], + "spans": [ + { + "bbox": [ + 90, + 320, + 117, + 334 + ], + "score": 1.0, + "content": "Using", + "type": "text" + }, + { + "bbox": [ + 117, + 321, + 135, + 333 + ], + "score": 0.9, + "content": "p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 320, + 455, + 334 + ], + "score": 1.0, + "content": "we can introduce the following log-likelihood objective function for the model:", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "interline_equation", + "bbox": [ + 229, + 341, + 383, + 360 + ], + "lines": [ + { + "bbox": [ + 229, + 341, + 383, + 360 + ], + "spans": [ + { + "bbox": [ + 229, + 341, + 383, + 360 + ], + "score": 0.91, + "content": "\\operatorname* { m a x } _ { \\psi , \\theta } \\mathbb { E } _ { p _ { d } ( \\boldsymbol { x } ) } \\mathbb { E } _ { p ( \\boldsymbol { b } ) } \\log p _ { \\psi , \\theta } \\big ( x _ { b } | \\boldsymbol { x } _ { 1 - \\boldsymbol { b } } , \\boldsymbol { b } \\big )", + "type": "interline_equation", + "image_path": "6eacc0f28bc2e238ba172a9f915eb5066998bc34e8c21e5ae8d8543613577821.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 229, + 341, + 383, + 360 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 90, + 369, + 521, + 393 + ], + "lines": [ + { + "bbox": [ + 90, + 369, + 521, + 382 + ], + "spans": [ + { + "bbox": [ + 90, + 369, + 353, + 382 + ], + "score": 1.0, + "content": "The special cases of the objective (4) are variational autoencoder", + "type": "text" + }, + { + "bbox": [ + 354, + 370, + 454, + 382 + ], + "score": 0.9, + "content": "( b _ { i } = 1 \\forall i \\in \\{ 1 , \\ldots , D \\} )", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 369, + 521, + 382 + ], + "score": 1.0, + "content": "and conditional", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 90, + 380, + 247, + 393 + ], + "spans": [ + { + "bbox": [ + 90, + 380, + 189, + 393 + ], + "score": 1.0, + "content": "variational autoencoder", + "type": "text" + }, + { + "bbox": [ + 190, + 382, + 195, + 391 + ], + "score": 0.7, + "content": "^ { \\textit { b } }", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 380, + 247, + 393 + ], + "score": 1.0, + "content": "is constant).", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5 + }, + { + "type": "title", + "bbox": [ + 91, + 410, + 210, + 421 + ], + "lines": [ + { + "bbox": [ + 90, + 410, + 210, + 422 + ], + "spans": [ + { + "bbox": [ + 90, + 410, + 210, + 422 + ], + "score": 1.0, + "content": "4.2 MODEL DESCRIPTION", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 90, + 432, + 521, + 476 + ], + "lines": [ + { + "bbox": [ + 89, + 431, + 521, + 445 + ], + "spans": [ + { + "bbox": [ + 89, + 431, + 521, + 445 + ], + "score": 1.0, + "content": "The generative process of our model is similar to the generative process of CVAE: for each ob-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 88, + 442, + 522, + 456 + ], + "spans": [ + { + "bbox": [ + 88, + 442, + 190, + 456 + ], + "score": 1.0, + "content": "ject firstly we generate", + "type": "text" + }, + { + "bbox": [ + 191, + 443, + 272, + 455 + ], + "score": 0.91, + "content": "z \\sim p _ { \\psi } ( z | x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 442, + 522, + 456 + ], + "score": 1.0, + "content": "using prior network, and then sample unobserved features", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 91, + 453, + 522, + 468 + ], + "spans": [ + { + "bbox": [ + 91, + 454, + 190, + 466 + ], + "score": 0.91, + "content": "x _ { b } ~ \\sim ~ p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 453, + 522, + 468 + ], + "score": 1.0, + "content": "using generative network. This process induces the following model distribution", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 89, + 464, + 196, + 477 + ], + "spans": [ + { + "bbox": [ + 89, + 464, + 196, + 477 + ], + "score": 1.0, + "content": "over unobserved features:", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22.5 + }, + { + "type": "interline_equation", + "bbox": [ + 199, + 485, + 412, + 500 + ], + "lines": [ + { + "bbox": [ + 199, + 485, + 412, + 500 + ], + "spans": [ + { + "bbox": [ + 199, + 485, + 412, + 500 + ], + "score": 0.86, + "content": "p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b ) = \\mathbb { E } _ { z \\sim p _ { \\psi } ( z | x _ { 1 - b } , b ) } p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b )", + "type": "interline_equation", + "image_path": "1010c9b588784d89851c33cee48ac5e4d9b7fc2302b1356bbdce195c382c31ce.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 199, + 485, + 412, + 500 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 90, + 515, + 521, + 586 + ], + "lines": [ + { + "bbox": [ + 90, + 514, + 522, + 529 + ], + "spans": [ + { + "bbox": [ + 90, + 514, + 125, + 529 + ], + "score": 1.0, + "content": "We use", + "type": "text" + }, + { + "bbox": [ + 126, + 515, + 164, + 527 + ], + "score": 0.91, + "content": "z ~ \\in ~ \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 514, + 280, + 529 + ], + "score": 1.0, + "content": ", and Gaussian distribution", + "type": "text" + }, + { + "bbox": [ + 280, + 518, + 293, + 528 + ], + "score": 0.82, + "content": "p _ { \\psi }", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 514, + 318, + 529 + ], + "score": 1.0, + "content": "over", + "type": "text" + }, + { + "bbox": [ + 318, + 519, + 324, + 526 + ], + "score": 0.74, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 514, + 522, + 529 + ], + "score": 1.0, + "content": ", with parameters from a neural network with", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 90, + 527, + 521, + 541 + ], + "spans": [ + { + "bbox": [ + 90, + 527, + 125, + 540 + ], + "score": 1.0, + "content": "weights", + "type": "text" + }, + { + "bbox": [ + 125, + 528, + 133, + 540 + ], + "score": 0.75, + "content": "\\psi", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 527, + 138, + 540 + ], + "score": 1.0, + "content": ":", + "type": "text" + }, + { + "bbox": [ + 139, + 527, + 356, + 541 + ], + "score": 0.87, + "content": "p _ { \\psi } ( z | x _ { 1 - b } , b , \\psi ) = \\mathcal { N } ( z | \\mu _ { \\psi } ( x _ { 1 - b } , b ) , \\sigma _ { \\psi } ^ { 2 } ( x _ { 1 - b } , b ) I )", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 527, + 521, + 540 + ], + "score": 1.0, + "content": ". The real-valued components of distri-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 90, + 540, + 520, + 553 + ], + "spans": [ + { + "bbox": [ + 90, + 540, + 118, + 553 + ], + "score": 1.0, + "content": "bution", + "type": "text" + }, + { + "bbox": [ + 118, + 541, + 187, + 552 + ], + "score": 0.92, + "content": "p _ { \\theta } ( x _ { b } | \\boldsymbol { z } , x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 540, + 387, + 553 + ], + "score": 1.0, + "content": "are defined likewise. Each categorical component", + "type": "text" + }, + { + "bbox": [ + 387, + 541, + 392, + 550 + ], + "score": 0.72, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 540, + 452, + 553 + ], + "score": 1.0, + "content": "of distribution", + "type": "text" + }, + { + "bbox": [ + 452, + 540, + 520, + 552 + ], + "score": 0.92, + "content": "p _ { \\theta } ( x _ { i } | \\boldsymbol { z } , x _ { 1 - b } , b )", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 88, + 551, + 522, + 565 + ], + "spans": [ + { + "bbox": [ + 88, + 551, + 215, + 565 + ], + "score": 1.0, + "content": "is parameterized by a function", + "type": "text" + }, + { + "bbox": [ + 215, + 552, + 278, + 563 + ], + "score": 0.91, + "content": "w _ { i , \\theta } ( z , x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 551, + 522, + 565 + ], + "score": 1.0, + "content": ", whose outputs are logits of probabilities for each category:", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 91, + 561, + 521, + 576 + ], + "spans": [ + { + "bbox": [ + 91, + 563, + 245, + 574 + ], + "score": 0.89, + "content": "x _ { i } \\sim \\mathrm { C a t } [ \\mathrm { S o f t m a x } ( w _ { i , \\theta } ( z , x _ { 1 - b } , b ) ) ]", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 561, + 442, + 576 + ], + "score": 1.0, + "content": ". Therefore the components of the latent vector", + "type": "text" + }, + { + "bbox": [ + 442, + 565, + 449, + 572 + ], + "score": 0.75, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 561, + 521, + 576 + ], + "score": 1.0, + "content": "are conditionally", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 89, + 573, + 522, + 587 + ], + "spans": [ + { + "bbox": [ + 89, + 573, + 165, + 587 + ], + "score": 1.0, + "content": "independent given", + "type": "text" + }, + { + "bbox": [ + 165, + 576, + 186, + 585 + ], + "score": 0.89, + "content": "x _ { 1 - b }", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 573, + 204, + 587 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 205, + 575, + 210, + 583 + ], + "score": 0.77, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 573, + 306, + 587 + ], + "score": 1.0, + "content": ", and the components of", + "type": "text" + }, + { + "bbox": [ + 306, + 575, + 317, + 585 + ], + "score": 0.85, + "content": "x _ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 573, + 462, + 587 + ], + "score": 1.0, + "content": "are conditionally independent given", + "type": "text" + }, + { + "bbox": [ + 462, + 575, + 494, + 585 + ], + "score": 0.51, + "content": "z , x _ { 1 - b }", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 573, + 512, + 587 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 512, + 574, + 518, + 583 + ], + "score": 0.75, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 518, + 573, + 522, + 587 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 91, + 590, + 521, + 635 + ], + "lines": [ + { + "bbox": [ + 90, + 590, + 522, + 604 + ], + "spans": [ + { + "bbox": [ + 90, + 590, + 147, + 604 + ], + "score": 1.0, + "content": "The variables", + "type": "text" + }, + { + "bbox": [ + 147, + 592, + 158, + 601 + ], + "score": 0.84, + "content": "x _ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 590, + 177, + 604 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 177, + 592, + 198, + 602 + ], + "score": 0.88, + "content": "x _ { 1 - b }", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 590, + 350, + 604 + ], + "score": 1.0, + "content": "have variable length that depends on", + "type": "text" + }, + { + "bbox": [ + 350, + 591, + 355, + 600 + ], + "score": 0.69, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 590, + 522, + 604 + ], + "score": 1.0, + "content": ". So in order to use architectures such as", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 90, + 601, + 522, + 614 + ], + "spans": [ + { + "bbox": [ + 90, + 601, + 376, + 614 + ], + "score": 1.0, + "content": "multi-layer perceptron and convolutional neural network we consider", + "type": "text" + }, + { + "bbox": [ + 376, + 601, + 461, + 613 + ], + "score": 0.92, + "content": "x _ { 1 - b } = x \\circ ( 1 - b )", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 601, + 490, + 614 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 491, + 603, + 497, + 611 + ], + "score": 0.71, + "content": "\\circ", + "type": "inline_equation" + }, + { + "bbox": [ + 498, + 601, + 522, + 614 + ], + "score": 1.0, + "content": "is an", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 89, + 611, + 522, + 626 + ], + "spans": [ + { + "bbox": [ + 89, + 611, + 274, + 626 + ], + "score": 1.0, + "content": "element-wise product. So in implementation", + "type": "text" + }, + { + "bbox": [ + 274, + 613, + 295, + 624 + ], + "score": 0.86, + "content": "x _ { 1 - b }", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 611, + 522, + 626 + ], + "score": 1.0, + "content": "has fixed length. The output of the generative network", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 90, + 623, + 440, + 636 + ], + "spans": [ + { + "bbox": [ + 90, + 623, + 440, + 636 + ], + "score": 1.0, + "content": "also has a fixed length, but we use only unobserved components to compute likelihood.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33.5 + }, + { + "type": "text", + "bbox": [ + 90, + 640, + 354, + 652 + ], + "lines": [ + { + "bbox": [ + 90, + 639, + 356, + 653 + ], + "spans": [ + { + "bbox": [ + 90, + 639, + 356, + 653 + ], + "score": 1.0, + "content": "The theoretical analysis of the model is available in appendix B.1.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36 + } + ], + "page_idx": 3, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 39, + 277, + 51 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 286, + 671, + 293, + 680 + ], + "lines": [ + { + "bbox": [ + 285, + 670, + 294, + 682 + ], + "spans": [ + { + "bbox": [ + 285, + 670, + 294, + 682 + ], + "score": 1.0, + "content": "4", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 90, + 94, + 441, + 108 + ], + "lines": [ + { + "bbox": [ + 88, + 93, + 442, + 109 + ], + "spans": [ + { + "bbox": [ + 88, + 93, + 442, + 109 + ], + "score": 1.0, + "content": "4 VARIATIONAL AUTOENCODER WITH ARBITRARY CONDITIONING", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 91, + 121, + 212, + 133 + ], + "lines": [ + { + "bbox": [ + 90, + 122, + 212, + 133 + ], + "spans": [ + { + "bbox": [ + 90, + 122, + 212, + 133 + ], + "score": 1.0, + "content": "4.1 PROBLEM STATEMENT", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 90, + 143, + 520, + 166 + ], + "lines": [ + { + "bbox": [ + 89, + 142, + 521, + 156 + ], + "spans": [ + { + "bbox": [ + 89, + 142, + 187, + 156 + ], + "score": 1.0, + "content": "Consider a distribution", + "type": "text" + }, + { + "bbox": [ + 188, + 144, + 212, + 156 + ], + "score": 0.93, + "content": "p _ { d } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 142, + 244, + 156 + ], + "score": 1.0, + "content": "over a", + "type": "text" + }, + { + "bbox": [ + 244, + 144, + 253, + 154 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 142, + 336, + 156 + ], + "score": 1.0, + "content": "-dimensional vector", + "type": "text" + }, + { + "bbox": [ + 337, + 146, + 344, + 154 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 344, + 142, + 521, + 156 + ], + "score": 1.0, + "content": "with real or categorical components. The", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 90, + 155, + 271, + 167 + ], + "spans": [ + { + "bbox": [ + 90, + 155, + 271, + 167 + ], + "score": 1.0, + "content": "components of the vector are called features.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5, + "bbox_fs": [ + 89, + 142, + 521, + 167 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 171, + 522, + 205 + ], + "lines": [ + { + "bbox": [ + 90, + 171, + 521, + 184 + ], + "spans": [ + { + "bbox": [ + 90, + 171, + 163, + 184 + ], + "score": 1.0, + "content": "Let binary vector", + "type": "text" + }, + { + "bbox": [ + 163, + 171, + 213, + 184 + ], + "score": 0.93, + "content": "b \\in \\{ 0 , 1 \\} ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 213, + 171, + 521, + 184 + ], + "score": 1.0, + "content": "be the binary mask of unobserved features of the object. Then we describe", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 88, + 181, + 523, + 198 + ], + "spans": [ + { + "bbox": [ + 88, + 181, + 239, + 198 + ], + "score": 1.0, + "content": "the vector of unobserved features as", + "type": "text" + }, + { + "bbox": [ + 240, + 183, + 303, + 195 + ], + "score": 0.93, + "content": "x _ { b } = \\{ x _ { i : b _ { i } = 1 } \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 181, + 366, + 198 + ], + "score": 1.0, + "content": ". For example,", + "type": "text" + }, + { + "bbox": [ + 366, + 183, + 471, + 195 + ], + "score": 0.91, + "content": "x _ { ( 0 , 1 , 1 , 0 , 1 ) } = ( x _ { 2 } , x _ { 3 } , x _ { 5 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 471, + 181, + 523, + 198 + ], + "score": 1.0, + "content": ". Using this", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 89, + 192, + 322, + 206 + ], + "spans": [ + { + "bbox": [ + 89, + 192, + 168, + 206 + ], + "score": 1.0, + "content": "notation we denote", + "type": "text" + }, + { + "bbox": [ + 169, + 195, + 190, + 205 + ], + "score": 0.89, + "content": "x _ { 1 - b }", + "type": "inline_equation" + }, + { + "bbox": [ + 190, + 192, + 322, + 206 + ], + "score": 1.0, + "content": "as a vector of observed features.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5, + "bbox_fs": [ + 88, + 171, + 523, + 206 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 210, + 520, + 233 + ], + "lines": [ + { + "bbox": [ + 89, + 209, + 521, + 224 + ], + "spans": [ + { + "bbox": [ + 89, + 209, + 320, + 224 + ], + "score": 1.0, + "content": "Our goal is to build a model of the conditional distribution", + "type": "text" + }, + { + "bbox": [ + 320, + 210, + 459, + 223 + ], + "score": 0.92, + "content": "p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b ) \\approx p _ { d } ( x _ { b } | x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 459, + 209, + 521, + 224 + ], + "score": 1.0, + "content": "for an arbitrary", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 91, + 220, + 406, + 235 + ], + "spans": [ + { + "bbox": [ + 91, + 222, + 96, + 231 + ], + "score": 0.74, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 96, + 220, + 126, + 235 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 126, + 222, + 135, + 233 + ], + "score": 0.85, + "content": "\\psi", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 220, + 153, + 235 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 153, + 222, + 159, + 231 + ], + "score": 0.79, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 220, + 406, + 235 + ], + "score": 1.0, + "content": "are parameters that are used in our model at the testing stage.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 7.5, + "bbox_fs": [ + 89, + 209, + 521, + 235 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 237, + 521, + 316 + ], + "lines": [ + { + "bbox": [ + 89, + 237, + 522, + 252 + ], + "spans": [ + { + "bbox": [ + 89, + 237, + 212, + 252 + ], + "score": 1.0, + "content": "However, the true distribution", + "type": "text" + }, + { + "bbox": [ + 212, + 238, + 271, + 250 + ], + "score": 0.93, + "content": "p _ { d } ( x _ { b } | x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 237, + 463, + 252 + ], + "score": 1.0, + "content": "is intractable without strong assumptions about", + "type": "text" + }, + { + "bbox": [ + 463, + 238, + 487, + 250 + ], + "score": 0.91, + "content": "p _ { d } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 488, + 237, + 522, + 252 + ], + "score": 1.0, + "content": ". There-", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 89, + 249, + 522, + 263 + ], + "spans": [ + { + "bbox": [ + 89, + 249, + 154, + 263 + ], + "score": 1.0, + "content": "fore, our model", + "type": "text" + }, + { + "bbox": [ + 155, + 249, + 221, + 262 + ], + "score": 0.92, + "content": "p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 249, + 352, + 263 + ], + "score": 1.0, + "content": "has to be more precise for some", + "type": "text" + }, + { + "bbox": [ + 352, + 250, + 358, + 259 + ], + "score": 0.69, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 249, + 522, + 263 + ], + "score": 1.0, + "content": "and less precise for others. To formalize", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 89, + 260, + 521, + 273 + ], + "spans": [ + { + "bbox": [ + 89, + 260, + 417, + 273 + ], + "score": 1.0, + "content": "our requirements about the accuracy of our model we introduce the distribution", + "type": "text" + }, + { + "bbox": [ + 417, + 260, + 435, + 272 + ], + "score": 0.91, + "content": "p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 436, + 260, + 521, + 273 + ], + "score": 1.0, + "content": "over different unob-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 89, + 271, + 521, + 284 + ], + "spans": [ + { + "bbox": [ + 89, + 271, + 250, + 284 + ], + "score": 1.0, + "content": "served feature masks. The distribution", + "type": "text" + }, + { + "bbox": [ + 250, + 271, + 268, + 283 + ], + "score": 0.91, + "content": "p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 268, + 271, + 521, + 284 + ], + "score": 1.0, + "content": "is arbitrary and may be defined by the user depending on the", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 88, + 280, + 522, + 297 + ], + "spans": [ + { + "bbox": [ + 88, + 280, + 300, + 297 + ], + "score": 1.0, + "content": "problem. Generally it should have full support over", + "type": "text" + }, + { + "bbox": [ + 300, + 282, + 333, + 294 + ], + "score": 0.94, + "content": "\\{ 0 , 1 \\} ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 333, + 280, + 363, + 297 + ], + "score": 1.0, + "content": "so that", + "type": "text" + }, + { + "bbox": [ + 364, + 282, + 430, + 294 + ], + "score": 0.92, + "content": "p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 280, + 522, + 297 + ], + "score": 1.0, + "content": "can evaluate arbitrary", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 89, + 292, + 522, + 306 + ], + "spans": [ + { + "bbox": [ + 89, + 292, + 522, + 306 + ], + "score": 1.0, + "content": "conditioning. Nevertheless, it is not necessary if the model is used for specific kinds of conditioning (as we", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 89, + 304, + 164, + 317 + ], + "spans": [ + { + "bbox": [ + 89, + 304, + 164, + 317 + ], + "score": 1.0, + "content": "do in section 5.2).", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 12, + "bbox_fs": [ + 88, + 237, + 522, + 317 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 320, + 454, + 333 + ], + "lines": [ + { + "bbox": [ + 90, + 320, + 455, + 334 + ], + "spans": [ + { + "bbox": [ + 90, + 320, + 117, + 334 + ], + "score": 1.0, + "content": "Using", + "type": "text" + }, + { + "bbox": [ + 117, + 321, + 135, + 333 + ], + "score": 0.9, + "content": "p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 320, + 455, + 334 + ], + "score": 1.0, + "content": "we can introduce the following log-likelihood objective function for the model:", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16, + "bbox_fs": [ + 90, + 320, + 455, + 334 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 229, + 341, + 383, + 360 + ], + "lines": [ + { + "bbox": [ + 229, + 341, + 383, + 360 + ], + "spans": [ + { + "bbox": [ + 229, + 341, + 383, + 360 + ], + "score": 0.91, + "content": "\\operatorname* { m a x } _ { \\psi , \\theta } \\mathbb { E } _ { p _ { d } ( \\boldsymbol { x } ) } \\mathbb { E } _ { p ( \\boldsymbol { b } ) } \\log p _ { \\psi , \\theta } \\big ( x _ { b } | \\boldsymbol { x } _ { 1 - \\boldsymbol { b } } , \\boldsymbol { b } \\big )", + "type": "interline_equation", + "image_path": "6eacc0f28bc2e238ba172a9f915eb5066998bc34e8c21e5ae8d8543613577821.jpg" + } + ] + } + ], + "index": 17, + "virtual_lines": [ + { + "bbox": [ + 229, + 341, + 383, + 360 + ], + "spans": [], + "index": 17 + } + ] + }, + { + "type": "text", + "bbox": [ + 90, + 369, + 521, + 393 + ], + "lines": [ + { + "bbox": [ + 90, + 369, + 521, + 382 + ], + "spans": [ + { + "bbox": [ + 90, + 369, + 353, + 382 + ], + "score": 1.0, + "content": "The special cases of the objective (4) are variational autoencoder", + "type": "text" + }, + { + "bbox": [ + 354, + 370, + 454, + 382 + ], + "score": 0.9, + "content": "( b _ { i } = 1 \\forall i \\in \\{ 1 , \\ldots , D \\} )", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 369, + 521, + 382 + ], + "score": 1.0, + "content": "and conditional", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 90, + 380, + 247, + 393 + ], + "spans": [ + { + "bbox": [ + 90, + 380, + 189, + 393 + ], + "score": 1.0, + "content": "variational autoencoder", + "type": "text" + }, + { + "bbox": [ + 190, + 382, + 195, + 391 + ], + "score": 0.7, + "content": "^ { \\textit { b } }", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 380, + 247, + 393 + ], + "score": 1.0, + "content": "is constant).", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18.5, + "bbox_fs": [ + 90, + 369, + 521, + 393 + ] + }, + { + "type": "title", + "bbox": [ + 91, + 410, + 210, + 421 + ], + "lines": [ + { + "bbox": [ + 90, + 410, + 210, + 422 + ], + "spans": [ + { + "bbox": [ + 90, + 410, + 210, + 422 + ], + "score": 1.0, + "content": "4.2 MODEL DESCRIPTION", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 90, + 432, + 521, + 476 + ], + "lines": [ + { + "bbox": [ + 89, + 431, + 521, + 445 + ], + "spans": [ + { + "bbox": [ + 89, + 431, + 521, + 445 + ], + "score": 1.0, + "content": "The generative process of our model is similar to the generative process of CVAE: for each ob-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 88, + 442, + 522, + 456 + ], + "spans": [ + { + "bbox": [ + 88, + 442, + 190, + 456 + ], + "score": 1.0, + "content": "ject firstly we generate", + "type": "text" + }, + { + "bbox": [ + 191, + 443, + 272, + 455 + ], + "score": 0.91, + "content": "z \\sim p _ { \\psi } ( z | x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 272, + 442, + 522, + 456 + ], + "score": 1.0, + "content": "using prior network, and then sample unobserved features", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 91, + 453, + 522, + 468 + ], + "spans": [ + { + "bbox": [ + 91, + 454, + 190, + 466 + ], + "score": 0.91, + "content": "x _ { b } ~ \\sim ~ p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 453, + 522, + 468 + ], + "score": 1.0, + "content": "using generative network. This process induces the following model distribution", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 89, + 464, + 196, + 477 + ], + "spans": [ + { + "bbox": [ + 89, + 464, + 196, + 477 + ], + "score": 1.0, + "content": "over unobserved features:", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22.5, + "bbox_fs": [ + 88, + 431, + 522, + 477 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 199, + 485, + 412, + 500 + ], + "lines": [ + { + "bbox": [ + 199, + 485, + 412, + 500 + ], + "spans": [ + { + "bbox": [ + 199, + 485, + 412, + 500 + ], + "score": 0.86, + "content": "p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b ) = \\mathbb { E } _ { z \\sim p _ { \\psi } ( z | x _ { 1 - b } , b ) } p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b )", + "type": "interline_equation", + "image_path": "1010c9b588784d89851c33cee48ac5e4d9b7fc2302b1356bbdce195c382c31ce.jpg" + } + ] + } + ], + "index": 25, + "virtual_lines": [ + { + "bbox": [ + 199, + 485, + 412, + 500 + ], + "spans": [], + "index": 25 + } + ] + }, + { + "type": "text", + "bbox": [ + 90, + 515, + 521, + 586 + ], + "lines": [ + { + "bbox": [ + 90, + 514, + 522, + 529 + ], + "spans": [ + { + "bbox": [ + 90, + 514, + 125, + 529 + ], + "score": 1.0, + "content": "We use", + "type": "text" + }, + { + "bbox": [ + 126, + 515, + 164, + 527 + ], + "score": 0.91, + "content": "z ~ \\in ~ \\mathbb { R } ^ { d }", + "type": "inline_equation" + }, + { + "bbox": [ + 164, + 514, + 280, + 529 + ], + "score": 1.0, + "content": ", and Gaussian distribution", + "type": "text" + }, + { + "bbox": [ + 280, + 518, + 293, + 528 + ], + "score": 0.82, + "content": "p _ { \\psi }", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 514, + 318, + 529 + ], + "score": 1.0, + "content": "over", + "type": "text" + }, + { + "bbox": [ + 318, + 519, + 324, + 526 + ], + "score": 0.74, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 325, + 514, + 522, + 529 + ], + "score": 1.0, + "content": ", with parameters from a neural network with", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 90, + 527, + 521, + 541 + ], + "spans": [ + { + "bbox": [ + 90, + 527, + 125, + 540 + ], + "score": 1.0, + "content": "weights", + "type": "text" + }, + { + "bbox": [ + 125, + 528, + 133, + 540 + ], + "score": 0.75, + "content": "\\psi", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 527, + 138, + 540 + ], + "score": 1.0, + "content": ":", + "type": "text" + }, + { + "bbox": [ + 139, + 527, + 356, + 541 + ], + "score": 0.87, + "content": "p _ { \\psi } ( z | x _ { 1 - b } , b , \\psi ) = \\mathcal { N } ( z | \\mu _ { \\psi } ( x _ { 1 - b } , b ) , \\sigma _ { \\psi } ^ { 2 } ( x _ { 1 - b } , b ) I )", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 527, + 521, + 540 + ], + "score": 1.0, + "content": ". The real-valued components of distri-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 90, + 540, + 520, + 553 + ], + "spans": [ + { + "bbox": [ + 90, + 540, + 118, + 553 + ], + "score": 1.0, + "content": "bution", + "type": "text" + }, + { + "bbox": [ + 118, + 541, + 187, + 552 + ], + "score": 0.92, + "content": "p _ { \\theta } ( x _ { b } | \\boldsymbol { z } , x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 540, + 387, + 553 + ], + "score": 1.0, + "content": "are defined likewise. Each categorical component", + "type": "text" + }, + { + "bbox": [ + 387, + 541, + 392, + 550 + ], + "score": 0.72, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 540, + 452, + 553 + ], + "score": 1.0, + "content": "of distribution", + "type": "text" + }, + { + "bbox": [ + 452, + 540, + 520, + 552 + ], + "score": 0.92, + "content": "p _ { \\theta } ( x _ { i } | \\boldsymbol { z } , x _ { 1 - b } , b )", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 88, + 551, + 522, + 565 + ], + "spans": [ + { + "bbox": [ + 88, + 551, + 215, + 565 + ], + "score": 1.0, + "content": "is parameterized by a function", + "type": "text" + }, + { + "bbox": [ + 215, + 552, + 278, + 563 + ], + "score": 0.91, + "content": "w _ { i , \\theta } ( z , x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 279, + 551, + 522, + 565 + ], + "score": 1.0, + "content": ", whose outputs are logits of probabilities for each category:", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 91, + 561, + 521, + 576 + ], + "spans": [ + { + "bbox": [ + 91, + 563, + 245, + 574 + ], + "score": 0.89, + "content": "x _ { i } \\sim \\mathrm { C a t } [ \\mathrm { S o f t m a x } ( w _ { i , \\theta } ( z , x _ { 1 - b } , b ) ) ]", + "type": "inline_equation" + }, + { + "bbox": [ + 245, + 561, + 442, + 576 + ], + "score": 1.0, + "content": ". Therefore the components of the latent vector", + "type": "text" + }, + { + "bbox": [ + 442, + 565, + 449, + 572 + ], + "score": 0.75, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 449, + 561, + 521, + 576 + ], + "score": 1.0, + "content": "are conditionally", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 89, + 573, + 522, + 587 + ], + "spans": [ + { + "bbox": [ + 89, + 573, + 165, + 587 + ], + "score": 1.0, + "content": "independent given", + "type": "text" + }, + { + "bbox": [ + 165, + 576, + 186, + 585 + ], + "score": 0.89, + "content": "x _ { 1 - b }", + "type": "inline_equation" + }, + { + "bbox": [ + 187, + 573, + 204, + 587 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 205, + 575, + 210, + 583 + ], + "score": 0.77, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 210, + 573, + 306, + 587 + ], + "score": 1.0, + "content": ", and the components of", + "type": "text" + }, + { + "bbox": [ + 306, + 575, + 317, + 585 + ], + "score": 0.85, + "content": "x _ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 573, + 462, + 587 + ], + "score": 1.0, + "content": "are conditionally independent given", + "type": "text" + }, + { + "bbox": [ + 462, + 575, + 494, + 585 + ], + "score": 0.51, + "content": "z , x _ { 1 - b }", + "type": "inline_equation" + }, + { + "bbox": [ + 494, + 573, + 512, + 587 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 512, + 574, + 518, + 583 + ], + "score": 0.75, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 518, + 573, + 522, + 587 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28.5, + "bbox_fs": [ + 88, + 514, + 522, + 587 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 590, + 521, + 635 + ], + "lines": [ + { + "bbox": [ + 90, + 590, + 522, + 604 + ], + "spans": [ + { + "bbox": [ + 90, + 590, + 147, + 604 + ], + "score": 1.0, + "content": "The variables", + "type": "text" + }, + { + "bbox": [ + 147, + 592, + 158, + 601 + ], + "score": 0.84, + "content": "x _ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 590, + 177, + 604 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 177, + 592, + 198, + 602 + ], + "score": 0.88, + "content": "x _ { 1 - b }", + "type": "inline_equation" + }, + { + "bbox": [ + 198, + 590, + 350, + 604 + ], + "score": 1.0, + "content": "have variable length that depends on", + "type": "text" + }, + { + "bbox": [ + 350, + 591, + 355, + 600 + ], + "score": 0.69, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 590, + 522, + 604 + ], + "score": 1.0, + "content": ". So in order to use architectures such as", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 90, + 601, + 522, + 614 + ], + "spans": [ + { + "bbox": [ + 90, + 601, + 376, + 614 + ], + "score": 1.0, + "content": "multi-layer perceptron and convolutional neural network we consider", + "type": "text" + }, + { + "bbox": [ + 376, + 601, + 461, + 613 + ], + "score": 0.92, + "content": "x _ { 1 - b } = x \\circ ( 1 - b )", + "type": "inline_equation" + }, + { + "bbox": [ + 461, + 601, + 490, + 614 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 491, + 603, + 497, + 611 + ], + "score": 0.71, + "content": "\\circ", + "type": "inline_equation" + }, + { + "bbox": [ + 498, + 601, + 522, + 614 + ], + "score": 1.0, + "content": "is an", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 89, + 611, + 522, + 626 + ], + "spans": [ + { + "bbox": [ + 89, + 611, + 274, + 626 + ], + "score": 1.0, + "content": "element-wise product. So in implementation", + "type": "text" + }, + { + "bbox": [ + 274, + 613, + 295, + 624 + ], + "score": 0.86, + "content": "x _ { 1 - b }", + "type": "inline_equation" + }, + { + "bbox": [ + 295, + 611, + 522, + 626 + ], + "score": 1.0, + "content": "has fixed length. The output of the generative network", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 90, + 623, + 440, + 636 + ], + "spans": [ + { + "bbox": [ + 90, + 623, + 440, + 636 + ], + "score": 1.0, + "content": "also has a fixed length, but we use only unobserved components to compute likelihood.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 33.5, + "bbox_fs": [ + 89, + 590, + 522, + 636 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 640, + 354, + 652 + ], + "lines": [ + { + "bbox": [ + 90, + 639, + 356, + 653 + ], + "spans": [ + { + "bbox": [ + 90, + 639, + 356, + 653 + ], + "score": 1.0, + "content": "The theoretical analysis of the model is available in appendix B.1.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 36, + "bbox_fs": [ + 90, + 639, + 356, + 653 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "title", + "bbox": [ + 89, + 95, + 439, + 108 + ], + "lines": [ + { + "bbox": [ + 89, + 95, + 441, + 108 + ], + "spans": [ + { + "bbox": [ + 89, + 95, + 441, + 108 + ], + "score": 1.0, + "content": "4.3 LEARNING VARIATIONAL AUTOENCODER WITH ARBITRARY CONDITIONING", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 91, + 116, + 253, + 128 + ], + "lines": [ + { + "bbox": [ + 89, + 116, + 253, + 129 + ], + "spans": [ + { + "bbox": [ + 89, + 116, + 253, + 129 + ], + "score": 1.0, + "content": "4.3.1 VARIATIONAL LOWER BOUND", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 90, + 136, + 429, + 149 + ], + "lines": [ + { + "bbox": [ + 90, + 136, + 430, + 150 + ], + "spans": [ + { + "bbox": [ + 90, + 136, + 222, + 150 + ], + "score": 1.0, + "content": "We can derive a lower bound for", + "type": "text" + }, + { + "bbox": [ + 223, + 137, + 304, + 149 + ], + "score": 0.91, + "content": "\\log p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 136, + 430, + 150 + ], + "score": 1.0, + "content": "as for variational autoencoder:", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2 + }, + { + "type": "interline_equation", + "bbox": [ + 100, + 159, + 499, + 203 + ], + "lines": [ + { + "bbox": [ + 100, + 159, + 499, + 203 + ], + "spans": [ + { + "bbox": [ + 100, + 159, + 499, + 203 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\log p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b ) = \\mathbb { E } _ { q _ { \\phi } ( z | x , b ) } \\log \\frac { p _ { \\psi , \\theta } ( x _ { b } , z | x _ { 1 - b } , b ) } { q _ { \\phi } ( z | x , b ) } + D _ { \\mathrm { K L } } ( q _ { \\phi } ( z | x , b ) | | p _ { \\psi , \\theta } ( z | x , b ) ) } \\\\ & { \\qquad \\geq \\mathbb { E } _ { q _ { \\phi } ( z | x , b ) } \\log p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b ) - D _ { \\mathrm { K L } } ( q _ { \\phi } ( z | x , b ) | | p _ { \\psi } ( z | x _ { 1 - b } , b ) ) = L _ { V A E A C } ( x , b ; \\theta , \\psi , \\phi ) } \\end{array}", + "type": "interline_equation", + "image_path": "23cab9595cb2df09a6fe520b5c3c4ac610eb31ffc336c62a0f70f0e3846d00aa.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 100, + 159, + 499, + 173.66666666666666 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 100, + 173.66666666666666, + 499, + 188.33333333333331 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 100, + 188.33333333333331, + 499, + 202.99999999999997 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 90, + 207, + 411, + 219 + ], + "lines": [ + { + "bbox": [ + 89, + 205, + 412, + 222 + ], + "spans": [ + { + "bbox": [ + 89, + 205, + 412, + 222 + ], + "score": 1.0, + "content": "Therefore we have the following variational lower bound optimization problem:", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 226, + 388, + 245 + ], + "lines": [ + { + "bbox": [ + 225, + 226, + 388, + 245 + ], + "spans": [ + { + "bbox": [ + 225, + 226, + 388, + 245 + ], + "score": 0.92, + "content": "\\operatorname* { m a x } _ { \\theta , \\psi , \\phi } \\mathbb { E } _ { p _ { d } ( x ) } \\mathbb { E } _ { p ( b ) } L _ { V A E A C } ( x , b ; \\theta , \\psi , \\phi )", + "type": "interline_equation", + "image_path": "0adebae96daeaa3c98aa6ef63b82c1e9fa0b29000c3c6a3af4e0beff92f397c1.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 225, + 226, + 388, + 245 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 91, + 252, + 521, + 275 + ], + "lines": [ + { + "bbox": [ + 90, + 252, + 521, + 265 + ], + "spans": [ + { + "bbox": [ + 90, + 252, + 315, + 265 + ], + "score": 1.0, + "content": "We use fully-factorized Gaussian proposal distribution", + "type": "text" + }, + { + "bbox": [ + 315, + 254, + 326, + 264 + ], + "score": 0.86, + "content": "q _ { \\phi }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 252, + 521, + 265 + ], + "score": 1.0, + "content": "which allows us to perform reparameterization", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 90, + 263, + 372, + 276 + ], + "spans": [ + { + "bbox": [ + 90, + 263, + 372, + 276 + ], + "score": 1.0, + "content": "trick and compute KL divergence analytically in order to optimize (7).", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "title", + "bbox": [ + 91, + 288, + 231, + 299 + ], + "lines": [ + { + "bbox": [ + 89, + 287, + 231, + 301 + ], + "spans": [ + { + "bbox": [ + 89, + 287, + 231, + 301 + ], + "score": 1.0, + "content": "4.3.2 PRIOR IN LATENT SPACE", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 90, + 308, + 521, + 364 + ], + "lines": [ + { + "bbox": [ + 89, + 308, + 522, + 321 + ], + "spans": [ + { + "bbox": [ + 89, + 308, + 317, + 321 + ], + "score": 1.0, + "content": "During the optimization of objective (7) the parameters", + "type": "text" + }, + { + "bbox": [ + 317, + 311, + 330, + 321 + ], + "score": 0.86, + "content": "\\mu _ { \\psi }", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 308, + 349, + 321 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 349, + 311, + 362, + 321 + ], + "score": 0.87, + "content": "\\sigma _ { \\psi }", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 308, + 473, + 321 + ], + "score": 1.0, + "content": "of the prior distribution of", + "type": "text" + }, + { + "bbox": [ + 473, + 311, + 480, + 318 + ], + "score": 0.7, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 308, + 522, + 321 + ], + "score": 1.0, + "content": "may tend", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 89, + 320, + 521, + 331 + ], + "spans": [ + { + "bbox": [ + 89, + 320, + 521, + 331 + ], + "score": 1.0, + "content": "to infinity, since there is no penalty for large values of those parameters. We usually observe the growth", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 89, + 330, + 523, + 343 + ], + "spans": [ + { + "bbox": [ + 89, + 330, + 102, + 343 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 103, + 331, + 122, + 342 + ], + "score": 0.91, + "content": "\\left. z \\right. _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 123, + 330, + 523, + 343 + ], + "score": 1.0, + "content": "during training, though it is slow enough. To prevent potential numerical instabilities, we put a", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 88, + 340, + 522, + 355 + ], + "spans": [ + { + "bbox": [ + 88, + 340, + 522, + 355 + ], + "score": 1.0, + "content": "Normal-Gamma prior on the parameters of the prior distribution to prevent the divergence. Formally, we", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 88, + 351, + 229, + 366 + ], + "spans": [ + { + "bbox": [ + 88, + 351, + 125, + 366 + ], + "score": 1.0, + "content": "redefine", + "type": "text" + }, + { + "bbox": [ + 125, + 353, + 181, + 365 + ], + "score": 0.93, + "content": "p _ { \\psi } ( z | x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 351, + 229, + 366 + ], + "score": 1.0, + "content": "as follows:", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13 + }, + { + "type": "interline_equation", + "bbox": [ + 161, + 371, + 450, + 387 + ], + "lines": [ + { + "bbox": [ + 161, + 371, + 450, + 387 + ], + "spans": [ + { + "bbox": [ + 161, + 371, + 450, + 387 + ], + "score": 0.89, + "content": "p _ { \\psi } ( z , \\mu _ { \\psi } , \\sigma _ { \\psi } | x _ { 1 - b } , b ) = \\mathcal { N } ( z | \\mu _ { \\psi } , \\sigma _ { \\psi } ^ { 2 } ) \\mathcal { N } ( \\mu _ { \\psi } | 0 , \\sigma _ { \\mu } ) \\mathrm { G a m m a } ( \\sigma _ { \\psi } | 2 , \\sigma _ { \\sigma } )", + "type": "interline_equation", + "image_path": "42bd1f2e7f266115ecbfdf42ffc85525498becdfe4a210dad151bb0713eff63a.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 161, + 371, + 450, + 387 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 90, + 394, + 520, + 437 + ], + "lines": [ + { + "bbox": [ + 90, + 392, + 520, + 418 + ], + "spans": [ + { + "bbox": [ + 90, + 398, + 203, + 411 + ], + "score": 1.0, + "content": "As a result, the regularizers", + "type": "text" + }, + { + "bbox": [ + 203, + 394, + 227, + 415 + ], + "score": 0.92, + "content": "- \\frac { \\mu _ { \\psi } ^ { 2 } } { 2 \\sigma _ { \\mu } ^ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 392, + 246, + 418 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 246, + 398, + 323, + 412 + ], + "score": 0.91, + "content": "\\sigma _ { \\sigma } ( \\log ( \\sigma _ { \\psi } ) - \\sigma _ { \\psi } )", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 392, + 520, + 418 + ], + "score": 1.0, + "content": "are added to the model log-likelihood. Hyperpa-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 89, + 414, + 522, + 427 + ], + "spans": [ + { + "bbox": [ + 89, + 414, + 123, + 427 + ], + "score": 1.0, + "content": "rameter", + "type": "text" + }, + { + "bbox": [ + 123, + 416, + 135, + 427 + ], + "score": 0.84, + "content": "\\sigma _ { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 414, + 218, + 427 + ], + "score": 1.0, + "content": "is chosen to be large", + "type": "text" + }, + { + "bbox": [ + 218, + 414, + 240, + 426 + ], + "score": 0.86, + "content": "( 1 0 ^ { 4 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 414, + 257, + 427 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 258, + 416, + 270, + 425 + ], + "score": 0.86, + "content": "\\sigma _ { \\sigma }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 414, + 421, + 427 + ], + "score": 1.0, + "content": "is taken to be a small positive number", + "type": "text" + }, + { + "bbox": [ + 421, + 414, + 448, + 426 + ], + "score": 0.86, + "content": "( 1 0 ^ { - 4 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 414, + 522, + 427 + ], + "score": 1.0, + "content": ". This distribution", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 88, + 424, + 420, + 438 + ], + "spans": [ + { + "bbox": [ + 88, + 424, + 420, + 438 + ], + "score": 1.0, + "content": "is close to uniform near zero, so it doesn’t affect the learning process significantly.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18 + }, + { + "type": "title", + "bbox": [ + 91, + 450, + 209, + 462 + ], + "lines": [ + { + "bbox": [ + 90, + 451, + 209, + 462 + ], + "spans": [ + { + "bbox": [ + 90, + 451, + 209, + 462 + ], + "score": 1.0, + "content": "4.3.3 MISSING FEATURES", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 91, + 470, + 521, + 515 + ], + "lines": [ + { + "bbox": [ + 90, + 470, + 521, + 483 + ], + "spans": [ + { + "bbox": [ + 90, + 470, + 521, + 483 + ], + "score": 1.0, + "content": "The optimization objective (7) requires all features of each object at the training stage: some of the features", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 89, + 482, + 521, + 494 + ], + "spans": [ + { + "bbox": [ + 89, + 482, + 521, + 494 + ], + "score": 1.0, + "content": "will be observed variables at the input of the model and other will be unobserved features used to evaluate the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 89, + 492, + 521, + 505 + ], + "spans": [ + { + "bbox": [ + 89, + 492, + 521, + 505 + ], + "score": 1.0, + "content": "model. Nevertheless, in some problem settings the training data contains missing features too. We propose", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 90, + 504, + 460, + 516 + ], + "spans": [ + { + "bbox": [ + 90, + 504, + 460, + 516 + ], + "score": 1.0, + "content": "the following slight modification of the problem (7) in order to cover such problems as well.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 90, + 520, + 521, + 564 + ], + "lines": [ + { + "bbox": [ + 89, + 520, + 522, + 534 + ], + "spans": [ + { + "bbox": [ + 89, + 520, + 263, + 534 + ], + "score": 1.0, + "content": "The missing values cannot be observed so", + "type": "text" + }, + { + "bbox": [ + 264, + 521, + 341, + 532 + ], + "score": 0.91, + "content": "x _ { i } = \\omega \\Rightarrow b _ { i } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 520, + 372, + 534 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 372, + 523, + 381, + 531 + ], + "score": 0.77, + "content": "\\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 520, + 522, + 534 + ], + "score": 1.0, + "content": "describes the missing value in the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 89, + 531, + 521, + 544 + ], + "spans": [ + { + "bbox": [ + 89, + 531, + 448, + 544 + ], + "score": 1.0, + "content": "data. In order to meet this requirement, we redefine mask distribution as conditioned on", + "type": "text" + }, + { + "bbox": [ + 448, + 533, + 455, + 541 + ], + "score": 0.35, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 531, + 460, + 544 + ], + "score": 1.0, + "content": ":", + "type": "text" + }, + { + "bbox": [ + 460, + 532, + 479, + 543 + ], + "score": 0.7, + "content": "p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 531, + 521, + 544 + ], + "score": 1.0, + "content": "turns into", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 91, + 542, + 521, + 556 + ], + "spans": [ + { + "bbox": [ + 91, + 542, + 117, + 555 + ], + "score": 0.92, + "content": "p ( b | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 542, + 521, + 556 + ], + "score": 1.0, + "content": "in (4) and (7). In the reconstruction loss (5) we simply omit the missing features, i. e. marginalize", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 89, + 554, + 131, + 565 + ], + "spans": [ + { + "bbox": [ + 89, + 554, + 131, + 565 + ], + "score": 1.0, + "content": "them out:", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5 + }, + { + "type": "interline_equation", + "bbox": [ + 193, + 562, + 419, + 591 + ], + "lines": [ + { + "bbox": [ + 193, + 562, + 419, + 591 + ], + "spans": [ + { + "bbox": [ + 193, + 562, + 419, + 591 + ], + "score": 0.91, + "content": "\\log p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b ) = \\sum _ { \\substack { i : b _ { i } = 1 , x _ { i } \\neq \\omega } } \\log p _ { \\theta } ( x _ { i } | z , x _ { 1 - b } , b )", + "type": "interline_equation", + "image_path": "fd9e44e756fd89b81c9ee7ee19237edf2b2753f3ffd55278ac0c898fa8a3729d.jpg" + } + ] + } + ], + "index": 29.5, + "virtual_lines": [ + { + "bbox": [ + 193, + 562, + 419, + 576.5 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 193, + 576.5, + 419, + 591.0 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 91, + 601, + 521, + 635 + ], + "lines": [ + { + "bbox": [ + 90, + 600, + 522, + 614 + ], + "spans": [ + { + "bbox": [ + 90, + 600, + 522, + 614 + ], + "score": 1.0, + "content": "The proposal network must be able to determine which features came from real object and which are just", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 89, + 612, + 521, + 625 + ], + "spans": [ + { + "bbox": [ + 89, + 612, + 521, + 625 + ], + "score": 1.0, + "content": "missing. So we use additional missing features mask which is fed to proposal network together with unob-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 89, + 622, + 238, + 636 + ], + "spans": [ + { + "bbox": [ + 89, + 622, + 176, + 636 + ], + "score": 1.0, + "content": "served features mask", + "type": "text" + }, + { + "bbox": [ + 176, + 624, + 182, + 633 + ], + "score": 0.76, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 622, + 227, + 636 + ], + "score": 1.0, + "content": "and object", + "type": "text" + }, + { + "bbox": [ + 227, + 628, + 233, + 633 + ], + "score": 0.8, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 622, + 238, + 636 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32 + }, + { + "type": "text", + "bbox": [ + 91, + 640, + 316, + 652 + ], + "lines": [ + { + "bbox": [ + 90, + 639, + 317, + 653 + ], + "spans": [ + { + "bbox": [ + 90, + 639, + 317, + 653 + ], + "score": 1.0, + "content": "The proposed modifications are evaluated in section 5.1.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34 + } + ], + "page_idx": 4, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 39, + 277, + 51 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 278, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 278, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 286, + 671, + 292, + 680 + ], + "lines": [ + { + "bbox": [ + 285, + 671, + 293, + 682 + ], + "spans": [ + { + "bbox": [ + 285, + 671, + 293, + 682 + ], + "score": 1.0, + "content": "5", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "title", + "bbox": [ + 89, + 95, + 439, + 108 + ], + "lines": [ + { + "bbox": [ + 89, + 95, + 441, + 108 + ], + "spans": [ + { + "bbox": [ + 89, + 95, + 441, + 108 + ], + "score": 1.0, + "content": "4.3 LEARNING VARIATIONAL AUTOENCODER WITH ARBITRARY CONDITIONING", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "title", + "bbox": [ + 91, + 116, + 253, + 128 + ], + "lines": [ + { + "bbox": [ + 89, + 116, + 253, + 129 + ], + "spans": [ + { + "bbox": [ + 89, + 116, + 253, + 129 + ], + "score": 1.0, + "content": "4.3.1 VARIATIONAL LOWER BOUND", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 1 + }, + { + "type": "text", + "bbox": [ + 90, + 136, + 429, + 149 + ], + "lines": [ + { + "bbox": [ + 90, + 136, + 430, + 150 + ], + "spans": [ + { + "bbox": [ + 90, + 136, + 222, + 150 + ], + "score": 1.0, + "content": "We can derive a lower bound for", + "type": "text" + }, + { + "bbox": [ + 223, + 137, + 304, + 149 + ], + "score": 0.91, + "content": "\\log p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 304, + 136, + 430, + 150 + ], + "score": 1.0, + "content": "as for variational autoencoder:", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 2, + "bbox_fs": [ + 90, + 136, + 430, + 150 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 100, + 159, + 499, + 203 + ], + "lines": [ + { + "bbox": [ + 100, + 159, + 499, + 203 + ], + "spans": [ + { + "bbox": [ + 100, + 159, + 499, + 203 + ], + "score": 0.95, + "content": "\\begin{array} { r l } & { \\log p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b ) = \\mathbb { E } _ { q _ { \\phi } ( z | x , b ) } \\log \\frac { p _ { \\psi , \\theta } ( x _ { b } , z | x _ { 1 - b } , b ) } { q _ { \\phi } ( z | x , b ) } + D _ { \\mathrm { K L } } ( q _ { \\phi } ( z | x , b ) | | p _ { \\psi , \\theta } ( z | x , b ) ) } \\\\ & { \\qquad \\geq \\mathbb { E } _ { q _ { \\phi } ( z | x , b ) } \\log p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b ) - D _ { \\mathrm { K L } } ( q _ { \\phi } ( z | x , b ) | | p _ { \\psi } ( z | x _ { 1 - b } , b ) ) = L _ { V A E A C } ( x , b ; \\theta , \\psi , \\phi ) } \\end{array}", + "type": "interline_equation", + "image_path": "23cab9595cb2df09a6fe520b5c3c4ac610eb31ffc336c62a0f70f0e3846d00aa.jpg" + } + ] + } + ], + "index": 4, + "virtual_lines": [ + { + "bbox": [ + 100, + 159, + 499, + 173.66666666666666 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 100, + 173.66666666666666, + 499, + 188.33333333333331 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 100, + 188.33333333333331, + 499, + 202.99999999999997 + ], + "spans": [], + "index": 5 + } + ] + }, + { + "type": "text", + "bbox": [ + 90, + 207, + 411, + 219 + ], + "lines": [ + { + "bbox": [ + 89, + 205, + 412, + 222 + ], + "spans": [ + { + "bbox": [ + 89, + 205, + 412, + 222 + ], + "score": 1.0, + "content": "Therefore we have the following variational lower bound optimization problem:", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6, + "bbox_fs": [ + 89, + 205, + 412, + 222 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 225, + 226, + 388, + 245 + ], + "lines": [ + { + "bbox": [ + 225, + 226, + 388, + 245 + ], + "spans": [ + { + "bbox": [ + 225, + 226, + 388, + 245 + ], + "score": 0.92, + "content": "\\operatorname* { m a x } _ { \\theta , \\psi , \\phi } \\mathbb { E } _ { p _ { d } ( x ) } \\mathbb { E } _ { p ( b ) } L _ { V A E A C } ( x , b ; \\theta , \\psi , \\phi )", + "type": "interline_equation", + "image_path": "0adebae96daeaa3c98aa6ef63b82c1e9fa0b29000c3c6a3af4e0beff92f397c1.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 225, + 226, + 388, + 245 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "text", + "bbox": [ + 91, + 252, + 521, + 275 + ], + "lines": [ + { + "bbox": [ + 90, + 252, + 521, + 265 + ], + "spans": [ + { + "bbox": [ + 90, + 252, + 315, + 265 + ], + "score": 1.0, + "content": "We use fully-factorized Gaussian proposal distribution", + "type": "text" + }, + { + "bbox": [ + 315, + 254, + 326, + 264 + ], + "score": 0.86, + "content": "q _ { \\phi }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 252, + 521, + 265 + ], + "score": 1.0, + "content": "which allows us to perform reparameterization", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 90, + 263, + 372, + 276 + ], + "spans": [ + { + "bbox": [ + 90, + 263, + 372, + 276 + ], + "score": 1.0, + "content": "trick and compute KL divergence analytically in order to optimize (7).", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 90, + 252, + 521, + 276 + ] + }, + { + "type": "title", + "bbox": [ + 91, + 288, + 231, + 299 + ], + "lines": [ + { + "bbox": [ + 89, + 287, + 231, + 301 + ], + "spans": [ + { + "bbox": [ + 89, + 287, + 231, + 301 + ], + "score": 1.0, + "content": "4.3.2 PRIOR IN LATENT SPACE", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 90, + 308, + 521, + 364 + ], + "lines": [ + { + "bbox": [ + 89, + 308, + 522, + 321 + ], + "spans": [ + { + "bbox": [ + 89, + 308, + 317, + 321 + ], + "score": 1.0, + "content": "During the optimization of objective (7) the parameters", + "type": "text" + }, + { + "bbox": [ + 317, + 311, + 330, + 321 + ], + "score": 0.86, + "content": "\\mu _ { \\psi }", + "type": "inline_equation" + }, + { + "bbox": [ + 330, + 308, + 349, + 321 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 349, + 311, + 362, + 321 + ], + "score": 0.87, + "content": "\\sigma _ { \\psi }", + "type": "inline_equation" + }, + { + "bbox": [ + 363, + 308, + 473, + 321 + ], + "score": 1.0, + "content": "of the prior distribution of", + "type": "text" + }, + { + "bbox": [ + 473, + 311, + 480, + 318 + ], + "score": 0.7, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 480, + 308, + 522, + 321 + ], + "score": 1.0, + "content": "may tend", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 89, + 320, + 521, + 331 + ], + "spans": [ + { + "bbox": [ + 89, + 320, + 521, + 331 + ], + "score": 1.0, + "content": "to infinity, since there is no penalty for large values of those parameters. We usually observe the growth", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 89, + 330, + 523, + 343 + ], + "spans": [ + { + "bbox": [ + 89, + 330, + 102, + 343 + ], + "score": 1.0, + "content": "of", + "type": "text" + }, + { + "bbox": [ + 103, + 331, + 122, + 342 + ], + "score": 0.91, + "content": "\\left. z \\right. _ { 2 }", + "type": "inline_equation" + }, + { + "bbox": [ + 123, + 330, + 523, + 343 + ], + "score": 1.0, + "content": "during training, though it is slow enough. To prevent potential numerical instabilities, we put a", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 88, + 340, + 522, + 355 + ], + "spans": [ + { + "bbox": [ + 88, + 340, + 522, + 355 + ], + "score": 1.0, + "content": "Normal-Gamma prior on the parameters of the prior distribution to prevent the divergence. Formally, we", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 88, + 351, + 229, + 366 + ], + "spans": [ + { + "bbox": [ + 88, + 351, + 125, + 366 + ], + "score": 1.0, + "content": "redefine", + "type": "text" + }, + { + "bbox": [ + 125, + 353, + 181, + 365 + ], + "score": 0.93, + "content": "p _ { \\psi } ( z | x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 351, + 229, + 366 + ], + "score": 1.0, + "content": "as follows:", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13, + "bbox_fs": [ + 88, + 308, + 523, + 366 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 161, + 371, + 450, + 387 + ], + "lines": [ + { + "bbox": [ + 161, + 371, + 450, + 387 + ], + "spans": [ + { + "bbox": [ + 161, + 371, + 450, + 387 + ], + "score": 0.89, + "content": "p _ { \\psi } ( z , \\mu _ { \\psi } , \\sigma _ { \\psi } | x _ { 1 - b } , b ) = \\mathcal { N } ( z | \\mu _ { \\psi } , \\sigma _ { \\psi } ^ { 2 } ) \\mathcal { N } ( \\mu _ { \\psi } | 0 , \\sigma _ { \\mu } ) \\mathrm { G a m m a } ( \\sigma _ { \\psi } | 2 , \\sigma _ { \\sigma } )", + "type": "interline_equation", + "image_path": "42bd1f2e7f266115ecbfdf42ffc85525498becdfe4a210dad151bb0713eff63a.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 161, + 371, + 450, + 387 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 90, + 394, + 520, + 437 + ], + "lines": [ + { + "bbox": [ + 90, + 392, + 520, + 418 + ], + "spans": [ + { + "bbox": [ + 90, + 398, + 203, + 411 + ], + "score": 1.0, + "content": "As a result, the regularizers", + "type": "text" + }, + { + "bbox": [ + 203, + 394, + 227, + 415 + ], + "score": 0.92, + "content": "- \\frac { \\mu _ { \\psi } ^ { 2 } } { 2 \\sigma _ { \\mu } ^ { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 228, + 392, + 246, + 418 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 246, + 398, + 323, + 412 + ], + "score": 0.91, + "content": "\\sigma _ { \\sigma } ( \\log ( \\sigma _ { \\psi } ) - \\sigma _ { \\psi } )", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 392, + 520, + 418 + ], + "score": 1.0, + "content": "are added to the model log-likelihood. Hyperpa-", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 89, + 414, + 522, + 427 + ], + "spans": [ + { + "bbox": [ + 89, + 414, + 123, + 427 + ], + "score": 1.0, + "content": "rameter", + "type": "text" + }, + { + "bbox": [ + 123, + 416, + 135, + 427 + ], + "score": 0.84, + "content": "\\sigma _ { \\mu }", + "type": "inline_equation" + }, + { + "bbox": [ + 135, + 414, + 218, + 427 + ], + "score": 1.0, + "content": "is chosen to be large", + "type": "text" + }, + { + "bbox": [ + 218, + 414, + 240, + 426 + ], + "score": 0.86, + "content": "( 1 0 ^ { 4 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 240, + 414, + 257, + 427 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 258, + 416, + 270, + 425 + ], + "score": 0.86, + "content": "\\sigma _ { \\sigma }", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 414, + 421, + 427 + ], + "score": 1.0, + "content": "is taken to be a small positive number", + "type": "text" + }, + { + "bbox": [ + 421, + 414, + 448, + 426 + ], + "score": 0.86, + "content": "( 1 0 ^ { - 4 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 448, + 414, + 522, + 427 + ], + "score": 1.0, + "content": ". This distribution", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 88, + 424, + 420, + 438 + ], + "spans": [ + { + "bbox": [ + 88, + 424, + 420, + 438 + ], + "score": 1.0, + "content": "is close to uniform near zero, so it doesn’t affect the learning process significantly.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18, + "bbox_fs": [ + 88, + 392, + 522, + 438 + ] + }, + { + "type": "title", + "bbox": [ + 91, + 450, + 209, + 462 + ], + "lines": [ + { + "bbox": [ + 90, + 451, + 209, + 462 + ], + "spans": [ + { + "bbox": [ + 90, + 451, + 209, + 462 + ], + "score": 1.0, + "content": "4.3.3 MISSING FEATURES", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 91, + 470, + 521, + 515 + ], + "lines": [ + { + "bbox": [ + 90, + 470, + 521, + 483 + ], + "spans": [ + { + "bbox": [ + 90, + 470, + 521, + 483 + ], + "score": 1.0, + "content": "The optimization objective (7) requires all features of each object at the training stage: some of the features", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 89, + 482, + 521, + 494 + ], + "spans": [ + { + "bbox": [ + 89, + 482, + 521, + 494 + ], + "score": 1.0, + "content": "will be observed variables at the input of the model and other will be unobserved features used to evaluate the", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 89, + 492, + 521, + 505 + ], + "spans": [ + { + "bbox": [ + 89, + 492, + 521, + 505 + ], + "score": 1.0, + "content": "model. Nevertheless, in some problem settings the training data contains missing features too. We propose", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 90, + 504, + 460, + 516 + ], + "spans": [ + { + "bbox": [ + 90, + 504, + 460, + 516 + ], + "score": 1.0, + "content": "the following slight modification of the problem (7) in order to cover such problems as well.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22.5, + "bbox_fs": [ + 89, + 470, + 521, + 516 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 520, + 521, + 564 + ], + "lines": [ + { + "bbox": [ + 89, + 520, + 522, + 534 + ], + "spans": [ + { + "bbox": [ + 89, + 520, + 263, + 534 + ], + "score": 1.0, + "content": "The missing values cannot be observed so", + "type": "text" + }, + { + "bbox": [ + 264, + 521, + 341, + 532 + ], + "score": 0.91, + "content": "x _ { i } = \\omega \\Rightarrow b _ { i } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 341, + 520, + 372, + 534 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 372, + 523, + 381, + 531 + ], + "score": 0.77, + "content": "\\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 381, + 520, + 522, + 534 + ], + "score": 1.0, + "content": "describes the missing value in the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 89, + 531, + 521, + 544 + ], + "spans": [ + { + "bbox": [ + 89, + 531, + 448, + 544 + ], + "score": 1.0, + "content": "data. In order to meet this requirement, we redefine mask distribution as conditioned on", + "type": "text" + }, + { + "bbox": [ + 448, + 533, + 455, + 541 + ], + "score": 0.35, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 455, + 531, + 460, + 544 + ], + "score": 1.0, + "content": ":", + "type": "text" + }, + { + "bbox": [ + 460, + 532, + 479, + 543 + ], + "score": 0.7, + "content": "p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 479, + 531, + 521, + 544 + ], + "score": 1.0, + "content": "turns into", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 91, + 542, + 521, + 556 + ], + "spans": [ + { + "bbox": [ + 91, + 542, + 117, + 555 + ], + "score": 0.92, + "content": "p ( b | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 542, + 521, + 556 + ], + "score": 1.0, + "content": "in (4) and (7). In the reconstruction loss (5) we simply omit the missing features, i. e. marginalize", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 89, + 554, + 131, + 565 + ], + "spans": [ + { + "bbox": [ + 89, + 554, + 131, + 565 + ], + "score": 1.0, + "content": "them out:", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5, + "bbox_fs": [ + 89, + 520, + 522, + 565 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 193, + 562, + 419, + 591 + ], + "lines": [ + { + "bbox": [ + 193, + 562, + 419, + 591 + ], + "spans": [ + { + "bbox": [ + 193, + 562, + 419, + 591 + ], + "score": 0.91, + "content": "\\log p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b ) = \\sum _ { \\substack { i : b _ { i } = 1 , x _ { i } \\neq \\omega } } \\log p _ { \\theta } ( x _ { i } | z , x _ { 1 - b } , b )", + "type": "interline_equation", + "image_path": "fd9e44e756fd89b81c9ee7ee19237edf2b2753f3ffd55278ac0c898fa8a3729d.jpg" + } + ] + } + ], + "index": 29.5, + "virtual_lines": [ + { + "bbox": [ + 193, + 562, + 419, + 576.5 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 193, + 576.5, + 419, + 591.0 + ], + "spans": [], + "index": 30 + } + ] + }, + { + "type": "text", + "bbox": [ + 91, + 601, + 521, + 635 + ], + "lines": [ + { + "bbox": [ + 90, + 600, + 522, + 614 + ], + "spans": [ + { + "bbox": [ + 90, + 600, + 522, + 614 + ], + "score": 1.0, + "content": "The proposal network must be able to determine which features came from real object and which are just", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 89, + 612, + 521, + 625 + ], + "spans": [ + { + "bbox": [ + 89, + 612, + 521, + 625 + ], + "score": 1.0, + "content": "missing. So we use additional missing features mask which is fed to proposal network together with unob-", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 89, + 622, + 238, + 636 + ], + "spans": [ + { + "bbox": [ + 89, + 622, + 176, + 636 + ], + "score": 1.0, + "content": "served features mask", + "type": "text" + }, + { + "bbox": [ + 176, + 624, + 182, + 633 + ], + "score": 0.76, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 622, + 227, + 636 + ], + "score": 1.0, + "content": "and object", + "type": "text" + }, + { + "bbox": [ + 227, + 628, + 233, + 633 + ], + "score": 0.8, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 234, + 622, + 238, + 636 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 32, + "bbox_fs": [ + 89, + 600, + 522, + 636 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 640, + 316, + 652 + ], + "lines": [ + { + "bbox": [ + 90, + 639, + 317, + 653 + ], + "spans": [ + { + "bbox": [ + 90, + 639, + 317, + 653 + ], + "score": 1.0, + "content": "The proposed modifications are evaluated in section 5.1.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 34, + "bbox_fs": [ + 90, + 639, + 317, + 653 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 91, + 113, + 525, + 171 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 97, + 93, + 511, + 105 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 98, + 91, + 512, + 107 + ], + "spans": [ + { + "bbox": [ + 98, + 91, + 512, + 107 + ], + "score": 1.0, + "content": "Table 1: NRMSE (for continuous datasets) or PFC (for categorical ones) of imputations. Less is better.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 91, + 113, + 525, + 171 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 91, + 113, + 525, + 171 + ], + "spans": [ + { + "bbox": [ + 91, + 113, + 525, + 171 + ], + "score": 0.976, + "html": "
Method/DatasetWhiteWineYeastMushroomZooPhishing
MICE0.964± 0.0071.01 ± 0.010.334± 0.0020.19±0.030.422± 0.006
MissForest0.878 ± 0.0091.02 ± 0.060.249 ± 0.0060.16 ±0.020.422 ± 0.009
GAIN0.97 ± 0.020.99 ± 0.030.271 ± 0.0030.20± 0.020.427 ± 0.010
VAEAC0.850 ± 0.0070.94 ± 0.010.244 ± 0.0020.16 ± 0.020.394± 0.006
", + "type": "table", + "image_path": "8a4721c4c0839da51aa1143ab97f04e2031a1758359b3d6ea12919686c3c0741.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 91, + 113, + 525, + 132.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 91, + 132.33333333333334, + 525, + 151.66666666666669 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 91, + 151.66666666666669, + 525, + 171.00000000000003 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "title", + "bbox": [ + 91, + 196, + 184, + 209 + ], + "lines": [ + { + "bbox": [ + 88, + 195, + 185, + 211 + ], + "spans": [ + { + "bbox": [ + 88, + 195, + 185, + 211 + ], + "score": 1.0, + "content": "5 EXPERIMENTS", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 91, + 226, + 521, + 347 + ], + "lines": [ + { + "bbox": [ + 89, + 226, + 522, + 239 + ], + "spans": [ + { + "bbox": [ + 89, + 226, + 522, + 239 + ], + "score": 1.0, + "content": "In this section we validate the performance of VAEAC using several real-world datasets. In the first set", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 89, + 237, + 522, + 249 + ], + "spans": [ + { + "bbox": [ + 89, + 237, + 522, + 249 + ], + "score": 1.0, + "content": "of experiments we evaluate VAEAC missing features imputation performance using various UCI datasets", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 89, + 248, + 522, + 260 + ], + "spans": [ + { + "bbox": [ + 89, + 248, + 522, + 260 + ], + "score": 1.0, + "content": "(Lichman, 2013). We compare imputations from our model with imputations from such classical methods as", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 90, + 260, + 521, + 271 + ], + "spans": [ + { + "bbox": [ + 90, + 260, + 521, + 271 + ], + "score": 1.0, + "content": "MICE (Buuren & Groothuis-Oudshoorn, 2010) and MissForest (Stekhoven & Buhlmann, 2011) and recently ¨", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 88, + 269, + 523, + 283 + ], + "spans": [ + { + "bbox": [ + 88, + 269, + 523, + 283 + ], + "score": 1.0, + "content": "proposed GANs-based method GAIN (Yoon et al., 2018). In the second set of experiments we use VAEAC", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 88, + 280, + 522, + 294 + ], + "spans": [ + { + "bbox": [ + 88, + 280, + 522, + 294 + ], + "score": 1.0, + "content": "to solve image inpainting problem. We show inpainitngs generated by VAEAC and compare our model with", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 88, + 291, + 522, + 304 + ], + "spans": [ + { + "bbox": [ + 88, + 291, + 522, + 304 + ], + "score": 1.0, + "content": "models from papers Pathak et al. (2016), Yeh et al. (2017) and Li et al. (2017) in terms of peak signal-to-noise", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 89, + 303, + 522, + 315 + ], + "spans": [ + { + "bbox": [ + 89, + 303, + 522, + 315 + ], + "score": 1.0, + "content": "ratio (PSNR) of obtained inpaintings on CelebA dataset (Liu et al., 2015) . And finally, we evaluate VAEAC", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 88, + 313, + 521, + 326 + ], + "spans": [ + { + "bbox": [ + 88, + 313, + 521, + 326 + ], + "score": 1.0, + "content": "against the competing method called Universal Marginalizer (Douglas et al., 2017). Additional experiments", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 89, + 325, + 520, + 338 + ], + "spans": [ + { + "bbox": [ + 89, + 325, + 520, + 338 + ], + "score": 1.0, + "content": "can be found in appendices C and D. The code is available at https://github.com/tigvarts/", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 90, + 337, + 124, + 347 + ], + "spans": [ + { + "bbox": [ + 90, + 337, + 124, + 347 + ], + "score": 1.0, + "content": "vaeac.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 10 + }, + { + "type": "title", + "bbox": [ + 92, + 367, + 256, + 379 + ], + "lines": [ + { + "bbox": [ + 90, + 367, + 258, + 380 + ], + "spans": [ + { + "bbox": [ + 90, + 367, + 258, + 380 + ], + "score": 1.0, + "content": "5.1 MISSING FEATURES IMPUTATION", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 91, + 391, + 521, + 435 + ], + "lines": [ + { + "bbox": [ + 90, + 391, + 521, + 403 + ], + "spans": [ + { + "bbox": [ + 90, + 391, + 392, + 403 + ], + "score": 1.0, + "content": "The datasets with missing features are widespread. Consider a dataset with", + "type": "text" + }, + { + "bbox": [ + 392, + 392, + 402, + 401 + ], + "score": 0.79, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 391, + 486, + 403 + ], + "score": 1.0, + "content": "-dimensional objects", + "type": "text" + }, + { + "bbox": [ + 486, + 393, + 493, + 401 + ], + "score": 0.72, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 391, + 521, + 403 + ], + "score": 1.0, + "content": "where", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 90, + 402, + 521, + 414 + ], + "spans": [ + { + "bbox": [ + 90, + 402, + 303, + 414 + ], + "score": 1.0, + "content": "each feature may be missing (which we denote by", + "type": "text" + }, + { + "bbox": [ + 303, + 403, + 339, + 413 + ], + "score": 0.88, + "content": "x _ { i } ~ = ~ \\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 402, + 438, + 414 + ], + "score": 1.0, + "content": ") and their target values", + "type": "text" + }, + { + "bbox": [ + 439, + 404, + 446, + 414 + ], + "score": 0.78, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 402, + 521, + 414 + ], + "score": 1.0, + "content": ". The majority of", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 89, + 412, + 522, + 427 + ], + "spans": [ + { + "bbox": [ + 89, + 412, + 522, + 427 + ], + "score": 1.0, + "content": "discriminative methods do not support missing values in the objects. The procedure of filling in the missing", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 89, + 423, + 303, + 437 + ], + "spans": [ + { + "bbox": [ + 89, + 423, + 303, + 437 + ], + "score": 1.0, + "content": "features values is called missing features imputation.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 91, + 441, + 520, + 496 + ], + "lines": [ + { + "bbox": [ + 89, + 441, + 522, + 454 + ], + "spans": [ + { + "bbox": [ + 89, + 441, + 522, + 454 + ], + "score": 1.0, + "content": "In this section we evaluate the quality of imputations produced by VAEAC. For evaluation we use datasets", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 90, + 451, + 522, + 465 + ], + "spans": [ + { + "bbox": [ + 90, + 451, + 381, + 465 + ], + "score": 1.0, + "content": "from UCI repository (Lichman, 2013). Before training we drop randomly", + "type": "text" + }, + { + "bbox": [ + 381, + 452, + 401, + 462 + ], + "score": 0.87, + "content": "50 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 451, + 522, + 465 + ], + "score": 1.0, + "content": "of values both in train and test", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 88, + 462, + 522, + 475 + ], + "spans": [ + { + "bbox": [ + 88, + 462, + 522, + 475 + ], + "score": 1.0, + "content": "set. After that we impute missing features using MICE (Buuren & Groothuis-Oudshoorn, 2010), MissForest", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 89, + 473, + 523, + 486 + ], + "spans": [ + { + "bbox": [ + 89, + 473, + 523, + 486 + ], + "score": 1.0, + "content": "(Stekhoven & Buhlmann, 2011), GAIN (Yoon et al., 2018) and VAEAC trained on the observed data. The ¨", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 89, + 484, + 347, + 497 + ], + "spans": [ + { + "bbox": [ + 89, + 484, + 347, + 497 + ], + "score": 1.0, + "content": "details of GAIN implementation are described in appendix A.4.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 91, + 501, + 520, + 546 + ], + "lines": [ + { + "bbox": [ + 90, + 501, + 522, + 515 + ], + "spans": [ + { + "bbox": [ + 90, + 501, + 522, + 515 + ], + "score": 1.0, + "content": "Our model learns the distribution of the imputations, so it is able to sample from this distribution. We replace", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 90, + 513, + 522, + 525 + ], + "spans": [ + { + "bbox": [ + 90, + 513, + 242, + 525 + ], + "score": 1.0, + "content": "each object with missing features by", + "type": "text" + }, + { + "bbox": [ + 242, + 513, + 275, + 523 + ], + "score": 0.89, + "content": "n = 1 0", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 513, + 522, + 525 + ], + "score": 1.0, + "content": "objects with sampled imputations, so the size of the dataset", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 90, + 523, + 521, + 536 + ], + "spans": [ + { + "bbox": [ + 90, + 523, + 142, + 536 + ], + "score": 1.0, + "content": "increases by", + "type": "text" + }, + { + "bbox": [ + 142, + 526, + 150, + 533 + ], + "score": 0.62, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 523, + 521, + 536 + ], + "score": 1.0, + "content": "times. This procedure is called missing features multiple imputation. MICE and GAIN are", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 90, + 534, + 522, + 547 + ], + "spans": [ + { + "bbox": [ + 90, + 534, + 267, + 547 + ], + "score": 1.0, + "content": "also capable of multiple imputation (we use", + "type": "text" + }, + { + "bbox": [ + 267, + 535, + 298, + 545 + ], + "score": 0.89, + "content": "n = 1 0", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 534, + 522, + 547 + ], + "score": 1.0, + "content": "for them in experiments as well), but MissForest is not.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 91, + 551, + 411, + 563 + ], + "lines": [ + { + "bbox": [ + 89, + 551, + 413, + 565 + ], + "spans": [ + { + "bbox": [ + 89, + 551, + 413, + 565 + ], + "score": 1.0, + "content": "For more details about the experimental setup see appendices A.1, A.2, and A.4.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 91, + 568, + 521, + 613 + ], + "lines": [ + { + "bbox": [ + 89, + 568, + 522, + 581 + ], + "spans": [ + { + "bbox": [ + 89, + 568, + 522, + 581 + ], + "score": 1.0, + "content": "In table 1 we report NRMSE (i.e. RMSE normalized by the standard deviation of each feature and then", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 89, + 579, + 521, + 591 + ], + "spans": [ + { + "bbox": [ + 89, + 579, + 521, + 591 + ], + "score": 1.0, + "content": "averaged over all features) of imputations for continuous datasets and proportion of falsely classified (PFC)", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 90, + 590, + 521, + 603 + ], + "spans": [ + { + "bbox": [ + 90, + 590, + 521, + 603 + ], + "score": 1.0, + "content": "for categorical ones. For multiple imputation methods we average imputations of continuous variables and", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 90, + 601, + 357, + 614 + ], + "spans": [ + { + "bbox": [ + 90, + 601, + 357, + 614 + ], + "score": 1.0, + "content": "take most frequent imputation for categorical ones for each object.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 92, + 618, + 520, + 651 + ], + "lines": [ + { + "bbox": [ + 89, + 616, + 521, + 633 + ], + "spans": [ + { + "bbox": [ + 89, + 616, + 521, + 633 + ], + "score": 1.0, + "content": "We also learn linear or logistic regression and report the regression or classification performance after ap-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 90, + 629, + 521, + 642 + ], + "spans": [ + { + "bbox": [ + 90, + 629, + 521, + 642 + ], + "score": 1.0, + "content": "plying imputations of different methods in table 2. For multiple imputation methods we average predictions", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 89, + 639, + 501, + 653 + ], + "spans": [ + { + "bbox": [ + 89, + 639, + 501, + 653 + ], + "score": 1.0, + "content": "for continuous targets and take most frequent prediction for categorical ones for each object in test set.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36 + } + ], + "page_idx": 5, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 40, + 277, + 50 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 286, + 671, + 293, + 680 + ], + "lines": [ + { + "bbox": [ + 286, + 671, + 294, + 682 + ], + "spans": [ + { + "bbox": [ + 286, + 671, + 294, + 682 + ], + "score": 1.0, + "content": "6", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 91, + 113, + 525, + 171 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 97, + 93, + 511, + 105 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 98, + 91, + 512, + 107 + ], + "spans": [ + { + "bbox": [ + 98, + 91, + 512, + 107 + ], + "score": 1.0, + "content": "Table 1: NRMSE (for continuous datasets) or PFC (for categorical ones) of imputations. Less is better.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 91, + 113, + 525, + 171 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 91, + 113, + 525, + 171 + ], + "spans": [ + { + "bbox": [ + 91, + 113, + 525, + 171 + ], + "score": 0.976, + "html": "
Method/DatasetWhiteWineYeastMushroomZooPhishing
MICE0.964± 0.0071.01 ± 0.010.334± 0.0020.19±0.030.422± 0.006
MissForest0.878 ± 0.0091.02 ± 0.060.249 ± 0.0060.16 ±0.020.422 ± 0.009
GAIN0.97 ± 0.020.99 ± 0.030.271 ± 0.0030.20± 0.020.427 ± 0.010
VAEAC0.850 ± 0.0070.94 ± 0.010.244 ± 0.0020.16 ± 0.020.394± 0.006
", + "type": "table", + "image_path": "8a4721c4c0839da51aa1143ab97f04e2031a1758359b3d6ea12919686c3c0741.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 91, + 113, + 525, + 132.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 91, + 132.33333333333334, + 525, + 151.66666666666669 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 91, + 151.66666666666669, + 525, + 171.00000000000003 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "title", + "bbox": [ + 91, + 196, + 184, + 209 + ], + "lines": [ + { + "bbox": [ + 88, + 195, + 185, + 211 + ], + "spans": [ + { + "bbox": [ + 88, + 195, + 185, + 211 + ], + "score": 1.0, + "content": "5 EXPERIMENTS", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 91, + 226, + 521, + 347 + ], + "lines": [ + { + "bbox": [ + 89, + 226, + 522, + 239 + ], + "spans": [ + { + "bbox": [ + 89, + 226, + 522, + 239 + ], + "score": 1.0, + "content": "In this section we validate the performance of VAEAC using several real-world datasets. In the first set", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 89, + 237, + 522, + 249 + ], + "spans": [ + { + "bbox": [ + 89, + 237, + 522, + 249 + ], + "score": 1.0, + "content": "of experiments we evaluate VAEAC missing features imputation performance using various UCI datasets", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 89, + 248, + 522, + 260 + ], + "spans": [ + { + "bbox": [ + 89, + 248, + 522, + 260 + ], + "score": 1.0, + "content": "(Lichman, 2013). We compare imputations from our model with imputations from such classical methods as", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 90, + 260, + 521, + 271 + ], + "spans": [ + { + "bbox": [ + 90, + 260, + 521, + 271 + ], + "score": 1.0, + "content": "MICE (Buuren & Groothuis-Oudshoorn, 2010) and MissForest (Stekhoven & Buhlmann, 2011) and recently ¨", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 88, + 269, + 523, + 283 + ], + "spans": [ + { + "bbox": [ + 88, + 269, + 523, + 283 + ], + "score": 1.0, + "content": "proposed GANs-based method GAIN (Yoon et al., 2018). In the second set of experiments we use VAEAC", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 88, + 280, + 522, + 294 + ], + "spans": [ + { + "bbox": [ + 88, + 280, + 522, + 294 + ], + "score": 1.0, + "content": "to solve image inpainting problem. We show inpainitngs generated by VAEAC and compare our model with", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 88, + 291, + 522, + 304 + ], + "spans": [ + { + "bbox": [ + 88, + 291, + 522, + 304 + ], + "score": 1.0, + "content": "models from papers Pathak et al. (2016), Yeh et al. (2017) and Li et al. (2017) in terms of peak signal-to-noise", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 89, + 303, + 522, + 315 + ], + "spans": [ + { + "bbox": [ + 89, + 303, + 522, + 315 + ], + "score": 1.0, + "content": "ratio (PSNR) of obtained inpaintings on CelebA dataset (Liu et al., 2015) . And finally, we evaluate VAEAC", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 88, + 313, + 521, + 326 + ], + "spans": [ + { + "bbox": [ + 88, + 313, + 521, + 326 + ], + "score": 1.0, + "content": "against the competing method called Universal Marginalizer (Douglas et al., 2017). Additional experiments", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 89, + 325, + 520, + 338 + ], + "spans": [ + { + "bbox": [ + 89, + 325, + 520, + 338 + ], + "score": 1.0, + "content": "can be found in appendices C and D. The code is available at https://github.com/tigvarts/", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 90, + 337, + 124, + 347 + ], + "spans": [ + { + "bbox": [ + 90, + 337, + 124, + 347 + ], + "score": 1.0, + "content": "vaeac.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 10, + "bbox_fs": [ + 88, + 226, + 523, + 347 + ] + }, + { + "type": "title", + "bbox": [ + 92, + 367, + 256, + 379 + ], + "lines": [ + { + "bbox": [ + 90, + 367, + 258, + 380 + ], + "spans": [ + { + "bbox": [ + 90, + 367, + 258, + 380 + ], + "score": 1.0, + "content": "5.1 MISSING FEATURES IMPUTATION", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 91, + 391, + 521, + 435 + ], + "lines": [ + { + "bbox": [ + 90, + 391, + 521, + 403 + ], + "spans": [ + { + "bbox": [ + 90, + 391, + 392, + 403 + ], + "score": 1.0, + "content": "The datasets with missing features are widespread. Consider a dataset with", + "type": "text" + }, + { + "bbox": [ + 392, + 392, + 402, + 401 + ], + "score": 0.79, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 402, + 391, + 486, + 403 + ], + "score": 1.0, + "content": "-dimensional objects", + "type": "text" + }, + { + "bbox": [ + 486, + 393, + 493, + 401 + ], + "score": 0.72, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 391, + 521, + 403 + ], + "score": 1.0, + "content": "where", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 90, + 402, + 521, + 414 + ], + "spans": [ + { + "bbox": [ + 90, + 402, + 303, + 414 + ], + "score": 1.0, + "content": "each feature may be missing (which we denote by", + "type": "text" + }, + { + "bbox": [ + 303, + 403, + 339, + 413 + ], + "score": 0.88, + "content": "x _ { i } ~ = ~ \\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 339, + 402, + 438, + 414 + ], + "score": 1.0, + "content": ") and their target values", + "type": "text" + }, + { + "bbox": [ + 439, + 404, + 446, + 414 + ], + "score": 0.78, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 402, + 521, + 414 + ], + "score": 1.0, + "content": ". The majority of", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 89, + 412, + 522, + 427 + ], + "spans": [ + { + "bbox": [ + 89, + 412, + 522, + 427 + ], + "score": 1.0, + "content": "discriminative methods do not support missing values in the objects. The procedure of filling in the missing", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 89, + 423, + 303, + 437 + ], + "spans": [ + { + "bbox": [ + 89, + 423, + 303, + 437 + ], + "score": 1.0, + "content": "features values is called missing features imputation.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5, + "bbox_fs": [ + 89, + 391, + 522, + 437 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 441, + 520, + 496 + ], + "lines": [ + { + "bbox": [ + 89, + 441, + 522, + 454 + ], + "spans": [ + { + "bbox": [ + 89, + 441, + 522, + 454 + ], + "score": 1.0, + "content": "In this section we evaluate the quality of imputations produced by VAEAC. For evaluation we use datasets", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 90, + 451, + 522, + 465 + ], + "spans": [ + { + "bbox": [ + 90, + 451, + 381, + 465 + ], + "score": 1.0, + "content": "from UCI repository (Lichman, 2013). Before training we drop randomly", + "type": "text" + }, + { + "bbox": [ + 381, + 452, + 401, + 462 + ], + "score": 0.87, + "content": "50 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 401, + 451, + 522, + 465 + ], + "score": 1.0, + "content": "of values both in train and test", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 88, + 462, + 522, + 475 + ], + "spans": [ + { + "bbox": [ + 88, + 462, + 522, + 475 + ], + "score": 1.0, + "content": "set. After that we impute missing features using MICE (Buuren & Groothuis-Oudshoorn, 2010), MissForest", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 89, + 473, + 523, + 486 + ], + "spans": [ + { + "bbox": [ + 89, + 473, + 523, + 486 + ], + "score": 1.0, + "content": "(Stekhoven & Buhlmann, 2011), GAIN (Yoon et al., 2018) and VAEAC trained on the observed data. The ¨", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 89, + 484, + 347, + 497 + ], + "spans": [ + { + "bbox": [ + 89, + 484, + 347, + 497 + ], + "score": 1.0, + "content": "details of GAIN implementation are described in appendix A.4.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23, + "bbox_fs": [ + 88, + 441, + 523, + 497 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 501, + 520, + 546 + ], + "lines": [ + { + "bbox": [ + 90, + 501, + 522, + 515 + ], + "spans": [ + { + "bbox": [ + 90, + 501, + 522, + 515 + ], + "score": 1.0, + "content": "Our model learns the distribution of the imputations, so it is able to sample from this distribution. We replace", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 90, + 513, + 522, + 525 + ], + "spans": [ + { + "bbox": [ + 90, + 513, + 242, + 525 + ], + "score": 1.0, + "content": "each object with missing features by", + "type": "text" + }, + { + "bbox": [ + 242, + 513, + 275, + 523 + ], + "score": 0.89, + "content": "n = 1 0", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 513, + 522, + 525 + ], + "score": 1.0, + "content": "objects with sampled imputations, so the size of the dataset", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 90, + 523, + 521, + 536 + ], + "spans": [ + { + "bbox": [ + 90, + 523, + 142, + 536 + ], + "score": 1.0, + "content": "increases by", + "type": "text" + }, + { + "bbox": [ + 142, + 526, + 150, + 533 + ], + "score": 0.62, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 150, + 523, + 521, + 536 + ], + "score": 1.0, + "content": "times. This procedure is called missing features multiple imputation. MICE and GAIN are", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 90, + 534, + 522, + 547 + ], + "spans": [ + { + "bbox": [ + 90, + 534, + 267, + 547 + ], + "score": 1.0, + "content": "also capable of multiple imputation (we use", + "type": "text" + }, + { + "bbox": [ + 267, + 535, + 298, + 545 + ], + "score": 0.89, + "content": "n = 1 0", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 534, + 522, + 547 + ], + "score": 1.0, + "content": "for them in experiments as well), but MissForest is not.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 27.5, + "bbox_fs": [ + 90, + 501, + 522, + 547 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 551, + 411, + 563 + ], + "lines": [ + { + "bbox": [ + 89, + 551, + 413, + 565 + ], + "spans": [ + { + "bbox": [ + 89, + 551, + 413, + 565 + ], + "score": 1.0, + "content": "For more details about the experimental setup see appendices A.1, A.2, and A.4.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30, + "bbox_fs": [ + 89, + 551, + 413, + 565 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 568, + 521, + 613 + ], + "lines": [ + { + "bbox": [ + 89, + 568, + 522, + 581 + ], + "spans": [ + { + "bbox": [ + 89, + 568, + 522, + 581 + ], + "score": 1.0, + "content": "In table 1 we report NRMSE (i.e. RMSE normalized by the standard deviation of each feature and then", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 89, + 579, + 521, + 591 + ], + "spans": [ + { + "bbox": [ + 89, + 579, + 521, + 591 + ], + "score": 1.0, + "content": "averaged over all features) of imputations for continuous datasets and proportion of falsely classified (PFC)", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 90, + 590, + 521, + 603 + ], + "spans": [ + { + "bbox": [ + 90, + 590, + 521, + 603 + ], + "score": 1.0, + "content": "for categorical ones. For multiple imputation methods we average imputations of continuous variables and", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 90, + 601, + 357, + 614 + ], + "spans": [ + { + "bbox": [ + 90, + 601, + 357, + 614 + ], + "score": 1.0, + "content": "take most frequent imputation for categorical ones for each object.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32.5, + "bbox_fs": [ + 89, + 568, + 522, + 614 + ] + }, + { + "type": "text", + "bbox": [ + 92, + 618, + 520, + 651 + ], + "lines": [ + { + "bbox": [ + 89, + 616, + 521, + 633 + ], + "spans": [ + { + "bbox": [ + 89, + 616, + 521, + 633 + ], + "score": 1.0, + "content": "We also learn linear or logistic regression and report the regression or classification performance after ap-", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 90, + 629, + 521, + 642 + ], + "spans": [ + { + "bbox": [ + 90, + 629, + 521, + 642 + ], + "score": 1.0, + "content": "plying imputations of different methods in table 2. For multiple imputation methods we average predictions", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 89, + 639, + 501, + 653 + ], + "spans": [ + { + "bbox": [ + 89, + 639, + 501, + 653 + ], + "score": 1.0, + "content": "for continuous targets and take most frequent prediction for categorical ones for each object in test set.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36, + "bbox_fs": [ + 89, + 616, + 521, + 653 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 97, + 124, + 514, + 183 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 88, + 93, + 519, + 116 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 89, + 92, + 521, + 106 + ], + "spans": [ + { + "bbox": [ + 89, + 92, + 521, + 106 + ], + "score": 1.0, + "content": "Table 2: R2-score (for continuous targets) or accuracy (for categorical ones) of post-imputation regression", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 89, + 104, + 226, + 117 + ], + "spans": [ + { + "bbox": [ + 89, + 104, + 226, + 117 + ], + "score": 1.0, + "content": "or classification. Higher is better.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 97, + 124, + 514, + 183 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 97, + 124, + 514, + 183 + ], + "spans": [ + { + "bbox": [ + 97, + 124, + 514, + 183 + ], + "score": 0.973, + "html": "
Method /DatasetWhiteWineYeastMushroomZ00Phishing
MICE0.13±0.020.41 ±0.020.92± 0.010.78± 0.050.75 ±0.02
MissForest0.17 ± 0.010.42 ± 0.020.972 ± 0.0030.71 ± 0.070.73 ± 0.02
GAIN0.11 ± 0.010.39 ± 0.060.969 ± 0.0050.67 ± 0.060.74 ± 0.03
VAEAC0.17 ± 0.010.43 ± 0.010.983 ± 0.0020.8 ± 0.10.74 ± 0.02
", + "type": "table", + "image_path": "f713f50af0782fe9152224004ad7b2409e899b7731c683b2a4bdd8d12d03d1da.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 97, + 124, + 514, + 143.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 97, + 143.66666666666666, + 514, + 163.33333333333331 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 97, + 163.33333333333331, + 514, + 182.99999999999997 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 1.75 + }, + { + "type": "text", + "bbox": [ + 91, + 207, + 521, + 263 + ], + "lines": [ + { + "bbox": [ + 89, + 206, + 522, + 220 + ], + "spans": [ + { + "bbox": [ + 89, + 206, + 522, + 220 + ], + "score": 1.0, + "content": "As can be seen from the tables 1 and 2, VAEAC can learn joint data distribution and use it for missing", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 90, + 218, + 521, + 230 + ], + "spans": [ + { + "bbox": [ + 90, + 218, + 521, + 230 + ], + "score": 1.0, + "content": "feature imputation. The imputations are competitive with current state of the art imputation methods in", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 90, + 229, + 522, + 241 + ], + "spans": [ + { + "bbox": [ + 90, + 229, + 522, + 241 + ], + "score": 1.0, + "content": "terms of RMSE, PFC, post-imputation regression R2-score and classification accuracy. Nevertheless, we", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 90, + 240, + 522, + 252 + ], + "spans": [ + { + "bbox": [ + 90, + 240, + 522, + 252 + ], + "score": 1.0, + "content": "don’t claim that our method is state of the art in missing features imputation; for some datasets MICE or", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 90, + 251, + 429, + 263 + ], + "spans": [ + { + "bbox": [ + 90, + 251, + 429, + 263 + ], + "score": 1.0, + "content": "MissForest outperform it. The additional experiments can be found in appendix D.2.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 7 + }, + { + "type": "title", + "bbox": [ + 92, + 281, + 198, + 292 + ], + "lines": [ + { + "bbox": [ + 89, + 280, + 200, + 293 + ], + "spans": [ + { + "bbox": [ + 89, + 280, + 200, + 293 + ], + "score": 1.0, + "content": "5.2 IMAGE INPAINTING", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 91, + 303, + 520, + 358 + ], + "lines": [ + { + "bbox": [ + 90, + 304, + 522, + 316 + ], + "spans": [ + { + "bbox": [ + 90, + 304, + 522, + 316 + ], + "score": 1.0, + "content": "The image inpainting problem has a number of different formulations. The formulation of our interest is as", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 89, + 313, + 522, + 327 + ], + "spans": [ + { + "bbox": [ + 89, + 313, + 522, + 327 + ], + "score": 1.0, + "content": "follows: some of the pixels of an image are unobserved and we want to restore them in a natural way. Unlike", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 89, + 325, + 521, + 338 + ], + "spans": [ + { + "bbox": [ + 89, + 325, + 521, + 338 + ], + "score": 1.0, + "content": "the majority of papers, we want to restore not just one most probable inpainting, but the distribution over all", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 88, + 336, + 522, + 349 + ], + "spans": [ + { + "bbox": [ + 88, + 336, + 522, + 349 + ], + "score": 1.0, + "content": "possible inpaintings from which we can sample. This distribution is extremely multi-modal because often", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 89, + 347, + 331, + 361 + ], + "spans": [ + { + "bbox": [ + 89, + 347, + 331, + 361 + ], + "score": 1.0, + "content": "there is a lot of different possible ways to inpaint the image.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 90, + 364, + 519, + 387 + ], + "lines": [ + { + "bbox": [ + 88, + 362, + 522, + 378 + ], + "spans": [ + { + "bbox": [ + 88, + 362, + 522, + 378 + ], + "score": 1.0, + "content": "Unlike the previous subsection, here we have uncorrupted images without missing features in the training", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 89, + 374, + 179, + 388 + ], + "spans": [ + { + "bbox": [ + 89, + 374, + 118, + 388 + ], + "score": 1.0, + "content": "set, so", + "type": "text" + }, + { + "bbox": [ + 118, + 375, + 174, + 388 + ], + "score": 0.93, + "content": "p ( b | x ) = p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 374, + 179, + 388 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 91, + 392, + 521, + 447 + ], + "lines": [ + { + "bbox": [ + 90, + 392, + 521, + 404 + ], + "spans": [ + { + "bbox": [ + 90, + 392, + 521, + 404 + ], + "score": 1.0, + "content": "As we show in section 2, state of the art results use different adversarial losses to achieve more sharp and", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 88, + 402, + 522, + 416 + ], + "spans": [ + { + "bbox": [ + 88, + 402, + 522, + 416 + ], + "score": 1.0, + "content": "realistic samples. VAEAC can be adapted to the image inpainting problem by using a combination of those", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 89, + 414, + 522, + 427 + ], + "spans": [ + { + "bbox": [ + 89, + 414, + 285, + 427 + ], + "score": 1.0, + "content": "adversarial losses as a part of reconstruction loss", + "type": "text" + }, + { + "bbox": [ + 286, + 414, + 354, + 426 + ], + "score": 0.91, + "content": "p _ { \\theta } ( x _ { b } | \\boldsymbol { z } , x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 414, + 522, + 427 + ], + "score": 1.0, + "content": ". Nevertheless, such construction is out of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 89, + 425, + 521, + 437 + ], + "spans": [ + { + "bbox": [ + 89, + 425, + 521, + 437 + ], + "score": 1.0, + "content": "scope for this research, so we leave it for the future work. In the current work we show that the model can", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 89, + 435, + 278, + 450 + ], + "spans": [ + { + "bbox": [ + 89, + 435, + 278, + 450 + ], + "score": 1.0, + "content": "generate both diverse and realistic inpaintings.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 91, + 453, + 521, + 486 + ], + "lines": [ + { + "bbox": [ + 90, + 452, + 521, + 466 + ], + "spans": [ + { + "bbox": [ + 90, + 452, + 521, + 466 + ], + "score": 1.0, + "content": "In figures 1, 2, 3 and 4 we visualize image inpaintings produced by VAEAC on binarized MNIST (LeCun", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 90, + 463, + 521, + 477 + ], + "spans": [ + { + "bbox": [ + 90, + 463, + 521, + 477 + ], + "score": 1.0, + "content": "et al., 1998), Omniglot (Lake et al., 2015) and CelebA (Liu et al., 2015). The details of learning procedure", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 90, + 475, + 362, + 487 + ], + "spans": [ + { + "bbox": [ + 90, + 475, + 362, + 487 + ], + "score": 1.0, + "content": "and description of datasets are available in appendixes A.1 and A.3.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 90, + 491, + 521, + 580 + ], + "lines": [ + { + "bbox": [ + 89, + 491, + 521, + 505 + ], + "spans": [ + { + "bbox": [ + 89, + 491, + 521, + 505 + ], + "score": 1.0, + "content": "To the best of our knowledge, the most modern inpainting papers don’t consider the diverse inpainting", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 89, + 502, + 522, + 516 + ], + "spans": [ + { + "bbox": [ + 89, + 502, + 522, + 516 + ], + "score": 1.0, + "content": "problem, where the goal is to build diverse image inpaintings, so there is no straightforward way to compare", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 88, + 513, + 521, + 527 + ], + "spans": [ + { + "bbox": [ + 88, + 513, + 521, + 527 + ], + "score": 1.0, + "content": "with these models. Nevertheless, we compute peak signal-to-noise ratio (PSNR) for one random inpainting", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 88, + 523, + 522, + 537 + ], + "spans": [ + { + "bbox": [ + 88, + 523, + 522, + 537 + ], + "score": 1.0, + "content": "from VAEAC and the best PSNR among 10 random inpaintings from VAEAC. One inpainting might not", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 89, + 536, + 522, + 548 + ], + "spans": [ + { + "bbox": [ + 89, + 536, + 522, + 548 + ], + "score": 1.0, + "content": "be similar to the original image, so we also measure how good the inpainting which is most similar to the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 89, + 547, + 522, + 559 + ], + "spans": [ + { + "bbox": [ + 89, + 547, + 522, + 559 + ], + "score": 1.0, + "content": "original image reconstructs it. We compare these two metrics computed for certain masks with the PSNRs", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 89, + 558, + 522, + 570 + ], + "spans": [ + { + "bbox": [ + 89, + 558, + 522, + 570 + ], + "score": 1.0, + "content": "for the same masks on CelebA from papers Yeh et al. (2017) and Li et al. (2017). The results are available", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 89, + 568, + 161, + 580 + ], + "spans": [ + { + "bbox": [ + 89, + 568, + 161, + 580 + ], + "score": 1.0, + "content": "in tables 3 and 4.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 91, + 585, + 521, + 651 + ], + "lines": [ + { + "bbox": [ + 89, + 585, + 522, + 598 + ], + "spans": [ + { + "bbox": [ + 89, + 585, + 522, + 598 + ], + "score": 1.0, + "content": "We observe that for the majority of proposed masks our model outperforms the competing methods in terms", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 89, + 595, + 522, + 609 + ], + "spans": [ + { + "bbox": [ + 89, + 595, + 522, + 609 + ], + "score": 1.0, + "content": "of PSNR even with one sample, and for the rest (where the inpaintings are significantly diverse) the best", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 89, + 607, + 522, + 620 + ], + "spans": [ + { + "bbox": [ + 89, + 607, + 522, + 620 + ], + "score": 1.0, + "content": "PSNR over 10 inpaintings is larger than the same PSNR of the competing models. Even if PSNR does", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 89, + 618, + 522, + 631 + ], + "spans": [ + { + "bbox": [ + 89, + 618, + 522, + 631 + ], + "score": 1.0, + "content": "not reflect completely the visual quality of images and tends to encourage blurry VAE samples instead of", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 89, + 629, + 521, + 642 + ], + "spans": [ + { + "bbox": [ + 89, + 629, + 521, + 642 + ], + "score": 1.0, + "content": "realistic GANs samples, the results show that VAEAC is able to solve inpainting problem comparably to the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 89, + 640, + 522, + 653 + ], + "spans": [ + { + "bbox": [ + 89, + 640, + 522, + 653 + ], + "score": 1.0, + "content": "state of the art methods. The disadvantage of VAEAC compared to Yeh et al. (2017) and Li et al. (2017) (but", + "type": "text" + } + ], + "index": 39 + } + ], + "index": 36.5 + } + ], + "page_idx": 6, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 40, + 277, + 51 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 286, + 671, + 292, + 679 + ], + "lines": [ + { + "bbox": [ + 286, + 670, + 293, + 682 + ], + "spans": [ + { + "bbox": [ + 286, + 670, + 293, + 682 + ], + "score": 1.0, + "content": "7", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 97, + 124, + 514, + 183 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 88, + 93, + 519, + 116 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 89, + 92, + 521, + 106 + ], + "spans": [ + { + "bbox": [ + 89, + 92, + 521, + 106 + ], + "score": 1.0, + "content": "Table 2: R2-score (for continuous targets) or accuracy (for categorical ones) of post-imputation regression", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 89, + 104, + 226, + 117 + ], + "spans": [ + { + "bbox": [ + 89, + 104, + 226, + 117 + ], + "score": 1.0, + "content": "or classification. Higher is better.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 97, + 124, + 514, + 183 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 97, + 124, + 514, + 183 + ], + "spans": [ + { + "bbox": [ + 97, + 124, + 514, + 183 + ], + "score": 0.973, + "html": "
Method /DatasetWhiteWineYeastMushroomZ00Phishing
MICE0.13±0.020.41 ±0.020.92± 0.010.78± 0.050.75 ±0.02
MissForest0.17 ± 0.010.42 ± 0.020.972 ± 0.0030.71 ± 0.070.73 ± 0.02
GAIN0.11 ± 0.010.39 ± 0.060.969 ± 0.0050.67 ± 0.060.74 ± 0.03
VAEAC0.17 ± 0.010.43 ± 0.010.983 ± 0.0020.8 ± 0.10.74 ± 0.02
", + "type": "table", + "image_path": "f713f50af0782fe9152224004ad7b2409e899b7731c683b2a4bdd8d12d03d1da.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 97, + 124, + 514, + 143.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 97, + 143.66666666666666, + 514, + 163.33333333333331 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 97, + 163.33333333333331, + 514, + 182.99999999999997 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 1.75 + }, + { + "type": "text", + "bbox": [ + 91, + 207, + 521, + 263 + ], + "lines": [ + { + "bbox": [ + 89, + 206, + 522, + 220 + ], + "spans": [ + { + "bbox": [ + 89, + 206, + 522, + 220 + ], + "score": 1.0, + "content": "As can be seen from the tables 1 and 2, VAEAC can learn joint data distribution and use it for missing", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 90, + 218, + 521, + 230 + ], + "spans": [ + { + "bbox": [ + 90, + 218, + 521, + 230 + ], + "score": 1.0, + "content": "feature imputation. The imputations are competitive with current state of the art imputation methods in", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 90, + 229, + 522, + 241 + ], + "spans": [ + { + "bbox": [ + 90, + 229, + 522, + 241 + ], + "score": 1.0, + "content": "terms of RMSE, PFC, post-imputation regression R2-score and classification accuracy. Nevertheless, we", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 90, + 240, + 522, + 252 + ], + "spans": [ + { + "bbox": [ + 90, + 240, + 522, + 252 + ], + "score": 1.0, + "content": "don’t claim that our method is state of the art in missing features imputation; for some datasets MICE or", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 90, + 251, + 429, + 263 + ], + "spans": [ + { + "bbox": [ + 90, + 251, + 429, + 263 + ], + "score": 1.0, + "content": "MissForest outperform it. The additional experiments can be found in appendix D.2.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 7, + "bbox_fs": [ + 89, + 206, + 522, + 263 + ] + }, + { + "type": "title", + "bbox": [ + 92, + 281, + 198, + 292 + ], + "lines": [ + { + "bbox": [ + 89, + 280, + 200, + 293 + ], + "spans": [ + { + "bbox": [ + 89, + 280, + 200, + 293 + ], + "score": 1.0, + "content": "5.2 IMAGE INPAINTING", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 91, + 303, + 520, + 358 + ], + "lines": [ + { + "bbox": [ + 90, + 304, + 522, + 316 + ], + "spans": [ + { + "bbox": [ + 90, + 304, + 522, + 316 + ], + "score": 1.0, + "content": "The image inpainting problem has a number of different formulations. The formulation of our interest is as", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 89, + 313, + 522, + 327 + ], + "spans": [ + { + "bbox": [ + 89, + 313, + 522, + 327 + ], + "score": 1.0, + "content": "follows: some of the pixels of an image are unobserved and we want to restore them in a natural way. Unlike", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 89, + 325, + 521, + 338 + ], + "spans": [ + { + "bbox": [ + 89, + 325, + 521, + 338 + ], + "score": 1.0, + "content": "the majority of papers, we want to restore not just one most probable inpainting, but the distribution over all", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 88, + 336, + 522, + 349 + ], + "spans": [ + { + "bbox": [ + 88, + 336, + 522, + 349 + ], + "score": 1.0, + "content": "possible inpaintings from which we can sample. This distribution is extremely multi-modal because often", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 89, + 347, + 331, + 361 + ], + "spans": [ + { + "bbox": [ + 89, + 347, + 331, + 361 + ], + "score": 1.0, + "content": "there is a lot of different possible ways to inpaint the image.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13, + "bbox_fs": [ + 88, + 304, + 522, + 361 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 364, + 519, + 387 + ], + "lines": [ + { + "bbox": [ + 88, + 362, + 522, + 378 + ], + "spans": [ + { + "bbox": [ + 88, + 362, + 522, + 378 + ], + "score": 1.0, + "content": "Unlike the previous subsection, here we have uncorrupted images without missing features in the training", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 89, + 374, + 179, + 388 + ], + "spans": [ + { + "bbox": [ + 89, + 374, + 118, + 388 + ], + "score": 1.0, + "content": "set, so", + "type": "text" + }, + { + "bbox": [ + 118, + 375, + 174, + 388 + ], + "score": 0.93, + "content": "p ( b | x ) = p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 175, + 374, + 179, + 388 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 88, + 362, + 522, + 388 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 392, + 521, + 447 + ], + "lines": [ + { + "bbox": [ + 90, + 392, + 521, + 404 + ], + "spans": [ + { + "bbox": [ + 90, + 392, + 521, + 404 + ], + "score": 1.0, + "content": "As we show in section 2, state of the art results use different adversarial losses to achieve more sharp and", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 88, + 402, + 522, + 416 + ], + "spans": [ + { + "bbox": [ + 88, + 402, + 522, + 416 + ], + "score": 1.0, + "content": "realistic samples. VAEAC can be adapted to the image inpainting problem by using a combination of those", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 89, + 414, + 522, + 427 + ], + "spans": [ + { + "bbox": [ + 89, + 414, + 285, + 427 + ], + "score": 1.0, + "content": "adversarial losses as a part of reconstruction loss", + "type": "text" + }, + { + "bbox": [ + 286, + 414, + 354, + 426 + ], + "score": 0.91, + "content": "p _ { \\theta } ( x _ { b } | \\boldsymbol { z } , x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 354, + 414, + 522, + 427 + ], + "score": 1.0, + "content": ". Nevertheless, such construction is out of", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 89, + 425, + 521, + 437 + ], + "spans": [ + { + "bbox": [ + 89, + 425, + 521, + 437 + ], + "score": 1.0, + "content": "scope for this research, so we leave it for the future work. In the current work we show that the model can", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 89, + 435, + 278, + 450 + ], + "spans": [ + { + "bbox": [ + 89, + 435, + 278, + 450 + ], + "score": 1.0, + "content": "generate both diverse and realistic inpaintings.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 20, + "bbox_fs": [ + 88, + 392, + 522, + 450 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 453, + 521, + 486 + ], + "lines": [ + { + "bbox": [ + 90, + 452, + 521, + 466 + ], + "spans": [ + { + "bbox": [ + 90, + 452, + 521, + 466 + ], + "score": 1.0, + "content": "In figures 1, 2, 3 and 4 we visualize image inpaintings produced by VAEAC on binarized MNIST (LeCun", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 90, + 463, + 521, + 477 + ], + "spans": [ + { + "bbox": [ + 90, + 463, + 521, + 477 + ], + "score": 1.0, + "content": "et al., 1998), Omniglot (Lake et al., 2015) and CelebA (Liu et al., 2015). The details of learning procedure", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 90, + 475, + 362, + 487 + ], + "spans": [ + { + "bbox": [ + 90, + 475, + 362, + 487 + ], + "score": 1.0, + "content": "and description of datasets are available in appendixes A.1 and A.3.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24, + "bbox_fs": [ + 90, + 452, + 521, + 487 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 491, + 521, + 580 + ], + "lines": [ + { + "bbox": [ + 89, + 491, + 521, + 505 + ], + "spans": [ + { + "bbox": [ + 89, + 491, + 521, + 505 + ], + "score": 1.0, + "content": "To the best of our knowledge, the most modern inpainting papers don’t consider the diverse inpainting", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 89, + 502, + 522, + 516 + ], + "spans": [ + { + "bbox": [ + 89, + 502, + 522, + 516 + ], + "score": 1.0, + "content": "problem, where the goal is to build diverse image inpaintings, so there is no straightforward way to compare", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 88, + 513, + 521, + 527 + ], + "spans": [ + { + "bbox": [ + 88, + 513, + 521, + 527 + ], + "score": 1.0, + "content": "with these models. Nevertheless, we compute peak signal-to-noise ratio (PSNR) for one random inpainting", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 88, + 523, + 522, + 537 + ], + "spans": [ + { + "bbox": [ + 88, + 523, + 522, + 537 + ], + "score": 1.0, + "content": "from VAEAC and the best PSNR among 10 random inpaintings from VAEAC. One inpainting might not", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 89, + 536, + 522, + 548 + ], + "spans": [ + { + "bbox": [ + 89, + 536, + 522, + 548 + ], + "score": 1.0, + "content": "be similar to the original image, so we also measure how good the inpainting which is most similar to the", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 89, + 547, + 522, + 559 + ], + "spans": [ + { + "bbox": [ + 89, + 547, + 522, + 559 + ], + "score": 1.0, + "content": "original image reconstructs it. We compare these two metrics computed for certain masks with the PSNRs", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 89, + 558, + 522, + 570 + ], + "spans": [ + { + "bbox": [ + 89, + 558, + 522, + 570 + ], + "score": 1.0, + "content": "for the same masks on CelebA from papers Yeh et al. (2017) and Li et al. (2017). The results are available", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 89, + 568, + 161, + 580 + ], + "spans": [ + { + "bbox": [ + 89, + 568, + 161, + 580 + ], + "score": 1.0, + "content": "in tables 3 and 4.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 29.5, + "bbox_fs": [ + 88, + 491, + 522, + 580 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 585, + 521, + 651 + ], + "lines": [ + { + "bbox": [ + 89, + 585, + 522, + 598 + ], + "spans": [ + { + "bbox": [ + 89, + 585, + 522, + 598 + ], + "score": 1.0, + "content": "We observe that for the majority of proposed masks our model outperforms the competing methods in terms", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 89, + 595, + 522, + 609 + ], + "spans": [ + { + "bbox": [ + 89, + 595, + 522, + 609 + ], + "score": 1.0, + "content": "of PSNR even with one sample, and for the rest (where the inpaintings are significantly diverse) the best", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 89, + 607, + 522, + 620 + ], + "spans": [ + { + "bbox": [ + 89, + 607, + 522, + 620 + ], + "score": 1.0, + "content": "PSNR over 10 inpaintings is larger than the same PSNR of the competing models. Even if PSNR does", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 89, + 618, + 522, + 631 + ], + "spans": [ + { + "bbox": [ + 89, + 618, + 522, + 631 + ], + "score": 1.0, + "content": "not reflect completely the visual quality of images and tends to encourage blurry VAE samples instead of", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 89, + 629, + 521, + 642 + ], + "spans": [ + { + "bbox": [ + 89, + 629, + 521, + 642 + ], + "score": 1.0, + "content": "realistic GANs samples, the results show that VAEAC is able to solve inpainting problem comparably to the", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 89, + 640, + 522, + 653 + ], + "spans": [ + { + "bbox": [ + 89, + 640, + 522, + 653 + ], + "score": 1.0, + "content": "state of the art methods. The disadvantage of VAEAC compared to Yeh et al. (2017) and Li et al. (2017) (but", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 89, + 321, + 522, + 334 + ], + "spans": [ + { + "bbox": [ + 89, + 321, + 522, + 334 + ], + "score": 1.0, + "content": "not Pathak et al. (2016)) is that it needs the distribution over masks at the training stage to be similar to the", + "type": "text", + "cross_page": true + } + ], + "index": 10 + }, + { + "bbox": [ + 89, + 331, + 504, + 345 + ], + "spans": [ + { + "bbox": [ + 89, + 331, + 504, + 345 + ], + "score": 1.0, + "content": "distribution over them at the test stage. However, it is not a very strict limitation for the practical usage.", + "type": "text", + "cross_page": true + } + ], + "index": 11 + } + ], + "index": 36.5, + "bbox_fs": [ + 89, + 585, + 522, + 653 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 180, + 129, + 430, + 188 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 89, + 92, + 521, + 117 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 89, + 93, + 522, + 106 + ], + "spans": [ + { + "bbox": [ + 89, + 93, + 522, + 106 + ], + "score": 1.0, + "content": "Table 3: PSNR of inpaintings for different masks for Context Encoder (Pathak et al., 2016), model from", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 89, + 103, + 522, + 118 + ], + "spans": [ + { + "bbox": [ + 89, + 103, + 522, + 118 + ], + "score": 1.0, + "content": "“Semantic Image Inpainting with Deep Generative Models” (Yeh et al., 2017) and VAEAC. Higher is better.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 180, + 129, + 430, + 188 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 180, + 129, + 430, + 188 + ], + "spans": [ + { + "bbox": [ + 180, + 129, + 430, + 188 + ], + "score": 0.975, + "html": "
Method/MasksCenterPatternRandomHalf
Context Encoder 121.319.220.615.5
SIIDGM 119.417.422.813.7
VAEAC, 1 sample22.121.429.314.9
VAEAC,10 samples23.723.329.317.4
", + "type": "table", + "image_path": "5026e89f974837461ecf99590c5cc0d8aa5c4f5fb91a1c67e40c826d645b2e9c.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 180, + 129, + 430, + 148.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 180, + 148.66666666666666, + 430, + 168.33333333333331 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 180, + 168.33333333333331, + 430, + 187.99999999999997 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 1.75 + }, + { + "type": "table", + "bbox": [ + 169, + 239, + 442, + 298 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 88, + 202, + 520, + 226 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 89, + 201, + 522, + 216 + ], + "spans": [ + { + "bbox": [ + 89, + 201, + 522, + 216 + ], + "score": 1.0, + "content": "Table 4: PSNR of inpaintings for different masks for Context Encoder (Pathak et al., 2016), model from", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 91, + 214, + 401, + 226 + ], + "spans": [ + { + "bbox": [ + 91, + 214, + 401, + 226 + ], + "score": 1.0, + "content": "“Generative Face Completion” (Li et al., 2017) and VAEAC. Higher is better.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "table_body", + "bbox": [ + 169, + 239, + 442, + 298 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 169, + 239, + 442, + 298 + ], + "spans": [ + { + "bbox": [ + 169, + 239, + 442, + 298 + ], + "score": 0.971, + "html": "
Method/Masks010203040506
Context Encoder218.618.417.919.019.119.3
GFC ²20.019.818.819.719.520.2
VAEAC,1 sample20.821.019.520.320.321.0
VAEAC,10 samples22.022.220.821.721.822.2
", + "type": "table", + "image_path": "651996553ca128a2271022b3f6a402952fe4db1b266c3fc41a688b8cbaf6c735.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 169, + 239, + 442, + 258.6666666666667 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 169, + 258.6666666666667, + 442, + 278.33333333333337 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 169, + 278.33333333333337, + 442, + 298.00000000000006 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 6.75 + }, + { + "type": "text", + "bbox": [ + 92, + 321, + 521, + 344 + ], + "lines": [ + { + "bbox": [ + 89, + 321, + 522, + 334 + ], + "spans": [ + { + "bbox": [ + 89, + 321, + 522, + 334 + ], + "score": 1.0, + "content": "not Pathak et al. (2016)) is that it needs the distribution over masks at the training stage to be similar to the", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 89, + 331, + 504, + 345 + ], + "spans": [ + { + "bbox": [ + 89, + 331, + 504, + 345 + ], + "score": 1.0, + "content": "distribution over them at the test stage. However, it is not a very strict limitation for the practical usage.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "title", + "bbox": [ + 91, + 357, + 237, + 369 + ], + "lines": [ + { + "bbox": [ + 90, + 357, + 238, + 369 + ], + "spans": [ + { + "bbox": [ + 90, + 357, + 238, + 369 + ], + "score": 1.0, + "content": "5.3 UNIVERSAL MARGINALIZER", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 91, + 377, + 520, + 400 + ], + "lines": [ + { + "bbox": [ + 90, + 377, + 522, + 390 + ], + "spans": [ + { + "bbox": [ + 90, + 377, + 522, + 390 + ], + "score": 1.0, + "content": "Universal Marginalizer (Douglas et al., 2017) (UM) is a model which uses a single neural network to estimate", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 90, + 388, + 474, + 402 + ], + "spans": [ + { + "bbox": [ + 90, + 388, + 474, + 402 + ], + "score": 1.0, + "content": "the marginal distributions over the unobserved features. So it optimizes the following objective:", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "interline_equation", + "bbox": [ + 211, + 403, + 401, + 438 + ], + "lines": [ + { + "bbox": [ + 211, + 403, + 401, + 438 + ], + "spans": [ + { + "bbox": [ + 211, + 403, + 401, + 438 + ], + "score": 0.94, + "content": "\\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { x \\sim p _ { d } ( x ) } \\mathbb { E } _ { b \\sim p ( b ) } \\sum _ { i = 1 } ^ { D } b _ { i } \\log p _ { \\theta } \\big ( x _ { i } | x _ { 1 - b } , b \\big )", + "type": "interline_equation", + "image_path": "acbf51b909c72b7e6a56e480067a9402f1912d3b957be1fdb0f7435430250ce2.jpg" + } + ] + } + ], + "index": 15.5, + "virtual_lines": [ + { + "bbox": [ + 211, + 403, + 401, + 420.5 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 211, + 420.5, + 401, + 438.0 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 90, + 446, + 522, + 480 + ], + "lines": [ + { + "bbox": [ + 89, + 446, + 522, + 459 + ], + "spans": [ + { + "bbox": [ + 89, + 446, + 153, + 459 + ], + "score": 1.0, + "content": "For given mask", + "type": "text" + }, + { + "bbox": [ + 154, + 447, + 159, + 456 + ], + "score": 0.77, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 446, + 367, + 459 + ], + "score": 1.0, + "content": "we fix a permutation of its unobserved components:", + "type": "text" + }, + { + "bbox": [ + 368, + 446, + 430, + 459 + ], + "score": 0.92, + "content": "( i _ { 1 } , i _ { 2 } , \\dots , i _ { | b | } )", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 446, + 461, + 459 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 461, + 446, + 472, + 458 + ], + "score": 0.88, + "content": "| b |", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 446, + 522, + 459 + ], + "score": 1.0, + "content": "is a number", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 89, + 457, + 522, + 470 + ], + "spans": [ + { + "bbox": [ + 89, + 457, + 522, + 470 + ], + "score": 1.0, + "content": "of unobserved components. Using the learned model and the permutation we can generate objects from joint", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 90, + 468, + 326, + 480 + ], + "spans": [ + { + "bbox": [ + 90, + 468, + 326, + 480 + ], + "score": 1.0, + "content": "distribution and estimate their probability using chain rule.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18 + }, + { + "type": "interline_equation", + "bbox": [ + 172, + 483, + 441, + 519 + ], + "lines": [ + { + "bbox": [ + 172, + 483, + 441, + 519 + ], + "spans": [ + { + "bbox": [ + 172, + 483, + 441, + 519 + ], + "score": 0.92, + "content": "\\log p _ { \\theta } ( x _ { b } | x _ { 1 - b } , b ) = \\sum _ { j = 1 } ^ { | b | } \\log p _ { \\theta } ( x _ { i _ { j } } | x _ { 1 - ( b - \\sum _ { k = 1 } ^ { j - 1 } e _ { i _ { k } } ) } , b - \\sum _ { k = 1 } ^ { j - 1 } e _ { i _ { k } } )", + "type": "interline_equation", + "image_path": "29a588d63f7f9b56a9a4a873e8bb8c472a9609e06bfe1de20bf3ead78d6521b2.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 172, + 483, + 441, + 495.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 172, + 495.0, + 441, + 507.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 172, + 507.0, + 441, + 519.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 91, + 522, + 456, + 536 + ], + "lines": [ + { + "bbox": [ + 88, + 520, + 456, + 537 + ], + "spans": [ + { + "bbox": [ + 88, + 520, + 144, + 537 + ], + "score": 1.0, + "content": "For example,", + "type": "text" + }, + { + "bbox": [ + 145, + 522, + 456, + 535 + ], + "score": 0.72, + "content": "p _ { \\theta } ( x _ { 1 } , x _ { 4 } , x _ { 5 } | x _ { 2 } , x _ { 3 } ) = p _ { \\theta } ( x _ { 4 } | x _ { 2 } , x _ { 3 } ) p _ { \\theta } ( x _ { 1 } | x _ { 2 } , x _ { 3 } , x _ { 4 } ) p _ { \\theta } ( x _ { 5 } | x _ { 1 } , x _ { 2 } , x _ { 3 } , x _ { 4 } ) .", + "type": "inline_equation", + "image_path": "8856627f2edacd04d09da432d21e523edd0fdba8ee175da45ab1eb0502524149.jpg" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 90, + 539, + 521, + 584 + ], + "lines": [ + { + "bbox": [ + 90, + 538, + 521, + 552 + ], + "spans": [ + { + "bbox": [ + 90, + 538, + 413, + 552 + ], + "score": 1.0, + "content": "Conditional sampling or conditional likelihood estimation for one object requires", + "type": "text" + }, + { + "bbox": [ + 414, + 540, + 424, + 551 + ], + "score": 0.89, + "content": "| b |", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 538, + 521, + 552 + ], + "score": 1.0, + "content": "requests to UM to com-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 89, + 551, + 521, + 563 + ], + "spans": [ + { + "bbox": [ + 89, + 551, + 110, + 562 + ], + "score": 1.0, + "content": "pute", + "type": "text" + }, + { + "bbox": [ + 111, + 551, + 169, + 563 + ], + "score": 0.93, + "content": "p _ { \\theta } ( x _ { i } | x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 551, + 521, + 562 + ], + "score": 1.0, + "content": ". Each request is a forward pass through the neural network. In the case of conditional", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 89, + 561, + 522, + 575 + ], + "spans": [ + { + "bbox": [ + 89, + 561, + 522, + 575 + ], + "score": 1.0, + "content": "sampling those requests even cannot be paralleled because the input of the next request contains the output", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 89, + 571, + 172, + 586 + ], + "spans": [ + { + "bbox": [ + 89, + 571, + 172, + 586 + ], + "score": 1.0, + "content": "of the previous one.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 91, + 589, + 521, + 622 + ], + "lines": [ + { + "bbox": [ + 90, + 588, + 521, + 602 + ], + "spans": [ + { + "bbox": [ + 90, + 588, + 521, + 602 + ], + "score": 1.0, + "content": "We propose a slight modification of the original UM training procedure which allows learning UM efficiently", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 90, + 600, + 522, + 613 + ], + "spans": [ + { + "bbox": [ + 90, + 600, + 522, + 613 + ], + "score": 1.0, + "content": "for any kind of masks including those considered in this paper. The details of the modification are described", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 88, + 611, + 159, + 624 + ], + "spans": [ + { + "bbox": [ + 88, + 611, + 159, + 624 + ], + "score": 1.0, + "content": "in appendix B.3.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29 + } + ], + "page_idx": 7, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 102, + 630, + 280, + 652 + ], + "lines": [ + { + "bbox": [ + 102, + 627, + 281, + 642 + ], + "spans": [ + { + "bbox": [ + 102, + 627, + 281, + 642 + ], + "score": 1.0, + "content": "1The results are from the paper (Yeh et al., 2017)", + "type": "text" + } + ] + }, + { + "bbox": [ + 102, + 638, + 276, + 654 + ], + "spans": [ + { + "bbox": [ + 102, + 638, + 276, + 654 + ], + "score": 1.0, + "content": "2The results are from the paper (Li et al., 2017)", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 91, + 39, + 277, + 51 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 286, + 671, + 292, + 680 + ], + "lines": [ + { + "bbox": [ + 284, + 670, + 293, + 681 + ], + "spans": [ + { + "bbox": [ + 284, + 670, + 293, + 681 + ], + "score": 1.0, + "content": "8", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 180, + 129, + 430, + 188 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 89, + 92, + 521, + 117 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 89, + 93, + 522, + 106 + ], + "spans": [ + { + "bbox": [ + 89, + 93, + 522, + 106 + ], + "score": 1.0, + "content": "Table 3: PSNR of inpaintings for different masks for Context Encoder (Pathak et al., 2016), model from", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 89, + 103, + 522, + 118 + ], + "spans": [ + { + "bbox": [ + 89, + 103, + 522, + 118 + ], + "score": 1.0, + "content": "“Semantic Image Inpainting with Deep Generative Models” (Yeh et al., 2017) and VAEAC. Higher is better.", + "type": "text" + } + ], + "index": 1 + } + ], + "index": 0.5 + }, + { + "type": "table_body", + "bbox": [ + 180, + 129, + 430, + 188 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 180, + 129, + 430, + 188 + ], + "spans": [ + { + "bbox": [ + 180, + 129, + 430, + 188 + ], + "score": 0.975, + "html": "
Method/MasksCenterPatternRandomHalf
Context Encoder 121.319.220.615.5
SIIDGM 119.417.422.813.7
VAEAC, 1 sample22.121.429.314.9
VAEAC,10 samples23.723.329.317.4
", + "type": "table", + "image_path": "5026e89f974837461ecf99590c5cc0d8aa5c4f5fb91a1c67e40c826d645b2e9c.jpg" + } + ] + } + ], + "index": 3, + "virtual_lines": [ + { + "bbox": [ + 180, + 129, + 430, + 148.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 180, + 148.66666666666666, + 430, + 168.33333333333331 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 180, + 168.33333333333331, + 430, + 187.99999999999997 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 1.75 + }, + { + "type": "table", + "bbox": [ + 169, + 239, + 442, + 298 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 88, + 202, + 520, + 226 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 89, + 201, + 522, + 216 + ], + "spans": [ + { + "bbox": [ + 89, + 201, + 522, + 216 + ], + "score": 1.0, + "content": "Table 4: PSNR of inpaintings for different masks for Context Encoder (Pathak et al., 2016), model from", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 91, + 214, + 401, + 226 + ], + "spans": [ + { + "bbox": [ + 91, + 214, + 401, + 226 + ], + "score": 1.0, + "content": "“Generative Face Completion” (Li et al., 2017) and VAEAC. Higher is better.", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 5.5 + }, + { + "type": "table_body", + "bbox": [ + 169, + 239, + 442, + 298 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 169, + 239, + 442, + 298 + ], + "spans": [ + { + "bbox": [ + 169, + 239, + 442, + 298 + ], + "score": 0.971, + "html": "
Method/Masks010203040506
Context Encoder218.618.417.919.019.119.3
GFC ²20.019.818.819.719.520.2
VAEAC,1 sample20.821.019.520.320.321.0
VAEAC,10 samples22.022.220.821.721.822.2
", + "type": "table", + "image_path": "651996553ca128a2271022b3f6a402952fe4db1b266c3fc41a688b8cbaf6c735.jpg" + } + ] + } + ], + "index": 8, + "virtual_lines": [ + { + "bbox": [ + 169, + 239, + 442, + 258.6666666666667 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 169, + 258.6666666666667, + 442, + 278.33333333333337 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 169, + 278.33333333333337, + 442, + 298.00000000000006 + ], + "spans": [], + "index": 9 + } + ] + } + ], + "index": 6.75 + }, + { + "type": "text", + "bbox": [ + 92, + 321, + 521, + 344 + ], + "lines": [], + "index": 10.5, + "bbox_fs": [ + 89, + 321, + 522, + 345 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 91, + 357, + 237, + 369 + ], + "lines": [ + { + "bbox": [ + 90, + 357, + 238, + 369 + ], + "spans": [ + { + "bbox": [ + 90, + 357, + 238, + 369 + ], + "score": 1.0, + "content": "5.3 UNIVERSAL MARGINALIZER", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 91, + 377, + 520, + 400 + ], + "lines": [ + { + "bbox": [ + 90, + 377, + 522, + 390 + ], + "spans": [ + { + "bbox": [ + 90, + 377, + 522, + 390 + ], + "score": 1.0, + "content": "Universal Marginalizer (Douglas et al., 2017) (UM) is a model which uses a single neural network to estimate", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 90, + 388, + 474, + 402 + ], + "spans": [ + { + "bbox": [ + 90, + 388, + 474, + 402 + ], + "score": 1.0, + "content": "the marginal distributions over the unobserved features. So it optimizes the following objective:", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5, + "bbox_fs": [ + 90, + 377, + 522, + 402 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 211, + 403, + 401, + 438 + ], + "lines": [ + { + "bbox": [ + 211, + 403, + 401, + 438 + ], + "spans": [ + { + "bbox": [ + 211, + 403, + 401, + 438 + ], + "score": 0.94, + "content": "\\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { x \\sim p _ { d } ( x ) } \\mathbb { E } _ { b \\sim p ( b ) } \\sum _ { i = 1 } ^ { D } b _ { i } \\log p _ { \\theta } \\big ( x _ { i } | x _ { 1 - b } , b \\big )", + "type": "interline_equation", + "image_path": "acbf51b909c72b7e6a56e480067a9402f1912d3b957be1fdb0f7435430250ce2.jpg" + } + ] + } + ], + "index": 15.5, + "virtual_lines": [ + { + "bbox": [ + 211, + 403, + 401, + 420.5 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 211, + 420.5, + 401, + 438.0 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 90, + 446, + 522, + 480 + ], + "lines": [ + { + "bbox": [ + 89, + 446, + 522, + 459 + ], + "spans": [ + { + "bbox": [ + 89, + 446, + 153, + 459 + ], + "score": 1.0, + "content": "For given mask", + "type": "text" + }, + { + "bbox": [ + 154, + 447, + 159, + 456 + ], + "score": 0.77, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 159, + 446, + 367, + 459 + ], + "score": 1.0, + "content": "we fix a permutation of its unobserved components:", + "type": "text" + }, + { + "bbox": [ + 368, + 446, + 430, + 459 + ], + "score": 0.92, + "content": "( i _ { 1 } , i _ { 2 } , \\dots , i _ { | b | } )", + "type": "inline_equation" + }, + { + "bbox": [ + 431, + 446, + 461, + 459 + ], + "score": 1.0, + "content": ", where", + "type": "text" + }, + { + "bbox": [ + 461, + 446, + 472, + 458 + ], + "score": 0.88, + "content": "| b |", + "type": "inline_equation" + }, + { + "bbox": [ + 472, + 446, + 522, + 459 + ], + "score": 1.0, + "content": "is a number", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 89, + 457, + 522, + 470 + ], + "spans": [ + { + "bbox": [ + 89, + 457, + 522, + 470 + ], + "score": 1.0, + "content": "of unobserved components. Using the learned model and the permutation we can generate objects from joint", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 90, + 468, + 326, + 480 + ], + "spans": [ + { + "bbox": [ + 90, + 468, + 326, + 480 + ], + "score": 1.0, + "content": "distribution and estimate their probability using chain rule.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 18, + "bbox_fs": [ + 89, + 446, + 522, + 480 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 172, + 483, + 441, + 519 + ], + "lines": [ + { + "bbox": [ + 172, + 483, + 441, + 519 + ], + "spans": [ + { + "bbox": [ + 172, + 483, + 441, + 519 + ], + "score": 0.92, + "content": "\\log p _ { \\theta } ( x _ { b } | x _ { 1 - b } , b ) = \\sum _ { j = 1 } ^ { | b | } \\log p _ { \\theta } ( x _ { i _ { j } } | x _ { 1 - ( b - \\sum _ { k = 1 } ^ { j - 1 } e _ { i _ { k } } ) } , b - \\sum _ { k = 1 } ^ { j - 1 } e _ { i _ { k } } )", + "type": "interline_equation", + "image_path": "29a588d63f7f9b56a9a4a873e8bb8c472a9609e06bfe1de20bf3ead78d6521b2.jpg" + } + ] + } + ], + "index": 21, + "virtual_lines": [ + { + "bbox": [ + 172, + 483, + 441, + 495.0 + ], + "spans": [], + "index": 20 + }, + { + "bbox": [ + 172, + 495.0, + 441, + 507.0 + ], + "spans": [], + "index": 21 + }, + { + "bbox": [ + 172, + 507.0, + 441, + 519.0 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 91, + 522, + 456, + 536 + ], + "lines": [ + { + "bbox": [ + 88, + 520, + 456, + 537 + ], + "spans": [ + { + "bbox": [ + 88, + 520, + 144, + 537 + ], + "score": 1.0, + "content": "For example,", + "type": "text" + }, + { + "bbox": [ + 145, + 522, + 456, + 535 + ], + "score": 0.72, + "content": "p _ { \\theta } ( x _ { 1 } , x _ { 4 } , x _ { 5 } | x _ { 2 } , x _ { 3 } ) = p _ { \\theta } ( x _ { 4 } | x _ { 2 } , x _ { 3 } ) p _ { \\theta } ( x _ { 1 } | x _ { 2 } , x _ { 3 } , x _ { 4 } ) p _ { \\theta } ( x _ { 5 } | x _ { 1 } , x _ { 2 } , x _ { 3 } , x _ { 4 } ) .", + "type": "inline_equation", + "image_path": "8856627f2edacd04d09da432d21e523edd0fdba8ee175da45ab1eb0502524149.jpg" + } + ], + "index": 23 + } + ], + "index": 23, + "bbox_fs": [ + 88, + 520, + 456, + 537 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 539, + 521, + 584 + ], + "lines": [ + { + "bbox": [ + 90, + 538, + 521, + 552 + ], + "spans": [ + { + "bbox": [ + 90, + 538, + 413, + 552 + ], + "score": 1.0, + "content": "Conditional sampling or conditional likelihood estimation for one object requires", + "type": "text" + }, + { + "bbox": [ + 414, + 540, + 424, + 551 + ], + "score": 0.89, + "content": "| b |", + "type": "inline_equation" + }, + { + "bbox": [ + 424, + 538, + 521, + 552 + ], + "score": 1.0, + "content": "requests to UM to com-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 89, + 551, + 521, + 563 + ], + "spans": [ + { + "bbox": [ + 89, + 551, + 110, + 562 + ], + "score": 1.0, + "content": "pute", + "type": "text" + }, + { + "bbox": [ + 111, + 551, + 169, + 563 + ], + "score": 0.93, + "content": "p _ { \\theta } ( x _ { i } | x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 169, + 551, + 521, + 562 + ], + "score": 1.0, + "content": ". Each request is a forward pass through the neural network. In the case of conditional", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 89, + 561, + 522, + 575 + ], + "spans": [ + { + "bbox": [ + 89, + 561, + 522, + 575 + ], + "score": 1.0, + "content": "sampling those requests even cannot be paralleled because the input of the next request contains the output", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 89, + 571, + 172, + 586 + ], + "spans": [ + { + "bbox": [ + 89, + 571, + 172, + 586 + ], + "score": 1.0, + "content": "of the previous one.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5, + "bbox_fs": [ + 89, + 538, + 522, + 586 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 589, + 521, + 622 + ], + "lines": [ + { + "bbox": [ + 90, + 588, + 521, + 602 + ], + "spans": [ + { + "bbox": [ + 90, + 588, + 521, + 602 + ], + "score": 1.0, + "content": "We propose a slight modification of the original UM training procedure which allows learning UM efficiently", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 90, + 600, + 522, + 613 + ], + "spans": [ + { + "bbox": [ + 90, + 600, + 522, + 613 + ], + "score": 1.0, + "content": "for any kind of masks including those considered in this paper. The details of the modification are described", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 88, + 611, + 159, + 624 + ], + "spans": [ + { + "bbox": [ + 88, + 611, + 159, + 624 + ], + "score": 1.0, + "content": "in appendix B.3.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29, + "bbox_fs": [ + 88, + 588, + 522, + 624 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 109, + 295, + 264 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 109, + 295, + 264 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 109, + 295, + 264 + ], + "spans": [ + { + "bbox": [ + 108, + 109, + 295, + 264 + ], + "score": 0.972, + "type": "image", + "image_path": "d8e3304c44f6641f3fcf46844d8ec076bf559af92021199814e73a3e8e86af57.jpg" + } + ] + } + ], + "index": 4.5, + "virtual_lines": [ + { + "bbox": [ + 108, + 109, + 295, + 124.5 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 124.5, + 295, + 140.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 140.0, + 295, + 155.5 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 108, + 155.5, + 295, + 171.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 108, + 171.0, + 295, + 186.5 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 108, + 186.5, + 295, + 202.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 108, + 202.0, + 295, + 217.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 108, + 217.5, + 295, + 233.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 108, + 233.0, + 295, + 248.5 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 108, + 248.5, + 295, + 264.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 137, + 272, + 258, + 284 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 137, + 272, + 259, + 286 + ], + "spans": [ + { + "bbox": [ + 137, + 272, + 259, + 286 + ], + "score": 1.0, + "content": "Figure 1: MNIST inpaintings.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + } + ], + "index": 7.25 + }, + { + "type": "image", + "bbox": [ + 324, + 109, + 509, + 264 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 324, + 109, + 509, + 264 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 324, + 109, + 509, + 264 + ], + "spans": [ + { + "bbox": [ + 324, + 109, + 509, + 264 + ], + "score": 0.973, + "type": "image", + "image_path": "696a7457eff5fb7c33faf339e1df45fda1cdfba4264c7adccc10fee5a551d327.jpg" + } + ] + } + ], + "index": 15.5, + "virtual_lines": [ + { + "bbox": [ + 324, + 109, + 509, + 124.5 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 324, + 124.5, + 509, + 140.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 324, + 140.0, + 509, + 155.5 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 324, + 155.5, + 509, + 171.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 324, + 171.0, + 509, + 186.5 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 324, + 186.5, + 509, + 202.0 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 324, + 202.0, + 509, + 217.5 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 324, + 217.5, + 509, + 233.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 324, + 233.0, + 509, + 248.5 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 324, + 248.5, + 509, + 264.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 351, + 275, + 480, + 286 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 350, + 273, + 480, + 289 + ], + "spans": [ + { + "bbox": [ + 350, + 273, + 480, + 289 + ], + "score": 1.0, + "content": "Figure 2: Omniglot inpaintings.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + } + ], + "index": 18.25 + }, + { + "type": "image", + "bbox": [ + 99, + 307, + 301, + 585 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 99, + 307, + 301, + 585 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 99, + 307, + 301, + 585 + ], + "spans": [ + { + "bbox": [ + 99, + 307, + 301, + 585 + ], + "score": 0.973, + "type": "image", + "image_path": "22952789d3ea86e1798ed0fcb93d6b5ec6a1f7740b49980530ccffa387686d74.jpg" + } + ] + } + ], + "index": 30.5, + "virtual_lines": [ + { + "bbox": [ + 99, + 307, + 301, + 322.44444444444446 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 99, + 322.44444444444446, + 301, + 337.8888888888889 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 99, + 337.8888888888889, + 301, + 353.33333333333337 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 99, + 353.33333333333337, + 301, + 368.7777777777778 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 99, + 368.7777777777778, + 301, + 384.2222222222223 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 99, + 384.2222222222223, + 301, + 399.66666666666674 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 99, + 399.66666666666674, + 301, + 415.1111111111112 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 99, + 415.1111111111112, + 301, + 430.55555555555566 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 99, + 430.55555555555566, + 301, + 446.0000000000001 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 99, + 446.0000000000001, + 301, + 461.44444444444457 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 99, + 461.44444444444457, + 301, + 476.888888888889 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 99, + 476.888888888889, + 301, + 492.3333333333335 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 99, + 492.3333333333335, + 301, + 507.77777777777794 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 99, + 507.77777777777794, + 301, + 523.2222222222224 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 99, + 523.2222222222224, + 301, + 538.6666666666669 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 99, + 538.6666666666669, + 301, + 554.1111111111113 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 99, + 554.1111111111113, + 301, + 569.5555555555558 + ], + "spans": [], + "index": 38 + }, + { + "bbox": [ + 99, + 569.5555555555558, + 301, + 585.0000000000002 + ], + "spans": [], + "index": 39 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 139, + 596, + 259, + 607 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 137, + 594, + 260, + 610 + ], + "spans": [ + { + "bbox": [ + 137, + 594, + 260, + 610 + ], + "score": 1.0, + "content": "Figure 3: CelebA inpaintings.", + "type": "text" + } + ], + "index": 58 + } + ], + "index": 58 + } + ], + "index": 44.25 + }, + { + "type": "image", + "bbox": [ + 318, + 305, + 514, + 584 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 318, + 305, + 514, + 584 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 318, + 305, + 514, + 584 + ], + "spans": [ + { + "bbox": [ + 318, + 305, + 514, + 584 + ], + "score": 0.974, + "type": "image", + "image_path": "1a5cbe58c9eb61a7543d31d330e9a0fdd262a2099e64f42864b8627f3f719b25.jpg" + } + ] + } + ], + "index": 48.5, + "virtual_lines": [ + { + "bbox": [ + 318, + 305, + 514, + 320.5 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 318, + 320.5, + 514, + 336.0 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 318, + 336.0, + 514, + 351.5 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 318, + 351.5, + 514, + 367.0 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 318, + 367.0, + 514, + 382.5 + ], + "spans": [], + "index": 44 + }, + { + "bbox": [ + 318, + 382.5, + 514, + 398.0 + ], + "spans": [], + "index": 45 + }, + { + "bbox": [ + 318, + 398.0, + 514, + 413.5 + ], + "spans": [], + "index": 46 + }, + { + "bbox": [ + 318, + 413.5, + 514, + 429.0 + ], + "spans": [], + "index": 47 + }, + { + "bbox": [ + 318, + 429.0, + 514, + 444.5 + ], + "spans": [], + "index": 48 + }, + { + "bbox": [ + 318, + 444.5, + 514, + 460.0 + ], + "spans": [], + "index": 49 + }, + { + "bbox": [ + 318, + 460.0, + 514, + 475.5 + ], + "spans": [], + "index": 50 + }, + { + "bbox": [ + 318, + 475.5, + 514, + 491.0 + ], + "spans": [], + "index": 51 + }, + { + "bbox": [ + 318, + 491.0, + 514, + 506.5 + ], + "spans": [], + "index": 52 + }, + { + "bbox": [ + 318, + 506.5, + 514, + 522.0 + ], + "spans": [], + "index": 53 + }, + { + "bbox": [ + 318, + 522.0, + 514, + 537.5 + ], + "spans": [], + "index": 54 + }, + { + "bbox": [ + 318, + 537.5, + 514, + 553.0 + ], + "spans": [], + "index": 55 + }, + { + "bbox": [ + 318, + 553.0, + 514, + 568.5 + ], + "spans": [], + "index": 56 + }, + { + "bbox": [ + 318, + 568.5, + 514, + 584.0 + ], + "spans": [], + "index": 57 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 309, + 596, + 521, + 618 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 309, + 595, + 521, + 608 + ], + "spans": [ + { + "bbox": [ + 309, + 595, + 521, + 608 + ], + "score": 1.0, + "content": "Figure 4: CelebA inpaintings with masks from (Yeh", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 309, + 606, + 363, + 619 + ], + "spans": [ + { + "bbox": [ + 309, + 606, + 363, + 619 + ], + "score": 1.0, + "content": "et al., 2017).", + "type": "text" + } + ], + "index": 60 + } + ], + "index": 59.5 + } + ], + "index": 54.0 + }, + { + "type": "text", + "bbox": [ + 107, + 625, + 499, + 637 + ], + "lines": [ + { + "bbox": [ + 109, + 623, + 501, + 640 + ], + "spans": [ + { + "bbox": [ + 109, + 623, + 501, + 640 + ], + "score": 1.0, + "content": "Left: input. The gray pixels are unobserved. Middle: samples from VAEAC. Right: ground truth.", + "type": "text" + } + ], + "index": 61 + } + ], + "index": 61 + } + ], + "page_idx": 8, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 40, + 277, + 50 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 286, + 671, + 293, + 680 + ], + "lines": [ + { + "bbox": [ + 285, + 670, + 293, + 681 + ], + "spans": [ + { + "bbox": [ + 285, + 670, + 293, + 681 + ], + "score": 1.0, + "content": "9", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 108, + 109, + 295, + 264 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 108, + 109, + 295, + 264 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 108, + 109, + 295, + 264 + ], + "spans": [ + { + "bbox": [ + 108, + 109, + 295, + 264 + ], + "score": 0.972, + "type": "image", + "image_path": "d8e3304c44f6641f3fcf46844d8ec076bf559af92021199814e73a3e8e86af57.jpg" + } + ] + } + ], + "index": 4.5, + "virtual_lines": [ + { + "bbox": [ + 108, + 109, + 295, + 124.5 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 108, + 124.5, + 295, + 140.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 108, + 140.0, + 295, + 155.5 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 108, + 155.5, + 295, + 171.0 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 108, + 171.0, + 295, + 186.5 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 108, + 186.5, + 295, + 202.0 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 108, + 202.0, + 295, + 217.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 108, + 217.5, + 295, + 233.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 108, + 233.0, + 295, + 248.5 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 108, + 248.5, + 295, + 264.0 + ], + "spans": [], + "index": 9 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 137, + 272, + 258, + 284 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 137, + 272, + 259, + 286 + ], + "spans": [ + { + "bbox": [ + 137, + 272, + 259, + 286 + ], + "score": 1.0, + "content": "Figure 1: MNIST inpaintings.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + } + ], + "index": 7.25 + }, + { + "type": "image", + "bbox": [ + 324, + 109, + 509, + 264 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 324, + 109, + 509, + 264 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 324, + 109, + 509, + 264 + ], + "spans": [ + { + "bbox": [ + 324, + 109, + 509, + 264 + ], + "score": 0.973, + "type": "image", + "image_path": "696a7457eff5fb7c33faf339e1df45fda1cdfba4264c7adccc10fee5a551d327.jpg" + } + ] + } + ], + "index": 15.5, + "virtual_lines": [ + { + "bbox": [ + 324, + 109, + 509, + 124.5 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 324, + 124.5, + 509, + 140.0 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 324, + 140.0, + 509, + 155.5 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 324, + 155.5, + 509, + 171.0 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 324, + 171.0, + 509, + 186.5 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 324, + 186.5, + 509, + 202.0 + ], + "spans": [], + "index": 16 + }, + { + "bbox": [ + 324, + 202.0, + 509, + 217.5 + ], + "spans": [], + "index": 17 + }, + { + "bbox": [ + 324, + 217.5, + 509, + 233.0 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 324, + 233.0, + 509, + 248.5 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 324, + 248.5, + 509, + 264.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 351, + 275, + 480, + 286 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 350, + 273, + 480, + 289 + ], + "spans": [ + { + "bbox": [ + 350, + 273, + 480, + 289 + ], + "score": 1.0, + "content": "Figure 2: Omniglot inpaintings.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + } + ], + "index": 18.25 + }, + { + "type": "image", + "bbox": [ + 99, + 307, + 301, + 585 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 99, + 307, + 301, + 585 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 99, + 307, + 301, + 585 + ], + "spans": [ + { + "bbox": [ + 99, + 307, + 301, + 585 + ], + "score": 0.973, + "type": "image", + "image_path": "22952789d3ea86e1798ed0fcb93d6b5ec6a1f7740b49980530ccffa387686d74.jpg" + } + ] + } + ], + "index": 30.5, + "virtual_lines": [ + { + "bbox": [ + 99, + 307, + 301, + 322.44444444444446 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 99, + 322.44444444444446, + 301, + 337.8888888888889 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 99, + 337.8888888888889, + 301, + 353.33333333333337 + ], + "spans": [], + "index": 24 + }, + { + "bbox": [ + 99, + 353.33333333333337, + 301, + 368.7777777777778 + ], + "spans": [], + "index": 25 + }, + { + "bbox": [ + 99, + 368.7777777777778, + 301, + 384.2222222222223 + ], + "spans": [], + "index": 26 + }, + { + "bbox": [ + 99, + 384.2222222222223, + 301, + 399.66666666666674 + ], + "spans": [], + "index": 27 + }, + { + "bbox": [ + 99, + 399.66666666666674, + 301, + 415.1111111111112 + ], + "spans": [], + "index": 28 + }, + { + "bbox": [ + 99, + 415.1111111111112, + 301, + 430.55555555555566 + ], + "spans": [], + "index": 29 + }, + { + "bbox": [ + 99, + 430.55555555555566, + 301, + 446.0000000000001 + ], + "spans": [], + "index": 30 + }, + { + "bbox": [ + 99, + 446.0000000000001, + 301, + 461.44444444444457 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 99, + 461.44444444444457, + 301, + 476.888888888889 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 99, + 476.888888888889, + 301, + 492.3333333333335 + ], + "spans": [], + "index": 33 + }, + { + "bbox": [ + 99, + 492.3333333333335, + 301, + 507.77777777777794 + ], + "spans": [], + "index": 34 + }, + { + "bbox": [ + 99, + 507.77777777777794, + 301, + 523.2222222222224 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 99, + 523.2222222222224, + 301, + 538.6666666666669 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 99, + 538.6666666666669, + 301, + 554.1111111111113 + ], + "spans": [], + "index": 37 + }, + { + "bbox": [ + 99, + 554.1111111111113, + 301, + 569.5555555555558 + ], + "spans": [], + "index": 38 + }, + { + "bbox": [ + 99, + 569.5555555555558, + 301, + 585.0000000000002 + ], + "spans": [], + "index": 39 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 139, + 596, + 259, + 607 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 137, + 594, + 260, + 610 + ], + "spans": [ + { + "bbox": [ + 137, + 594, + 260, + 610 + ], + "score": 1.0, + "content": "Figure 3: CelebA inpaintings.", + "type": "text" + } + ], + "index": 58 + } + ], + "index": 58 + } + ], + "index": 44.25 + }, + { + "type": "image", + "bbox": [ + 318, + 305, + 514, + 584 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 318, + 305, + 514, + 584 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 318, + 305, + 514, + 584 + ], + "spans": [ + { + "bbox": [ + 318, + 305, + 514, + 584 + ], + "score": 0.974, + "type": "image", + "image_path": "1a5cbe58c9eb61a7543d31d330e9a0fdd262a2099e64f42864b8627f3f719b25.jpg" + } + ] + } + ], + "index": 48.5, + "virtual_lines": [ + { + "bbox": [ + 318, + 305, + 514, + 320.5 + ], + "spans": [], + "index": 40 + }, + { + "bbox": [ + 318, + 320.5, + 514, + 336.0 + ], + "spans": [], + "index": 41 + }, + { + "bbox": [ + 318, + 336.0, + 514, + 351.5 + ], + "spans": [], + "index": 42 + }, + { + "bbox": [ + 318, + 351.5, + 514, + 367.0 + ], + "spans": [], + "index": 43 + }, + { + "bbox": [ + 318, + 367.0, + 514, + 382.5 + ], + "spans": [], + "index": 44 + }, + { + "bbox": [ + 318, + 382.5, + 514, + 398.0 + ], + "spans": [], + "index": 45 + }, + { + "bbox": [ + 318, + 398.0, + 514, + 413.5 + ], + "spans": [], + "index": 46 + }, + { + "bbox": [ + 318, + 413.5, + 514, + 429.0 + ], + "spans": [], + "index": 47 + }, + { + "bbox": [ + 318, + 429.0, + 514, + 444.5 + ], + "spans": [], + "index": 48 + }, + { + "bbox": [ + 318, + 444.5, + 514, + 460.0 + ], + "spans": [], + "index": 49 + }, + { + "bbox": [ + 318, + 460.0, + 514, + 475.5 + ], + "spans": [], + "index": 50 + }, + { + "bbox": [ + 318, + 475.5, + 514, + 491.0 + ], + "spans": [], + "index": 51 + }, + { + "bbox": [ + 318, + 491.0, + 514, + 506.5 + ], + "spans": [], + "index": 52 + }, + { + "bbox": [ + 318, + 506.5, + 514, + 522.0 + ], + "spans": [], + "index": 53 + }, + { + "bbox": [ + 318, + 522.0, + 514, + 537.5 + ], + "spans": [], + "index": 54 + }, + { + "bbox": [ + 318, + 537.5, + 514, + 553.0 + ], + "spans": [], + "index": 55 + }, + { + "bbox": [ + 318, + 553.0, + 514, + 568.5 + ], + "spans": [], + "index": 56 + }, + { + "bbox": [ + 318, + 568.5, + 514, + 584.0 + ], + "spans": [], + "index": 57 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 309, + 596, + 521, + 618 + ], + "group_id": 3, + "lines": [ + { + "bbox": [ + 309, + 595, + 521, + 608 + ], + "spans": [ + { + "bbox": [ + 309, + 595, + 521, + 608 + ], + "score": 1.0, + "content": "Figure 4: CelebA inpaintings with masks from (Yeh", + "type": "text" + } + ], + "index": 59 + }, + { + "bbox": [ + 309, + 606, + 363, + 619 + ], + "spans": [ + { + "bbox": [ + 309, + 606, + 363, + 619 + ], + "score": 1.0, + "content": "et al., 2017).", + "type": "text" + } + ], + "index": 60 + } + ], + "index": 59.5 + } + ], + "index": 54.0 + }, + { + "type": "text", + "bbox": [ + 107, + 625, + 499, + 637 + ], + "lines": [ + { + "bbox": [ + 109, + 623, + 501, + 640 + ], + "spans": [ + { + "bbox": [ + 109, + 623, + 501, + 640 + ], + "score": 1.0, + "content": "Left: input. The gray pixels are unobserved. Middle: samples from VAEAC. Right: ground truth.", + "type": "text" + } + ], + "index": 61 + } + ], + "index": 61, + "bbox_fs": [ + 109, + 623, + 501, + 640 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 182, + 112, + 431, + 160 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 205, + 93, + 405, + 105 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 203, + 92, + 407, + 106 + ], + "spans": [ + { + "bbox": [ + 203, + 92, + 407, + 106 + ], + "score": 1.0, + "content": "Table 5: VAEAC and UM comparison on MNIST.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 182, + 112, + 431, + 160 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 182, + 112, + 431, + 160 + ], + "spans": [ + { + "bbox": [ + 182, + 112, + 431, + 160 + ], + "score": 0.97, + "html": "
MethodVAEACUM
Negative log-likelihood6141
Training time (30 epochs)5min 47s3min 14s
Test time (1OO samples generation)0.7ms1s
", + "type": "table", + "image_path": "cbb12bac85d7349bfd6184e4aa519378b2ab3130d2d4cea064e8fd8f042f3272.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 182, + 112, + 431, + 128.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 182, + 128.0, + 431, + 144.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 182, + 144.0, + 431, + 160.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "text", + "bbox": [ + 90, + 182, + 521, + 271 + ], + "lines": [ + { + "bbox": [ + 90, + 183, + 522, + 195 + ], + "spans": [ + { + "bbox": [ + 90, + 183, + 522, + 195 + ], + "score": 1.0, + "content": "The results of using this modification of UM are provided in table 5. We can say that the relation between", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 89, + 192, + 523, + 206 + ], + "spans": [ + { + "bbox": [ + 89, + 192, + 523, + 206 + ], + "score": 1.0, + "content": "VAEAC and UM is similar to the relation between VAE and PixelCNN. The second one is much slower", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 89, + 205, + 522, + 217 + ], + "spans": [ + { + "bbox": [ + 89, + 205, + 522, + 217 + ], + "score": 1.0, + "content": "at the testing stage, but it easily takes into account local dependencies in data while the first one is faster", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 89, + 216, + 521, + 228 + ], + "spans": [ + { + "bbox": [ + 89, + 216, + 521, + 228 + ], + "score": 1.0, + "content": "but assumes conditional independence of the outputs. Nevertheless, there are a number of cases where", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 90, + 226, + 522, + 238 + ], + "spans": [ + { + "bbox": [ + 90, + 226, + 522, + 238 + ], + "score": 1.0, + "content": "UM cannot learn the distribution well while VAEAC can. For example, when the data is real-valued and", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 89, + 237, + 522, + 250 + ], + "spans": [ + { + "bbox": [ + 89, + 237, + 522, + 250 + ], + "score": 1.0, + "content": "marginal distributions have many local optima, there is no straightforward parametrization which allows UM", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 89, + 248, + 522, + 261 + ], + "spans": [ + { + "bbox": [ + 89, + 248, + 522, + 261 + ], + "score": 1.0, + "content": "to approximate them, and, therefore also the conditioned joint distribution. An example of such distribution", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 89, + 259, + 446, + 272 + ], + "spans": [ + { + "bbox": [ + 89, + 259, + 446, + 272 + ], + "score": 1.0, + "content": "and more illustrations for comparison of VAEAC and UM are available in appendix D.5.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 7.5 + }, + { + "type": "title", + "bbox": [ + 91, + 288, + 179, + 301 + ], + "lines": [ + { + "bbox": [ + 88, + 286, + 181, + 304 + ], + "spans": [ + { + "bbox": [ + 88, + 286, + 181, + 304 + ], + "score": 1.0, + "content": "6 CONCLUSION", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 91, + 314, + 521, + 392 + ], + "lines": [ + { + "bbox": [ + 89, + 315, + 522, + 327 + ], + "spans": [ + { + "bbox": [ + 89, + 315, + 522, + 327 + ], + "score": 1.0, + "content": "In this paper we consider the problem of simultaneous learning of all conditional distributions for a vector.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 89, + 324, + 522, + 339 + ], + "spans": [ + { + "bbox": [ + 89, + 324, + 522, + 339 + ], + "score": 1.0, + "content": "This problem has a number of different special cases with practical applications. We propose neural network", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 89, + 335, + 522, + 350 + ], + "spans": [ + { + "bbox": [ + 89, + 335, + 522, + 350 + ], + "score": 1.0, + "content": "based probabilistic model for distribution conditioning learning with Gaussian latent variables. This model is", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 88, + 346, + 522, + 361 + ], + "spans": [ + { + "bbox": [ + 88, + 346, + 522, + 361 + ], + "score": 1.0, + "content": "scalable and efficient in inference and learning. We propose several tricks to improve optimization and give", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 89, + 358, + 521, + 371 + ], + "spans": [ + { + "bbox": [ + 89, + 358, + 521, + 371 + ], + "score": 1.0, + "content": "recommendations about hyperparameters choice. The model is successfully applied to feature imputation", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 90, + 369, + 522, + 382 + ], + "spans": [ + { + "bbox": [ + 90, + 369, + 522, + 382 + ], + "score": 1.0, + "content": "and inpainting tasks. The experimental results show that the model is competitive with state of the art", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 89, + 380, + 403, + 393 + ], + "spans": [ + { + "bbox": [ + 89, + 380, + 403, + 393 + ], + "score": 1.0, + "content": "methods for both missing features imputation and image inpainting problems.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 16 + }, + { + "type": "title", + "bbox": [ + 91, + 410, + 159, + 422 + ], + "lines": [ + { + "bbox": [ + 90, + 410, + 160, + 424 + ], + "spans": [ + { + "bbox": [ + 90, + 410, + 160, + 424 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 89, + 430, + 520, + 452 + ], + "lines": [ + { + "bbox": [ + 89, + 430, + 522, + 443 + ], + "spans": [ + { + "bbox": [ + 89, + 430, + 522, + 443 + ], + "score": 1.0, + "content": "Philip Bachman and Doina Precup. Data generation as sequential decision making. In Advances in Neural", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 99, + 441, + 324, + 453 + ], + "spans": [ + { + "bbox": [ + 99, + 441, + 324, + 453 + ], + "score": 1.0, + "content": "Information Processing Systems, pp. 3249–3257, 2015.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 90, + 462, + 521, + 507 + ], + "lines": [ + { + "bbox": [ + 89, + 462, + 522, + 477 + ], + "spans": [ + { + "bbox": [ + 89, + 462, + 522, + 477 + ], + "score": 1.0, + "content": "Marcelo Bertalmio, Guillermo Sapiro, Vincent Caselles, and Coloma Ballester. Image inpainting. In Pro-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 100, + 474, + 522, + 487 + ], + "spans": [ + { + "bbox": [ + 100, + 474, + 522, + 487 + ], + "score": 1.0, + "content": "ceedings of the 27th Annual Conference on Computer Graphics and Interactive Techniques, SIGGRAPH", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 100, + 484, + 522, + 498 + ], + "spans": [ + { + "bbox": [ + 100, + 484, + 522, + 498 + ], + "score": 1.0, + "content": "’00, pp. 417–424, New York, NY, USA, 2000. ACM Press/Addison-Wesley Publishing Co. ISBN 1-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 100, + 495, + 157, + 509 + ], + "spans": [ + { + "bbox": [ + 100, + 495, + 157, + 509 + ], + "score": 1.0, + "content": "58113-208-5.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5 + }, + { + "type": "text", + "bbox": [ + 88, + 518, + 520, + 541 + ], + "lines": [ + { + "bbox": [ + 89, + 518, + 522, + 532 + ], + "spans": [ + { + "bbox": [ + 89, + 518, + 522, + 532 + ], + "score": 1.0, + "content": "Florian Bordes, Sina Honari, and Pascal Vincent. Learning to generate samples from noise through infusion", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 99, + 529, + 397, + 542 + ], + "spans": [ + { + "bbox": [ + 99, + 529, + 397, + 542 + ], + "score": 1.0, + "content": "training. In International Conference on Learning Representations, 2017.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 89, + 551, + 519, + 574 + ], + "lines": [ + { + "bbox": [ + 90, + 550, + 521, + 564 + ], + "spans": [ + { + "bbox": [ + 90, + 550, + 521, + 564 + ], + "score": 1.0, + "content": "Yuri Burda, Roger Grosse, and Ruslan Salakhutdinov. Importance weighted autoencoders. arXiv preprint", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 99, + 563, + 203, + 573 + ], + "spans": [ + { + "bbox": [ + 99, + 563, + 203, + 573 + ], + "score": 1.0, + "content": "arXiv:1509.00519, 2015.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5 + }, + { + "type": "text", + "bbox": [ + 88, + 585, + 519, + 608 + ], + "lines": [ + { + "bbox": [ + 89, + 583, + 522, + 598 + ], + "spans": [ + { + "bbox": [ + 89, + 583, + 522, + 598 + ], + "score": 1.0, + "content": "S van Buuren and Karin Groothuis-Oudshoorn. mice: Multivariate imputation by chained equations in r.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 99, + 596, + 289, + 609 + ], + "spans": [ + { + "bbox": [ + 99, + 596, + 289, + 609 + ], + "score": 1.0, + "content": "Journal of statistical software, pp. 1–68, 2010.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 91, + 618, + 521, + 651 + ], + "lines": [ + { + "bbox": [ + 90, + 617, + 521, + 631 + ], + "spans": [ + { + "bbox": [ + 90, + 617, + 521, + 631 + ], + "score": 1.0, + "content": "Laura Douglas, Iliyan Zarov, Konstantinos Gourgoulias, Chris Lucas, Chris Hart, Adam Baker, Maneesh", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 100, + 629, + 521, + 642 + ], + "spans": [ + { + "bbox": [ + 100, + 629, + 521, + 642 + ], + "score": 1.0, + "content": "Sahani, Yura Perov, and Saurabh Johri. A universal marginalizer for amortized inference in generative", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 99, + 640, + 297, + 653 + ], + "spans": [ + { + "bbox": [ + 99, + 640, + 297, + 653 + ], + "score": 1.0, + "content": "models. arXiv preprint arXiv:1711.00695, 2017.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34 + } + ], + "page_idx": 9, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 40, + 277, + 50 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 284, + 671, + 295, + 680 + ], + "lines": [ + { + "bbox": [ + 283, + 670, + 297, + 684 + ], + "spans": [ + { + "bbox": [ + 283, + 670, + 297, + 684 + ], + "score": 1.0, + "content": "10", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 182, + 112, + 431, + 160 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 205, + 93, + 405, + 105 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 203, + 92, + 407, + 106 + ], + "spans": [ + { + "bbox": [ + 203, + 92, + 407, + 106 + ], + "score": 1.0, + "content": "Table 5: VAEAC and UM comparison on MNIST.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 182, + 112, + 431, + 160 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 182, + 112, + 431, + 160 + ], + "spans": [ + { + "bbox": [ + 182, + 112, + 431, + 160 + ], + "score": 0.97, + "html": "
MethodVAEACUM
Negative log-likelihood6141
Training time (30 epochs)5min 47s3min 14s
Test time (1OO samples generation)0.7ms1s
", + "type": "table", + "image_path": "cbb12bac85d7349bfd6184e4aa519378b2ab3130d2d4cea064e8fd8f042f3272.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 182, + 112, + 431, + 128.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 182, + 128.0, + 431, + 144.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 182, + 144.0, + 431, + 160.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "text", + "bbox": [ + 90, + 182, + 521, + 271 + ], + "lines": [ + { + "bbox": [ + 90, + 183, + 522, + 195 + ], + "spans": [ + { + "bbox": [ + 90, + 183, + 522, + 195 + ], + "score": 1.0, + "content": "The results of using this modification of UM are provided in table 5. We can say that the relation between", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 89, + 192, + 523, + 206 + ], + "spans": [ + { + "bbox": [ + 89, + 192, + 523, + 206 + ], + "score": 1.0, + "content": "VAEAC and UM is similar to the relation between VAE and PixelCNN. The second one is much slower", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 89, + 205, + 522, + 217 + ], + "spans": [ + { + "bbox": [ + 89, + 205, + 522, + 217 + ], + "score": 1.0, + "content": "at the testing stage, but it easily takes into account local dependencies in data while the first one is faster", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 89, + 216, + 521, + 228 + ], + "spans": [ + { + "bbox": [ + 89, + 216, + 521, + 228 + ], + "score": 1.0, + "content": "but assumes conditional independence of the outputs. Nevertheless, there are a number of cases where", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 90, + 226, + 522, + 238 + ], + "spans": [ + { + "bbox": [ + 90, + 226, + 522, + 238 + ], + "score": 1.0, + "content": "UM cannot learn the distribution well while VAEAC can. For example, when the data is real-valued and", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 89, + 237, + 522, + 250 + ], + "spans": [ + { + "bbox": [ + 89, + 237, + 522, + 250 + ], + "score": 1.0, + "content": "marginal distributions have many local optima, there is no straightforward parametrization which allows UM", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 89, + 248, + 522, + 261 + ], + "spans": [ + { + "bbox": [ + 89, + 248, + 522, + 261 + ], + "score": 1.0, + "content": "to approximate them, and, therefore also the conditioned joint distribution. An example of such distribution", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 89, + 259, + 446, + 272 + ], + "spans": [ + { + "bbox": [ + 89, + 259, + 446, + 272 + ], + "score": 1.0, + "content": "and more illustrations for comparison of VAEAC and UM are available in appendix D.5.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 7.5, + "bbox_fs": [ + 89, + 183, + 523, + 272 + ] + }, + { + "type": "title", + "bbox": [ + 91, + 288, + 179, + 301 + ], + "lines": [ + { + "bbox": [ + 88, + 286, + 181, + 304 + ], + "spans": [ + { + "bbox": [ + 88, + 286, + 181, + 304 + ], + "score": 1.0, + "content": "6 CONCLUSION", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 91, + 314, + 521, + 392 + ], + "lines": [ + { + "bbox": [ + 89, + 315, + 522, + 327 + ], + "spans": [ + { + "bbox": [ + 89, + 315, + 522, + 327 + ], + "score": 1.0, + "content": "In this paper we consider the problem of simultaneous learning of all conditional distributions for a vector.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 89, + 324, + 522, + 339 + ], + "spans": [ + { + "bbox": [ + 89, + 324, + 522, + 339 + ], + "score": 1.0, + "content": "This problem has a number of different special cases with practical applications. We propose neural network", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 89, + 335, + 522, + 350 + ], + "spans": [ + { + "bbox": [ + 89, + 335, + 522, + 350 + ], + "score": 1.0, + "content": "based probabilistic model for distribution conditioning learning with Gaussian latent variables. This model is", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 88, + 346, + 522, + 361 + ], + "spans": [ + { + "bbox": [ + 88, + 346, + 522, + 361 + ], + "score": 1.0, + "content": "scalable and efficient in inference and learning. We propose several tricks to improve optimization and give", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 89, + 358, + 521, + 371 + ], + "spans": [ + { + "bbox": [ + 89, + 358, + 521, + 371 + ], + "score": 1.0, + "content": "recommendations about hyperparameters choice. The model is successfully applied to feature imputation", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 90, + 369, + 522, + 382 + ], + "spans": [ + { + "bbox": [ + 90, + 369, + 522, + 382 + ], + "score": 1.0, + "content": "and inpainting tasks. The experimental results show that the model is competitive with state of the art", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 89, + 380, + 403, + 393 + ], + "spans": [ + { + "bbox": [ + 89, + 380, + 403, + 393 + ], + "score": 1.0, + "content": "methods for both missing features imputation and image inpainting problems.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 16, + "bbox_fs": [ + 88, + 315, + 522, + 393 + ] + }, + { + "type": "title", + "bbox": [ + 91, + 410, + 159, + 422 + ], + "lines": [ + { + "bbox": [ + 90, + 410, + 160, + 424 + ], + "spans": [ + { + "bbox": [ + 90, + 410, + 160, + 424 + ], + "score": 1.0, + "content": "REFERENCES", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 89, + 430, + 520, + 452 + ], + "lines": [ + { + "bbox": [ + 89, + 430, + 522, + 443 + ], + "spans": [ + { + "bbox": [ + 89, + 430, + 522, + 443 + ], + "score": 1.0, + "content": "Philip Bachman and Doina Precup. Data generation as sequential decision making. In Advances in Neural", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 99, + 441, + 324, + 453 + ], + "spans": [ + { + "bbox": [ + 99, + 441, + 324, + 453 + ], + "score": 1.0, + "content": "Information Processing Systems, pp. 3249–3257, 2015.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5, + "bbox_fs": [ + 89, + 430, + 522, + 453 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 462, + 521, + 507 + ], + "lines": [ + { + "bbox": [ + 89, + 462, + 522, + 477 + ], + "spans": [ + { + "bbox": [ + 89, + 462, + 522, + 477 + ], + "score": 1.0, + "content": "Marcelo Bertalmio, Guillermo Sapiro, Vincent Caselles, and Coloma Ballester. Image inpainting. In Pro-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 100, + 474, + 522, + 487 + ], + "spans": [ + { + "bbox": [ + 100, + 474, + 522, + 487 + ], + "score": 1.0, + "content": "ceedings of the 27th Annual Conference on Computer Graphics and Interactive Techniques, SIGGRAPH", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 100, + 484, + 522, + 498 + ], + "spans": [ + { + "bbox": [ + 100, + 484, + 522, + 498 + ], + "score": 1.0, + "content": "’00, pp. 417–424, New York, NY, USA, 2000. ACM Press/Addison-Wesley Publishing Co. ISBN 1-", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 100, + 495, + 157, + 509 + ], + "spans": [ + { + "bbox": [ + 100, + 495, + 157, + 509 + ], + "score": 1.0, + "content": "58113-208-5.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24.5, + "bbox_fs": [ + 89, + 462, + 522, + 509 + ] + }, + { + "type": "text", + "bbox": [ + 88, + 518, + 520, + 541 + ], + "lines": [ + { + "bbox": [ + 89, + 518, + 522, + 532 + ], + "spans": [ + { + "bbox": [ + 89, + 518, + 522, + 532 + ], + "score": 1.0, + "content": "Florian Bordes, Sina Honari, and Pascal Vincent. Learning to generate samples from noise through infusion", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 99, + 529, + 397, + 542 + ], + "spans": [ + { + "bbox": [ + 99, + 529, + 397, + 542 + ], + "score": 1.0, + "content": "training. In International Conference on Learning Representations, 2017.", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5, + "bbox_fs": [ + 89, + 518, + 522, + 542 + ] + }, + { + "type": "text", + "bbox": [ + 89, + 551, + 519, + 574 + ], + "lines": [ + { + "bbox": [ + 90, + 550, + 521, + 564 + ], + "spans": [ + { + "bbox": [ + 90, + 550, + 521, + 564 + ], + "score": 1.0, + "content": "Yuri Burda, Roger Grosse, and Ruslan Salakhutdinov. Importance weighted autoencoders. arXiv preprint", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 99, + 563, + 203, + 573 + ], + "spans": [ + { + "bbox": [ + 99, + 563, + 203, + 573 + ], + "score": 1.0, + "content": "arXiv:1509.00519, 2015.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29.5, + "bbox_fs": [ + 90, + 550, + 521, + 573 + ] + }, + { + "type": "text", + "bbox": [ + 88, + 585, + 519, + 608 + ], + "lines": [ + { + "bbox": [ + 89, + 583, + 522, + 598 + ], + "spans": [ + { + "bbox": [ + 89, + 583, + 522, + 598 + ], + "score": 1.0, + "content": "S van Buuren and Karin Groothuis-Oudshoorn. mice: Multivariate imputation by chained equations in r.", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 99, + 596, + 289, + 609 + ], + "spans": [ + { + "bbox": [ + 99, + 596, + 289, + 609 + ], + "score": 1.0, + "content": "Journal of statistical software, pp. 1–68, 2010.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 89, + 583, + 522, + 609 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 618, + 521, + 651 + ], + "lines": [ + { + "bbox": [ + 90, + 617, + 521, + 631 + ], + "spans": [ + { + "bbox": [ + 90, + 617, + 521, + 631 + ], + "score": 1.0, + "content": "Laura Douglas, Iliyan Zarov, Konstantinos Gourgoulias, Chris Lucas, Chris Hart, Adam Baker, Maneesh", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 100, + 629, + 521, + 642 + ], + "spans": [ + { + "bbox": [ + 100, + 629, + 521, + 642 + ], + "score": 1.0, + "content": "Sahani, Yura Perov, and Saurabh Johri. A universal marginalizer for amortized inference in generative", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 99, + 640, + 297, + 653 + ], + "spans": [ + { + "bbox": [ + 99, + 640, + 297, + 653 + ], + "score": 1.0, + "content": "models. arXiv preprint arXiv:1711.00695, 2017.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34, + "bbox_fs": [ + 90, + 617, + 521, + 653 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 90, + 95, + 521, + 150 + ], + "lines": [ + { + "bbox": [ + 90, + 96, + 521, + 108 + ], + "spans": [ + { + "bbox": [ + 90, + 96, + 521, + 108 + ], + "score": 1.0, + "content": "Mathieu Germain, Karol Gregor, Iain Murray, and Hugo Larochelle. Made: Masked autoencoder for distri-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 100, + 107, + 521, + 119 + ], + "spans": [ + { + "bbox": [ + 100, + 107, + 521, + 119 + ], + "score": 1.0, + "content": "bution estimation. In Francis Bach and David Blei (eds.), Proceedings of the 32nd International Confer-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 99, + 117, + 522, + 131 + ], + "spans": [ + { + "bbox": [ + 99, + 117, + 522, + 131 + ], + "score": 1.0, + "content": "ence on Machine Learning, volume 37 of Proceedings of Machine Learning Research, pp. 881–889, Lille,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 99, + 128, + 521, + 142 + ], + "spans": [ + { + "bbox": [ + 99, + 128, + 521, + 142 + ], + "score": 1.0, + "content": "France, 07–09 Jul 2015. PMLR. URL http://proceedings.mlr.press/v37/germain15.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 99, + 139, + 130, + 152 + ], + "spans": [ + { + "bbox": [ + 99, + 139, + 130, + 152 + ], + "score": 1.0, + "content": "html.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 90, + 158, + 522, + 192 + ], + "lines": [ + { + "bbox": [ + 89, + 157, + 522, + 172 + ], + "spans": [ + { + "bbox": [ + 89, + 157, + 522, + 172 + ], + "score": 1.0, + "content": "Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 99, + 167, + 522, + 184 + ], + "spans": [ + { + "bbox": [ + 99, + 167, + 522, + 184 + ], + "score": 1.0, + "content": "Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in neural information processing", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 99, + 181, + 225, + 192 + ], + "spans": [ + { + "bbox": [ + 99, + 181, + 225, + 192 + ], + "score": 1.0, + "content": "systems, pp. 2672–2680, 2014.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 91, + 199, + 519, + 233 + ], + "lines": [ + { + "bbox": [ + 90, + 199, + 522, + 211 + ], + "spans": [ + { + "bbox": [ + 90, + 199, + 522, + 211 + ], + "score": 1.0, + "content": "Anirudh Goyal Alias Parth Goyal, Nan Ke, Surya Ganguli, and Yoshua Bengio. Variational walkback:", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 98, + 209, + 522, + 223 + ], + "spans": [ + { + "bbox": [ + 98, + 209, + 522, + 223 + ], + "score": 1.0, + "content": "Learning a transition operator as a stochastic recurrent net. In Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 99, + 221, + 227, + 233 + ], + "spans": [ + { + "bbox": [ + 99, + 221, + 227, + 233 + ], + "score": 1.0, + "content": "Systems, pp. 4392–4402, 2017.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 88, + 239, + 519, + 263 + ], + "lines": [ + { + "bbox": [ + 90, + 240, + 521, + 253 + ], + "spans": [ + { + "bbox": [ + 90, + 240, + 521, + 253 + ], + "score": 1.0, + "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 99, + 251, + 516, + 264 + ], + "spans": [ + { + "bbox": [ + 99, + 251, + 516, + 264 + ], + "score": 1.0, + "content": "In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770–778, 2016.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 89, + 270, + 520, + 293 + ], + "lines": [ + { + "bbox": [ + 88, + 268, + 522, + 284 + ], + "spans": [ + { + "bbox": [ + 88, + 268, + 522, + 284 + ], + "score": 1.0, + "content": "Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 99, + 281, + 198, + 292 + ], + "spans": [ + { + "bbox": [ + 99, + 281, + 198, + 292 + ], + "score": 1.0, + "content": "arXiv:1412.6980, 2014.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + }, + { + "type": "text", + "bbox": [ + 92, + 299, + 499, + 312 + ], + "lines": [ + { + "bbox": [ + 90, + 299, + 500, + 313 + ], + "spans": [ + { + "bbox": [ + 90, + 299, + 500, + 313 + ], + "score": 1.0, + "content": "Diederik P. Kingma and Max Welling. Auto-encoding variational bayes. CoRR, abs/1312.6114, 2013.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 90, + 318, + 520, + 342 + ], + "lines": [ + { + "bbox": [ + 89, + 317, + 521, + 332 + ], + "spans": [ + { + "bbox": [ + 89, + 317, + 521, + 332 + ], + "score": 1.0, + "content": "Brenden M Lake, Ruslan Salakhutdinov, and Joshua B Tenenbaum. Human-level concept learning through", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 99, + 330, + 388, + 342 + ], + "spans": [ + { + "bbox": [ + 99, + 330, + 388, + 342 + ], + "score": 1.0, + "content": "probabilistic program induction. Science, 350(6266):1332–1338, 2015.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "text", + "bbox": [ + 91, + 348, + 518, + 361 + ], + "lines": [ + { + "bbox": [ + 90, + 347, + 516, + 361 + ], + "spans": [ + { + "bbox": [ + 90, + 347, + 516, + 361 + ], + "score": 1.0, + "content": "Yann LeCun, Corinna Cortes, and Christopher JC Burges. The mnist database of handwritten digits, 1998.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 90, + 367, + 520, + 391 + ], + "lines": [ + { + "bbox": [ + 90, + 367, + 522, + 381 + ], + "spans": [ + { + "bbox": [ + 90, + 367, + 522, + 381 + ], + "score": 1.0, + "content": "Yijun Li, Sifei Liu, Jimei Yang, and Ming-Hsuan Yang. Generative face completion. In The IEEE Conference", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 99, + 379, + 414, + 391 + ], + "spans": [ + { + "bbox": [ + 99, + 379, + 414, + 391 + ], + "score": 1.0, + "content": "on Computer Vision and Pattern Recognition (CVPR), volume 1, pp. 3, 2017.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 91, + 397, + 508, + 410 + ], + "lines": [ + { + "bbox": [ + 89, + 395, + 509, + 412 + ], + "spans": [ + { + "bbox": [ + 89, + 395, + 509, + 412 + ], + "score": 1.0, + "content": "M. Lichman. UCI machine learning repository, 2013. URL http://archive.ics.uci.edu/ml.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 90, + 416, + 519, + 440 + ], + "lines": [ + { + "bbox": [ + 89, + 416, + 521, + 429 + ], + "spans": [ + { + "bbox": [ + 89, + 416, + 521, + 429 + ], + "score": 1.0, + "content": "Ziwei Liu, Ping Luo, Xiaogang Wang, and Xiaoou Tang. Deep learning face attributes in the wild. In", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 100, + 428, + 451, + 441 + ], + "spans": [ + { + "bbox": [ + 100, + 428, + 451, + 441 + ], + "score": 1.0, + "content": "Proceedings of International Conference on Computer Vision (ICCV), December 2015.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 91, + 446, + 519, + 480 + ], + "lines": [ + { + "bbox": [ + 90, + 446, + 521, + 459 + ], + "spans": [ + { + "bbox": [ + 90, + 446, + 521, + 459 + ], + "score": 1.0, + "content": "Alireza Makhzani, Jonathon Shlens, Navdeep Jaitly, and Ian Goodfellow. Adversarial autoencoders. In In-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 99, + 458, + 520, + 470 + ], + "spans": [ + { + "bbox": [ + 99, + 458, + 520, + 470 + ], + "score": 1.0, + "content": "ternational Conference on Learning Representations, 2016. URL http://arxiv.org/abs/1511.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 99, + 469, + 137, + 480 + ], + "spans": [ + { + "bbox": [ + 99, + 469, + 137, + 480 + ], + "score": 1.0, + "content": "05644.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25 + }, + { + "type": "text", + "bbox": [ + 90, + 487, + 519, + 521 + ], + "lines": [ + { + "bbox": [ + 89, + 486, + 521, + 500 + ], + "spans": [ + { + "bbox": [ + 89, + 486, + 521, + 500 + ], + "score": 1.0, + "content": "Xiaojiao Mao, Chunhua Shen, and Yu-Bin Yang. Image restoration using very deep convolutional encoder-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 99, + 497, + 521, + 511 + ], + "spans": [ + { + "bbox": [ + 99, + 497, + 521, + 511 + ], + "score": 1.0, + "content": "decoder networks with symmetric skip connections. In Advances in neural information processing sys-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 99, + 509, + 213, + 521 + ], + "spans": [ + { + "bbox": [ + 99, + 509, + 213, + 521 + ], + "score": 1.0, + "content": "tems, pp. 2802–2810, 2016.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 89, + 527, + 515, + 541 + ], + "lines": [ + { + "bbox": [ + 90, + 528, + 511, + 541 + ], + "spans": [ + { + "bbox": [ + 90, + 528, + 511, + 541 + ], + "score": 1.0, + "content": "Mehdi Mirza and Simon Osindero. Conditional generative adversarial nets. CoRR, abs/1411.1784, 2014.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30 + }, + { + "type": "text", + "bbox": [ + 90, + 546, + 519, + 570 + ], + "lines": [ + { + "bbox": [ + 90, + 547, + 521, + 560 + ], + "spans": [ + { + "bbox": [ + 90, + 547, + 521, + 560 + ], + "score": 1.0, + "content": "Deepak Pathak, Philipp Krahenb ¨ uhl, Jeff Donahue, Trevor Darrell, and Alexei Efros. Context encoders: ¨", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 100, + 559, + 468, + 570 + ], + "spans": [ + { + "bbox": [ + 100, + 559, + 468, + 570 + ], + "score": 1.0, + "content": "Feature learning by inpainting. In Computer Vision and Pattern Recognition (CVPR), 2016.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5 + }, + { + "type": "text", + "bbox": [ + 91, + 577, + 519, + 611 + ], + "lines": [ + { + "bbox": [ + 89, + 576, + 521, + 590 + ], + "spans": [ + { + "bbox": [ + 89, + 576, + 521, + 590 + ], + "score": 1.0, + "content": "Danilo Jimenez Rezende, Shakir Mohamed, and Daan Wierstra. Stochastic backpropagation and approxi-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 98, + 587, + 522, + 601 + ], + "spans": [ + { + "bbox": [ + 98, + 587, + 522, + 601 + ], + "score": 1.0, + "content": "mate inference in deep generative models. In International Conference on Machine Learning, pp. 1278–", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 100, + 598, + 151, + 611 + ], + "spans": [ + { + "bbox": [ + 100, + 598, + 151, + 611 + ], + "score": 1.0, + "content": "1286, 2014.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34 + }, + { + "type": "text", + "bbox": [ + 91, + 618, + 521, + 652 + ], + "lines": [ + { + "bbox": [ + 89, + 615, + 521, + 632 + ], + "spans": [ + { + "bbox": [ + 89, + 615, + 521, + 632 + ], + "score": 1.0, + "content": "Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 99, + 628, + 521, + 644 + ], + "spans": [ + { + "bbox": [ + 99, + 628, + 521, + 644 + ], + "score": 1.0, + "content": "segmentation. In International Conference on Medical image computing and computer-assisted interven-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 99, + 640, + 240, + 652 + ], + "spans": [ + { + "bbox": [ + 99, + 640, + 240, + 652 + ], + "score": 1.0, + "content": "tion, pp. 234–241. Springer, 2015.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37 + } + ], + "page_idx": 10, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 40, + 277, + 51 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 284, + 671, + 294, + 680 + ], + "lines": [ + { + "bbox": [ + 283, + 670, + 296, + 684 + ], + "spans": [ + { + "bbox": [ + 283, + 670, + 296, + 684 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 90, + 95, + 521, + 150 + ], + "lines": [ + { + "bbox": [ + 90, + 96, + 521, + 108 + ], + "spans": [ + { + "bbox": [ + 90, + 96, + 521, + 108 + ], + "score": 1.0, + "content": "Mathieu Germain, Karol Gregor, Iain Murray, and Hugo Larochelle. Made: Masked autoencoder for distri-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 100, + 107, + 521, + 119 + ], + "spans": [ + { + "bbox": [ + 100, + 107, + 521, + 119 + ], + "score": 1.0, + "content": "bution estimation. In Francis Bach and David Blei (eds.), Proceedings of the 32nd International Confer-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 99, + 117, + 522, + 131 + ], + "spans": [ + { + "bbox": [ + 99, + 117, + 522, + 131 + ], + "score": 1.0, + "content": "ence on Machine Learning, volume 37 of Proceedings of Machine Learning Research, pp. 881–889, Lille,", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 99, + 128, + 521, + 142 + ], + "spans": [ + { + "bbox": [ + 99, + 128, + 521, + 142 + ], + "score": 1.0, + "content": "France, 07–09 Jul 2015. PMLR. URL http://proceedings.mlr.press/v37/germain15.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 99, + 139, + 130, + 152 + ], + "spans": [ + { + "bbox": [ + 99, + 139, + 130, + 152 + ], + "score": 1.0, + "content": "html.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2, + "bbox_fs": [ + 90, + 96, + 522, + 152 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 158, + 522, + 192 + ], + "lines": [ + { + "bbox": [ + 89, + 157, + 522, + 172 + ], + "spans": [ + { + "bbox": [ + 89, + 157, + 522, + 172 + ], + "score": 1.0, + "content": "Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 99, + 167, + 522, + 184 + ], + "spans": [ + { + "bbox": [ + 99, + 167, + 522, + 184 + ], + "score": 1.0, + "content": "Courville, and Yoshua Bengio. Generative adversarial nets. In Advances in neural information processing", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 99, + 181, + 225, + 192 + ], + "spans": [ + { + "bbox": [ + 99, + 181, + 225, + 192 + ], + "score": 1.0, + "content": "systems, pp. 2672–2680, 2014.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6, + "bbox_fs": [ + 89, + 157, + 522, + 192 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 199, + 519, + 233 + ], + "lines": [ + { + "bbox": [ + 90, + 199, + 522, + 211 + ], + "spans": [ + { + "bbox": [ + 90, + 199, + 522, + 211 + ], + "score": 1.0, + "content": "Anirudh Goyal Alias Parth Goyal, Nan Ke, Surya Ganguli, and Yoshua Bengio. Variational walkback:", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 98, + 209, + 522, + 223 + ], + "spans": [ + { + "bbox": [ + 98, + 209, + 522, + 223 + ], + "score": 1.0, + "content": "Learning a transition operator as a stochastic recurrent net. In Advances in Neural Information Processing", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 99, + 221, + 227, + 233 + ], + "spans": [ + { + "bbox": [ + 99, + 221, + 227, + 233 + ], + "score": 1.0, + "content": "Systems, pp. 4392–4402, 2017.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 90, + 199, + 522, + 233 + ] + }, + { + "type": "text", + "bbox": [ + 88, + 239, + 519, + 263 + ], + "lines": [ + { + "bbox": [ + 90, + 240, + 521, + 253 + ], + "spans": [ + { + "bbox": [ + 90, + 240, + 521, + 253 + ], + "score": 1.0, + "content": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition.", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 99, + 251, + 516, + 264 + ], + "spans": [ + { + "bbox": [ + 99, + 251, + 516, + 264 + ], + "score": 1.0, + "content": "In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770–778, 2016.", + "type": "text" + } + ], + "index": 12 + } + ], + "index": 11.5, + "bbox_fs": [ + 90, + 240, + 521, + 264 + ] + }, + { + "type": "text", + "bbox": [ + 89, + 270, + 520, + 293 + ], + "lines": [ + { + "bbox": [ + 88, + 268, + 522, + 284 + ], + "spans": [ + { + "bbox": [ + 88, + 268, + 522, + 284 + ], + "score": 1.0, + "content": "Diederik P Kingma and Jimmy Ba. Adam: A method for stochastic optimization. arXiv preprint", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 99, + 281, + 198, + 292 + ], + "spans": [ + { + "bbox": [ + 99, + 281, + 198, + 292 + ], + "score": 1.0, + "content": "arXiv:1412.6980, 2014.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5, + "bbox_fs": [ + 88, + 268, + 522, + 292 + ] + }, + { + "type": "text", + "bbox": [ + 92, + 299, + 499, + 312 + ], + "lines": [ + { + "bbox": [ + 90, + 299, + 500, + 313 + ], + "spans": [ + { + "bbox": [ + 90, + 299, + 500, + 313 + ], + "score": 1.0, + "content": "Diederik P. Kingma and Max Welling. Auto-encoding variational bayes. CoRR, abs/1312.6114, 2013.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15, + "bbox_fs": [ + 90, + 299, + 500, + 313 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 318, + 520, + 342 + ], + "lines": [ + { + "bbox": [ + 89, + 317, + 521, + 332 + ], + "spans": [ + { + "bbox": [ + 89, + 317, + 521, + 332 + ], + "score": 1.0, + "content": "Brenden M Lake, Ruslan Salakhutdinov, and Joshua B Tenenbaum. Human-level concept learning through", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 99, + 330, + 388, + 342 + ], + "spans": [ + { + "bbox": [ + 99, + 330, + 388, + 342 + ], + "score": 1.0, + "content": "probabilistic program induction. Science, 350(6266):1332–1338, 2015.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 89, + 317, + 521, + 342 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 348, + 518, + 361 + ], + "lines": [ + { + "bbox": [ + 90, + 347, + 516, + 361 + ], + "spans": [ + { + "bbox": [ + 90, + 347, + 516, + 361 + ], + "score": 1.0, + "content": "Yann LeCun, Corinna Cortes, and Christopher JC Burges. The mnist database of handwritten digits, 1998.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18, + "bbox_fs": [ + 90, + 347, + 516, + 361 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 367, + 520, + 391 + ], + "lines": [ + { + "bbox": [ + 90, + 367, + 522, + 381 + ], + "spans": [ + { + "bbox": [ + 90, + 367, + 522, + 381 + ], + "score": 1.0, + "content": "Yijun Li, Sifei Liu, Jimei Yang, and Ming-Hsuan Yang. Generative face completion. In The IEEE Conference", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 99, + 379, + 414, + 391 + ], + "spans": [ + { + "bbox": [ + 99, + 379, + 414, + 391 + ], + "score": 1.0, + "content": "on Computer Vision and Pattern Recognition (CVPR), volume 1, pp. 3, 2017.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 90, + 367, + 522, + 391 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 397, + 508, + 410 + ], + "lines": [ + { + "bbox": [ + 89, + 395, + 509, + 412 + ], + "spans": [ + { + "bbox": [ + 89, + 395, + 509, + 412 + ], + "score": 1.0, + "content": "M. Lichman. UCI machine learning repository, 2013. URL http://archive.ics.uci.edu/ml.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21, + "bbox_fs": [ + 89, + 395, + 509, + 412 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 416, + 519, + 440 + ], + "lines": [ + { + "bbox": [ + 89, + 416, + 521, + 429 + ], + "spans": [ + { + "bbox": [ + 89, + 416, + 521, + 429 + ], + "score": 1.0, + "content": "Ziwei Liu, Ping Luo, Xiaogang Wang, and Xiaoou Tang. Deep learning face attributes in the wild. In", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 100, + 428, + 451, + 441 + ], + "spans": [ + { + "bbox": [ + 100, + 428, + 451, + 441 + ], + "score": 1.0, + "content": "Proceedings of International Conference on Computer Vision (ICCV), December 2015.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 22.5, + "bbox_fs": [ + 89, + 416, + 521, + 441 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 446, + 519, + 480 + ], + "lines": [ + { + "bbox": [ + 90, + 446, + 521, + 459 + ], + "spans": [ + { + "bbox": [ + 90, + 446, + 521, + 459 + ], + "score": 1.0, + "content": "Alireza Makhzani, Jonathon Shlens, Navdeep Jaitly, and Ian Goodfellow. Adversarial autoencoders. In In-", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 99, + 458, + 520, + 470 + ], + "spans": [ + { + "bbox": [ + 99, + 458, + 520, + 470 + ], + "score": 1.0, + "content": "ternational Conference on Learning Representations, 2016. URL http://arxiv.org/abs/1511.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 99, + 469, + 137, + 480 + ], + "spans": [ + { + "bbox": [ + 99, + 469, + 137, + 480 + ], + "score": 1.0, + "content": "05644.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25, + "bbox_fs": [ + 90, + 446, + 521, + 480 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 487, + 519, + 521 + ], + "lines": [ + { + "bbox": [ + 89, + 486, + 521, + 500 + ], + "spans": [ + { + "bbox": [ + 89, + 486, + 521, + 500 + ], + "score": 1.0, + "content": "Xiaojiao Mao, Chunhua Shen, and Yu-Bin Yang. Image restoration using very deep convolutional encoder-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 99, + 497, + 521, + 511 + ], + "spans": [ + { + "bbox": [ + 99, + 497, + 521, + 511 + ], + "score": 1.0, + "content": "decoder networks with symmetric skip connections. In Advances in neural information processing sys-", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 99, + 509, + 213, + 521 + ], + "spans": [ + { + "bbox": [ + 99, + 509, + 213, + 521 + ], + "score": 1.0, + "content": "tems, pp. 2802–2810, 2016.", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 28, + "bbox_fs": [ + 89, + 486, + 521, + 521 + ] + }, + { + "type": "text", + "bbox": [ + 89, + 527, + 515, + 541 + ], + "lines": [ + { + "bbox": [ + 90, + 528, + 511, + 541 + ], + "spans": [ + { + "bbox": [ + 90, + 528, + 511, + 541 + ], + "score": 1.0, + "content": "Mehdi Mirza and Simon Osindero. Conditional generative adversarial nets. CoRR, abs/1411.1784, 2014.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 30, + "bbox_fs": [ + 90, + 528, + 511, + 541 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 546, + 519, + 570 + ], + "lines": [ + { + "bbox": [ + 90, + 547, + 521, + 560 + ], + "spans": [ + { + "bbox": [ + 90, + 547, + 521, + 560 + ], + "score": 1.0, + "content": "Deepak Pathak, Philipp Krahenb ¨ uhl, Jeff Donahue, Trevor Darrell, and Alexei Efros. Context encoders: ¨", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 100, + 559, + 468, + 570 + ], + "spans": [ + { + "bbox": [ + 100, + 559, + 468, + 570 + ], + "score": 1.0, + "content": "Feature learning by inpainting. In Computer Vision and Pattern Recognition (CVPR), 2016.", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31.5, + "bbox_fs": [ + 90, + 547, + 521, + 570 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 577, + 519, + 611 + ], + "lines": [ + { + "bbox": [ + 89, + 576, + 521, + 590 + ], + "spans": [ + { + "bbox": [ + 89, + 576, + 521, + 590 + ], + "score": 1.0, + "content": "Danilo Jimenez Rezende, Shakir Mohamed, and Daan Wierstra. Stochastic backpropagation and approxi-", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 98, + 587, + 522, + 601 + ], + "spans": [ + { + "bbox": [ + 98, + 587, + 522, + 601 + ], + "score": 1.0, + "content": "mate inference in deep generative models. In International Conference on Machine Learning, pp. 1278–", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 100, + 598, + 151, + 611 + ], + "spans": [ + { + "bbox": [ + 100, + 598, + 151, + 611 + ], + "score": 1.0, + "content": "1286, 2014.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34, + "bbox_fs": [ + 89, + 576, + 522, + 611 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 618, + 521, + 652 + ], + "lines": [ + { + "bbox": [ + 89, + 615, + 521, + 632 + ], + "spans": [ + { + "bbox": [ + 89, + 615, + 521, + 632 + ], + "score": 1.0, + "content": "Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-net: Convolutional networks for biomedical image", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 99, + 628, + 521, + 644 + ], + "spans": [ + { + "bbox": [ + 99, + 628, + 521, + 644 + ], + "score": 1.0, + "content": "segmentation. In International Conference on Medical image computing and computer-assisted interven-", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 99, + 640, + 240, + 652 + ], + "spans": [ + { + "bbox": [ + 99, + 640, + 240, + 652 + ], + "score": 1.0, + "content": "tion, pp. 234–241. Springer, 2015.", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37, + "bbox_fs": [ + 89, + 615, + 521, + 652 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 91, + 96, + 521, + 150 + ], + "lines": [ + { + "bbox": [ + 89, + 95, + 521, + 109 + ], + "spans": [ + { + "bbox": [ + 89, + 95, + 521, + 109 + ], + "score": 1.0, + "content": "Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 100, + 107, + 521, + 119 + ], + "spans": [ + { + "bbox": [ + 100, + 107, + 521, + 119 + ], + "score": 1.0, + "content": "using nonequilibrium thermodynamics. In Francis Bach and David Blei (eds.), Proceedings of the 32nd", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 99, + 117, + 521, + 131 + ], + "spans": [ + { + "bbox": [ + 99, + 117, + 521, + 131 + ], + "score": 1.0, + "content": "International Conference on Machine Learning, volume 37 of Proceedings of Machine Learning Re-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 99, + 128, + 520, + 141 + ], + "spans": [ + { + "bbox": [ + 99, + 128, + 520, + 141 + ], + "score": 1.0, + "content": "search, pp. 2256–2265, Lille, France, 07–09 Jul 2015. PMLR. URL http://proceedings.mlr.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 98, + 139, + 292, + 152 + ], + "spans": [ + { + "bbox": [ + 98, + 139, + 292, + 152 + ], + "score": 1.0, + "content": "press/v37/sohl-dickstein15.html.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2 + }, + { + "type": "text", + "bbox": [ + 90, + 159, + 521, + 192 + ], + "lines": [ + { + "bbox": [ + 89, + 158, + 520, + 172 + ], + "spans": [ + { + "bbox": [ + 89, + 158, + 520, + 172 + ], + "score": 1.0, + "content": "Kihyuk Sohn, Honglak Lee, and Xinchen Yan. Learning structured output representation using deep condi-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 99, + 168, + 521, + 183 + ], + "spans": [ + { + "bbox": [ + 99, + 168, + 521, + 183 + ], + "score": 1.0, + "content": "tional generative models. In C. Cortes, N. D. Lawrence, D. D. Lee, M. Sugiyama, and R. Garnett (eds.),", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 99, + 180, + 515, + 194 + ], + "spans": [ + { + "bbox": [ + 99, + 180, + 515, + 194 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems 28, pp. 3483–3491. Curran Associates, Inc., 2015.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 87, + 199, + 520, + 222 + ], + "lines": [ + { + "bbox": [ + 90, + 199, + 521, + 212 + ], + "spans": [ + { + "bbox": [ + 90, + 199, + 521, + 212 + ], + "score": 1.0, + "content": "Casper Kaae Sønderby, Tapani Raiko, Lars Maaløe, Søren Kaae Sønderby, and Ole Winther. Ladder varia-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 99, + 210, + 494, + 223 + ], + "spans": [ + { + "bbox": [ + 99, + 210, + 494, + 223 + ], + "score": 1.0, + "content": "tional autoencoders. In Advances in neural information processing systems, pp. 3738–3746, 2016.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 88, + 230, + 520, + 253 + ], + "lines": [ + { + "bbox": [ + 89, + 229, + 520, + 243 + ], + "spans": [ + { + "bbox": [ + 89, + 229, + 520, + 243 + ], + "score": 1.0, + "content": "Daniel J Stekhoven and Peter Buhlmann. Missforest - non-parametric missing value imputation for mixed-¨", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 99, + 241, + 296, + 253 + ], + "spans": [ + { + "bbox": [ + 99, + 241, + 296, + 253 + ], + "score": 1.0, + "content": "type data. Bioinformatics, 28(1):112–118, 2011.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 89, + 259, + 520, + 283 + ], + "lines": [ + { + "bbox": [ + 89, + 259, + 521, + 273 + ], + "spans": [ + { + "bbox": [ + 89, + 259, + 521, + 273 + ], + "score": 1.0, + "content": "Benigno Uria, Marc-Alexandre Cotˆ e, Karol Gregor, Iain Murray, and Hugo Larochelle. Neural autoregres- ´", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 100, + 271, + 492, + 283 + ], + "spans": [ + { + "bbox": [ + 100, + 271, + 492, + 283 + ], + "score": 1.0, + "content": "sive distribution estimation. The Journal of Machine Learning Research, 17(1):7184–7220, 2016.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5 + }, + { + "type": "text", + "bbox": [ + 91, + 289, + 520, + 324 + ], + "lines": [ + { + "bbox": [ + 90, + 289, + 520, + 303 + ], + "spans": [ + { + "bbox": [ + 90, + 289, + 520, + 303 + ], + "score": 1.0, + "content": "C. Yang, X. Lu, Z. Lin, E. Shechtman, O. Wang, and H. Li. High-resolution image inpainting using multi-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 99, + 300, + 522, + 314 + ], + "spans": [ + { + "bbox": [ + 99, + 300, + 522, + 314 + ], + "score": 1.0, + "content": "scale neural patch synthesis. In 2017 IEEE Conference on Computer Vision and Pattern Recognition", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 99, + 311, + 371, + 325 + ], + "spans": [ + { + "bbox": [ + 99, + 311, + 371, + 325 + ], + "score": 1.0, + "content": "(CVPR), pp. 4076–4084, July 2017. doi: 10.1109/CVPR.2017.434.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 88, + 331, + 520, + 353 + ], + "lines": [ + { + "bbox": [ + 88, + 330, + 522, + 345 + ], + "spans": [ + { + "bbox": [ + 88, + 330, + 522, + 345 + ], + "score": 1.0, + "content": "Raymond Yeh, Chen Chen, Teck-Yian Lim, Mark Hasegawa-Johnson, and Minh N. Do. Semantic image", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 99, + 342, + 420, + 354 + ], + "spans": [ + { + "bbox": [ + 99, + 342, + 420, + 354 + ], + "score": 1.0, + "content": "inpainting with perceptual and contextual losses. CoRR, abs/1607.07539, 2016.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 91, + 361, + 520, + 395 + ], + "lines": [ + { + "bbox": [ + 90, + 359, + 522, + 375 + ], + "spans": [ + { + "bbox": [ + 90, + 359, + 522, + 375 + ], + "score": 1.0, + "content": "Raymond A Yeh, Chen Chen, Teck Yian Lim, Alexander G Schwing, Mark Hasegawa-Johnson, and Minh N", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 100, + 372, + 522, + 385 + ], + "spans": [ + { + "bbox": [ + 100, + 372, + 522, + 385 + ], + "score": 1.0, + "content": "Do. Semantic image inpainting with deep generative models. In Proceedings of the IEEE Conference on", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 100, + 383, + 362, + 396 + ], + "spans": [ + { + "bbox": [ + 100, + 383, + 362, + 396 + ], + "score": 1.0, + "content": "Computer Vision and Pattern Recognition, pp. 5485–5493, 2017.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "text", + "bbox": [ + 91, + 402, + 521, + 458 + ], + "lines": [ + { + "bbox": [ + 89, + 402, + 522, + 415 + ], + "spans": [ + { + "bbox": [ + 89, + 402, + 522, + 415 + ], + "score": 1.0, + "content": "Jinsung Yoon, James Jordon, and Mihaela van der Schaar. GAIN: Missing data imputation using generative", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 99, + 412, + 521, + 426 + ], + "spans": [ + { + "bbox": [ + 99, + 412, + 521, + 426 + ], + "score": 1.0, + "content": "adversarial nets. In Jennifer Dy and Andreas Krause (eds.), Proceedings of the 35th International Con-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 99, + 425, + 521, + 437 + ], + "spans": [ + { + "bbox": [ + 99, + 425, + 521, + 437 + ], + "score": 1.0, + "content": "ference on Machine Learning, volume 80 of Proceedings of Machine Learning Research, pp. 5689–5698,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 99, + 434, + 521, + 448 + ], + "spans": [ + { + "bbox": [ + 99, + 434, + 521, + 448 + ], + "score": 1.0, + "content": "Stockholmsmssan, Stockholm Sweden, 10–15 Jul 2018. PMLR. URL http://proceedings.mlr.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 98, + 447, + 237, + 459 + ], + "spans": [ + { + "bbox": [ + 98, + 447, + 237, + 459 + ], + "score": 1.0, + "content": "press/v80/yoon18a.html.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24 + }, + { + "type": "title", + "bbox": [ + 92, + 479, + 145, + 491 + ], + "lines": [ + { + "bbox": [ + 90, + 479, + 146, + 493 + ], + "spans": [ + { + "bbox": [ + 90, + 479, + 146, + 493 + ], + "score": 1.0, + "content": "APPENDIX", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "title", + "bbox": [ + 92, + 506, + 241, + 519 + ], + "lines": [ + { + "bbox": [ + 90, + 505, + 243, + 520 + ], + "spans": [ + { + "bbox": [ + 90, + 505, + 243, + 520 + ], + "score": 1.0, + "content": "A EXPERIMENTAL DETAILS", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 92, + 531, + 277, + 542 + ], + "lines": [ + { + "bbox": [ + 90, + 530, + 278, + 543 + ], + "spans": [ + { + "bbox": [ + 90, + 530, + 278, + 543 + ], + "score": 1.0, + "content": "A.1 NEURAL NETWORK ARCHITECTURES", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 91, + 551, + 521, + 585 + ], + "lines": [ + { + "bbox": [ + 90, + 551, + 522, + 564 + ], + "spans": [ + { + "bbox": [ + 90, + 551, + 522, + 564 + ], + "score": 1.0, + "content": "In all experiments we use optimization method Adam (Kingma & Ba, 2014), skip-connections between prior", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 87, + 561, + 522, + 577 + ], + "spans": [ + { + "bbox": [ + 87, + 561, + 522, + 577 + ], + "score": 1.0, + "content": "network and generative network inspired by (Mao et al., 2016), (Sønderby et al., 2016) and (Ronneberger", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 89, + 573, + 451, + 586 + ], + "spans": [ + { + "bbox": [ + 89, + 573, + 451, + 586 + ], + "score": 1.0, + "content": "et al., 2015), and convolutional neural networks based on ResNet blocks (He et al., 2016).", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 91, + 590, + 521, + 635 + ], + "lines": [ + { + "bbox": [ + 90, + 590, + 521, + 603 + ], + "spans": [ + { + "bbox": [ + 90, + 590, + 521, + 603 + ], + "score": 1.0, + "content": "Without skip-connections all information for decoder goes through the latent variables. In image inpainting", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 89, + 600, + 521, + 614 + ], + "spans": [ + { + "bbox": [ + 89, + 600, + 521, + 614 + ], + "score": 1.0, + "content": "we found skip-connections very useful in both terms of log-likelihood improvement and the image realism,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 90, + 612, + 521, + 625 + ], + "spans": [ + { + "bbox": [ + 90, + 612, + 521, + 625 + ], + "score": 1.0, + "content": "because latent variables are responsible for the global information only while the local information passes", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 90, + 623, + 513, + 636 + ], + "spans": [ + { + "bbox": [ + 90, + 623, + 513, + 636 + ], + "score": 1.0, + "content": "through skip-connections. Therefore the border between image and inpainting becomes less conspicuous.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 90, + 640, + 366, + 651 + ], + "lines": [ + { + "bbox": [ + 89, + 639, + 369, + 653 + ], + "spans": [ + { + "bbox": [ + 89, + 639, + 369, + 653 + ], + "score": 1.0, + "content": "The main idea of neural networks architecture is reflected in figure 5.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37 + } + ], + "page_idx": 11, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 40, + 277, + 50 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 284, + 671, + 295, + 680 + ], + "lines": [ + { + "bbox": [ + 283, + 670, + 297, + 684 + ], + "spans": [ + { + "bbox": [ + 283, + 670, + 297, + 684 + ], + "score": 1.0, + "content": "12", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 91, + 96, + 521, + 150 + ], + "lines": [ + { + "bbox": [ + 89, + 95, + 521, + 109 + ], + "spans": [ + { + "bbox": [ + 89, + 95, + 521, + 109 + ], + "score": 1.0, + "content": "Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, and Surya Ganguli. Deep unsupervised learning", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 100, + 107, + 521, + 119 + ], + "spans": [ + { + "bbox": [ + 100, + 107, + 521, + 119 + ], + "score": 1.0, + "content": "using nonequilibrium thermodynamics. In Francis Bach and David Blei (eds.), Proceedings of the 32nd", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 99, + 117, + 521, + 131 + ], + "spans": [ + { + "bbox": [ + 99, + 117, + 521, + 131 + ], + "score": 1.0, + "content": "International Conference on Machine Learning, volume 37 of Proceedings of Machine Learning Re-", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 99, + 128, + 520, + 141 + ], + "spans": [ + { + "bbox": [ + 99, + 128, + 520, + 141 + ], + "score": 1.0, + "content": "search, pp. 2256–2265, Lille, France, 07–09 Jul 2015. PMLR. URL http://proceedings.mlr.", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 98, + 139, + 292, + 152 + ], + "spans": [ + { + "bbox": [ + 98, + 139, + 292, + 152 + ], + "score": 1.0, + "content": "press/v37/sohl-dickstein15.html.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 2, + "bbox_fs": [ + 89, + 95, + 521, + 152 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 159, + 521, + 192 + ], + "lines": [ + { + "bbox": [ + 89, + 158, + 520, + 172 + ], + "spans": [ + { + "bbox": [ + 89, + 158, + 520, + 172 + ], + "score": 1.0, + "content": "Kihyuk Sohn, Honglak Lee, and Xinchen Yan. Learning structured output representation using deep condi-", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 99, + 168, + 521, + 183 + ], + "spans": [ + { + "bbox": [ + 99, + 168, + 521, + 183 + ], + "score": 1.0, + "content": "tional generative models. In C. Cortes, N. D. Lawrence, D. D. Lee, M. Sugiyama, and R. Garnett (eds.),", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 99, + 180, + 515, + 194 + ], + "spans": [ + { + "bbox": [ + 99, + 180, + 515, + 194 + ], + "score": 1.0, + "content": "Advances in Neural Information Processing Systems 28, pp. 3483–3491. Curran Associates, Inc., 2015.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6, + "bbox_fs": [ + 89, + 158, + 521, + 194 + ] + }, + { + "type": "text", + "bbox": [ + 87, + 199, + 520, + 222 + ], + "lines": [ + { + "bbox": [ + 90, + 199, + 521, + 212 + ], + "spans": [ + { + "bbox": [ + 90, + 199, + 521, + 212 + ], + "score": 1.0, + "content": "Casper Kaae Sønderby, Tapani Raiko, Lars Maaløe, Søren Kaae Sønderby, and Ole Winther. Ladder varia-", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 99, + 210, + 494, + 223 + ], + "spans": [ + { + "bbox": [ + 99, + 210, + 494, + 223 + ], + "score": 1.0, + "content": "tional autoencoders. In Advances in neural information processing systems, pp. 3738–3746, 2016.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 90, + 199, + 521, + 223 + ] + }, + { + "type": "text", + "bbox": [ + 88, + 230, + 520, + 253 + ], + "lines": [ + { + "bbox": [ + 89, + 229, + 520, + 243 + ], + "spans": [ + { + "bbox": [ + 89, + 229, + 520, + 243 + ], + "score": 1.0, + "content": "Daniel J Stekhoven and Peter Buhlmann. Missforest - non-parametric missing value imputation for mixed-¨", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 99, + 241, + 296, + 253 + ], + "spans": [ + { + "bbox": [ + 99, + 241, + 296, + 253 + ], + "score": 1.0, + "content": "type data. Bioinformatics, 28(1):112–118, 2011.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5, + "bbox_fs": [ + 89, + 229, + 520, + 253 + ] + }, + { + "type": "text", + "bbox": [ + 89, + 259, + 520, + 283 + ], + "lines": [ + { + "bbox": [ + 89, + 259, + 521, + 273 + ], + "spans": [ + { + "bbox": [ + 89, + 259, + 521, + 273 + ], + "score": 1.0, + "content": "Benigno Uria, Marc-Alexandre Cotˆ e, Karol Gregor, Iain Murray, and Hugo Larochelle. Neural autoregres- ´", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 100, + 271, + 492, + 283 + ], + "spans": [ + { + "bbox": [ + 100, + 271, + 492, + 283 + ], + "score": 1.0, + "content": "sive distribution estimation. The Journal of Machine Learning Research, 17(1):7184–7220, 2016.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12.5, + "bbox_fs": [ + 89, + 259, + 521, + 283 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 289, + 520, + 324 + ], + "lines": [ + { + "bbox": [ + 90, + 289, + 520, + 303 + ], + "spans": [ + { + "bbox": [ + 90, + 289, + 520, + 303 + ], + "score": 1.0, + "content": "C. Yang, X. Lu, Z. Lin, E. Shechtman, O. Wang, and H. Li. High-resolution image inpainting using multi-", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 99, + 300, + 522, + 314 + ], + "spans": [ + { + "bbox": [ + 99, + 300, + 522, + 314 + ], + "score": 1.0, + "content": "scale neural patch synthesis. In 2017 IEEE Conference on Computer Vision and Pattern Recognition", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 99, + 311, + 371, + 325 + ], + "spans": [ + { + "bbox": [ + 99, + 311, + 371, + 325 + ], + "score": 1.0, + "content": "(CVPR), pp. 4076–4084, July 2017. doi: 10.1109/CVPR.2017.434.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15, + "bbox_fs": [ + 90, + 289, + 522, + 325 + ] + }, + { + "type": "text", + "bbox": [ + 88, + 331, + 520, + 353 + ], + "lines": [ + { + "bbox": [ + 88, + 330, + 522, + 345 + ], + "spans": [ + { + "bbox": [ + 88, + 330, + 522, + 345 + ], + "score": 1.0, + "content": "Raymond Yeh, Chen Chen, Teck-Yian Lim, Mark Hasegawa-Johnson, and Minh N. Do. Semantic image", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 99, + 342, + 420, + 354 + ], + "spans": [ + { + "bbox": [ + 99, + 342, + 420, + 354 + ], + "score": 1.0, + "content": "inpainting with perceptual and contextual losses. CoRR, abs/1607.07539, 2016.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 88, + 330, + 522, + 354 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 361, + 520, + 395 + ], + "lines": [ + { + "bbox": [ + 90, + 359, + 522, + 375 + ], + "spans": [ + { + "bbox": [ + 90, + 359, + 522, + 375 + ], + "score": 1.0, + "content": "Raymond A Yeh, Chen Chen, Teck Yian Lim, Alexander G Schwing, Mark Hasegawa-Johnson, and Minh N", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 100, + 372, + 522, + 385 + ], + "spans": [ + { + "bbox": [ + 100, + 372, + 522, + 385 + ], + "score": 1.0, + "content": "Do. Semantic image inpainting with deep generative models. In Proceedings of the IEEE Conference on", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 100, + 383, + 362, + 396 + ], + "spans": [ + { + "bbox": [ + 100, + 383, + 362, + 396 + ], + "score": 1.0, + "content": "Computer Vision and Pattern Recognition, pp. 5485–5493, 2017.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 90, + 359, + 522, + 396 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 402, + 521, + 458 + ], + "lines": [ + { + "bbox": [ + 89, + 402, + 522, + 415 + ], + "spans": [ + { + "bbox": [ + 89, + 402, + 522, + 415 + ], + "score": 1.0, + "content": "Jinsung Yoon, James Jordon, and Mihaela van der Schaar. GAIN: Missing data imputation using generative", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 99, + 412, + 521, + 426 + ], + "spans": [ + { + "bbox": [ + 99, + 412, + 521, + 426 + ], + "score": 1.0, + "content": "adversarial nets. In Jennifer Dy and Andreas Krause (eds.), Proceedings of the 35th International Con-", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 99, + 425, + 521, + 437 + ], + "spans": [ + { + "bbox": [ + 99, + 425, + 521, + 437 + ], + "score": 1.0, + "content": "ference on Machine Learning, volume 80 of Proceedings of Machine Learning Research, pp. 5689–5698,", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 99, + 434, + 521, + 448 + ], + "spans": [ + { + "bbox": [ + 99, + 434, + 521, + 448 + ], + "score": 1.0, + "content": "Stockholmsmssan, Stockholm Sweden, 10–15 Jul 2018. PMLR. URL http://proceedings.mlr.", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 98, + 447, + 237, + 459 + ], + "spans": [ + { + "bbox": [ + 98, + 447, + 237, + 459 + ], + "score": 1.0, + "content": "press/v80/yoon18a.html.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 24, + "bbox_fs": [ + 89, + 402, + 522, + 459 + ] + }, + { + "type": "title", + "bbox": [ + 92, + 479, + 145, + 491 + ], + "lines": [ + { + "bbox": [ + 90, + 479, + 146, + 493 + ], + "spans": [ + { + "bbox": [ + 90, + 479, + 146, + 493 + ], + "score": 1.0, + "content": "APPENDIX", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 27 + }, + { + "type": "title", + "bbox": [ + 92, + 506, + 241, + 519 + ], + "lines": [ + { + "bbox": [ + 90, + 505, + 243, + 520 + ], + "spans": [ + { + "bbox": [ + 90, + 505, + 243, + 520 + ], + "score": 1.0, + "content": "A EXPERIMENTAL DETAILS", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 92, + 531, + 277, + 542 + ], + "lines": [ + { + "bbox": [ + 90, + 530, + 278, + 543 + ], + "spans": [ + { + "bbox": [ + 90, + 530, + 278, + 543 + ], + "score": 1.0, + "content": "A.1 NEURAL NETWORK ARCHITECTURES", + "type": "text" + } + ], + "index": 29 + } + ], + "index": 29, + "bbox_fs": [ + 90, + 530, + 278, + 543 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 551, + 521, + 585 + ], + "lines": [ + { + "bbox": [ + 90, + 551, + 522, + 564 + ], + "spans": [ + { + "bbox": [ + 90, + 551, + 522, + 564 + ], + "score": 1.0, + "content": "In all experiments we use optimization method Adam (Kingma & Ba, 2014), skip-connections between prior", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 87, + 561, + 522, + 577 + ], + "spans": [ + { + "bbox": [ + 87, + 561, + 522, + 577 + ], + "score": 1.0, + "content": "network and generative network inspired by (Mao et al., 2016), (Sønderby et al., 2016) and (Ronneberger", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 89, + 573, + 451, + 586 + ], + "spans": [ + { + "bbox": [ + 89, + 573, + 451, + 586 + ], + "score": 1.0, + "content": "et al., 2015), and convolutional neural networks based on ResNet blocks (He et al., 2016).", + "type": "text" + } + ], + "index": 32 + } + ], + "index": 31, + "bbox_fs": [ + 87, + 551, + 522, + 586 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 590, + 521, + 635 + ], + "lines": [ + { + "bbox": [ + 90, + 590, + 521, + 603 + ], + "spans": [ + { + "bbox": [ + 90, + 590, + 521, + 603 + ], + "score": 1.0, + "content": "Without skip-connections all information for decoder goes through the latent variables. In image inpainting", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 89, + 600, + 521, + 614 + ], + "spans": [ + { + "bbox": [ + 89, + 600, + 521, + 614 + ], + "score": 1.0, + "content": "we found skip-connections very useful in both terms of log-likelihood improvement and the image realism,", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 90, + 612, + 521, + 625 + ], + "spans": [ + { + "bbox": [ + 90, + 612, + 521, + 625 + ], + "score": 1.0, + "content": "because latent variables are responsible for the global information only while the local information passes", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 90, + 623, + 513, + 636 + ], + "spans": [ + { + "bbox": [ + 90, + 623, + 513, + 636 + ], + "score": 1.0, + "content": "through skip-connections. Therefore the border between image and inpainting becomes less conspicuous.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 34.5, + "bbox_fs": [ + 89, + 590, + 521, + 636 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 640, + 366, + 651 + ], + "lines": [ + { + "bbox": [ + 89, + 639, + 369, + 653 + ], + "spans": [ + { + "bbox": [ + 89, + 639, + 369, + 653 + ], + "score": 1.0, + "content": "The main idea of neural networks architecture is reflected in figure 5.", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 37, + "bbox_fs": [ + 89, + 639, + 369, + 653 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 99, + 109, + 510, + 279 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 99, + 109, + 510, + 279 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 99, + 109, + 510, + 279 + ], + "spans": [ + { + "bbox": [ + 99, + 109, + 510, + 279 + ], + "score": 0.972, + "type": "image", + "image_path": "5b67d91960752eb18692b5b61ef536a092947e8d387abb82437625c2d41f393a.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 99, + 109, + 510, + 165.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 99, + 165.66666666666666, + 510, + 222.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 99, + 222.33333333333331, + 510, + 279.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 199, + 295, + 412, + 307 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 198, + 293, + 412, + 309 + ], + "spans": [ + { + "bbox": [ + 198, + 293, + 412, + 309 + ], + "score": 1.0, + "content": "Figure 5: Neural network architecture for inpainting.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 91, + 338, + 384, + 350 + ], + "lines": [ + { + "bbox": [ + 90, + 337, + 385, + 352 + ], + "spans": [ + { + "bbox": [ + 90, + 337, + 385, + 352 + ], + "score": 1.0, + "content": "The number of hidden layers, their widths and structure may be different.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4 + }, + { + "type": "text", + "bbox": [ + 91, + 355, + 522, + 389 + ], + "lines": [ + { + "bbox": [ + 90, + 356, + 521, + 367 + ], + "spans": [ + { + "bbox": [ + 90, + 356, + 521, + 367 + ], + "score": 1.0, + "content": "The neural networks we used for image inpainting have He-Uniform initialization of convolutional ResNet", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 90, + 366, + 522, + 379 + ], + "spans": [ + { + "bbox": [ + 90, + 366, + 522, + 379 + ], + "score": 1.0, + "content": "blocks, and the skip-connections are implemented using concatenation, not addition. The proposal network", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 90, + 378, + 388, + 390 + ], + "spans": [ + { + "bbox": [ + 90, + 378, + 388, + 390 + ], + "score": 1.0, + "content": "structure is exactly the same as the prior network except skip-connections.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 90, + 394, + 520, + 417 + ], + "lines": [ + { + "bbox": [ + 90, + 394, + 522, + 407 + ], + "spans": [ + { + "bbox": [ + 90, + 394, + 522, + 407 + ], + "score": 1.0, + "content": "Also one could use much simpler fully-connected networks with one hidden layer as a proposal, prior and", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 89, + 405, + 394, + 417 + ], + "spans": [ + { + "bbox": [ + 89, + 405, + 394, + 417 + ], + "score": 1.0, + "content": "generative networks in VAEAC and still obtain nice inpaintings on MNIST.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5 + }, + { + "type": "title", + "bbox": [ + 92, + 443, + 258, + 454 + ], + "lines": [ + { + "bbox": [ + 90, + 443, + 260, + 456 + ], + "spans": [ + { + "bbox": [ + 90, + 443, + 260, + 456 + ], + "score": 1.0, + "content": "A.2 MISSING FEATURES IMPUTATION", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 90, + 468, + 524, + 502 + ], + "lines": [ + { + "bbox": [ + 90, + 468, + 522, + 481 + ], + "spans": [ + { + "bbox": [ + 90, + 468, + 488, + 481 + ], + "score": 1.0, + "content": "We split the dataset into train and test set with size ratio 3:1. Before training we drop randomly", + "type": "text" + }, + { + "bbox": [ + 489, + 469, + 509, + 479 + ], + "score": 0.86, + "content": "50 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 509, + 468, + 522, + 481 + ], + "score": 1.0, + "content": "of", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 89, + 478, + 522, + 492 + ], + "spans": [ + { + "bbox": [ + 89, + 478, + 522, + 492 + ], + "score": 1.0, + "content": "values both in train and test set. We repeat each experiment 5 times with different train-test splits and", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 90, + 491, + 410, + 503 + ], + "spans": [ + { + "bbox": [ + 90, + 491, + 410, + 503 + ], + "score": 1.0, + "content": "dropped features and then average results and compute their standard deviation.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 90, + 507, + 522, + 541 + ], + "lines": [ + { + "bbox": [ + 89, + 507, + 522, + 520 + ], + "spans": [ + { + "bbox": [ + 89, + 507, + 522, + 520 + ], + "score": 1.0, + "content": "As we show in appendix B.2, the better results can be achieved when the model learns the concatenation of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 90, + 519, + 520, + 531 + ], + "spans": [ + { + "bbox": [ + 90, + 519, + 155, + 531 + ], + "score": 1.0, + "content": "objects features", + "type": "text" + }, + { + "bbox": [ + 155, + 520, + 162, + 528 + ], + "score": 0.72, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 519, + 209, + 531 + ], + "score": 1.0, + "content": "and targets", + "type": "text" + }, + { + "bbox": [ + 210, + 520, + 216, + 530 + ], + "score": 0.72, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 519, + 268, + 531 + ], + "score": 1.0, + "content": ". So we treat", + "type": "text" + }, + { + "bbox": [ + 269, + 520, + 276, + 530 + ], + "score": 0.79, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 519, + 520, + 531 + ], + "score": 1.0, + "content": "as an additional feature that is always unobserved during the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 90, + 528, + 141, + 543 + ], + "spans": [ + { + "bbox": [ + 90, + 528, + 141, + 543 + ], + "score": 1.0, + "content": "testing time.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 91, + 546, + 520, + 591 + ], + "lines": [ + { + "bbox": [ + 88, + 545, + 522, + 560 + ], + "spans": [ + { + "bbox": [ + 88, + 545, + 250, + 560 + ], + "score": 1.0, + "content": "To train our model we use distribution", + "type": "text" + }, + { + "bbox": [ + 250, + 546, + 280, + 558 + ], + "score": 0.93, + "content": "p ( b _ { i } | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 545, + 320, + 560 + ], + "score": 1.0, + "content": "in which", + "type": "text" + }, + { + "bbox": [ + 321, + 546, + 398, + 558 + ], + "score": 0.91, + "content": "p ( b _ { i } | x _ { i } = \\omega ) = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 545, + 417, + 560 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 417, + 546, + 476, + 559 + ], + "score": 0.93, + "content": "p ( b _ { i } | x ) = 0 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 476, + 545, + 522, + 560 + ], + "score": 1.0, + "content": "otherwise.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 90, + 557, + 522, + 570 + ], + "spans": [ + { + "bbox": [ + 90, + 557, + 345, + 570 + ], + "score": 1.0, + "content": "Also for VAEAC trainig we normalize real-valued features, fix", + "type": "text" + }, + { + "bbox": [ + 345, + 559, + 376, + 568 + ], + "score": 0.89, + "content": "\\sigma _ { \\theta } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 557, + 522, + 570 + ], + "score": 1.0, + "content": "in the generative model of VAEAC", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 90, + 568, + 522, + 581 + ], + "spans": [ + { + "bbox": [ + 90, + 568, + 235, + 581 + ], + "score": 1.0, + "content": "in order to optimize RMSE, and use", + "type": "text" + }, + { + "bbox": [ + 235, + 569, + 255, + 579 + ], + "score": 0.87, + "content": "2 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 568, + 522, + 581 + ], + "score": 1.0, + "content": "of training data as validation set to select the best model among all", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 89, + 577, + 168, + 594 + ], + "spans": [ + { + "bbox": [ + 89, + 577, + 168, + 594 + ], + "score": 1.0, + "content": "epochs of training.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5 + }, + { + "type": "text", + "bbox": [ + 91, + 596, + 521, + 652 + ], + "lines": [ + { + "bbox": [ + 90, + 596, + 522, + 608 + ], + "spans": [ + { + "bbox": [ + 90, + 596, + 353, + 608 + ], + "score": 1.0, + "content": "For the test set, the classifier or regressor is applied to each of the", + "type": "text" + }, + { + "bbox": [ + 354, + 598, + 361, + 606 + ], + "score": 0.74, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 596, + 522, + 608 + ], + "score": 1.0, + "content": "imputed objects and the predictions are", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 89, + 606, + 523, + 621 + ], + "spans": [ + { + "bbox": [ + 89, + 606, + 523, + 621 + ], + "score": 1.0, + "content": "combined. For regression problems we report R2-score of combined predictions, so we use averaging as a", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 90, + 618, + 521, + 631 + ], + "spans": [ + { + "bbox": [ + 90, + 618, + 521, + 631 + ], + "score": 1.0, + "content": "combination method. For classification problem we report accuracy, and therefore choose the mode. We", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 90, + 630, + 521, + 641 + ], + "spans": [ + { + "bbox": [ + 90, + 630, + 297, + 641 + ], + "score": 1.0, + "content": "consider the workflow where the imputed values of", + "type": "text" + }, + { + "bbox": [ + 298, + 631, + 304, + 641 + ], + "score": 0.81, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 630, + 521, + 641 + ], + "score": 1.0, + "content": "are not fed to the classifier or regressor to make a fair", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 89, + 640, + 257, + 654 + ], + "spans": [ + { + "bbox": [ + 89, + 640, + 257, + 654 + ], + "score": 1.0, + "content": "comparison of feature imputation quality.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23 + } + ], + "page_idx": 12, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 40, + 277, + 50 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 284, + 671, + 295, + 680 + ], + "lines": [ + { + "bbox": [ + 283, + 670, + 297, + 684 + ], + "spans": [ + { + "bbox": [ + 283, + 670, + 297, + 684 + ], + "score": 1.0, + "content": "13", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 99, + 109, + 510, + 279 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 99, + 109, + 510, + 279 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 99, + 109, + 510, + 279 + ], + "spans": [ + { + "bbox": [ + 99, + 109, + 510, + 279 + ], + "score": 0.972, + "type": "image", + "image_path": "5b67d91960752eb18692b5b61ef536a092947e8d387abb82437625c2d41f393a.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 99, + 109, + 510, + 165.66666666666666 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 99, + 165.66666666666666, + 510, + 222.33333333333331 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 99, + 222.33333333333331, + 510, + 279.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 199, + 295, + 412, + 307 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 198, + 293, + 412, + 309 + ], + "spans": [ + { + "bbox": [ + 198, + 293, + 412, + 309 + ], + "score": 1.0, + "content": "Figure 5: Neural network architecture for inpainting.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + }, + { + "type": "text", + "bbox": [ + 91, + 338, + 384, + 350 + ], + "lines": [ + { + "bbox": [ + 90, + 337, + 385, + 352 + ], + "spans": [ + { + "bbox": [ + 90, + 337, + 385, + 352 + ], + "score": 1.0, + "content": "The number of hidden layers, their widths and structure may be different.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 4, + "bbox_fs": [ + 90, + 337, + 385, + 352 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 355, + 522, + 389 + ], + "lines": [ + { + "bbox": [ + 90, + 356, + 521, + 367 + ], + "spans": [ + { + "bbox": [ + 90, + 356, + 521, + 367 + ], + "score": 1.0, + "content": "The neural networks we used for image inpainting have He-Uniform initialization of convolutional ResNet", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 90, + 366, + 522, + 379 + ], + "spans": [ + { + "bbox": [ + 90, + 366, + 522, + 379 + ], + "score": 1.0, + "content": "blocks, and the skip-connections are implemented using concatenation, not addition. The proposal network", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 90, + 378, + 388, + 390 + ], + "spans": [ + { + "bbox": [ + 90, + 378, + 388, + 390 + ], + "score": 1.0, + "content": "structure is exactly the same as the prior network except skip-connections.", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 6, + "bbox_fs": [ + 90, + 356, + 522, + 390 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 394, + 520, + 417 + ], + "lines": [ + { + "bbox": [ + 90, + 394, + 522, + 407 + ], + "spans": [ + { + "bbox": [ + 90, + 394, + 522, + 407 + ], + "score": 1.0, + "content": "Also one could use much simpler fully-connected networks with one hidden layer as a proposal, prior and", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 89, + 405, + 394, + 417 + ], + "spans": [ + { + "bbox": [ + 89, + 405, + 394, + 417 + ], + "score": 1.0, + "content": "generative networks in VAEAC and still obtain nice inpaintings on MNIST.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8.5, + "bbox_fs": [ + 89, + 394, + 522, + 417 + ] + }, + { + "type": "title", + "bbox": [ + 92, + 443, + 258, + 454 + ], + "lines": [ + { + "bbox": [ + 90, + 443, + 260, + 456 + ], + "spans": [ + { + "bbox": [ + 90, + 443, + 260, + 456 + ], + "score": 1.0, + "content": "A.2 MISSING FEATURES IMPUTATION", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 90, + 468, + 524, + 502 + ], + "lines": [ + { + "bbox": [ + 90, + 468, + 522, + 481 + ], + "spans": [ + { + "bbox": [ + 90, + 468, + 488, + 481 + ], + "score": 1.0, + "content": "We split the dataset into train and test set with size ratio 3:1. Before training we drop randomly", + "type": "text" + }, + { + "bbox": [ + 489, + 469, + 509, + 479 + ], + "score": 0.86, + "content": "50 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 509, + 468, + 522, + 481 + ], + "score": 1.0, + "content": "of", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 89, + 478, + 522, + 492 + ], + "spans": [ + { + "bbox": [ + 89, + 478, + 522, + 492 + ], + "score": 1.0, + "content": "values both in train and test set. We repeat each experiment 5 times with different train-test splits and", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 90, + 491, + 410, + 503 + ], + "spans": [ + { + "bbox": [ + 90, + 491, + 410, + 503 + ], + "score": 1.0, + "content": "dropped features and then average results and compute their standard deviation.", + "type": "text" + } + ], + "index": 13 + } + ], + "index": 12, + "bbox_fs": [ + 89, + 468, + 522, + 503 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 507, + 522, + 541 + ], + "lines": [ + { + "bbox": [ + 89, + 507, + 522, + 520 + ], + "spans": [ + { + "bbox": [ + 89, + 507, + 522, + 520 + ], + "score": 1.0, + "content": "As we show in appendix B.2, the better results can be achieved when the model learns the concatenation of", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 90, + 519, + 520, + 531 + ], + "spans": [ + { + "bbox": [ + 90, + 519, + 155, + 531 + ], + "score": 1.0, + "content": "objects features", + "type": "text" + }, + { + "bbox": [ + 155, + 520, + 162, + 528 + ], + "score": 0.72, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 519, + 209, + 531 + ], + "score": 1.0, + "content": "and targets", + "type": "text" + }, + { + "bbox": [ + 210, + 520, + 216, + 530 + ], + "score": 0.72, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 217, + 519, + 268, + 531 + ], + "score": 1.0, + "content": ". So we treat", + "type": "text" + }, + { + "bbox": [ + 269, + 520, + 276, + 530 + ], + "score": 0.79, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 276, + 519, + 520, + 531 + ], + "score": 1.0, + "content": "as an additional feature that is always unobserved during the", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 90, + 528, + 141, + 543 + ], + "spans": [ + { + "bbox": [ + 90, + 528, + 141, + 543 + ], + "score": 1.0, + "content": "testing time.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15, + "bbox_fs": [ + 89, + 507, + 522, + 543 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 546, + 520, + 591 + ], + "lines": [ + { + "bbox": [ + 88, + 545, + 522, + 560 + ], + "spans": [ + { + "bbox": [ + 88, + 545, + 250, + 560 + ], + "score": 1.0, + "content": "To train our model we use distribution", + "type": "text" + }, + { + "bbox": [ + 250, + 546, + 280, + 558 + ], + "score": 0.93, + "content": "p ( b _ { i } | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 545, + 320, + 560 + ], + "score": 1.0, + "content": "in which", + "type": "text" + }, + { + "bbox": [ + 321, + 546, + 398, + 558 + ], + "score": 0.91, + "content": "p ( b _ { i } | x _ { i } = \\omega ) = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 545, + 417, + 560 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 417, + 546, + 476, + 559 + ], + "score": 0.93, + "content": "p ( b _ { i } | x ) = 0 . 2", + "type": "inline_equation" + }, + { + "bbox": [ + 476, + 545, + 522, + 560 + ], + "score": 1.0, + "content": "otherwise.", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 90, + 557, + 522, + 570 + ], + "spans": [ + { + "bbox": [ + 90, + 557, + 345, + 570 + ], + "score": 1.0, + "content": "Also for VAEAC trainig we normalize real-valued features, fix", + "type": "text" + }, + { + "bbox": [ + 345, + 559, + 376, + 568 + ], + "score": 0.89, + "content": "\\sigma _ { \\theta } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 376, + 557, + 522, + 570 + ], + "score": 1.0, + "content": "in the generative model of VAEAC", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 90, + 568, + 522, + 581 + ], + "spans": [ + { + "bbox": [ + 90, + 568, + 235, + 581 + ], + "score": 1.0, + "content": "in order to optimize RMSE, and use", + "type": "text" + }, + { + "bbox": [ + 235, + 569, + 255, + 579 + ], + "score": 0.87, + "content": "2 5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 255, + 568, + 522, + 581 + ], + "score": 1.0, + "content": "of training data as validation set to select the best model among all", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 89, + 577, + 168, + 594 + ], + "spans": [ + { + "bbox": [ + 89, + 577, + 168, + 594 + ], + "score": 1.0, + "content": "epochs of training.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5, + "bbox_fs": [ + 88, + 545, + 522, + 594 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 596, + 521, + 652 + ], + "lines": [ + { + "bbox": [ + 90, + 596, + 522, + 608 + ], + "spans": [ + { + "bbox": [ + 90, + 596, + 353, + 608 + ], + "score": 1.0, + "content": "For the test set, the classifier or regressor is applied to each of the", + "type": "text" + }, + { + "bbox": [ + 354, + 598, + 361, + 606 + ], + "score": 0.74, + "content": "n", + "type": "inline_equation" + }, + { + "bbox": [ + 361, + 596, + 522, + 608 + ], + "score": 1.0, + "content": "imputed objects and the predictions are", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 89, + 606, + 523, + 621 + ], + "spans": [ + { + "bbox": [ + 89, + 606, + 523, + 621 + ], + "score": 1.0, + "content": "combined. For regression problems we report R2-score of combined predictions, so we use averaging as a", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 90, + 618, + 521, + 631 + ], + "spans": [ + { + "bbox": [ + 90, + 618, + 521, + 631 + ], + "score": 1.0, + "content": "combination method. For classification problem we report accuracy, and therefore choose the mode. We", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 90, + 630, + 521, + 641 + ], + "spans": [ + { + "bbox": [ + 90, + 630, + 297, + 641 + ], + "score": 1.0, + "content": "consider the workflow where the imputed values of", + "type": "text" + }, + { + "bbox": [ + 298, + 631, + 304, + 641 + ], + "score": 0.81, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 630, + 521, + 641 + ], + "score": 1.0, + "content": "are not fed to the classifier or regressor to make a fair", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 89, + 640, + 257, + 654 + ], + "spans": [ + { + "bbox": [ + 89, + 640, + 257, + 654 + ], + "score": 1.0, + "content": "comparison of feature imputation quality.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 23, + "bbox_fs": [ + 89, + 596, + 523, + 654 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 191, + 113, + 420, + 194 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 135, + 93, + 475, + 105 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 134, + 92, + 477, + 107 + ], + "spans": [ + { + "bbox": [ + 134, + 92, + 477, + 107 + ], + "score": 1.0, + "content": "Table 6: Generative Face Completion (Li et al., 2017) masks. Image size is 128x128.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 191, + 113, + 420, + 194 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 191, + 113, + 420, + 194 + ], + "spans": [ + { + "bbox": [ + 191, + 113, + 420, + 194 + ], + "score": 0.97, + "html": "
MaskMeaningX1x2y1y2
01Left half of the face337052115
02Right half of the face577095115
03Two eyes29985273
04Left eye29665273
05Right eye61995273
06Lower half of the face408786123
", + "type": "table", + "image_path": "04d4687dcf9699d512eac4a44124a37465506a687d19b17f4d1f42a5bbb655e0.jpg" + } + ] + } + ], + "index": 3.5, + "virtual_lines": [ + { + "bbox": [ + 191, + 113, + 420, + 126.5 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 191, + 126.5, + 420, + 140.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 191, + 140.0, + 420, + 153.5 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 191, + 153.5, + 420, + 167.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 191, + 167.0, + 420, + 180.5 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 191, + 180.5, + 420, + 194.0 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 1.75 + }, + { + "type": "text", + "bbox": [ + 91, + 214, + 520, + 248 + ], + "lines": [ + { + "bbox": [ + 90, + 213, + 521, + 227 + ], + "spans": [ + { + "bbox": [ + 90, + 213, + 521, + 227 + ], + "score": 1.0, + "content": "NRMSE or PFC for dataset is computed as an average of NRMSE or PFC of all features of this dataset.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 90, + 224, + 522, + 238 + ], + "spans": [ + { + "bbox": [ + 90, + 224, + 522, + 238 + ], + "score": 1.0, + "content": "NRMSE of a feature is just RMSE of imputations divided by the standard deviation of this feature. PFC of", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 89, + 236, + 331, + 249 + ], + "spans": [ + { + "bbox": [ + 89, + 236, + 331, + 249 + ], + "score": 1.0, + "content": "a feature is a proportion of imputations which are incorrect.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8 + }, + { + "type": "title", + "bbox": [ + 92, + 261, + 302, + 273 + ], + "lines": [ + { + "bbox": [ + 90, + 261, + 303, + 274 + ], + "spans": [ + { + "bbox": [ + 90, + 261, + 303, + 274 + ], + "score": 1.0, + "content": "A.3 IMAGE INPAINTING DATASETS AND MASKS", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 90, + 282, + 521, + 338 + ], + "lines": [ + { + "bbox": [ + 88, + 281, + 521, + 295 + ], + "spans": [ + { + "bbox": [ + 88, + 281, + 491, + 295 + ], + "score": 1.0, + "content": "MNIST is a dataset of 60000 train and 10000 test grayscale images of digits from 0 to 9 of size", + "type": "text" + }, + { + "bbox": [ + 491, + 283, + 518, + 293 + ], + "score": 0.77, + "content": "2 8 \\mathbf { x } 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 518, + 281, + 521, + 295 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 90, + 293, + 521, + 305 + ], + "spans": [ + { + "bbox": [ + 90, + 293, + 521, + 305 + ], + "score": 1.0, + "content": "We binarize all images in the dataset. For MNIST we consider Bernoulli log-likelihood as the reconstruction", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 87, + 302, + 524, + 321 + ], + "spans": [ + { + "bbox": [ + 87, + 302, + 110, + 321 + ], + "score": 1.0, + "content": "loss:", + "type": "text" + }, + { + "bbox": [ + 111, + 304, + 379, + 317 + ], + "score": 0.78, + "content": "\\begin{array} { r } { \\log p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b ) = \\sum _ { i : b _ { i } = 1 } \\log \\mathrm { B e r n o u l l i } ( x _ { i } | p _ { \\theta , i } ( z , x _ { 1 - b } , b ) ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 302, + 408, + 321 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 408, + 304, + 470, + 317 + ], + "score": 0.93, + "content": "p _ { \\theta , i } ( z , x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 302, + 524, + 321 + ], + "score": 1.0, + "content": "is an output", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 89, + 315, + 522, + 328 + ], + "spans": [ + { + "bbox": [ + 89, + 315, + 522, + 328 + ], + "score": 1.0, + "content": "of the generative neural network. We use 16 latent variables. In the mask for this dataset the observed pixels", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 88, + 325, + 409, + 339 + ], + "spans": [ + { + "bbox": [ + 88, + 325, + 409, + 339 + ], + "score": 1.0, + "content": "form a three pixels wide horizontal line which position is distributed uniformly.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 90, + 350, + 520, + 406 + ], + "lines": [ + { + "bbox": [ + 90, + 351, + 521, + 362 + ], + "spans": [ + { + "bbox": [ + 90, + 351, + 521, + 362 + ], + "score": 1.0, + "content": "Omniglot is a dataset of 19280 train and 13180 test black-and-white images of different alphabets symbols", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 89, + 361, + 522, + 373 + ], + "spans": [ + { + "bbox": [ + 89, + 361, + 119, + 373 + ], + "score": 1.0, + "content": "of size", + "type": "text" + }, + { + "bbox": [ + 120, + 362, + 155, + 372 + ], + "score": 0.27, + "content": "1 0 5 \\mathrm { x } 1 0 5", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 361, + 522, + 373 + ], + "score": 1.0, + "content": ". As in previous section, the brightness of each pixel is treated as a Bernoulli probability of", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 89, + 373, + 520, + 383 + ], + "spans": [ + { + "bbox": [ + 89, + 373, + 520, + 383 + ], + "score": 1.0, + "content": "it to be 1. The mask we use is a random rectangular which is described below. We use 64 latent variables.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 90, + 383, + 521, + 395 + ], + "spans": [ + { + "bbox": [ + 90, + 383, + 521, + 395 + ], + "score": 1.0, + "content": "We train model for 50 epochs and choose best model according to IWAE log-likelihood estimation on the", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 90, + 395, + 216, + 407 + ], + "spans": [ + { + "bbox": [ + 90, + 395, + 216, + 407 + ], + "score": 1.0, + "content": "validation set after each epoch.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 91, + 418, + 520, + 463 + ], + "lines": [ + { + "bbox": [ + 89, + 418, + 522, + 431 + ], + "spans": [ + { + "bbox": [ + 89, + 418, + 522, + 431 + ], + "score": 1.0, + "content": "CelebA is a dataset of 162770 train, 19867 validation and 19962 test color images of faces of celebrities", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 89, + 429, + 522, + 443 + ], + "spans": [ + { + "bbox": [ + 89, + 429, + 119, + 443 + ], + "score": 1.0, + "content": "of size", + "type": "text" + }, + { + "bbox": [ + 119, + 430, + 155, + 441 + ], + "score": 0.6, + "content": "1 7 8 \\mathrm { x } 2 1 8", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 429, + 522, + 443 + ], + "score": 1.0, + "content": ". Before learning we normalize the channels in dataset. We use logarithm of fully-factorized", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 90, + 441, + 522, + 453 + ], + "spans": [ + { + "bbox": [ + 90, + 441, + 522, + 453 + ], + "score": 1.0, + "content": "Gaussian distribution as reconstruction loss. The mask we use is a random rectangular which is describe", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 89, + 451, + 229, + 464 + ], + "spans": [ + { + "bbox": [ + 89, + 451, + 229, + 464 + ], + "score": 1.0, + "content": "below. We use 32 latent variables.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22.5 + }, + { + "type": "text", + "bbox": [ + 91, + 475, + 521, + 509 + ], + "lines": [ + { + "bbox": [ + 90, + 475, + 521, + 489 + ], + "spans": [ + { + "bbox": [ + 90, + 475, + 521, + 489 + ], + "score": 1.0, + "content": "Rectangular mask is the common shape of unobserved region in image inpainting. We use such mask for", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 91, + 486, + 521, + 500 + ], + "spans": [ + { + "bbox": [ + 91, + 486, + 521, + 500 + ], + "score": 1.0, + "content": "Omniglot and Celeba. We sample the corner points of rectangles uniprobably on the image, but reject those", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 89, + 497, + 337, + 511 + ], + "spans": [ + { + "bbox": [ + 89, + 497, + 337, + 511 + ], + "score": 1.0, + "content": "rectangles which area is less than a quarter of the image area.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26 + }, + { + "type": "text", + "bbox": [ + 91, + 522, + 520, + 555 + ], + "lines": [ + { + "bbox": [ + 89, + 520, + 522, + 534 + ], + "spans": [ + { + "bbox": [ + 89, + 520, + 522, + 534 + ], + "score": 1.0, + "content": "In Li et al. (2017) six different masks O1–O6 are used on the testing stage. We reconstruct the positions", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 89, + 532, + 522, + 546 + ], + "spans": [ + { + "bbox": [ + 89, + 532, + 522, + 546 + ], + "score": 1.0, + "content": "of masks from the illustrations in the paper and give their coordinates in table 6. The visualizations of the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 90, + 544, + 221, + 556 + ], + "spans": [ + { + "bbox": [ + 90, + 544, + 221, + 556 + ], + "score": 1.0, + "content": "masks are available in figure 10.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29 + }, + { + "type": "text", + "bbox": [ + 91, + 560, + 520, + 605 + ], + "lines": [ + { + "bbox": [ + 89, + 560, + 522, + 574 + ], + "spans": [ + { + "bbox": [ + 89, + 560, + 522, + 574 + ], + "score": 1.0, + "content": "At the training stage we used a rectangle mask with uniprobable random corners. We reject masks with", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 90, + 572, + 522, + 585 + ], + "spans": [ + { + "bbox": [ + 90, + 572, + 522, + 585 + ], + "score": 1.0, + "content": "width or height less than 16pt. We use 64 latent variables and take the best model over 50 epochs based on", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 89, + 582, + 522, + 596 + ], + "spans": [ + { + "bbox": [ + 89, + 582, + 522, + 596 + ], + "score": 1.0, + "content": "the validation IWAE log-likelihood estimation. We can obtain slightly higher PSNR values than reported in", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 88, + 592, + 307, + 608 + ], + "spans": [ + { + "bbox": [ + 88, + 592, + 307, + 608 + ], + "score": 1.0, + "content": "table 4 if use only masks O1–O6 at the training stage.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32.5 + }, + { + "type": "text", + "bbox": [ + 91, + 618, + 521, + 652 + ], + "lines": [ + { + "bbox": [ + 89, + 617, + 521, + 631 + ], + "spans": [ + { + "bbox": [ + 89, + 617, + 440, + 631 + ], + "score": 1.0, + "content": "In Yeh et al. (2017) four types of masks are used. Center mask is just an unobserved", + "type": "text" + }, + { + "bbox": [ + 440, + 618, + 466, + 629 + ], + "score": 0.44, + "content": "3 2 \\mathrm { x } 3 2 ", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 617, + 521, + 631 + ], + "score": 1.0, + "content": "square in the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 89, + 629, + 520, + 642 + ], + "spans": [ + { + "bbox": [ + 89, + 629, + 520, + 642 + ], + "score": 1.0, + "content": "center of 64x64 image. Half mask mean that one of upper, lower, left or right half of the image is unobserved.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 90, + 640, + 520, + 652 + ], + "spans": [ + { + "bbox": [ + 90, + 640, + 520, + 652 + ], + "score": 1.0, + "content": "All these types of a half are equiprobable. Random mask means that we use pixelwise-independent Bernoulli", + "type": "text" + } + ], + "index": 37 + } + ], + "index": 36 + } + ], + "page_idx": 13, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 40, + 277, + 51 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 284, + 671, + 295, + 680 + ], + "lines": [ + { + "bbox": [ + 283, + 670, + 297, + 684 + ], + "spans": [ + { + "bbox": [ + 283, + 670, + 297, + 684 + ], + "score": 1.0, + "content": "14", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 191, + 113, + 420, + 194 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 135, + 93, + 475, + 105 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 134, + 92, + 477, + 107 + ], + "spans": [ + { + "bbox": [ + 134, + 92, + 477, + 107 + ], + "score": 1.0, + "content": "Table 6: Generative Face Completion (Li et al., 2017) masks. Image size is 128x128.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 191, + 113, + 420, + 194 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 191, + 113, + 420, + 194 + ], + "spans": [ + { + "bbox": [ + 191, + 113, + 420, + 194 + ], + "score": 0.97, + "html": "
MaskMeaningX1x2y1y2
01Left half of the face337052115
02Right half of the face577095115
03Two eyes29985273
04Left eye29665273
05Right eye61995273
06Lower half of the face408786123
", + "type": "table", + "image_path": "04d4687dcf9699d512eac4a44124a37465506a687d19b17f4d1f42a5bbb655e0.jpg" + } + ] + } + ], + "index": 3.5, + "virtual_lines": [ + { + "bbox": [ + 191, + 113, + 420, + 126.5 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 191, + 126.5, + 420, + 140.0 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 191, + 140.0, + 420, + 153.5 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 191, + 153.5, + 420, + 167.0 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 191, + 167.0, + 420, + 180.5 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 191, + 180.5, + 420, + 194.0 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 1.75 + }, + { + "type": "text", + "bbox": [ + 91, + 214, + 520, + 248 + ], + "lines": [ + { + "bbox": [ + 90, + 213, + 521, + 227 + ], + "spans": [ + { + "bbox": [ + 90, + 213, + 521, + 227 + ], + "score": 1.0, + "content": "NRMSE or PFC for dataset is computed as an average of NRMSE or PFC of all features of this dataset.", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 90, + 224, + 522, + 238 + ], + "spans": [ + { + "bbox": [ + 90, + 224, + 522, + 238 + ], + "score": 1.0, + "content": "NRMSE of a feature is just RMSE of imputations divided by the standard deviation of this feature. PFC of", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 89, + 236, + 331, + 249 + ], + "spans": [ + { + "bbox": [ + 89, + 236, + 331, + 249 + ], + "score": 1.0, + "content": "a feature is a proportion of imputations which are incorrect.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 8, + "bbox_fs": [ + 89, + 213, + 522, + 249 + ] + }, + { + "type": "title", + "bbox": [ + 92, + 261, + 302, + 273 + ], + "lines": [ + { + "bbox": [ + 90, + 261, + 303, + 274 + ], + "spans": [ + { + "bbox": [ + 90, + 261, + 303, + 274 + ], + "score": 1.0, + "content": "A.3 IMAGE INPAINTING DATASETS AND MASKS", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 90, + 282, + 521, + 338 + ], + "lines": [ + { + "bbox": [ + 88, + 281, + 521, + 295 + ], + "spans": [ + { + "bbox": [ + 88, + 281, + 491, + 295 + ], + "score": 1.0, + "content": "MNIST is a dataset of 60000 train and 10000 test grayscale images of digits from 0 to 9 of size", + "type": "text" + }, + { + "bbox": [ + 491, + 283, + 518, + 293 + ], + "score": 0.77, + "content": "2 8 \\mathbf { x } 2 8", + "type": "inline_equation" + }, + { + "bbox": [ + 518, + 281, + 521, + 295 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 90, + 293, + 521, + 305 + ], + "spans": [ + { + "bbox": [ + 90, + 293, + 521, + 305 + ], + "score": 1.0, + "content": "We binarize all images in the dataset. For MNIST we consider Bernoulli log-likelihood as the reconstruction", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 87, + 302, + 524, + 321 + ], + "spans": [ + { + "bbox": [ + 87, + 302, + 110, + 321 + ], + "score": 1.0, + "content": "loss:", + "type": "text" + }, + { + "bbox": [ + 111, + 304, + 379, + 317 + ], + "score": 0.78, + "content": "\\begin{array} { r } { \\log p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b ) = \\sum _ { i : b _ { i } = 1 } \\log \\mathrm { B e r n o u l l i } ( x _ { i } | p _ { \\theta , i } ( z , x _ { 1 - b } , b ) ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 380, + 302, + 408, + 321 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 408, + 304, + 470, + 317 + ], + "score": 0.93, + "content": "p _ { \\theta , i } ( z , x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 302, + 524, + 321 + ], + "score": 1.0, + "content": "is an output", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 89, + 315, + 522, + 328 + ], + "spans": [ + { + "bbox": [ + 89, + 315, + 522, + 328 + ], + "score": 1.0, + "content": "of the generative neural network. We use 16 latent variables. In the mask for this dataset the observed pixels", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 88, + 325, + 409, + 339 + ], + "spans": [ + { + "bbox": [ + 88, + 325, + 409, + 339 + ], + "score": 1.0, + "content": "form a three pixels wide horizontal line which position is distributed uniformly.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13, + "bbox_fs": [ + 87, + 281, + 524, + 339 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 350, + 520, + 406 + ], + "lines": [ + { + "bbox": [ + 90, + 351, + 521, + 362 + ], + "spans": [ + { + "bbox": [ + 90, + 351, + 521, + 362 + ], + "score": 1.0, + "content": "Omniglot is a dataset of 19280 train and 13180 test black-and-white images of different alphabets symbols", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 89, + 361, + 522, + 373 + ], + "spans": [ + { + "bbox": [ + 89, + 361, + 119, + 373 + ], + "score": 1.0, + "content": "of size", + "type": "text" + }, + { + "bbox": [ + 120, + 362, + 155, + 372 + ], + "score": 0.27, + "content": "1 0 5 \\mathrm { x } 1 0 5", + "type": "inline_equation" + }, + { + "bbox": [ + 156, + 361, + 522, + 373 + ], + "score": 1.0, + "content": ". As in previous section, the brightness of each pixel is treated as a Bernoulli probability of", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 89, + 373, + 520, + 383 + ], + "spans": [ + { + "bbox": [ + 89, + 373, + 520, + 383 + ], + "score": 1.0, + "content": "it to be 1. The mask we use is a random rectangular which is described below. We use 64 latent variables.", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 90, + 383, + 521, + 395 + ], + "spans": [ + { + "bbox": [ + 90, + 383, + 521, + 395 + ], + "score": 1.0, + "content": "We train model for 50 epochs and choose best model according to IWAE log-likelihood estimation on the", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 90, + 395, + 216, + 407 + ], + "spans": [ + { + "bbox": [ + 90, + 395, + 216, + 407 + ], + "score": 1.0, + "content": "validation set after each epoch.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18, + "bbox_fs": [ + 89, + 351, + 522, + 407 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 418, + 520, + 463 + ], + "lines": [ + { + "bbox": [ + 89, + 418, + 522, + 431 + ], + "spans": [ + { + "bbox": [ + 89, + 418, + 522, + 431 + ], + "score": 1.0, + "content": "CelebA is a dataset of 162770 train, 19867 validation and 19962 test color images of faces of celebrities", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 89, + 429, + 522, + 443 + ], + "spans": [ + { + "bbox": [ + 89, + 429, + 119, + 443 + ], + "score": 1.0, + "content": "of size", + "type": "text" + }, + { + "bbox": [ + 119, + 430, + 155, + 441 + ], + "score": 0.6, + "content": "1 7 8 \\mathrm { x } 2 1 8", + "type": "inline_equation" + }, + { + "bbox": [ + 155, + 429, + 522, + 443 + ], + "score": 1.0, + "content": ". Before learning we normalize the channels in dataset. We use logarithm of fully-factorized", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 90, + 441, + 522, + 453 + ], + "spans": [ + { + "bbox": [ + 90, + 441, + 522, + 453 + ], + "score": 1.0, + "content": "Gaussian distribution as reconstruction loss. The mask we use is a random rectangular which is describe", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 89, + 451, + 229, + 464 + ], + "spans": [ + { + "bbox": [ + 89, + 451, + 229, + 464 + ], + "score": 1.0, + "content": "below. We use 32 latent variables.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 22.5, + "bbox_fs": [ + 89, + 418, + 522, + 464 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 475, + 521, + 509 + ], + "lines": [ + { + "bbox": [ + 90, + 475, + 521, + 489 + ], + "spans": [ + { + "bbox": [ + 90, + 475, + 521, + 489 + ], + "score": 1.0, + "content": "Rectangular mask is the common shape of unobserved region in image inpainting. We use such mask for", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 91, + 486, + 521, + 500 + ], + "spans": [ + { + "bbox": [ + 91, + 486, + 521, + 500 + ], + "score": 1.0, + "content": "Omniglot and Celeba. We sample the corner points of rectangles uniprobably on the image, but reject those", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 89, + 497, + 337, + 511 + ], + "spans": [ + { + "bbox": [ + 89, + 497, + 337, + 511 + ], + "score": 1.0, + "content": "rectangles which area is less than a quarter of the image area.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 26, + "bbox_fs": [ + 89, + 475, + 521, + 511 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 522, + 520, + 555 + ], + "lines": [ + { + "bbox": [ + 89, + 520, + 522, + 534 + ], + "spans": [ + { + "bbox": [ + 89, + 520, + 522, + 534 + ], + "score": 1.0, + "content": "In Li et al. (2017) six different masks O1–O6 are used on the testing stage. We reconstruct the positions", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 89, + 532, + 522, + 546 + ], + "spans": [ + { + "bbox": [ + 89, + 532, + 522, + 546 + ], + "score": 1.0, + "content": "of masks from the illustrations in the paper and give their coordinates in table 6. The visualizations of the", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 90, + 544, + 221, + 556 + ], + "spans": [ + { + "bbox": [ + 90, + 544, + 221, + 556 + ], + "score": 1.0, + "content": "masks are available in figure 10.", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 29, + "bbox_fs": [ + 89, + 520, + 522, + 556 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 560, + 520, + 605 + ], + "lines": [ + { + "bbox": [ + 89, + 560, + 522, + 574 + ], + "spans": [ + { + "bbox": [ + 89, + 560, + 522, + 574 + ], + "score": 1.0, + "content": "At the training stage we used a rectangle mask with uniprobable random corners. We reject masks with", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 90, + 572, + 522, + 585 + ], + "spans": [ + { + "bbox": [ + 90, + 572, + 522, + 585 + ], + "score": 1.0, + "content": "width or height less than 16pt. We use 64 latent variables and take the best model over 50 epochs based on", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 89, + 582, + 522, + 596 + ], + "spans": [ + { + "bbox": [ + 89, + 582, + 522, + 596 + ], + "score": 1.0, + "content": "the validation IWAE log-likelihood estimation. We can obtain slightly higher PSNR values than reported in", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 88, + 592, + 307, + 608 + ], + "spans": [ + { + "bbox": [ + 88, + 592, + 307, + 608 + ], + "score": 1.0, + "content": "table 4 if use only masks O1–O6 at the training stage.", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 32.5, + "bbox_fs": [ + 88, + 560, + 522, + 608 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 618, + 521, + 652 + ], + "lines": [ + { + "bbox": [ + 89, + 617, + 521, + 631 + ], + "spans": [ + { + "bbox": [ + 89, + 617, + 440, + 631 + ], + "score": 1.0, + "content": "In Yeh et al. (2017) four types of masks are used. Center mask is just an unobserved", + "type": "text" + }, + { + "bbox": [ + 440, + 618, + 466, + 629 + ], + "score": 0.44, + "content": "3 2 \\mathrm { x } 3 2 ", + "type": "inline_equation" + }, + { + "bbox": [ + 467, + 617, + 521, + 631 + ], + "score": 1.0, + "content": "square in the", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 89, + 629, + 520, + 642 + ], + "spans": [ + { + "bbox": [ + 89, + 629, + 520, + 642 + ], + "score": 1.0, + "content": "center of 64x64 image. Half mask mean that one of upper, lower, left or right half of the image is unobserved.", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 90, + 640, + 520, + 652 + ], + "spans": [ + { + "bbox": [ + 90, + 640, + 520, + 652 + ], + "score": 1.0, + "content": "All these types of a half are equiprobable. Random mask means that we use pixelwise-independent Bernoulli", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 90, + 96, + 521, + 108 + ], + "spans": [ + { + "bbox": [ + 90, + 96, + 521, + 108 + ], + "score": 1.0, + "content": "distribution with probability 0.8 to form a mask of unobserved pixels. Pattern mask is proposed in Pathak", + "type": "text", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 89, + 106, + 520, + 119 + ], + "spans": [ + { + "bbox": [ + 89, + 106, + 483, + 119 + ], + "score": 1.0, + "content": "et al. (2016). As we deduced from the code 3, the generation process is follows: firstly we generate", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 484, + 107, + 520, + 117 + ], + "score": 0.54, + "content": "6 0 0 \\times 6 0 0", + "type": "inline_equation", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 90, + 118, + 521, + 130 + ], + "spans": [ + { + "bbox": [ + 90, + 118, + 521, + 130 + ], + "score": 1.0, + "content": "one-channel image with uniform distribution over pixels, then bicubically interpolate it to image of size", + "type": "text", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 91, + 128, + 522, + 141 + ], + "spans": [ + { + "bbox": [ + 91, + 129, + 146, + 139 + ], + "score": 0.7, + "content": "1 0 0 0 0 \\mathrm { x } 1 0 0 0 0", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 146, + 128, + 307, + 141 + ], + "score": 1.0, + "content": ", and then apply Heaviside step function", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 307, + 128, + 359, + 141 + ], + "score": 0.92, + "content": "H ( x - 0 . 2 5 )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 359, + 128, + 522, + 141 + ], + "score": 1.0, + "content": "(i. e. all points with value less than 0.25", + "type": "text", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 89, + 139, + 521, + 152 + ], + "spans": [ + { + "bbox": [ + 89, + 139, + 436, + 152 + ], + "score": 1.0, + "content": "are considered as unobserved). To sample a mask we sample a random position in this", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 437, + 140, + 492, + 150 + ], + "score": 0.33, + "content": "1 0 0 0 0 \\mathrm { x } 1 0 0 0 0", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 493, + 139, + 521, + 152 + ], + "score": 1.0, + "content": "binary", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 89, + 150, + 522, + 163 + ], + "spans": [ + { + "bbox": [ + 89, + 150, + 155, + 163 + ], + "score": 1.0, + "content": "image and crop", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 155, + 151, + 182, + 161 + ], + "score": 0.63, + "content": "6 4 \\mathrm { x } 6 4", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 182, + 150, + 257, + 163 + ], + "score": 1.0, + "content": "mask. If less than", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 257, + 151, + 277, + 161 + ], + "score": 0.87, + "content": "20 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 277, + 150, + 332, + 163 + ], + "score": 1.0, + "content": "or more than", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 333, + 151, + 352, + 161 + ], + "score": 0.87, + "content": "30 \\%", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 353, + 150, + 522, + 163 + ], + "score": 1.0, + "content": "of pixel are unobserved, than the mask is", + "type": "text", + "cross_page": true + } + ], + "index": 5 + }, + { + "bbox": [ + 89, + 162, + 522, + 174 + ], + "spans": [ + { + "bbox": [ + 89, + 162, + 522, + 174 + ], + "score": 1.0, + "content": "rejected and the position is sampled again. In comparison with this paper in section 5.2 we use the same", + "type": "text", + "cross_page": true + } + ], + "index": 6 + }, + { + "bbox": [ + 89, + 172, + 522, + 185 + ], + "spans": [ + { + "bbox": [ + 89, + 172, + 522, + 185 + ], + "score": 1.0, + "content": "distribution over masks at training and testing stages. We use VAEAC with 64 latent variables and take the", + "type": "text", + "cross_page": true + } + ], + "index": 7 + }, + { + "bbox": [ + 89, + 183, + 426, + 196 + ], + "spans": [ + { + "bbox": [ + 89, + 183, + 426, + 196 + ], + "score": 1.0, + "content": "best model over 50 epochs based on the validation IWAE log-likelihood estimation.", + "type": "text", + "cross_page": true + } + ], + "index": 8 + } + ], + "index": 36, + "bbox_fs": [ + 89, + 617, + 521, + 652 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 91, + 95, + 521, + 195 + ], + "lines": [ + { + "bbox": [ + 90, + 96, + 521, + 108 + ], + "spans": [ + { + "bbox": [ + 90, + 96, + 521, + 108 + ], + "score": 1.0, + "content": "distribution with probability 0.8 to form a mask of unobserved pixels. Pattern mask is proposed in Pathak", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 89, + 106, + 520, + 119 + ], + "spans": [ + { + "bbox": [ + 89, + 106, + 483, + 119 + ], + "score": 1.0, + "content": "et al. (2016). As we deduced from the code 3, the generation process is follows: firstly we generate", + "type": "text" + }, + { + "bbox": [ + 484, + 107, + 520, + 117 + ], + "score": 0.54, + "content": "6 0 0 \\times 6 0 0", + "type": "inline_equation" + } + ], + "index": 1 + }, + { + "bbox": [ + 90, + 118, + 521, + 130 + ], + "spans": [ + { + "bbox": [ + 90, + 118, + 521, + 130 + ], + "score": 1.0, + "content": "one-channel image with uniform distribution over pixels, then bicubically interpolate it to image of size", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 91, + 128, + 522, + 141 + ], + "spans": [ + { + "bbox": [ + 91, + 129, + 146, + 139 + ], + "score": 0.7, + "content": "1 0 0 0 0 \\mathrm { x } 1 0 0 0 0", + "type": "inline_equation" + }, + { + "bbox": [ + 146, + 128, + 307, + 141 + ], + "score": 1.0, + "content": ", and then apply Heaviside step function", + "type": "text" + }, + { + "bbox": [ + 307, + 128, + 359, + 141 + ], + "score": 0.92, + "content": "H ( x - 0 . 2 5 )", + "type": "inline_equation" + }, + { + "bbox": [ + 359, + 128, + 522, + 141 + ], + "score": 1.0, + "content": "(i. e. all points with value less than 0.25", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 89, + 139, + 521, + 152 + ], + "spans": [ + { + "bbox": [ + 89, + 139, + 436, + 152 + ], + "score": 1.0, + "content": "are considered as unobserved). To sample a mask we sample a random position in this", + "type": "text" + }, + { + "bbox": [ + 437, + 140, + 492, + 150 + ], + "score": 0.33, + "content": "1 0 0 0 0 \\mathrm { x } 1 0 0 0 0", + "type": "inline_equation" + }, + { + "bbox": [ + 493, + 139, + 521, + 152 + ], + "score": 1.0, + "content": "binary", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 89, + 150, + 522, + 163 + ], + "spans": [ + { + "bbox": [ + 89, + 150, + 155, + 163 + ], + "score": 1.0, + "content": "image and crop", + "type": "text" + }, + { + "bbox": [ + 155, + 151, + 182, + 161 + ], + "score": 0.63, + "content": "6 4 \\mathrm { x } 6 4", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 150, + 257, + 163 + ], + "score": 1.0, + "content": "mask. If less than", + "type": "text" + }, + { + "bbox": [ + 257, + 151, + 277, + 161 + ], + "score": 0.87, + "content": "20 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 277, + 150, + 332, + 163 + ], + "score": 1.0, + "content": "or more than", + "type": "text" + }, + { + "bbox": [ + 333, + 151, + 352, + 161 + ], + "score": 0.87, + "content": "30 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 150, + 522, + 163 + ], + "score": 1.0, + "content": "of pixel are unobserved, than the mask is", + "type": "text" + } + ], + "index": 5 + }, + { + "bbox": [ + 89, + 162, + 522, + 174 + ], + "spans": [ + { + "bbox": [ + 89, + 162, + 522, + 174 + ], + "score": 1.0, + "content": "rejected and the position is sampled again. In comparison with this paper in section 5.2 we use the same", + "type": "text" + } + ], + "index": 6 + }, + { + "bbox": [ + 89, + 172, + 522, + 185 + ], + "spans": [ + { + "bbox": [ + 89, + 172, + 522, + 185 + ], + "score": 1.0, + "content": "distribution over masks at training and testing stages. We use VAEAC with 64 latent variables and take the", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 89, + 183, + 426, + 196 + ], + "spans": [ + { + "bbox": [ + 89, + 183, + 426, + 196 + ], + "score": 1.0, + "content": "best model over 50 epochs based on the validation IWAE log-likelihood estimation.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 4 + }, + { + "type": "title", + "bbox": [ + 92, + 215, + 264, + 226 + ], + "lines": [ + { + "bbox": [ + 90, + 214, + 266, + 228 + ], + "spans": [ + { + "bbox": [ + 90, + 214, + 266, + 228 + ], + "score": 1.0, + "content": "A.4 GAIN IMPLEMENTATION DETAILS", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 90, + 238, + 517, + 262 + ], + "lines": [ + { + "bbox": [ + 90, + 239, + 518, + 250 + ], + "spans": [ + { + "bbox": [ + 90, + 239, + 518, + 250 + ], + "score": 1.0, + "content": "For missing feature imputation we reimplemented GAIN in PyTorch based on the paper (Yoon et al., 2018)", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 89, + 248, + 352, + 264 + ], + "spans": [ + { + "bbox": [ + 89, + 248, + 352, + 264 + ], + "score": 1.0, + "content": "and the available TensorFlow source code for image inpainting 4.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5 + }, + { + "type": "text", + "bbox": [ + 91, + 266, + 521, + 300 + ], + "lines": [ + { + "bbox": [ + 89, + 266, + 522, + 279 + ], + "spans": [ + { + "bbox": [ + 89, + 266, + 522, + 279 + ], + "score": 1.0, + "content": "For categorical features we use one-hot encoding. We observe in experiments that it works better in terms of", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 90, + 277, + 521, + 290 + ], + "spans": [ + { + "bbox": [ + 90, + 277, + 521, + 290 + ], + "score": 1.0, + "content": "NRMSE and PFC than processing categorical features in GAIN as continuous ones and then rounding them", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 89, + 288, + 184, + 303 + ], + "spans": [ + { + "bbox": [ + 89, + 288, + 184, + 303 + ], + "score": 1.0, + "content": "to the nearest category.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 91, + 306, + 520, + 344 + ], + "lines": [ + { + "bbox": [ + 88, + 305, + 510, + 323 + ], + "spans": [ + { + "bbox": [ + 88, + 305, + 317, + 323 + ], + "score": 1.0, + "content": "For categorical features we also use reconstruction loss", + "type": "text" + }, + { + "bbox": [ + 318, + 306, + 486, + 323 + ], + "score": 0.92, + "content": "\\begin{array} { r } { L _ { M } ( x _ { i } , x _ { i } ^ { \\prime } ) = - \\frac { 1 } { | X _ { i } | } \\sum _ { j = 1 } ^ { | X _ { i } | } x _ { i , j } \\log ( x _ { i , j } ^ { \\prime } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 492, + 307, + 510, + 320 + ], + "score": 0.84, + "content": "\\left| X _ { i } \\right|", + "type": "inline_equation" + } + ], + "index": 15 + }, + { + "bbox": [ + 90, + 321, + 522, + 334 + ], + "spans": [ + { + "bbox": [ + 90, + 321, + 214, + 334 + ], + "score": 1.0, + "content": "the number of categories of the", + "type": "text" + }, + { + "bbox": [ + 215, + 322, + 219, + 331 + ], + "score": 0.7, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 321, + 280, + 334 + ], + "score": 1.0, + "content": "-th feature, and", + "type": "text" + }, + { + "bbox": [ + 280, + 322, + 296, + 334 + ], + "score": 0.88, + "content": "x _ { i , j }", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 321, + 320, + 334 + ], + "score": 1.0, + "content": "is the", + "type": "text" + }, + { + "bbox": [ + 321, + 322, + 326, + 333 + ], + "score": 0.8, + "content": "j", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 321, + 522, + 334 + ], + "score": 1.0, + "content": "-th component of one-hot encoding of the feature", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 91, + 331, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 91, + 334, + 100, + 343 + ], + "score": 0.8, + "content": "x _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 101, + 331, + 127, + 344 + ], + "score": 1.0, + "content": ". Such", + "type": "text" + }, + { + "bbox": [ + 127, + 332, + 144, + 343 + ], + "score": 0.89, + "content": "L _ { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 331, + 505, + 344 + ], + "score": 1.0, + "content": "enforces equal contribution of each categorical feature into the whole reconstruction loss.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 91, + 348, + 519, + 393 + ], + "lines": [ + { + "bbox": [ + 88, + 346, + 520, + 363 + ], + "spans": [ + { + "bbox": [ + 88, + 346, + 227, + 363 + ], + "score": 1.0, + "content": "We use one more modification of", + "type": "text" + }, + { + "bbox": [ + 227, + 348, + 270, + 361 + ], + "score": 0.93, + "content": "L _ { M } ( x , x ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 346, + 503, + 363 + ], + "score": 1.0, + "content": "for binary and categorical features. Cross-entropy loss in", + "type": "text" + }, + { + "bbox": [ + 503, + 349, + 520, + 360 + ], + "score": 0.87, + "content": "L _ { M }", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 89, + 358, + 520, + 373 + ], + "spans": [ + { + "bbox": [ + 89, + 358, + 520, + 373 + ], + "score": 1.0, + "content": "penalizes incorrect reconstructions of categorical and binary features much more than incorrect reconstruc-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 88, + 369, + 521, + 384 + ], + "spans": [ + { + "bbox": [ + 88, + 369, + 521, + 384 + ], + "score": 1.0, + "content": "tions for continuous ones. To avoid such imbalance we mixed L2 and cross-entropy reconstruction losses", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 90, + 382, + 383, + 394 + ], + "spans": [ + { + "bbox": [ + 90, + 382, + 383, + 394 + ], + "score": 1.0, + "content": "for binary and categorical features with weights 0.8 and 0.2 respectively:", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5 + }, + { + "type": "interline_equation", + "bbox": [ + 127, + 405, + 468, + 438 + ], + "lines": [], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 127, + 405, + 468, + 416.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 127, + 416.0, + 468, + 427.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 127, + 427.0, + 468, + 438.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 93, + 449, + 521, + 472 + ], + "lines": [ + { + "bbox": [ + 91, + 448, + 522, + 461 + ], + "spans": [ + { + "bbox": [ + 91, + 448, + 522, + 461 + ], + "score": 1.0, + "content": "We observe in experiments that this modification also works better in terms of NRMSE and PFC than the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 90, + 460, + 154, + 472 + ], + "spans": [ + { + "bbox": [ + 90, + 460, + 154, + 472 + ], + "score": 1.0, + "content": "original model.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + }, + { + "type": "text", + "bbox": [ + 90, + 477, + 520, + 499 + ], + "lines": [ + { + "bbox": [ + 89, + 475, + 521, + 490 + ], + "spans": [ + { + "bbox": [ + 89, + 475, + 248, + 490 + ], + "score": 1.0, + "content": "We use validation set which contains", + "type": "text" + }, + { + "bbox": [ + 249, + 477, + 263, + 487 + ], + "score": 0.85, + "content": "5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 475, + 521, + 490 + ], + "score": 1.0, + "content": "of the observed features for the best model selection (hyper-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 88, + 488, + 245, + 500 + ], + "spans": [ + { + "bbox": [ + 88, + 488, + 245, + 500 + ], + "score": 1.0, + "content": "parameter is the number of iterations).", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5 + }, + { + "type": "text", + "bbox": [ + 91, + 505, + 521, + 560 + ], + "lines": [ + { + "bbox": [ + 88, + 504, + 522, + 518 + ], + "spans": [ + { + "bbox": [ + 88, + 504, + 522, + 518 + ], + "score": 1.0, + "content": "In the original GAIN paper authors propose to use cross-validation for hyper-parameter", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 90, + 516, + 520, + 528 + ], + "spans": [ + { + "bbox": [ + 90, + 516, + 207, + 528 + ], + "score": 0.87, + "content": "\\alpha \\in \\{ 0 . 1 , 0 . 5 , 1 , 2 , 1 0 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 516, + 316, + 528 + ], + "score": 1.0, + "content": ". We observe that using", + "type": "text" + }, + { + "bbox": [ + 316, + 516, + 355, + 526 + ], + "score": 0.89, + "content": "\\alpha ~ = ~ 1 0", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 516, + 406, + 528 + ], + "score": 1.0, + "content": "and a hint", + "type": "text" + }, + { + "bbox": [ + 406, + 516, + 520, + 528 + ], + "score": 0.91, + "content": "h \\ = \\ b \\circ \\ m \\ + \\ 0 . 5 ( 1 \\ - \\ b )", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 90, + 527, + 522, + 538 + ], + "spans": [ + { + "bbox": [ + 90, + 527, + 146, + 538 + ], + "score": 1.0, + "content": "where vector", + "type": "text" + }, + { + "bbox": [ + 146, + 528, + 153, + 537 + ], + "score": 0.7, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 527, + 336, + 538 + ], + "score": 1.0, + "content": "is sampled from Bernoulli distribution with", + "type": "text" + }, + { + "bbox": [ + 336, + 527, + 378, + 538 + ], + "score": 0.89, + "content": "p = 0 . 0 1", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 527, + 522, + 538 + ], + "score": 1.0, + "content": "provides better results in terms of", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 89, + 537, + 522, + 550 + ], + "spans": [ + { + "bbox": [ + 89, + 537, + 305, + 550 + ], + "score": 1.0, + "content": "NRMSE and PFC than the original model with every", + "type": "text" + }, + { + "bbox": [ + 306, + 538, + 399, + 550 + ], + "score": 0.9, + "content": "\\alpha \\in \\{ 0 . 1 , 0 . 5 , 1 , 2 , 1 0 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 537, + 522, + 550 + ], + "score": 1.0, + "content": ". Such hint distribution makes", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 88, + 548, + 379, + 562 + ], + "spans": [ + { + "bbox": [ + 88, + 548, + 379, + 562 + ], + "score": 1.0, + "content": "model theoretically inconsistent but works well in practice (see table 7).", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 91, + 565, + 520, + 599 + ], + "lines": [ + { + "bbox": [ + 89, + 564, + 522, + 578 + ], + "spans": [ + { + "bbox": [ + 89, + 564, + 522, + 578 + ], + "score": 1.0, + "content": "Table 7 shows that our modifications provide consistently not worse or even better imputations than the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 88, + 575, + 522, + 591 + ], + "spans": [ + { + "bbox": [ + 88, + 575, + 522, + 591 + ], + "score": 1.0, + "content": "original GAIN (in terms of NRMSE and PFC, on the considered datasets). So in this paper for the missing", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 90, + 587, + 405, + 600 + ], + "spans": [ + { + "bbox": [ + 90, + 587, + 405, + 600 + ], + "score": 1.0, + "content": "feature imputation problem we report the results of our modification of GAIN.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35 + } + ], + "page_idx": 14, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 620, + 502, + 651 + ], + "lines": [ + { + "bbox": [ + 101, + 618, + 502, + 632 + ], + "spans": [ + { + "bbox": [ + 101, + 618, + 502, + 632 + ], + "score": 1.0, + "content": "3https://github.com/pathak22/context-encoder/blob/master/train_random.lua#", + "type": "text" + } + ] + }, + { + "bbox": [ + 90, + 630, + 114, + 641 + ], + "spans": [ + { + "bbox": [ + 90, + 630, + 114, + 641 + ], + "score": 1.0, + "content": "L273", + "type": "text" + } + ] + }, + { + "bbox": [ + 101, + 638, + 244, + 654 + ], + "spans": [ + { + "bbox": [ + 101, + 638, + 244, + 654 + ], + "score": 1.0, + "content": "4https://github.com/jsyoon0823/GAIN", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 91, + 40, + 277, + 50 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 284, + 671, + 295, + 680 + ], + "lines": [ + { + "bbox": [ + 283, + 670, + 297, + 684 + ], + "spans": [ + { + "bbox": [ + 283, + 670, + 297, + 684 + ], + "score": 1.0, + "content": "15", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 91, + 95, + 521, + 195 + ], + "lines": [], + "index": 4, + "bbox_fs": [ + 89, + 96, + 522, + 196 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 92, + 215, + 264, + 226 + ], + "lines": [ + { + "bbox": [ + 90, + 214, + 266, + 228 + ], + "spans": [ + { + "bbox": [ + 90, + 214, + 266, + 228 + ], + "score": 1.0, + "content": "A.4 GAIN IMPLEMENTATION DETAILS", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 90, + 238, + 517, + 262 + ], + "lines": [ + { + "bbox": [ + 90, + 239, + 518, + 250 + ], + "spans": [ + { + "bbox": [ + 90, + 239, + 518, + 250 + ], + "score": 1.0, + "content": "For missing feature imputation we reimplemented GAIN in PyTorch based on the paper (Yoon et al., 2018)", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 89, + 248, + 352, + 264 + ], + "spans": [ + { + "bbox": [ + 89, + 248, + 352, + 264 + ], + "score": 1.0, + "content": "and the available TensorFlow source code for image inpainting 4.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10.5, + "bbox_fs": [ + 89, + 239, + 518, + 264 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 266, + 521, + 300 + ], + "lines": [ + { + "bbox": [ + 89, + 266, + 522, + 279 + ], + "spans": [ + { + "bbox": [ + 89, + 266, + 522, + 279 + ], + "score": 1.0, + "content": "For categorical features we use one-hot encoding. We observe in experiments that it works better in terms of", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 90, + 277, + 521, + 290 + ], + "spans": [ + { + "bbox": [ + 90, + 277, + 521, + 290 + ], + "score": 1.0, + "content": "NRMSE and PFC than processing categorical features in GAIN as continuous ones and then rounding them", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 89, + 288, + 184, + 303 + ], + "spans": [ + { + "bbox": [ + 89, + 288, + 184, + 303 + ], + "score": 1.0, + "content": "to the nearest category.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13, + "bbox_fs": [ + 89, + 266, + 522, + 303 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 306, + 520, + 344 + ], + "lines": [ + { + "bbox": [ + 88, + 305, + 510, + 323 + ], + "spans": [ + { + "bbox": [ + 88, + 305, + 317, + 323 + ], + "score": 1.0, + "content": "For categorical features we also use reconstruction loss", + "type": "text" + }, + { + "bbox": [ + 318, + 306, + 486, + 323 + ], + "score": 0.92, + "content": "\\begin{array} { r } { L _ { M } ( x _ { i } , x _ { i } ^ { \\prime } ) = - \\frac { 1 } { | X _ { i } | } \\sum _ { j = 1 } ^ { | X _ { i } | } x _ { i , j } \\log ( x _ { i , j } ^ { \\prime } ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 492, + 307, + 510, + 320 + ], + "score": 0.84, + "content": "\\left| X _ { i } \\right|", + "type": "inline_equation" + } + ], + "index": 15 + }, + { + "bbox": [ + 90, + 321, + 522, + 334 + ], + "spans": [ + { + "bbox": [ + 90, + 321, + 214, + 334 + ], + "score": 1.0, + "content": "the number of categories of the", + "type": "text" + }, + { + "bbox": [ + 215, + 322, + 219, + 331 + ], + "score": 0.7, + "content": "i", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 321, + 280, + 334 + ], + "score": 1.0, + "content": "-th feature, and", + "type": "text" + }, + { + "bbox": [ + 280, + 322, + 296, + 334 + ], + "score": 0.88, + "content": "x _ { i , j }", + "type": "inline_equation" + }, + { + "bbox": [ + 297, + 321, + 320, + 334 + ], + "score": 1.0, + "content": "is the", + "type": "text" + }, + { + "bbox": [ + 321, + 322, + 326, + 333 + ], + "score": 0.8, + "content": "j", + "type": "inline_equation" + }, + { + "bbox": [ + 326, + 321, + 522, + 334 + ], + "score": 1.0, + "content": "-th component of one-hot encoding of the feature", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 91, + 331, + 505, + 344 + ], + "spans": [ + { + "bbox": [ + 91, + 334, + 100, + 343 + ], + "score": 0.8, + "content": "x _ { i }", + "type": "inline_equation" + }, + { + "bbox": [ + 101, + 331, + 127, + 344 + ], + "score": 1.0, + "content": ". Such", + "type": "text" + }, + { + "bbox": [ + 127, + 332, + 144, + 343 + ], + "score": 0.89, + "content": "L _ { M }", + "type": "inline_equation" + }, + { + "bbox": [ + 144, + 331, + 505, + 344 + ], + "score": 1.0, + "content": "enforces equal contribution of each categorical feature into the whole reconstruction loss.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16, + "bbox_fs": [ + 88, + 305, + 522, + 344 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 348, + 519, + 393 + ], + "lines": [ + { + "bbox": [ + 88, + 346, + 520, + 363 + ], + "spans": [ + { + "bbox": [ + 88, + 346, + 227, + 363 + ], + "score": 1.0, + "content": "We use one more modification of", + "type": "text" + }, + { + "bbox": [ + 227, + 348, + 270, + 361 + ], + "score": 0.93, + "content": "L _ { M } ( x , x ^ { \\prime } )", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 346, + 503, + 363 + ], + "score": 1.0, + "content": "for binary and categorical features. Cross-entropy loss in", + "type": "text" + }, + { + "bbox": [ + 503, + 349, + 520, + 360 + ], + "score": 0.87, + "content": "L _ { M }", + "type": "inline_equation" + } + ], + "index": 18 + }, + { + "bbox": [ + 89, + 358, + 520, + 373 + ], + "spans": [ + { + "bbox": [ + 89, + 358, + 520, + 373 + ], + "score": 1.0, + "content": "penalizes incorrect reconstructions of categorical and binary features much more than incorrect reconstruc-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 88, + 369, + 521, + 384 + ], + "spans": [ + { + "bbox": [ + 88, + 369, + 521, + 384 + ], + "score": 1.0, + "content": "tions for continuous ones. To avoid such imbalance we mixed L2 and cross-entropy reconstruction losses", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 90, + 382, + 383, + 394 + ], + "spans": [ + { + "bbox": [ + 90, + 382, + 383, + 394 + ], + "score": 1.0, + "content": "for binary and categorical features with weights 0.8 and 0.2 respectively:", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 19.5, + "bbox_fs": [ + 88, + 346, + 521, + 394 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 127, + 405, + 468, + 438 + ], + "lines": [], + "index": 23, + "virtual_lines": [ + { + "bbox": [ + 127, + 405, + 468, + 416.0 + ], + "spans": [], + "index": 22 + }, + { + "bbox": [ + 127, + 416.0, + 468, + 427.0 + ], + "spans": [], + "index": 23 + }, + { + "bbox": [ + 127, + 427.0, + 468, + 438.0 + ], + "spans": [], + "index": 24 + } + ] + }, + { + "type": "text", + "bbox": [ + 93, + 449, + 521, + 472 + ], + "lines": [ + { + "bbox": [ + 91, + 448, + 522, + 461 + ], + "spans": [ + { + "bbox": [ + 91, + 448, + 522, + 461 + ], + "score": 1.0, + "content": "We observe in experiments that this modification also works better in terms of NRMSE and PFC than the", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 90, + 460, + 154, + 472 + ], + "spans": [ + { + "bbox": [ + 90, + 460, + 154, + 472 + ], + "score": 1.0, + "content": "original model.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 90, + 448, + 522, + 472 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 477, + 520, + 499 + ], + "lines": [ + { + "bbox": [ + 89, + 475, + 521, + 490 + ], + "spans": [ + { + "bbox": [ + 89, + 475, + 248, + 490 + ], + "score": 1.0, + "content": "We use validation set which contains", + "type": "text" + }, + { + "bbox": [ + 249, + 477, + 263, + 487 + ], + "score": 0.85, + "content": "5 \\%", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 475, + 521, + 490 + ], + "score": 1.0, + "content": "of the observed features for the best model selection (hyper-", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 88, + 488, + 245, + 500 + ], + "spans": [ + { + "bbox": [ + 88, + 488, + 245, + 500 + ], + "score": 1.0, + "content": "parameter is the number of iterations).", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 27.5, + "bbox_fs": [ + 88, + 475, + 521, + 500 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 505, + 521, + 560 + ], + "lines": [ + { + "bbox": [ + 88, + 504, + 522, + 518 + ], + "spans": [ + { + "bbox": [ + 88, + 504, + 522, + 518 + ], + "score": 1.0, + "content": "In the original GAIN paper authors propose to use cross-validation for hyper-parameter", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 90, + 516, + 520, + 528 + ], + "spans": [ + { + "bbox": [ + 90, + 516, + 207, + 528 + ], + "score": 0.87, + "content": "\\alpha \\in \\{ 0 . 1 , 0 . 5 , 1 , 2 , 1 0 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 207, + 516, + 316, + 528 + ], + "score": 1.0, + "content": ". We observe that using", + "type": "text" + }, + { + "bbox": [ + 316, + 516, + 355, + 526 + ], + "score": 0.89, + "content": "\\alpha ~ = ~ 1 0", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 516, + 406, + 528 + ], + "score": 1.0, + "content": "and a hint", + "type": "text" + }, + { + "bbox": [ + 406, + 516, + 520, + 528 + ], + "score": 0.91, + "content": "h \\ = \\ b \\circ \\ m \\ + \\ 0 . 5 ( 1 \\ - \\ b )", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 90, + 527, + 522, + 538 + ], + "spans": [ + { + "bbox": [ + 90, + 527, + 146, + 538 + ], + "score": 1.0, + "content": "where vector", + "type": "text" + }, + { + "bbox": [ + 146, + 528, + 153, + 537 + ], + "score": 0.7, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 153, + 527, + 336, + 538 + ], + "score": 1.0, + "content": "is sampled from Bernoulli distribution with", + "type": "text" + }, + { + "bbox": [ + 336, + 527, + 378, + 538 + ], + "score": 0.89, + "content": "p = 0 . 0 1", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 527, + 522, + 538 + ], + "score": 1.0, + "content": "provides better results in terms of", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 89, + 537, + 522, + 550 + ], + "spans": [ + { + "bbox": [ + 89, + 537, + 305, + 550 + ], + "score": 1.0, + "content": "NRMSE and PFC than the original model with every", + "type": "text" + }, + { + "bbox": [ + 306, + 538, + 399, + 550 + ], + "score": 0.9, + "content": "\\alpha \\in \\{ 0 . 1 , 0 . 5 , 1 , 2 , 1 0 \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 537, + 522, + 550 + ], + "score": 1.0, + "content": ". Such hint distribution makes", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 88, + 548, + 379, + 562 + ], + "spans": [ + { + "bbox": [ + 88, + 548, + 379, + 562 + ], + "score": 1.0, + "content": "model theoretically inconsistent but works well in practice (see table 7).", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31, + "bbox_fs": [ + 88, + 504, + 522, + 562 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 565, + 520, + 599 + ], + "lines": [ + { + "bbox": [ + 89, + 564, + 522, + 578 + ], + "spans": [ + { + "bbox": [ + 89, + 564, + 522, + 578 + ], + "score": 1.0, + "content": "Table 7 shows that our modifications provide consistently not worse or even better imputations than the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 88, + 575, + 522, + 591 + ], + "spans": [ + { + "bbox": [ + 88, + 575, + 522, + 591 + ], + "score": 1.0, + "content": "original GAIN (in terms of NRMSE and PFC, on the considered datasets). So in this paper for the missing", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 90, + 587, + 405, + 600 + ], + "spans": [ + { + "bbox": [ + 90, + 587, + 405, + 600 + ], + "score": 1.0, + "content": "feature imputation problem we report the results of our modification of GAIN.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35, + "bbox_fs": [ + 88, + 564, + 522, + 600 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 92, + 146, + 519, + 310 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 90, + 93, + 522, + 138 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 90, + 93, + 521, + 106 + ], + "spans": [ + { + "bbox": [ + 90, + 93, + 521, + 106 + ], + "score": 1.0, + "content": "Table 7: NRMSE (for continuous datasets) or PFC (for categorical ones) of imputations for different GAIN", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 90, + 104, + 522, + 118 + ], + "spans": [ + { + "bbox": [ + 90, + 104, + 422, + 118 + ], + "score": 1.0, + "content": "modifications. Less is better. “Our modification” includes the reconstruction loss", + "type": "text" + }, + { + "bbox": [ + 422, + 104, + 439, + 117 + ], + "score": 0.9, + "content": "L _ { M } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 104, + 522, + 118 + ], + "score": 1.0, + "content": "(12), Bernoulli dis-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 90, + 115, + 520, + 127 + ], + "spans": [ + { + "bbox": [ + 90, + 116, + 147, + 127 + ], + "score": 1.0, + "content": "tribution over", + "type": "text" + }, + { + "bbox": [ + 148, + 116, + 154, + 125 + ], + "score": 0.48, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 116, + 331, + 127 + ], + "score": 1.0, + "content": "in the hint generation procedure, and fixed", + "type": "text" + }, + { + "bbox": [ + 331, + 115, + 365, + 126 + ], + "score": 0.9, + "content": "\\alpha = 1 0", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 116, + 520, + 127 + ], + "score": 1.0, + "content": ". Other columns refers original GAIN", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 90, + 126, + 327, + 139 + ], + "spans": [ + { + "bbox": [ + 90, + 126, + 315, + 139 + ], + "score": 1.0, + "content": "without these modifications and with different values of", + "type": "text" + }, + { + "bbox": [ + 315, + 129, + 322, + 136 + ], + "score": 0.67, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 126, + 327, + 139 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "table_body", + "bbox": [ + 92, + 146, + 519, + 310 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 92, + 146, + 519, + 310 + ], + "spans": [ + { + "bbox": [ + 92, + 146, + 519, + 310 + ], + "score": 0.98, + "html": "
DatasetOur modificationα=10α=2α=1α= 0.5α=0.1
Boston0.78±0.030.87±0.021.0 ± 0.11.0 ± 0.11.02 ± 0.051.6±0.2
Breast0.67 ± 0.010.80±0.051.00 ± 0.051.10 ± 0.071.19 ± 0.051.52 ± 0.06
Concrete0.96 ± 0.010.98 ± 0.021.02 ± 0.021.13 ± 0.061.17 ± 0.041.3 ± 0.1
Diabetes0.911 ± 0.0090.93 ±0.031.05 ± 0.041.07 ± 0.071.21 ± 0.071.6 ± 0.1
Digits0.79 ± 0.020.88 ± 0.011.05 ± 0.021.13 ± 0.021.24 ± 0.081.4± 0.2
Glass1.06 ± 0.051.04 ± 0.051.19 ± 0.061.4 ± 0.21.6 ± 0.11.81 ± 0.10
Iris0.72 ±0.040.73±0.060.83 ±0.080.97 ± 0.091.2 ± 0.21.3±0.2
Mushroom0.271 ± 0.0030.404 ± 0.0040.52 ± 0.050.55 ± 0.010.56 ± 0.030.64± 0.06
Orthopedic0.91 ± 0.030.91 ±0.081.1 ± 0.11.2 ± 0.11.34 ± 0.081.6 ± 0.2
Phishing0.427 ± 0.0100.52 ±0.020.54±0.020.543 ± 0.0100.56 ± 0.010.57 ± 0.04
WallRobot0.907 ± 0.0050.924± 0.0050.933 ± 0.0080.95 ± 0.011.00 ± 0.021.26 ± 0.04
WhiteWine0.97±0.021.02 ± 0.041.2 ± 0.11.3 ± 0.11.6 ± 0.11.86 ± 0.08
Yeast0.99 ±0.031.3±0.21.6 ± 0.11.83 ± 0.091.9 ±0.12.4± 0.4
Zoo0.20 ±0.020.24± 0.050.35 ± 0.060.36 ± 0.030.43 ± 0.040.433 ± 0.004
", + "type": "table", + "image_path": "bd6b6b98454a1eb53c64484cc4462e2c757911704352b0336a4bd11fa6d9a4c0.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 92, + 146, + 519, + 200.66666666666666 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 92, + 200.66666666666666, + 519, + 255.33333333333331 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 92, + 255.33333333333331, + 519, + 310.0 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 3.25 + }, + { + "type": "title", + "bbox": [ + 91, + 329, + 155, + 342 + ], + "lines": [ + { + "bbox": [ + 88, + 327, + 157, + 344 + ], + "spans": [ + { + "bbox": [ + 88, + 327, + 157, + 344 + ], + "score": 1.0, + "content": "B THEORY", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "title", + "bbox": [ + 91, + 354, + 220, + 366 + ], + "lines": [ + { + "bbox": [ + 89, + 353, + 222, + 367 + ], + "spans": [ + { + "bbox": [ + 89, + 353, + 222, + 367 + ], + "score": 1.0, + "content": "B.1 VAEAC UNIVERSALITY", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 91, + 375, + 521, + 409 + ], + "lines": [ + { + "bbox": [ + 90, + 374, + 522, + 389 + ], + "spans": [ + { + "bbox": [ + 90, + 374, + 522, + 389 + ], + "score": 1.0, + "content": "The theoretical guarantees that VAEAC can model arbitrary distribution are based on the same guarantees", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 90, + 385, + 521, + 399 + ], + "spans": [ + { + "bbox": [ + 90, + 385, + 521, + 399 + ], + "score": 1.0, + "content": "for Condtitional Variational Autoencoder (CVAE). We prove below that if CVAE can model each of the", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 90, + 397, + 387, + 411 + ], + "spans": [ + { + "bbox": [ + 90, + 397, + 190, + 411 + ], + "score": 1.0, + "content": "conditional distributions", + "type": "text" + }, + { + "bbox": [ + 190, + 397, + 236, + 410 + ], + "score": 0.93, + "content": "p ( x _ { b } | x _ { 1 - b } )", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 397, + 387, + 411 + ], + "score": 1.0, + "content": ", then VAEAC can model all of them.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10 + }, + { + "type": "text", + "bbox": [ + 92, + 414, + 520, + 448 + ], + "lines": [ + { + "bbox": [ + 90, + 413, + 520, + 428 + ], + "spans": [ + { + "bbox": [ + 90, + 413, + 158, + 428 + ], + "score": 1.0, + "content": "We can imagine", + "type": "text" + }, + { + "bbox": [ + 158, + 413, + 172, + 425 + ], + "score": 0.86, + "content": "2 ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 413, + 520, + 428 + ], + "score": 1.0, + "content": "CVAEs learned each for the certain mask. Because neural networks are universal ap-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 89, + 424, + 522, + 438 + ], + "spans": [ + { + "bbox": [ + 89, + 424, + 522, + 438 + ], + "score": 1.0, + "content": "proximators, VAEAC networks could model the union of CVAE networks, so that VAEAC network performs", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 90, + 436, + 457, + 450 + ], + "spans": [ + { + "bbox": [ + 90, + 436, + 457, + 450 + ], + "score": 1.0, + "content": "transformation defined by the same network of the corresponding to the given mask CVAE.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13 + }, + { + "type": "interline_equation", + "bbox": [ + 193, + 454, + 415, + 468 + ], + "lines": [ + { + "bbox": [ + 193, + 454, + 415, + 468 + ], + "spans": [ + { + "bbox": [ + 193, + 454, + 415, + 468 + ], + "score": 0.79, + "content": "p _ { \\psi , V A E A C } ( z | x _ { 1 - b } , b ) = p _ { \\psi , C V A E , 1 - b } ( z | x _ { 1 - b } ) \\forall x , b", + "type": "interline_equation", + "image_path": "ebd07ceb7235754b921ee91da23d856f5d6aa8b4c7359835c4a8bda147c1fb4c.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 193, + 454, + 415, + 468 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "interline_equation", + "bbox": [ + 180, + 473, + 432, + 487 + ], + "lines": [ + { + "bbox": [ + 180, + 473, + 432, + 487 + ], + "spans": [ + { + "bbox": [ + 180, + 473, + 432, + 487 + ], + "score": 0.84, + "content": "p _ { \\theta , V A E A C } ( x _ { b } | z , x _ { 1 - b } , b ) = p _ { \\theta , C V A E , 1 - b } ( x _ { b } | z , x _ { 1 - b } ) \\forall z , x , b", + "type": "interline_equation", + "image_path": "6dbaafed2c4133461a5e6db047a62653e52e4a0973ef592faa2ba8d413f8f487.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 180, + 473, + 432, + 487 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 91, + 489, + 336, + 501 + ], + "lines": [ + { + "bbox": [ + 90, + 488, + 336, + 503 + ], + "spans": [ + { + "bbox": [ + 90, + 488, + 236, + 503 + ], + "score": 1.0, + "content": "So if CVAE models any distribution", + "type": "text" + }, + { + "bbox": [ + 236, + 489, + 264, + 502 + ], + "score": 0.92, + "content": "p ( x | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 488, + 336, + 503 + ], + "score": 1.0, + "content": ", VAEAC also do.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 91, + 506, + 522, + 573 + ], + "lines": [ + { + "bbox": [ + 90, + 506, + 521, + 518 + ], + "spans": [ + { + "bbox": [ + 90, + 506, + 521, + 518 + ], + "score": 1.0, + "content": "The guarantees for CVAE in the case of continuous variables are based on the point that every smooth dis-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 89, + 516, + 522, + 529 + ], + "spans": [ + { + "bbox": [ + 89, + 516, + 522, + 529 + ], + "score": 1.0, + "content": "tribution can be approximated with a large enough mixture of Gaussians, which is a special case of CVAE’s", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 88, + 528, + 522, + 541 + ], + "spans": [ + { + "bbox": [ + 88, + 528, + 522, + 541 + ], + "score": 1.0, + "content": "generative model. These guarantees can be extended on the case of categorical-continuous variables also.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 89, + 539, + 522, + 551 + ], + "spans": [ + { + "bbox": [ + 89, + 539, + 522, + 551 + ], + "score": 1.0, + "content": "Actually, there are distributions over categorical variables which CVAE with Gaussian prior and proposal", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 89, + 549, + 522, + 563 + ], + "spans": [ + { + "bbox": [ + 89, + 549, + 522, + 563 + ], + "score": 1.0, + "content": "distributions cannot learn. Nevertheless, this kind of limitation is not fundamental and is caused by poor", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 89, + 561, + 205, + 573 + ], + "spans": [ + { + "bbox": [ + 89, + 561, + 205, + 573 + ], + "score": 1.0, + "content": "proposal distribution family.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20.5 + }, + { + "type": "title", + "bbox": [ + 91, + 586, + 452, + 598 + ], + "lines": [ + { + "bbox": [ + 90, + 586, + 452, + 598 + ], + "spans": [ + { + "bbox": [ + 90, + 586, + 452, + 598 + ], + "score": 1.0, + "content": "B.2 WHY VAEAC NEEDS TARGET VALUES FOR MISSING FEATURES IMPUTATION?", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 91, + 607, + 521, + 652 + ], + "lines": [ + { + "bbox": [ + 90, + 607, + 521, + 620 + ], + "spans": [ + { + "bbox": [ + 90, + 607, + 187, + 620 + ], + "score": 1.0, + "content": "Consider a dataset with", + "type": "text" + }, + { + "bbox": [ + 187, + 608, + 197, + 617 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 607, + 281, + 620 + ], + "score": 1.0, + "content": "-dimensional objects", + "type": "text" + }, + { + "bbox": [ + 282, + 610, + 289, + 617 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 607, + 521, + 620 + ], + "score": 1.0, + "content": "where each feature may be missing (which we denote by", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 91, + 618, + 522, + 631 + ], + "spans": [ + { + "bbox": [ + 91, + 619, + 124, + 630 + ], + "score": 0.87, + "content": "x _ { i } = \\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 618, + 223, + 631 + ], + "score": 1.0, + "content": ") and their target values", + "type": "text" + }, + { + "bbox": [ + 223, + 620, + 230, + 630 + ], + "score": 0.76, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 618, + 522, + 631 + ], + "score": 1.0, + "content": ". In this section we show that the better results are achieved when our", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 90, + 629, + 522, + 642 + ], + "spans": [ + { + "bbox": [ + 90, + 629, + 293, + 642 + ], + "score": 1.0, + "content": "model learns the concatenation of objects features", + "type": "text" + }, + { + "bbox": [ + 293, + 632, + 299, + 639 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 629, + 347, + 642 + ], + "score": 1.0, + "content": "and targets", + "type": "text" + }, + { + "bbox": [ + 347, + 631, + 353, + 641 + ], + "score": 0.73, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 629, + 522, + 642 + ], + "score": 1.0, + "content": ". The example that shows the necessity of", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 88, + 639, + 522, + 653 + ], + "spans": [ + { + "bbox": [ + 88, + 639, + 256, + 653 + ], + "score": 1.0, + "content": "it is following. Consider a dataset where", + "type": "text" + }, + { + "bbox": [ + 257, + 641, + 287, + 651 + ], + "score": 0.88, + "content": "x _ { 1 } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 639, + 291, + 653 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 291, + 640, + 362, + 652 + ], + "score": 0.89, + "content": "x _ { 2 } \\sim \\mathcal { N } ( \\bar { x } _ { 2 } | y , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 639, + 366, + 653 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 366, + 640, + 488, + 652 + ], + "score": 0.9, + "content": "p _ { d } ( y = \\mathrm { { 0 } ) = { { p } ( y = 5 ) = 0 . 5 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 639, + 522, + 653 + ], + "score": 1.0, + "content": ". In this", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 26.5 + } + ], + "page_idx": 15, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 40, + 277, + 51 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 284, + 671, + 295, + 680 + ], + "lines": [ + { + "bbox": [ + 283, + 670, + 297, + 684 + ], + "spans": [ + { + "bbox": [ + 283, + 670, + 297, + 684 + ], + "score": 1.0, + "content": "16", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 92, + 146, + 519, + 310 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 90, + 93, + 522, + 138 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 90, + 93, + 521, + 106 + ], + "spans": [ + { + "bbox": [ + 90, + 93, + 521, + 106 + ], + "score": 1.0, + "content": "Table 7: NRMSE (for continuous datasets) or PFC (for categorical ones) of imputations for different GAIN", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 90, + 104, + 522, + 118 + ], + "spans": [ + { + "bbox": [ + 90, + 104, + 422, + 118 + ], + "score": 1.0, + "content": "modifications. Less is better. “Our modification” includes the reconstruction loss", + "type": "text" + }, + { + "bbox": [ + 422, + 104, + 439, + 117 + ], + "score": 0.9, + "content": "L _ { M } ^ { \\prime }", + "type": "inline_equation" + }, + { + "bbox": [ + 439, + 104, + 522, + 118 + ], + "score": 1.0, + "content": "(12), Bernoulli dis-", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 90, + 115, + 520, + 127 + ], + "spans": [ + { + "bbox": [ + 90, + 116, + 147, + 127 + ], + "score": 1.0, + "content": "tribution over", + "type": "text" + }, + { + "bbox": [ + 148, + 116, + 154, + 125 + ], + "score": 0.48, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 154, + 116, + 331, + 127 + ], + "score": 1.0, + "content": "in the hint generation procedure, and fixed", + "type": "text" + }, + { + "bbox": [ + 331, + 115, + 365, + 126 + ], + "score": 0.9, + "content": "\\alpha = 1 0", + "type": "inline_equation" + }, + { + "bbox": [ + 365, + 116, + 520, + 127 + ], + "score": 1.0, + "content": ". Other columns refers original GAIN", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 90, + 126, + 327, + 139 + ], + "spans": [ + { + "bbox": [ + 90, + 126, + 315, + 139 + ], + "score": 1.0, + "content": "without these modifications and with different values of", + "type": "text" + }, + { + "bbox": [ + 315, + 129, + 322, + 136 + ], + "score": 0.67, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 323, + 126, + 327, + 139 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "table_body", + "bbox": [ + 92, + 146, + 519, + 310 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 92, + 146, + 519, + 310 + ], + "spans": [ + { + "bbox": [ + 92, + 146, + 519, + 310 + ], + "score": 0.98, + "html": "
DatasetOur modificationα=10α=2α=1α= 0.5α=0.1
Boston0.78±0.030.87±0.021.0 ± 0.11.0 ± 0.11.02 ± 0.051.6±0.2
Breast0.67 ± 0.010.80±0.051.00 ± 0.051.10 ± 0.071.19 ± 0.051.52 ± 0.06
Concrete0.96 ± 0.010.98 ± 0.021.02 ± 0.021.13 ± 0.061.17 ± 0.041.3 ± 0.1
Diabetes0.911 ± 0.0090.93 ±0.031.05 ± 0.041.07 ± 0.071.21 ± 0.071.6 ± 0.1
Digits0.79 ± 0.020.88 ± 0.011.05 ± 0.021.13 ± 0.021.24 ± 0.081.4± 0.2
Glass1.06 ± 0.051.04 ± 0.051.19 ± 0.061.4 ± 0.21.6 ± 0.11.81 ± 0.10
Iris0.72 ±0.040.73±0.060.83 ±0.080.97 ± 0.091.2 ± 0.21.3±0.2
Mushroom0.271 ± 0.0030.404 ± 0.0040.52 ± 0.050.55 ± 0.010.56 ± 0.030.64± 0.06
Orthopedic0.91 ± 0.030.91 ±0.081.1 ± 0.11.2 ± 0.11.34 ± 0.081.6 ± 0.2
Phishing0.427 ± 0.0100.52 ±0.020.54±0.020.543 ± 0.0100.56 ± 0.010.57 ± 0.04
WallRobot0.907 ± 0.0050.924± 0.0050.933 ± 0.0080.95 ± 0.011.00 ± 0.021.26 ± 0.04
WhiteWine0.97±0.021.02 ± 0.041.2 ± 0.11.3 ± 0.11.6 ± 0.11.86 ± 0.08
Yeast0.99 ±0.031.3±0.21.6 ± 0.11.83 ± 0.091.9 ±0.12.4± 0.4
Zoo0.20 ±0.020.24± 0.050.35 ± 0.060.36 ± 0.030.43 ± 0.040.433 ± 0.004
", + "type": "table", + "image_path": "bd6b6b98454a1eb53c64484cc4462e2c757911704352b0336a4bd11fa6d9a4c0.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 92, + 146, + 519, + 200.66666666666666 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 92, + 200.66666666666666, + 519, + 255.33333333333331 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 92, + 255.33333333333331, + 519, + 310.0 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 3.25 + }, + { + "type": "title", + "bbox": [ + 91, + 329, + 155, + 342 + ], + "lines": [ + { + "bbox": [ + 88, + 327, + 157, + 344 + ], + "spans": [ + { + "bbox": [ + 88, + 327, + 157, + 344 + ], + "score": 1.0, + "content": "B THEORY", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "title", + "bbox": [ + 91, + 354, + 220, + 366 + ], + "lines": [ + { + "bbox": [ + 89, + 353, + 222, + 367 + ], + "spans": [ + { + "bbox": [ + 89, + 353, + 222, + 367 + ], + "score": 1.0, + "content": "B.1 VAEAC UNIVERSALITY", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + }, + { + "type": "text", + "bbox": [ + 91, + 375, + 521, + 409 + ], + "lines": [ + { + "bbox": [ + 90, + 374, + 522, + 389 + ], + "spans": [ + { + "bbox": [ + 90, + 374, + 522, + 389 + ], + "score": 1.0, + "content": "The theoretical guarantees that VAEAC can model arbitrary distribution are based on the same guarantees", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 90, + 385, + 521, + 399 + ], + "spans": [ + { + "bbox": [ + 90, + 385, + 521, + 399 + ], + "score": 1.0, + "content": "for Condtitional Variational Autoencoder (CVAE). We prove below that if CVAE can model each of the", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 90, + 397, + 387, + 411 + ], + "spans": [ + { + "bbox": [ + 90, + 397, + 190, + 411 + ], + "score": 1.0, + "content": "conditional distributions", + "type": "text" + }, + { + "bbox": [ + 190, + 397, + 236, + 410 + ], + "score": 0.93, + "content": "p ( x _ { b } | x _ { 1 - b } )", + "type": "inline_equation" + }, + { + "bbox": [ + 236, + 397, + 387, + 411 + ], + "score": 1.0, + "content": ", then VAEAC can model all of them.", + "type": "text" + } + ], + "index": 11 + } + ], + "index": 10, + "bbox_fs": [ + 90, + 374, + 522, + 411 + ] + }, + { + "type": "text", + "bbox": [ + 92, + 414, + 520, + 448 + ], + "lines": [ + { + "bbox": [ + 90, + 413, + 520, + 428 + ], + "spans": [ + { + "bbox": [ + 90, + 413, + 158, + 428 + ], + "score": 1.0, + "content": "We can imagine", + "type": "text" + }, + { + "bbox": [ + 158, + 413, + 172, + 425 + ], + "score": 0.86, + "content": "2 ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 172, + 413, + 520, + 428 + ], + "score": 1.0, + "content": "CVAEs learned each for the certain mask. Because neural networks are universal ap-", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 89, + 424, + 522, + 438 + ], + "spans": [ + { + "bbox": [ + 89, + 424, + 522, + 438 + ], + "score": 1.0, + "content": "proximators, VAEAC networks could model the union of CVAE networks, so that VAEAC network performs", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 90, + 436, + 457, + 450 + ], + "spans": [ + { + "bbox": [ + 90, + 436, + 457, + 450 + ], + "score": 1.0, + "content": "transformation defined by the same network of the corresponding to the given mask CVAE.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13, + "bbox_fs": [ + 89, + 413, + 522, + 450 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 193, + 454, + 415, + 468 + ], + "lines": [ + { + "bbox": [ + 193, + 454, + 415, + 468 + ], + "spans": [ + { + "bbox": [ + 193, + 454, + 415, + 468 + ], + "score": 0.79, + "content": "p _ { \\psi , V A E A C } ( z | x _ { 1 - b } , b ) = p _ { \\psi , C V A E , 1 - b } ( z | x _ { 1 - b } ) \\forall x , b", + "type": "interline_equation", + "image_path": "ebd07ceb7235754b921ee91da23d856f5d6aa8b4c7359835c4a8bda147c1fb4c.jpg" + } + ] + } + ], + "index": 15, + "virtual_lines": [ + { + "bbox": [ + 193, + 454, + 415, + 468 + ], + "spans": [], + "index": 15 + } + ] + }, + { + "type": "interline_equation", + "bbox": [ + 180, + 473, + 432, + 487 + ], + "lines": [ + { + "bbox": [ + 180, + 473, + 432, + 487 + ], + "spans": [ + { + "bbox": [ + 180, + 473, + 432, + 487 + ], + "score": 0.84, + "content": "p _ { \\theta , V A E A C } ( x _ { b } | z , x _ { 1 - b } , b ) = p _ { \\theta , C V A E , 1 - b } ( x _ { b } | z , x _ { 1 - b } ) \\forall z , x , b", + "type": "interline_equation", + "image_path": "6dbaafed2c4133461a5e6db047a62653e52e4a0973ef592faa2ba8d413f8f487.jpg" + } + ] + } + ], + "index": 16, + "virtual_lines": [ + { + "bbox": [ + 180, + 473, + 432, + 487 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "text", + "bbox": [ + 91, + 489, + 336, + 501 + ], + "lines": [ + { + "bbox": [ + 90, + 488, + 336, + 503 + ], + "spans": [ + { + "bbox": [ + 90, + 488, + 236, + 503 + ], + "score": 1.0, + "content": "So if CVAE models any distribution", + "type": "text" + }, + { + "bbox": [ + 236, + 489, + 264, + 502 + ], + "score": 0.92, + "content": "p ( x | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 488, + 336, + 503 + ], + "score": 1.0, + "content": ", VAEAC also do.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17, + "bbox_fs": [ + 90, + 488, + 336, + 503 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 506, + 522, + 573 + ], + "lines": [ + { + "bbox": [ + 90, + 506, + 521, + 518 + ], + "spans": [ + { + "bbox": [ + 90, + 506, + 521, + 518 + ], + "score": 1.0, + "content": "The guarantees for CVAE in the case of continuous variables are based on the point that every smooth dis-", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 89, + 516, + 522, + 529 + ], + "spans": [ + { + "bbox": [ + 89, + 516, + 522, + 529 + ], + "score": 1.0, + "content": "tribution can be approximated with a large enough mixture of Gaussians, which is a special case of CVAE’s", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 88, + 528, + 522, + 541 + ], + "spans": [ + { + "bbox": [ + 88, + 528, + 522, + 541 + ], + "score": 1.0, + "content": "generative model. These guarantees can be extended on the case of categorical-continuous variables also.", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 89, + 539, + 522, + 551 + ], + "spans": [ + { + "bbox": [ + 89, + 539, + 522, + 551 + ], + "score": 1.0, + "content": "Actually, there are distributions over categorical variables which CVAE with Gaussian prior and proposal", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 89, + 549, + 522, + 563 + ], + "spans": [ + { + "bbox": [ + 89, + 549, + 522, + 563 + ], + "score": 1.0, + "content": "distributions cannot learn. Nevertheless, this kind of limitation is not fundamental and is caused by poor", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 89, + 561, + 205, + 573 + ], + "spans": [ + { + "bbox": [ + 89, + 561, + 205, + 573 + ], + "score": 1.0, + "content": "proposal distribution family.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 20.5, + "bbox_fs": [ + 88, + 506, + 522, + 573 + ] + }, + { + "type": "title", + "bbox": [ + 91, + 586, + 452, + 598 + ], + "lines": [ + { + "bbox": [ + 90, + 586, + 452, + 598 + ], + "spans": [ + { + "bbox": [ + 90, + 586, + 452, + 598 + ], + "score": 1.0, + "content": "B.2 WHY VAEAC NEEDS TARGET VALUES FOR MISSING FEATURES IMPUTATION?", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 91, + 607, + 521, + 652 + ], + "lines": [ + { + "bbox": [ + 90, + 607, + 521, + 620 + ], + "spans": [ + { + "bbox": [ + 90, + 607, + 187, + 620 + ], + "score": 1.0, + "content": "Consider a dataset with", + "type": "text" + }, + { + "bbox": [ + 187, + 608, + 197, + 617 + ], + "score": 0.81, + "content": "D", + "type": "inline_equation" + }, + { + "bbox": [ + 197, + 607, + 281, + 620 + ], + "score": 1.0, + "content": "-dimensional objects", + "type": "text" + }, + { + "bbox": [ + 282, + 610, + 289, + 617 + ], + "score": 0.75, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 289, + 607, + 521, + 620 + ], + "score": 1.0, + "content": "where each feature may be missing (which we denote by", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 91, + 618, + 522, + 631 + ], + "spans": [ + { + "bbox": [ + 91, + 619, + 124, + 630 + ], + "score": 0.87, + "content": "x _ { i } = \\omega", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 618, + 223, + 631 + ], + "score": 1.0, + "content": ") and their target values", + "type": "text" + }, + { + "bbox": [ + 223, + 620, + 230, + 630 + ], + "score": 0.76, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 230, + 618, + 522, + 631 + ], + "score": 1.0, + "content": ". In this section we show that the better results are achieved when our", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 90, + 629, + 522, + 642 + ], + "spans": [ + { + "bbox": [ + 90, + 629, + 293, + 642 + ], + "score": 1.0, + "content": "model learns the concatenation of objects features", + "type": "text" + }, + { + "bbox": [ + 293, + 632, + 299, + 639 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 300, + 629, + 347, + 642 + ], + "score": 1.0, + "content": "and targets", + "type": "text" + }, + { + "bbox": [ + 347, + 631, + 353, + 641 + ], + "score": 0.73, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 353, + 629, + 522, + 642 + ], + "score": 1.0, + "content": ". The example that shows the necessity of", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 88, + 639, + 522, + 653 + ], + "spans": [ + { + "bbox": [ + 88, + 639, + 256, + 653 + ], + "score": 1.0, + "content": "it is following. Consider a dataset where", + "type": "text" + }, + { + "bbox": [ + 257, + 641, + 287, + 651 + ], + "score": 0.88, + "content": "x _ { 1 } = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 639, + 291, + 653 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 291, + 640, + 362, + 652 + ], + "score": 0.89, + "content": "x _ { 2 } \\sim \\mathcal { N } ( \\bar { x } _ { 2 } | y , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 362, + 639, + 366, + 653 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 366, + 640, + 488, + 652 + ], + "score": 0.9, + "content": "p _ { d } ( y = \\mathrm { { 0 } ) = { { p } ( y = 5 ) = 0 . 5 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 489, + 639, + 522, + 653 + ], + "score": 1.0, + "content": ". In this", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 89, + 95, + 520, + 109 + ], + "spans": [ + { + "bbox": [ + 89, + 95, + 110, + 109 + ], + "score": 1.0, + "content": "case", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 111, + 95, + 317, + 108 + ], + "score": 0.87, + "content": "p _ { d } ( x _ { 2 } | x _ { 1 } = 1 ) = 0 . 5 \\mathcal { N } ( x _ { 2 } | 0 , 1 ) + 0 . 5 \\mathcal { N } ( x _ { 2 } | 5 , 1 )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 317, + 95, + 478, + 109 + ], + "score": 1.0, + "content": ". We can see that generating data from", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 479, + 96, + 520, + 108 + ], + "score": 0.91, + "content": "p _ { d } ( x _ { 2 } | x _ { 1 } )", + "type": "inline_equation", + "cross_page": true + } + ], + "index": 0 + }, + { + "bbox": [ + 89, + 106, + 523, + 120 + ], + "spans": [ + { + "bbox": [ + 89, + 106, + 395, + 120 + ], + "score": 1.0, + "content": "may only confuse the classifier, because with probability 0.5 it generates", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 395, + 106, + 455, + 119 + ], + "score": 0.93, + "content": "x _ { 2 } \\sim \\bar { \\mathcal { N } } ( 0 , 1 )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 456, + 106, + 473, + 120 + ], + "score": 1.0, + "content": "for", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 473, + 108, + 502, + 118 + ], + "score": 0.89, + "content": "y = 5", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 502, + 106, + 523, + 120 + ], + "score": 1.0, + "content": "and", + "type": "text", + "cross_page": true + } + ], + "index": 1 + }, + { + "bbox": [ + 91, + 117, + 520, + 131 + ], + "spans": [ + { + "bbox": [ + 91, + 118, + 148, + 130 + ], + "score": 0.93, + "content": "x _ { 2 } \\sim \\mathcal { N } ( 5 , 1 )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 149, + 117, + 164, + 131 + ], + "score": 1.0, + "content": "for", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 165, + 118, + 191, + 129 + ], + "score": 0.91, + "content": "y = 0", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 192, + 117, + 276, + 131 + ], + "score": 1.0, + "content": ". On the other hand,", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 276, + 118, + 387, + 130 + ], + "score": 0.93, + "content": "p _ { d } ( x _ { 2 } | x _ { 1 } , y ) = \\mathcal { N } ( x _ { 2 } | y , 1 )", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 388, + 117, + 468, + 131 + ], + "score": 1.0, + "content": ". Filling gaps using", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 468, + 118, + 520, + 129 + ], + "score": 0.88, + "content": "p _ { d } ( x _ { 2 } | x _ { 1 } , y )", + "type": "inline_equation", + "cross_page": true + } + ], + "index": 2 + }, + { + "bbox": [ + 88, + 127, + 520, + 142 + ], + "spans": [ + { + "bbox": [ + 88, + 127, + 486, + 142 + ], + "score": 1.0, + "content": "may only improve classifier or regressor by giving it some information from the joint distribution", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 486, + 129, + 520, + 141 + ], + "score": 0.9, + "content": "p _ { d } ( x , y )", + "type": "inline_equation", + "cross_page": true + } + ], + "index": 3 + }, + { + "bbox": [ + 89, + 139, + 522, + 152 + ], + "spans": [ + { + "bbox": [ + 89, + 139, + 419, + 152 + ], + "score": 1.0, + "content": "and thus simplifying the dependence to be learned at the training time. So we treat", + "type": "text", + "cross_page": true + }, + { + "bbox": [ + 419, + 141, + 426, + 151 + ], + "score": 0.8, + "content": "y", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 426, + 139, + 522, + 152 + ], + "score": 1.0, + "content": "as an additional feature", + "type": "text", + "cross_page": true + } + ], + "index": 4 + }, + { + "bbox": [ + 89, + 150, + 289, + 163 + ], + "spans": [ + { + "bbox": [ + 89, + 150, + 289, + 163 + ], + "score": 1.0, + "content": "that is always unobserved during the testing time.", + "type": "text", + "cross_page": true + } + ], + "index": 5 + } + ], + "index": 26.5, + "bbox_fs": [ + 88, + 607, + 522, + 653 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 91, + 95, + 521, + 162 + ], + "lines": [ + { + "bbox": [ + 89, + 95, + 520, + 109 + ], + "spans": [ + { + "bbox": [ + 89, + 95, + 110, + 109 + ], + "score": 1.0, + "content": "case", + "type": "text" + }, + { + "bbox": [ + 111, + 95, + 317, + 108 + ], + "score": 0.87, + "content": "p _ { d } ( x _ { 2 } | x _ { 1 } = 1 ) = 0 . 5 \\mathcal { N } ( x _ { 2 } | 0 , 1 ) + 0 . 5 \\mathcal { N } ( x _ { 2 } | 5 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 317, + 95, + 478, + 109 + ], + "score": 1.0, + "content": ". We can see that generating data from", + "type": "text" + }, + { + "bbox": [ + 479, + 96, + 520, + 108 + ], + "score": 0.91, + "content": "p _ { d } ( x _ { 2 } | x _ { 1 } )", + "type": "inline_equation" + } + ], + "index": 0 + }, + { + "bbox": [ + 89, + 106, + 523, + 120 + ], + "spans": [ + { + "bbox": [ + 89, + 106, + 395, + 120 + ], + "score": 1.0, + "content": "may only confuse the classifier, because with probability 0.5 it generates", + "type": "text" + }, + { + "bbox": [ + 395, + 106, + 455, + 119 + ], + "score": 0.93, + "content": "x _ { 2 } \\sim \\bar { \\mathcal { N } } ( 0 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 456, + 106, + 473, + 120 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 473, + 108, + 502, + 118 + ], + "score": 0.89, + "content": "y = 5", + "type": "inline_equation" + }, + { + "bbox": [ + 502, + 106, + 523, + 120 + ], + "score": 1.0, + "content": "and", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 91, + 117, + 520, + 131 + ], + "spans": [ + { + "bbox": [ + 91, + 118, + 148, + 130 + ], + "score": 0.93, + "content": "x _ { 2 } \\sim \\mathcal { N } ( 5 , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 149, + 117, + 164, + 131 + ], + "score": 1.0, + "content": "for", + "type": "text" + }, + { + "bbox": [ + 165, + 118, + 191, + 129 + ], + "score": 0.91, + "content": "y = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 192, + 117, + 276, + 131 + ], + "score": 1.0, + "content": ". On the other hand,", + "type": "text" + }, + { + "bbox": [ + 276, + 118, + 387, + 130 + ], + "score": 0.93, + "content": "p _ { d } ( x _ { 2 } | x _ { 1 } , y ) = \\mathcal { N } ( x _ { 2 } | y , 1 )", + "type": "inline_equation" + }, + { + "bbox": [ + 388, + 117, + 468, + 131 + ], + "score": 1.0, + "content": ". Filling gaps using", + "type": "text" + }, + { + "bbox": [ + 468, + 118, + 520, + 129 + ], + "score": 0.88, + "content": "p _ { d } ( x _ { 2 } | x _ { 1 } , y )", + "type": "inline_equation" + } + ], + "index": 2 + }, + { + "bbox": [ + 88, + 127, + 520, + 142 + ], + "spans": [ + { + "bbox": [ + 88, + 127, + 486, + 142 + ], + "score": 1.0, + "content": "may only improve classifier or regressor by giving it some information from the joint distribution", + "type": "text" + }, + { + "bbox": [ + 486, + 129, + 520, + 141 + ], + "score": 0.9, + "content": "p _ { d } ( x , y )", + "type": "inline_equation" + } + ], + "index": 3 + }, + { + "bbox": [ + 89, + 139, + 522, + 152 + ], + "spans": [ + { + "bbox": [ + 89, + 139, + 419, + 152 + ], + "score": 1.0, + "content": "and thus simplifying the dependence to be learned at the training time. So we treat", + "type": "text" + }, + { + "bbox": [ + 419, + 141, + 426, + 151 + ], + "score": 0.8, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 426, + 139, + 522, + 152 + ], + "score": 1.0, + "content": "as an additional feature", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 89, + 150, + 289, + 163 + ], + "spans": [ + { + "bbox": [ + 89, + 150, + 289, + 163 + ], + "score": 1.0, + "content": "that is always unobserved during the testing time.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 2.5 + }, + { + "type": "title", + "bbox": [ + 90, + 175, + 411, + 187 + ], + "lines": [ + { + "bbox": [ + 89, + 175, + 413, + 187 + ], + "spans": [ + { + "bbox": [ + 89, + 175, + 413, + 187 + ], + "score": 1.0, + "content": "B.3 UNIVERSAL MARGINALIZER: TRAINING PROCEDURE MODIFICATION", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 91, + 195, + 521, + 241 + ], + "lines": [ + { + "bbox": [ + 89, + 194, + 521, + 209 + ], + "spans": [ + { + "bbox": [ + 89, + 194, + 521, + 209 + ], + "score": 1.0, + "content": "The problem authors did not address in the original paper is the relation between the distribution of un-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 89, + 205, + 521, + 220 + ], + "spans": [ + { + "bbox": [ + 89, + 205, + 181, + 220 + ], + "score": 1.0, + "content": "observed components", + "type": "text" + }, + { + "bbox": [ + 181, + 207, + 199, + 218 + ], + "score": 0.91, + "content": "p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 205, + 499, + 220 + ], + "score": 1.0, + "content": "at the testing stage and the distribution of masks in the requests to UM", + "type": "text" + }, + { + "bbox": [ + 500, + 207, + 517, + 218 + ], + "score": 0.86, + "content": "{ \\hat { p } } ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 518, + 205, + 521, + 220 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 90, + 217, + 522, + 231 + ], + "spans": [ + { + "bbox": [ + 90, + 217, + 208, + 231 + ], + "score": 1.0, + "content": "The distribution over masks", + "type": "text" + }, + { + "bbox": [ + 208, + 218, + 226, + 230 + ], + "score": 0.91, + "content": "p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 217, + 327, + 231 + ], + "score": 1.0, + "content": "induces the distribution", + "type": "text" + }, + { + "bbox": [ + 327, + 218, + 345, + 230 + ], + "score": 0.87, + "content": "{ \\hat { p } } ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 217, + 443, + 231 + ], + "score": 1.0, + "content": ", and in the most cases", + "type": "text" + }, + { + "bbox": [ + 443, + 218, + 495, + 230 + ], + "score": 0.93, + "content": "p ( b ) \\neq { \\hat { p } } ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 496, + 217, + 522, + 231 + ], + "score": 1.0, + "content": ". The", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 89, + 227, + 486, + 243 + ], + "spans": [ + { + "bbox": [ + 89, + 227, + 138, + 243 + ], + "score": 1.0, + "content": "distribution", + "type": "text" + }, + { + "bbox": [ + 139, + 228, + 156, + 241 + ], + "score": 0.91, + "content": "{ \\hat { p } } ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 227, + 294, + 243 + ], + "score": 1.0, + "content": "also depends on the permutations", + "type": "text" + }, + { + "bbox": [ + 294, + 229, + 356, + 242 + ], + "score": 0.92, + "content": "( i _ { 1 } , i _ { 2 } , \\dots , i _ { | b | } )", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 227, + 486, + 243 + ], + "score": 1.0, + "content": "that we use to generate objects.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8.5 + }, + { + "type": "text", + "bbox": [ + 90, + 246, + 521, + 308 + ], + "lines": [ + { + "bbox": [ + 89, + 246, + 521, + 261 + ], + "spans": [ + { + "bbox": [ + 89, + 246, + 444, + 261 + ], + "score": 1.0, + "content": "We observed in experiments, that UM must be trained using unobserved mask distribution", + "type": "text" + }, + { + "bbox": [ + 445, + 247, + 462, + 259 + ], + "score": 0.9, + "content": "{ \\hat { p } } ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 463, + 246, + 521, + 261 + ], + "score": 1.0, + "content": ". For example,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 88, + 256, + 523, + 274 + ], + "spans": [ + { + "bbox": [ + 88, + 256, + 165, + 274 + ], + "score": 1.0, + "content": "if all masks from", + "type": "text" + }, + { + "bbox": [ + 165, + 259, + 183, + 271 + ], + "score": 0.91, + "content": "p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 256, + 418, + 274 + ], + "score": 1.0, + "content": "have a fixed number of unobserved components (e. g.,", + "type": "text" + }, + { + "bbox": [ + 419, + 258, + 429, + 272 + ], + "score": 0.7, + "content": "\\begin{array} { l } { { \\frac { D } { 2 } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 256, + 523, + 274 + ], + "score": 1.0, + "content": "), then UM will never", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 87, + 267, + 524, + 290 + ], + "spans": [ + { + "bbox": [ + 87, + 267, + 214, + 290 + ], + "score": 1.0, + "content": "see an example of mask with", + "type": "text" + }, + { + "bbox": [ + 214, + 271, + 280, + 285 + ], + "score": 0.93, + "content": "\\begin{array} { r } { { 1 , 2 , \\ldots , \\frac { D } { 2 } - 1 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 267, + 524, + 290 + ], + "score": 1.0, + "content": "unobserved components, which is necessary to generate a", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 87, + 282, + 523, + 301 + ], + "spans": [ + { + "bbox": [ + 87, + 282, + 184, + 301 + ], + "score": 1.0, + "content": "sample conditioned on", + "type": "text" + }, + { + "bbox": [ + 184, + 284, + 194, + 298 + ], + "score": 0.88, + "content": "\\textstyle { \\frac { D } { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 282, + 523, + 301 + ], + "score": 1.0, + "content": "components. That leads to drastically low likelihood estimate for the test set and", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 89, + 296, + 171, + 309 + ], + "spans": [ + { + "bbox": [ + 89, + 296, + 171, + 309 + ], + "score": 1.0, + "content": "unrealistic samples.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 91, + 312, + 521, + 361 + ], + "lines": [ + { + "bbox": [ + 89, + 311, + 522, + 326 + ], + "spans": [ + { + "bbox": [ + 89, + 311, + 268, + 326 + ], + "score": 1.0, + "content": "We developed an easy generative process for", + "type": "text" + }, + { + "bbox": [ + 268, + 313, + 286, + 325 + ], + "score": 0.91, + "content": "{ \\hat { p } } ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 311, + 338, + 326 + ], + "score": 1.0, + "content": "for arbitrary", + "type": "text" + }, + { + "bbox": [ + 338, + 313, + 356, + 325 + ], + "score": 0.91, + "content": "p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 311, + 522, + 326 + ], + "score": 1.0, + "content": "if the permutation of unobserved compo-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 89, + 323, + 522, + 338 + ], + "spans": [ + { + "bbox": [ + 89, + 323, + 114, + 338 + ], + "score": 1.0, + "content": "nents", + "type": "text" + }, + { + "bbox": [ + 114, + 324, + 177, + 336 + ], + "score": 0.91, + "content": "( i _ { 1 } , i _ { 2 } , \\dots , i _ { | b | } )", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 323, + 406, + 338 + ], + "score": 1.0, + "content": "is chosen randomly and equiprobably: firstly we generate", + "type": "text" + }, + { + "bbox": [ + 407, + 324, + 446, + 336 + ], + "score": 0.82, + "content": "b _ { 0 } \\sim p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 323, + 450, + 338 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 450, + 324, + 498, + 336 + ], + "score": 0.83, + "content": "u \\sim U [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 498, + 323, + 522, + 338 + ], + "score": 1.0, + "content": ", then", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 91, + 335, + 522, + 351 + ], + "spans": [ + { + "bbox": [ + 91, + 337, + 181, + 349 + ], + "score": 0.88, + "content": "b _ { 1 } \\sim ( \\mathrm { B e r n o u l l i } ( u ) ) ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 335, + 199, + 351 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 199, + 336, + 244, + 348 + ], + "score": 0.9, + "content": "b = b _ { 0 } \\circ b _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 335, + 522, + 351 + ], + "score": 1.0, + "content": ". More complicated generative process exists for a sorted permutation", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 89, + 347, + 228, + 362 + ], + "spans": [ + { + "bbox": [ + 89, + 347, + 117, + 362 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 118, + 349, + 224, + 360 + ], + "score": 0.89, + "content": "i _ { j - 1 } < i _ { j } \\forall j : 2 \\le j \\le | b |", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 347, + 228, + 362 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 91, + 364, + 355, + 376 + ], + "lines": [ + { + "bbox": [ + 88, + 363, + 357, + 379 + ], + "spans": [ + { + "bbox": [ + 88, + 363, + 357, + 379 + ], + "score": 1.0, + "content": "In experiments we use uniform distribution over the permutations.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20 + }, + { + "type": "title", + "bbox": [ + 91, + 392, + 339, + 405 + ], + "lines": [ + { + "bbox": [ + 89, + 390, + 341, + 407 + ], + "spans": [ + { + "bbox": [ + 89, + 390, + 341, + 407 + ], + "score": 1.0, + "content": "C GAUSSIAN STOCHASTIC NEURAL NETWORK", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 90, + 416, + 521, + 518 + ], + "lines": [ + { + "bbox": [ + 89, + 416, + 522, + 430 + ], + "spans": [ + { + "bbox": [ + 89, + 416, + 522, + 430 + ], + "score": 1.0, + "content": "Gaussian stochastic neural network (13) and hybrid model (14) are originally proposed in the paper on", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 89, + 426, + 522, + 441 + ], + "spans": [ + { + "bbox": [ + 89, + 426, + 522, + 441 + ], + "score": 1.0, + "content": "Conditional VAE (Sohn et al., 2015). The motivation authors mention in the paper is as follows. During", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 88, + 437, + 522, + 453 + ], + "spans": [ + { + "bbox": [ + 88, + 437, + 222, + 453 + ], + "score": 1.0, + "content": "training the proposal distribution", + "type": "text" + }, + { + "bbox": [ + 222, + 439, + 263, + 451 + ], + "score": 0.93, + "content": "q _ { \\phi } ( z | x , y )", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 437, + 415, + 453 + ], + "score": 1.0, + "content": "is used to generate the latent variables", + "type": "text" + }, + { + "bbox": [ + 416, + 441, + 422, + 449 + ], + "score": 0.72, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 437, + 522, + 453 + ], + "score": 1.0, + "content": ", while during the testing", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 89, + 449, + 522, + 464 + ], + "spans": [ + { + "bbox": [ + 89, + 449, + 149, + 464 + ], + "score": 1.0, + "content": "stage the prior", + "type": "text" + }, + { + "bbox": [ + 149, + 450, + 181, + 462 + ], + "score": 0.92, + "content": "p _ { \\psi } ( z | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 449, + 522, + 464 + ], + "score": 1.0, + "content": "is used. KL divergence tries to close the gap between two distributions but, according", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 89, + 461, + 521, + 473 + ], + "spans": [ + { + "bbox": [ + 89, + 461, + 521, + 473 + ], + "score": 1.0, + "content": "to authors, it is not enough. To overcome the issue authors propose to use a hybrid model (14), a weighted", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 88, + 471, + 521, + 484 + ], + "spans": [ + { + "bbox": [ + 88, + 471, + 521, + 484 + ], + "score": 1.0, + "content": "mixture of variational lower bound (3) and a single-sample Monte-Carlo estimation of log-likelihood (13).", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 89, + 482, + 522, + 496 + ], + "spans": [ + { + "bbox": [ + 89, + 482, + 522, + 496 + ], + "score": 1.0, + "content": "The model corresponding to the second term is called Gaussian Stochastic Neural Network (13), because", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 89, + 493, + 523, + 506 + ], + "spans": [ + { + "bbox": [ + 89, + 493, + 523, + 506 + ], + "score": 1.0, + "content": "it is a feed-forward neural network with a single Gaussian stochastic layer in the middle. Also GSNN is a", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 90, + 505, + 297, + 518 + ], + "spans": [ + { + "bbox": [ + 90, + 505, + 206, + 518 + ], + "score": 1.0, + "content": "special case of CVAE where", + "type": "text" + }, + { + "bbox": [ + 206, + 505, + 293, + 517 + ], + "score": 0.93, + "content": "q _ { \\phi } ( z | x , y ) = p _ { \\psi } ( z | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 505, + 297, + 518 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 26 + }, + { + "type": "interline_equation", + "bbox": [ + 132, + 531, + 479, + 563 + ], + "lines": [ + { + "bbox": [ + 132, + 531, + 479, + 563 + ], + "spans": [ + { + "bbox": [ + 132, + 531, + 479, + 563 + ], + "score": 0.92, + "content": "\\begin{array} { r } { L _ { G S N N } ( x , y ; \\theta , \\psi ) = \\mathbb { E } _ { p _ { \\psi } ( z | y ) } \\log p _ { \\theta } ( x | z , y ) \\qquad } \\\\ { L ( x , y ; \\theta , \\psi , \\phi ) = \\alpha L _ { C V A E } ( x , y ; \\theta , \\psi , \\phi ) + ( 1 - \\alpha ) L _ { G S N N } ( x , y ; \\theta , \\psi ) , \\quad \\alpha \\in [ 0 , 1 ] } \\end{array}", + "type": "interline_equation", + "image_path": "6eda0f59e7f92ff41e13fc03a857ebaa2e07dbc1dc3cfeb2358c4de8661097dd.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 132, + 531, + 479, + 541.6666666666666 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 132, + 541.6666666666666, + 479, + 552.3333333333333 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 132, + 552.3333333333333, + 479, + 562.9999999999999 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 90, + 568, + 522, + 591 + ], + "lines": [ + { + "bbox": [ + 90, + 568, + 522, + 581 + ], + "spans": [ + { + "bbox": [ + 90, + 568, + 522, + 581 + ], + "score": 1.0, + "content": "Authors report that hybrid model and GSNN outperform CVAE in terms of segmentation accuracy on the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 88, + 579, + 174, + 591 + ], + "spans": [ + { + "bbox": [ + 88, + 579, + 174, + 591 + ], + "score": 1.0, + "content": "majority of datasets.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5 + }, + { + "type": "text", + "bbox": [ + 91, + 596, + 521, + 652 + ], + "lines": [ + { + "bbox": [ + 90, + 596, + 522, + 609 + ], + "spans": [ + { + "bbox": [ + 90, + 596, + 522, + 609 + ], + "score": 1.0, + "content": "We can also add that this technique seems to soften the “holes problem” (Makhzani et al., 2016). In", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 89, + 606, + 522, + 620 + ], + "spans": [ + { + "bbox": [ + 89, + 606, + 298, + 620 + ], + "score": 1.0, + "content": "Makhzani et al. (2016) authors observe that vectors", + "type": "text" + }, + { + "bbox": [ + 299, + 611, + 305, + 617 + ], + "score": 0.67, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 606, + 522, + 620 + ], + "score": 1.0, + "content": "from prior distribution may be different enough from", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 89, + 618, + 522, + 631 + ], + "spans": [ + { + "bbox": [ + 89, + 618, + 134, + 631 + ], + "score": 1.0, + "content": "all vectors", + "type": "text" + }, + { + "bbox": [ + 134, + 620, + 141, + 628 + ], + "score": 0.76, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 618, + 522, + 631 + ], + "score": 1.0, + "content": "from the proposal distribution at the training stage, so the generator network may be confused", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 89, + 628, + 522, + 643 + ], + "spans": [ + { + "bbox": [ + 89, + 628, + 422, + 643 + ], + "score": 1.0, + "content": "at the testing stage. Due to this problem CVAE can have good reconstructions of", + "type": "text" + }, + { + "bbox": [ + 422, + 631, + 429, + 641 + ], + "score": 0.79, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 628, + 455, + 643 + ], + "score": 1.0, + "content": "given", + "type": "text" + }, + { + "bbox": [ + 455, + 629, + 517, + 642 + ], + "score": 0.93, + "content": "z \\sim q _ { \\phi } ( z | x , y )", + "type": "inline_equation" + }, + { + "bbox": [ + 518, + 628, + 522, + 643 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 89, + 640, + 312, + 653 + ], + "spans": [ + { + "bbox": [ + 89, + 640, + 160, + 653 + ], + "score": 1.0, + "content": "while samples of", + "type": "text" + }, + { + "bbox": [ + 160, + 642, + 167, + 652 + ], + "score": 0.8, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 640, + 192, + 653 + ], + "score": 1.0, + "content": "given", + "type": "text" + }, + { + "bbox": [ + 192, + 641, + 244, + 652 + ], + "score": 0.93, + "content": "z \\sim p _ { \\psi } ( z | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 640, + 312, + 653 + ], + "score": 1.0, + "content": "are not realistic.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38 + } + ], + "page_idx": 16, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 40, + 277, + 50 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 284, + 671, + 294, + 680 + ], + "lines": [ + { + "bbox": [ + 283, + 670, + 296, + 684 + ], + "spans": [ + { + "bbox": [ + 283, + 670, + 296, + 684 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 14, + "width": 13 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 91, + 95, + 521, + 162 + ], + "lines": [], + "index": 2.5, + "bbox_fs": [ + 88, + 95, + 523, + 163 + ], + "lines_deleted": true + }, + { + "type": "title", + "bbox": [ + 90, + 175, + 411, + 187 + ], + "lines": [ + { + "bbox": [ + 89, + 175, + 413, + 187 + ], + "spans": [ + { + "bbox": [ + 89, + 175, + 413, + 187 + ], + "score": 1.0, + "content": "B.3 UNIVERSAL MARGINALIZER: TRAINING PROCEDURE MODIFICATION", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "text", + "bbox": [ + 91, + 195, + 521, + 241 + ], + "lines": [ + { + "bbox": [ + 89, + 194, + 521, + 209 + ], + "spans": [ + { + "bbox": [ + 89, + 194, + 521, + 209 + ], + "score": 1.0, + "content": "The problem authors did not address in the original paper is the relation between the distribution of un-", + "type": "text" + } + ], + "index": 7 + }, + { + "bbox": [ + 89, + 205, + 521, + 220 + ], + "spans": [ + { + "bbox": [ + 89, + 205, + 181, + 220 + ], + "score": 1.0, + "content": "observed components", + "type": "text" + }, + { + "bbox": [ + 181, + 207, + 199, + 218 + ], + "score": 0.91, + "content": "p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 205, + 499, + 220 + ], + "score": 1.0, + "content": "at the testing stage and the distribution of masks in the requests to UM", + "type": "text" + }, + { + "bbox": [ + 500, + 207, + 517, + 218 + ], + "score": 0.86, + "content": "{ \\hat { p } } ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 518, + 205, + 521, + 220 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 90, + 217, + 522, + 231 + ], + "spans": [ + { + "bbox": [ + 90, + 217, + 208, + 231 + ], + "score": 1.0, + "content": "The distribution over masks", + "type": "text" + }, + { + "bbox": [ + 208, + 218, + 226, + 230 + ], + "score": 0.91, + "content": "p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 226, + 217, + 327, + 231 + ], + "score": 1.0, + "content": "induces the distribution", + "type": "text" + }, + { + "bbox": [ + 327, + 218, + 345, + 230 + ], + "score": 0.87, + "content": "{ \\hat { p } } ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 346, + 217, + 443, + 231 + ], + "score": 1.0, + "content": ", and in the most cases", + "type": "text" + }, + { + "bbox": [ + 443, + 218, + 495, + 230 + ], + "score": 0.93, + "content": "p ( b ) \\neq { \\hat { p } } ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 496, + 217, + 522, + 231 + ], + "score": 1.0, + "content": ". The", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 89, + 227, + 486, + 243 + ], + "spans": [ + { + "bbox": [ + 89, + 227, + 138, + 243 + ], + "score": 1.0, + "content": "distribution", + "type": "text" + }, + { + "bbox": [ + 139, + 228, + 156, + 241 + ], + "score": 0.91, + "content": "{ \\hat { p } } ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 157, + 227, + 294, + 243 + ], + "score": 1.0, + "content": "also depends on the permutations", + "type": "text" + }, + { + "bbox": [ + 294, + 229, + 356, + 242 + ], + "score": 0.92, + "content": "( i _ { 1 } , i _ { 2 } , \\dots , i _ { | b | } )", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 227, + 486, + 243 + ], + "score": 1.0, + "content": "that we use to generate objects.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 8.5, + "bbox_fs": [ + 89, + 194, + 522, + 243 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 246, + 521, + 308 + ], + "lines": [ + { + "bbox": [ + 89, + 246, + 521, + 261 + ], + "spans": [ + { + "bbox": [ + 89, + 246, + 444, + 261 + ], + "score": 1.0, + "content": "We observed in experiments, that UM must be trained using unobserved mask distribution", + "type": "text" + }, + { + "bbox": [ + 445, + 247, + 462, + 259 + ], + "score": 0.9, + "content": "{ \\hat { p } } ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 463, + 246, + 521, + 261 + ], + "score": 1.0, + "content": ". For example,", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 88, + 256, + 523, + 274 + ], + "spans": [ + { + "bbox": [ + 88, + 256, + 165, + 274 + ], + "score": 1.0, + "content": "if all masks from", + "type": "text" + }, + { + "bbox": [ + 165, + 259, + 183, + 271 + ], + "score": 0.91, + "content": "p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 184, + 256, + 418, + 274 + ], + "score": 1.0, + "content": "have a fixed number of unobserved components (e. g.,", + "type": "text" + }, + { + "bbox": [ + 419, + 258, + 429, + 272 + ], + "score": 0.7, + "content": "\\begin{array} { l } { { \\frac { D } { 2 } } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 256, + 523, + 274 + ], + "score": 1.0, + "content": "), then UM will never", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 87, + 267, + 524, + 290 + ], + "spans": [ + { + "bbox": [ + 87, + 267, + 214, + 290 + ], + "score": 1.0, + "content": "see an example of mask with", + "type": "text" + }, + { + "bbox": [ + 214, + 271, + 280, + 285 + ], + "score": 0.93, + "content": "\\begin{array} { r } { { 1 , 2 , \\ldots , \\frac { D } { 2 } - 1 } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 280, + 267, + 524, + 290 + ], + "score": 1.0, + "content": "unobserved components, which is necessary to generate a", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 87, + 282, + 523, + 301 + ], + "spans": [ + { + "bbox": [ + 87, + 282, + 184, + 301 + ], + "score": 1.0, + "content": "sample conditioned on", + "type": "text" + }, + { + "bbox": [ + 184, + 284, + 194, + 298 + ], + "score": 0.88, + "content": "\\textstyle { \\frac { D } { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 194, + 282, + 523, + 301 + ], + "score": 1.0, + "content": "components. That leads to drastically low likelihood estimate for the test set and", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 89, + 296, + 171, + 309 + ], + "spans": [ + { + "bbox": [ + 89, + 296, + 171, + 309 + ], + "score": 1.0, + "content": "unrealistic samples.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13, + "bbox_fs": [ + 87, + 246, + 524, + 309 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 312, + 521, + 361 + ], + "lines": [ + { + "bbox": [ + 89, + 311, + 522, + 326 + ], + "spans": [ + { + "bbox": [ + 89, + 311, + 268, + 326 + ], + "score": 1.0, + "content": "We developed an easy generative process for", + "type": "text" + }, + { + "bbox": [ + 268, + 313, + 286, + 325 + ], + "score": 0.91, + "content": "{ \\hat { p } } ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 287, + 311, + 338, + 326 + ], + "score": 1.0, + "content": "for arbitrary", + "type": "text" + }, + { + "bbox": [ + 338, + 313, + 356, + 325 + ], + "score": 0.91, + "content": "p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 356, + 311, + 522, + 326 + ], + "score": 1.0, + "content": "if the permutation of unobserved compo-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 89, + 323, + 522, + 338 + ], + "spans": [ + { + "bbox": [ + 89, + 323, + 114, + 338 + ], + "score": 1.0, + "content": "nents", + "type": "text" + }, + { + "bbox": [ + 114, + 324, + 177, + 336 + ], + "score": 0.91, + "content": "( i _ { 1 } , i _ { 2 } , \\dots , i _ { | b | } )", + "type": "inline_equation" + }, + { + "bbox": [ + 177, + 323, + 406, + 338 + ], + "score": 1.0, + "content": "is chosen randomly and equiprobably: firstly we generate", + "type": "text" + }, + { + "bbox": [ + 407, + 324, + 446, + 336 + ], + "score": 0.82, + "content": "b _ { 0 } \\sim p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 446, + 323, + 450, + 338 + ], + "score": 1.0, + "content": ",", + "type": "text" + }, + { + "bbox": [ + 450, + 324, + 498, + 336 + ], + "score": 0.83, + "content": "u \\sim U [ 0 , 1 ]", + "type": "inline_equation" + }, + { + "bbox": [ + 498, + 323, + 522, + 338 + ], + "score": 1.0, + "content": ", then", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 91, + 335, + 522, + 351 + ], + "spans": [ + { + "bbox": [ + 91, + 337, + 181, + 349 + ], + "score": 0.88, + "content": "b _ { 1 } \\sim ( \\mathrm { B e r n o u l l i } ( u ) ) ^ { D }", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 335, + 199, + 351 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 199, + 336, + 244, + 348 + ], + "score": 0.9, + "content": "b = b _ { 0 } \\circ b _ { 1 }", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 335, + 522, + 351 + ], + "score": 1.0, + "content": ". More complicated generative process exists for a sorted permutation", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 89, + 347, + 228, + 362 + ], + "spans": [ + { + "bbox": [ + 89, + 347, + 117, + 362 + ], + "score": 1.0, + "content": "where", + "type": "text" + }, + { + "bbox": [ + 118, + 349, + 224, + 360 + ], + "score": 0.89, + "content": "i _ { j - 1 } < i _ { j } \\forall j : 2 \\le j \\le | b |", + "type": "inline_equation" + }, + { + "bbox": [ + 225, + 347, + 228, + 362 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17.5, + "bbox_fs": [ + 89, + 311, + 522, + 362 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 364, + 355, + 376 + ], + "lines": [ + { + "bbox": [ + 88, + 363, + 357, + 379 + ], + "spans": [ + { + "bbox": [ + 88, + 363, + 357, + 379 + ], + "score": 1.0, + "content": "In experiments we use uniform distribution over the permutations.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 20, + "bbox_fs": [ + 88, + 363, + 357, + 379 + ] + }, + { + "type": "title", + "bbox": [ + 91, + 392, + 339, + 405 + ], + "lines": [ + { + "bbox": [ + 89, + 390, + 341, + 407 + ], + "spans": [ + { + "bbox": [ + 89, + 390, + 341, + 407 + ], + "score": 1.0, + "content": "C GAUSSIAN STOCHASTIC NEURAL NETWORK", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "text", + "bbox": [ + 90, + 416, + 521, + 518 + ], + "lines": [ + { + "bbox": [ + 89, + 416, + 522, + 430 + ], + "spans": [ + { + "bbox": [ + 89, + 416, + 522, + 430 + ], + "score": 1.0, + "content": "Gaussian stochastic neural network (13) and hybrid model (14) are originally proposed in the paper on", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 89, + 426, + 522, + 441 + ], + "spans": [ + { + "bbox": [ + 89, + 426, + 522, + 441 + ], + "score": 1.0, + "content": "Conditional VAE (Sohn et al., 2015). The motivation authors mention in the paper is as follows. During", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 88, + 437, + 522, + 453 + ], + "spans": [ + { + "bbox": [ + 88, + 437, + 222, + 453 + ], + "score": 1.0, + "content": "training the proposal distribution", + "type": "text" + }, + { + "bbox": [ + 222, + 439, + 263, + 451 + ], + "score": 0.93, + "content": "q _ { \\phi } ( z | x , y )", + "type": "inline_equation" + }, + { + "bbox": [ + 264, + 437, + 415, + 453 + ], + "score": 1.0, + "content": "is used to generate the latent variables", + "type": "text" + }, + { + "bbox": [ + 416, + 441, + 422, + 449 + ], + "score": 0.72, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 423, + 437, + 522, + 453 + ], + "score": 1.0, + "content": ", while during the testing", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 89, + 449, + 522, + 464 + ], + "spans": [ + { + "bbox": [ + 89, + 449, + 149, + 464 + ], + "score": 1.0, + "content": "stage the prior", + "type": "text" + }, + { + "bbox": [ + 149, + 450, + 181, + 462 + ], + "score": 0.92, + "content": "p _ { \\psi } ( z | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 182, + 449, + 522, + 464 + ], + "score": 1.0, + "content": "is used. KL divergence tries to close the gap between two distributions but, according", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 89, + 461, + 521, + 473 + ], + "spans": [ + { + "bbox": [ + 89, + 461, + 521, + 473 + ], + "score": 1.0, + "content": "to authors, it is not enough. To overcome the issue authors propose to use a hybrid model (14), a weighted", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 88, + 471, + 521, + 484 + ], + "spans": [ + { + "bbox": [ + 88, + 471, + 521, + 484 + ], + "score": 1.0, + "content": "mixture of variational lower bound (3) and a single-sample Monte-Carlo estimation of log-likelihood (13).", + "type": "text" + } + ], + "index": 27 + }, + { + "bbox": [ + 89, + 482, + 522, + 496 + ], + "spans": [ + { + "bbox": [ + 89, + 482, + 522, + 496 + ], + "score": 1.0, + "content": "The model corresponding to the second term is called Gaussian Stochastic Neural Network (13), because", + "type": "text" + } + ], + "index": 28 + }, + { + "bbox": [ + 89, + 493, + 523, + 506 + ], + "spans": [ + { + "bbox": [ + 89, + 493, + 523, + 506 + ], + "score": 1.0, + "content": "it is a feed-forward neural network with a single Gaussian stochastic layer in the middle. Also GSNN is a", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 90, + 505, + 297, + 518 + ], + "spans": [ + { + "bbox": [ + 90, + 505, + 206, + 518 + ], + "score": 1.0, + "content": "special case of CVAE where", + "type": "text" + }, + { + "bbox": [ + 206, + 505, + 293, + 517 + ], + "score": 0.93, + "content": "q _ { \\phi } ( z | x , y ) = p _ { \\psi } ( z | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 293, + 505, + 297, + 518 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 30 + } + ], + "index": 26, + "bbox_fs": [ + 88, + 416, + 523, + 518 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 132, + 531, + 479, + 563 + ], + "lines": [ + { + "bbox": [ + 132, + 531, + 479, + 563 + ], + "spans": [ + { + "bbox": [ + 132, + 531, + 479, + 563 + ], + "score": 0.92, + "content": "\\begin{array} { r } { L _ { G S N N } ( x , y ; \\theta , \\psi ) = \\mathbb { E } _ { p _ { \\psi } ( z | y ) } \\log p _ { \\theta } ( x | z , y ) \\qquad } \\\\ { L ( x , y ; \\theta , \\psi , \\phi ) = \\alpha L _ { C V A E } ( x , y ; \\theta , \\psi , \\phi ) + ( 1 - \\alpha ) L _ { G S N N } ( x , y ; \\theta , \\psi ) , \\quad \\alpha \\in [ 0 , 1 ] } \\end{array}", + "type": "interline_equation", + "image_path": "6eda0f59e7f92ff41e13fc03a857ebaa2e07dbc1dc3cfeb2358c4de8661097dd.jpg" + } + ] + } + ], + "index": 32, + "virtual_lines": [ + { + "bbox": [ + 132, + 531, + 479, + 541.6666666666666 + ], + "spans": [], + "index": 31 + }, + { + "bbox": [ + 132, + 541.6666666666666, + 479, + 552.3333333333333 + ], + "spans": [], + "index": 32 + }, + { + "bbox": [ + 132, + 552.3333333333333, + 479, + 562.9999999999999 + ], + "spans": [], + "index": 33 + } + ] + }, + { + "type": "text", + "bbox": [ + 90, + 568, + 522, + 591 + ], + "lines": [ + { + "bbox": [ + 90, + 568, + 522, + 581 + ], + "spans": [ + { + "bbox": [ + 90, + 568, + 522, + 581 + ], + "score": 1.0, + "content": "Authors report that hybrid model and GSNN outperform CVAE in terms of segmentation accuracy on the", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 88, + 579, + 174, + 591 + ], + "spans": [ + { + "bbox": [ + 88, + 579, + 174, + 591 + ], + "score": 1.0, + "content": "majority of datasets.", + "type": "text" + } + ], + "index": 35 + } + ], + "index": 34.5, + "bbox_fs": [ + 88, + 568, + 522, + 591 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 596, + 521, + 652 + ], + "lines": [ + { + "bbox": [ + 90, + 596, + 522, + 609 + ], + "spans": [ + { + "bbox": [ + 90, + 596, + 522, + 609 + ], + "score": 1.0, + "content": "We can also add that this technique seems to soften the “holes problem” (Makhzani et al., 2016). In", + "type": "text" + } + ], + "index": 36 + }, + { + "bbox": [ + 89, + 606, + 522, + 620 + ], + "spans": [ + { + "bbox": [ + 89, + 606, + 298, + 620 + ], + "score": 1.0, + "content": "Makhzani et al. (2016) authors observe that vectors", + "type": "text" + }, + { + "bbox": [ + 299, + 611, + 305, + 617 + ], + "score": 0.67, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 305, + 606, + 522, + 620 + ], + "score": 1.0, + "content": "from prior distribution may be different enough from", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 89, + 618, + 522, + 631 + ], + "spans": [ + { + "bbox": [ + 89, + 618, + 134, + 631 + ], + "score": 1.0, + "content": "all vectors", + "type": "text" + }, + { + "bbox": [ + 134, + 620, + 141, + 628 + ], + "score": 0.76, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 141, + 618, + 522, + 631 + ], + "score": 1.0, + "content": "from the proposal distribution at the training stage, so the generator network may be confused", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 89, + 628, + 522, + 643 + ], + "spans": [ + { + "bbox": [ + 89, + 628, + 422, + 643 + ], + "score": 1.0, + "content": "at the testing stage. Due to this problem CVAE can have good reconstructions of", + "type": "text" + }, + { + "bbox": [ + 422, + 631, + 429, + 641 + ], + "score": 0.79, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 628, + 455, + 643 + ], + "score": 1.0, + "content": "given", + "type": "text" + }, + { + "bbox": [ + 455, + 629, + 517, + 642 + ], + "score": 0.93, + "content": "z \\sim q _ { \\phi } ( z | x , y )", + "type": "inline_equation" + }, + { + "bbox": [ + 518, + 628, + 522, + 643 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 39 + }, + { + "bbox": [ + 89, + 640, + 312, + 653 + ], + "spans": [ + { + "bbox": [ + 89, + 640, + 160, + 653 + ], + "score": 1.0, + "content": "while samples of", + "type": "text" + }, + { + "bbox": [ + 160, + 642, + 167, + 652 + ], + "score": 0.8, + "content": "y", + "type": "inline_equation" + }, + { + "bbox": [ + 167, + 640, + 192, + 653 + ], + "score": 1.0, + "content": "given", + "type": "text" + }, + { + "bbox": [ + 192, + 641, + 244, + 652 + ], + "score": 0.93, + "content": "z \\sim p _ { \\psi } ( z | x )", + "type": "inline_equation" + }, + { + "bbox": [ + 244, + 640, + 312, + 653 + ], + "score": 1.0, + "content": "are not realistic.", + "type": "text" + } + ], + "index": 40 + } + ], + "index": 38, + "bbox_fs": [ + 89, + 596, + 522, + 653 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 90, + 95, + 291, + 108 + ], + "lines": [ + { + "bbox": [ + 90, + 95, + 291, + 108 + ], + "spans": [ + { + "bbox": [ + 90, + 95, + 291, + 108 + ], + "score": 1.0, + "content": "The same trick is applicable to our model as well:", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "interline_equation", + "bbox": [ + 131, + 113, + 480, + 144 + ], + "lines": [ + { + "bbox": [ + 131, + 113, + 480, + 144 + ], + "spans": [ + { + "bbox": [ + 131, + 113, + 480, + 144 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { L _ { G S N N } ( x , b ; \\theta , \\psi ) = \\mathbb { E } _ { p _ { \\psi } ( z | x _ { 1 - b } , b ) } \\log p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b ) } \\\\ & { L ( x , b ; \\theta , \\psi , \\phi ) = \\alpha L _ { V A E A C } ( x , b ; \\theta , \\psi , \\phi ) + ( 1 - \\alpha ) L _ { G S N N } ( x , b ; \\theta , \\psi ) , \\quad \\alpha \\in [ 0 , 1 ] } \\end{array}", + "type": "interline_equation", + "image_path": "c33502d1bb26f985e83664b8c446d3988a4e59347df95247f5e7bef837ca1711.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 131, + 113, + 480, + 123.33333333333333 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 131, + 123.33333333333333, + 480, + 133.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 131, + 133.66666666666666, + 480, + 144.0 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 88, + 154, + 524, + 177 + ], + "lines": [ + { + "bbox": [ + 89, + 154, + 522, + 167 + ], + "spans": [ + { + "bbox": [ + 89, + 154, + 292, + 167 + ], + "score": 1.0, + "content": "In order to reflect the difference between sampling", + "type": "text" + }, + { + "bbox": [ + 292, + 156, + 299, + 164 + ], + "score": 0.78, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 154, + 522, + 167 + ], + "score": 1.0, + "content": "from prior and proposal distributions, authors of CVAE", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 89, + 165, + 277, + 179 + ], + "spans": [ + { + "bbox": [ + 89, + 165, + 277, + 179 + ], + "score": 1.0, + "content": "use two methods of log-likelihood estimation:", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "interline_equation", + "bbox": [ + 194, + 183, + 419, + 218 + ], + "lines": [ + { + "bbox": [ + 194, + 183, + 419, + 218 + ], + "spans": [ + { + "bbox": [ + 194, + 183, + 419, + 218 + ], + "score": 0.93, + "content": "\\log p _ { \\theta , \\psi } ( x | y ) \\approx \\log \\frac { 1 } { S } \\sum _ { i = 1 } ^ { S } p _ { \\theta } ( x | z _ { i } , y ) , ~ z _ { i } \\sim p _ { \\psi } ( z | y )", + "type": "interline_equation", + "image_path": "583a13fe422be7a69bc3cb86f05f54ddd22119c286cd1b5e0c02498f08abc599.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 194, + 183, + 419, + 200.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 194, + 200.5, + 419, + 218.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "interline_equation", + "bbox": [ + 171, + 223, + 442, + 258 + ], + "lines": [ + { + "bbox": [ + 171, + 223, + 442, + 258 + ], + "spans": [ + { + "bbox": [ + 171, + 223, + 442, + 258 + ], + "score": 0.94, + "content": "\\log p _ { \\theta , \\psi } ( x | y ) \\approx \\log \\frac { 1 } { S } \\sum _ { i = 1 } ^ { S } \\frac { p _ { \\theta } ( x | z _ { i } , y ) p _ { \\psi } ( z _ { i } | y ) } { q _ { \\phi } ( z _ { i } | x , y ) } , ~ z _ { i } \\sim q _ { \\phi } ( z | x , y )", + "type": "interline_equation", + "image_path": "6ccb2cb6db773a0184ddddcd940622b1174f4ffcf1db0c6a0ee7b8d59e0ea61b.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 171, + 223, + 442, + 234.66666666666666 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 171, + 234.66666666666666, + 442, + 246.33333333333331 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 171, + 246.33333333333331, + 442, + 258.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 90, + 261, + 522, + 317 + ], + "lines": [ + { + "bbox": [ + 90, + 260, + 521, + 273 + ], + "spans": [ + { + "bbox": [ + 90, + 260, + 521, + 273 + ], + "score": 1.0, + "content": "The first estimator is called Monte-Carlo estimator and the second one is called Importance Sampling esti-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 89, + 272, + 522, + 284 + ], + "spans": [ + { + "bbox": [ + 89, + 272, + 522, + 284 + ], + "score": 1.0, + "content": "mator (also known as IWAE). They are asymptotically equivalent, but in practice the Monte-Carlo estimator", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 89, + 284, + 522, + 295 + ], + "spans": [ + { + "bbox": [ + 89, + 284, + 402, + 295 + ], + "score": 1.0, + "content": "requires much more samples to obtain the same accuracy of estimation. Small", + "type": "text" + }, + { + "bbox": [ + 402, + 284, + 410, + 293 + ], + "score": 0.82, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 284, + 522, + 295 + ], + "score": 1.0, + "content": "leads to underestimation of", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 88, + 293, + 522, + 307 + ], + "spans": [ + { + "bbox": [ + 88, + 293, + 522, + 307 + ], + "score": 1.0, + "content": "the log-likelihood for both Monte-Carlo and Importance Sampling (Burda et al., 2015), but for Monte-Carlo", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 89, + 304, + 284, + 319 + ], + "spans": [ + { + "bbox": [ + 89, + 304, + 284, + 319 + ], + "score": 1.0, + "content": "the underestimation is expressed much stronger.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 90, + 321, + 521, + 344 + ], + "lines": [ + { + "bbox": [ + 90, + 321, + 521, + 334 + ], + "spans": [ + { + "bbox": [ + 90, + 321, + 521, + 334 + ], + "score": 1.0, + "content": "We perform an additional study of GSNN and hybrid model and show that they have drawbacks when the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 90, + 333, + 361, + 345 + ], + "spans": [ + { + "bbox": [ + 90, + 333, + 163, + 345 + ], + "score": 1.0, + "content": "target distribution", + "type": "text" + }, + { + "bbox": [ + 164, + 333, + 191, + 345 + ], + "score": 0.92, + "content": "p ( x | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 333, + 361, + 345 + ], + "score": 1.0, + "content": "is has multiple different local maximums.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5 + }, + { + "type": "title", + "bbox": [ + 91, + 358, + 211, + 370 + ], + "lines": [ + { + "bbox": [ + 90, + 357, + 212, + 371 + ], + "spans": [ + { + "bbox": [ + 90, + 357, + 212, + 371 + ], + "score": 1.0, + "content": "C.1 THEORETICAL STUDY", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 89, + 379, + 521, + 403 + ], + "lines": [ + { + "bbox": [ + 88, + 378, + 523, + 393 + ], + "spans": [ + { + "bbox": [ + 88, + 378, + 523, + 393 + ], + "score": 1.0, + "content": "In this section we show why GSNN cannot learn distributions with several different modes and leads to a", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 89, + 390, + 181, + 404 + ], + "spans": [ + { + "bbox": [ + 89, + 390, + 181, + 404 + ], + "score": 1.0, + "content": "blurry image samples.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5 + }, + { + "type": "text", + "bbox": [ + 92, + 407, + 413, + 419 + ], + "lines": [ + { + "bbox": [ + 89, + 406, + 412, + 421 + ], + "spans": [ + { + "bbox": [ + 89, + 406, + 412, + 421 + ], + "score": 1.0, + "content": "For the simplicity of the notation we consider hybrid model for a standard VAE:", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21 + }, + { + "type": "interline_equation", + "bbox": [ + 146, + 425, + 465, + 452 + ], + "lines": [ + { + "bbox": [ + 146, + 425, + 465, + 452 + ], + "spans": [ + { + "bbox": [ + 146, + 425, + 465, + 452 + ], + "score": 0.92, + "content": "L ( x ; \\phi , \\psi , \\theta ) = \\alpha \\mathbb { E } _ { z \\sim q _ { \\phi } ( z \\mid x ) } \\log \\frac { p _ { \\theta } ( x \\mid z ) p _ { \\psi } ( z ) } { q _ { \\phi } ( z \\mid x ) } + ( 1 - \\alpha ) \\mathbb { E } _ { z \\sim p _ { \\psi } ( z ) } \\log p _ { \\theta } ( x \\mid z )", + "type": "interline_equation", + "image_path": "c1559493b96d28a0f47430d3d517775906770a2762939c39f8bd304ea066eecd.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 146, + 425, + 465, + 452 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 90, + 457, + 523, + 491 + ], + "lines": [ + { + "bbox": [ + 91, + 457, + 521, + 470 + ], + "spans": [ + { + "bbox": [ + 91, + 457, + 396, + 470 + ], + "score": 1.0, + "content": "The hybrid model (16) for VAEAC can be obtained from (19) by replacing", + "type": "text" + }, + { + "bbox": [ + 396, + 460, + 403, + 468 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 457, + 425, + 470 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 425, + 459, + 437, + 469 + ], + "score": 0.84, + "content": "x _ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 457, + 521, + 470 + ], + "score": 1.0, + "content": "and conditioning all", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 89, + 468, + 522, + 483 + ], + "spans": [ + { + "bbox": [ + 89, + 468, + 155, + 483 + ], + "score": 1.0, + "content": "distributions on", + "type": "text" + }, + { + "bbox": [ + 155, + 470, + 176, + 480 + ], + "score": 0.89, + "content": "x _ { 1 - b }", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 468, + 194, + 483 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 194, + 469, + 200, + 479 + ], + "score": 0.72, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 468, + 522, + 483 + ], + "score": 1.0, + "content": ". The validity of the further equations and conclusions remains for VAEAC after", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 90, + 478, + 161, + 492 + ], + "spans": [ + { + "bbox": [ + 90, + 478, + 161, + 492 + ], + "score": 1.0, + "content": "this replacement.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24 + }, + { + "type": "text", + "bbox": [ + 90, + 496, + 521, + 564 + ], + "lines": [ + { + "bbox": [ + 90, + 496, + 521, + 507 + ], + "spans": [ + { + "bbox": [ + 90, + 496, + 263, + 507 + ], + "score": 1.0, + "content": "Consider now a categorical latent variable", + "type": "text" + }, + { + "bbox": [ + 263, + 499, + 270, + 506 + ], + "score": 0.76, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 496, + 363, + 507 + ], + "score": 1.0, + "content": "which can take one of", + "type": "text" + }, + { + "bbox": [ + 364, + 497, + 375, + 506 + ], + "score": 0.83, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 496, + 424, + 507 + ], + "score": 1.0, + "content": "values. Let", + "type": "text" + }, + { + "bbox": [ + 425, + 499, + 432, + 506 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 496, + 521, + 507 + ], + "score": 1.0, + "content": "be a random variable", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 86, + 502, + 520, + 532 + ], + "spans": [ + { + "bbox": [ + 86, + 502, + 104, + 532 + ], + "score": 1.0, + "content": "witfor", + "type": "text" + }, + { + "bbox": [ + 177, + 508, + 201, + 519 + ], + "score": 0.92, + "content": "p _ { d } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 502, + 254, + 532 + ], + "score": 1.0, + "content": "to be modele some values", + "type": "text" + }, + { + "bbox": [ + 316, + 502, + 431, + 532 + ], + "score": 1.0, + "content": "following true data distribut. So the true distribution has", + "type": "text" + }, + { + "bbox": [ + 442, + 502, + 450, + 532 + ], + "score": 1.0, + "content": "n: d", + "type": "text" + }, + { + "bbox": [ + 450, + 507, + 520, + 520 + ], + "score": 0.91, + "content": "\\begin{array} { r } { p _ { d } ( x = x _ { i } ) = \\frac { 1 } { K } } \\end{array}", + "type": "inline_equation" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 519, + 442, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 183, + 531 + ], + "score": 0.92, + "content": "i \\in \\{ 1 , 2 , \\ldots , K \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 520, + 315, + 530 + ], + "score": 0.86, + "content": "x _ { 1 } , x _ { 2 } , \\dotsc , x _ { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 519, + 442, + 528 + ], + "score": 0.82, + "content": "K", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 89, + 529, + 521, + 542 + ], + "spans": [ + { + "bbox": [ + 89, + 529, + 275, + 542 + ], + "score": 1.0, + "content": "able modes. Suppose the generator network", + "type": "text" + }, + { + "bbox": [ + 275, + 530, + 298, + 541 + ], + "score": 0.88, + "content": "N N _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 529, + 421, + 542 + ], + "score": 1.0, + "content": "which models mapping from", + "type": "text" + }, + { + "bbox": [ + 421, + 532, + 428, + 540 + ], + "score": 0.77, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 529, + 521, + 542 + ], + "score": 1.0, + "content": "to some vector of pa-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 88, + 540, + 522, + 554 + ], + "spans": [ + { + "bbox": [ + 88, + 540, + 128, + 554 + ], + "score": 1.0, + "content": "rameters", + "type": "text" + }, + { + "bbox": [ + 128, + 540, + 195, + 552 + ], + "score": 0.93, + "content": "\\begin{array} { r l r } { v _ { z } } & { { } = } & { \\bar { N } N _ { \\theta } ( z ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 540, + 522, + 554 + ], + "score": 1.0, + "content": ". Thus, we define generative distribution as some function of these parameters:", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 90, + 550, + 425, + 566 + ], + "spans": [ + { + "bbox": [ + 90, + 551, + 173, + 564 + ], + "score": 0.91, + "content": "p _ { \\theta } ( x | z ) \\ : = \\ : f ( x , v _ { z } )", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 550, + 283, + 566 + ], + "score": 1.0, + "content": ". Therefore, the parameters", + "type": "text" + }, + { + "bbox": [ + 284, + 552, + 290, + 561 + ], + "score": 0.8, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 550, + 361, + 566 + ], + "score": 1.0, + "content": "are just the set of", + "type": "text" + }, + { + "bbox": [ + 362, + 553, + 420, + 563 + ], + "score": 0.89, + "content": "v _ { 1 } , v _ { 2 } , \\dotsc , v _ { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 550, + 425, + 566 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28.5 + }, + { + "type": "text", + "bbox": [ + 90, + 567, + 521, + 609 + ], + "lines": [ + { + "bbox": [ + 87, + 566, + 522, + 582 + ], + "spans": [ + { + "bbox": [ + 87, + 566, + 262, + 582 + ], + "score": 1.0, + "content": "For the simplicity of the model we assume", + "type": "text" + }, + { + "bbox": [ + 262, + 567, + 309, + 581 + ], + "score": 0.93, + "content": "\\begin{array} { r } { p _ { \\psi } ( z ) = \\frac { 1 } { K } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 566, + 395, + 582 + ], + "score": 1.0, + "content": ". Taking into account", + "type": "text" + }, + { + "bbox": [ + 396, + 567, + 443, + 581 + ], + "score": 0.93, + "content": "\\begin{array} { r } { p _ { \\psi } ( z ) = \\frac { 1 } { K } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 443, + 566, + 522, + 582 + ], + "score": 1.0, + "content": ", we obtain optimal", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 91, + 580, + 521, + 600 + ], + "spans": [ + { + "bbox": [ + 91, + 580, + 200, + 600 + ], + "score": 0.94, + "content": "\\begin{array} { r } { q ( z = i | x ) = \\frac { f ( x , v _ { i } ) } { \\sum _ { j = 1 } ^ { K } f ( x , v _ { j } ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 581, + 364, + 596 + ], + "score": 1.0, + "content": "Using (19) and the above formulas for", + "type": "text" + }, + { + "bbox": [ + 364, + 585, + 391, + 595 + ], + "score": 0.3, + "content": "q _ { \\phi } , p _ { \\psi }", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 581, + 410, + 596 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 410, + 585, + 421, + 595 + ], + "score": 0.84, + "content": "p _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 581, + 521, + 596 + ], + "score": 1.0, + "content": "we obtain the following", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 90, + 598, + 182, + 610 + ], + "spans": [ + { + "bbox": [ + 90, + 598, + 182, + 610 + ], + "score": 1.0, + "content": "optimization problem:", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33 + }, + { + "type": "interline_equation", + "bbox": [ + 109, + 615, + 485, + 655 + ], + "lines": [ + { + "bbox": [ + 109, + 615, + 485, + 655 + ], + "spans": [ + { + "bbox": [ + 109, + 615, + 485, + 655 + ], + "score": 0.93, + "content": "\\operatorname* { m a x } _ { v _ { 1 } , v _ { 2 } , \\ldots , v _ { K } } \\frac { 1 } { K } \\sum _ { i = 1 } ^ { K } \\left[ \\alpha \\sum _ { j = 1 } ^ { K } \\frac { f ( x _ { i } , v _ { j } ) } { \\sum _ { k = 1 } ^ { K } f ( x _ { i } , v _ { k } ) } \\log \\frac { f ( x _ { i } , v _ { j } ) \\frac { 1 } { K } } { \\frac { f ( x _ { i } , v _ { j } ) } { \\sum _ { k = 1 } ^ { K } f ( x _ { i } , v _ { k } ) } } + ( 1 - \\alpha ) \\sum _ { j = 1 } ^ { K } \\frac { 1 } { K } \\log f ( x _ { i } , v _ { j } ) \\right]", + "type": "interline_equation", + "image_path": "07b4e4827ba4de27da9506a427292dfd27100f30c89238d1b1a7b66b439dfe73.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 109, + 615, + 485, + 628.3333333333334 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 109, + 628.3333333333334, + 485, + 641.6666666666667 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 109, + 641.6666666666667, + 485, + 655.0000000000001 + ], + "spans": [], + "index": 37 + } + ] + } + ], + "page_idx": 17, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 39, + 277, + 51 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 278, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 278, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 284, + 671, + 295, + 680 + ], + "lines": [ + { + "bbox": [ + 283, + 670, + 297, + 683 + ], + "spans": [ + { + "bbox": [ + 283, + 670, + 297, + 683 + ], + "score": 1.0, + "content": "18", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 90, + 95, + 291, + 108 + ], + "lines": [ + { + "bbox": [ + 90, + 95, + 291, + 108 + ], + "spans": [ + { + "bbox": [ + 90, + 95, + 291, + 108 + ], + "score": 1.0, + "content": "The same trick is applicable to our model as well:", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0, + "bbox_fs": [ + 90, + 95, + 291, + 108 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 131, + 113, + 480, + 144 + ], + "lines": [ + { + "bbox": [ + 131, + 113, + 480, + 144 + ], + "spans": [ + { + "bbox": [ + 131, + 113, + 480, + 144 + ], + "score": 0.93, + "content": "\\begin{array} { r l } & { L _ { G S N N } ( x , b ; \\theta , \\psi ) = \\mathbb { E } _ { p _ { \\psi } ( z | x _ { 1 - b } , b ) } \\log p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b ) } \\\\ & { L ( x , b ; \\theta , \\psi , \\phi ) = \\alpha L _ { V A E A C } ( x , b ; \\theta , \\psi , \\phi ) + ( 1 - \\alpha ) L _ { G S N N } ( x , b ; \\theta , \\psi ) , \\quad \\alpha \\in [ 0 , 1 ] } \\end{array}", + "type": "interline_equation", + "image_path": "c33502d1bb26f985e83664b8c446d3988a4e59347df95247f5e7bef837ca1711.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 131, + 113, + 480, + 123.33333333333333 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 131, + 123.33333333333333, + 480, + 133.66666666666666 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 131, + 133.66666666666666, + 480, + 144.0 + ], + "spans": [], + "index": 3 + } + ] + }, + { + "type": "text", + "bbox": [ + 88, + 154, + 524, + 177 + ], + "lines": [ + { + "bbox": [ + 89, + 154, + 522, + 167 + ], + "spans": [ + { + "bbox": [ + 89, + 154, + 292, + 167 + ], + "score": 1.0, + "content": "In order to reflect the difference between sampling", + "type": "text" + }, + { + "bbox": [ + 292, + 156, + 299, + 164 + ], + "score": 0.78, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 299, + 154, + 522, + 167 + ], + "score": 1.0, + "content": "from prior and proposal distributions, authors of CVAE", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 89, + 165, + 277, + 179 + ], + "spans": [ + { + "bbox": [ + 89, + 165, + 277, + 179 + ], + "score": 1.0, + "content": "use two methods of log-likelihood estimation:", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5, + "bbox_fs": [ + 89, + 154, + 522, + 179 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 194, + 183, + 419, + 218 + ], + "lines": [ + { + "bbox": [ + 194, + 183, + 419, + 218 + ], + "spans": [ + { + "bbox": [ + 194, + 183, + 419, + 218 + ], + "score": 0.93, + "content": "\\log p _ { \\theta , \\psi } ( x | y ) \\approx \\log \\frac { 1 } { S } \\sum _ { i = 1 } ^ { S } p _ { \\theta } ( x | z _ { i } , y ) , ~ z _ { i } \\sim p _ { \\psi } ( z | y )", + "type": "interline_equation", + "image_path": "583a13fe422be7a69bc3cb86f05f54ddd22119c286cd1b5e0c02498f08abc599.jpg" + } + ] + } + ], + "index": 6.5, + "virtual_lines": [ + { + "bbox": [ + 194, + 183, + 419, + 200.5 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 194, + 200.5, + 419, + 218.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "interline_equation", + "bbox": [ + 171, + 223, + 442, + 258 + ], + "lines": [ + { + "bbox": [ + 171, + 223, + 442, + 258 + ], + "spans": [ + { + "bbox": [ + 171, + 223, + 442, + 258 + ], + "score": 0.94, + "content": "\\log p _ { \\theta , \\psi } ( x | y ) \\approx \\log \\frac { 1 } { S } \\sum _ { i = 1 } ^ { S } \\frac { p _ { \\theta } ( x | z _ { i } , y ) p _ { \\psi } ( z _ { i } | y ) } { q _ { \\phi } ( z _ { i } | x , y ) } , ~ z _ { i } \\sim q _ { \\phi } ( z | x , y )", + "type": "interline_equation", + "image_path": "6ccb2cb6db773a0184ddddcd940622b1174f4ffcf1db0c6a0ee7b8d59e0ea61b.jpg" + } + ] + } + ], + "index": 9, + "virtual_lines": [ + { + "bbox": [ + 171, + 223, + 442, + 234.66666666666666 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 171, + 234.66666666666666, + 442, + 246.33333333333331 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 171, + 246.33333333333331, + 442, + 258.0 + ], + "spans": [], + "index": 10 + } + ] + }, + { + "type": "text", + "bbox": [ + 90, + 261, + 522, + 317 + ], + "lines": [ + { + "bbox": [ + 90, + 260, + 521, + 273 + ], + "spans": [ + { + "bbox": [ + 90, + 260, + 521, + 273 + ], + "score": 1.0, + "content": "The first estimator is called Monte-Carlo estimator and the second one is called Importance Sampling esti-", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 89, + 272, + 522, + 284 + ], + "spans": [ + { + "bbox": [ + 89, + 272, + 522, + 284 + ], + "score": 1.0, + "content": "mator (also known as IWAE). They are asymptotically equivalent, but in practice the Monte-Carlo estimator", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 89, + 284, + 522, + 295 + ], + "spans": [ + { + "bbox": [ + 89, + 284, + 402, + 295 + ], + "score": 1.0, + "content": "requires much more samples to obtain the same accuracy of estimation. Small", + "type": "text" + }, + { + "bbox": [ + 402, + 284, + 410, + 293 + ], + "score": 0.82, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 410, + 284, + 522, + 295 + ], + "score": 1.0, + "content": "leads to underestimation of", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 88, + 293, + 522, + 307 + ], + "spans": [ + { + "bbox": [ + 88, + 293, + 522, + 307 + ], + "score": 1.0, + "content": "the log-likelihood for both Monte-Carlo and Importance Sampling (Burda et al., 2015), but for Monte-Carlo", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 89, + 304, + 284, + 319 + ], + "spans": [ + { + "bbox": [ + 89, + 304, + 284, + 319 + ], + "score": 1.0, + "content": "the underestimation is expressed much stronger.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13, + "bbox_fs": [ + 88, + 260, + 522, + 319 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 321, + 521, + 344 + ], + "lines": [ + { + "bbox": [ + 90, + 321, + 521, + 334 + ], + "spans": [ + { + "bbox": [ + 90, + 321, + 521, + 334 + ], + "score": 1.0, + "content": "We perform an additional study of GSNN and hybrid model and show that they have drawbacks when the", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 90, + 333, + 361, + 345 + ], + "spans": [ + { + "bbox": [ + 90, + 333, + 163, + 345 + ], + "score": 1.0, + "content": "target distribution", + "type": "text" + }, + { + "bbox": [ + 164, + 333, + 191, + 345 + ], + "score": 0.92, + "content": "p ( x | y )", + "type": "inline_equation" + }, + { + "bbox": [ + 191, + 333, + 361, + 345 + ], + "score": 1.0, + "content": "is has multiple different local maximums.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 16.5, + "bbox_fs": [ + 90, + 321, + 521, + 345 + ] + }, + { + "type": "title", + "bbox": [ + 91, + 358, + 211, + 370 + ], + "lines": [ + { + "bbox": [ + 90, + 357, + 212, + 371 + ], + "spans": [ + { + "bbox": [ + 90, + 357, + 212, + 371 + ], + "score": 1.0, + "content": "C.1 THEORETICAL STUDY", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 18 + }, + { + "type": "text", + "bbox": [ + 89, + 379, + 521, + 403 + ], + "lines": [ + { + "bbox": [ + 88, + 378, + 523, + 393 + ], + "spans": [ + { + "bbox": [ + 88, + 378, + 523, + 393 + ], + "score": 1.0, + "content": "In this section we show why GSNN cannot learn distributions with several different modes and leads to a", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 89, + 390, + 181, + 404 + ], + "spans": [ + { + "bbox": [ + 89, + 390, + 181, + 404 + ], + "score": 1.0, + "content": "blurry image samples.", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 19.5, + "bbox_fs": [ + 88, + 378, + 523, + 404 + ] + }, + { + "type": "text", + "bbox": [ + 92, + 407, + 413, + 419 + ], + "lines": [ + { + "bbox": [ + 89, + 406, + 412, + 421 + ], + "spans": [ + { + "bbox": [ + 89, + 406, + 412, + 421 + ], + "score": 1.0, + "content": "For the simplicity of the notation we consider hybrid model for a standard VAE:", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 21, + "bbox_fs": [ + 89, + 406, + 412, + 421 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 146, + 425, + 465, + 452 + ], + "lines": [ + { + "bbox": [ + 146, + 425, + 465, + 452 + ], + "spans": [ + { + "bbox": [ + 146, + 425, + 465, + 452 + ], + "score": 0.92, + "content": "L ( x ; \\phi , \\psi , \\theta ) = \\alpha \\mathbb { E } _ { z \\sim q _ { \\phi } ( z \\mid x ) } \\log \\frac { p _ { \\theta } ( x \\mid z ) p _ { \\psi } ( z ) } { q _ { \\phi } ( z \\mid x ) } + ( 1 - \\alpha ) \\mathbb { E } _ { z \\sim p _ { \\psi } ( z ) } \\log p _ { \\theta } ( x \\mid z )", + "type": "interline_equation", + "image_path": "c1559493b96d28a0f47430d3d517775906770a2762939c39f8bd304ea066eecd.jpg" + } + ] + } + ], + "index": 22, + "virtual_lines": [ + { + "bbox": [ + 146, + 425, + 465, + 452 + ], + "spans": [], + "index": 22 + } + ] + }, + { + "type": "text", + "bbox": [ + 90, + 457, + 523, + 491 + ], + "lines": [ + { + "bbox": [ + 91, + 457, + 521, + 470 + ], + "spans": [ + { + "bbox": [ + 91, + 457, + 396, + 470 + ], + "score": 1.0, + "content": "The hybrid model (16) for VAEAC can be obtained from (19) by replacing", + "type": "text" + }, + { + "bbox": [ + 396, + 460, + 403, + 468 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 404, + 457, + 425, + 470 + ], + "score": 1.0, + "content": "with", + "type": "text" + }, + { + "bbox": [ + 425, + 459, + 437, + 469 + ], + "score": 0.84, + "content": "x _ { b }", + "type": "inline_equation" + }, + { + "bbox": [ + 437, + 457, + 521, + 470 + ], + "score": 1.0, + "content": "and conditioning all", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 89, + 468, + 522, + 483 + ], + "spans": [ + { + "bbox": [ + 89, + 468, + 155, + 483 + ], + "score": 1.0, + "content": "distributions on", + "type": "text" + }, + { + "bbox": [ + 155, + 470, + 176, + 480 + ], + "score": 0.89, + "content": "x _ { 1 - b }", + "type": "inline_equation" + }, + { + "bbox": [ + 176, + 468, + 194, + 483 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 194, + 469, + 200, + 479 + ], + "score": 0.72, + "content": "b", + "type": "inline_equation" + }, + { + "bbox": [ + 200, + 468, + 522, + 483 + ], + "score": 1.0, + "content": ". The validity of the further equations and conclusions remains for VAEAC after", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 90, + 478, + 161, + 492 + ], + "spans": [ + { + "bbox": [ + 90, + 478, + 161, + 492 + ], + "score": 1.0, + "content": "this replacement.", + "type": "text" + } + ], + "index": 25 + } + ], + "index": 24, + "bbox_fs": [ + 89, + 457, + 522, + 492 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 496, + 521, + 564 + ], + "lines": [ + { + "bbox": [ + 90, + 496, + 521, + 507 + ], + "spans": [ + { + "bbox": [ + 90, + 496, + 263, + 507 + ], + "score": 1.0, + "content": "Consider now a categorical latent variable", + "type": "text" + }, + { + "bbox": [ + 263, + 499, + 270, + 506 + ], + "score": 0.76, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 270, + 496, + 363, + 507 + ], + "score": 1.0, + "content": "which can take one of", + "type": "text" + }, + { + "bbox": [ + 364, + 497, + 375, + 506 + ], + "score": 0.83, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 375, + 496, + 424, + 507 + ], + "score": 1.0, + "content": "values. Let", + "type": "text" + }, + { + "bbox": [ + 425, + 499, + 432, + 506 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 496, + 521, + 507 + ], + "score": 1.0, + "content": "be a random variable", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 86, + 502, + 520, + 532 + ], + "spans": [ + { + "bbox": [ + 86, + 502, + 104, + 532 + ], + "score": 1.0, + "content": "witfor", + "type": "text" + }, + { + "bbox": [ + 177, + 508, + 201, + 519 + ], + "score": 0.92, + "content": "p _ { d } ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 201, + 502, + 254, + 532 + ], + "score": 1.0, + "content": "to be modele some values", + "type": "text" + }, + { + "bbox": [ + 316, + 502, + 431, + 532 + ], + "score": 1.0, + "content": "following true data distribut. So the true distribution has", + "type": "text" + }, + { + "bbox": [ + 442, + 502, + 450, + 532 + ], + "score": 1.0, + "content": "n: d", + "type": "text" + }, + { + "bbox": [ + 450, + 507, + 520, + 520 + ], + "score": 0.91, + "content": "\\begin{array} { r } { p _ { d } ( x = x _ { i } ) = \\frac { 1 } { K } } \\end{array}", + "type": "inline_equation" + } + ], + "index": 27 + }, + { + "bbox": [ + 105, + 519, + 442, + 531 + ], + "spans": [ + { + "bbox": [ + 105, + 519, + 183, + 531 + ], + "score": 0.92, + "content": "i \\in \\{ 1 , 2 , \\ldots , K \\}", + "type": "inline_equation" + }, + { + "bbox": [ + 254, + 520, + 315, + 530 + ], + "score": 0.86, + "content": "x _ { 1 } , x _ { 2 } , \\dotsc , x _ { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 519, + 442, + 528 + ], + "score": 0.82, + "content": "K", + "type": "inline_equation" + } + ], + "index": 28 + }, + { + "bbox": [ + 89, + 529, + 521, + 542 + ], + "spans": [ + { + "bbox": [ + 89, + 529, + 275, + 542 + ], + "score": 1.0, + "content": "able modes. Suppose the generator network", + "type": "text" + }, + { + "bbox": [ + 275, + 530, + 298, + 541 + ], + "score": 0.88, + "content": "N N _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 298, + 529, + 421, + 542 + ], + "score": 1.0, + "content": "which models mapping from", + "type": "text" + }, + { + "bbox": [ + 421, + 532, + 428, + 540 + ], + "score": 0.77, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 429, + 529, + 521, + 542 + ], + "score": 1.0, + "content": "to some vector of pa-", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 88, + 540, + 522, + 554 + ], + "spans": [ + { + "bbox": [ + 88, + 540, + 128, + 554 + ], + "score": 1.0, + "content": "rameters", + "type": "text" + }, + { + "bbox": [ + 128, + 540, + 195, + 552 + ], + "score": 0.93, + "content": "\\begin{array} { r l r } { v _ { z } } & { { } = } & { \\bar { N } N _ { \\theta } ( z ) } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 196, + 540, + 522, + 554 + ], + "score": 1.0, + "content": ". Thus, we define generative distribution as some function of these parameters:", + "type": "text" + } + ], + "index": 30 + }, + { + "bbox": [ + 90, + 550, + 425, + 566 + ], + "spans": [ + { + "bbox": [ + 90, + 551, + 173, + 564 + ], + "score": 0.91, + "content": "p _ { \\theta } ( x | z ) \\ : = \\ : f ( x , v _ { z } )", + "type": "inline_equation" + }, + { + "bbox": [ + 174, + 550, + 283, + 566 + ], + "score": 1.0, + "content": ". Therefore, the parameters", + "type": "text" + }, + { + "bbox": [ + 284, + 552, + 290, + 561 + ], + "score": 0.8, + "content": "\\theta", + "type": "inline_equation" + }, + { + "bbox": [ + 290, + 550, + 361, + 566 + ], + "score": 1.0, + "content": "are just the set of", + "type": "text" + }, + { + "bbox": [ + 362, + 553, + 420, + 563 + ], + "score": 0.89, + "content": "v _ { 1 } , v _ { 2 } , \\dotsc , v _ { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 420, + 550, + 425, + 566 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 31 + } + ], + "index": 28.5, + "bbox_fs": [ + 86, + 496, + 522, + 566 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 567, + 521, + 609 + ], + "lines": [ + { + "bbox": [ + 87, + 566, + 522, + 582 + ], + "spans": [ + { + "bbox": [ + 87, + 566, + 262, + 582 + ], + "score": 1.0, + "content": "For the simplicity of the model we assume", + "type": "text" + }, + { + "bbox": [ + 262, + 567, + 309, + 581 + ], + "score": 0.93, + "content": "\\begin{array} { r } { p _ { \\psi } ( z ) = \\frac { 1 } { K } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 310, + 566, + 395, + 582 + ], + "score": 1.0, + "content": ". Taking into account", + "type": "text" + }, + { + "bbox": [ + 396, + 567, + 443, + 581 + ], + "score": 0.93, + "content": "\\begin{array} { r } { p _ { \\psi } ( z ) = \\frac { 1 } { K } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 443, + 566, + 522, + 582 + ], + "score": 1.0, + "content": ", we obtain optimal", + "type": "text" + } + ], + "index": 32 + }, + { + "bbox": [ + 91, + 580, + 521, + 600 + ], + "spans": [ + { + "bbox": [ + 91, + 580, + 200, + 600 + ], + "score": 0.94, + "content": "\\begin{array} { r } { q ( z = i | x ) = \\frac { f ( x , v _ { i } ) } { \\sum _ { j = 1 } ^ { K } f ( x , v _ { j } ) } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 204, + 581, + 364, + 596 + ], + "score": 1.0, + "content": "Using (19) and the above formulas for", + "type": "text" + }, + { + "bbox": [ + 364, + 585, + 391, + 595 + ], + "score": 0.3, + "content": "q _ { \\phi } , p _ { \\psi }", + "type": "inline_equation" + }, + { + "bbox": [ + 392, + 581, + 410, + 596 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 410, + 585, + 421, + 595 + ], + "score": 0.84, + "content": "p _ { \\theta }", + "type": "inline_equation" + }, + { + "bbox": [ + 421, + 581, + 521, + 596 + ], + "score": 1.0, + "content": "we obtain the following", + "type": "text" + } + ], + "index": 33 + }, + { + "bbox": [ + 90, + 598, + 182, + 610 + ], + "spans": [ + { + "bbox": [ + 90, + 598, + 182, + 610 + ], + "score": 1.0, + "content": "optimization problem:", + "type": "text" + } + ], + "index": 34 + } + ], + "index": 33, + "bbox_fs": [ + 87, + 566, + 522, + 610 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 109, + 615, + 485, + 655 + ], + "lines": [ + { + "bbox": [ + 109, + 615, + 485, + 655 + ], + "spans": [ + { + "bbox": [ + 109, + 615, + 485, + 655 + ], + "score": 0.93, + "content": "\\operatorname* { m a x } _ { v _ { 1 } , v _ { 2 } , \\ldots , v _ { K } } \\frac { 1 } { K } \\sum _ { i = 1 } ^ { K } \\left[ \\alpha \\sum _ { j = 1 } ^ { K } \\frac { f ( x _ { i } , v _ { j } ) } { \\sum _ { k = 1 } ^ { K } f ( x _ { i } , v _ { k } ) } \\log \\frac { f ( x _ { i } , v _ { j } ) \\frac { 1 } { K } } { \\frac { f ( x _ { i } , v _ { j } ) } { \\sum _ { k = 1 } ^ { K } f ( x _ { i } , v _ { k } ) } } + ( 1 - \\alpha ) \\sum _ { j = 1 } ^ { K } \\frac { 1 } { K } \\log f ( x _ { i } , v _ { j } ) \\right]", + "type": "interline_equation", + "image_path": "07b4e4827ba4de27da9506a427292dfd27100f30c89238d1b1a7b66b439dfe73.jpg" + } + ] + } + ], + "index": 36, + "virtual_lines": [ + { + "bbox": [ + 109, + 615, + 485, + 628.3333333333334 + ], + "spans": [], + "index": 35 + }, + { + "bbox": [ + 109, + 628.3333333333334, + 485, + 641.6666666666667 + ], + "spans": [], + "index": 36 + }, + { + "bbox": [ + 109, + 641.6666666666667, + 485, + 655.0000000000001 + ], + "spans": [], + "index": 37 + } + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "text", + "bbox": [ + 90, + 93, + 521, + 127 + ], + "lines": [ + { + "bbox": [ + 90, + 92, + 521, + 106 + ], + "spans": [ + { + "bbox": [ + 90, + 92, + 446, + 106 + ], + "score": 1.0, + "content": "Table 8: Negative log-likelihood estimation of a hybrid model on the synthetic data. IS-", + "type": "text" + }, + { + "bbox": [ + 446, + 93, + 454, + 104 + ], + "score": 0.69, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 92, + 521, + 106 + ], + "score": 1.0, + "content": "refers to Impor-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 90, + 104, + 521, + 117 + ], + "spans": [ + { + "bbox": [ + 90, + 104, + 277, + 117 + ], + "score": 1.0, + "content": "tance Sampling log-likelihood estimation with", + "type": "text" + }, + { + "bbox": [ + 277, + 105, + 285, + 114 + ], + "score": 0.77, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 104, + 423, + 117 + ], + "score": 1.0, + "content": "samples for each object (18). MC-", + "type": "text" + }, + { + "bbox": [ + 423, + 105, + 431, + 115 + ], + "score": 0.73, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 104, + 521, + 117 + ], + "score": 1.0, + "content": "refers to Monte-Carlo", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 90, + 115, + 339, + 127 + ], + "spans": [ + { + "bbox": [ + 90, + 115, + 213, + 127 + ], + "score": 1.0, + "content": "log-likelihood estimation with", + "type": "text" + }, + { + "bbox": [ + 213, + 116, + 221, + 126 + ], + "score": 0.8, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 115, + 339, + 127 + ], + "score": 1.0, + "content": "samples for each object (17).", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1 + }, + { + "type": "table", + "bbox": [ + 230, + 135, + 381, + 182 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 230, + 135, + 381, + 182 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 230, + 135, + 381, + 182 + ], + "spans": [ + { + "bbox": [ + 230, + 135, + 381, + 182 + ], + "score": 0.96, + "html": "
VAEAC weightIS-10MC-10
a=10.2285
α = 0.990.3511
α = 0.90.621.7
", + "type": "table", + "image_path": "65cc93c875903136313a61e9ac8afdb60ebc425608483cce53104e3064921176.jpg" + } + ] + } + ], + "index": 3.5, + "virtual_lines": [ + { + "bbox": [ + 230, + 135, + 381, + 158.5 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 230, + 158.5, + 381, + 182.0 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 89, + 200, + 407, + 213 + ], + "lines": [ + { + "bbox": [ + 88, + 199, + 408, + 216 + ], + "spans": [ + { + "bbox": [ + 88, + 199, + 408, + 216 + ], + "score": 1.0, + "content": "It is easy to show that (20) is equivalent to the following optimization problem:", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "interline_equation", + "bbox": [ + 162, + 216, + 450, + 256 + ], + "lines": [ + { + "bbox": [ + 162, + 216, + 450, + 256 + ], + "spans": [ + { + "bbox": [ + 162, + 216, + 450, + 256 + ], + "score": 0.93, + "content": "\\operatorname* { m a x } _ { v _ { 1 } , v _ { 2 } , \\ldots , v _ { K } } \\sum _ { i = 1 } ^ { K } \\left[ \\alpha \\log \\frac { \\sum _ { j = 1 } ^ { K } f ( x _ { i } , v _ { j } ) } { K } + ( 1 - \\alpha ) \\sum _ { j = 1 } ^ { K } \\frac { 1 } { K } \\log f ( x _ { i } , v _ { j } ) \\right]", + "type": "interline_equation", + "image_path": "fba50a965f88da607ef696227097c6d0af74dd0f5711b3ed679ed890002f6b06.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 162, + 216, + 450, + 229.33333333333334 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 162, + 229.33333333333334, + 450, + 242.66666666666669 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 162, + 242.66666666666669, + 450, + 256.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "text", + "bbox": [ + 90, + 264, + 521, + 333 + ], + "lines": [ + { + "bbox": [ + 90, + 265, + 521, + 277 + ], + "spans": [ + { + "bbox": [ + 90, + 265, + 212, + 277 + ], + "score": 1.0, + "content": "It is clear from (21) that when", + "type": "text" + }, + { + "bbox": [ + 212, + 266, + 237, + 275 + ], + "score": 0.9, + "content": "\\alpha = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 265, + 521, + 277 + ], + "score": 1.0, + "content": "the log-likelihood of the initial model is optimized. On the other hand,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 85, + 271, + 527, + 302 + ], + "spans": [ + { + "bbox": [ + 85, + 271, + 115, + 302 + ], + "score": 1.0, + "content": "when influe", + "type": "text" + }, + { + "bbox": [ + 115, + 278, + 145, + 288 + ], + "score": 0.9, + "content": "\\alpha = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 271, + 230, + 302 + ], + "score": 1.0, + "content": "the optimal point is generative process, a", + "type": "text" + }, + { + "bbox": [ + 230, + 276, + 453, + 291 + ], + "score": 0.76, + "content": "v _ { 1 } = v _ { 2 } = \\cdots = v _ { K } = \\operatorname { a r g m a x } _ { v } \\sum _ { i = 1 } ^ { K } \\log f ( x _ { i } , v ) .", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 271, + 480, + 302 + ], + "score": 1.0, + "content": ", i. e. mizes", + "type": "text" + }, + { + "bbox": [ + 481, + 280, + 488, + 288 + ], + "score": 0.68, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 488, + 271, + 527, + 302 + ], + "score": 1.0, + "content": "doesn’telihood", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 274, + 291, + 405, + 299 + ], + "spans": [ + { + "bbox": [ + 274, + 291, + 281, + 299 + ], + "score": 0.75, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 292, + 405, + 299 + ], + "score": 0.71, + "content": "v", + "type": "inline_equation" + } + ], + "index": 11 + }, + { + "bbox": [ + 90, + 300, + 522, + 313 + ], + "spans": [ + { + "bbox": [ + 90, + 300, + 228, + 313 + ], + "score": 1.0, + "content": "estimation of the generative model", + "type": "text" + }, + { + "bbox": [ + 228, + 300, + 258, + 312 + ], + "score": 0.93, + "content": "f ( x , v )", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 300, + 350, + 313 + ], + "score": 1.0, + "content": "for the given dataset of", + "type": "text" + }, + { + "bbox": [ + 350, + 302, + 357, + 310 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 300, + 522, + 313 + ], + "score": 1.0, + "content": "’s. For Bernoulli and Gaussian generative", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 89, + 310, + 522, + 325 + ], + "spans": [ + { + "bbox": [ + 89, + 310, + 143, + 325 + ], + "score": 1.0, + "content": "distributions", + "type": "text" + }, + { + "bbox": [ + 143, + 311, + 150, + 322 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 310, + 173, + 325 + ], + "score": 1.0, + "content": "such", + "type": "text" + }, + { + "bbox": [ + 173, + 313, + 180, + 321 + ], + "score": 0.72, + "content": "v", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 310, + 297, + 325 + ], + "score": 1.0, + "content": "is just average of all modes", + "type": "text" + }, + { + "bbox": [ + 297, + 313, + 358, + 322 + ], + "score": 0.9, + "content": "x _ { 1 } , x _ { 2 } , \\dotsc , x _ { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 310, + 522, + 325 + ], + "score": 1.0, + "content": ". That explains why further we observe", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 90, + 322, + 256, + 335 + ], + "spans": [ + { + "bbox": [ + 90, + 322, + 256, + 335 + ], + "score": 1.0, + "content": "blurry images when using GSNN model.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 11.5 + }, + { + "type": "text", + "bbox": [ + 91, + 338, + 521, + 394 + ], + "lines": [ + { + "bbox": [ + 90, + 339, + 521, + 350 + ], + "spans": [ + { + "bbox": [ + 90, + 339, + 471, + 350 + ], + "score": 1.0, + "content": "The same conclusion holds for for continuous latent variables instead of categorical. Given", + "type": "text" + }, + { + "bbox": [ + 472, + 339, + 483, + 349 + ], + "score": 0.8, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 339, + 521, + 350 + ], + "score": 1.0, + "content": "different", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 89, + 349, + 521, + 362 + ], + "spans": [ + { + "bbox": [ + 89, + 349, + 459, + 362 + ], + "score": 1.0, + "content": "modes in true data distribution, VAE uses proposal network to separate prior distribution into", + "type": "text" + }, + { + "bbox": [ + 459, + 350, + 470, + 360 + ], + "score": 0.83, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 349, + 521, + 362 + ], + "score": 1.0, + "content": "components", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 88, + 360, + 521, + 373 + ], + "spans": [ + { + "bbox": [ + 88, + 360, + 521, + 373 + ], + "score": 1.0, + "content": "(i. e. regions in the latent space), so that each region corresponds to one mode. On the other hand, in GSNN", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 90, + 371, + 522, + 385 + ], + "spans": [ + { + "bbox": [ + 90, + 374, + 97, + 381 + ], + "score": 0.73, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 97, + 371, + 457, + 385 + ], + "score": 1.0, + "content": "is sampled independently on the mode which is to be reconstructed from it, so for each", + "type": "text" + }, + { + "bbox": [ + 457, + 374, + 464, + 381 + ], + "score": 0.74, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 371, + 522, + 385 + ], + "score": 1.0, + "content": "the generator", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 89, + 383, + 293, + 395 + ], + "spans": [ + { + "bbox": [ + 89, + 383, + 293, + 395 + ], + "score": 1.0, + "content": "have to produce parameters suitable for all modes.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17 + }, + { + "type": "text", + "bbox": [ + 91, + 399, + 520, + 444 + ], + "lines": [ + { + "bbox": [ + 90, + 399, + 522, + 412 + ], + "spans": [ + { + "bbox": [ + 90, + 399, + 522, + 412 + ], + "score": 1.0, + "content": "From this point of view, there is no difference between VAE and VAEAC. If the true conditional distribution", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 89, + 410, + 522, + 423 + ], + "spans": [ + { + "bbox": [ + 89, + 410, + 522, + 423 + ], + "score": 1.0, + "content": "has several different modes, then VAEAC can fit them all, while GSNN learns their average. If true condi-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 89, + 420, + 522, + 434 + ], + "spans": [ + { + "bbox": [ + 89, + 420, + 522, + 434 + ], + "score": 1.0, + "content": "tional distribution has one mode, GSNN and VAEAC are equal, and GSNN may even learn faster because it", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 89, + 432, + 172, + 446 + ], + "spans": [ + { + "bbox": [ + 89, + 432, + 172, + 446 + ], + "score": 1.0, + "content": "has less parameters.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21.5 + }, + { + "type": "text", + "bbox": [ + 91, + 448, + 520, + 493 + ], + "lines": [ + { + "bbox": [ + 90, + 449, + 522, + 462 + ], + "spans": [ + { + "bbox": [ + 90, + 449, + 368, + 462 + ], + "score": 1.0, + "content": "Hybrid model is a trade-off between VAEAC and GSNN: the closer", + "type": "text" + }, + { + "bbox": [ + 369, + 452, + 376, + 459 + ], + "score": 0.78, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 449, + 522, + 462 + ], + "score": 1.0, + "content": "to zero, the more blurry and closer", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 90, + 460, + 522, + 473 + ], + "spans": [ + { + "bbox": [ + 90, + 460, + 483, + 473 + ], + "score": 1.0, + "content": "to the average is the distribution of the model. The exact dependence of the model distribution on", + "type": "text" + }, + { + "bbox": [ + 484, + 462, + 492, + 471 + ], + "score": 0.74, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 492, + 460, + 522, + 473 + ], + "score": 1.0, + "content": "can be", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 89, + 470, + 520, + 484 + ], + "spans": [ + { + "bbox": [ + 89, + 470, + 520, + 484 + ], + "score": 1.0, + "content": "derived analytically for the simple data distributions or evaluated experimentally. We perform such experi-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 89, + 482, + 244, + 495 + ], + "spans": [ + { + "bbox": [ + 89, + 482, + 244, + 495 + ], + "score": 1.0, + "content": "mental evaluation in the next sections.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5 + }, + { + "type": "title", + "bbox": [ + 91, + 506, + 193, + 518 + ], + "lines": [ + { + "bbox": [ + 90, + 506, + 194, + 520 + ], + "spans": [ + { + "bbox": [ + 90, + 506, + 194, + 520 + ], + "score": 1.0, + "content": "C.2 SYNTHETIC DATA", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "text", + "bbox": [ + 91, + 527, + 521, + 585 + ], + "lines": [ + { + "bbox": [ + 89, + 527, + 521, + 540 + ], + "spans": [ + { + "bbox": [ + 89, + 527, + 521, + 540 + ], + "score": 1.0, + "content": "In this section we show that VAEAC is capable of learning a complex multimodal distribution of synthetic", + "type": "text" + } + ], + "index": 29 + }, + { + "bbox": [ + 89, + 537, + 520, + 552 + ], + "spans": [ + { + "bbox": [ + 89, + 537, + 293, + 552 + ], + "score": 1.0, + "content": "data while GSNN and hybrid model are not. Let", + "type": "text" + }, + { + "bbox": [ + 293, + 538, + 326, + 549 + ], + "score": 0.91, + "content": "x \\in \\mathbb { R } ^ { \\bar { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 537, + 346, + 552 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 347, + 538, + 478, + 551 + ], + "score": 0.89, + "content": "p ( \\bar { b } _ { 1 } = 1 ) = p ( b _ { 2 } = 1 ) = 0 . 5", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 537, + 483, + 552 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 483, + 538, + 520, + 551 + ], + "score": 0.82, + "content": "p _ { d } ( x ) =", + "type": "inline_equation" + } + ], + "index": 30 + }, + { + "bbox": [ + 91, + 544, + 525, + 575 + ], + "spans": [ + { + "bbox": [ + 91, + 550, + 180, + 564 + ], + "score": 0.89, + "content": "\\begin{array} { r l } { \\frac { 1 } { 8 } \\sum _ { i = 1 } ^ { 8 } \\mathcal { N } ( x | \\mu _ { i } , \\frac { 1 } { 1 0 } I ) } & { { } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 544, + 210, + 575 + ], + "score": 1.0, + "content": "where s samp", + "type": "text" + }, + { + "bbox": [ + 210, + 551, + 282, + 563 + ], + "score": 0.91, + "content": "\\mu _ { i } \\sim \\mathcal N ( \\mu _ { i } | 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 544, + 357, + 575 + ], + "score": 1.0, + "content": ". The distribution e use multi-layer p", + "type": "text" + }, + { + "bbox": [ + 357, + 551, + 377, + 564 + ], + "score": 0.92, + "content": "p ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 544, + 525, + 575 + ], + "score": 1.0, + "content": "is plotted in figure 6. The datasetptron with four ReLU layers of size", + "type": "text" + } + ], + "index": 31 + }, + { + "bbox": [ + 244, + 563, + 268, + 574 + ], + "spans": [ + { + "bbox": [ + 244, + 563, + 268, + 574 + ], + "score": 0.91, + "content": "p _ { d } ( x )", + "type": "inline_equation" + } + ], + "index": 32 + }, + { + "bbox": [ + 89, + 573, + 329, + 586 + ], + "spans": [ + { + "bbox": [ + 89, + 573, + 329, + 586 + ], + "score": 1.0, + "content": "400-200-100-50, 25-dimensional Gaussian latent variables.", + "type": "text" + } + ], + "index": 33 + } + ], + "index": 31 + }, + { + "type": "text", + "bbox": [ + 91, + 590, + 521, + 624 + ], + "lines": [ + { + "bbox": [ + 88, + 587, + 522, + 605 + ], + "spans": [ + { + "bbox": [ + 88, + 587, + 233, + 605 + ], + "score": 1.0, + "content": "For different mixture coefficients", + "type": "text" + }, + { + "bbox": [ + 233, + 593, + 241, + 600 + ], + "score": 0.75, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 587, + 466, + 605 + ], + "score": 1.0, + "content": "we visualize samples from the learned distributions", + "type": "text" + }, + { + "bbox": [ + 466, + 590, + 517, + 603 + ], + "score": 0.91, + "content": "p _ { \\psi , \\theta } ( x _ { 1 } , x _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 517, + 587, + 522, + 605 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 90, + 600, + 522, + 615 + ], + "spans": [ + { + "bbox": [ + 90, + 601, + 140, + 614 + ], + "score": 0.92, + "content": "p _ { \\psi , \\theta } ( x _ { 1 } | x _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 600, + 162, + 615 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 162, + 601, + 211, + 613 + ], + "score": 0.92, + "content": "p _ { \\psi , \\theta } ( x _ { 2 } | x _ { 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 600, + 522, + 615 + ], + "score": 1.0, + "content": ". The observed features for the conditional distributions are generated from", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 89, + 611, + 314, + 626 + ], + "spans": [ + { + "bbox": [ + 89, + 611, + 195, + 626 + ], + "score": 1.0, + "content": "the marginal distributions", + "type": "text" + }, + { + "bbox": [ + 195, + 613, + 219, + 624 + ], + "score": 0.91, + "content": "p ( x _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 611, + 237, + 626 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 237, + 613, + 261, + 624 + ], + "score": 0.92, + "content": "p ( x _ { 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 611, + 314, + 626 + ], + "score": 1.0, + "content": "respectively.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35 + }, + { + "type": "text", + "bbox": [ + 90, + 629, + 520, + 652 + ], + "lines": [ + { + "bbox": [ + 89, + 627, + 522, + 643 + ], + "spans": [ + { + "bbox": [ + 89, + 627, + 522, + 643 + ], + "score": 1.0, + "content": "We see in table 8 and in figure 7, that even with very small weight GSNN prevents model from learning", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 90, + 640, + 521, + 653 + ], + "spans": [ + { + "bbox": [ + 90, + 640, + 521, + 653 + ], + "score": 1.0, + "content": "distributions with several local optimas. GSNN also increases Monte-Carlo log-likelihood estimation with", + "type": "text" + } + ], + "index": 38 + } + ], + "index": 37.5 + } + ], + "page_idx": 18, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 39, + 277, + 51 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 284, + 671, + 295, + 680 + ], + "lines": [ + { + "bbox": [ + 283, + 670, + 297, + 684 + ], + "spans": [ + { + "bbox": [ + 283, + 670, + 297, + 684 + ], + "score": 1.0, + "content": "19", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "text", + "bbox": [ + 90, + 93, + 521, + 127 + ], + "lines": [ + { + "bbox": [ + 90, + 92, + 521, + 106 + ], + "spans": [ + { + "bbox": [ + 90, + 92, + 446, + 106 + ], + "score": 1.0, + "content": "Table 8: Negative log-likelihood estimation of a hybrid model on the synthetic data. IS-", + "type": "text" + }, + { + "bbox": [ + 446, + 93, + 454, + 104 + ], + "score": 0.69, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 454, + 92, + 521, + 106 + ], + "score": 1.0, + "content": "refers to Impor-", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 90, + 104, + 521, + 117 + ], + "spans": [ + { + "bbox": [ + 90, + 104, + 277, + 117 + ], + "score": 1.0, + "content": "tance Sampling log-likelihood estimation with", + "type": "text" + }, + { + "bbox": [ + 277, + 105, + 285, + 114 + ], + "score": 0.77, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 286, + 104, + 423, + 117 + ], + "score": 1.0, + "content": "samples for each object (18). MC-", + "type": "text" + }, + { + "bbox": [ + 423, + 105, + 431, + 115 + ], + "score": 0.73, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 432, + 104, + 521, + 117 + ], + "score": 1.0, + "content": "refers to Monte-Carlo", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 90, + 115, + 339, + 127 + ], + "spans": [ + { + "bbox": [ + 90, + 115, + 213, + 127 + ], + "score": 1.0, + "content": "log-likelihood estimation with", + "type": "text" + }, + { + "bbox": [ + 213, + 116, + 221, + 126 + ], + "score": 0.8, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 222, + 115, + 339, + 127 + ], + "score": 1.0, + "content": "samples for each object (17).", + "type": "text" + } + ], + "index": 2 + } + ], + "index": 1, + "bbox_fs": [ + 90, + 92, + 521, + 127 + ] + }, + { + "type": "table", + "bbox": [ + 230, + 135, + 381, + 182 + ], + "blocks": [ + { + "type": "table_body", + "bbox": [ + 230, + 135, + 381, + 182 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 230, + 135, + 381, + 182 + ], + "spans": [ + { + "bbox": [ + 230, + 135, + 381, + 182 + ], + "score": 0.96, + "html": "
VAEAC weightIS-10MC-10
a=10.2285
α = 0.990.3511
α = 0.90.621.7
", + "type": "table", + "image_path": "65cc93c875903136313a61e9ac8afdb60ebc425608483cce53104e3064921176.jpg" + } + ] + } + ], + "index": 3.5, + "virtual_lines": [ + { + "bbox": [ + 230, + 135, + 381, + 158.5 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 230, + 158.5, + 381, + 182.0 + ], + "spans": [], + "index": 4 + } + ] + } + ], + "index": 3.5 + }, + { + "type": "text", + "bbox": [ + 89, + 200, + 407, + 213 + ], + "lines": [ + { + "bbox": [ + 88, + 199, + 408, + 216 + ], + "spans": [ + { + "bbox": [ + 88, + 199, + 408, + 216 + ], + "score": 1.0, + "content": "It is easy to show that (20) is equivalent to the following optimization problem:", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5, + "bbox_fs": [ + 88, + 199, + 408, + 216 + ] + }, + { + "type": "interline_equation", + "bbox": [ + 162, + 216, + 450, + 256 + ], + "lines": [ + { + "bbox": [ + 162, + 216, + 450, + 256 + ], + "spans": [ + { + "bbox": [ + 162, + 216, + 450, + 256 + ], + "score": 0.93, + "content": "\\operatorname* { m a x } _ { v _ { 1 } , v _ { 2 } , \\ldots , v _ { K } } \\sum _ { i = 1 } ^ { K } \\left[ \\alpha \\log \\frac { \\sum _ { j = 1 } ^ { K } f ( x _ { i } , v _ { j } ) } { K } + ( 1 - \\alpha ) \\sum _ { j = 1 } ^ { K } \\frac { 1 } { K } \\log f ( x _ { i } , v _ { j } ) \\right]", + "type": "interline_equation", + "image_path": "fba50a965f88da607ef696227097c6d0af74dd0f5711b3ed679ed890002f6b06.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 162, + 216, + 450, + 229.33333333333334 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 162, + 229.33333333333334, + 450, + 242.66666666666669 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 162, + 242.66666666666669, + 450, + 256.0 + ], + "spans": [], + "index": 8 + } + ] + }, + { + "type": "list", + "bbox": [ + 90, + 264, + 521, + 333 + ], + "lines": [ + { + "bbox": [ + 90, + 265, + 521, + 277 + ], + "spans": [ + { + "bbox": [ + 90, + 265, + 212, + 277 + ], + "score": 1.0, + "content": "It is clear from (21) that when", + "type": "text" + }, + { + "bbox": [ + 212, + 266, + 237, + 275 + ], + "score": 0.9, + "content": "\\alpha = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 238, + 265, + 521, + 277 + ], + "score": 1.0, + "content": "the log-likelihood of the initial model is optimized. On the other hand,", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 85, + 271, + 527, + 302 + ], + "spans": [ + { + "bbox": [ + 85, + 271, + 115, + 302 + ], + "score": 1.0, + "content": "when influe", + "type": "text" + }, + { + "bbox": [ + 115, + 278, + 145, + 288 + ], + "score": 0.9, + "content": "\\alpha = 0", + "type": "inline_equation" + }, + { + "bbox": [ + 145, + 271, + 230, + 302 + ], + "score": 1.0, + "content": "the optimal point is generative process, a", + "type": "text" + }, + { + "bbox": [ + 230, + 276, + 453, + 291 + ], + "score": 0.76, + "content": "v _ { 1 } = v _ { 2 } = \\cdots = v _ { K } = \\operatorname { a r g m a x } _ { v } \\sum _ { i = 1 } ^ { K } \\log f ( x _ { i } , v ) .", + "type": "inline_equation" + }, + { + "bbox": [ + 453, + 271, + 480, + 302 + ], + "score": 1.0, + "content": ", i. e. mizes", + "type": "text" + }, + { + "bbox": [ + 481, + 280, + 488, + 288 + ], + "score": 0.68, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 488, + 271, + 527, + 302 + ], + "score": 1.0, + "content": "doesn’telihood", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 274, + 291, + 405, + 299 + ], + "spans": [ + { + "bbox": [ + 274, + 291, + 281, + 299 + ], + "score": 0.75, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 398, + 292, + 405, + 299 + ], + "score": 0.71, + "content": "v", + "type": "inline_equation" + } + ], + "index": 11, + "is_list_end_line": true + }, + { + "bbox": [ + 90, + 300, + 522, + 313 + ], + "spans": [ + { + "bbox": [ + 90, + 300, + 228, + 313 + ], + "score": 1.0, + "content": "estimation of the generative model", + "type": "text" + }, + { + "bbox": [ + 228, + 300, + 258, + 312 + ], + "score": 0.93, + "content": "f ( x , v )", + "type": "inline_equation" + }, + { + "bbox": [ + 258, + 300, + 350, + 313 + ], + "score": 1.0, + "content": "for the given dataset of", + "type": "text" + }, + { + "bbox": [ + 350, + 302, + 357, + 310 + ], + "score": 0.76, + "content": "x", + "type": "inline_equation" + }, + { + "bbox": [ + 357, + 300, + 522, + 313 + ], + "score": 1.0, + "content": "’s. For Bernoulli and Gaussian generative", + "type": "text" + } + ], + "index": 12, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 310, + 522, + 325 + ], + "spans": [ + { + "bbox": [ + 89, + 310, + 143, + 325 + ], + "score": 1.0, + "content": "distributions", + "type": "text" + }, + { + "bbox": [ + 143, + 311, + 150, + 322 + ], + "score": 0.85, + "content": "f", + "type": "inline_equation" + }, + { + "bbox": [ + 151, + 310, + 173, + 325 + ], + "score": 1.0, + "content": "such", + "type": "text" + }, + { + "bbox": [ + 173, + 313, + 180, + 321 + ], + "score": 0.72, + "content": "v", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 310, + 297, + 325 + ], + "score": 1.0, + "content": "is just average of all modes", + "type": "text" + }, + { + "bbox": [ + 297, + 313, + 358, + 322 + ], + "score": 0.9, + "content": "x _ { 1 } , x _ { 2 } , \\dotsc , x _ { K }", + "type": "inline_equation" + }, + { + "bbox": [ + 358, + 310, + 522, + 325 + ], + "score": 1.0, + "content": ". That explains why further we observe", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 90, + 322, + 256, + 335 + ], + "spans": [ + { + "bbox": [ + 90, + 322, + 256, + 335 + ], + "score": 1.0, + "content": "blurry images when using GSNN model.", + "type": "text" + } + ], + "index": 14, + "is_list_end_line": true + } + ], + "index": 11.5, + "bbox_fs": [ + 85, + 265, + 527, + 335 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 338, + 521, + 394 + ], + "lines": [ + { + "bbox": [ + 90, + 339, + 521, + 350 + ], + "spans": [ + { + "bbox": [ + 90, + 339, + 471, + 350 + ], + "score": 1.0, + "content": "The same conclusion holds for for continuous latent variables instead of categorical. Given", + "type": "text" + }, + { + "bbox": [ + 472, + 339, + 483, + 349 + ], + "score": 0.8, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 483, + 339, + 521, + 350 + ], + "score": 1.0, + "content": "different", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 89, + 349, + 521, + 362 + ], + "spans": [ + { + "bbox": [ + 89, + 349, + 459, + 362 + ], + "score": 1.0, + "content": "modes in true data distribution, VAE uses proposal network to separate prior distribution into", + "type": "text" + }, + { + "bbox": [ + 459, + 350, + 470, + 360 + ], + "score": 0.83, + "content": "K", + "type": "inline_equation" + }, + { + "bbox": [ + 470, + 349, + 521, + 362 + ], + "score": 1.0, + "content": "components", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 88, + 360, + 521, + 373 + ], + "spans": [ + { + "bbox": [ + 88, + 360, + 521, + 373 + ], + "score": 1.0, + "content": "(i. e. regions in the latent space), so that each region corresponds to one mode. On the other hand, in GSNN", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 90, + 371, + 522, + 385 + ], + "spans": [ + { + "bbox": [ + 90, + 374, + 97, + 381 + ], + "score": 0.73, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 97, + 371, + 457, + 385 + ], + "score": 1.0, + "content": "is sampled independently on the mode which is to be reconstructed from it, so for each", + "type": "text" + }, + { + "bbox": [ + 457, + 374, + 464, + 381 + ], + "score": 0.74, + "content": "z", + "type": "inline_equation" + }, + { + "bbox": [ + 465, + 371, + 522, + 385 + ], + "score": 1.0, + "content": "the generator", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 89, + 383, + 293, + 395 + ], + "spans": [ + { + "bbox": [ + 89, + 383, + 293, + 395 + ], + "score": 1.0, + "content": "have to produce parameters suitable for all modes.", + "type": "text" + } + ], + "index": 19 + } + ], + "index": 17, + "bbox_fs": [ + 88, + 339, + 522, + 395 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 399, + 520, + 444 + ], + "lines": [ + { + "bbox": [ + 90, + 399, + 522, + 412 + ], + "spans": [ + { + "bbox": [ + 90, + 399, + 522, + 412 + ], + "score": 1.0, + "content": "From this point of view, there is no difference between VAE and VAEAC. If the true conditional distribution", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 89, + 410, + 522, + 423 + ], + "spans": [ + { + "bbox": [ + 89, + 410, + 522, + 423 + ], + "score": 1.0, + "content": "has several different modes, then VAEAC can fit them all, while GSNN learns their average. If true condi-", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 89, + 420, + 522, + 434 + ], + "spans": [ + { + "bbox": [ + 89, + 420, + 522, + 434 + ], + "score": 1.0, + "content": "tional distribution has one mode, GSNN and VAEAC are equal, and GSNN may even learn faster because it", + "type": "text" + } + ], + "index": 22 + }, + { + "bbox": [ + 89, + 432, + 172, + 446 + ], + "spans": [ + { + "bbox": [ + 89, + 432, + 172, + 446 + ], + "score": 1.0, + "content": "has less parameters.", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 21.5, + "bbox_fs": [ + 89, + 399, + 522, + 446 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 448, + 520, + 493 + ], + "lines": [ + { + "bbox": [ + 90, + 449, + 522, + 462 + ], + "spans": [ + { + "bbox": [ + 90, + 449, + 368, + 462 + ], + "score": 1.0, + "content": "Hybrid model is a trade-off between VAEAC and GSNN: the closer", + "type": "text" + }, + { + "bbox": [ + 369, + 452, + 376, + 459 + ], + "score": 0.78, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 377, + 449, + 522, + 462 + ], + "score": 1.0, + "content": "to zero, the more blurry and closer", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 90, + 460, + 522, + 473 + ], + "spans": [ + { + "bbox": [ + 90, + 460, + 483, + 473 + ], + "score": 1.0, + "content": "to the average is the distribution of the model. The exact dependence of the model distribution on", + "type": "text" + }, + { + "bbox": [ + 484, + 462, + 492, + 471 + ], + "score": 0.74, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 492, + 460, + 522, + 473 + ], + "score": 1.0, + "content": "can be", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 89, + 470, + 520, + 484 + ], + "spans": [ + { + "bbox": [ + 89, + 470, + 520, + 484 + ], + "score": 1.0, + "content": "derived analytically for the simple data distributions or evaluated experimentally. We perform such experi-", + "type": "text" + } + ], + "index": 26 + }, + { + "bbox": [ + 89, + 482, + 244, + 495 + ], + "spans": [ + { + "bbox": [ + 89, + 482, + 244, + 495 + ], + "score": 1.0, + "content": "mental evaluation in the next sections.", + "type": "text" + } + ], + "index": 27 + } + ], + "index": 25.5, + "bbox_fs": [ + 89, + 449, + 522, + 495 + ] + }, + { + "type": "title", + "bbox": [ + 91, + 506, + 193, + 518 + ], + "lines": [ + { + "bbox": [ + 90, + 506, + 194, + 520 + ], + "spans": [ + { + "bbox": [ + 90, + 506, + 194, + 520 + ], + "score": 1.0, + "content": "C.2 SYNTHETIC DATA", + "type": "text" + } + ], + "index": 28 + } + ], + "index": 28 + }, + { + "type": "list", + "bbox": [ + 91, + 527, + 521, + 585 + ], + "lines": [ + { + "bbox": [ + 89, + 527, + 521, + 540 + ], + "spans": [ + { + "bbox": [ + 89, + 527, + 521, + 540 + ], + "score": 1.0, + "content": "In this section we show that VAEAC is capable of learning a complex multimodal distribution of synthetic", + "type": "text" + } + ], + "index": 29, + "is_list_start_line": true + }, + { + "bbox": [ + 89, + 537, + 520, + 552 + ], + "spans": [ + { + "bbox": [ + 89, + 537, + 293, + 552 + ], + "score": 1.0, + "content": "data while GSNN and hybrid model are not. Let", + "type": "text" + }, + { + "bbox": [ + 293, + 538, + 326, + 549 + ], + "score": 0.91, + "content": "x \\in \\mathbb { R } ^ { \\bar { 2 } }", + "type": "inline_equation" + }, + { + "bbox": [ + 327, + 537, + 346, + 552 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 347, + 538, + 478, + 551 + ], + "score": 0.89, + "content": "p ( \\bar { b } _ { 1 } = 1 ) = p ( b _ { 2 } = 1 ) = 0 . 5", + "type": "inline_equation" + }, + { + "bbox": [ + 478, + 537, + 483, + 552 + ], + "score": 1.0, + "content": ".", + "type": "text" + }, + { + "bbox": [ + 483, + 538, + 520, + 551 + ], + "score": 0.82, + "content": "p _ { d } ( x ) =", + "type": "inline_equation" + } + ], + "index": 30, + "is_list_start_line": true + }, + { + "bbox": [ + 91, + 544, + 525, + 575 + ], + "spans": [ + { + "bbox": [ + 91, + 550, + 180, + 564 + ], + "score": 0.89, + "content": "\\begin{array} { r l } { \\frac { 1 } { 8 } \\sum _ { i = 1 } ^ { 8 } \\mathcal { N } ( x | \\mu _ { i } , \\frac { 1 } { 1 0 } I ) } & { { } } \\end{array}", + "type": "inline_equation" + }, + { + "bbox": [ + 181, + 544, + 210, + 575 + ], + "score": 1.0, + "content": "where s samp", + "type": "text" + }, + { + "bbox": [ + 210, + 551, + 282, + 563 + ], + "score": 0.91, + "content": "\\mu _ { i } \\sim \\mathcal N ( \\mu _ { i } | 0 , I )", + "type": "inline_equation" + }, + { + "bbox": [ + 282, + 544, + 357, + 575 + ], + "score": 1.0, + "content": ". The distribution e use multi-layer p", + "type": "text" + }, + { + "bbox": [ + 357, + 551, + 377, + 564 + ], + "score": 0.92, + "content": "p ( x )", + "type": "inline_equation" + }, + { + "bbox": [ + 378, + 544, + 525, + 575 + ], + "score": 1.0, + "content": "is plotted in figure 6. The datasetptron with four ReLU layers of size", + "type": "text" + } + ], + "index": 31, + "is_list_start_line": true + }, + { + "bbox": [ + 244, + 563, + 268, + 574 + ], + "spans": [ + { + "bbox": [ + 244, + 563, + 268, + 574 + ], + "score": 0.91, + "content": "p _ { d } ( x )", + "type": "inline_equation" + } + ], + "index": 32, + "is_list_end_line": true + }, + { + "bbox": [ + 89, + 573, + 329, + 586 + ], + "spans": [ + { + "bbox": [ + 89, + 573, + 329, + 586 + ], + "score": 1.0, + "content": "400-200-100-50, 25-dimensional Gaussian latent variables.", + "type": "text" + } + ], + "index": 33, + "is_list_start_line": true, + "is_list_end_line": true + } + ], + "index": 31, + "bbox_fs": [ + 89, + 527, + 525, + 586 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 590, + 521, + 624 + ], + "lines": [ + { + "bbox": [ + 88, + 587, + 522, + 605 + ], + "spans": [ + { + "bbox": [ + 88, + 587, + 233, + 605 + ], + "score": 1.0, + "content": "For different mixture coefficients", + "type": "text" + }, + { + "bbox": [ + 233, + 593, + 241, + 600 + ], + "score": 0.75, + "content": "\\alpha", + "type": "inline_equation" + }, + { + "bbox": [ + 241, + 587, + 466, + 605 + ], + "score": 1.0, + "content": "we visualize samples from the learned distributions", + "type": "text" + }, + { + "bbox": [ + 466, + 590, + 517, + 603 + ], + "score": 0.91, + "content": "p _ { \\psi , \\theta } ( x _ { 1 } , x _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 517, + 587, + 522, + 605 + ], + "score": 1.0, + "content": ",", + "type": "text" + } + ], + "index": 34 + }, + { + "bbox": [ + 90, + 600, + 522, + 615 + ], + "spans": [ + { + "bbox": [ + 90, + 601, + 140, + 614 + ], + "score": 0.92, + "content": "p _ { \\psi , \\theta } ( x _ { 1 } | x _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 140, + 600, + 162, + 615 + ], + "score": 1.0, + "content": ", and", + "type": "text" + }, + { + "bbox": [ + 162, + 601, + 211, + 613 + ], + "score": 0.92, + "content": "p _ { \\psi , \\theta } ( x _ { 2 } | x _ { 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 212, + 600, + 522, + 615 + ], + "score": 1.0, + "content": ". The observed features for the conditional distributions are generated from", + "type": "text" + } + ], + "index": 35 + }, + { + "bbox": [ + 89, + 611, + 314, + 626 + ], + "spans": [ + { + "bbox": [ + 89, + 611, + 195, + 626 + ], + "score": 1.0, + "content": "the marginal distributions", + "type": "text" + }, + { + "bbox": [ + 195, + 613, + 219, + 624 + ], + "score": 0.91, + "content": "p ( x _ { 2 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 219, + 611, + 237, + 626 + ], + "score": 1.0, + "content": "and", + "type": "text" + }, + { + "bbox": [ + 237, + 613, + 261, + 624 + ], + "score": 0.92, + "content": "p ( x _ { 1 } )", + "type": "inline_equation" + }, + { + "bbox": [ + 261, + 611, + 314, + 626 + ], + "score": 1.0, + "content": "respectively.", + "type": "text" + } + ], + "index": 36 + } + ], + "index": 35, + "bbox_fs": [ + 88, + 587, + 522, + 626 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 629, + 520, + 652 + ], + "lines": [ + { + "bbox": [ + 89, + 627, + 522, + 643 + ], + "spans": [ + { + "bbox": [ + 89, + 627, + 522, + 643 + ], + "score": 1.0, + "content": "We see in table 8 and in figure 7, that even with very small weight GSNN prevents model from learning", + "type": "text" + } + ], + "index": 37 + }, + { + "bbox": [ + 90, + 640, + 521, + 653 + ], + "spans": [ + { + "bbox": [ + 90, + 640, + 521, + 653 + ], + "score": 1.0, + "content": "distributions with several local optimas. GSNN also increases Monte-Carlo log-likelihood estimation with", + "type": "text" + } + ], + "index": 38 + }, + { + "bbox": [ + 89, + 601, + 522, + 614 + ], + "spans": [ + { + "bbox": [ + 89, + 601, + 522, + 614 + ], + "score": 1.0, + "content": "a few samples and decreases much more precise Importance Sampling log-likelihood estimation. When", + "type": "text", + "cross_page": true + } + ], + "index": 23 + }, + { + "bbox": [ + 90, + 613, + 281, + 624 + ], + "spans": [ + { + "bbox": [ + 90, + 613, + 124, + 623 + ], + "score": 0.88, + "content": "\\alpha = 0 . 9", + "type": "inline_equation", + "cross_page": true + }, + { + "bbox": [ + 124, + 613, + 281, + 624 + ], + "score": 1.0, + "content": "the whole distribution structure is lost.", + "type": "text", + "cross_page": true + } + ], + "index": 24 + } + ], + "index": 37.5, + "bbox_fs": [ + 89, + 627, + 522, + 653 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 133, + 147, + 213, + 212 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 133, + 147, + 213, + 212 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 133, + 147, + 213, + 212 + ], + "spans": [ + { + "bbox": [ + 133, + 147, + 213, + 212 + ], + "score": 0.479, + "type": "image", + "image_path": "7407e71d027399c997ec577ddb914586b76471425f58bff6404a5cd035d567fc.jpg" + } + ] + } + ], + "index": 0.5, + "virtual_lines": [ + { + "bbox": [ + 133, + 147, + 213, + 179.5 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 133, + 179.5, + 213, + 212.0 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 90, + 222, + 256, + 245 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 90, + 221, + 256, + 235 + ], + "spans": [ + { + "bbox": [ + 90, + 221, + 256, + 235 + ], + "score": 1.0, + "content": "Figure 6: Probability density function of", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 90, + 233, + 199, + 245 + ], + "spans": [ + { + "bbox": [ + 90, + 233, + 199, + 245 + ], + "score": 1.0, + "content": "synthetic data distribution.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + } + ], + "index": 1.5 + }, + { + "type": "image", + "bbox": [ + 330, + 95, + 513, + 275 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 330, + 95, + 513, + 275 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 330, + 95, + 513, + 275 + ], + "spans": [ + { + "bbox": [ + 330, + 95, + 513, + 275 + ], + "score": 0.94, + "type": "image", + "image_path": "a4bd4e815fbae18fb3e5afb3429dcbb43b07bb985c3a0166b18e1666e16f7080.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 330, + 95, + 513, + 108.84615384615384 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 330, + 108.84615384615384, + 513, + 122.69230769230768 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 330, + 122.69230769230768, + 513, + 136.53846153846152 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 330, + 136.53846153846152, + 513, + 150.38461538461536 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 330, + 150.38461538461536, + 513, + 164.2307692307692 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 330, + 164.2307692307692, + 513, + 178.07692307692304 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 330, + 178.07692307692304, + 513, + 191.92307692307688 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 330, + 191.92307692307688, + 513, + 205.76923076923072 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 330, + 205.76923076923072, + 513, + 219.61538461538456 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 330, + 219.61538461538456, + 513, + 233.4615384615384 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 330, + 233.4615384615384, + 513, + 247.30769230769224 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 330, + 247.30769230769224, + 513, + 261.1538461538461 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 330, + 261.1538461538461, + 513, + 274.99999999999994 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 348, + 285, + 495, + 297 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 346, + 283, + 495, + 299 + ], + "spans": [ + { + "bbox": [ + 346, + 283, + 495, + 299 + ], + "score": 1.0, + "content": "Figure 7: VAEAC for synthetic data.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + } + ], + "index": 13.5 + }, + { + "type": "image", + "bbox": [ + 106, + 340, + 511, + 522 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 340, + 511, + 522 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 340, + 511, + 522 + ], + "spans": [ + { + "bbox": [ + 106, + 340, + 511, + 522 + ], + "score": 0.954, + "type": "image", + "image_path": "54463305da3093916463d301746c0407e7d8d2daa9a2e48291c407dc4d828109.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 106, + 340, + 511, + 400.6666666666667 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 106, + 400.6666666666667, + 511, + 461.33333333333337 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 106, + 461.33333333333337, + 511, + 522.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 97, + 537, + 506, + 560 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 244, + 536, + 366, + 549 + ], + "spans": [ + { + "bbox": [ + 244, + 536, + 366, + 549 + ], + "score": 1.0, + "content": "Figure 8: MNIST inpaintings.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 546, + 504, + 561 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 504, + 561 + ], + "score": 1.0, + "content": "Left: input. The gray pixels are unobserved. Middle: samples from the model. Right: ground truth.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + } + ], + "index": 20.25 + }, + { + "type": "text", + "bbox": [ + 91, + 601, + 520, + 624 + ], + "lines": [ + { + "bbox": [ + 89, + 601, + 522, + 614 + ], + "spans": [ + { + "bbox": [ + 89, + 601, + 522, + 614 + ], + "score": 1.0, + "content": "a few samples and decreases much more precise Importance Sampling log-likelihood estimation. When", + "type": "text" + } + ], + "index": 23 + }, + { + "bbox": [ + 90, + 613, + 281, + 624 + ], + "spans": [ + { + "bbox": [ + 90, + 613, + 124, + 623 + ], + "score": 0.88, + "content": "\\alpha = 0 . 9", + "type": "inline_equation" + }, + { + "bbox": [ + 124, + 613, + 281, + 624 + ], + "score": 1.0, + "content": "the whole distribution structure is lost.", + "type": "text" + } + ], + "index": 24 + } + ], + "index": 23.5 + }, + { + "type": "text", + "bbox": [ + 90, + 629, + 518, + 652 + ], + "lines": [ + { + "bbox": [ + 89, + 628, + 521, + 642 + ], + "spans": [ + { + "bbox": [ + 89, + 628, + 164, + 642 + ], + "score": 1.0, + "content": "We see that using", + "type": "text" + }, + { + "bbox": [ + 165, + 630, + 192, + 641 + ], + "score": 0.9, + "content": "\\alpha \\neq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 628, + 521, + 642 + ], + "score": 1.0, + "content": "ruins multimodality of the restored distribution, so we highly recommend to use", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 90, + 640, + 189, + 651 + ], + "spans": [ + { + "bbox": [ + 90, + 641, + 116, + 650 + ], + "score": 0.89, + "content": "\\alpha = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 640, + 158, + 651 + ], + "score": 1.0, + "content": "or at least", + "type": "text" + }, + { + "bbox": [ + 159, + 641, + 185, + 650 + ], + "score": 0.88, + "content": "\\alpha \\approx 1", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 640, + 189, + 651 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5 + } + ], + "page_idx": 19, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 90, + 40, + 277, + 50 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 278, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 278, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 284, + 671, + 295, + 680 + ], + "lines": [ + { + "bbox": [ + 281, + 669, + 297, + 684 + ], + "spans": [ + { + "bbox": [ + 281, + 669, + 297, + 684 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 16 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 133, + 147, + 213, + 212 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 133, + 147, + 213, + 212 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 133, + 147, + 213, + 212 + ], + "spans": [ + { + "bbox": [ + 133, + 147, + 213, + 212 + ], + "score": 0.479, + "type": "image", + "image_path": "7407e71d027399c997ec577ddb914586b76471425f58bff6404a5cd035d567fc.jpg" + } + ] + } + ], + "index": 0.5, + "virtual_lines": [ + { + "bbox": [ + 133, + 147, + 213, + 179.5 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 133, + 179.5, + 213, + 212.0 + ], + "spans": [], + "index": 1 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 90, + 222, + 256, + 245 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 90, + 221, + 256, + 235 + ], + "spans": [ + { + "bbox": [ + 90, + 221, + 256, + 235 + ], + "score": 1.0, + "content": "Figure 6: Probability density function of", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 90, + 233, + 199, + 245 + ], + "spans": [ + { + "bbox": [ + 90, + 233, + 199, + 245 + ], + "score": 1.0, + "content": "synthetic data distribution.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 2.5 + } + ], + "index": 1.5 + }, + { + "type": "image", + "bbox": [ + 330, + 95, + 513, + 275 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 330, + 95, + 513, + 275 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 330, + 95, + 513, + 275 + ], + "spans": [ + { + "bbox": [ + 330, + 95, + 513, + 275 + ], + "score": 0.94, + "type": "image", + "image_path": "a4bd4e815fbae18fb3e5afb3429dcbb43b07bb985c3a0166b18e1666e16f7080.jpg" + } + ] + } + ], + "index": 10, + "virtual_lines": [ + { + "bbox": [ + 330, + 95, + 513, + 108.84615384615384 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 330, + 108.84615384615384, + 513, + 122.69230769230768 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 330, + 122.69230769230768, + 513, + 136.53846153846152 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 330, + 136.53846153846152, + 513, + 150.38461538461536 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 330, + 150.38461538461536, + 513, + 164.2307692307692 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 330, + 164.2307692307692, + 513, + 178.07692307692304 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 330, + 178.07692307692304, + 513, + 191.92307692307688 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 330, + 191.92307692307688, + 513, + 205.76923076923072 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 330, + 205.76923076923072, + 513, + 219.61538461538456 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 330, + 219.61538461538456, + 513, + 233.4615384615384 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 330, + 233.4615384615384, + 513, + 247.30769230769224 + ], + "spans": [], + "index": 14 + }, + { + "bbox": [ + 330, + 247.30769230769224, + 513, + 261.1538461538461 + ], + "spans": [], + "index": 15 + }, + { + "bbox": [ + 330, + 261.1538461538461, + 513, + 274.99999999999994 + ], + "spans": [], + "index": 16 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 348, + 285, + 495, + 297 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 346, + 283, + 495, + 299 + ], + "spans": [ + { + "bbox": [ + 346, + 283, + 495, + 299 + ], + "score": 1.0, + "content": "Figure 7: VAEAC for synthetic data.", + "type": "text" + } + ], + "index": 17 + } + ], + "index": 17 + } + ], + "index": 13.5 + }, + { + "type": "image", + "bbox": [ + 106, + 340, + 511, + 522 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 106, + 340, + 511, + 522 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 106, + 340, + 511, + 522 + ], + "spans": [ + { + "bbox": [ + 106, + 340, + 511, + 522 + ], + "score": 0.954, + "type": "image", + "image_path": "54463305da3093916463d301746c0407e7d8d2daa9a2e48291c407dc4d828109.jpg" + } + ] + } + ], + "index": 19, + "virtual_lines": [ + { + "bbox": [ + 106, + 340, + 511, + 400.6666666666667 + ], + "spans": [], + "index": 18 + }, + { + "bbox": [ + 106, + 400.6666666666667, + 511, + 461.33333333333337 + ], + "spans": [], + "index": 19 + }, + { + "bbox": [ + 106, + 461.33333333333337, + 511, + 522.0 + ], + "spans": [], + "index": 20 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 97, + 537, + 506, + 560 + ], + "group_id": 2, + "lines": [ + { + "bbox": [ + 244, + 536, + 366, + 549 + ], + "spans": [ + { + "bbox": [ + 244, + 536, + 366, + 549 + ], + "score": 1.0, + "content": "Figure 8: MNIST inpaintings.", + "type": "text" + } + ], + "index": 21 + }, + { + "bbox": [ + 106, + 546, + 504, + 561 + ], + "spans": [ + { + "bbox": [ + 106, + 546, + 504, + 561 + ], + "score": 1.0, + "content": "Left: input. The gray pixels are unobserved. Middle: samples from the model. Right: ground truth.", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 21.5 + } + ], + "index": 20.25 + }, + { + "type": "text", + "bbox": [ + 91, + 601, + 520, + 624 + ], + "lines": [], + "index": 23.5, + "bbox_fs": [ + 89, + 601, + 522, + 624 + ], + "lines_deleted": true + }, + { + "type": "text", + "bbox": [ + 90, + 629, + 518, + 652 + ], + "lines": [ + { + "bbox": [ + 89, + 628, + 521, + 642 + ], + "spans": [ + { + "bbox": [ + 89, + 628, + 164, + 642 + ], + "score": 1.0, + "content": "We see that using", + "type": "text" + }, + { + "bbox": [ + 165, + 630, + 192, + 641 + ], + "score": 0.9, + "content": "\\alpha \\neq 1", + "type": "inline_equation" + }, + { + "bbox": [ + 193, + 628, + 521, + 642 + ], + "score": 1.0, + "content": "ruins multimodality of the restored distribution, so we highly recommend to use", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 90, + 640, + 189, + 651 + ], + "spans": [ + { + "bbox": [ + 90, + 641, + 116, + 650 + ], + "score": 0.89, + "content": "\\alpha = 1", + "type": "inline_equation" + }, + { + "bbox": [ + 117, + 640, + 158, + 651 + ], + "score": 1.0, + "content": "or at least", + "type": "text" + }, + { + "bbox": [ + 159, + 641, + 185, + 650 + ], + "score": 0.88, + "content": "\\alpha \\approx 1", + "type": "inline_equation" + }, + { + "bbox": [ + 185, + 640, + 189, + 651 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25.5, + "bbox_fs": [ + 89, + 628, + 521, + 651 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 177, + 146, + 434, + 228 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 90, + 93, + 521, + 137 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 88, + 92, + 521, + 107 + ], + "spans": [ + { + "bbox": [ + 88, + 92, + 390, + 107 + ], + "score": 1.0, + "content": "Table 9: Average negative log-likelihood of inpaintings for 1000 objects. IS-", + "type": "text" + }, + { + "bbox": [ + 390, + 93, + 398, + 104 + ], + "score": 0.61, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 92, + 521, + 107 + ], + "score": 1.0, + "content": "refers to Importance Sampling", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 90, + 105, + 521, + 117 + ], + "spans": [ + { + "bbox": [ + 90, + 105, + 214, + 117 + ], + "score": 1.0, + "content": "log-likelihood estimation with", + "type": "text" + }, + { + "bbox": [ + 214, + 105, + 222, + 115 + ], + "score": 0.77, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 105, + 363, + 117 + ], + "score": 1.0, + "content": "samples for each object (18). MC-", + "type": "text" + }, + { + "bbox": [ + 363, + 105, + 371, + 115 + ], + "score": 0.69, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 105, + 521, + 117 + ], + "score": 1.0, + "content": "refers to Monte-Carlo log-likelihood", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 89, + 115, + 521, + 128 + ], + "spans": [ + { + "bbox": [ + 89, + 115, + 154, + 128 + ], + "score": 1.0, + "content": "estimation with", + "type": "text" + }, + { + "bbox": [ + 154, + 116, + 162, + 126 + ], + "score": 0.79, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 115, + 521, + 128 + ], + "score": 1.0, + "content": "samples for each object (17). Naive Bayes is a baseline method which assumes pixels and", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 89, + 126, + 177, + 140 + ], + "spans": [ + { + "bbox": [ + 89, + 126, + 177, + 140 + ], + "score": 1.0, + "content": "colors independence.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "table_body", + "bbox": [ + 177, + 146, + 434, + 228 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 177, + 146, + 434, + 228 + ], + "spans": [ + { + "bbox": [ + 177, + 146, + 434, + 228 + ], + "score": 0.929, + "html": "
MethodMNISTOmniglotCelebA
VAEAC IS-10261±1275±1734035 ± 1609
VAEAC MC-10494±41452 ± 10941513 ± 2163
VAEAC MC-102156 ±12203 ± 15053904 ± 3121
GSNN MC-104141 ±71199 ± 6253427 ± 2208
GSNN MC-10²141 ±11200 ± 6253486 ± 2210
Naive Bayes2052490269480
", + "type": "table", + "image_path": "a38ab45727edbf8bab4aa92d10957fbb17674c382278cea78615c2779533612a.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 177, + 146, + 434, + 173.33333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 177, + 173.33333333333334, + 434, + 200.66666666666669 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 177, + 200.66666666666669, + 434, + 228.00000000000003 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 3.25 + }, + { + "type": "image", + "bbox": [ + 208, + 261, + 387, + 380 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 208, + 261, + 387, + 380 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 208, + 261, + 387, + 380 + ], + "spans": [ + { + "bbox": [ + 208, + 261, + 387, + 380 + ], + "score": 0.914, + "type": "image", + "image_path": "3126337f9f6dd982a9196015b6247be9dab5083bb45c2e9eb0945400c4c56f31.jpg" + } + ] + } + ], + "index": 10.5, + "virtual_lines": [ + { + "bbox": [ + 208, + 261, + 387, + 275.875 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 208, + 275.875, + 387, + 290.75 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 208, + 290.75, + 387, + 305.625 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 208, + 305.625, + 387, + 320.5 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 208, + 320.5, + 387, + 335.375 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 208, + 335.375, + 387, + 350.25 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 208, + 350.25, + 387, + 365.125 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 208, + 365.125, + 387, + 380.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 177, + 403, + 433, + 415 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 176, + 402, + 434, + 417 + ], + "spans": [ + { + "bbox": [ + 176, + 402, + 434, + 417 + ], + "score": 1.0, + "content": "Figure 9: Convergence of VAE and VAEAC on MNIST dataset.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + } + ], + "index": 12.75 + }, + { + "type": "text", + "bbox": [ + 90, + 446, + 339, + 458 + ], + "lines": [ + { + "bbox": [ + 90, + 446, + 340, + 459 + ], + "spans": [ + { + "bbox": [ + 90, + 446, + 340, + 459 + ], + "score": 1.0, + "content": "C.3 COMPARISON ON THE IMAGE INPAINTING PROBLEM", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 90, + 471, + 520, + 494 + ], + "lines": [ + { + "bbox": [ + 88, + 470, + 522, + 485 + ], + "spans": [ + { + "bbox": [ + 88, + 470, + 522, + 485 + ], + "score": 1.0, + "content": "In figure 8 we can see that the inpaintings produced by GSNN are smooth, blurry and not diverse compared", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 89, + 483, + 148, + 495 + ], + "spans": [ + { + "bbox": [ + 89, + 483, + 148, + 495 + ], + "score": 1.0, + "content": "with VAEAC.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + }, + { + "type": "text", + "bbox": [ + 91, + 499, + 521, + 533 + ], + "lines": [ + { + "bbox": [ + 88, + 497, + 521, + 514 + ], + "spans": [ + { + "bbox": [ + 88, + 497, + 521, + 514 + ], + "score": 1.0, + "content": "Table 9 shows that VAEAC learns distribution over inpaintings better than GSNN in terms of test log-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 90, + 511, + 521, + 523 + ], + "spans": [ + { + "bbox": [ + 90, + 511, + 521, + 523 + ], + "score": 1.0, + "content": "likelihood. Nevertheless, Monte-Carlo estimations with a small number of samples sometimes are better for", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 90, + 521, + 493, + 534 + ], + "spans": [ + { + "bbox": [ + 90, + 521, + 493, + 534 + ], + "score": 1.0, + "content": "GSNN, which means less local modes in the learned distribution and more blurriness in the samples.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20 + }, + { + "type": "title", + "bbox": [ + 91, + 561, + 256, + 574 + ], + "lines": [ + { + "bbox": [ + 89, + 560, + 258, + 575 + ], + "spans": [ + { + "bbox": [ + 89, + 560, + 258, + 575 + ], + "score": 1.0, + "content": "D ADDITIONAL EXPERIMENTS", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "title", + "bbox": [ + 91, + 592, + 214, + 604 + ], + "lines": [ + { + "bbox": [ + 90, + 592, + 216, + 605 + ], + "spans": [ + { + "bbox": [ + 90, + 592, + 216, + 605 + ], + "score": 1.0, + "content": "D.1 CONVERGENCE SPEED", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 91, + 618, + 521, + 651 + ], + "lines": [ + { + "bbox": [ + 89, + 617, + 521, + 630 + ], + "spans": [ + { + "bbox": [ + 89, + 617, + 521, + 630 + ], + "score": 1.0, + "content": "In figure 9 one can see that VAEAC has similar convergence speed to VAE in terms of iterations on MNIST", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 90, + 628, + 521, + 642 + ], + "spans": [ + { + "bbox": [ + 90, + 628, + 521, + 642 + ], + "score": 1.0, + "content": "dataset. In our experiments we observed the same behaviour for other datasets. Each iteration of VAEAC is", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 89, + 639, + 410, + 653 + ], + "spans": [ + { + "bbox": [ + 89, + 639, + 410, + 653 + ], + "score": 1.0, + "content": "about 1.5 times slower than VAE due to usage of three networks instead of two.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25 + } + ], + "page_idx": 20, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 40, + 277, + 50 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 284, + 671, + 294, + 680 + ], + "lines": [ + { + "bbox": [ + 282, + 669, + 296, + 684 + ], + "spans": [ + { + "bbox": [ + 282, + 669, + 296, + 684 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 14 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 177, + 146, + 434, + 228 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 90, + 93, + 521, + 137 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 88, + 92, + 521, + 107 + ], + "spans": [ + { + "bbox": [ + 88, + 92, + 390, + 107 + ], + "score": 1.0, + "content": "Table 9: Average negative log-likelihood of inpaintings for 1000 objects. IS-", + "type": "text" + }, + { + "bbox": [ + 390, + 93, + 398, + 104 + ], + "score": 0.61, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 399, + 92, + 521, + 107 + ], + "score": 1.0, + "content": "refers to Importance Sampling", + "type": "text" + } + ], + "index": 0 + }, + { + "bbox": [ + 90, + 105, + 521, + 117 + ], + "spans": [ + { + "bbox": [ + 90, + 105, + 214, + 117 + ], + "score": 1.0, + "content": "log-likelihood estimation with", + "type": "text" + }, + { + "bbox": [ + 214, + 105, + 222, + 115 + ], + "score": 0.77, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 223, + 105, + 363, + 117 + ], + "score": 1.0, + "content": "samples for each object (18). MC-", + "type": "text" + }, + { + "bbox": [ + 363, + 105, + 371, + 115 + ], + "score": 0.69, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 371, + 105, + 521, + 117 + ], + "score": 1.0, + "content": "refers to Monte-Carlo log-likelihood", + "type": "text" + } + ], + "index": 1 + }, + { + "bbox": [ + 89, + 115, + 521, + 128 + ], + "spans": [ + { + "bbox": [ + 89, + 115, + 154, + 128 + ], + "score": 1.0, + "content": "estimation with", + "type": "text" + }, + { + "bbox": [ + 154, + 116, + 162, + 126 + ], + "score": 0.79, + "content": "S", + "type": "inline_equation" + }, + { + "bbox": [ + 163, + 115, + 521, + 128 + ], + "score": 1.0, + "content": "samples for each object (17). Naive Bayes is a baseline method which assumes pixels and", + "type": "text" + } + ], + "index": 2 + }, + { + "bbox": [ + 89, + 126, + 177, + 140 + ], + "spans": [ + { + "bbox": [ + 89, + 126, + 177, + 140 + ], + "score": 1.0, + "content": "colors independence.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 1.5 + }, + { + "type": "table_body", + "bbox": [ + 177, + 146, + 434, + 228 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 177, + 146, + 434, + 228 + ], + "spans": [ + { + "bbox": [ + 177, + 146, + 434, + 228 + ], + "score": 0.929, + "html": "
MethodMNISTOmniglotCelebA
VAEAC IS-10261±1275±1734035 ± 1609
VAEAC MC-10494±41452 ± 10941513 ± 2163
VAEAC MC-102156 ±12203 ± 15053904 ± 3121
GSNN MC-104141 ±71199 ± 6253427 ± 2208
GSNN MC-10²141 ±11200 ± 6253486 ± 2210
Naive Bayes2052490269480
", + "type": "table", + "image_path": "a38ab45727edbf8bab4aa92d10957fbb17674c382278cea78615c2779533612a.jpg" + } + ] + } + ], + "index": 5, + "virtual_lines": [ + { + "bbox": [ + 177, + 146, + 434, + 173.33333333333334 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 177, + 173.33333333333334, + 434, + 200.66666666666669 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 177, + 200.66666666666669, + 434, + 228.00000000000003 + ], + "spans": [], + "index": 6 + } + ] + } + ], + "index": 3.25 + }, + { + "type": "image", + "bbox": [ + 208, + 261, + 387, + 380 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 208, + 261, + 387, + 380 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 208, + 261, + 387, + 380 + ], + "spans": [ + { + "bbox": [ + 208, + 261, + 387, + 380 + ], + "score": 0.914, + "type": "image", + "image_path": "3126337f9f6dd982a9196015b6247be9dab5083bb45c2e9eb0945400c4c56f31.jpg" + } + ] + } + ], + "index": 10.5, + "virtual_lines": [ + { + "bbox": [ + 208, + 261, + 387, + 275.875 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 208, + 275.875, + 387, + 290.75 + ], + "spans": [], + "index": 8 + }, + { + "bbox": [ + 208, + 290.75, + 387, + 305.625 + ], + "spans": [], + "index": 9 + }, + { + "bbox": [ + 208, + 305.625, + 387, + 320.5 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 208, + 320.5, + 387, + 335.375 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 208, + 335.375, + 387, + 350.25 + ], + "spans": [], + "index": 12 + }, + { + "bbox": [ + 208, + 350.25, + 387, + 365.125 + ], + "spans": [], + "index": 13 + }, + { + "bbox": [ + 208, + 365.125, + 387, + 380.0 + ], + "spans": [], + "index": 14 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 177, + 403, + 433, + 415 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 176, + 402, + 434, + 417 + ], + "spans": [ + { + "bbox": [ + 176, + 402, + 434, + 417 + ], + "score": 1.0, + "content": "Figure 9: Convergence of VAE and VAEAC on MNIST dataset.", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + } + ], + "index": 12.75 + }, + { + "type": "text", + "bbox": [ + 90, + 446, + 339, + 458 + ], + "lines": [ + { + "bbox": [ + 90, + 446, + 340, + 459 + ], + "spans": [ + { + "bbox": [ + 90, + 446, + 340, + 459 + ], + "score": 1.0, + "content": "C.3 COMPARISON ON THE IMAGE INPAINTING PROBLEM", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16, + "bbox_fs": [ + 90, + 446, + 340, + 459 + ] + }, + { + "type": "text", + "bbox": [ + 90, + 471, + 520, + 494 + ], + "lines": [ + { + "bbox": [ + 88, + 470, + 522, + 485 + ], + "spans": [ + { + "bbox": [ + 88, + 470, + 522, + 485 + ], + "score": 1.0, + "content": "In figure 8 we can see that the inpaintings produced by GSNN are smooth, blurry and not diverse compared", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 89, + 483, + 148, + 495 + ], + "spans": [ + { + "bbox": [ + 89, + 483, + 148, + 495 + ], + "score": 1.0, + "content": "with VAEAC.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 88, + 470, + 522, + 495 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 499, + 521, + 533 + ], + "lines": [ + { + "bbox": [ + 88, + 497, + 521, + 514 + ], + "spans": [ + { + "bbox": [ + 88, + 497, + 521, + 514 + ], + "score": 1.0, + "content": "Table 9 shows that VAEAC learns distribution over inpaintings better than GSNN in terms of test log-", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 90, + 511, + 521, + 523 + ], + "spans": [ + { + "bbox": [ + 90, + 511, + 521, + 523 + ], + "score": 1.0, + "content": "likelihood. Nevertheless, Monte-Carlo estimations with a small number of samples sometimes are better for", + "type": "text" + } + ], + "index": 20 + }, + { + "bbox": [ + 90, + 521, + 493, + 534 + ], + "spans": [ + { + "bbox": [ + 90, + 521, + 493, + 534 + ], + "score": 1.0, + "content": "GSNN, which means less local modes in the learned distribution and more blurriness in the samples.", + "type": "text" + } + ], + "index": 21 + } + ], + "index": 20, + "bbox_fs": [ + 88, + 497, + 521, + 534 + ] + }, + { + "type": "title", + "bbox": [ + 91, + 561, + 256, + 574 + ], + "lines": [ + { + "bbox": [ + 89, + 560, + 258, + 575 + ], + "spans": [ + { + "bbox": [ + 89, + 560, + 258, + 575 + ], + "score": 1.0, + "content": "D ADDITIONAL EXPERIMENTS", + "type": "text" + } + ], + "index": 22 + } + ], + "index": 22 + }, + { + "type": "title", + "bbox": [ + 91, + 592, + 214, + 604 + ], + "lines": [ + { + "bbox": [ + 90, + 592, + 216, + 605 + ], + "spans": [ + { + "bbox": [ + 90, + 592, + 216, + 605 + ], + "score": 1.0, + "content": "D.1 CONVERGENCE SPEED", + "type": "text" + } + ], + "index": 23 + } + ], + "index": 23 + }, + { + "type": "text", + "bbox": [ + 91, + 618, + 521, + 651 + ], + "lines": [ + { + "bbox": [ + 89, + 617, + 521, + 630 + ], + "spans": [ + { + "bbox": [ + 89, + 617, + 521, + 630 + ], + "score": 1.0, + "content": "In figure 9 one can see that VAEAC has similar convergence speed to VAE in terms of iterations on MNIST", + "type": "text" + } + ], + "index": 24 + }, + { + "bbox": [ + 90, + 628, + 521, + 642 + ], + "spans": [ + { + "bbox": [ + 90, + 628, + 521, + 642 + ], + "score": 1.0, + "content": "dataset. In our experiments we observed the same behaviour for other datasets. Each iteration of VAEAC is", + "type": "text" + } + ], + "index": 25 + }, + { + "bbox": [ + 89, + 639, + 410, + 653 + ], + "spans": [ + { + "bbox": [ + 89, + 639, + 410, + 653 + ], + "score": 1.0, + "content": "about 1.5 times slower than VAE due to usage of three networks instead of two.", + "type": "text" + } + ], + "index": 26 + } + ], + "index": 25, + "bbox_fs": [ + 89, + 617, + 521, + 653 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "table", + "bbox": [ + 95, + 112, + 519, + 272 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 97, + 93, + 514, + 105 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 96, + 91, + 515, + 107 + ], + "spans": [ + { + "bbox": [ + 96, + 91, + 515, + 107 + ], + "score": 1.0, + "content": "Table 10: NRMSE (for continuous datasets) or PFC (for categorical ones) of imputations. Less is better.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 95, + 112, + 519, + 272 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 95, + 112, + 519, + 272 + ], + "spans": [ + { + "bbox": [ + 95, + 112, + 519, + 272 + ], + "score": 0.979, + "html": "
DatasetMICEMissForestGAINVAEACGSNNNN
Boston0.69 ± 0.020.58±0.020.78 ± 0.030.71 ± 0.020.70 ± 0.010.69 ± 0.01
Breast0.58 ±0.020.515 ± 0.0080.67 ±0.010.55±0.020.55 ± 0.020.52 ±0.02
Concrete0.850 ± 0.0070.78 ± 0.010.96 ±0.010.84 ±0.020.85 ± 0.012±3
Diabetes0.80 ±0.010.84±0.020.911 ± 0.0090.90 ±0.030.91 ± 0.030.90 ±0.02
Digits0.69 ±0.020.61± 0.020.79±0.020.69 ±0.020.69 ± 0.020.67 ±0.02
Glass0.91 ±0.020.83± 0.041.06 ±0.050.91 ±0.040.91 ± 0.050.87 ± 0.04
Iris0.59 ± 0.020.62 ± 0.040.72 ± 0.040.64± 0.040.62 ± 0.040.61± 0.02
Mushroom0.334 ± 0.0020.249 ±0.0060.271 ±0.0030.241 ± 0.0020.2412 ± 0.00090.239 ± 0.001
Orthopedic0.76 ±0.020.79±0.030.91±0.030.80±0.030.81 ±0.030.81 ±0.02
Phishing0.422 ± 0.0060.422 ±0.0090.427 ± 0.0100.397 ± 0.0100.392 ±0.0090.41 ± 0.01
WallRobot0.885 ± 0.0030.640 ± 0.0030.907 ± 0.0050.78 ±0.010.776 ±0.0070.757± 0.005
WhiteWine0.964 ± 0.0070.878 ±0.0090.97 ± 0.020.850 ± 0.0050.848 ± 0.0070.85 ± 0.01
Yeast0.98±0.021.00 ± 0.020.99 ±0.030.95 ± 0.010.958 ± 0.0070.97 ± 0.03
Zoo0.19 ± 0.030.16 ±0.020.20 ±0.020.16±0.020.17 ±0.020.16 ±0.01
", + "type": "table", + "image_path": "a28b110ac71efa360491f823a261c429125aef3652390412ba5272709177d2d9.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 95, + 112, + 519, + 165.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 95, + 165.33333333333334, + 519, + 218.66666666666669 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 95, + 218.66666666666669, + 519, + 272.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "table", + "bbox": [ + 93, + 314, + 519, + 476 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 88, + 283, + 518, + 306 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 90, + 282, + 519, + 297 + ], + "spans": [ + { + "bbox": [ + 90, + 282, + 519, + 297 + ], + "score": 1.0, + "content": "Table 11: R2-score (for continuous targets) or accuracy (for categorical ones) of post-imputation regression", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 90, + 295, + 225, + 307 + ], + "spans": [ + { + "bbox": [ + 90, + 295, + 225, + 307 + ], + "score": 1.0, + "content": "or classification. Higher is better.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "table_body", + "bbox": [ + 93, + 314, + 519, + 476 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 93, + 314, + 519, + 476 + ], + "spans": [ + { + "bbox": [ + 93, + 314, + 519, + 476 + ], + "score": 0.98, + "html": "
DatasetMICEMissForestGAINVAEACGSNNNN
Boston0.57 ± 0.080.6 ± 0.10.50 ± 0.100.5 ± 0.10.5± 0.10.50±0.09
Breast0.96 ± 0.020.95 ± 0.020.94± 0.010.95 ± 0.020.96 ± 0.020.95 ± 0.02
Concrete0.35 ± 0.050.33 ± 0.040.28±0.060.30 ± 0.080.32 ± 0.050±1
Diabetes0.37 ± 0.060.34±0.060.34±0.030.34± 0.040.33 ±0.040.27±0.06
Digits0.86±0.020.887 ±0.0080.83 ±0.030.892 ± 0.0100.895 ± 0.0100.912 ± 0.010
Glass0.44 ±0.080.53 ±0.050.37±0.050.49 ±0.090.47 ±0.090.48 ±0.09
Iris0.81± 0.020.84±0.020.66 ±0.060.84±0.050.82±0.060.73±0.09
Mushroom0.92 ±0.010.972 ± 0.0030.969 ± 0.0050.987 ± 0.0010.986 ± 0.0020.989 ± 0.003
Orthopedic0.71 ± 0.020.72 ± 0.030.60±0.030.71 ±0.020.70± 0.040.61±0.04
Phishing0.75 ± 0.020.73±0.030.74± 0.030.75 ± 0.010.74±0.040.73±0.02
WallRobot0.55 ±0.010.697 ± 0.0050.56 ±0.010.62±0.020.62 ± 0.010.64±0.02
WhiteWine0.13 ± 0.020.17 ± 0.010.11± 0.010.18 ±0.020.17 ± 0.010.15 ± 0.03
Yeast0.42 ±0.020.41 ±0.020.39 ±0.060.42 ± 0.010.425 ± 0.0100.33±0.03
Zoo0.78 ± 0.060.71 ±0.080.67±0.060.77 ± 0.090.8±0.10.83 ± 0.08
", + "type": "table", + "image_path": "7d3a3cbee5bc4c33e77c2d6020bcaa263208c0156dec7db727d7d8e3202cc1d5.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 93, + 314, + 519, + 368.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 93, + 368.0, + 519, + 422.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 93, + 422.0, + 519, + 476.0 + ], + "spans": [], + "index": 8 + } + ] + } + ], + "index": 5.75 + }, + { + "type": "title", + "bbox": [ + 91, + 497, + 259, + 509 + ], + "lines": [ + { + "bbox": [ + 89, + 497, + 260, + 510 + ], + "spans": [ + { + "bbox": [ + 89, + 497, + 260, + 510 + ], + "score": 1.0, + "content": "D.2 MISSING FEATURES IMPUTATION", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 91, + 518, + 521, + 574 + ], + "lines": [ + { + "bbox": [ + 90, + 519, + 521, + 531 + ], + "spans": [ + { + "bbox": [ + 90, + 519, + 521, + 531 + ], + "score": 1.0, + "content": "We evaluate the quality of imputations on different datasets (mostly from UCI (Lichman, 2013)). The", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 90, + 530, + 521, + 542 + ], + "spans": [ + { + "bbox": [ + 90, + 530, + 521, + 542 + ], + "score": 1.0, + "content": "evaluation is performed for VAEAC, GSNN (15) and NN (neural network; can be considered as a special case", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 88, + 540, + 522, + 553 + ], + "spans": [ + { + "bbox": [ + 88, + 540, + 159, + 553 + ], + "score": 1.0, + "content": "of GSNN where", + "type": "text" + }, + { + "bbox": [ + 159, + 540, + 214, + 552 + ], + "score": 0.93, + "content": "p _ { \\theta } ( z | x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 540, + 522, + 553 + ], + "score": 1.0, + "content": "is delta-function; produces single imputation). We compare these methods", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 90, + 551, + 521, + 563 + ], + "spans": [ + { + "bbox": [ + 90, + 551, + 521, + 563 + ], + "score": 1.0, + "content": "with MICE (Buuren & Groothuis-Oudshoorn, 2010), MissForest (Stekhoven & Buhlmann, 2011), and GAIN ¨", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 90, + 562, + 169, + 575 + ], + "spans": [ + { + "bbox": [ + 90, + 562, + 169, + 575 + ], + "score": 1.0, + "content": "(Yoon et al., 2018).", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12 + }, + { + "type": "text", + "bbox": [ + 89, + 579, + 521, + 602 + ], + "lines": [ + { + "bbox": [ + 89, + 578, + 522, + 592 + ], + "spans": [ + { + "bbox": [ + 89, + 578, + 522, + 592 + ], + "score": 1.0, + "content": "We see that for some datasets MICE and MissForest outperform VAEAC, GSNN and NN. The reason is that", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 90, + 590, + 403, + 602 + ], + "spans": [ + { + "bbox": [ + 90, + 590, + 403, + 602 + ], + "score": 1.0, + "content": "for some datasets random forest is more natural structure than neural network.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5 + }, + { + "type": "text", + "bbox": [ + 91, + 607, + 521, + 652 + ], + "lines": [ + { + "bbox": [ + 90, + 606, + 522, + 619 + ], + "spans": [ + { + "bbox": [ + 90, + 606, + 522, + 619 + ], + "score": 1.0, + "content": "The results also show that VAEAC, GSNN and NN show similar imputation performance in terms of", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 89, + 617, + 522, + 630 + ], + "spans": [ + { + "bbox": [ + 89, + 617, + 522, + 630 + ], + "score": 1.0, + "content": "NRMSE, PFC, post-imputation R2-score and accuracy. Given the result from appendix C we can take", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 89, + 628, + 522, + 641 + ], + "spans": [ + { + "bbox": [ + 89, + 628, + 522, + 641 + ], + "score": 1.0, + "content": "this as a weak evidence that the distribution of imputations has only one local maximum for datasets from", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 89, + 640, + 162, + 653 + ], + "spans": [ + { + "bbox": [ + 89, + 640, + 162, + 653 + ], + "score": 1.0, + "content": "(Lichman, 2013).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5 + } + ], + "page_idx": 21, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 40, + 276, + 50 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 284, + 671, + 295, + 680 + ], + "lines": [ + { + "bbox": [ + 282, + 670, + 296, + 684 + ], + "spans": [ + { + "bbox": [ + 282, + 670, + 296, + 684 + ], + "score": 1.0, + "content": "22", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "table", + "bbox": [ + 95, + 112, + 519, + 272 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 97, + 93, + 514, + 105 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 96, + 91, + 515, + 107 + ], + "spans": [ + { + "bbox": [ + 96, + 91, + 515, + 107 + ], + "score": 1.0, + "content": "Table 10: NRMSE (for continuous datasets) or PFC (for categorical ones) of imputations. Less is better.", + "type": "text" + } + ], + "index": 0 + } + ], + "index": 0 + }, + { + "type": "table_body", + "bbox": [ + 95, + 112, + 519, + 272 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 95, + 112, + 519, + 272 + ], + "spans": [ + { + "bbox": [ + 95, + 112, + 519, + 272 + ], + "score": 0.979, + "html": "
DatasetMICEMissForestGAINVAEACGSNNNN
Boston0.69 ± 0.020.58±0.020.78 ± 0.030.71 ± 0.020.70 ± 0.010.69 ± 0.01
Breast0.58 ±0.020.515 ± 0.0080.67 ±0.010.55±0.020.55 ± 0.020.52 ±0.02
Concrete0.850 ± 0.0070.78 ± 0.010.96 ±0.010.84 ±0.020.85 ± 0.012±3
Diabetes0.80 ±0.010.84±0.020.911 ± 0.0090.90 ±0.030.91 ± 0.030.90 ±0.02
Digits0.69 ±0.020.61± 0.020.79±0.020.69 ±0.020.69 ± 0.020.67 ±0.02
Glass0.91 ±0.020.83± 0.041.06 ±0.050.91 ±0.040.91 ± 0.050.87 ± 0.04
Iris0.59 ± 0.020.62 ± 0.040.72 ± 0.040.64± 0.040.62 ± 0.040.61± 0.02
Mushroom0.334 ± 0.0020.249 ±0.0060.271 ±0.0030.241 ± 0.0020.2412 ± 0.00090.239 ± 0.001
Orthopedic0.76 ±0.020.79±0.030.91±0.030.80±0.030.81 ±0.030.81 ±0.02
Phishing0.422 ± 0.0060.422 ±0.0090.427 ± 0.0100.397 ± 0.0100.392 ±0.0090.41 ± 0.01
WallRobot0.885 ± 0.0030.640 ± 0.0030.907 ± 0.0050.78 ±0.010.776 ±0.0070.757± 0.005
WhiteWine0.964 ± 0.0070.878 ±0.0090.97 ± 0.020.850 ± 0.0050.848 ± 0.0070.85 ± 0.01
Yeast0.98±0.021.00 ± 0.020.99 ±0.030.95 ± 0.010.958 ± 0.0070.97 ± 0.03
Zoo0.19 ± 0.030.16 ±0.020.20 ±0.020.16±0.020.17 ±0.020.16 ±0.01
", + "type": "table", + "image_path": "a28b110ac71efa360491f823a261c429125aef3652390412ba5272709177d2d9.jpg" + } + ] + } + ], + "index": 2, + "virtual_lines": [ + { + "bbox": [ + 95, + 112, + 519, + 165.33333333333334 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 95, + 165.33333333333334, + 519, + 218.66666666666669 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 95, + 218.66666666666669, + 519, + 272.0 + ], + "spans": [], + "index": 3 + } + ] + } + ], + "index": 1.0 + }, + { + "type": "table", + "bbox": [ + 93, + 314, + 519, + 476 + ], + "blocks": [ + { + "type": "table_caption", + "bbox": [ + 88, + 283, + 518, + 306 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 90, + 282, + 519, + 297 + ], + "spans": [ + { + "bbox": [ + 90, + 282, + 519, + 297 + ], + "score": 1.0, + "content": "Table 11: R2-score (for continuous targets) or accuracy (for categorical ones) of post-imputation regression", + "type": "text" + } + ], + "index": 4 + }, + { + "bbox": [ + 90, + 295, + 225, + 307 + ], + "spans": [ + { + "bbox": [ + 90, + 295, + 225, + 307 + ], + "score": 1.0, + "content": "or classification. Higher is better.", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 4.5 + }, + { + "type": "table_body", + "bbox": [ + 93, + 314, + 519, + 476 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 93, + 314, + 519, + 476 + ], + "spans": [ + { + "bbox": [ + 93, + 314, + 519, + 476 + ], + "score": 0.98, + "html": "
DatasetMICEMissForestGAINVAEACGSNNNN
Boston0.57 ± 0.080.6 ± 0.10.50 ± 0.100.5 ± 0.10.5± 0.10.50±0.09
Breast0.96 ± 0.020.95 ± 0.020.94± 0.010.95 ± 0.020.96 ± 0.020.95 ± 0.02
Concrete0.35 ± 0.050.33 ± 0.040.28±0.060.30 ± 0.080.32 ± 0.050±1
Diabetes0.37 ± 0.060.34±0.060.34±0.030.34± 0.040.33 ±0.040.27±0.06
Digits0.86±0.020.887 ±0.0080.83 ±0.030.892 ± 0.0100.895 ± 0.0100.912 ± 0.010
Glass0.44 ±0.080.53 ±0.050.37±0.050.49 ±0.090.47 ±0.090.48 ±0.09
Iris0.81± 0.020.84±0.020.66 ±0.060.84±0.050.82±0.060.73±0.09
Mushroom0.92 ±0.010.972 ± 0.0030.969 ± 0.0050.987 ± 0.0010.986 ± 0.0020.989 ± 0.003
Orthopedic0.71 ± 0.020.72 ± 0.030.60±0.030.71 ±0.020.70± 0.040.61±0.04
Phishing0.75 ± 0.020.73±0.030.74± 0.030.75 ± 0.010.74±0.040.73±0.02
WallRobot0.55 ±0.010.697 ± 0.0050.56 ±0.010.62±0.020.62 ± 0.010.64±0.02
WhiteWine0.13 ± 0.020.17 ± 0.010.11± 0.010.18 ±0.020.17 ± 0.010.15 ± 0.03
Yeast0.42 ±0.020.41 ±0.020.39 ±0.060.42 ± 0.010.425 ± 0.0100.33±0.03
Zoo0.78 ± 0.060.71 ±0.080.67±0.060.77 ± 0.090.8±0.10.83 ± 0.08
", + "type": "table", + "image_path": "7d3a3cbee5bc4c33e77c2d6020bcaa263208c0156dec7db727d7d8e3202cc1d5.jpg" + } + ] + } + ], + "index": 7, + "virtual_lines": [ + { + "bbox": [ + 93, + 314, + 519, + 368.0 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 93, + 368.0, + 519, + 422.0 + ], + "spans": [], + "index": 7 + }, + { + "bbox": [ + 93, + 422.0, + 519, + 476.0 + ], + "spans": [], + "index": 8 + } + ] + } + ], + "index": 5.75 + }, + { + "type": "title", + "bbox": [ + 91, + 497, + 259, + 509 + ], + "lines": [ + { + "bbox": [ + 89, + 497, + 260, + 510 + ], + "spans": [ + { + "bbox": [ + 89, + 497, + 260, + 510 + ], + "score": 1.0, + "content": "D.2 MISSING FEATURES IMPUTATION", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 91, + 518, + 521, + 574 + ], + "lines": [ + { + "bbox": [ + 90, + 519, + 521, + 531 + ], + "spans": [ + { + "bbox": [ + 90, + 519, + 521, + 531 + ], + "score": 1.0, + "content": "We evaluate the quality of imputations on different datasets (mostly from UCI (Lichman, 2013)). The", + "type": "text" + } + ], + "index": 10 + }, + { + "bbox": [ + 90, + 530, + 521, + 542 + ], + "spans": [ + { + "bbox": [ + 90, + 530, + 521, + 542 + ], + "score": 1.0, + "content": "evaluation is performed for VAEAC, GSNN (15) and NN (neural network; can be considered as a special case", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 88, + 540, + 522, + 553 + ], + "spans": [ + { + "bbox": [ + 88, + 540, + 159, + 553 + ], + "score": 1.0, + "content": "of GSNN where", + "type": "text" + }, + { + "bbox": [ + 159, + 540, + 214, + 552 + ], + "score": 0.93, + "content": "p _ { \\theta } ( z | x _ { 1 - b } , b )", + "type": "inline_equation" + }, + { + "bbox": [ + 214, + 540, + 522, + 553 + ], + "score": 1.0, + "content": "is delta-function; produces single imputation). We compare these methods", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 90, + 551, + 521, + 563 + ], + "spans": [ + { + "bbox": [ + 90, + 551, + 521, + 563 + ], + "score": 1.0, + "content": "with MICE (Buuren & Groothuis-Oudshoorn, 2010), MissForest (Stekhoven & Buhlmann, 2011), and GAIN ¨", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 90, + 562, + 169, + 575 + ], + "spans": [ + { + "bbox": [ + 90, + 562, + 169, + 575 + ], + "score": 1.0, + "content": "(Yoon et al., 2018).", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 12, + "bbox_fs": [ + 88, + 519, + 522, + 575 + ] + }, + { + "type": "text", + "bbox": [ + 89, + 579, + 521, + 602 + ], + "lines": [ + { + "bbox": [ + 89, + 578, + 522, + 592 + ], + "spans": [ + { + "bbox": [ + 89, + 578, + 522, + 592 + ], + "score": 1.0, + "content": "We see that for some datasets MICE and MissForest outperform VAEAC, GSNN and NN. The reason is that", + "type": "text" + } + ], + "index": 15 + }, + { + "bbox": [ + 90, + 590, + 403, + 602 + ], + "spans": [ + { + "bbox": [ + 90, + 590, + 403, + 602 + ], + "score": 1.0, + "content": "for some datasets random forest is more natural structure than neural network.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 15.5, + "bbox_fs": [ + 89, + 578, + 522, + 602 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 607, + 521, + 652 + ], + "lines": [ + { + "bbox": [ + 90, + 606, + 522, + 619 + ], + "spans": [ + { + "bbox": [ + 90, + 606, + 522, + 619 + ], + "score": 1.0, + "content": "The results also show that VAEAC, GSNN and NN show similar imputation performance in terms of", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 89, + 617, + 522, + 630 + ], + "spans": [ + { + "bbox": [ + 89, + 617, + 522, + 630 + ], + "score": 1.0, + "content": "NRMSE, PFC, post-imputation R2-score and accuracy. Given the result from appendix C we can take", + "type": "text" + } + ], + "index": 18 + }, + { + "bbox": [ + 89, + 628, + 522, + 641 + ], + "spans": [ + { + "bbox": [ + 89, + 628, + 522, + 641 + ], + "score": 1.0, + "content": "this as a weak evidence that the distribution of imputations has only one local maximum for datasets from", + "type": "text" + } + ], + "index": 19 + }, + { + "bbox": [ + 89, + 640, + 162, + 653 + ], + "spans": [ + { + "bbox": [ + 89, + 640, + 162, + 653 + ], + "score": 1.0, + "content": "(Lichman, 2013).", + "type": "text" + } + ], + "index": 20 + } + ], + "index": 18.5, + "bbox_fs": [ + 89, + 606, + 522, + 653 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 150, + 95, + 465, + 363 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 150, + 95, + 465, + 363 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 150, + 95, + 465, + 363 + ], + "spans": [ + { + "bbox": [ + 150, + 95, + 465, + 363 + ], + "score": 0.978, + "type": "image", + "image_path": "21041c7b3c7fc1baa4707c17e3944ec8fc55aadb337ff273f13ec39db52954c7.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 150, + 95, + 465, + 184.33333333333331 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 150, + 184.33333333333331, + 465, + 273.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 150, + 273.66666666666663, + 465, + 362.99999999999994 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 123, + 377, + 495, + 399 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 175, + 377, + 435, + 389 + ], + "spans": [ + { + "bbox": [ + 175, + 377, + 435, + 389 + ], + "score": 1.0, + "content": "Figure 10: CelebA inpaintings with masks from (Li et al., 2017).", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 120, + 387, + 497, + 401 + ], + "spans": [ + { + "bbox": [ + 120, + 387, + 497, + 401 + ], + "score": 1.0, + "content": "ft: input. The gray pixels are unobserved. Middle: samples from VAEAC. Right: ground truth.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "title", + "bbox": [ + 91, + 426, + 198, + 438 + ], + "lines": [ + { + "bbox": [ + 90, + 426, + 199, + 439 + ], + "spans": [ + { + "bbox": [ + 90, + 426, + 199, + 439 + ], + "score": 1.0, + "content": "D.3 FACE INPAINTINGS", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 92, + 447, + 500, + 459 + ], + "lines": [ + { + "bbox": [ + 89, + 446, + 502, + 461 + ], + "spans": [ + { + "bbox": [ + 89, + 446, + 502, + 461 + ], + "score": 1.0, + "content": "In figure 10 we provide samples of VAEAC on the CelebA dataset for the masks from (Li et al., 2017).", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6 + }, + { + "type": "title", + "bbox": [ + 91, + 474, + 249, + 486 + ], + "lines": [ + { + "bbox": [ + 90, + 474, + 250, + 487 + ], + "spans": [ + { + "bbox": [ + 90, + 474, + 250, + 487 + ], + "score": 1.0, + "content": "D.4 GAIN FOR IMAGE INPAINTING", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 91, + 496, + 520, + 529 + ], + "lines": [ + { + "bbox": [ + 89, + 493, + 522, + 510 + ], + "spans": [ + { + "bbox": [ + 89, + 493, + 522, + 510 + ], + "score": 1.0, + "content": "GAIN (Yoon et al., 2018) doesnt use unobserved data during training, which makes it easier to apply to", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 90, + 506, + 521, + 520 + ], + "spans": [ + { + "bbox": [ + 90, + 506, + 521, + 520 + ], + "score": 1.0, + "content": "the missing features imputation problem. Nevertheless, it turns into a disadvantage when the fully-observed", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 89, + 517, + 395, + 531 + ], + "spans": [ + { + "bbox": [ + 89, + 517, + 395, + 531 + ], + "score": 1.0, + "content": "training data is available but the missingness rate at the testing stage is high.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9 + }, + { + "type": "text", + "bbox": [ + 91, + 534, + 520, + 590 + ], + "lines": [ + { + "bbox": [ + 90, + 534, + 522, + 547 + ], + "spans": [ + { + "bbox": [ + 90, + 534, + 522, + 547 + ], + "score": 1.0, + "content": "We consider the horizontal line mask for MNIST which is described in appendix A.3. We use the released", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 89, + 544, + 522, + 558 + ], + "spans": [ + { + "bbox": [ + 89, + 544, + 522, + 558 + ], + "score": 1.0, + "content": "GAIN code 5 with a different mask generator. The inpaintings from VAEAC which uses the unobserved", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 88, + 556, + 522, + 569 + ], + "spans": [ + { + "bbox": [ + 88, + 556, + 522, + 569 + ], + "score": 1.0, + "content": "pixels during training are available in figure 1. The inpaintings from GAIN which ignores unobserved pixels", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 88, + 567, + 522, + 581 + ], + "spans": [ + { + "bbox": [ + 88, + 567, + 522, + 581 + ], + "score": 1.0, + "content": "are provided in figure 11. As can be seen in figure 11, GAIN fails to learn conditional distribution for given", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 89, + 577, + 184, + 592 + ], + "spans": [ + { + "bbox": [ + 89, + 577, + 162, + 592 + ], + "score": 1.0, + "content": "mask distribution", + "type": "text" + }, + { + "bbox": [ + 162, + 578, + 180, + 591 + ], + "score": 0.92, + "content": "{ \\dot { p } } ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 577, + 184, + 592 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13 + }, + { + "type": "text", + "bbox": [ + 91, + 595, + 520, + 629 + ], + "lines": [ + { + "bbox": [ + 89, + 594, + 520, + 608 + ], + "spans": [ + { + "bbox": [ + 89, + 594, + 520, + 608 + ], + "score": 1.0, + "content": "Nevertheless, we don’t claim that GAIN is not suitable for image inpainting. As it was shown in the supple-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 90, + 606, + 521, + 619 + ], + "spans": [ + { + "bbox": [ + 90, + 606, + 521, + 619 + ], + "score": 1.0, + "content": "mentary of (Yoon et al., 2018) and in the corresponding code, GAIN is able to learn conditional distributions", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 90, + 617, + 411, + 631 + ], + "spans": [ + { + "bbox": [ + 90, + 617, + 114, + 631 + ], + "score": 1.0, + "content": "when", + "type": "text" + }, + { + "bbox": [ + 114, + 618, + 132, + 630 + ], + "score": 0.92, + "content": "p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 617, + 411, + 631 + ], + "score": 1.0, + "content": "is pixel-wise independent Bernoulli distribution with probability 0.5.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17 + } + ], + "page_idx": 22, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 103, + 641, + 241, + 651 + ], + "lines": [ + { + "bbox": [ + 102, + 639, + 244, + 654 + ], + "spans": [ + { + "bbox": [ + 102, + 639, + 244, + 654 + ], + "score": 1.0, + "content": "5https://github.com/jsyoon0823/GAIN", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 90, + 40, + 277, + 50 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 284, + 671, + 295, + 680 + ], + "lines": [ + { + "bbox": [ + 282, + 669, + 296, + 683 + ], + "spans": [ + { + "bbox": [ + 282, + 669, + 296, + 683 + ], + "score": 1.0, + "content": "23", + "type": "text" + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 150, + 95, + 465, + 363 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 150, + 95, + 465, + 363 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 150, + 95, + 465, + 363 + ], + "spans": [ + { + "bbox": [ + 150, + 95, + 465, + 363 + ], + "score": 0.978, + "type": "image", + "image_path": "21041c7b3c7fc1baa4707c17e3944ec8fc55aadb337ff273f13ec39db52954c7.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 150, + 95, + 465, + 184.33333333333331 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 150, + 184.33333333333331, + 465, + 273.66666666666663 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 150, + 273.66666666666663, + 465, + 362.99999999999994 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 123, + 377, + 495, + 399 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 175, + 377, + 435, + 389 + ], + "spans": [ + { + "bbox": [ + 175, + 377, + 435, + 389 + ], + "score": 1.0, + "content": "Figure 10: CelebA inpaintings with masks from (Li et al., 2017).", + "type": "text" + } + ], + "index": 3 + }, + { + "bbox": [ + 120, + 387, + 497, + 401 + ], + "spans": [ + { + "bbox": [ + 120, + 387, + 497, + 401 + ], + "score": 1.0, + "content": "ft: input. The gray pixels are unobserved. Middle: samples from VAEAC. Right: ground truth.", + "type": "text" + } + ], + "index": 4 + } + ], + "index": 3.5 + } + ], + "index": 2.25 + }, + { + "type": "title", + "bbox": [ + 91, + 426, + 198, + 438 + ], + "lines": [ + { + "bbox": [ + 90, + 426, + 199, + 439 + ], + "spans": [ + { + "bbox": [ + 90, + 426, + 199, + 439 + ], + "score": 1.0, + "content": "D.3 FACE INPAINTINGS", + "type": "text" + } + ], + "index": 5 + } + ], + "index": 5 + }, + { + "type": "text", + "bbox": [ + 92, + 447, + 500, + 459 + ], + "lines": [ + { + "bbox": [ + 89, + 446, + 502, + 461 + ], + "spans": [ + { + "bbox": [ + 89, + 446, + 502, + 461 + ], + "score": 1.0, + "content": "In figure 10 we provide samples of VAEAC on the CelebA dataset for the masks from (Li et al., 2017).", + "type": "text" + } + ], + "index": 6 + } + ], + "index": 6, + "bbox_fs": [ + 89, + 446, + 502, + 461 + ] + }, + { + "type": "title", + "bbox": [ + 91, + 474, + 249, + 486 + ], + "lines": [ + { + "bbox": [ + 90, + 474, + 250, + 487 + ], + "spans": [ + { + "bbox": [ + 90, + 474, + 250, + 487 + ], + "score": 1.0, + "content": "D.4 GAIN FOR IMAGE INPAINTING", + "type": "text" + } + ], + "index": 7 + } + ], + "index": 7 + }, + { + "type": "text", + "bbox": [ + 91, + 496, + 520, + 529 + ], + "lines": [ + { + "bbox": [ + 89, + 493, + 522, + 510 + ], + "spans": [ + { + "bbox": [ + 89, + 493, + 522, + 510 + ], + "score": 1.0, + "content": "GAIN (Yoon et al., 2018) doesnt use unobserved data during training, which makes it easier to apply to", + "type": "text" + } + ], + "index": 8 + }, + { + "bbox": [ + 90, + 506, + 521, + 520 + ], + "spans": [ + { + "bbox": [ + 90, + 506, + 521, + 520 + ], + "score": 1.0, + "content": "the missing features imputation problem. Nevertheless, it turns into a disadvantage when the fully-observed", + "type": "text" + } + ], + "index": 9 + }, + { + "bbox": [ + 89, + 517, + 395, + 531 + ], + "spans": [ + { + "bbox": [ + 89, + 517, + 395, + 531 + ], + "score": 1.0, + "content": "training data is available but the missingness rate at the testing stage is high.", + "type": "text" + } + ], + "index": 10 + } + ], + "index": 9, + "bbox_fs": [ + 89, + 493, + 522, + 531 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 534, + 520, + 590 + ], + "lines": [ + { + "bbox": [ + 90, + 534, + 522, + 547 + ], + "spans": [ + { + "bbox": [ + 90, + 534, + 522, + 547 + ], + "score": 1.0, + "content": "We consider the horizontal line mask for MNIST which is described in appendix A.3. We use the released", + "type": "text" + } + ], + "index": 11 + }, + { + "bbox": [ + 89, + 544, + 522, + 558 + ], + "spans": [ + { + "bbox": [ + 89, + 544, + 522, + 558 + ], + "score": 1.0, + "content": "GAIN code 5 with a different mask generator. The inpaintings from VAEAC which uses the unobserved", + "type": "text" + } + ], + "index": 12 + }, + { + "bbox": [ + 88, + 556, + 522, + 569 + ], + "spans": [ + { + "bbox": [ + 88, + 556, + 522, + 569 + ], + "score": 1.0, + "content": "pixels during training are available in figure 1. The inpaintings from GAIN which ignores unobserved pixels", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 88, + 567, + 522, + 581 + ], + "spans": [ + { + "bbox": [ + 88, + 567, + 522, + 581 + ], + "score": 1.0, + "content": "are provided in figure 11. As can be seen in figure 11, GAIN fails to learn conditional distribution for given", + "type": "text" + } + ], + "index": 14 + }, + { + "bbox": [ + 89, + 577, + 184, + 592 + ], + "spans": [ + { + "bbox": [ + 89, + 577, + 162, + 592 + ], + "score": 1.0, + "content": "mask distribution", + "type": "text" + }, + { + "bbox": [ + 162, + 578, + 180, + 591 + ], + "score": 0.92, + "content": "{ \\dot { p } } ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 180, + 577, + 184, + 592 + ], + "score": 1.0, + "content": ".", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 13, + "bbox_fs": [ + 88, + 534, + 522, + 592 + ] + }, + { + "type": "text", + "bbox": [ + 91, + 595, + 520, + 629 + ], + "lines": [ + { + "bbox": [ + 89, + 594, + 520, + 608 + ], + "spans": [ + { + "bbox": [ + 89, + 594, + 520, + 608 + ], + "score": 1.0, + "content": "Nevertheless, we don’t claim that GAIN is not suitable for image inpainting. As it was shown in the supple-", + "type": "text" + } + ], + "index": 16 + }, + { + "bbox": [ + 90, + 606, + 521, + 619 + ], + "spans": [ + { + "bbox": [ + 90, + 606, + 521, + 619 + ], + "score": 1.0, + "content": "mentary of (Yoon et al., 2018) and in the corresponding code, GAIN is able to learn conditional distributions", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 90, + 617, + 411, + 631 + ], + "spans": [ + { + "bbox": [ + 90, + 617, + 114, + 631 + ], + "score": 1.0, + "content": "when", + "type": "text" + }, + { + "bbox": [ + 114, + 618, + 132, + 630 + ], + "score": 0.92, + "content": "p ( b )", + "type": "inline_equation" + }, + { + "bbox": [ + 133, + 617, + 411, + 631 + ], + "score": 1.0, + "content": "is pixel-wise independent Bernoulli distribution with probability 0.5.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17, + "bbox_fs": [ + 89, + 594, + 521, + 631 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 213, + 97, + 398, + 230 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 213, + 97, + 398, + 230 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 213, + 97, + 398, + 230 + ], + "spans": [ + { + "bbox": [ + 213, + 97, + 398, + 230 + ], + "score": 0.966, + "type": "image", + "image_path": "77978a748fdfece774bbdba6a454e4caa9d264c220caf33b0d01dfee4003f59b.jpg" + } + ] + } + ], + "index": 3.5, + "virtual_lines": [ + { + "bbox": [ + 213, + 97, + 398, + 113.625 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 213, + 113.625, + 398, + 130.25 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 213, + 130.25, + 398, + 146.875 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 213, + 146.875, + 398, + 163.5 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 213, + 163.5, + 398, + 180.125 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 213, + 180.125, + 398, + 196.75 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 213, + 196.75, + 398, + 213.375 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 213, + 213.375, + 398, + 230.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 219, + 243, + 393, + 254 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 217, + 242, + 394, + 257 + ], + "spans": [ + { + "bbox": [ + 217, + 242, + 394, + 257 + ], + "score": 1.0, + "content": "Figure 11: MNIST inpaintings from GAIN.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + } + ], + "index": 5.75 + }, + { + "type": "text", + "bbox": [ + 102, + 254, + 504, + 266 + ], + "lines": [ + { + "bbox": [ + 105, + 252, + 504, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 504, + 267 + ], + "score": 1.0, + "content": "Left: input. The gray pixels are unobserved. Middle: samples from the model. Right: ground truth.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9 + }, + { + "type": "image", + "bbox": [ + 107, + 280, + 511, + 460 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 280, + 511, + 460 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 107, + 280, + 511, + 460 + ], + "spans": [ + { + "bbox": [ + 107, + 280, + 511, + 460 + ], + "score": 0.97, + "type": "image", + "image_path": "3addba9b5d9e7527dc59c64b67c88b55ee2d758e97d2f9f04dcf482aada3e5b6.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 107, + 280, + 511, + 340.0 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 107, + 340.0, + 511, + 400.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 107, + 400.0, + 511, + 460.0 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 104, + 475, + 506, + 499 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 241, + 474, + 370, + 490 + ], + "spans": [ + { + "bbox": [ + 241, + 474, + 370, + 490 + ], + "score": 1.0, + "content": "Figure 12: MNIST inpaintings.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 485, + 504, + 500 + ], + "spans": [ + { + "bbox": [ + 106, + 485, + 504, + 500 + ], + "score": 1.0, + "content": "Left: input. The gray pixels are unobserved. Middle: samples from the model. Right: ground truth.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + } + ], + "index": 12.25 + }, + { + "type": "title", + "bbox": [ + 90, + 518, + 313, + 530 + ], + "lines": [ + { + "bbox": [ + 90, + 518, + 315, + 531 + ], + "spans": [ + { + "bbox": [ + 90, + 518, + 315, + 531 + ], + "score": 1.0, + "content": "D.5 UNIVERSAL MARGINALIZER: ILLUSTRATIONS", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 91, + 538, + 485, + 551 + ], + "lines": [ + { + "bbox": [ + 89, + 537, + 485, + 552 + ], + "spans": [ + { + "bbox": [ + 89, + 537, + 485, + 552 + ], + "score": 1.0, + "content": "In figure 12 we provide samples of Universal Marginalizer (UM) and VAEAC for the same inputs.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16 + }, + { + "type": "text", + "bbox": [ + 88, + 555, + 522, + 579 + ], + "lines": [ + { + "bbox": [ + 91, + 556, + 521, + 568 + ], + "spans": [ + { + "bbox": [ + 91, + 556, + 521, + 568 + ], + "score": 1.0, + "content": "Consider the case when UM marginal distributions are parametrized with Gaussians. The most simple", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 88, + 567, + 445, + 579 + ], + "spans": [ + { + "bbox": [ + 88, + 567, + 445, + 579 + ], + "score": 1.0, + "content": "example of a distribution, which UM cannot learn but VAEAC can, is given in figure 13.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5 + } + ], + "page_idx": 23, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 40, + 277, + 50 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 284, + 671, + 295, + 680 + ], + "lines": [ + { + "bbox": [ + 282, + 669, + 297, + 684 + ], + "spans": [ + { + "bbox": [ + 282, + 669, + 297, + 684 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 213, + 97, + 398, + 230 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 213, + 97, + 398, + 230 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 213, + 97, + 398, + 230 + ], + "spans": [ + { + "bbox": [ + 213, + 97, + 398, + 230 + ], + "score": 0.966, + "type": "image", + "image_path": "77978a748fdfece774bbdba6a454e4caa9d264c220caf33b0d01dfee4003f59b.jpg" + } + ] + } + ], + "index": 3.5, + "virtual_lines": [ + { + "bbox": [ + 213, + 97, + 398, + 113.625 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 213, + 113.625, + 398, + 130.25 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 213, + 130.25, + 398, + 146.875 + ], + "spans": [], + "index": 2 + }, + { + "bbox": [ + 213, + 146.875, + 398, + 163.5 + ], + "spans": [], + "index": 3 + }, + { + "bbox": [ + 213, + 163.5, + 398, + 180.125 + ], + "spans": [], + "index": 4 + }, + { + "bbox": [ + 213, + 180.125, + 398, + 196.75 + ], + "spans": [], + "index": 5 + }, + { + "bbox": [ + 213, + 196.75, + 398, + 213.375 + ], + "spans": [], + "index": 6 + }, + { + "bbox": [ + 213, + 213.375, + 398, + 230.0 + ], + "spans": [], + "index": 7 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 219, + 243, + 393, + 254 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 217, + 242, + 394, + 257 + ], + "spans": [ + { + "bbox": [ + 217, + 242, + 394, + 257 + ], + "score": 1.0, + "content": "Figure 11: MNIST inpaintings from GAIN.", + "type": "text" + } + ], + "index": 8 + } + ], + "index": 8 + } + ], + "index": 5.75 + }, + { + "type": "text", + "bbox": [ + 102, + 254, + 504, + 266 + ], + "lines": [ + { + "bbox": [ + 105, + 252, + 504, + 267 + ], + "spans": [ + { + "bbox": [ + 105, + 252, + 504, + 267 + ], + "score": 1.0, + "content": "Left: input. The gray pixels are unobserved. Middle: samples from the model. Right: ground truth.", + "type": "text" + } + ], + "index": 9 + } + ], + "index": 9, + "bbox_fs": [ + 105, + 252, + 504, + 267 + ] + }, + { + "type": "image", + "bbox": [ + 107, + 280, + 511, + 460 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 107, + 280, + 511, + 460 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 107, + 280, + 511, + 460 + ], + "spans": [ + { + "bbox": [ + 107, + 280, + 511, + 460 + ], + "score": 0.97, + "type": "image", + "image_path": "3addba9b5d9e7527dc59c64b67c88b55ee2d758e97d2f9f04dcf482aada3e5b6.jpg" + } + ] + } + ], + "index": 11, + "virtual_lines": [ + { + "bbox": [ + 107, + 280, + 511, + 340.0 + ], + "spans": [], + "index": 10 + }, + { + "bbox": [ + 107, + 340.0, + 511, + 400.0 + ], + "spans": [], + "index": 11 + }, + { + "bbox": [ + 107, + 400.0, + 511, + 460.0 + ], + "spans": [], + "index": 12 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 104, + 475, + 506, + 499 + ], + "group_id": 1, + "lines": [ + { + "bbox": [ + 241, + 474, + 370, + 490 + ], + "spans": [ + { + "bbox": [ + 241, + 474, + 370, + 490 + ], + "score": 1.0, + "content": "Figure 12: MNIST inpaintings.", + "type": "text" + } + ], + "index": 13 + }, + { + "bbox": [ + 106, + 485, + 504, + 500 + ], + "spans": [ + { + "bbox": [ + 106, + 485, + 504, + 500 + ], + "score": 1.0, + "content": "Left: input. The gray pixels are unobserved. Middle: samples from the model. Right: ground truth.", + "type": "text" + } + ], + "index": 14 + } + ], + "index": 13.5 + } + ], + "index": 12.25 + }, + { + "type": "title", + "bbox": [ + 90, + 518, + 313, + 530 + ], + "lines": [ + { + "bbox": [ + 90, + 518, + 315, + 531 + ], + "spans": [ + { + "bbox": [ + 90, + 518, + 315, + 531 + ], + "score": 1.0, + "content": "D.5 UNIVERSAL MARGINALIZER: ILLUSTRATIONS", + "type": "text" + } + ], + "index": 15 + } + ], + "index": 15 + }, + { + "type": "text", + "bbox": [ + 91, + 538, + 485, + 551 + ], + "lines": [ + { + "bbox": [ + 89, + 537, + 485, + 552 + ], + "spans": [ + { + "bbox": [ + 89, + 537, + 485, + 552 + ], + "score": 1.0, + "content": "In figure 12 we provide samples of Universal Marginalizer (UM) and VAEAC for the same inputs.", + "type": "text" + } + ], + "index": 16 + } + ], + "index": 16, + "bbox_fs": [ + 89, + 537, + 485, + 552 + ] + }, + { + "type": "text", + "bbox": [ + 88, + 555, + 522, + 579 + ], + "lines": [ + { + "bbox": [ + 91, + 556, + 521, + 568 + ], + "spans": [ + { + "bbox": [ + 91, + 556, + 521, + 568 + ], + "score": 1.0, + "content": "Consider the case when UM marginal distributions are parametrized with Gaussians. The most simple", + "type": "text" + } + ], + "index": 17 + }, + { + "bbox": [ + 88, + 567, + 445, + 579 + ], + "spans": [ + { + "bbox": [ + 88, + 567, + 445, + 579 + ], + "score": 1.0, + "content": "example of a distribution, which UM cannot learn but VAEAC can, is given in figure 13.", + "type": "text" + } + ], + "index": 18 + } + ], + "index": 17.5, + "bbox_fs": [ + 88, + 556, + 521, + 579 + ] + } + ] + }, + { + "preproc_blocks": [ + { + "type": "image", + "bbox": [ + 97, + 297, + 514, + 420 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 97, + 297, + 514, + 420 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 97, + 297, + 514, + 420 + ], + "spans": [ + { + "bbox": [ + 97, + 297, + 514, + 420 + ], + "score": 0.953, + "type": "image", + "image_path": "34b2efedea65ec8dd1cd6662835f9b38b5a648676328c4f707b65cec76fc2de2.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 97, + 297, + 514, + 338.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 97, + 338.0, + 514, + 379.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 97, + 379.0, + 514, + 420.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 204, + 436, + 406, + 448 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 205, + 435, + 406, + 450 + ], + "spans": [ + { + "bbox": [ + 205, + 435, + 406, + 450 + ], + "score": 1.0, + "content": "Figure 13: Distribution learning: VAEAC vs UM.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ], + "page_idx": 24, + "page_size": [ + 612, + 792 + ], + "discarded_blocks": [ + { + "type": "discarded", + "bbox": [ + 91, + 40, + 277, + 51 + ], + "lines": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "spans": [ + { + "bbox": [ + 90, + 39, + 277, + 52 + ], + "score": 1.0, + "content": "Published as a conference paper at ICLR 2019", + "type": "text" + } + ] + } + ] + }, + { + "type": "discarded", + "bbox": [ + 284, + 671, + 295, + 680 + ], + "lines": [ + { + "bbox": [ + 282, + 669, + 297, + 684 + ], + "spans": [ + { + "bbox": [ + 282, + 669, + 297, + 684 + ], + "score": 1.0, + "content": "", + "type": "text", + "height": 15, + "width": 15 + } + ] + } + ] + } + ], + "para_blocks": [ + { + "type": "image", + "bbox": [ + 97, + 297, + 514, + 420 + ], + "blocks": [ + { + "type": "image_body", + "bbox": [ + 97, + 297, + 514, + 420 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 97, + 297, + 514, + 420 + ], + "spans": [ + { + "bbox": [ + 97, + 297, + 514, + 420 + ], + "score": 0.953, + "type": "image", + "image_path": "34b2efedea65ec8dd1cd6662835f9b38b5a648676328c4f707b65cec76fc2de2.jpg" + } + ] + } + ], + "index": 1, + "virtual_lines": [ + { + "bbox": [ + 97, + 297, + 514, + 338.0 + ], + "spans": [], + "index": 0 + }, + { + "bbox": [ + 97, + 338.0, + 514, + 379.0 + ], + "spans": [], + "index": 1 + }, + { + "bbox": [ + 97, + 379.0, + 514, + 420.0 + ], + "spans": [], + "index": 2 + } + ] + }, + { + "type": "image_caption", + "bbox": [ + 204, + 436, + 406, + 448 + ], + "group_id": 0, + "lines": [ + { + "bbox": [ + 205, + 435, + 406, + 450 + ], + "spans": [ + { + "bbox": [ + 205, + 435, + 406, + 450 + ], + "score": 1.0, + "content": "Figure 13: Distribution learning: VAEAC vs UM.", + "type": "text" + } + ], + "index": 3 + } + ], + "index": 3 + } + ], + "index": 2.0 + } + ] + } + ], + "_backend": "pipeline", + "_version_name": "2.2.2" +} \ No newline at end of file diff --git a/parse/train/SyxtJh0qYm/SyxtJh0qYm_model.json b/parse/train/SyxtJh0qYm/SyxtJh0qYm_model.json new file mode 100644 index 0000000000000000000000000000000000000000..aa96ca05d65e101977b9a44ccd6799033c147eb3 --- /dev/null +++ b/parse/train/SyxtJh0qYm/SyxtJh0qYm_model.json @@ -0,0 +1,30947 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 253, + 1351, + 1448, + 1351, + 1448, + 1567, + 253, + 1567 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 353, + 772, + 1348, + 772, + 1348, + 957, + 353, + 957 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 253, + 1582, + 1449, + 1582, + 1449, + 1766, + 253, + 1766 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 253, + 1213, + 1447, + 1213, + 1447, + 1336, + 253, + 1336 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 254, + 1075, + 1446, + 1075, + 1446, + 1198, + 254, + 1198 + ], + "score": 0.973 + }, + { + "category_id": 0, + "poly": [ + 253, + 259, + 951, + 259, + 951, + 361, + 253, + 361 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 1060, + 414, + 1390, + 414, + 1390, + 627, + 1060, + 627 + ], + "score": 0.943 + }, + { + "category_id": 1, + "poly": [ + 268, + 413, + 593, + 413, + 593, + 537, + 268, + 537 + ], + "score": 0.942 + }, + { + "category_id": 0, + "poly": [ + 257, + 1007, + 529, + 1007, + 529, + 1043, + 257, + 1043 + ], + "score": 0.896 + }, + { + "category_id": 2, + "poly": [ + 287, + 1783, + 582, + 1783, + 582, + 1810, + 287, + 1810 + ], + "score": 0.895 + }, + { + "category_id": 2, + "poly": [ + 254, + 112, + 771, + 112, + 771, + 141, + 254, + 141 + ], + "score": 0.882 + }, + { + "category_id": 0, + "poly": [ + 772, + 708, + 927, + 708, + 927, + 741, + 772, + 741 + ], + "score": 0.845 + }, + { + "category_id": 1, + "poly": [ + 660, + 412, + 1003, + 412, + 1003, + 567, + 660, + 567 + ], + "score": 0.832 + }, + { + "category_id": 2, + "poly": [ + 796, + 1866, + 812, + 1866, + 812, + 1889, + 796, + 1889 + ], + "score": 0.761 + }, + { + "category_id": 0, + "poly": [ + 1062, + 415, + 1234, + 415, + 1234, + 444, + 1062, + 444 + ], + "score": 0.17 + }, + { + "category_id": 0, + "poly": [ + 662, + 414, + 875, + 414, + 875, + 444, + 662, + 444 + ], + "score": 0.143 + }, + { + "category_id": 13, + "poly": [ + 949, + 1414, + 1025, + 1414, + 1025, + 1446, + 949, + 1446 + ], + "score": 0.94, + "latex": "p ( x | y )" + }, + { + "category_id": 13, + "poly": [ + 1030, + 1214, + 1105, + 1214, + 1105, + 1248, + 1030, + 1248 + ], + "score": 0.92, + "latex": "p ( x | y )" + }, + { + "category_id": 13, + "poly": [ + 1271, + 1137, + 1325, + 1137, + 1325, + 1170, + 1271, + 1170 + ], + "score": 0.92, + "latex": "p ( x )" + }, + { + "category_id": 13, + "poly": [ + 1307, + 1352, + 1438, + 1352, + 1438, + 1388, + 1307, + 1388 + ], + "score": 0.91, + "latex": "p ( x _ { I } | x _ { U \\setminus I } )" + }, + { + "category_id": 13, + "poly": [ + 443, + 1413, + 497, + 1413, + 497, + 1446, + 443, + 1446 + ], + "score": 0.91, + "latex": "p ( x )" + }, + { + "category_id": 13, + "poly": [ + 328, + 1384, + 354, + 1384, + 354, + 1410, + 328, + 1410 + ], + "score": 0.79, + "latex": "U" + }, + { + "category_id": 13, + "poly": [ + 1293, + 1221, + 1312, + 1221, + 1312, + 1242, + 1293, + 1242 + ], + "score": 0.77, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 365, + 1250, + 384, + 1250, + 384, + 1276, + 365, + 1276 + ], + "score": 0.76, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 681, + 1384, + 699, + 1384, + 699, + 1410, + 681, + 1410 + ], + "score": 0.68, + "latex": "I" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 254.0, + 872.0, + 254.0, + 872.0, + 308.0, + 251.0, + 308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 309.0, + 955.0, + 309.0, + 955.0, + 365.0, + 250.0, + 365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1003.0, + 534.0, + 1003.0, + 534.0, + 1050.0, + 249.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 1777.0, + 588.0, + 1777.0, + 588.0, + 1814.0, + 286.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 109.0, + 772.0, + 109.0, + 772.0, + 145.0, + 250.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 707.0, + 932.0, + 707.0, + 932.0, + 744.0, + 769.0, + 744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 1865.0, + 816.0, + 1865.0, + 816.0, + 1895.0, + 795.0, + 1895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 412.0, + 1238.0, + 412.0, + 1238.0, + 449.0, + 1060.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 410.0, + 879.0, + 410.0, + 879.0, + 452.0, + 657.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1346.0, + 1306.0, + 1346.0, + 1306.0, + 1393.0, + 246.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1439.0, + 1346.0, + 1451.0, + 1346.0, + 1451.0, + 1393.0, + 1439.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1380.0, + 327.0, + 1380.0, + 327.0, + 1417.0, + 249.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1380.0, + 680.0, + 1380.0, + 680.0, + 1417.0, + 355.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 700.0, + 1380.0, + 1451.0, + 1380.0, + 1451.0, + 1417.0, + 700.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1411.0, + 442.0, + 1411.0, + 442.0, + 1451.0, + 249.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 1411.0, + 948.0, + 1411.0, + 948.0, + 1451.0, + 498.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1026.0, + 1411.0, + 1452.0, + 1411.0, + 1452.0, + 1451.0, + 1026.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1441.0, + 1451.0, + 1441.0, + 1451.0, + 1479.0, + 249.0, + 1479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1470.0, + 1452.0, + 1470.0, + 1452.0, + 1511.0, + 246.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1503.0, + 1452.0, + 1503.0, + 1452.0, + 1539.0, + 247.0, + 1539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1532.0, + 1045.0, + 1532.0, + 1045.0, + 1572.0, + 249.0, + 1572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 773.0, + 1351.0, + 773.0, + 1351.0, + 808.0, + 350.0, + 808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 800.0, + 1350.0, + 800.0, + 1350.0, + 840.0, + 348.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 829.0, + 1351.0, + 829.0, + 1351.0, + 871.0, + 348.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 863.0, + 1351.0, + 863.0, + 1351.0, + 900.0, + 348.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 895.0, + 1351.0, + 895.0, + 1351.0, + 929.0, + 349.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 923.0, + 1226.0, + 923.0, + 1226.0, + 959.0, + 349.0, + 959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1580.0, + 1450.0, + 1580.0, + 1450.0, + 1617.0, + 249.0, + 1617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1611.0, + 1450.0, + 1611.0, + 1450.0, + 1647.0, + 248.0, + 1647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1645.0, + 1449.0, + 1645.0, + 1449.0, + 1677.0, + 250.0, + 1677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1674.0, + 1450.0, + 1674.0, + 1450.0, + 1710.0, + 249.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1702.0, + 1449.0, + 1702.0, + 1449.0, + 1737.0, + 249.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1735.0, + 1074.0, + 1735.0, + 1074.0, + 1770.0, + 249.0, + 1770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1213.0, + 1029.0, + 1213.0, + 1029.0, + 1248.0, + 249.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1106.0, + 1213.0, + 1292.0, + 1213.0, + 1292.0, + 1248.0, + 1106.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1313.0, + 1213.0, + 1448.0, + 1213.0, + 1448.0, + 1248.0, + 1313.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1242.0, + 364.0, + 1242.0, + 364.0, + 1280.0, + 249.0, + 1280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 385.0, + 1242.0, + 1450.0, + 1242.0, + 1450.0, + 1280.0, + 385.0, + 1280.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1273.0, + 1450.0, + 1273.0, + 1450.0, + 1311.0, + 249.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1304.0, + 892.0, + 1304.0, + 892.0, + 1340.0, + 248.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1074.0, + 1449.0, + 1074.0, + 1449.0, + 1109.0, + 251.0, + 1109.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1104.0, + 1449.0, + 1104.0, + 1449.0, + 1139.0, + 250.0, + 1139.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1137.0, + 1270.0, + 1137.0, + 1270.0, + 1171.0, + 250.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1326.0, + 1137.0, + 1449.0, + 1137.0, + 1449.0, + 1171.0, + 1326.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1164.0, + 609.0, + 1164.0, + 609.0, + 1205.0, + 249.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 412.0, + 1241.0, + 412.0, + 1241.0, + 446.0, + 1058.0, + 446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 441.0, + 1366.0, + 441.0, + 1366.0, + 479.0, + 1059.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 470.0, + 1392.0, + 470.0, + 1392.0, + 509.0, + 1056.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 504.0, + 1387.0, + 504.0, + 1387.0, + 537.0, + 1058.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 536.0, + 1386.0, + 536.0, + 1386.0, + 567.0, + 1059.0, + 567.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 564.0, + 1249.0, + 564.0, + 1249.0, + 598.0, + 1058.0, + 598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1059.0, + 597.0, + 1350.0, + 597.0, + 1350.0, + 631.0, + 1059.0, + 631.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 266.0, + 412.0, + 419.0, + 412.0, + 419.0, + 447.0, + 266.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 268.0, + 445.0, + 594.0, + 445.0, + 594.0, + 477.0, + 268.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 266.0, + 473.0, + 457.0, + 473.0, + 457.0, + 507.0, + 266.0, + 507.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 268.0, + 507.0, + 575.0, + 507.0, + 575.0, + 539.0, + 268.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 411.0, + 880.0, + 411.0, + 880.0, + 449.0, + 657.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 440.0, + 995.0, + 440.0, + 995.0, + 479.0, + 657.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 656.0, + 473.0, + 1007.0, + 473.0, + 1007.0, + 505.0, + 656.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 503.0, + 849.0, + 503.0, + 849.0, + 537.0, + 658.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.0, + 535.0, + 982.0, + 535.0, + 982.0, + 569.0, + 658.0, + 569.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 253, + 920, + 1448, + 920, + 1448, + 1136, + 253, + 1136 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 253, + 1457, + 1447, + 1457, + 1447, + 1673, + 253, + 1673 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 253, + 266, + 1448, + 266, + 1448, + 512, + 253, + 512 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 253, + 1150, + 1447, + 1150, + 1447, + 1335, + 253, + 1335 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 254, + 782, + 1448, + 782, + 1448, + 905, + 254, + 905 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 255, + 1688, + 1447, + 1688, + 1447, + 1810, + 255, + 1810 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 255, + 527, + 1447, + 527, + 1447, + 650, + 255, + 650 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 254, + 1350, + 1446, + 1350, + 1446, + 1442, + 254, + 1442 + ], + "score": 0.972 + }, + { + "category_id": 2, + "poly": [ + 254, + 113, + 770, + 113, + 770, + 141, + 254, + 141 + ], + "score": 0.906 + }, + { + "category_id": 0, + "poly": [ + 256, + 706, + 542, + 706, + 542, + 741, + 256, + 741 + ], + "score": 0.896 + }, + { + "category_id": 2, + "poly": [ + 796, + 1866, + 813, + 1866, + 813, + 1889, + 796, + 1889 + ], + "score": 0.777 + }, + { + "category_id": 15, + "poly": [ + 250.0, + 110.0, + 771.0, + 110.0, + 771.0, + 145.0, + 250.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 703.0, + 548.0, + 703.0, + 548.0, + 747.0, + 247.0, + 747.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 794.0, + 1863.0, + 817.0, + 1863.0, + 817.0, + 1898.0, + 794.0, + 1898.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 921.0, + 1447.0, + 921.0, + 1447.0, + 956.0, + 250.0, + 956.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 950.0, + 1451.0, + 950.0, + 1451.0, + 989.0, + 248.0, + 989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 981.0, + 1449.0, + 981.0, + 1449.0, + 1017.0, + 250.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1011.0, + 1451.0, + 1011.0, + 1451.0, + 1051.0, + 249.0, + 1051.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1044.0, + 1449.0, + 1044.0, + 1449.0, + 1079.0, + 249.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1070.0, + 1449.0, + 1070.0, + 1449.0, + 1111.0, + 248.0, + 1111.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1103.0, + 937.0, + 1103.0, + 937.0, + 1141.0, + 249.0, + 1141.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1458.0, + 1450.0, + 1458.0, + 1450.0, + 1493.0, + 249.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1487.0, + 1451.0, + 1487.0, + 1451.0, + 1523.0, + 247.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1517.0, + 1453.0, + 1517.0, + 1453.0, + 1557.0, + 249.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1548.0, + 1451.0, + 1548.0, + 1451.0, + 1588.0, + 247.0, + 1588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1581.0, + 1451.0, + 1581.0, + 1451.0, + 1616.0, + 249.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1609.0, + 1451.0, + 1609.0, + 1451.0, + 1645.0, + 247.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1640.0, + 1291.0, + 1640.0, + 1291.0, + 1678.0, + 249.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 267.0, + 1448.0, + 267.0, + 1448.0, + 301.0, + 252.0, + 301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 293.0, + 1449.0, + 293.0, + 1449.0, + 335.0, + 248.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 328.0, + 1451.0, + 328.0, + 1451.0, + 362.0, + 249.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 359.0, + 1451.0, + 359.0, + 1451.0, + 393.0, + 249.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 388.0, + 1449.0, + 388.0, + 1449.0, + 421.0, + 250.0, + 421.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 420.0, + 1448.0, + 420.0, + 1448.0, + 454.0, + 249.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 447.0, + 1452.0, + 447.0, + 1452.0, + 486.0, + 249.0, + 486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 478.0, + 736.0, + 478.0, + 736.0, + 517.0, + 248.0, + 517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1152.0, + 1450.0, + 1152.0, + 1450.0, + 1184.0, + 250.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1179.0, + 1451.0, + 1179.0, + 1451.0, + 1217.0, + 249.0, + 1217.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1212.0, + 1451.0, + 1212.0, + 1451.0, + 1248.0, + 249.0, + 1248.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1242.0, + 1451.0, + 1242.0, + 1451.0, + 1278.0, + 249.0, + 1278.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1273.0, + 1451.0, + 1273.0, + 1451.0, + 1309.0, + 249.0, + 1309.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1304.0, + 1176.0, + 1304.0, + 1176.0, + 1339.0, + 247.0, + 1339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 253.0, + 784.0, + 1449.0, + 784.0, + 1449.0, + 818.0, + 253.0, + 818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 811.0, + 1452.0, + 811.0, + 1452.0, + 849.0, + 247.0, + 849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 842.0, + 1451.0, + 842.0, + 1451.0, + 880.0, + 246.0, + 880.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 874.0, + 1313.0, + 874.0, + 1313.0, + 909.0, + 251.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1681.0, + 1451.0, + 1681.0, + 1451.0, + 1728.0, + 247.0, + 1728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1718.0, + 1449.0, + 1718.0, + 1449.0, + 1753.0, + 251.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1748.0, + 1450.0, + 1748.0, + 1450.0, + 1785.0, + 248.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1780.0, + 1198.0, + 1780.0, + 1198.0, + 1814.0, + 251.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 527.0, + 1450.0, + 527.0, + 1450.0, + 562.0, + 251.0, + 562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 555.0, + 1451.0, + 555.0, + 1451.0, + 594.0, + 249.0, + 594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 586.0, + 1451.0, + 586.0, + 1451.0, + 625.0, + 249.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 616.0, + 1414.0, + 616.0, + 1414.0, + 655.0, + 249.0, + 655.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1346.0, + 1452.0, + 1346.0, + 1452.0, + 1388.0, + 250.0, + 1388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1381.0, + 1448.0, + 1381.0, + 1448.0, + 1414.0, + 251.0, + 1414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1411.0, + 783.0, + 1411.0, + 783.0, + 1448.0, + 248.0, + 1448.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 251, + 389, + 1450, + 389, + 1450, + 579, + 251, + 579 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 253, + 874, + 1450, + 874, + 1450, + 1060, + 253, + 1060 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 252, + 1291, + 1449, + 1291, + 1449, + 1447, + 252, + 1447 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 254, + 591, + 1447, + 591, + 1447, + 717, + 254, + 717 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 253, + 1609, + 1450, + 1609, + 1450, + 1704, + 253, + 1704 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 255, + 1716, + 1447, + 1716, + 1447, + 1813, + 255, + 1813 + ], + "score": 0.971 + }, + { + "category_id": 8, + "poly": [ + 278, + 741, + 1385, + 741, + 1385, + 859, + 278, + 859 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 244, + 1459, + 1446, + 1459, + 1446, + 1526, + 244, + 1526 + ], + "score": 0.939 + }, + { + "category_id": 8, + "poly": [ + 357, + 1075, + 1342, + 1075, + 1342, + 1147, + 357, + 1147 + ], + "score": 0.925 + }, + { + "category_id": 0, + "poly": [ + 254, + 262, + 512, + 262, + 512, + 299, + 254, + 299 + ], + "score": 0.922 + }, + { + "category_id": 2, + "poly": [ + 253, + 111, + 770, + 111, + 770, + 142, + 253, + 142 + ], + "score": 0.917 + }, + { + "category_id": 1, + "poly": [ + 251, + 1163, + 1249, + 1163, + 1249, + 1197, + 251, + 1197 + ], + "score": 0.915 + }, + { + "category_id": 8, + "poly": [ + 313, + 1540, + 1375, + 1540, + 1375, + 1582, + 313, + 1582 + ], + "score": 0.905 + }, + { + "category_id": 0, + "poly": [ + 254, + 332, + 676, + 332, + 676, + 365, + 254, + 365 + ], + "score": 0.899 + }, + { + "category_id": 0, + "poly": [ + 253, + 1234, + 856, + 1234, + 856, + 1267, + 253, + 1267 + ], + "score": 0.889 + }, + { + "category_id": 9, + "poly": [ + 1410, + 1097, + 1445, + 1097, + 1445, + 1128, + 1410, + 1128 + ], + "score": 0.873 + }, + { + "category_id": 9, + "poly": [ + 1412, + 821, + 1445, + 821, + 1445, + 852, + 1412, + 852 + ], + "score": 0.863 + }, + { + "category_id": 9, + "poly": [ + 1412, + 1544, + 1445, + 1544, + 1445, + 1575, + 1412, + 1575 + ], + "score": 0.849 + }, + { + "category_id": 2, + "poly": [ + 796, + 1865, + 814, + 1865, + 814, + 1890, + 796, + 1890 + ], + "score": 0.796 + }, + { + "category_id": 14, + "poly": [ + 281, + 739, + 1387, + 739, + 1387, + 864, + 281, + 864 + ], + "score": 0.95, + "latex": "\\begin{array} { r l } & { \\log p _ { \\theta } ( x ) = \\mathbb { E } _ { q _ { \\phi } ( z | x ) } \\log \\frac { p _ { \\theta } ( x , z ) } { q _ { \\phi } ( z | x ) } + D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( z | x ) \\| p ( z | x , \\theta ) \\big ) } \\\\ & { \\qquad \\quad \\ge \\mathbb { E } _ { q _ { \\phi } ( z | x ) } \\log p _ { \\theta } ( x | z ) - D _ { \\mathrm { K L } } \\big ( q _ { \\phi } ( z | x ) \\| p ( z ) \\big ) = L _ { V A E } \\big ( x ; \\theta , \\phi \\big ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 303, + 1642, + 393, + 1642, + 393, + 1676, + 303, + 1676 + ], + "score": 0.93, + "latex": "p _ { \\psi } ( z | y )" + }, + { + "category_id": 13, + "poly": [ + 440, + 515, + 530, + 515, + 530, + 549, + 440, + 549 + ], + "score": 0.93, + "latex": "p _ { \\theta } ( x | z )" + }, + { + "category_id": 13, + "poly": [ + 1012, + 1324, + 1101, + 1324, + 1101, + 1357, + 1012, + 1357 + ], + "score": 0.93, + "latex": "p _ { d } ( x | y )" + }, + { + "category_id": 13, + "poly": [ + 316, + 875, + 404, + 875, + 404, + 909, + 316, + 909 + ], + "score": 0.93, + "latex": "q _ { \\phi } ( z | x )" + }, + { + "category_id": 13, + "poly": [ + 1199, + 591, + 1290, + 591, + 1290, + 627, + 1199, + 627 + ], + "score": 0.93, + "latex": "\\{ x _ { i } \\} _ { i = 1 } ^ { N }" + }, + { + "category_id": 13, + "poly": [ + 253, + 1324, + 341, + 1324, + 341, + 1357, + 253, + 1357 + ], + "score": 0.92, + "latex": "p _ { d } ( x | y )" + }, + { + "category_id": 13, + "poly": [ + 951, + 936, + 1089, + 936, + 1089, + 970, + 951, + 970 + ], + "score": 0.92, + "latex": "L _ { V A E } ( \\theta , \\phi )" + }, + { + "category_id": 13, + "poly": [ + 448, + 941, + 548, + 941, + 548, + 970, + 448, + 970 + ], + "score": 0.92, + "latex": "p ( z | x , \\theta )" + }, + { + "category_id": 13, + "poly": [ + 289, + 1164, + 427, + 1164, + 427, + 1198, + 289, + 1198 + ], + "score": 0.92, + "latex": "L _ { V A E } ( \\theta , \\phi )" + }, + { + "category_id": 13, + "poly": [ + 253, + 997, + 396, + 997, + 396, + 1031, + 253, + 1031 + ], + "score": 0.92, + "latex": "\\varepsilon \\sim \\mathcal { N } ( 0 , I )" + }, + { + "category_id": 13, + "poly": [ + 1190, + 483, + 1439, + 483, + 1439, + 518, + 1190, + 518 + ], + "score": 0.92, + "latex": "p _ { \\theta } ( x ) = \\mathbb { E } _ { p ( z ) } p _ { \\theta } ( x | z )" + }, + { + "category_id": 13, + "poly": [ + 445, + 624, + 512, + 624, + 512, + 657, + 445, + 657 + ], + "score": 0.92, + "latex": "p _ { d } ( x )" + }, + { + "category_id": 14, + "poly": [ + 358, + 1075, + 1341, + 1075, + 1341, + 1149, + 358, + 1149 + ], + "score": 0.92, + "latex": "\\frac { \\partial L _ { V A E } ( x ; \\theta , \\phi ) } { \\partial \\phi } = \\mathbb { E } _ { \\varepsilon \\sim \\mathcal { N } ( 0 , I ) } \\frac { \\partial } { \\partial \\phi } \\log p _ { \\theta } ( x | \\mu _ { \\phi } ( x ) + \\varepsilon \\sigma _ { \\phi } ( x ) ) - \\frac { \\partial } { \\partial \\phi } D _ { \\mathrm { K L } } ( q _ { \\phi } ( z | x ) \\| p ( z ) ) ." + }, + { + "category_id": 13, + "poly": [ + 326, + 938, + 414, + 938, + 414, + 971, + 326, + 971 + ], + "score": 0.92, + "latex": "q _ { \\phi } ( z | x )" + }, + { + "category_id": 13, + "poly": [ + 1118, + 966, + 1369, + 966, + 1369, + 1002, + 1118, + 1002 + ], + "score": 0.92, + "latex": "z = \\bar { \\mu } _ { \\phi } ( x ) + \\bar { \\varepsilon } \\sigma _ { \\phi } ( x )" + }, + { + "category_id": 13, + "poly": [ + 556, + 452, + 608, + 452, + 608, + 486, + 556, + 486 + ], + "score": 0.92, + "latex": "p ( z )" + }, + { + "category_id": 14, + "poly": [ + 318, + 1539, + 1381, + 1539, + 1381, + 1582, + 318, + 1582 + ], + "score": 0.91, + "latex": "L _ { C V A E } ( x , y ; \\theta , \\psi , \\phi ) = \\mathbb { E } _ { q _ { \\phi } ( z | x , y ) } \\log p _ { \\theta } ( x | z , y ) - D _ { \\mathrm { K L } } ( q _ { \\phi } ( z | x , y ) | | p _ { \\psi } ( z | y ) ) \\leq \\log p _ { \\theta , \\psi } ( x | y )" + }, + { + "category_id": 13, + "poly": [ + 475, + 909, + 574, + 909, + 574, + 939, + 475, + 939 + ], + "score": 0.91, + "latex": "p ( z | x , \\theta )" + }, + { + "category_id": 13, + "poly": [ + 1350, + 452, + 1437, + 452, + 1437, + 484, + 1350, + 484 + ], + "score": 0.9, + "latex": "p _ { \\theta } ( x | z )" + }, + { + "category_id": 13, + "poly": [ + 1272, + 877, + 1292, + 877, + 1292, + 907, + 1272, + 907 + ], + "score": 0.86, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 1154, + 1166, + 1174, + 1166, + 1174, + 1197, + 1154, + 1197 + ], + "score": 0.86, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 1129, + 515, + 1182, + 515, + 1182, + 549, + 1129, + 549 + ], + "score": 0.86, + "latex": "p ( z )" + }, + { + "category_id": 13, + "poly": [ + 1210, + 1643, + 1232, + 1643, + 1232, + 1674, + 1210, + 1674 + ], + "score": 0.86, + "latex": "\\psi" + }, + { + "category_id": 13, + "poly": [ + 536, + 1002, + 568, + 1002, + 568, + 1032, + 536, + 1032 + ], + "score": 0.84, + "latex": "\\sigma _ { \\phi }" + }, + { + "category_id": 13, + "poly": [ + 449, + 1001, + 482, + 1001, + 482, + 1032, + 449, + 1032 + ], + "score": 0.84, + "latex": "\\mu _ { \\phi }" + }, + { + "category_id": 13, + "poly": [ + 707, + 969, + 727, + 969, + 727, + 999, + 707, + 999 + ], + "score": 0.83, + "latex": "\\phi" + }, + { + "category_id": 13, + "poly": [ + 818, + 1358, + 837, + 1358, + 837, + 1381, + 818, + 1381 + ], + "score": 0.8, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 432, + 595, + 450, + 595, + 450, + 621, + 432, + 621 + ], + "score": 0.79, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 326, + 484, + 344, + 484, + 344, + 510, + 326, + 510 + ], + "score": 0.79, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 1225, + 1166, + 1241, + 1166, + 1241, + 1192, + 1225, + 1192 + ], + "score": 0.79, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 1127, + 1390, + 1146, + 1390, + 1146, + 1412, + 1127, + 1412 + ], + "score": 0.78, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 598, + 1647, + 616, + 1647, + 616, + 1674, + 598, + 1674 + ], + "score": 0.77, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 821, + 458, + 841, + 458, + 841, + 480, + 821, + 480 + ], + "score": 0.77, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 1286, + 427, + 1305, + 427, + 1305, + 449, + 1286, + 449 + ], + "score": 0.76, + "latex": "z" + }, + { + "category_id": 13, + "poly": [ + 531, + 1421, + 549, + 1421, + 549, + 1442, + 531, + 1442 + ], + "score": 0.75, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 349, + 1500, + 365, + 1500, + 365, + 1525, + 349, + 1525 + ], + "score": 0.72, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 497, + 1359, + 515, + 1359, + 515, + 1386, + 497, + 1386 + ], + "score": 0.64, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 1374, + 1329, + 1405, + 1329, + 1405, + 1352, + 1374, + 1352 + ], + "score": 0.64, + "latex": "x \\mathbf { s }" + }, + { + "category_id": 13, + "poly": [ + 1094, + 516, + 1111, + 516, + 1111, + 543, + 1094, + 543 + ], + "score": 0.53, + "latex": "\\theta" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 259.0, + 516.0, + 259.0, + 516.0, + 306.0, + 246.0, + 306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 109.0, + 772.0, + 109.0, + 772.0, + 145.0, + 251.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 333.0, + 680.0, + 333.0, + 680.0, + 366.0, + 251.0, + 366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1232.0, + 859.0, + 1232.0, + 859.0, + 1271.0, + 249.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 794.0, + 1863.0, + 815.0, + 1863.0, + 815.0, + 1894.0, + 794.0, + 1894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 388.0, + 1450.0, + 388.0, + 1450.0, + 426.0, + 250.0, + 426.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 422.0, + 1285.0, + 422.0, + 1285.0, + 454.0, + 251.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1306.0, + 422.0, + 1450.0, + 422.0, + 1450.0, + 454.0, + 1306.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 452.0, + 555.0, + 452.0, + 555.0, + 488.0, + 250.0, + 488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 609.0, + 452.0, + 820.0, + 452.0, + 820.0, + 488.0, + 609.0, + 488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 842.0, + 452.0, + 1349.0, + 452.0, + 1349.0, + 488.0, + 842.0, + 488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1438.0, + 452.0, + 1450.0, + 452.0, + 1450.0, + 488.0, + 1438.0, + 488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 479.0, + 325.0, + 479.0, + 325.0, + 519.0, + 248.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 345.0, + 479.0, + 1189.0, + 479.0, + 1189.0, + 519.0, + 345.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1440.0, + 479.0, + 1450.0, + 479.0, + 1450.0, + 519.0, + 1440.0, + 519.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 512.0, + 439.0, + 512.0, + 439.0, + 550.0, + 248.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 512.0, + 1093.0, + 512.0, + 1093.0, + 550.0, + 531.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1112.0, + 512.0, + 1128.0, + 512.0, + 1128.0, + 550.0, + 1112.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1183.0, + 512.0, + 1451.0, + 512.0, + 1451.0, + 550.0, + 1183.0, + 550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 543.0, + 393.0, + 543.0, + 393.0, + 580.0, + 249.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 871.0, + 315.0, + 871.0, + 315.0, + 912.0, + 246.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 871.0, + 1271.0, + 871.0, + 1271.0, + 912.0, + 405.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1293.0, + 871.0, + 1451.0, + 871.0, + 1451.0, + 912.0, + 1293.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 904.0, + 474.0, + 904.0, + 474.0, + 940.0, + 249.0, + 940.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 904.0, + 1450.0, + 904.0, + 1450.0, + 940.0, + 575.0, + 940.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 938.0, + 325.0, + 938.0, + 325.0, + 970.0, + 252.0, + 970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 938.0, + 447.0, + 938.0, + 447.0, + 970.0, + 415.0, + 970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 938.0, + 950.0, + 938.0, + 950.0, + 970.0, + 549.0, + 970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 938.0, + 1449.0, + 938.0, + 1449.0, + 970.0, + 1090.0, + 970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 964.0, + 706.0, + 964.0, + 706.0, + 1001.0, + 248.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 728.0, + 964.0, + 1117.0, + 964.0, + 1117.0, + 1001.0, + 728.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1370.0, + 964.0, + 1450.0, + 964.0, + 1450.0, + 1001.0, + 1370.0, + 1001.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 996.0, + 252.0, + 996.0, + 252.0, + 1032.0, + 248.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 397.0, + 996.0, + 448.0, + 996.0, + 448.0, + 1032.0, + 397.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 483.0, + 996.0, + 535.0, + 996.0, + 535.0, + 1032.0, + 483.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 569.0, + 996.0, + 1450.0, + 996.0, + 1450.0, + 1032.0, + 569.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1027.0, + 1438.0, + 1027.0, + 1438.0, + 1063.0, + 249.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1292.0, + 1452.0, + 1292.0, + 1452.0, + 1329.0, + 251.0, + 1329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1321.0, + 252.0, + 1321.0, + 252.0, + 1360.0, + 247.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 1321.0, + 1011.0, + 1321.0, + 1011.0, + 1360.0, + 342.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1102.0, + 1321.0, + 1373.0, + 1321.0, + 1373.0, + 1360.0, + 1102.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1406.0, + 1321.0, + 1452.0, + 1321.0, + 1452.0, + 1360.0, + 1406.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 245.0, + 1352.0, + 496.0, + 1352.0, + 496.0, + 1390.0, + 245.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 1352.0, + 817.0, + 1352.0, + 817.0, + 1390.0, + 516.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 1352.0, + 1452.0, + 1352.0, + 1452.0, + 1390.0, + 838.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1382.0, + 1126.0, + 1382.0, + 1126.0, + 1418.0, + 247.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1147.0, + 1382.0, + 1452.0, + 1382.0, + 1452.0, + 1418.0, + 1147.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1412.0, + 530.0, + 1412.0, + 530.0, + 1450.0, + 247.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 550.0, + 1412.0, + 1190.0, + 1412.0, + 1190.0, + 1450.0, + 550.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 585.0, + 431.0, + 585.0, + 431.0, + 637.0, + 246.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 585.0, + 1198.0, + 585.0, + 1198.0, + 637.0, + 451.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1291.0, + 585.0, + 1455.0, + 585.0, + 1455.0, + 637.0, + 1291.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 619.0, + 444.0, + 619.0, + 444.0, + 661.0, + 249.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 513.0, + 619.0, + 1451.0, + 619.0, + 1451.0, + 661.0, + 513.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 653.0, + 1451.0, + 653.0, + 1451.0, + 692.0, + 248.0, + 692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 686.0, + 562.0, + 686.0, + 562.0, + 718.0, + 251.0, + 718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1606.0, + 1449.0, + 1606.0, + 1449.0, + 1647.0, + 249.0, + 1647.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1640.0, + 302.0, + 1640.0, + 302.0, + 1678.0, + 250.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1640.0, + 597.0, + 1640.0, + 597.0, + 1678.0, + 394.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 617.0, + 1640.0, + 1209.0, + 1640.0, + 1209.0, + 1678.0, + 617.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 1640.0, + 1450.0, + 1640.0, + 1450.0, + 1678.0, + 1233.0, + 1678.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1669.0, + 904.0, + 1669.0, + 904.0, + 1708.0, + 249.0, + 1708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1718.0, + 1449.0, + 1718.0, + 1449.0, + 1752.0, + 251.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1747.0, + 1451.0, + 1747.0, + 1451.0, + 1783.0, + 249.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1779.0, + 785.0, + 1779.0, + 785.0, + 1814.0, + 251.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1457.0, + 1447.0, + 1457.0, + 1447.0, + 1496.0, + 250.0, + 1496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1488.0, + 348.0, + 1488.0, + 348.0, + 1532.0, + 248.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 366.0, + 1488.0, + 379.0, + 1488.0, + 379.0, + 1532.0, + 366.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1162.0, + 288.0, + 1162.0, + 288.0, + 1202.0, + 249.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 428.0, + 1162.0, + 1153.0, + 1162.0, + 1153.0, + 1202.0, + 428.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1175.0, + 1162.0, + 1224.0, + 1162.0, + 1224.0, + 1202.0, + 1175.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1242.0, + 1162.0, + 1253.0, + 1162.0, + 1253.0, + 1202.0, + 1242.0, + 1202.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 252, + 661, + 1448, + 661, + 1448, + 879, + 252, + 879 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 251, + 1432, + 1448, + 1432, + 1448, + 1628, + 251, + 1628 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 252, + 1200, + 1448, + 1200, + 1448, + 1324, + 252, + 1324 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 253, + 1640, + 1448, + 1640, + 1448, + 1765, + 253, + 1765 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 251, + 476, + 1450, + 476, + 1450, + 571, + 251, + 571 + ], + "score": 0.959 + }, + { + "category_id": 1, + "poly": [ + 252, + 1027, + 1448, + 1027, + 1448, + 1092, + 252, + 1092 + ], + "score": 0.954 + }, + { + "category_id": 1, + "poly": [ + 253, + 585, + 1445, + 585, + 1445, + 649, + 253, + 649 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 250, + 398, + 1447, + 398, + 1447, + 463, + 250, + 463 + ], + "score": 0.949 + }, + { + "category_id": 8, + "poly": [ + 552, + 1349, + 1146, + 1349, + 1146, + 1392, + 552, + 1392 + ], + "score": 0.939 + }, + { + "category_id": 1, + "poly": [ + 251, + 1778, + 985, + 1778, + 985, + 1812, + 251, + 1812 + ], + "score": 0.932 + }, + { + "category_id": 8, + "poly": [ + 641, + 949, + 1063, + 949, + 1063, + 1001, + 641, + 1001 + ], + "score": 0.927 + }, + { + "category_id": 0, + "poly": [ + 255, + 1139, + 584, + 1139, + 584, + 1172, + 255, + 1172 + ], + "score": 0.926 + }, + { + "category_id": 1, + "poly": [ + 251, + 891, + 1262, + 891, + 1262, + 927, + 251, + 927 + ], + "score": 0.926 + }, + { + "category_id": 2, + "poly": [ + 253, + 111, + 770, + 111, + 770, + 142, + 253, + 142 + ], + "score": 0.918 + }, + { + "category_id": 0, + "poly": [ + 255, + 338, + 589, + 338, + 589, + 371, + 255, + 371 + ], + "score": 0.913 + }, + { + "category_id": 0, + "poly": [ + 251, + 263, + 1225, + 263, + 1225, + 300, + 251, + 300 + ], + "score": 0.911 + }, + { + "category_id": 9, + "poly": [ + 1411, + 1353, + 1445, + 1353, + 1445, + 1383, + 1411, + 1383 + ], + "score": 0.887 + }, + { + "category_id": 9, + "poly": [ + 1411, + 953, + 1445, + 953, + 1445, + 983, + 1411, + 983 + ], + "score": 0.881 + }, + { + "category_id": 2, + "poly": [ + 796, + 1866, + 814, + 1866, + 814, + 1889, + 796, + 1889 + ], + "score": 0.799 + }, + { + "category_id": 13, + "poly": [ + 836, + 784, + 925, + 784, + 925, + 819, + 836, + 819 + ], + "score": 0.94, + "latex": "\\{ 0 , 1 \\} ^ { D }" + }, + { + "category_id": 13, + "poly": [ + 454, + 476, + 593, + 476, + 593, + 512, + 454, + 512 + ], + "score": 0.93, + "latex": "b \\in \\{ 0 , 1 \\} ^ { D }" + }, + { + "category_id": 13, + "poly": [ + 590, + 663, + 755, + 663, + 755, + 697, + 590, + 697 + ], + "score": 0.93, + "latex": "p _ { d } ( x _ { b } | x _ { 1 - b } , b )" + }, + { + "category_id": 13, + "poly": [ + 667, + 509, + 844, + 509, + 844, + 542, + 667, + 542 + ], + "score": 0.93, + "latex": "x _ { b } = \\{ x _ { i : b _ { i } = 1 } \\}" + }, + { + "category_id": 13, + "poly": [ + 523, + 401, + 590, + 401, + 590, + 434, + 523, + 434 + ], + "score": 0.93, + "latex": "p _ { d } ( x )" + }, + { + "category_id": 13, + "poly": [ + 890, + 586, + 1275, + 586, + 1275, + 620, + 890, + 620 + ], + "score": 0.92, + "latex": "p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b ) \\approx p _ { d } ( x _ { b } | x _ { 1 - b } , b )" + }, + { + "category_id": 13, + "poly": [ + 431, + 694, + 616, + 694, + 616, + 729, + 431, + 729 + ], + "score": 0.92, + "latex": "p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b )" + }, + { + "category_id": 13, + "poly": [ + 1012, + 785, + 1197, + 785, + 1197, + 819, + 1012, + 819 + ], + "score": 0.92, + "latex": "p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b )" + }, + { + "category_id": 13, + "poly": [ + 1257, + 1501, + 1445, + 1501, + 1445, + 1536, + 1257, + 1536 + ], + "score": 0.92, + "latex": "p _ { \\theta } ( x _ { i } | \\boldsymbol { z } , x _ { 1 - b } , b )" + }, + { + "category_id": 13, + "poly": [ + 330, + 1503, + 520, + 1503, + 520, + 1536, + 330, + 1536 + ], + "score": 0.92, + "latex": "p _ { \\theta } ( x _ { b } | \\boldsymbol { z } , x _ { 1 - b } , b )" + }, + { + "category_id": 13, + "poly": [ + 1046, + 1672, + 1282, + 1672, + 1282, + 1705, + 1046, + 1705 + ], + "score": 0.92, + "latex": "x _ { 1 - b } = x \\circ ( 1 - b )" + }, + { + "category_id": 13, + "poly": [ + 531, + 1231, + 756, + 1231, + 756, + 1265, + 531, + 1265 + ], + "score": 0.91, + "latex": "z \\sim p _ { \\psi } ( z | x _ { 1 - b } , b )" + }, + { + "category_id": 13, + "poly": [ + 696, + 755, + 746, + 755, + 746, + 788, + 696, + 788 + ], + "score": 0.91, + "latex": "p ( b )" + }, + { + "category_id": 13, + "poly": [ + 1161, + 724, + 1211, + 724, + 1211, + 758, + 1161, + 758 + ], + "score": 0.91, + "latex": "p ( b )" + }, + { + "category_id": 13, + "poly": [ + 1288, + 663, + 1355, + 663, + 1355, + 696, + 1288, + 696 + ], + "score": 0.91, + "latex": "p _ { d } ( x )" + }, + { + "category_id": 13, + "poly": [ + 599, + 1534, + 774, + 1534, + 774, + 1566, + 599, + 1566 + ], + "score": 0.91, + "latex": "w _ { i , \\theta } ( z , x _ { 1 - b } , b )" + }, + { + "category_id": 13, + "poly": [ + 350, + 1432, + 456, + 1432, + 456, + 1464, + 350, + 1464 + ], + "score": 0.91, + "latex": "z ~ \\in ~ \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 1019, + 509, + 1310, + 509, + 1310, + 544, + 1019, + 544 + ], + "score": 0.91, + "latex": "x _ { ( 0 , 1 , 1 , 0 , 1 ) } = ( x _ { 2 } , x _ { 3 } , x _ { 5 } )" + }, + { + "category_id": 14, + "poly": [ + 638, + 949, + 1064, + 949, + 1064, + 1002, + 638, + 1002 + ], + "score": 0.91, + "latex": "\\operatorname* { m a x } _ { \\psi , \\theta } \\mathbb { E } _ { p _ { d } ( \\boldsymbol { x } ) } \\mathbb { E } _ { p ( \\boldsymbol { b } ) } \\log p _ { \\psi , \\theta } \\big ( x _ { b } | \\boldsymbol { x } _ { 1 - \\boldsymbol { b } } , \\boldsymbol { b } \\big )" + }, + { + "category_id": 13, + "poly": [ + 253, + 1262, + 530, + 1262, + 530, + 1296, + 253, + 1296 + ], + "score": 0.91, + "latex": "x _ { b } ~ \\sim ~ p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b )" + }, + { + "category_id": 13, + "poly": [ + 326, + 893, + 375, + 893, + 375, + 927, + 326, + 927 + ], + "score": 0.9, + "latex": "p ( b )" + }, + { + "category_id": 13, + "poly": [ + 984, + 1029, + 1263, + 1029, + 1263, + 1063, + 984, + 1063 + ], + "score": 0.9, + "latex": "( b _ { i } = 1 \\forall i \\in \\{ 1 , \\ldots , D \\} )" + }, + { + "category_id": 13, + "poly": [ + 470, + 544, + 528, + 544, + 528, + 572, + 470, + 572 + ], + "score": 0.89, + "latex": "x _ { 1 - b }" + }, + { + "category_id": 13, + "poly": [ + 461, + 1601, + 519, + 1601, + 519, + 1627, + 461, + 1627 + ], + "score": 0.89, + "latex": "x _ { 1 - b }" + }, + { + "category_id": 13, + "poly": [ + 253, + 1564, + 681, + 1564, + 681, + 1597, + 253, + 1597 + ], + "score": 0.89, + "latex": "x _ { i } \\sim \\mathrm { C a t } [ \\mathrm { S o f t m a x } ( w _ { i , \\theta } ( z , x _ { 1 - b } , b ) ) ]" + }, + { + "category_id": 13, + "poly": [ + 493, + 1647, + 551, + 1647, + 551, + 1673, + 493, + 1673 + ], + "score": 0.88, + "latex": "x _ { 1 - b }" + }, + { + "category_id": 13, + "poly": [ + 387, + 1466, + 991, + 1466, + 991, + 1503, + 387, + 1503 + ], + "score": 0.87, + "latex": "p _ { \\psi } ( z | x _ { 1 - b } , b , \\psi ) = \\mathcal { N } ( z | \\mu _ { \\psi } ( x _ { 1 - b } , b ) , \\sigma _ { \\psi } ^ { 2 } ( x _ { 1 - b } , b ) I )" + }, + { + "category_id": 14, + "poly": [ + 554, + 1349, + 1145, + 1349, + 1145, + 1390, + 554, + 1390 + ], + "score": 0.86, + "latex": "p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b ) = \\mathbb { E } _ { z \\sim p _ { \\psi } ( z | x _ { 1 - b } , b ) } p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b )" + }, + { + "category_id": 13, + "poly": [ + 763, + 1705, + 821, + 1705, + 821, + 1735, + 763, + 1735 + ], + "score": 0.86, + "latex": "x _ { 1 - b }" + }, + { + "category_id": 13, + "poly": [ + 852, + 1598, + 882, + 1598, + 882, + 1625, + 852, + 1625 + ], + "score": 0.85, + "latex": "x _ { b }" + }, + { + "category_id": 13, + "poly": [ + 352, + 618, + 375, + 618, + 375, + 649, + 352, + 649 + ], + "score": 0.85, + "latex": "\\psi" + }, + { + "category_id": 13, + "poly": [ + 410, + 1647, + 441, + 1647, + 441, + 1672, + 410, + 1672 + ], + "score": 0.84, + "latex": "x _ { b }" + }, + { + "category_id": 13, + "poly": [ + 780, + 1441, + 814, + 1441, + 814, + 1468, + 780, + 1468 + ], + "score": 0.82, + "latex": "p _ { \\psi }" + }, + { + "category_id": 13, + "poly": [ + 679, + 402, + 705, + 402, + 705, + 428, + 679, + 428 + ], + "score": 0.81, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 426, + 618, + 443, + 618, + 443, + 644, + 426, + 644 + ], + "score": 0.79, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 570, + 1598, + 584, + 1598, + 584, + 1622, + 570, + 1622 + ], + "score": 0.77, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 937, + 408, + 956, + 408, + 956, + 428, + 937, + 428 + ], + "score": 0.76, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 1424, + 1596, + 1439, + 1596, + 1439, + 1622, + 1424, + 1622 + ], + "score": 0.75, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 349, + 1469, + 370, + 1469, + 370, + 1500, + 349, + 1500 + ], + "score": 0.75, + "latex": "\\psi" + }, + { + "category_id": 13, + "poly": [ + 1229, + 1570, + 1248, + 1570, + 1248, + 1591, + 1229, + 1591 + ], + "score": 0.75, + "latex": "z" + }, + { + "category_id": 13, + "poly": [ + 885, + 1442, + 902, + 1442, + 902, + 1462, + 885, + 1462 + ], + "score": 0.74, + "latex": "z" + }, + { + "category_id": 13, + "poly": [ + 254, + 619, + 267, + 619, + 267, + 644, + 254, + 644 + ], + "score": 0.74, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 1077, + 1505, + 1090, + 1505, + 1090, + 1530, + 1077, + 1530 + ], + "score": 0.72, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 1364, + 1677, + 1383, + 1677, + 1383, + 1699, + 1364, + 1699 + ], + "score": 0.71, + "latex": "\\circ" + }, + { + "category_id": 13, + "poly": [ + 528, + 1062, + 544, + 1062, + 544, + 1087, + 528, + 1087 + ], + "score": 0.7, + "latex": "^ { \\textit { b } }" + }, + { + "category_id": 13, + "poly": [ + 974, + 1643, + 988, + 1643, + 988, + 1669, + 974, + 1669 + ], + "score": 0.69, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 980, + 695, + 995, + 695, + 995, + 721, + 980, + 721 + ], + "score": 0.69, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 1286, + 1598, + 1373, + 1598, + 1373, + 1627, + 1286, + 1627 + ], + "score": 0.51, + "latex": "z , x _ { 1 - b }" + }, + { + "category_id": 13, + "poly": [ + 1288, + 1601, + 1305, + 1601, + 1305, + 1622, + 1288, + 1622 + ], + "score": 0.34, + "latex": "z" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1140.0, + 586.0, + 1140.0, + 586.0, + 1173.0, + 250.0, + 1173.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 109.0, + 772.0, + 109.0, + 772.0, + 145.0, + 251.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 339.0, + 590.0, + 339.0, + 590.0, + 372.0, + 251.0, + 372.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 261.0, + 1228.0, + 261.0, + 1228.0, + 304.0, + 247.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 793.0, + 1863.0, + 817.0, + 1863.0, + 817.0, + 1896.0, + 793.0, + 1896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 659.0, + 589.0, + 659.0, + 589.0, + 702.0, + 248.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 756.0, + 659.0, + 1287.0, + 659.0, + 1287.0, + 702.0, + 756.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1356.0, + 659.0, + 1451.0, + 659.0, + 1451.0, + 702.0, + 1356.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 692.0, + 430.0, + 692.0, + 430.0, + 732.0, + 248.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 617.0, + 692.0, + 979.0, + 692.0, + 979.0, + 732.0, + 617.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 996.0, + 692.0, + 1452.0, + 692.0, + 1452.0, + 732.0, + 996.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 723.0, + 1160.0, + 723.0, + 1160.0, + 759.0, + 249.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1212.0, + 723.0, + 1448.0, + 723.0, + 1448.0, + 759.0, + 1212.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 754.0, + 695.0, + 754.0, + 695.0, + 790.0, + 249.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 754.0, + 1449.0, + 754.0, + 1449.0, + 790.0, + 747.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 245.0, + 780.0, + 835.0, + 780.0, + 835.0, + 825.0, + 245.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 926.0, + 780.0, + 1011.0, + 780.0, + 1011.0, + 825.0, + 926.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 780.0, + 1451.0, + 780.0, + 1451.0, + 825.0, + 1198.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 813.0, + 1451.0, + 813.0, + 1451.0, + 851.0, + 248.0, + 851.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 845.0, + 457.0, + 845.0, + 457.0, + 881.0, + 249.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1428.0, + 349.0, + 1428.0, + 349.0, + 1471.0, + 250.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 1428.0, + 779.0, + 1428.0, + 779.0, + 1471.0, + 457.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 815.0, + 1428.0, + 884.0, + 1428.0, + 884.0, + 1471.0, + 815.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 903.0, + 1428.0, + 1451.0, + 1428.0, + 1451.0, + 1471.0, + 903.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1465.0, + 348.0, + 1465.0, + 348.0, + 1502.0, + 250.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 1465.0, + 386.0, + 1465.0, + 386.0, + 1502.0, + 371.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 1465.0, + 1449.0, + 1465.0, + 1449.0, + 1502.0, + 992.0, + 1502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1501.0, + 329.0, + 1501.0, + 329.0, + 1538.0, + 250.0, + 1538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 521.0, + 1501.0, + 1076.0, + 1501.0, + 1076.0, + 1538.0, + 521.0, + 1538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 1501.0, + 1256.0, + 1501.0, + 1256.0, + 1538.0, + 1091.0, + 1538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1531.0, + 598.0, + 1531.0, + 598.0, + 1572.0, + 247.0, + 1572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 775.0, + 1531.0, + 1451.0, + 1531.0, + 1451.0, + 1572.0, + 775.0, + 1572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1560.0, + 252.0, + 1560.0, + 252.0, + 1600.0, + 246.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 1560.0, + 1228.0, + 1560.0, + 1228.0, + 1600.0, + 682.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1249.0, + 1560.0, + 1449.0, + 1560.0, + 1449.0, + 1600.0, + 1249.0, + 1600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1592.0, + 460.0, + 1592.0, + 460.0, + 1633.0, + 248.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 520.0, + 1592.0, + 569.0, + 1592.0, + 569.0, + 1633.0, + 520.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 1592.0, + 851.0, + 1592.0, + 851.0, + 1633.0, + 585.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 1592.0, + 1285.0, + 1592.0, + 1285.0, + 1633.0, + 883.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1374.0, + 1592.0, + 1423.0, + 1592.0, + 1423.0, + 1633.0, + 1374.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1440.0, + 1592.0, + 1451.0, + 1592.0, + 1451.0, + 1633.0, + 1440.0, + 1633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1199.0, + 1449.0, + 1199.0, + 1449.0, + 1238.0, + 249.0, + 1238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1230.0, + 530.0, + 1230.0, + 530.0, + 1269.0, + 247.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 757.0, + 1230.0, + 1452.0, + 1230.0, + 1452.0, + 1269.0, + 757.0, + 1269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1261.0, + 252.0, + 1261.0, + 252.0, + 1300.0, + 248.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 1261.0, + 1451.0, + 1261.0, + 1451.0, + 1300.0, + 531.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1291.0, + 545.0, + 1291.0, + 545.0, + 1326.0, + 249.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1639.0, + 409.0, + 1639.0, + 409.0, + 1679.0, + 250.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 442.0, + 1639.0, + 492.0, + 1639.0, + 492.0, + 1679.0, + 442.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 1639.0, + 973.0, + 1639.0, + 973.0, + 1679.0, + 552.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 1639.0, + 1451.0, + 1639.0, + 1451.0, + 1679.0, + 989.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1672.0, + 1045.0, + 1672.0, + 1045.0, + 1707.0, + 250.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1283.0, + 1672.0, + 1363.0, + 1672.0, + 1363.0, + 1707.0, + 1283.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1384.0, + 1672.0, + 1451.0, + 1672.0, + 1451.0, + 1707.0, + 1384.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1698.0, + 762.0, + 1698.0, + 762.0, + 1739.0, + 248.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 822.0, + 1698.0, + 1452.0, + 1698.0, + 1452.0, + 1739.0, + 822.0, + 1739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1732.0, + 1223.0, + 1732.0, + 1223.0, + 1767.0, + 250.0, + 1767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 475.0, + 453.0, + 475.0, + 453.0, + 513.0, + 250.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 594.0, + 475.0, + 1449.0, + 475.0, + 1449.0, + 513.0, + 594.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 504.0, + 666.0, + 504.0, + 666.0, + 551.0, + 247.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 504.0, + 1018.0, + 504.0, + 1018.0, + 551.0, + 845.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1311.0, + 504.0, + 1453.0, + 504.0, + 1453.0, + 551.0, + 1311.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 536.0, + 469.0, + 536.0, + 469.0, + 574.0, + 248.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 529.0, + 536.0, + 895.0, + 536.0, + 895.0, + 574.0, + 529.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 1027.0, + 983.0, + 1027.0, + 983.0, + 1063.0, + 252.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1264.0, + 1027.0, + 1449.0, + 1027.0, + 1449.0, + 1063.0, + 1264.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1058.0, + 527.0, + 1058.0, + 527.0, + 1094.0, + 251.0, + 1094.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 1058.0, + 687.0, + 1058.0, + 687.0, + 1094.0, + 545.0, + 1094.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 582.0, + 889.0, + 582.0, + 889.0, + 623.0, + 249.0, + 623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1276.0, + 582.0, + 1449.0, + 582.0, + 1449.0, + 623.0, + 1276.0, + 623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 612.0, + 253.0, + 612.0, + 253.0, + 654.0, + 247.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 268.0, + 612.0, + 351.0, + 612.0, + 351.0, + 654.0, + 268.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 612.0, + 425.0, + 612.0, + 425.0, + 654.0, + 376.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 444.0, + 612.0, + 1130.0, + 612.0, + 1130.0, + 654.0, + 444.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 397.0, + 522.0, + 397.0, + 522.0, + 436.0, + 249.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 591.0, + 397.0, + 678.0, + 397.0, + 678.0, + 436.0, + 591.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 397.0, + 936.0, + 397.0, + 936.0, + 436.0, + 706.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 957.0, + 397.0, + 1448.0, + 397.0, + 1448.0, + 436.0, + 957.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 433.0, + 754.0, + 433.0, + 754.0, + 465.0, + 251.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1775.0, + 990.0, + 1775.0, + 990.0, + 1816.0, + 250.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 890.0, + 325.0, + 890.0, + 325.0, + 929.0, + 251.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 376.0, + 890.0, + 1266.0, + 890.0, + 1266.0, + 929.0, + 376.0, + 929.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 252, + 856, + 1449, + 856, + 1449, + 1013, + 252, + 1013 + ], + "score": 0.981 + }, + { + "category_id": 1, + "poly": [ + 252, + 1445, + 1448, + 1445, + 1448, + 1568, + 252, + 1568 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 253, + 1307, + 1449, + 1307, + 1449, + 1433, + 253, + 1433 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 252, + 1097, + 1447, + 1097, + 1447, + 1216, + 252, + 1216 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 254, + 1670, + 1448, + 1670, + 1448, + 1765, + 254, + 1765 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 254, + 700, + 1449, + 700, + 1449, + 766, + 254, + 766 + ], + "score": 0.957 + }, + { + "category_id": 8, + "poly": [ + 279, + 442, + 1386, + 442, + 1386, + 562, + 279, + 562 + ], + "score": 0.956 + }, + { + "category_id": 8, + "poly": [ + 537, + 1568, + 1165, + 1568, + 1165, + 1643, + 537, + 1643 + ], + "score": 0.948 + }, + { + "category_id": 8, + "poly": [ + 628, + 628, + 1076, + 628, + 1076, + 682, + 628, + 682 + ], + "score": 0.942 + }, + { + "category_id": 0, + "poly": [ + 254, + 1252, + 583, + 1252, + 583, + 1285, + 254, + 1285 + ], + "score": 0.932 + }, + { + "category_id": 2, + "poly": [ + 253, + 111, + 771, + 111, + 771, + 142, + 253, + 142 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 253, + 1778, + 878, + 1778, + 878, + 1813, + 253, + 1813 + ], + "score": 0.926 + }, + { + "category_id": 1, + "poly": [ + 251, + 575, + 1142, + 575, + 1142, + 610, + 251, + 610 + ], + "score": 0.924 + }, + { + "category_id": 1, + "poly": [ + 251, + 379, + 1192, + 379, + 1192, + 416, + 251, + 416 + ], + "score": 0.922 + }, + { + "category_id": 0, + "poly": [ + 255, + 801, + 642, + 801, + 642, + 833, + 255, + 833 + ], + "score": 0.922 + }, + { + "category_id": 8, + "poly": [ + 445, + 1033, + 1251, + 1033, + 1251, + 1077, + 445, + 1077 + ], + "score": 0.912 + }, + { + "category_id": 9, + "poly": [ + 1410, + 632, + 1445, + 632, + 1445, + 663, + 1410, + 663 + ], + "score": 0.885 + }, + { + "category_id": 0, + "poly": [ + 254, + 324, + 704, + 324, + 704, + 358, + 254, + 358 + ], + "score": 0.884 + }, + { + "category_id": 9, + "poly": [ + 1411, + 1576, + 1445, + 1576, + 1445, + 1607, + 1411, + 1607 + ], + "score": 0.881 + }, + { + "category_id": 9, + "poly": [ + 1411, + 1039, + 1445, + 1039, + 1445, + 1070, + 1411, + 1070 + ], + "score": 0.863 + }, + { + "category_id": 9, + "poly": [ + 1411, + 525, + 1446, + 525, + 1446, + 555, + 1411, + 555 + ], + "score": 0.849 + }, + { + "category_id": 0, + "poly": [ + 248, + 266, + 1222, + 266, + 1222, + 300, + 248, + 300 + ], + "score": 0.828 + }, + { + "category_id": 2, + "poly": [ + 796, + 1865, + 813, + 1865, + 813, + 1890, + 796, + 1890 + ], + "score": 0.813 + }, + { + "category_id": 14, + "poly": [ + 279, + 442, + 1388, + 442, + 1388, + 565, + 279, + 565 + ], + "score": 0.95, + "latex": "\\begin{array} { r l } & { \\log p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b ) = \\mathbb { E } _ { q _ { \\phi } ( z | x , b ) } \\log \\frac { p _ { \\psi , \\theta } ( x _ { b } , z | x _ { 1 - b } , b ) } { q _ { \\phi } ( z | x , b ) } + D _ { \\mathrm { K L } } ( q _ { \\phi } ( z | x , b ) | | p _ { \\psi , \\theta } ( z | x , b ) ) } \\\\ & { \\qquad \\geq \\mathbb { E } _ { q _ { \\phi } ( z | x , b ) } \\log p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b ) - D _ { \\mathrm { K L } } ( q _ { \\phi } ( z | x , b ) | | p _ { \\psi } ( z | x _ { 1 - b } , b ) ) = L _ { V A E A C } ( x , b ; \\theta , \\psi , \\phi ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 349, + 981, + 504, + 981, + 504, + 1014, + 349, + 1014 + ], + "score": 0.93, + "latex": "p _ { \\psi } ( z | x _ { 1 - b } , b )" + }, + { + "category_id": 13, + "poly": [ + 565, + 1096, + 633, + 1096, + 633, + 1153, + 565, + 1153 + ], + "score": 0.92, + "latex": "- \\frac { \\mu _ { \\psi } ^ { 2 } } { 2 \\sigma _ { \\mu } ^ { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 253, + 1508, + 326, + 1508, + 326, + 1542, + 253, + 1542 + ], + "score": 0.92, + "latex": "p ( b | x )" + }, + { + "category_id": 14, + "poly": [ + 626, + 628, + 1078, + 628, + 1078, + 681, + 626, + 681 + ], + "score": 0.92, + "latex": "\\operatorname* { m a x } _ { \\theta , \\psi , \\phi } \\mathbb { E } _ { p _ { d } ( x ) } \\mathbb { E } _ { p ( b ) } L _ { V A E A C } ( x , b ; \\theta , \\psi , \\phi )" + }, + { + "category_id": 13, + "poly": [ + 287, + 922, + 341, + 922, + 341, + 952, + 287, + 952 + ], + "score": 0.91, + "latex": "\\left. z \\right. _ { 2 }" + }, + { + "category_id": 14, + "poly": [ + 538, + 1563, + 1165, + 1563, + 1165, + 1643, + 538, + 1643 + ], + "score": 0.91, + "latex": "\\log p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b ) = \\sum _ { \\substack { i : b _ { i } = 1 , x _ { i } \\neq \\omega } } \\log p _ { \\theta } ( x _ { i } | z , x _ { 1 - b } , b )" + }, + { + "category_id": 13, + "poly": [ + 686, + 1107, + 898, + 1107, + 898, + 1145, + 686, + 1145 + ], + "score": 0.91, + "latex": "\\sigma _ { \\sigma } ( \\log ( \\sigma _ { \\psi } ) - \\sigma _ { \\psi } )" + }, + { + "category_id": 13, + "poly": [ + 734, + 1448, + 948, + 1448, + 948, + 1478, + 734, + 1478 + ], + "score": 0.91, + "latex": "x _ { i } = \\omega \\Rightarrow b _ { i } = 1" + }, + { + "category_id": 13, + "poly": [ + 620, + 381, + 846, + 381, + 846, + 416, + 620, + 416 + ], + "score": 0.91, + "latex": "\\log p _ { \\psi , \\theta } ( x _ { b } | x _ { 1 - b } , b )" + }, + { + "category_id": 14, + "poly": [ + 448, + 1032, + 1252, + 1032, + 1252, + 1077, + 448, + 1077 + ], + "score": 0.89, + "latex": "p _ { \\psi } ( z , \\mu _ { \\psi } , \\sigma _ { \\psi } | x _ { 1 - b } , b ) = \\mathcal { N } ( z | \\mu _ { \\psi } , \\sigma _ { \\psi } ^ { 2 } ) \\mathcal { N } ( \\mu _ { \\psi } | 0 , \\sigma _ { \\mu } ) \\mathrm { G a m m a } ( \\sigma _ { \\psi } | 2 , \\sigma _ { \\sigma } )" + }, + { + "category_id": 13, + "poly": [ + 972, + 864, + 1008, + 864, + 1008, + 892, + 972, + 892 + ], + "score": 0.87, + "latex": "\\sigma _ { \\psi }" + }, + { + "category_id": 13, + "poly": [ + 883, + 865, + 918, + 865, + 918, + 892, + 883, + 892 + ], + "score": 0.86, + "latex": "\\mu _ { \\psi }" + }, + { + "category_id": 13, + "poly": [ + 1171, + 1150, + 1245, + 1150, + 1245, + 1184, + 1171, + 1184 + ], + "score": 0.86, + "latex": "( 1 0 ^ { - 4 } )" + }, + { + "category_id": 13, + "poly": [ + 717, + 1156, + 750, + 1156, + 750, + 1183, + 717, + 1183 + ], + "score": 0.86, + "latex": "\\sigma _ { \\sigma }" + }, + { + "category_id": 13, + "poly": [ + 877, + 707, + 908, + 707, + 908, + 736, + 877, + 736 + ], + "score": 0.86, + "latex": "q _ { \\phi }" + }, + { + "category_id": 13, + "poly": [ + 608, + 1152, + 667, + 1152, + 667, + 1184, + 608, + 1184 + ], + "score": 0.86, + "latex": "( 1 0 ^ { 4 } )" + }, + { + "category_id": 13, + "poly": [ + 343, + 1157, + 376, + 1157, + 376, + 1187, + 343, + 1187 + ], + "score": 0.84, + "latex": "\\sigma _ { \\mu }" + }, + { + "category_id": 13, + "poly": [ + 632, + 1745, + 649, + 1745, + 649, + 1760, + 632, + 1760 + ], + "score": 0.8, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 1036, + 1453, + 1059, + 1453, + 1059, + 1476, + 1036, + 1476 + ], + "score": 0.77, + "latex": "\\omega" + }, + { + "category_id": 13, + "poly": [ + 491, + 1734, + 506, + 1734, + 506, + 1760, + 491, + 1760 + ], + "score": 0.76, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 1316, + 865, + 1334, + 865, + 1334, + 886, + 1316, + 886 + ], + "score": 0.7, + "latex": "z" + }, + { + "category_id": 13, + "poly": [ + 1280, + 1478, + 1331, + 1478, + 1331, + 1511, + 1280, + 1511 + ], + "score": 0.7, + "latex": "p ( b )" + }, + { + "category_id": 13, + "poly": [ + 1247, + 1482, + 1265, + 1482, + 1265, + 1505, + 1247, + 1505 + ], + "score": 0.35, + "latex": "x" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1253.0, + 583.0, + 1253.0, + 583.0, + 1286.0, + 250.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 109.0, + 773.0, + 109.0, + 773.0, + 145.0, + 251.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 798.0, + 644.0, + 798.0, + 644.0, + 837.0, + 249.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 323.0, + 705.0, + 323.0, + 705.0, + 360.0, + 249.0, + 360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 265.0, + 1225.0, + 265.0, + 1225.0, + 301.0, + 249.0, + 301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 794.0, + 1864.0, + 815.0, + 1864.0, + 815.0, + 1895.0, + 794.0, + 1895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 857.0, + 882.0, + 857.0, + 882.0, + 894.0, + 249.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 857.0, + 971.0, + 857.0, + 971.0, + 894.0, + 919.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1009.0, + 857.0, + 1315.0, + 857.0, + 1315.0, + 894.0, + 1009.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1335.0, + 857.0, + 1452.0, + 857.0, + 1452.0, + 894.0, + 1335.0, + 894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 889.0, + 1449.0, + 889.0, + 1449.0, + 922.0, + 249.0, + 922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 917.0, + 286.0, + 917.0, + 286.0, + 955.0, + 248.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 342.0, + 917.0, + 1453.0, + 917.0, + 1453.0, + 955.0, + 342.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 945.0, + 1450.0, + 945.0, + 1450.0, + 987.0, + 247.0, + 987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 976.0, + 348.0, + 976.0, + 348.0, + 1018.0, + 247.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 976.0, + 637.0, + 976.0, + 637.0, + 1018.0, + 505.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1445.0, + 733.0, + 1445.0, + 733.0, + 1484.0, + 249.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 949.0, + 1445.0, + 1035.0, + 1445.0, + 1035.0, + 1484.0, + 949.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1060.0, + 1445.0, + 1451.0, + 1445.0, + 1451.0, + 1484.0, + 1060.0, + 1484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1477.0, + 1246.0, + 1477.0, + 1246.0, + 1512.0, + 249.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1266.0, + 1477.0, + 1279.0, + 1477.0, + 1279.0, + 1512.0, + 1266.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1332.0, + 1477.0, + 1449.0, + 1477.0, + 1449.0, + 1512.0, + 1332.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1506.0, + 252.0, + 1506.0, + 252.0, + 1546.0, + 248.0, + 1546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 327.0, + 1506.0, + 1449.0, + 1506.0, + 1449.0, + 1546.0, + 327.0, + 1546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1539.0, + 365.0, + 1539.0, + 365.0, + 1571.0, + 249.0, + 1571.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1306.0, + 1448.0, + 1306.0, + 1448.0, + 1342.0, + 250.0, + 1342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1339.0, + 1449.0, + 1339.0, + 1449.0, + 1373.0, + 249.0, + 1373.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1367.0, + 1449.0, + 1367.0, + 1449.0, + 1405.0, + 249.0, + 1405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1400.0, + 1279.0, + 1400.0, + 1279.0, + 1435.0, + 250.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1106.0, + 564.0, + 1106.0, + 564.0, + 1144.0, + 251.0, + 1144.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.0, + 1089.0, + 685.0, + 1089.0, + 685.0, + 1163.0, + 634.0, + 1163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 1089.0, + 1447.0, + 1089.0, + 1447.0, + 1163.0, + 899.0, + 1163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1151.0, + 342.0, + 1151.0, + 342.0, + 1187.0, + 248.0, + 1187.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 377.0, + 1151.0, + 607.0, + 1151.0, + 607.0, + 1187.0, + 377.0, + 1187.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 668.0, + 1151.0, + 716.0, + 1151.0, + 716.0, + 1187.0, + 668.0, + 1187.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 1151.0, + 1170.0, + 1151.0, + 1170.0, + 1187.0, + 751.0, + 1187.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1246.0, + 1151.0, + 1451.0, + 1151.0, + 1451.0, + 1187.0, + 1246.0, + 1187.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1180.0, + 1168.0, + 1180.0, + 1168.0, + 1218.0, + 247.0, + 1218.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1667.0, + 1451.0, + 1667.0, + 1451.0, + 1708.0, + 250.0, + 1708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1701.0, + 1448.0, + 1701.0, + 1448.0, + 1737.0, + 248.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1730.0, + 490.0, + 1730.0, + 490.0, + 1769.0, + 249.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 507.0, + 1730.0, + 631.0, + 1730.0, + 631.0, + 1769.0, + 507.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 650.0, + 1730.0, + 662.0, + 1730.0, + 662.0, + 1769.0, + 650.0, + 1769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 701.0, + 876.0, + 701.0, + 876.0, + 737.0, + 251.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 701.0, + 1449.0, + 701.0, + 1449.0, + 737.0, + 909.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 732.0, + 1036.0, + 732.0, + 1036.0, + 767.0, + 250.0, + 767.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1777.0, + 883.0, + 1777.0, + 883.0, + 1814.0, + 250.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 570.0, + 1145.0, + 570.0, + 1145.0, + 618.0, + 248.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 378.0, + 619.0, + 378.0, + 619.0, + 418.0, + 251.0, + 418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 378.0, + 1195.0, + 378.0, + 1195.0, + 418.0, + 847.0, + 418.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 253, + 628, + 1448, + 628, + 1448, + 965, + 253, + 965 + ], + "score": 0.982 + }, + { + "category_id": 5, + "poly": [ + 255, + 314, + 1460, + 314, + 1460, + 476, + 255, + 476 + ], + "score": 0.976, + "html": "
Method/DatasetWhiteWineYeastMushroomZooPhishing
MICE0.964± 0.0071.01 ± 0.010.334± 0.0020.19±0.030.422± 0.006
MissForest0.878 ± 0.0091.02 ± 0.060.249 ± 0.0060.16 ±0.020.422 ± 0.009
GAIN0.97 ± 0.020.99 ± 0.030.271 ± 0.0030.20± 0.020.427 ± 0.010
VAEAC0.850 ± 0.0070.94 ± 0.010.244 ± 0.0020.16 ± 0.020.394± 0.006
" + }, + { + "category_id": 1, + "poly": [ + 254, + 1226, + 1447, + 1226, + 1447, + 1379, + 254, + 1379 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 254, + 1087, + 1448, + 1087, + 1448, + 1209, + 254, + 1209 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 255, + 1579, + 1448, + 1579, + 1448, + 1703, + 255, + 1703 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 254, + 1393, + 1446, + 1393, + 1446, + 1518, + 254, + 1518 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 256, + 1718, + 1447, + 1718, + 1447, + 1811, + 256, + 1811 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 253, + 1533, + 1143, + 1533, + 1143, + 1565, + 253, + 1565 + ], + "score": 0.913 + }, + { + "category_id": 2, + "poly": [ + 254, + 112, + 770, + 112, + 770, + 141, + 254, + 141 + ], + "score": 0.91 + }, + { + "category_id": 0, + "poly": [ + 254, + 547, + 512, + 547, + 512, + 582, + 254, + 582 + ], + "score": 0.907 + }, + { + "category_id": 0, + "poly": [ + 257, + 1021, + 712, + 1021, + 712, + 1053, + 257, + 1053 + ], + "score": 0.887 + }, + { + "category_id": 6, + "poly": [ + 272, + 259, + 1422, + 259, + 1422, + 294, + 272, + 294 + ], + "score": 0.861 + }, + { + "category_id": 2, + "poly": [ + 795, + 1866, + 814, + 1866, + 814, + 1889, + 795, + 1889 + ], + "score": 0.799 + }, + { + "category_id": 13, + "poly": [ + 674, + 1427, + 766, + 1427, + 766, + 1454, + 674, + 1454 + ], + "score": 0.89, + "latex": "n = 1 0" + }, + { + "category_id": 13, + "poly": [ + 743, + 1487, + 828, + 1487, + 828, + 1515, + 743, + 1515 + ], + "score": 0.89, + "latex": "n = 1 0" + }, + { + "category_id": 13, + "poly": [ + 843, + 1121, + 942, + 1121, + 942, + 1149, + 843, + 1149 + ], + "score": 0.88, + "latex": "x _ { i } ~ = ~ \\omega" + }, + { + "category_id": 13, + "poly": [ + 1060, + 1257, + 1114, + 1257, + 1114, + 1286, + 1060, + 1286 + ], + "score": 0.87, + "latex": "50 \\%" + }, + { + "category_id": 13, + "poly": [ + 1091, + 1089, + 1118, + 1089, + 1118, + 1115, + 1091, + 1115 + ], + "score": 0.79, + "latex": "D" + }, + { + "category_id": 13, + "poly": [ + 1220, + 1124, + 1239, + 1124, + 1239, + 1151, + 1220, + 1151 + ], + "score": 0.78, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 1352, + 1094, + 1371, + 1094, + 1371, + 1115, + 1352, + 1115 + ], + "score": 0.72, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 397, + 1462, + 418, + 1462, + 418, + 1483, + 397, + 1483 + ], + "score": 0.62, + "latex": "n" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 109.0, + 772.0, + 109.0, + 772.0, + 145.0, + 251.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 544.0, + 516.0, + 544.0, + 516.0, + 588.0, + 247.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1021.0, + 717.0, + 1021.0, + 717.0, + 1057.0, + 250.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 274.0, + 255.0, + 1424.0, + 255.0, + 1424.0, + 298.0, + 274.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 1864.0, + 817.0, + 1864.0, + 817.0, + 1895.0, + 795.0, + 1895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 630.0, + 1452.0, + 630.0, + 1452.0, + 664.0, + 249.0, + 664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 660.0, + 1451.0, + 660.0, + 1451.0, + 694.0, + 249.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 690.0, + 1451.0, + 690.0, + 1451.0, + 724.0, + 249.0, + 724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 723.0, + 1448.0, + 723.0, + 1448.0, + 753.0, + 250.0, + 753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 749.0, + 1453.0, + 749.0, + 1453.0, + 787.0, + 246.0, + 787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 780.0, + 1451.0, + 780.0, + 1451.0, + 817.0, + 246.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 810.0, + 1452.0, + 810.0, + 1452.0, + 847.0, + 246.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 843.0, + 1451.0, + 843.0, + 1451.0, + 877.0, + 249.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 872.0, + 1449.0, + 872.0, + 1449.0, + 908.0, + 247.0, + 908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 904.0, + 1445.0, + 904.0, + 1445.0, + 939.0, + 249.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 937.0, + 347.0, + 937.0, + 347.0, + 964.0, + 250.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1225.0, + 1451.0, + 1225.0, + 1451.0, + 1262.0, + 248.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1255.0, + 1059.0, + 1255.0, + 1059.0, + 1292.0, + 250.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1115.0, + 1255.0, + 1451.0, + 1255.0, + 1451.0, + 1292.0, + 1115.0, + 1292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1285.0, + 1451.0, + 1285.0, + 1451.0, + 1321.0, + 246.0, + 1321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1316.0, + 1453.0, + 1316.0, + 1453.0, + 1351.0, + 248.0, + 1351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1346.0, + 965.0, + 1346.0, + 965.0, + 1383.0, + 248.0, + 1383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1087.0, + 1090.0, + 1087.0, + 1090.0, + 1122.0, + 251.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1119.0, + 1087.0, + 1351.0, + 1087.0, + 1351.0, + 1122.0, + 1119.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1372.0, + 1087.0, + 1449.0, + 1087.0, + 1449.0, + 1122.0, + 1372.0, + 1122.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1117.0, + 842.0, + 1117.0, + 842.0, + 1152.0, + 250.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 943.0, + 1117.0, + 1219.0, + 1117.0, + 1219.0, + 1152.0, + 943.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1240.0, + 1117.0, + 1449.0, + 1117.0, + 1449.0, + 1152.0, + 1240.0, + 1152.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1145.0, + 1450.0, + 1145.0, + 1450.0, + 1187.0, + 249.0, + 1187.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1177.0, + 843.0, + 1177.0, + 843.0, + 1215.0, + 249.0, + 1215.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1579.0, + 1451.0, + 1579.0, + 1451.0, + 1614.0, + 248.0, + 1614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1610.0, + 1448.0, + 1610.0, + 1448.0, + 1644.0, + 248.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1640.0, + 1448.0, + 1640.0, + 1448.0, + 1675.0, + 251.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1672.0, + 994.0, + 1672.0, + 994.0, + 1707.0, + 251.0, + 1707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1392.0, + 1452.0, + 1392.0, + 1452.0, + 1432.0, + 250.0, + 1432.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1425.0, + 673.0, + 1425.0, + 673.0, + 1460.0, + 250.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 767.0, + 1425.0, + 1450.0, + 1425.0, + 1450.0, + 1460.0, + 767.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1454.0, + 396.0, + 1454.0, + 396.0, + 1489.0, + 251.0, + 1489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 419.0, + 1454.0, + 1449.0, + 1454.0, + 1449.0, + 1489.0, + 419.0, + 1489.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1486.0, + 742.0, + 1486.0, + 742.0, + 1522.0, + 250.0, + 1522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 1486.0, + 1450.0, + 1486.0, + 1450.0, + 1522.0, + 829.0, + 1522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1712.0, + 1449.0, + 1712.0, + 1449.0, + 1759.0, + 249.0, + 1759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 1748.0, + 1449.0, + 1748.0, + 1449.0, + 1786.0, + 252.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1777.0, + 1392.0, + 1777.0, + 1392.0, + 1816.0, + 249.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1531.0, + 1149.0, + 1531.0, + 1149.0, + 1570.0, + 249.0, + 1570.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 252, + 1366, + 1448, + 1366, + 1448, + 1612, + 252, + 1612 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 253, + 1626, + 1448, + 1626, + 1448, + 1811, + 253, + 1811 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 253, + 843, + 1447, + 843, + 1447, + 997, + 253, + 997 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 254, + 575, + 1448, + 575, + 1448, + 731, + 254, + 731 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 254, + 1089, + 1448, + 1089, + 1448, + 1244, + 254, + 1244 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 254, + 1259, + 1448, + 1259, + 1448, + 1352, + 254, + 1352 + ], + "score": 0.975 + }, + { + "category_id": 5, + "poly": [ + 270, + 346, + 1428, + 346, + 1428, + 509, + 270, + 509 + ], + "score": 0.973, + "html": "
Method /DatasetWhiteWineYeastMushroomZ00Phishing
MICE0.13±0.020.41 ±0.020.92± 0.010.78± 0.050.75 ±0.02
MissForest0.17 ± 0.010.42 ± 0.020.972 ± 0.0030.71 ± 0.070.73 ± 0.02
GAIN0.11 ± 0.010.39 ± 0.060.969 ± 0.0050.67 ± 0.060.74 ± 0.03
VAEAC0.17 ± 0.010.43 ± 0.010.983 ± 0.0020.8 ± 0.10.74 ± 0.02
" + }, + { + "category_id": 1, + "poly": [ + 250, + 1012, + 1443, + 1012, + 1443, + 1076, + 250, + 1076 + ], + "score": 0.945 + }, + { + "category_id": 6, + "poly": [ + 245, + 260, + 1444, + 260, + 1444, + 324, + 245, + 324 + ], + "score": 0.919 + }, + { + "category_id": 2, + "poly": [ + 254, + 112, + 770, + 112, + 770, + 142, + 254, + 142 + ], + "score": 0.911 + }, + { + "category_id": 0, + "poly": [ + 256, + 781, + 550, + 781, + 550, + 812, + 256, + 812 + ], + "score": 0.909 + }, + { + "category_id": 2, + "poly": [ + 797, + 1865, + 813, + 1865, + 813, + 1888, + 797, + 1888 + ], + "score": 0.781 + }, + { + "category_id": 13, + "poly": [ + 329, + 1043, + 486, + 1043, + 486, + 1078, + 329, + 1078 + ], + "score": 0.93, + "latex": "p ( b | x ) = p ( b )" + }, + { + "category_id": 13, + "poly": [ + 795, + 1152, + 985, + 1152, + 985, + 1185, + 795, + 1185 + ], + "score": 0.91, + "latex": "p _ { \\theta } ( x _ { b } | \\boldsymbol { z } , x _ { 1 - b } , b )" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 256.0, + 1448.0, + 256.0, + 1448.0, + 297.0, + 248.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 289.0, + 628.0, + 289.0, + 628.0, + 327.0, + 249.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 110.0, + 772.0, + 110.0, + 772.0, + 145.0, + 250.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 778.0, + 556.0, + 778.0, + 556.0, + 815.0, + 248.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 1863.0, + 815.0, + 1863.0, + 815.0, + 1896.0, + 795.0, + 1896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1366.0, + 1449.0, + 1366.0, + 1449.0, + 1403.0, + 249.0, + 1403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1396.0, + 1452.0, + 1396.0, + 1452.0, + 1435.0, + 248.0, + 1435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1425.0, + 1449.0, + 1425.0, + 1449.0, + 1466.0, + 247.0, + 1466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1455.0, + 1452.0, + 1455.0, + 1452.0, + 1494.0, + 247.0, + 1494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1489.0, + 1451.0, + 1489.0, + 1451.0, + 1523.0, + 249.0, + 1523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1520.0, + 1451.0, + 1520.0, + 1451.0, + 1554.0, + 249.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1550.0, + 1451.0, + 1550.0, + 1451.0, + 1584.0, + 249.0, + 1584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1578.0, + 448.0, + 1578.0, + 448.0, + 1612.0, + 248.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1626.0, + 1451.0, + 1626.0, + 1451.0, + 1663.0, + 249.0, + 1663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1655.0, + 1452.0, + 1655.0, + 1452.0, + 1692.0, + 248.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1688.0, + 1452.0, + 1688.0, + 1452.0, + 1724.0, + 249.0, + 1724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1718.0, + 1452.0, + 1718.0, + 1452.0, + 1754.0, + 249.0, + 1754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1749.0, + 1449.0, + 1749.0, + 1449.0, + 1785.0, + 248.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1778.0, + 1452.0, + 1778.0, + 1452.0, + 1814.0, + 248.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 845.0, + 1451.0, + 845.0, + 1451.0, + 878.0, + 250.0, + 878.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 871.0, + 1451.0, + 871.0, + 1451.0, + 909.0, + 248.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 904.0, + 1448.0, + 904.0, + 1448.0, + 941.0, + 249.0, + 941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 935.0, + 1452.0, + 935.0, + 1452.0, + 972.0, + 247.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 964.0, + 920.0, + 964.0, + 920.0, + 1003.0, + 248.0, + 1003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 573.0, + 1451.0, + 573.0, + 1451.0, + 613.0, + 249.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 608.0, + 1449.0, + 608.0, + 1449.0, + 641.0, + 250.0, + 641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 638.0, + 1451.0, + 638.0, + 1451.0, + 672.0, + 250.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 668.0, + 1451.0, + 668.0, + 1451.0, + 701.0, + 250.0, + 701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 698.0, + 1192.0, + 698.0, + 1192.0, + 732.0, + 250.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1091.0, + 1449.0, + 1091.0, + 1449.0, + 1124.0, + 251.0, + 1124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1117.0, + 1452.0, + 1117.0, + 1452.0, + 1156.0, + 247.0, + 1156.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1151.0, + 794.0, + 1151.0, + 794.0, + 1188.0, + 248.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 1151.0, + 1452.0, + 1151.0, + 1452.0, + 1188.0, + 986.0, + 1188.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1182.0, + 1449.0, + 1182.0, + 1449.0, + 1214.0, + 248.0, + 1214.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1210.0, + 773.0, + 1210.0, + 773.0, + 1250.0, + 249.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1257.0, + 1449.0, + 1257.0, + 1449.0, + 1295.0, + 250.0, + 1295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1287.0, + 1448.0, + 1287.0, + 1448.0, + 1325.0, + 250.0, + 1325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1321.0, + 1007.0, + 1321.0, + 1007.0, + 1354.0, + 251.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1007.0, + 1450.0, + 1007.0, + 1450.0, + 1050.0, + 247.0, + 1050.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1041.0, + 328.0, + 1041.0, + 328.0, + 1079.0, + 248.0, + 1079.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 487.0, + 1041.0, + 498.0, + 1041.0, + 498.0, + 1079.0, + 487.0, + 1079.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 252, + 1239, + 1450, + 1239, + 1450, + 1334, + 252, + 1334 + ], + "score": 0.975 + }, + { + "category_id": 5, + "poly": [ + 502, + 361, + 1197, + 361, + 1197, + 524, + 502, + 524 + ], + "score": 0.975, + "html": "
Method/MasksCenterPatternRandomHalf
Context Encoder 121.319.220.615.5
SIIDGM 119.417.422.813.7
VAEAC, 1 sample22.121.429.314.9
VAEAC,10 samples23.723.329.317.4
" + }, + { + "category_id": 1, + "poly": [ + 252, + 1498, + 1448, + 1498, + 1448, + 1623, + 252, + 1623 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 253, + 1637, + 1449, + 1637, + 1449, + 1730, + 253, + 1730 + ], + "score": 0.972 + }, + { + "category_id": 5, + "poly": [ + 470, + 664, + 1228, + 664, + 1228, + 830, + 470, + 830 + ], + "score": 0.971, + "html": "
Method/Masks010203040506
Context Encoder218.618.417.919.019.119.3
GFC ²20.019.818.819.719.520.2
VAEAC,1 sample20.821.019.520.320.321.0
VAEAC,10 samples22.022.220.821.721.822.2
" + }, + { + "category_id": 8, + "poly": [ + 587, + 1125, + 1116, + 1125, + 1116, + 1215, + 587, + 1215 + ], + "score": 0.958 + }, + { + "category_id": 1, + "poly": [ + 256, + 893, + 1448, + 893, + 1448, + 957, + 256, + 957 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 253, + 1048, + 1445, + 1048, + 1445, + 1113, + 253, + 1113 + ], + "score": 0.95 + }, + { + "category_id": 8, + "poly": [ + 478, + 1346, + 1226, + 1346, + 1226, + 1441, + 478, + 1441 + ], + "score": 0.949 + }, + { + "category_id": 2, + "poly": [ + 285, + 1750, + 780, + 1750, + 780, + 1812, + 285, + 1812 + ], + "score": 0.945 + }, + { + "category_id": 2, + "poly": [ + 254, + 111, + 770, + 111, + 770, + 142, + 254, + 142 + ], + "score": 0.921 + }, + { + "category_id": 0, + "poly": [ + 255, + 992, + 659, + 992, + 659, + 1025, + 255, + 1025 + ], + "score": 0.913 + }, + { + "category_id": 9, + "poly": [ + 1397, + 1152, + 1445, + 1152, + 1445, + 1184, + 1397, + 1184 + ], + "score": 0.913 + }, + { + "category_id": 1, + "poly": [ + 253, + 1451, + 1267, + 1451, + 1267, + 1489, + 253, + 1489 + ], + "score": 0.904 + }, + { + "category_id": 6, + "poly": [ + 249, + 258, + 1448, + 258, + 1448, + 325, + 249, + 325 + ], + "score": 0.903 + }, + { + "category_id": 9, + "poly": [ + 1397, + 1377, + 1445, + 1377, + 1445, + 1409, + 1397, + 1409 + ], + "score": 0.899 + }, + { + "category_id": 2, + "poly": [ + 795, + 1865, + 813, + 1865, + 813, + 1889, + 795, + 1889 + ], + "score": 0.809 + }, + { + "category_id": 6, + "poly": [ + 247, + 563, + 1445, + 563, + 1445, + 629, + 247, + 629 + ], + "score": 0.795 + }, + { + "category_id": 1, + "poly": [ + 247, + 563, + 1445, + 563, + 1445, + 629, + 247, + 629 + ], + "score": 0.151 + }, + { + "category_id": 14, + "poly": [ + 587, + 1121, + 1116, + 1121, + 1116, + 1217, + 587, + 1217 + ], + "score": 0.94, + "latex": "\\operatorname* { m a x } _ { \\theta } \\mathbb { E } _ { x \\sim p _ { d } ( x ) } \\mathbb { E } _ { b \\sim p ( b ) } \\sum _ { i = 1 } ^ { D } b _ { i } \\log p _ { \\theta } \\big ( x _ { i } | x _ { 1 - b } , b \\big )" + }, + { + "category_id": 13, + "poly": [ + 309, + 1531, + 470, + 1531, + 470, + 1564, + 309, + 1564 + ], + "score": 0.93, + "latex": "p _ { \\theta } ( x _ { i } | x _ { 1 - b } , b )" + }, + { + "category_id": 14, + "poly": [ + 480, + 1342, + 1226, + 1342, + 1226, + 1443, + 480, + 1443 + ], + "score": 0.92, + "latex": "\\log p _ { \\theta } ( x _ { b } | x _ { 1 - b } , b ) = \\sum _ { j = 1 } ^ { | b | } \\log p _ { \\theta } ( x _ { i _ { j } } | x _ { 1 - ( b - \\sum _ { k = 1 } ^ { j - 1 } e _ { i _ { k } } ) } , b - \\sum _ { k = 1 } ^ { j - 1 } e _ { i _ { k } } )" + }, + { + "category_id": 13, + "poly": [ + 1023, + 1240, + 1197, + 1240, + 1197, + 1277, + 1023, + 1277 + ], + "score": 0.92, + "latex": "( i _ { 1 } , i _ { 2 } , \\dots , i _ { | b | } )" + }, + { + "category_id": 13, + "poly": [ + 1150, + 1500, + 1179, + 1500, + 1179, + 1533, + 1150, + 1533 + ], + "score": 0.89, + "latex": "| b |" + }, + { + "category_id": 13, + "poly": [ + 1282, + 1241, + 1312, + 1241, + 1312, + 1274, + 1282, + 1274 + ], + "score": 0.88, + "latex": "| b |" + }, + { + "category_id": 13, + "poly": [ + 428, + 1243, + 443, + 1243, + 443, + 1268, + 428, + 1268 + ], + "score": 0.77, + "latex": "b" + }, + { + "category_id": 14, + "poly": [ + 403, + 1452, + 1267, + 1452, + 1267, + 1488, + 403, + 1488 + ], + "score": 0.72, + "latex": "p _ { \\theta } ( x _ { 1 } , x _ { 4 } , x _ { 5 } | x _ { 2 } , x _ { 3 } ) = p _ { \\theta } ( x _ { 4 } | x _ { 2 } , x _ { 3 } ) p _ { \\theta } ( x _ { 1 } | x _ { 2 } , x _ { 3 } , x _ { 4 } ) p _ { \\theta } ( x _ { 5 } | x _ { 1 } , x _ { 2 } , x _ { 3 } , x _ { 4 } ) ." + }, + { + "category_id": 13, + "poly": [ + 405, + 1452, + 1267, + 1452, + 1267, + 1488, + 405, + 1488 + ], + "score": 0.67, + "latex": "p _ { \\theta } ( x _ { 1 } , x _ { 4 } , x _ { 5 } | x _ { 2 } , x _ { 3 } ) = p _ { \\theta } ( x _ { 4 } | x _ { 2 } , x _ { 3 } ) p _ { \\theta } ( x _ { 1 } | x _ { 2 } , x _ { 3 } , x _ { 4 } ) p _ { \\theta } ( x _ { 5 } | x _ { 1 } , x _ { 2 } , x _ { 3 } , x _ { 4 } ) ." + }, + { + "category_id": 15, + "poly": [ + 286.0, + 1744.0, + 782.0, + 1744.0, + 782.0, + 1786.0, + 286.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 284.0, + 1774.0, + 767.0, + 1774.0, + 767.0, + 1818.0, + 284.0, + 1818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 109.0, + 772.0, + 109.0, + 772.0, + 145.0, + 251.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 993.0, + 663.0, + 993.0, + 663.0, + 1026.0, + 251.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 259.0, + 1451.0, + 259.0, + 1451.0, + 295.0, + 249.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 288.0, + 1451.0, + 288.0, + 1451.0, + 329.0, + 248.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 791.0, + 1862.0, + 814.0, + 1862.0, + 814.0, + 1893.0, + 791.0, + 1893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 561.0, + 1450.0, + 561.0, + 1450.0, + 600.0, + 248.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 254.0, + 596.0, + 1115.0, + 596.0, + 1115.0, + 628.0, + 254.0, + 628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1239.0, + 427.0, + 1239.0, + 427.0, + 1277.0, + 249.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 444.0, + 1239.0, + 1022.0, + 1239.0, + 1022.0, + 1277.0, + 444.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 1239.0, + 1281.0, + 1239.0, + 1281.0, + 1277.0, + 1198.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1313.0, + 1239.0, + 1450.0, + 1239.0, + 1450.0, + 1277.0, + 1313.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1270.0, + 1450.0, + 1270.0, + 1450.0, + 1308.0, + 249.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1302.0, + 906.0, + 1302.0, + 906.0, + 1335.0, + 251.0, + 1335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1497.0, + 1149.0, + 1497.0, + 1149.0, + 1535.0, + 251.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1180.0, + 1497.0, + 1448.0, + 1497.0, + 1448.0, + 1535.0, + 1180.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1531.0, + 308.0, + 1531.0, + 308.0, + 1563.0, + 249.0, + 1563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 471.0, + 1531.0, + 1449.0, + 1531.0, + 1449.0, + 1563.0, + 471.0, + 1563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1559.0, + 1452.0, + 1559.0, + 1452.0, + 1598.0, + 248.0, + 1598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1588.0, + 479.0, + 1588.0, + 479.0, + 1628.0, + 248.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1635.0, + 1449.0, + 1635.0, + 1449.0, + 1674.0, + 250.0, + 1674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1667.0, + 1450.0, + 1667.0, + 1450.0, + 1705.0, + 250.0, + 1705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1699.0, + 444.0, + 1699.0, + 444.0, + 1735.0, + 247.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 893.0, + 1450.0, + 893.0, + 1450.0, + 928.0, + 248.0, + 928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 921.0, + 1402.0, + 921.0, + 1402.0, + 961.0, + 249.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1048.0, + 1451.0, + 1048.0, + 1451.0, + 1086.0, + 251.0, + 1086.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1078.0, + 1319.0, + 1078.0, + 1319.0, + 1117.0, + 250.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1446.0, + 402.0, + 1446.0, + 402.0, + 1493.0, + 247.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 561.0, + 1450.0, + 561.0, + 1450.0, + 600.0, + 248.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 254.0, + 596.0, + 1115.0, + 596.0, + 1115.0, + 628.0, + 254.0, + 628.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 886, + 849, + 1429, + 849, + 1429, + 1624, + 886, + 1624 + ], + "score": 0.974 + }, + { + "category_id": 3, + "poly": [ + 900, + 305, + 1416, + 305, + 1416, + 735, + 900, + 735 + ], + "score": 0.973 + }, + { + "category_id": 3, + "poly": [ + 276, + 853, + 838, + 853, + 838, + 1626, + 276, + 1626 + ], + "score": 0.973 + }, + { + "category_id": 3, + "poly": [ + 301, + 305, + 821, + 305, + 821, + 734, + 301, + 734 + ], + "score": 0.972 + }, + { + "category_id": 4, + "poly": [ + 861, + 1657, + 1448, + 1657, + 1448, + 1718, + 861, + 1718 + ], + "score": 0.945 + }, + { + "category_id": 4, + "poly": [ + 976, + 764, + 1334, + 764, + 1334, + 797, + 976, + 797 + ], + "score": 0.92 + }, + { + "category_id": 4, + "poly": [ + 382, + 758, + 718, + 758, + 718, + 791, + 382, + 791 + ], + "score": 0.919 + }, + { + "category_id": 4, + "poly": [ + 387, + 1656, + 721, + 1656, + 721, + 1688, + 387, + 1688 + ], + "score": 0.914 + }, + { + "category_id": 2, + "poly": [ + 253, + 113, + 770, + 113, + 770, + 141, + 253, + 141 + ], + "score": 0.903 + }, + { + "category_id": 1, + "poly": [ + 298, + 1738, + 1388, + 1738, + 1388, + 1771, + 298, + 1771 + ], + "score": 0.882 + }, + { + "category_id": 2, + "poly": [ + 795, + 1865, + 814, + 1865, + 814, + 1889, + 795, + 1889 + ], + "score": 0.807 + }, + { + "category_id": 15, + "poly": [ + 1365.0, + 882.0, + 1399.0, + 882.0, + 1399.0, + 896.0, + 1365.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 933.0, + 1133.0, + 933.0, + 1133.0, + 1000.0, + 909.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1138.0, + 944.0, + 1297.0, + 944.0, + 1297.0, + 993.0, + 1138.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1316.0, + 942.0, + 1407.0, + 942.0, + 1407.0, + 991.0, + 1316.0, + 991.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1214.0, + 1420.0, + 1264.0, + 1420.0, + 1264.0, + 1448.0, + 1214.0, + 1448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1069.0, + 1506.0, + 1108.0, + 1506.0, + 1108.0, + 1529.0, + 1069.0, + 1529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 1471.0, + 1412.0, + 1471.0, + 1412.0, + 1549.0, + 1129.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1340.0, + 1571.0, + 1408.0, + 1571.0, + 1408.0, + 1610.0, + 1340.0, + 1610.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.0, + 301.0, + 1415.0, + 301.0, + 1415.0, + 355.0, + 935.0, + 355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 345.0, + 1032.0, + 345.0, + 1032.0, + 392.0, + 939.0, + 392.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 335.0, + 1415.0, + 335.0, + 1415.0, + 449.0, + 1033.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 936.0, + 387.0, + 1075.0, + 387.0, + 1075.0, + 438.0, + 936.0, + 438.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1193.0, + 378.0, + 1341.0, + 378.0, + 1341.0, + 440.0, + 1193.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 397.0, + 945.0, + 397.0, + 945.0, + 425.0, + 921.0, + 425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 400.0, + 1407.0, + 400.0, + 1407.0, + 424.0, + 1377.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 938.0, + 430.0, + 1163.0, + 430.0, + 1163.0, + 477.0, + 938.0, + 477.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1193.0, + 426.0, + 1413.0, + 426.0, + 1413.0, + 481.0, + 1193.0, + 481.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 476.0, + 1416.0, + 476.0, + 1416.0, + 528.0, + 899.0, + 528.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 939.0, + 517.0, + 1289.0, + 517.0, + 1289.0, + 569.0, + 939.0, + 569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1320.0, + 518.0, + 1416.0, + 518.0, + 1416.0, + 568.0, + 1320.0, + 568.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 561.0, + 1072.0, + 561.0, + 1072.0, + 607.0, + 915.0, + 607.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1074.0, + 562.0, + 1196.0, + 562.0, + 1196.0, + 608.0, + 1074.0, + 608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 560.0, + 1411.0, + 560.0, + 1411.0, + 609.0, + 1200.0, + 609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 913.0, + 603.0, + 1417.0, + 603.0, + 1417.0, + 653.0, + 913.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 942.0, + 639.0, + 1332.0, + 639.0, + 1332.0, + 740.0, + 942.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1370.0, + 646.0, + 1416.0, + 646.0, + 1416.0, + 691.0, + 1370.0, + 691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 955.0, + 694.0, + 983.0, + 694.0, + 983.0, + 729.0, + 955.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 691.0, + 1071.0, + 691.0, + 1071.0, + 733.0, + 985.0, + 733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1077.0, + 690.0, + 1112.0, + 690.0, + 1112.0, + 734.0, + 1077.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1117.0, + 691.0, + 1153.0, + 691.0, + 1153.0, + 733.0, + 1117.0, + 733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1161.0, + 691.0, + 1196.0, + 691.0, + 1196.0, + 734.0, + 1161.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1205.0, + 690.0, + 1239.0, + 690.0, + 1239.0, + 735.0, + 1205.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1287.0, + 691.0, + 1411.0, + 691.0, + 1411.0, + 736.0, + 1287.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 920.0, + 315.5, + 931.0, + 315.5, + 931.0, + 335.5, + 920.0, + 335.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.75, + 382.5, + 1100.75, + 382.5, + 1100.75, + 435.5, + 1090.75, + 435.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1139.25, + 380.5, + 1220.25, + 380.5, + 1220.25, + 436.0, + 1139.25, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1311.0, + 394.0, + 1387.0, + 394.0, + 1387.0, + 430.0, + 1311.0, + 430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1160.75, + 433.0, + 1202.75, + 433.0, + 1202.75, + 476.0, + 1160.75, + 476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1276.0, + 525.5, + 1345.0, + 525.5, + 1345.0, + 557.5, + 1276.0, + 557.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1313.0, + 645.0, + 1391.0, + 645.0, + 1391.0, + 688.5, + 1313.0, + 688.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 716.0, + 1068.0, + 724.0, + 1068.0, + 724.0, + 1078.0, + 716.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1128.0, + 818.0, + 1128.0, + 818.0, + 1182.0, + 697.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 640.0, + 1142.0, + 693.0, + 1142.0, + 693.0, + 1176.0, + 640.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 283.0, + 1539.0, + 836.0, + 1539.0, + 836.0, + 1625.0, + 283.0, + 1625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 422.0, + 299.0, + 485.0, + 299.0, + 485.0, + 731.0, + 422.0, + 731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 380.0, + 300.0, + 438.0, + 300.0, + 438.0, + 733.0, + 380.0, + 733.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 773.0, + 302.0, + 825.0, + 302.0, + 825.0, + 730.0, + 773.0, + 730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 468.0, + 303.0, + 530.0, + 303.0, + 530.0, + 738.0, + 468.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 305.0, + 396.0, + 305.0, + 396.0, + 731.0, + 338.0, + 731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 515.0, + 306.0, + 568.0, + 306.0, + 568.0, + 536.0, + 515.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 319.0, + 312.0, + 345.0, + 312.0, + 345.0, + 357.0, + 319.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 645.0, + 312.0, + 697.0, + 312.0, + 697.0, + 735.0, + 645.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 315.0, + 655.0, + 315.0, + 655.0, + 738.0, + 597.0, + 738.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 686.0, + 326.0, + 745.0, + 326.0, + 745.0, + 621.0, + 686.0, + 621.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 730.0, + 330.0, + 777.0, + 330.0, + 777.0, + 530.0, + 730.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 348.0, + 598.0, + 348.0, + 598.0, + 399.0, + 563.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 550.0, + 377.0, + 615.0, + 377.0, + 615.0, + 532.0, + 550.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 731.0, + 504.0, + 780.0, + 504.0, + 780.0, + 708.0, + 731.0, + 708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 514.0, + 508.0, + 567.0, + 508.0, + 567.0, + 736.0, + 514.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 512.0, + 611.0, + 512.0, + 611.0, + 736.0, + 555.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 692.0, + 595.0, + 738.0, + 595.0, + 738.0, + 720.0, + 692.0, + 720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 860.0, + 1655.0, + 1449.0, + 1655.0, + 1449.0, + 1691.0, + 860.0, + 1691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 859.0, + 1685.0, + 1010.0, + 1685.0, + 1010.0, + 1721.0, + 859.0, + 1721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 973.0, + 759.0, + 1335.0, + 759.0, + 1335.0, + 803.0, + 973.0, + 803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 381.0, + 756.0, + 720.0, + 756.0, + 720.0, + 795.0, + 381.0, + 795.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 383.0, + 1650.0, + 724.0, + 1650.0, + 724.0, + 1695.0, + 383.0, + 1695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 110.0, + 772.0, + 110.0, + 772.0, + 145.0, + 251.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 794.0, + 1863.0, + 816.0, + 1863.0, + 816.0, + 1894.0, + 794.0, + 1894.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1731.0, + 1392.0, + 1731.0, + 1392.0, + 1778.0, + 304.0, + 1778.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 252, + 506, + 1449, + 506, + 1449, + 753, + 252, + 753 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 253, + 874, + 1448, + 874, + 1448, + 1091, + 253, + 1091 + ], + "score": 0.979 + }, + { + "category_id": 5, + "poly": [ + 508, + 313, + 1198, + 313, + 1198, + 447, + 508, + 447 + ], + "score": 0.97, + "html": "
MethodVAEACUM
Negative log-likelihood6141
Training time (30 epochs)5min 47s3min 14s
Test time (1OO samples generation)0.7ms1s
" + }, + { + "category_id": 1, + "poly": [ + 249, + 1196, + 1446, + 1196, + 1446, + 1258, + 249, + 1258 + ], + "score": 0.915 + }, + { + "category_id": 6, + "poly": [ + 572, + 260, + 1126, + 260, + 1126, + 293, + 572, + 293 + ], + "score": 0.907 + }, + { + "category_id": 2, + "poly": [ + 253, + 112, + 771, + 112, + 771, + 141, + 253, + 141 + ], + "score": 0.906 + }, + { + "category_id": 0, + "poly": [ + 254, + 801, + 499, + 801, + 499, + 838, + 254, + 838 + ], + "score": 0.899 + }, + { + "category_id": 0, + "poly": [ + 254, + 1141, + 443, + 1141, + 443, + 1174, + 254, + 1174 + ], + "score": 0.886 + }, + { + "category_id": 1, + "poly": [ + 245, + 1625, + 1443, + 1625, + 1443, + 1689, + 245, + 1689 + ], + "score": 0.842 + }, + { + "category_id": 1, + "poly": [ + 249, + 1531, + 1442, + 1531, + 1442, + 1596, + 249, + 1596 + ], + "score": 0.84 + }, + { + "category_id": 1, + "poly": [ + 252, + 1286, + 1448, + 1286, + 1448, + 1410, + 252, + 1410 + ], + "score": 0.829 + }, + { + "category_id": 2, + "poly": [ + 791, + 1866, + 821, + 1866, + 821, + 1889, + 791, + 1889 + ], + "score": 0.819 + }, + { + "category_id": 1, + "poly": [ + 246, + 1440, + 1447, + 1440, + 1447, + 1505, + 246, + 1505 + ], + "score": 0.803 + }, + { + "category_id": 1, + "poly": [ + 254, + 1718, + 1449, + 1718, + 1449, + 1811, + 254, + 1811 + ], + "score": 0.769 + }, + { + "category_id": 15, + "poly": [ + 566.0, + 256.0, + 1132.0, + 256.0, + 1132.0, + 297.0, + 566.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 109.0, + 772.0, + 109.0, + 772.0, + 145.0, + 251.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 796.0, + 505.0, + 796.0, + 505.0, + 846.0, + 246.0, + 846.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 1141.0, + 445.0, + 1141.0, + 445.0, + 1178.0, + 252.0, + 1178.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 1862.0, + 825.0, + 1862.0, + 825.0, + 1901.0, + 787.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 509.0, + 1450.0, + 509.0, + 1450.0, + 543.0, + 251.0, + 543.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 535.0, + 1453.0, + 535.0, + 1453.0, + 574.0, + 249.0, + 574.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 570.0, + 1452.0, + 570.0, + 1452.0, + 604.0, + 249.0, + 604.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 600.0, + 1448.0, + 600.0, + 1448.0, + 634.0, + 249.0, + 634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 630.0, + 1450.0, + 630.0, + 1450.0, + 663.0, + 251.0, + 663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 661.0, + 1450.0, + 661.0, + 1450.0, + 695.0, + 249.0, + 695.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 689.0, + 1450.0, + 689.0, + 1450.0, + 727.0, + 248.0, + 727.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 720.0, + 1240.0, + 720.0, + 1240.0, + 758.0, + 248.0, + 758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 875.0, + 1451.0, + 875.0, + 1451.0, + 910.0, + 249.0, + 910.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 902.0, + 1452.0, + 902.0, + 1452.0, + 943.0, + 248.0, + 943.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 933.0, + 1452.0, + 933.0, + 1452.0, + 973.0, + 249.0, + 973.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 963.0, + 1451.0, + 963.0, + 1451.0, + 1004.0, + 246.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 997.0, + 1449.0, + 997.0, + 1449.0, + 1032.0, + 249.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1026.0, + 1451.0, + 1026.0, + 1451.0, + 1062.0, + 250.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1056.0, + 1120.0, + 1056.0, + 1120.0, + 1094.0, + 249.0, + 1094.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1195.0, + 1451.0, + 1195.0, + 1451.0, + 1231.0, + 249.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 276.0, + 1226.0, + 900.0, + 1226.0, + 900.0, + 1260.0, + 276.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1622.0, + 1450.0, + 1622.0, + 1450.0, + 1662.0, + 248.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 277.0, + 1656.0, + 803.0, + 1656.0, + 803.0, + 1692.0, + 277.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1530.0, + 1448.0, + 1530.0, + 1448.0, + 1569.0, + 250.0, + 1569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 276.0, + 1564.0, + 566.0, + 1564.0, + 566.0, + 1594.0, + 276.0, + 1594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1285.0, + 1451.0, + 1285.0, + 1451.0, + 1326.0, + 248.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 279.0, + 1319.0, + 1451.0, + 1319.0, + 1451.0, + 1354.0, + 279.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 279.0, + 1347.0, + 1451.0, + 1347.0, + 1451.0, + 1386.0, + 279.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 278.0, + 1377.0, + 437.0, + 1377.0, + 437.0, + 1414.0, + 278.0, + 1414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1439.0, + 1450.0, + 1439.0, + 1450.0, + 1478.0, + 249.0, + 1478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 277.0, + 1471.0, + 1104.0, + 1471.0, + 1104.0, + 1506.0, + 277.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1716.0, + 1449.0, + 1716.0, + 1449.0, + 1754.0, + 250.0, + 1754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 278.0, + 1748.0, + 1449.0, + 1748.0, + 1449.0, + 1786.0, + 278.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 277.0, + 1778.0, + 827.0, + 1778.0, + 827.0, + 1816.0, + 277.0, + 1816.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 254, + 112, + 770, + 112, + 770, + 142, + 254, + 142 + ], + "score": 0.896 + }, + { + "category_id": 1, + "poly": [ + 252, + 266, + 1449, + 266, + 1449, + 419, + 252, + 419 + ], + "score": 0.816 + }, + { + "category_id": 1, + "poly": [ + 251, + 885, + 1447, + 885, + 1447, + 952, + 251, + 952 + ], + "score": 0.807 + }, + { + "category_id": 2, + "poly": [ + 791, + 1865, + 818, + 1865, + 818, + 1890, + 791, + 1890 + ], + "score": 0.801 + }, + { + "category_id": 1, + "poly": [ + 255, + 1717, + 1449, + 1717, + 1449, + 1812, + 255, + 1812 + ], + "score": 0.799 + }, + { + "category_id": 1, + "poly": [ + 256, + 832, + 1388, + 832, + 1388, + 868, + 256, + 868 + ], + "score": 0.784 + }, + { + "category_id": 1, + "poly": [ + 252, + 1022, + 1446, + 1022, + 1446, + 1087, + 252, + 1087 + ], + "score": 0.783 + }, + { + "category_id": 1, + "poly": [ + 251, + 1353, + 1442, + 1353, + 1442, + 1448, + 251, + 1448 + ], + "score": 0.778 + }, + { + "category_id": 1, + "poly": [ + 251, + 1519, + 1442, + 1519, + 1442, + 1586, + 251, + 1586 + ], + "score": 0.777 + }, + { + "category_id": 1, + "poly": [ + 249, + 750, + 1446, + 750, + 1446, + 815, + 249, + 815 + ], + "score": 0.775 + }, + { + "category_id": 1, + "poly": [ + 254, + 1104, + 1412, + 1104, + 1412, + 1140, + 254, + 1140 + ], + "score": 0.772 + }, + { + "category_id": 1, + "poly": [ + 253, + 1240, + 1444, + 1240, + 1444, + 1335, + 253, + 1335 + ], + "score": 0.768 + }, + { + "category_id": 1, + "poly": [ + 253, + 968, + 1439, + 968, + 1439, + 1004, + 253, + 1004 + ], + "score": 0.766 + }, + { + "category_id": 1, + "poly": [ + 255, + 1603, + 1444, + 1603, + 1444, + 1698, + 255, + 1698 + ], + "score": 0.766 + }, + { + "category_id": 1, + "poly": [ + 249, + 1466, + 1431, + 1466, + 1431, + 1503, + 249, + 1503 + ], + "score": 0.766 + }, + { + "category_id": 1, + "poly": [ + 247, + 666, + 1444, + 666, + 1444, + 733, + 247, + 733 + ], + "score": 0.766 + }, + { + "category_id": 1, + "poly": [ + 253, + 553, + 1444, + 553, + 1444, + 648, + 253, + 648 + ], + "score": 0.758 + }, + { + "category_id": 1, + "poly": [ + 251, + 1157, + 1442, + 1157, + 1442, + 1223, + 251, + 1223 + ], + "score": 0.756 + }, + { + "category_id": 1, + "poly": [ + 251, + 440, + 1450, + 440, + 1450, + 535, + 251, + 535 + ], + "score": 0.642 + }, + { + "category_id": 15, + "poly": [ + 250.0, + 110.0, + 772.0, + 110.0, + 772.0, + 145.0, + 250.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 1862.0, + 823.0, + 1862.0, + 823.0, + 1902.0, + 787.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 267.0, + 1449.0, + 267.0, + 1449.0, + 300.0, + 251.0, + 300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 279.0, + 298.0, + 1448.0, + 298.0, + 1448.0, + 331.0, + 279.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 276.0, + 327.0, + 1450.0, + 327.0, + 1450.0, + 364.0, + 276.0, + 364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 275.0, + 356.0, + 1448.0, + 356.0, + 1448.0, + 395.0, + 275.0, + 395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 275.0, + 387.0, + 363.0, + 387.0, + 363.0, + 423.0, + 275.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 883.0, + 1448.0, + 883.0, + 1448.0, + 924.0, + 248.0, + 924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 277.0, + 919.0, + 1078.0, + 919.0, + 1078.0, + 952.0, + 277.0, + 952.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1711.0, + 1449.0, + 1711.0, + 1449.0, + 1758.0, + 248.0, + 1758.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 276.0, + 1746.0, + 1449.0, + 1746.0, + 1449.0, + 1789.0, + 276.0, + 1789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 276.0, + 1778.0, + 667.0, + 1778.0, + 667.0, + 1813.0, + 276.0, + 1813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 832.0, + 1391.0, + 832.0, + 1391.0, + 871.0, + 251.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1020.0, + 1450.0, + 1020.0, + 1450.0, + 1061.0, + 251.0, + 1061.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 276.0, + 1053.0, + 1150.0, + 1053.0, + 1150.0, + 1087.0, + 276.0, + 1087.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1352.0, + 1448.0, + 1352.0, + 1448.0, + 1391.0, + 248.0, + 1391.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 276.0, + 1383.0, + 1448.0, + 1383.0, + 1448.0, + 1422.0, + 276.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 276.0, + 1416.0, + 593.0, + 1416.0, + 593.0, + 1449.0, + 276.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1520.0, + 1448.0, + 1520.0, + 1448.0, + 1556.0, + 251.0, + 1556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 279.0, + 1554.0, + 1300.0, + 1554.0, + 1300.0, + 1586.0, + 279.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 745.0, + 1451.0, + 745.0, + 1451.0, + 789.0, + 246.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 276.0, + 781.0, + 551.0, + 781.0, + 551.0, + 812.0, + 276.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1099.0, + 1414.0, + 1099.0, + 1414.0, + 1145.0, + 249.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1241.0, + 1448.0, + 1241.0, + 1448.0, + 1275.0, + 250.0, + 1275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 277.0, + 1273.0, + 1446.0, + 1273.0, + 1446.0, + 1307.0, + 277.0, + 1307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 277.0, + 1303.0, + 382.0, + 1303.0, + 382.0, + 1335.0, + 277.0, + 1335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 965.0, + 1434.0, + 965.0, + 1434.0, + 1004.0, + 251.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1602.0, + 1448.0, + 1602.0, + 1448.0, + 1640.0, + 248.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 273.0, + 1633.0, + 1450.0, + 1633.0, + 1450.0, + 1670.0, + 273.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 278.0, + 1663.0, + 421.0, + 1663.0, + 421.0, + 1698.0, + 278.0, + 1698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1467.0, + 1421.0, + 1467.0, + 1421.0, + 1503.0, + 251.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 667.0, + 1448.0, + 667.0, + 1448.0, + 703.0, + 251.0, + 703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 277.0, + 698.0, + 1435.0, + 698.0, + 1435.0, + 735.0, + 277.0, + 735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 554.0, + 1450.0, + 554.0, + 1450.0, + 588.0, + 251.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 273.0, + 583.0, + 1451.0, + 583.0, + 1451.0, + 622.0, + 273.0, + 622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 277.0, + 616.0, + 633.0, + 616.0, + 633.0, + 649.0, + 277.0, + 649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1156.0, + 1448.0, + 1156.0, + 1448.0, + 1193.0, + 249.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 278.0, + 1189.0, + 1253.0, + 1189.0, + 1253.0, + 1225.0, + 278.0, + 1225.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 437.0, + 1450.0, + 437.0, + 1450.0, + 478.0, + 248.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 276.0, + 465.0, + 1450.0, + 465.0, + 1450.0, + 513.0, + 276.0, + 513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 275.0, + 503.0, + 626.0, + 503.0, + 626.0, + 535.0, + 275.0, + 535.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 254, + 1532, + 1448, + 1532, + 1448, + 1625, + 254, + 1625 + ], + "score": 0.964 + }, + { + "category_id": 1, + "poly": [ + 254, + 1641, + 1448, + 1641, + 1448, + 1764, + 254, + 1764 + ], + "score": 0.962 + }, + { + "category_id": 1, + "poly": [ + 252, + 1780, + 1019, + 1780, + 1019, + 1810, + 252, + 1810 + ], + "score": 0.903 + }, + { + "category_id": 1, + "poly": [ + 254, + 1118, + 1448, + 1118, + 1448, + 1273, + 254, + 1273 + ], + "score": 0.892 + }, + { + "category_id": 2, + "poly": [ + 254, + 112, + 770, + 112, + 770, + 141, + 254, + 141 + ], + "score": 0.878 + }, + { + "category_id": 1, + "poly": [ + 253, + 267, + 1449, + 267, + 1449, + 419, + 253, + 419 + ], + "score": 0.876 + }, + { + "category_id": 2, + "poly": [ + 791, + 1866, + 820, + 1866, + 820, + 1889, + 791, + 1889 + ], + "score": 0.8 + }, + { + "category_id": 1, + "poly": [ + 246, + 920, + 1445, + 920, + 1445, + 983, + 246, + 983 + ], + "score": 0.742 + }, + { + "category_id": 1, + "poly": [ + 256, + 1475, + 770, + 1475, + 770, + 1507, + 256, + 1507 + ], + "score": 0.736 + }, + { + "category_id": 1, + "poly": [ + 254, + 1004, + 1446, + 1004, + 1446, + 1098, + 254, + 1098 + ], + "score": 0.701 + }, + { + "category_id": 0, + "poly": [ + 258, + 1406, + 672, + 1406, + 672, + 1442, + 258, + 1442 + ], + "score": 0.699 + }, + { + "category_id": 1, + "poly": [ + 254, + 805, + 1447, + 805, + 1447, + 900, + 254, + 900 + ], + "score": 0.697 + }, + { + "category_id": 1, + "poly": [ + 248, + 721, + 1447, + 721, + 1447, + 787, + 248, + 787 + ], + "score": 0.694 + }, + { + "category_id": 1, + "poly": [ + 243, + 555, + 1445, + 555, + 1445, + 619, + 243, + 619 + ], + "score": 0.689 + }, + { + "category_id": 1, + "poly": [ + 247, + 639, + 1445, + 639, + 1445, + 703, + 247, + 703 + ], + "score": 0.687 + }, + { + "category_id": 1, + "poly": [ + 252, + 442, + 1448, + 442, + 1448, + 536, + 252, + 536 + ], + "score": 0.619 + }, + { + "category_id": 0, + "poly": [ + 256, + 1332, + 404, + 1332, + 404, + 1366, + 256, + 1366 + ], + "score": 0.592 + }, + { + "category_id": 1, + "poly": [ + 256, + 1332, + 404, + 1332, + 404, + 1366, + 256, + 1366 + ], + "score": 0.301 + }, + { + "category_id": 1, + "poly": [ + 258, + 1406, + 672, + 1406, + 672, + 1442, + 258, + 1442 + ], + "score": 0.183 + }, + { + "category_id": 0, + "poly": [ + 256, + 1475, + 770, + 1475, + 770, + 1507, + 256, + 1507 + ], + "score": 0.154 + }, + { + "category_id": 15, + "poly": [ + 251.0, + 109.0, + 772.0, + 109.0, + 772.0, + 145.0, + 251.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 1862.0, + 825.0, + 1862.0, + 825.0, + 1902.0, + 787.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1404.0, + 675.0, + 1404.0, + 675.0, + 1446.0, + 250.0, + 1446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 1332.0, + 408.0, + 1332.0, + 408.0, + 1370.0, + 252.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1474.0, + 774.0, + 1474.0, + 774.0, + 1511.0, + 251.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1532.0, + 1451.0, + 1532.0, + 1451.0, + 1569.0, + 250.0, + 1569.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 1559.0, + 1452.0, + 1559.0, + 1452.0, + 1603.0, + 244.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1592.0, + 1253.0, + 1592.0, + 1253.0, + 1630.0, + 248.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1640.0, + 1448.0, + 1640.0, + 1448.0, + 1677.0, + 250.0, + 1677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1668.0, + 1449.0, + 1668.0, + 1449.0, + 1708.0, + 249.0, + 1708.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1700.0, + 1448.0, + 1700.0, + 1448.0, + 1737.0, + 250.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1733.0, + 1425.0, + 1733.0, + 1425.0, + 1768.0, + 251.0, + 1768.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1775.0, + 1025.0, + 1775.0, + 1025.0, + 1815.0, + 249.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1118.0, + 1451.0, + 1118.0, + 1451.0, + 1153.0, + 249.0, + 1153.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 275.0, + 1146.0, + 1449.0, + 1146.0, + 1449.0, + 1185.0, + 275.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 275.0, + 1181.0, + 1448.0, + 1181.0, + 1448.0, + 1214.0, + 275.0, + 1214.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 275.0, + 1206.0, + 1448.0, + 1206.0, + 1448.0, + 1247.0, + 275.0, + 1247.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 274.0, + 1242.0, + 659.0, + 1242.0, + 659.0, + 1275.0, + 274.0, + 1275.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 264.0, + 1449.0, + 264.0, + 1449.0, + 304.0, + 248.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 280.0, + 298.0, + 1449.0, + 298.0, + 1449.0, + 331.0, + 280.0, + 331.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 276.0, + 326.0, + 1449.0, + 326.0, + 1449.0, + 364.0, + 276.0, + 364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 276.0, + 357.0, + 1446.0, + 357.0, + 1446.0, + 394.0, + 276.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 274.0, + 388.0, + 812.0, + 388.0, + 812.0, + 423.0, + 274.0, + 423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 917.0, + 1450.0, + 917.0, + 1450.0, + 959.0, + 247.0, + 959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 277.0, + 952.0, + 1167.0, + 952.0, + 1167.0, + 985.0, + 277.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1474.0, + 774.0, + 1474.0, + 774.0, + 1511.0, + 251.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 999.0, + 1450.0, + 999.0, + 1450.0, + 1043.0, + 250.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 278.0, + 1034.0, + 1450.0, + 1034.0, + 1450.0, + 1072.0, + 278.0, + 1072.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 278.0, + 1065.0, + 1006.0, + 1065.0, + 1006.0, + 1101.0, + 278.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 805.0, + 1446.0, + 805.0, + 1446.0, + 843.0, + 251.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 277.0, + 836.0, + 1450.0, + 836.0, + 1450.0, + 874.0, + 277.0, + 874.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 275.0, + 866.0, + 1032.0, + 866.0, + 1032.0, + 903.0, + 275.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 722.0, + 1448.0, + 722.0, + 1448.0, + 759.0, + 248.0, + 759.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 278.0, + 755.0, + 1368.0, + 755.0, + 1368.0, + 788.0, + 278.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 555.0, + 1449.0, + 555.0, + 1449.0, + 591.0, + 251.0, + 591.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 277.0, + 586.0, + 1374.0, + 586.0, + 1374.0, + 622.0, + 277.0, + 622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 638.0, + 1446.0, + 638.0, + 1446.0, + 676.0, + 248.0, + 676.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 277.0, + 671.0, + 824.0, + 671.0, + 824.0, + 703.0, + 277.0, + 703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 439.0, + 1447.0, + 439.0, + 1447.0, + 478.0, + 249.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 276.0, + 469.0, + 1448.0, + 469.0, + 1448.0, + 509.0, + 276.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 276.0, + 501.0, + 1431.0, + 501.0, + 1431.0, + 539.0, + 276.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 1332.0, + 408.0, + 1332.0, + 408.0, + 1370.0, + 252.0, + 1370.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1404.0, + 675.0, + 1404.0, + 675.0, + 1446.0, + 250.0, + 1446.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 254, + 1517, + 1447, + 1517, + 1447, + 1642, + 254, + 1642 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 253, + 1656, + 1449, + 1656, + 1449, + 1812, + 253, + 1812 + ], + "score": 0.974 + }, + { + "category_id": 3, + "poly": [ + 277, + 304, + 1419, + 304, + 1419, + 777, + 277, + 777 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 253, + 987, + 1450, + 987, + 1450, + 1081, + 253, + 1081 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 250, + 1302, + 1456, + 1302, + 1456, + 1396, + 250, + 1396 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 252, + 1096, + 1447, + 1096, + 1447, + 1159, + 252, + 1159 + ], + "score": 0.943 + }, + { + "category_id": 1, + "poly": [ + 252, + 1411, + 1450, + 1411, + 1450, + 1503, + 252, + 1503 + ], + "score": 0.928 + }, + { + "category_id": 4, + "poly": [ + 553, + 821, + 1145, + 821, + 1145, + 855, + 553, + 855 + ], + "score": 0.926 + }, + { + "category_id": 1, + "poly": [ + 255, + 940, + 1067, + 940, + 1067, + 973, + 255, + 973 + ], + "score": 0.923 + }, + { + "category_id": 2, + "poly": [ + 253, + 113, + 770, + 113, + 770, + 141, + 253, + 141 + ], + "score": 0.901 + }, + { + "category_id": 0, + "poly": [ + 257, + 1231, + 719, + 1231, + 719, + 1263, + 257, + 1263 + ], + "score": 0.882 + }, + { + "category_id": 2, + "poly": [ + 791, + 1865, + 820, + 1865, + 820, + 1890, + 791, + 1890 + ], + "score": 0.836 + }, + { + "category_id": 13, + "poly": [ + 696, + 1519, + 779, + 1519, + 779, + 1552, + 696, + 1552 + ], + "score": 0.93, + "latex": "p ( b _ { i } | x )" + }, + { + "category_id": 13, + "poly": [ + 1160, + 1519, + 1323, + 1519, + 1323, + 1553, + 1160, + 1553 + ], + "score": 0.93, + "latex": "p ( b _ { i } | x ) = 0 . 2" + }, + { + "category_id": 13, + "poly": [ + 892, + 1519, + 1106, + 1519, + 1106, + 1552, + 892, + 1552 + ], + "score": 0.91, + "latex": "p ( b _ { i } | x _ { i } = \\omega ) = 1" + }, + { + "category_id": 13, + "poly": [ + 961, + 1553, + 1046, + 1553, + 1046, + 1580, + 961, + 1580 + ], + "score": 0.89, + "latex": "\\sigma _ { \\theta } = 1" + }, + { + "category_id": 13, + "poly": [ + 654, + 1581, + 709, + 1581, + 709, + 1610, + 654, + 1610 + ], + "score": 0.87, + "latex": "2 5 \\%" + }, + { + "category_id": 13, + "poly": [ + 1359, + 1304, + 1414, + 1304, + 1414, + 1333, + 1359, + 1333 + ], + "score": 0.86, + "latex": "50 \\%" + }, + { + "category_id": 13, + "poly": [ + 828, + 1755, + 847, + 1755, + 847, + 1782, + 828, + 1782 + ], + "score": 0.81, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 748, + 1447, + 767, + 1447, + 767, + 1474, + 748, + 1474 + ], + "score": 0.79, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 984, + 1663, + 1004, + 1663, + 1004, + 1685, + 984, + 1685 + ], + "score": 0.74, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 433, + 1447, + 452, + 1447, + 452, + 1469, + 433, + 1469 + ], + "score": 0.72, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 584, + 1447, + 602, + 1447, + 602, + 1473, + 584, + 1473 + ], + "score": 0.72, + "latex": "y" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 309.0, + 662.0, + 309.0, + 662.0, + 346.0, + 450.0, + 346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1167.0, + 309.0, + 1354.0, + 309.0, + 1354.0, + 349.0, + 1167.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 454.0, + 348.0, + 601.0, + 348.0, + 601.0, + 385.0, + 454.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 343.0, + 961.0, + 343.0, + 961.0, + 387.0, + 780.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1230.0, + 338.0, + 1294.0, + 338.0, + 1294.0, + 380.0, + 1230.0, + 380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 382.0, + 734.0, + 382.0, + 734.0, + 411.0, + 585.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 390.0, + 852.0, + 390.0, + 852.0, + 405.0, + 836.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 685.0, + 413.0, + 811.0, + 413.0, + 811.0, + 458.0, + 685.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 414.0, + 1102.0, + 414.0, + 1102.0, + 433.0, + 1087.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 696.0, + 521.0, + 836.0, + 521.0, + 836.0, + 553.0, + 696.0, + 553.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 856.0, + 500.0, + 1017.0, + 500.0, + 1017.0, + 533.0, + 856.0, + 533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 512.0, + 1196.0, + 512.0, + 1196.0, + 545.0, + 1052.0, + 545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 637.0, + 546.0, + 669.0, + 546.0, + 669.0, + 602.0, + 637.0, + 602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 862.0, + 565.0, + 882.0, + 565.0, + 882.0, + 587.0, + 862.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1240.0, + 538.0, + 1418.0, + 538.0, + 1418.0, + 577.0, + 1240.0, + 577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 724.0, + 721.0, + 887.0, + 721.0, + 887.0, + 750.0, + 724.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 455.0, + 739.0, + 619.0, + 739.0, + 619.0, + 776.0, + 455.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1021.0, + 743.0, + 1254.0, + 743.0, + 1254.0, + 776.0, + 1021.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 658.75, + 448.0, + 661.75, + 448.0, + 661.75, + 459.0, + 658.75, + 459.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 890.0, + 566.0, + 894.0, + 566.0, + 894.0, + 578.0, + 890.0, + 578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 935.25, + 568.0, + 939.25, + 568.0, + 939.25, + 577.5, + 935.25, + 577.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 551.0, + 815.0, + 1147.0, + 815.0, + 1147.0, + 861.0, + 551.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 110.0, + 772.0, + 110.0, + 772.0, + 145.0, + 251.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 1231.0, + 723.0, + 1231.0, + 723.0, + 1267.0, + 252.0, + 1267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 1862.0, + 825.0, + 1862.0, + 825.0, + 1900.0, + 787.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1515.0, + 695.0, + 1515.0, + 695.0, + 1556.0, + 247.0, + 1556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 780.0, + 1515.0, + 891.0, + 1515.0, + 891.0, + 1556.0, + 780.0, + 1556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1107.0, + 1515.0, + 1159.0, + 1515.0, + 1159.0, + 1556.0, + 1107.0, + 1556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1324.0, + 1515.0, + 1450.0, + 1515.0, + 1450.0, + 1556.0, + 1324.0, + 1556.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1549.0, + 960.0, + 1549.0, + 960.0, + 1584.0, + 251.0, + 1584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1047.0, + 1549.0, + 1450.0, + 1549.0, + 1450.0, + 1584.0, + 1047.0, + 1584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1579.0, + 653.0, + 1579.0, + 653.0, + 1615.0, + 250.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 1579.0, + 1450.0, + 1579.0, + 1450.0, + 1615.0, + 710.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1605.0, + 467.0, + 1605.0, + 467.0, + 1651.0, + 248.0, + 1651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1658.0, + 983.0, + 1658.0, + 983.0, + 1691.0, + 250.0, + 1691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 1658.0, + 1450.0, + 1658.0, + 1450.0, + 1691.0, + 1005.0, + 1691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1686.0, + 1453.0, + 1686.0, + 1453.0, + 1725.0, + 248.0, + 1725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1719.0, + 1449.0, + 1719.0, + 1449.0, + 1753.0, + 250.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1750.0, + 827.0, + 1750.0, + 827.0, + 1783.0, + 250.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 848.0, + 1750.0, + 1449.0, + 1750.0, + 1449.0, + 1783.0, + 848.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1779.0, + 716.0, + 1779.0, + 716.0, + 1817.0, + 249.0, + 1817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 989.0, + 1449.0, + 989.0, + 1449.0, + 1022.0, + 252.0, + 1022.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1017.0, + 1450.0, + 1017.0, + 1450.0, + 1055.0, + 250.0, + 1055.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1051.0, + 1080.0, + 1051.0, + 1080.0, + 1084.0, + 250.0, + 1084.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 1300.0, + 1358.0, + 1300.0, + 1358.0, + 1338.0, + 252.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1415.0, + 1300.0, + 1450.0, + 1300.0, + 1450.0, + 1338.0, + 1415.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1329.0, + 1450.0, + 1329.0, + 1450.0, + 1369.0, + 249.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1366.0, + 1139.0, + 1366.0, + 1139.0, + 1399.0, + 250.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1096.0, + 1450.0, + 1096.0, + 1450.0, + 1131.0, + 251.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1126.0, + 1095.0, + 1126.0, + 1095.0, + 1161.0, + 249.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1409.0, + 1451.0, + 1409.0, + 1451.0, + 1445.0, + 249.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1442.0, + 432.0, + 1442.0, + 432.0, + 1475.0, + 251.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 1442.0, + 583.0, + 1442.0, + 583.0, + 1475.0, + 453.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 603.0, + 1442.0, + 747.0, + 1442.0, + 747.0, + 1475.0, + 603.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 1442.0, + 1447.0, + 1442.0, + 1447.0, + 1475.0, + 768.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1469.0, + 394.0, + 1469.0, + 394.0, + 1510.0, + 250.0, + 1510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 937.0, + 1070.0, + 937.0, + 1070.0, + 978.0, + 250.0, + 978.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 252, + 785, + 1448, + 785, + 1448, + 939, + 252, + 939 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 252, + 974, + 1447, + 974, + 1447, + 1129, + 252, + 1129 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 254, + 1558, + 1446, + 1558, + 1446, + 1683, + 254, + 1683 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 253, + 1322, + 1448, + 1322, + 1448, + 1416, + 253, + 1416 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 253, + 1163, + 1447, + 1163, + 1447, + 1287, + 253, + 1287 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 255, + 1718, + 1448, + 1718, + 1448, + 1812, + 255, + 1812 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 254, + 1450, + 1447, + 1450, + 1447, + 1544, + 254, + 1544 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 255, + 596, + 1445, + 596, + 1445, + 689, + 255, + 689 + ], + "score": 0.971 + }, + { + "category_id": 5, + "poly": [ + 531, + 316, + 1168, + 316, + 1168, + 539, + 531, + 539 + ], + "score": 0.97, + "html": "
MaskMeaningX1x2y1y2
01Left half of the face337052115
02Right half of the face577095115
03Two eyes29985273
04Left eye29665273
05Right eye61995273
06Lower half of the face408786123
" + }, + { + "category_id": 2, + "poly": [ + 253, + 112, + 770, + 112, + 770, + 142, + 253, + 142 + ], + "score": 0.907 + }, + { + "category_id": 0, + "poly": [ + 257, + 727, + 840, + 727, + 840, + 759, + 257, + 759 + ], + "score": 0.885 + }, + { + "category_id": 6, + "poly": [ + 375, + 259, + 1320, + 259, + 1320, + 294, + 375, + 294 + ], + "score": 0.874 + }, + { + "category_id": 2, + "poly": [ + 791, + 1866, + 820, + 1866, + 820, + 1889, + 791, + 1889 + ], + "score": 0.839 + }, + { + "category_id": 13, + "poly": [ + 1136, + 847, + 1306, + 847, + 1306, + 881, + 1136, + 881 + ], + "score": 0.93, + "latex": "p _ { \\theta , i } ( z , x _ { 1 - b } , b )" + }, + { + "category_id": 13, + "poly": [ + 309, + 847, + 1055, + 847, + 1055, + 882, + 309, + 882 + ], + "score": 0.78, + "latex": "\\begin{array} { r } { \\log p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b ) = \\sum _ { i : b _ { i } = 1 } \\log \\mathrm { B e r n o u l l i } ( x _ { i } | p _ { \\theta , i } ( z , x _ { 1 - b } , b ) ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1365, + 787, + 1439, + 787, + 1439, + 816, + 1365, + 816 + ], + "score": 0.77, + "latex": "2 8 \\mathbf { x } 2 8" + }, + { + "category_id": 13, + "poly": [ + 332, + 1195, + 431, + 1195, + 431, + 1225, + 332, + 1225 + ], + "score": 0.6, + "latex": "1 7 8 \\mathrm { x } 2 1 8" + }, + { + "category_id": 13, + "poly": [ + 1224, + 1719, + 1297, + 1719, + 1297, + 1748, + 1224, + 1748 + ], + "score": 0.44, + "latex": "3 2 \\mathrm { x } 3 2 " + }, + { + "category_id": 13, + "poly": [ + 334, + 1006, + 433, + 1006, + 433, + 1035, + 334, + 1035 + ], + "score": 0.27, + "latex": "1 0 5 \\mathrm { x } 1 0 5" + }, + { + "category_id": 13, + "poly": [ + 835, + 847, + 1054, + 847, + 1054, + 881, + 835, + 881 + ], + "score": 0.26, + "latex": "( x _ { i } | p _ { \\theta , i } ( z , x _ { 1 - b } , b ) )" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 110.0, + 772.0, + 110.0, + 772.0, + 145.0, + 251.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 726.0, + 843.0, + 726.0, + 843.0, + 763.0, + 252.0, + 763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 258.0, + 1325.0, + 258.0, + 1325.0, + 298.0, + 374.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 1862.0, + 826.0, + 1862.0, + 826.0, + 1901.0, + 787.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 783.0, + 1364.0, + 783.0, + 1364.0, + 821.0, + 247.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1440.0, + 783.0, + 1449.0, + 783.0, + 1449.0, + 821.0, + 1440.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 814.0, + 1449.0, + 814.0, + 1449.0, + 849.0, + 251.0, + 849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 243.0, + 841.0, + 308.0, + 841.0, + 308.0, + 892.0, + 243.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1056.0, + 841.0, + 1135.0, + 841.0, + 1135.0, + 892.0, + 1056.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1307.0, + 841.0, + 1456.0, + 841.0, + 1456.0, + 892.0, + 1307.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 875.0, + 1451.0, + 875.0, + 1451.0, + 912.0, + 248.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 904.0, + 1138.0, + 904.0, + 1138.0, + 944.0, + 247.0, + 944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 977.0, + 1448.0, + 977.0, + 1448.0, + 1006.0, + 252.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1005.0, + 333.0, + 1005.0, + 333.0, + 1038.0, + 249.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 1005.0, + 1451.0, + 1005.0, + 1451.0, + 1038.0, + 434.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1037.0, + 1447.0, + 1037.0, + 1447.0, + 1066.0, + 249.0, + 1066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1066.0, + 1448.0, + 1066.0, + 1448.0, + 1099.0, + 251.0, + 1099.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1098.0, + 600.0, + 1098.0, + 600.0, + 1131.0, + 251.0, + 1131.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1557.0, + 1450.0, + 1557.0, + 1450.0, + 1595.0, + 248.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1590.0, + 1450.0, + 1590.0, + 1450.0, + 1625.0, + 251.0, + 1625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1618.0, + 1450.0, + 1618.0, + 1450.0, + 1656.0, + 248.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1647.0, + 854.0, + 1647.0, + 854.0, + 1691.0, + 247.0, + 1691.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1322.0, + 1449.0, + 1322.0, + 1449.0, + 1360.0, + 250.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 253.0, + 1352.0, + 1449.0, + 1352.0, + 1449.0, + 1390.0, + 253.0, + 1390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1383.0, + 938.0, + 1383.0, + 938.0, + 1422.0, + 248.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1162.0, + 1450.0, + 1162.0, + 1450.0, + 1198.0, + 249.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1193.0, + 331.0, + 1193.0, + 331.0, + 1231.0, + 249.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 1193.0, + 1450.0, + 1193.0, + 1450.0, + 1231.0, + 432.0, + 1231.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1225.0, + 1450.0, + 1225.0, + 1450.0, + 1260.0, + 250.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1254.0, + 637.0, + 1254.0, + 637.0, + 1291.0, + 249.0, + 1291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1715.0, + 1223.0, + 1715.0, + 1223.0, + 1753.0, + 249.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1298.0, + 1715.0, + 1448.0, + 1715.0, + 1448.0, + 1753.0, + 1298.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1748.0, + 1447.0, + 1748.0, + 1447.0, + 1786.0, + 249.0, + 1786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1780.0, + 1447.0, + 1780.0, + 1447.0, + 1813.0, + 251.0, + 1813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1447.0, + 1450.0, + 1447.0, + 1450.0, + 1486.0, + 248.0, + 1486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1480.0, + 1450.0, + 1480.0, + 1450.0, + 1517.0, + 248.0, + 1517.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1512.0, + 615.0, + 1512.0, + 615.0, + 1546.0, + 251.0, + 1546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 594.0, + 1448.0, + 594.0, + 1448.0, + 632.0, + 251.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 624.0, + 1451.0, + 624.0, + 1451.0, + 662.0, + 251.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 656.0, + 920.0, + 656.0, + 920.0, + 692.0, + 248.0, + 692.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 13, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 253, + 266, + 1449, + 266, + 1449, + 542, + 253, + 542 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 254, + 1404, + 1448, + 1404, + 1448, + 1558, + 254, + 1558 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 254, + 968, + 1442, + 968, + 1442, + 1094, + 254, + 1094 + ], + "score": 0.974 + }, + { + "category_id": 1, + "poly": [ + 253, + 850, + 1447, + 850, + 1447, + 956, + 253, + 956 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 254, + 1572, + 1446, + 1572, + 1446, + 1665, + 254, + 1665 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 253, + 741, + 1448, + 741, + 1448, + 836, + 253, + 836 + ], + "score": 0.966 + }, + { + "category_id": 2, + "poly": [ + 254, + 1723, + 1396, + 1723, + 1396, + 1810, + 254, + 1810 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 259, + 1248, + 1448, + 1248, + 1448, + 1312, + 259, + 1312 + ], + "score": 0.946 + }, + { + "category_id": 1, + "poly": [ + 252, + 663, + 1437, + 663, + 1437, + 728, + 252, + 728 + ], + "score": 0.943 + }, + { + "category_id": 1, + "poly": [ + 251, + 1325, + 1446, + 1325, + 1446, + 1388, + 251, + 1388 + ], + "score": 0.943 + }, + { + "category_id": 8, + "poly": [ + 358, + 1125, + 1311, + 1125, + 1311, + 1217, + 358, + 1217 + ], + "score": 0.932 + }, + { + "category_id": 2, + "poly": [ + 254, + 112, + 770, + 112, + 770, + 141, + 254, + 141 + ], + "score": 0.905 + }, + { + "category_id": 9, + "poly": [ + 1398, + 1156, + 1444, + 1156, + 1444, + 1186, + 1398, + 1186 + ], + "score": 0.901 + }, + { + "category_id": 0, + "poly": [ + 256, + 598, + 734, + 598, + 734, + 630, + 256, + 630 + ], + "score": 0.885 + }, + { + "category_id": 2, + "poly": [ + 791, + 1865, + 820, + 1865, + 820, + 1890, + 791, + 1890 + ], + "score": 0.842 + }, + { + "category_id": 13, + "poly": [ + 632, + 969, + 750, + 969, + 750, + 1004, + 632, + 1004 + ], + "score": 0.93, + "latex": "L _ { M } ( x , x ^ { \\prime } )" + }, + { + "category_id": 13, + "poly": [ + 884, + 850, + 1352, + 850, + 1352, + 898, + 884, + 898 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { L _ { M } ( x _ { i } , x _ { i } ^ { \\prime } ) = - \\frac { 1 } { | X _ { i } | } \\sum _ { j = 1 } ^ { | X _ { i } | } x _ { i , j } \\log ( x _ { i , j } ^ { \\prime } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 854, + 358, + 998, + 358, + 998, + 392, + 854, + 392 + ], + "score": 0.92, + "latex": "H ( x - 0 . 2 5 )" + }, + { + "category_id": 13, + "poly": [ + 1130, + 1434, + 1446, + 1434, + 1446, + 1467, + 1130, + 1467 + ], + "score": 0.91, + "latex": "h \\ = \\ b \\circ \\ m \\ + \\ 0 . 5 ( 1 \\ - \\ b )" + }, + { + "category_id": 13, + "poly": [ + 850, + 1496, + 1109, + 1496, + 1109, + 1529, + 850, + 1529 + ], + "score": 0.9, + "latex": "\\alpha \\in \\{ 0 . 1 , 0 . 5 , 1 , 2 , 1 0 \\}" + }, + { + "category_id": 13, + "poly": [ + 879, + 1436, + 988, + 1436, + 988, + 1463, + 879, + 1463 + ], + "score": 0.89, + "latex": "\\alpha ~ = ~ 1 0" + }, + { + "category_id": 13, + "poly": [ + 355, + 924, + 401, + 924, + 401, + 954, + 355, + 954 + ], + "score": 0.89, + "latex": "L _ { M }" + }, + { + "category_id": 13, + "poly": [ + 935, + 1465, + 1050, + 1465, + 1050, + 1495, + 935, + 1495 + ], + "score": 0.89, + "latex": "p = 0 . 0 1" + }, + { + "category_id": 13, + "poly": [ + 780, + 896, + 824, + 896, + 824, + 928, + 780, + 928 + ], + "score": 0.88, + "latex": "x _ { i , j }" + }, + { + "category_id": 13, + "poly": [ + 252, + 1434, + 575, + 1434, + 575, + 1467, + 252, + 1467 + ], + "score": 0.87, + "latex": "\\alpha \\in \\{ 0 . 1 , 0 . 5 , 1 , 2 , 1 0 \\}" + }, + { + "category_id": 13, + "poly": [ + 716, + 420, + 770, + 420, + 770, + 449, + 716, + 449 + ], + "score": 0.87, + "latex": "20 \\%" + }, + { + "category_id": 13, + "poly": [ + 925, + 420, + 980, + 420, + 980, + 449, + 925, + 449 + ], + "score": 0.87, + "latex": "30 \\%" + }, + { + "category_id": 13, + "poly": [ + 1399, + 971, + 1445, + 971, + 1445, + 1001, + 1399, + 1001 + ], + "score": 0.87, + "latex": "L _ { M }" + }, + { + "category_id": 13, + "poly": [ + 692, + 1327, + 733, + 1327, + 733, + 1355, + 692, + 1355 + ], + "score": 0.85, + "latex": "5 \\%" + }, + { + "category_id": 13, + "poly": [ + 1367, + 854, + 1419, + 854, + 1419, + 891, + 1367, + 891 + ], + "score": 0.84, + "latex": "\\left| X _ { i } \\right|" + }, + { + "category_id": 13, + "poly": [ + 253, + 929, + 280, + 929, + 280, + 954, + 253, + 954 + ], + "score": 0.8, + "latex": "x _ { i }" + }, + { + "category_id": 13, + "poly": [ + 892, + 896, + 907, + 896, + 907, + 926, + 892, + 926 + ], + "score": 0.8, + "latex": "j" + }, + { + "category_id": 13, + "poly": [ + 972, + 1182, + 1000, + 1182, + 1000, + 1209, + 972, + 1209 + ], + "score": 0.75, + "latex": "x _ { i }" + }, + { + "category_id": 13, + "poly": [ + 254, + 359, + 407, + 359, + 407, + 388, + 254, + 388 + ], + "score": 0.7, + "latex": "1 0 0 0 0 \\mathrm { x } 1 0 0 0 0" + }, + { + "category_id": 13, + "poly": [ + 408, + 1467, + 425, + 1467, + 425, + 1492, + 408, + 1492 + ], + "score": 0.7, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 598, + 896, + 610, + 896, + 610, + 921, + 598, + 921 + ], + "score": 0.7, + "latex": "i" + }, + { + "category_id": 13, + "poly": [ + 432, + 420, + 506, + 420, + 506, + 449, + 432, + 449 + ], + "score": 0.63, + "latex": "6 4 \\mathrm { x } 6 4" + }, + { + "category_id": 13, + "poly": [ + 1345, + 298, + 1447, + 298, + 1447, + 327, + 1345, + 327 + ], + "score": 0.54, + "latex": "6 0 0 \\times 6 0 0" + }, + { + "category_id": 13, + "poly": [ + 1127, + 1142, + 1156, + 1142, + 1156, + 1168, + 1127, + 1168 + ], + "score": 0.52, + "latex": "x _ { i }" + }, + { + "category_id": 13, + "poly": [ + 353, + 1126, + 1291, + 1126, + 1291, + 1220, + 353, + 1220 + ], + "score": 0.5, + "latex": "L _ { M } ^ { \\prime } ( x _ { i } , x _ { i } ^ { \\prime } ) = 0 . 2 \\cdot L _ { M } ( x _ { i } , x _ { i } ^ { \\prime } ) + 0 . 8 \\cdot \\left\\{ \\frac { 1 } { | X _ { i } | } \\sum _ { j = 1 } ^ { | X _ { i } | } ( x _ { i , j } - x _ { i , j } ^ { \\prime } ) ^ { 2 } , \\mathrm { i f ~ } x _ { i } \\mathrm { ~ i s ~ c a t e g o r i c a l } - \\sum _ { j = 1 } ^ { J } \\mathrm { i f ~ } x _ { i } \\mathrm { ~ i s ~ b i n a r y } \\right." + }, + { + "category_id": 14, + "poly": [ + 355, + 1125, + 1302, + 1125, + 1302, + 1219, + 355, + 1219 + ], + "score": 0.42, + "latex": "L _ { M } ^ { \\prime } ( x _ { i } , x _ { i } ^ { \\prime } ) = 0 . 2 \\cdot L _ { M } ( x _ { i } , x _ { i } ^ { \\prime } ) + 0 . 8 \\cdot \\left\\{ \\frac { 1 } { | X _ { i } | } \\sum _ { j = 1 } ^ { | X _ { i } | } ( x _ { i , j } - x _ { i , j } ^ { \\prime } ) ^ { 2 } , \\mathrm { i f ~ } x _ { i } \\mathrm { ~ i s ~ c a t e g o r i c a t i o n } \\right\\} ." + }, + { + "category_id": 13, + "poly": [ + 807, + 1178, + 936, + 1178, + 936, + 1217, + 807, + 1217 + ], + "score": 0.39, + "latex": "\\lfloor ( x _ { i } - x _ { i } ^ { \\prime } ) ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1214, + 389, + 1369, + 389, + 1369, + 418, + 1214, + 418 + ], + "score": 0.33, + "latex": "1 0 0 0 0 \\mathrm { x } 1 0 0 0 0" + }, + { + "category_id": 15, + "poly": [ + 283.0, + 1718.0, + 1396.0, + 1718.0, + 1396.0, + 1756.0, + 283.0, + 1756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 1752.0, + 319.0, + 1752.0, + 319.0, + 1781.0, + 252.0, + 1781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 283.0, + 1774.0, + 679.0, + 1774.0, + 679.0, + 1818.0, + 283.0, + 1818.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 109.0, + 772.0, + 109.0, + 772.0, + 145.0, + 251.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 597.0, + 740.0, + 597.0, + 740.0, + 635.0, + 250.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 1862.0, + 825.0, + 1862.0, + 825.0, + 1900.0, + 787.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 268.0, + 1449.0, + 268.0, + 1449.0, + 301.0, + 250.0, + 301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 296.0, + 1344.0, + 296.0, + 1344.0, + 332.0, + 249.0, + 332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 330.0, + 1449.0, + 330.0, + 1449.0, + 362.0, + 250.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 357.0, + 253.0, + 357.0, + 253.0, + 394.0, + 250.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 357.0, + 853.0, + 357.0, + 853.0, + 394.0, + 408.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 999.0, + 357.0, + 1450.0, + 357.0, + 1450.0, + 394.0, + 999.0, + 394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 387.0, + 1213.0, + 387.0, + 1213.0, + 424.0, + 248.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1370.0, + 387.0, + 1449.0, + 387.0, + 1449.0, + 424.0, + 1370.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 418.0, + 431.0, + 418.0, + 431.0, + 454.0, + 248.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 507.0, + 418.0, + 715.0, + 418.0, + 715.0, + 454.0, + 507.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 771.0, + 418.0, + 924.0, + 418.0, + 924.0, + 454.0, + 771.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 418.0, + 1452.0, + 418.0, + 1452.0, + 454.0, + 981.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 451.0, + 1450.0, + 451.0, + 1450.0, + 484.0, + 249.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 480.0, + 1450.0, + 480.0, + 1450.0, + 516.0, + 249.0, + 516.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 510.0, + 1184.0, + 510.0, + 1184.0, + 546.0, + 249.0, + 546.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1402.0, + 1452.0, + 1402.0, + 1452.0, + 1440.0, + 247.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1435.0, + 251.0, + 1435.0, + 251.0, + 1468.0, + 247.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 576.0, + 1435.0, + 878.0, + 1435.0, + 878.0, + 1468.0, + 576.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 989.0, + 1435.0, + 1129.0, + 1435.0, + 1129.0, + 1468.0, + 989.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1464.0, + 407.0, + 1464.0, + 407.0, + 1497.0, + 251.0, + 1497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 426.0, + 1464.0, + 934.0, + 1464.0, + 934.0, + 1497.0, + 426.0, + 1497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 1464.0, + 1451.0, + 1464.0, + 1451.0, + 1497.0, + 1051.0, + 1497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1492.0, + 849.0, + 1492.0, + 849.0, + 1530.0, + 249.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1110.0, + 1492.0, + 1451.0, + 1492.0, + 1451.0, + 1530.0, + 1110.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1523.0, + 1055.0, + 1523.0, + 1055.0, + 1562.0, + 247.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 963.0, + 631.0, + 963.0, + 631.0, + 1011.0, + 247.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 751.0, + 963.0, + 1398.0, + 963.0, + 1398.0, + 1011.0, + 751.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1446.0, + 963.0, + 1449.0, + 963.0, + 1449.0, + 1011.0, + 1446.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 997.0, + 1446.0, + 997.0, + 1446.0, + 1037.0, + 248.0, + 1037.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1027.0, + 1449.0, + 1027.0, + 1449.0, + 1069.0, + 247.0, + 1069.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1062.0, + 1064.0, + 1062.0, + 1064.0, + 1097.0, + 250.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 848.0, + 883.0, + 848.0, + 883.0, + 899.0, + 246.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 893.0, + 597.0, + 893.0, + 597.0, + 929.0, + 250.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 611.0, + 893.0, + 779.0, + 893.0, + 779.0, + 929.0, + 611.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 893.0, + 891.0, + 893.0, + 891.0, + 929.0, + 825.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 908.0, + 893.0, + 1451.0, + 893.0, + 1451.0, + 929.0, + 908.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 922.0, + 252.0, + 922.0, + 252.0, + 958.0, + 247.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 281.0, + 922.0, + 354.0, + 922.0, + 354.0, + 958.0, + 281.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 922.0, + 1403.0, + 922.0, + 1403.0, + 958.0, + 402.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1063.75, + 852.0, + 1455.75, + 852.0, + 1455.75, + 901.0, + 1063.75, + 901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1569.0, + 1450.0, + 1569.0, + 1450.0, + 1608.0, + 248.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1599.0, + 1452.0, + 1599.0, + 1452.0, + 1644.0, + 247.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1631.0, + 1126.0, + 1631.0, + 1126.0, + 1668.0, + 250.0, + 1668.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 740.0, + 1452.0, + 740.0, + 1452.0, + 776.0, + 249.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 770.0, + 1449.0, + 770.0, + 1449.0, + 808.0, + 250.0, + 808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 801.0, + 513.0, + 801.0, + 513.0, + 842.0, + 248.0, + 842.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 253.0, + 1247.0, + 1450.0, + 1247.0, + 1450.0, + 1283.0, + 253.0, + 1283.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1279.0, + 429.0, + 1279.0, + 429.0, + 1313.0, + 250.0, + 1313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 665.0, + 1440.0, + 665.0, + 1440.0, + 697.0, + 252.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 690.0, + 980.0, + 690.0, + 980.0, + 734.0, + 248.0, + 734.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1321.0, + 691.0, + 1321.0, + 691.0, + 1363.0, + 248.0, + 1363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 734.0, + 1321.0, + 1449.0, + 1321.0, + 1449.0, + 1363.0, + 734.0, + 1363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1357.0, + 682.0, + 1357.0, + 682.0, + 1390.0, + 247.0, + 1390.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 14, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 253, + 1406, + 1451, + 1406, + 1451, + 1592, + 253, + 1592 + ], + "score": 0.981 + }, + { + "category_id": 5, + "poly": [ + 256, + 408, + 1443, + 408, + 1443, + 862, + 256, + 862 + ], + "score": 0.98, + "html": "
DatasetOur modificationα=10α=2α=1α= 0.5α=0.1
Boston0.78±0.030.87±0.021.0 ± 0.11.0 ± 0.11.02 ± 0.051.6±0.2
Breast0.67 ± 0.010.80±0.051.00 ± 0.051.10 ± 0.071.19 ± 0.051.52 ± 0.06
Concrete0.96 ± 0.010.98 ± 0.021.02 ± 0.021.13 ± 0.061.17 ± 0.041.3 ± 0.1
Diabetes0.911 ± 0.0090.93 ±0.031.05 ± 0.041.07 ± 0.071.21 ± 0.071.6 ± 0.1
Digits0.79 ± 0.020.88 ± 0.011.05 ± 0.021.13 ± 0.021.24 ± 0.081.4± 0.2
Glass1.06 ± 0.051.04 ± 0.051.19 ± 0.061.4 ± 0.21.6 ± 0.11.81 ± 0.10
Iris0.72 ±0.040.73±0.060.83 ±0.080.97 ± 0.091.2 ± 0.21.3±0.2
Mushroom0.271 ± 0.0030.404 ± 0.0040.52 ± 0.050.55 ± 0.010.56 ± 0.030.64± 0.06
Orthopedic0.91 ± 0.030.91 ±0.081.1 ± 0.11.2 ± 0.11.34 ± 0.081.6 ± 0.2
Phishing0.427 ± 0.0100.52 ±0.020.54±0.020.543 ± 0.0100.56 ± 0.010.57 ± 0.04
WallRobot0.907 ± 0.0050.924± 0.0050.933 ± 0.0080.95 ± 0.011.00 ± 0.021.26 ± 0.04
WhiteWine0.97±0.021.02 ± 0.041.2 ± 0.11.3 ± 0.11.6 ± 0.11.86 ± 0.08
Yeast0.99 ±0.031.3±0.21.6 ± 0.11.83 ± 0.091.9 ±0.12.4± 0.4
Zoo0.20 ±0.020.24± 0.050.35 ± 0.060.36 ± 0.030.43 ± 0.040.433 ± 0.004
" + }, + { + "category_id": 1, + "poly": [ + 254, + 1687, + 1448, + 1687, + 1448, + 1813, + 254, + 1813 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 256, + 1151, + 1447, + 1151, + 1447, + 1245, + 256, + 1245 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 254, + 1043, + 1449, + 1043, + 1449, + 1138, + 254, + 1138 + ], + "score": 0.975 + }, + { + "category_id": 0, + "poly": [ + 253, + 915, + 432, + 915, + 432, + 952, + 253, + 952 + ], + "score": 0.919 + }, + { + "category_id": 0, + "poly": [ + 255, + 984, + 613, + 984, + 613, + 1017, + 255, + 1017 + ], + "score": 0.914 + }, + { + "category_id": 2, + "poly": [ + 254, + 112, + 770, + 112, + 770, + 142, + 254, + 142 + ], + "score": 0.913 + }, + { + "category_id": 1, + "poly": [ + 253, + 1359, + 934, + 1359, + 934, + 1394, + 253, + 1394 + ], + "score": 0.908 + }, + { + "category_id": 6, + "poly": [ + 251, + 259, + 1450, + 259, + 1450, + 384, + 251, + 384 + ], + "score": 0.906 + }, + { + "category_id": 2, + "poly": [ + 791, + 1865, + 821, + 1865, + 821, + 1890, + 791, + 1890 + ], + "score": 0.841 + }, + { + "category_id": 0, + "poly": [ + 253, + 1628, + 1258, + 1628, + 1258, + 1663, + 253, + 1663 + ], + "score": 0.814 + }, + { + "category_id": 8, + "poly": [ + 499, + 1262, + 1195, + 1262, + 1195, + 1354, + 499, + 1354 + ], + "score": 0.797 + }, + { + "category_id": 8, + "poly": [ + 492, + 1315, + 1202, + 1315, + 1202, + 1353, + 492, + 1353 + ], + "score": 0.305 + }, + { + "category_id": 8, + "poly": [ + 506, + 1315, + 1195, + 1315, + 1195, + 1353, + 506, + 1353 + ], + "score": 0.221 + }, + { + "category_id": 8, + "poly": [ + 550, + 1262, + 1148, + 1262, + 1148, + 1301, + 550, + 1301 + ], + "score": 0.179 + }, + { + "category_id": 1, + "poly": [ + 253, + 1628, + 1258, + 1628, + 1258, + 1663, + 253, + 1663 + ], + "score": 0.112 + }, + { + "category_id": 13, + "poly": [ + 529, + 1105, + 656, + 1105, + 656, + 1139, + 529, + 1139 + ], + "score": 0.93, + "latex": "p ( x _ { b } | x _ { 1 - b } )" + }, + { + "category_id": 13, + "poly": [ + 658, + 1360, + 734, + 1360, + 734, + 1395, + 658, + 1395 + ], + "score": 0.92, + "latex": "p ( x | y )" + }, + { + "category_id": 13, + "poly": [ + 1019, + 1779, + 1358, + 1779, + 1358, + 1813, + 1019, + 1813 + ], + "score": 0.9, + "latex": "p _ { d } ( y = \\mathrm { { 0 } ) = { { p } ( y = 5 ) = 0 . 5 } }" + }, + { + "category_id": 13, + "poly": [ + 1174, + 291, + 1221, + 291, + 1221, + 325, + 1174, + 325 + ], + "score": 0.9, + "latex": "L _ { M } ^ { \\prime }" + }, + { + "category_id": 13, + "poly": [ + 922, + 322, + 1015, + 322, + 1015, + 350, + 922, + 350 + ], + "score": 0.9, + "latex": "\\alpha = 1 0" + }, + { + "category_id": 13, + "poly": [ + 811, + 1779, + 1006, + 1779, + 1006, + 1813, + 811, + 1813 + ], + "score": 0.89, + "latex": "x _ { 2 } \\sim \\mathcal { N } ( \\bar { x } _ { 2 } | y , 1 )" + }, + { + "category_id": 13, + "poly": [ + 714, + 1781, + 798, + 1781, + 798, + 1810, + 714, + 1810 + ], + "score": 0.88, + "latex": "x _ { 1 } = 1" + }, + { + "category_id": 13, + "poly": [ + 253, + 1720, + 346, + 1720, + 346, + 1750, + 253, + 1750 + ], + "score": 0.87, + "latex": "x _ { i } = \\omega" + }, + { + "category_id": 13, + "poly": [ + 441, + 1149, + 478, + 1149, + 478, + 1181, + 441, + 1181 + ], + "score": 0.86, + "latex": "2 ^ { D }" + }, + { + "category_id": 14, + "poly": [ + 502, + 1314, + 1201, + 1314, + 1201, + 1353, + 502, + 1353 + ], + "score": 0.84, + "latex": "p _ { \\theta , V A E A C } ( x _ { b } | z , x _ { 1 - b } , b ) = p _ { \\theta , C V A E , 1 - b } ( x _ { b } | z , x _ { 1 - b } ) \\forall z , x , b" + }, + { + "category_id": 13, + "poly": [ + 522, + 1690, + 548, + 1690, + 548, + 1716, + 522, + 1716 + ], + "score": 0.81, + "latex": "D" + }, + { + "category_id": 14, + "poly": [ + 538, + 1262, + 1155, + 1262, + 1155, + 1300, + 538, + 1300 + ], + "score": 0.79, + "latex": "p _ { \\psi , V A E A C } ( z | x _ { 1 - b } , b ) = p _ { \\psi , C V A E , 1 - b } ( z | x _ { 1 - b } ) \\forall x , b" + }, + { + "category_id": 13, + "poly": [ + 621, + 1724, + 639, + 1724, + 639, + 1750, + 621, + 1750 + ], + "score": 0.76, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 815, + 1756, + 833, + 1756, + 833, + 1777, + 815, + 1777 + ], + "score": 0.76, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 784, + 1695, + 803, + 1695, + 803, + 1716, + 784, + 1716 + ], + "score": 0.75, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 965, + 1755, + 982, + 1755, + 982, + 1781, + 965, + 1781 + ], + "score": 0.73, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 877, + 359, + 897, + 359, + 897, + 380, + 877, + 380 + ], + "score": 0.67, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 412, + 323, + 429, + 323, + 429, + 349, + 412, + 349 + ], + "score": 0.48, + "latex": "b" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 911.0, + 437.0, + 911.0, + 437.0, + 958.0, + 247.0, + 958.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 982.0, + 617.0, + 982.0, + 617.0, + 1021.0, + 248.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 110.0, + 772.0, + 110.0, + 772.0, + 145.0, + 250.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 260.0, + 1449.0, + 260.0, + 1449.0, + 295.0, + 251.0, + 295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 289.0, + 1173.0, + 289.0, + 1173.0, + 329.0, + 250.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1222.0, + 289.0, + 1450.0, + 289.0, + 1450.0, + 329.0, + 1222.0, + 329.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 323.0, + 411.0, + 323.0, + 411.0, + 354.0, + 251.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 323.0, + 921.0, + 323.0, + 921.0, + 354.0, + 430.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1016.0, + 323.0, + 1447.0, + 323.0, + 1447.0, + 354.0, + 1016.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 351.0, + 876.0, + 351.0, + 876.0, + 388.0, + 250.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 898.0, + 351.0, + 909.0, + 351.0, + 909.0, + 388.0, + 898.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 1862.0, + 825.0, + 1862.0, + 825.0, + 1901.0, + 787.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1630.0, + 1257.0, + 1630.0, + 1257.0, + 1662.0, + 250.0, + 1662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 1408.0, + 1448.0, + 1408.0, + 1448.0, + 1440.0, + 252.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1436.0, + 1451.0, + 1436.0, + 1451.0, + 1471.0, + 248.0, + 1471.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1467.0, + 1451.0, + 1467.0, + 1451.0, + 1503.0, + 246.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1498.0, + 1451.0, + 1498.0, + 1451.0, + 1533.0, + 249.0, + 1533.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1527.0, + 1451.0, + 1527.0, + 1451.0, + 1565.0, + 248.0, + 1565.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1560.0, + 572.0, + 1560.0, + 572.0, + 1593.0, + 249.0, + 1593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1688.0, + 521.0, + 1688.0, + 521.0, + 1723.0, + 251.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 549.0, + 1688.0, + 783.0, + 1688.0, + 783.0, + 1723.0, + 549.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 804.0, + 1688.0, + 1448.0, + 1688.0, + 1448.0, + 1723.0, + 804.0, + 1723.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1718.0, + 252.0, + 1718.0, + 252.0, + 1754.0, + 246.0, + 1754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1718.0, + 620.0, + 1718.0, + 620.0, + 1754.0, + 347.0, + 1754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 640.0, + 1718.0, + 1452.0, + 1718.0, + 1452.0, + 1754.0, + 640.0, + 1754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1749.0, + 814.0, + 1749.0, + 814.0, + 1784.0, + 250.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 834.0, + 1749.0, + 964.0, + 1749.0, + 964.0, + 1784.0, + 834.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 983.0, + 1749.0, + 1451.0, + 1749.0, + 1451.0, + 1784.0, + 983.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1777.0, + 713.0, + 1777.0, + 713.0, + 1816.0, + 247.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 799.0, + 1777.0, + 810.0, + 1777.0, + 810.0, + 1816.0, + 799.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1007.0, + 1777.0, + 1018.0, + 1777.0, + 1018.0, + 1816.0, + 1007.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1359.0, + 1777.0, + 1452.0, + 1777.0, + 1452.0, + 1816.0, + 1359.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1148.0, + 440.0, + 1148.0, + 440.0, + 1189.0, + 250.0, + 1189.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 1148.0, + 1447.0, + 1148.0, + 1447.0, + 1189.0, + 479.0, + 1189.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1180.0, + 1451.0, + 1180.0, + 1451.0, + 1219.0, + 248.0, + 1219.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1212.0, + 1270.0, + 1212.0, + 1270.0, + 1250.0, + 250.0, + 1250.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1040.0, + 1450.0, + 1040.0, + 1450.0, + 1081.0, + 250.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1072.0, + 1449.0, + 1072.0, + 1449.0, + 1110.0, + 250.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1104.0, + 528.0, + 1104.0, + 528.0, + 1142.0, + 250.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 1104.0, + 1075.0, + 1104.0, + 1075.0, + 1142.0, + 657.0, + 1142.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1358.0, + 657.0, + 1358.0, + 657.0, + 1398.0, + 251.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 1358.0, + 934.0, + 1358.0, + 934.0, + 1398.0, + 735.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1630.0, + 1257.0, + 1630.0, + 1257.0, + 1662.0, + 250.0, + 1662.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 15, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 252, + 1157, + 1449, + 1157, + 1449, + 1439, + 252, + 1439 + ], + "score": 0.983 + }, + { + "category_id": 1, + "poly": [ + 253, + 264, + 1449, + 264, + 1449, + 451, + 253, + 451 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 252, + 684, + 1449, + 684, + 1449, + 856, + 252, + 856 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 253, + 1656, + 1448, + 1656, + 1448, + 1813, + 253, + 1813 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 253, + 869, + 1449, + 869, + 1449, + 1003, + 253, + 1003 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 253, + 544, + 1448, + 544, + 1448, + 670, + 253, + 670 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 251, + 1579, + 1451, + 1579, + 1451, + 1643, + 251, + 1643 + ], + "score": 0.947 + }, + { + "category_id": 8, + "poly": [ + 596, + 1478, + 1103, + 1478, + 1103, + 1517, + 596, + 1517 + ], + "score": 0.93 + }, + { + "category_id": 1, + "poly": [ + 253, + 1013, + 988, + 1013, + 988, + 1046, + 253, + 1046 + ], + "score": 0.924 + }, + { + "category_id": 0, + "poly": [ + 254, + 1089, + 944, + 1089, + 944, + 1126, + 254, + 1126 + ], + "score": 0.919 + }, + { + "category_id": 2, + "poly": [ + 253, + 112, + 770, + 112, + 770, + 141, + 253, + 141 + ], + "score": 0.913 + }, + { + "category_id": 8, + "poly": [ + 375, + 1522, + 1273, + 1522, + 1273, + 1559, + 375, + 1559 + ], + "score": 0.903 + }, + { + "category_id": 9, + "poly": [ + 1397, + 1481, + 1444, + 1481, + 1444, + 1512, + 1397, + 1512 + ], + "score": 0.887 + }, + { + "category_id": 9, + "poly": [ + 1398, + 1525, + 1444, + 1525, + 1444, + 1555, + 1398, + 1555 + ], + "score": 0.878 + }, + { + "category_id": 2, + "poly": [ + 791, + 1865, + 819, + 1865, + 819, + 1889, + 791, + 1889 + ], + "score": 0.842 + }, + { + "category_id": 0, + "poly": [ + 252, + 487, + 1143, + 487, + 1143, + 520, + 252, + 520 + ], + "score": 0.521 + }, + { + "category_id": 1, + "poly": [ + 252, + 487, + 1143, + 487, + 1143, + 520, + 252, + 520 + ], + "score": 0.426 + }, + { + "category_id": 13, + "poly": [ + 597, + 753, + 778, + 753, + 778, + 793, + 597, + 793 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { { 1 , 2 , \\ldots , \\frac { D } { 2 } - 1 } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 536, + 1781, + 679, + 1781, + 679, + 1813, + 536, + 1813 + ], + "score": 0.93, + "latex": "z \\sim p _ { \\psi } ( z | x )" + }, + { + "category_id": 13, + "poly": [ + 1266, + 1749, + 1438, + 1749, + 1438, + 1784, + 1266, + 1784 + ], + "score": 0.93, + "latex": "z \\sim q _ { \\phi } ( z | x , y )" + }, + { + "category_id": 13, + "poly": [ + 619, + 1221, + 733, + 1221, + 733, + 1254, + 619, + 1254 + ], + "score": 0.93, + "latex": "q _ { \\phi } ( z | x , y )" + }, + { + "category_id": 13, + "poly": [ + 1099, + 297, + 1266, + 297, + 1266, + 331, + 1099, + 331 + ], + "score": 0.93, + "latex": "x _ { 2 } \\sim \\bar { \\mathcal { N } } ( 0 , 1 )" + }, + { + "category_id": 13, + "poly": [ + 574, + 1403, + 815, + 1403, + 815, + 1438, + 574, + 1438 + ], + "score": 0.93, + "latex": "q _ { \\phi } ( z | x , y ) = p _ { \\psi } ( z | y )" + }, + { + "category_id": 13, + "poly": [ + 1233, + 606, + 1377, + 606, + 1377, + 640, + 1233, + 640 + ], + "score": 0.93, + "latex": "p ( b ) \\neq { \\hat { p } } ( b )" + }, + { + "category_id": 13, + "poly": [ + 769, + 328, + 1077, + 328, + 1077, + 362, + 769, + 362 + ], + "score": 0.93, + "latex": "p _ { d } ( x _ { 2 } | x _ { 1 } , y ) = \\mathcal { N } ( x _ { 2 } | y , 1 )" + }, + { + "category_id": 13, + "poly": [ + 253, + 328, + 413, + 328, + 413, + 362, + 253, + 362 + ], + "score": 0.93, + "latex": "x _ { 2 } \\sim \\mathcal { N } ( 5 , 1 )" + }, + { + "category_id": 13, + "poly": [ + 415, + 1251, + 505, + 1251, + 505, + 1284, + 415, + 1284 + ], + "score": 0.92, + "latex": "p _ { \\psi } ( z | y )" + }, + { + "category_id": 13, + "poly": [ + 818, + 638, + 991, + 638, + 991, + 673, + 818, + 673 + ], + "score": 0.92, + "latex": "( i _ { 1 } , i _ { 2 } , \\dots , i _ { | b | } )" + }, + { + "category_id": 14, + "poly": [ + 367, + 1475, + 1332, + 1475, + 1332, + 1564, + 367, + 1564 + ], + "score": 0.92, + "latex": "\\begin{array} { r } { L _ { G S N N } ( x , y ; \\theta , \\psi ) = \\mathbb { E } _ { p _ { \\psi } ( z | y ) } \\log p _ { \\theta } ( x | z , y ) \\qquad } \\\\ { L ( x , y ; \\theta , \\psi , \\phi ) = \\alpha L _ { C V A E } ( x , y ; \\theta , \\psi , \\phi ) + ( 1 - \\alpha ) L _ { G S N N } ( x , y ; \\theta , \\psi ) , \\quad \\alpha \\in [ 0 , 1 ] } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 747, + 871, + 797, + 871, + 797, + 903, + 747, + 903 + ], + "score": 0.91, + "latex": "{ \\hat { p } } ( b )" + }, + { + "category_id": 13, + "poly": [ + 387, + 636, + 436, + 636, + 436, + 670, + 387, + 670 + ], + "score": 0.91, + "latex": "{ \\hat { p } } ( b )" + }, + { + "category_id": 13, + "poly": [ + 460, + 720, + 511, + 720, + 511, + 754, + 460, + 754 + ], + "score": 0.91, + "latex": "p ( b )" + }, + { + "category_id": 13, + "poly": [ + 505, + 576, + 555, + 576, + 555, + 608, + 505, + 608 + ], + "score": 0.91, + "latex": "p ( b )" + }, + { + "category_id": 13, + "poly": [ + 579, + 606, + 629, + 606, + 629, + 640, + 579, + 640 + ], + "score": 0.91, + "latex": "p ( b )" + }, + { + "category_id": 13, + "poly": [ + 318, + 901, + 492, + 901, + 492, + 936, + 318, + 936 + ], + "score": 0.91, + "latex": "( i _ { 1 } , i _ { 2 } , \\dots , i _ { | b | } )" + }, + { + "category_id": 13, + "poly": [ + 940, + 870, + 990, + 870, + 990, + 904, + 940, + 904 + ], + "score": 0.91, + "latex": "p ( b )" + }, + { + "category_id": 13, + "poly": [ + 459, + 330, + 533, + 330, + 533, + 360, + 459, + 360 + ], + "score": 0.91, + "latex": "y = 0" + }, + { + "category_id": 13, + "poly": [ + 1331, + 267, + 1445, + 267, + 1445, + 301, + 1331, + 301 + ], + "score": 0.91, + "latex": "p _ { d } ( x _ { 2 } | x _ { 1 } )" + }, + { + "category_id": 13, + "poly": [ + 1237, + 687, + 1286, + 687, + 1286, + 720, + 1237, + 720 + ], + "score": 0.9, + "latex": "{ \\hat { p } } ( b )" + }, + { + "category_id": 13, + "poly": [ + 555, + 936, + 678, + 936, + 678, + 968, + 555, + 968 + ], + "score": 0.9, + "latex": "b = b _ { 0 } \\circ b _ { 1 }" + }, + { + "category_id": 13, + "poly": [ + 1351, + 361, + 1445, + 361, + 1445, + 392, + 1351, + 392 + ], + "score": 0.9, + "latex": "p _ { d } ( x , y )" + }, + { + "category_id": 13, + "poly": [ + 1316, + 301, + 1396, + 301, + 1396, + 329, + 1316, + 329 + ], + "score": 0.89, + "latex": "y = 5" + }, + { + "category_id": 13, + "poly": [ + 328, + 970, + 624, + 970, + 624, + 1002, + 328, + 1002 + ], + "score": 0.89, + "latex": "i _ { j - 1 } < i _ { j } \\forall j : 2 \\le j \\le | b |" + }, + { + "category_id": 13, + "poly": [ + 253, + 937, + 504, + 937, + 504, + 970, + 253, + 970 + ], + "score": 0.88, + "latex": "b _ { 1 } \\sim ( \\mathrm { B e r n o u l l i } ( u ) ) ^ { D }" + }, + { + "category_id": 13, + "poly": [ + 513, + 790, + 540, + 790, + 540, + 829, + 513, + 829 + ], + "score": 0.88, + "latex": "\\textstyle { \\frac { D } { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 1302, + 330, + 1445, + 330, + 1445, + 361, + 1302, + 361 + ], + "score": 0.88, + "latex": "p _ { d } ( x _ { 2 } | x _ { 1 } , y )" + }, + { + "category_id": 13, + "poly": [ + 911, + 606, + 961, + 606, + 961, + 639, + 911, + 639 + ], + "score": 0.87, + "latex": "{ \\hat { p } } ( b )" + }, + { + "category_id": 13, + "poly": [ + 309, + 266, + 881, + 266, + 881, + 301, + 309, + 301 + ], + "score": 0.87, + "latex": "p _ { d } ( x _ { 2 } | x _ { 1 } = 1 ) = 0 . 5 \\mathcal { N } ( x _ { 2 } | 0 , 1 ) + 0 . 5 \\mathcal { N } ( x _ { 2 } | 5 , 1 )" + }, + { + "category_id": 13, + "poly": [ + 1389, + 575, + 1438, + 575, + 1438, + 608, + 1389, + 608 + ], + "score": 0.86, + "latex": "{ \\hat { p } } ( b )" + }, + { + "category_id": 13, + "poly": [ + 1251, + 901, + 1384, + 901, + 1384, + 935, + 1251, + 935 + ], + "score": 0.83, + "latex": "u \\sim U [ 0 , 1 ]" + }, + { + "category_id": 13, + "poly": [ + 1131, + 901, + 1240, + 901, + 1240, + 935, + 1131, + 935 + ], + "score": 0.82, + "latex": "b _ { 0 } \\sim p ( b )" + }, + { + "category_id": 13, + "poly": [ + 447, + 1785, + 465, + 1785, + 465, + 1812, + 447, + 1812 + ], + "score": 0.8, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 1165, + 394, + 1184, + 394, + 1184, + 420, + 1165, + 420 + ], + "score": 0.8, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 1174, + 1754, + 1193, + 1754, + 1193, + 1782, + 1174, + 1782 + ], + "score": 0.79, + "latex": "y" + }, + { + "category_id": 13, + "poly": [ + 374, + 1724, + 392, + 1724, + 392, + 1746, + 374, + 1746 + ], + "score": 0.76, + "latex": "z" + }, + { + "category_id": 13, + "poly": [ + 1156, + 1226, + 1174, + 1226, + 1174, + 1249, + 1156, + 1249 + ], + "score": 0.72, + "latex": "z" + }, + { + "category_id": 13, + "poly": [ + 1164, + 717, + 1193, + 717, + 1193, + 756, + 1164, + 756 + ], + "score": 0.7, + "latex": "\\begin{array} { l } { { \\frac { D } { 2 } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 831, + 1698, + 848, + 1698, + 848, + 1716, + 831, + 1716 + ], + "score": 0.67, + "latex": "z" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1084.0, + 949.0, + 1084.0, + 949.0, + 1133.0, + 249.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 109.0, + 772.0, + 109.0, + 772.0, + 145.0, + 251.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 1862.0, + 824.0, + 1862.0, + 824.0, + 1901.0, + 787.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 487.0, + 1148.0, + 487.0, + 1148.0, + 522.0, + 249.0, + 522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1156.0, + 1452.0, + 1156.0, + 1452.0, + 1196.0, + 249.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1185.0, + 1450.0, + 1185.0, + 1450.0, + 1226.0, + 248.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1216.0, + 618.0, + 1216.0, + 618.0, + 1260.0, + 247.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 734.0, + 1216.0, + 1155.0, + 1216.0, + 1155.0, + 1260.0, + 734.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1175.0, + 1216.0, + 1452.0, + 1216.0, + 1452.0, + 1260.0, + 1175.0, + 1260.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1248.0, + 414.0, + 1248.0, + 414.0, + 1289.0, + 248.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 506.0, + 1248.0, + 1450.0, + 1248.0, + 1450.0, + 1289.0, + 506.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1282.0, + 1449.0, + 1282.0, + 1449.0, + 1315.0, + 249.0, + 1315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1310.0, + 1449.0, + 1310.0, + 1449.0, + 1346.0, + 247.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1340.0, + 1452.0, + 1340.0, + 1452.0, + 1378.0, + 249.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1371.0, + 1453.0, + 1371.0, + 1453.0, + 1408.0, + 248.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1403.0, + 573.0, + 1403.0, + 573.0, + 1440.0, + 251.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 816.0, + 1403.0, + 826.0, + 1403.0, + 826.0, + 1440.0, + 816.0, + 1440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 264.0, + 308.0, + 264.0, + 308.0, + 304.0, + 248.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 882.0, + 264.0, + 1330.0, + 264.0, + 1330.0, + 304.0, + 882.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 295.0, + 1098.0, + 295.0, + 1098.0, + 334.0, + 248.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1267.0, + 295.0, + 1315.0, + 295.0, + 1315.0, + 334.0, + 1267.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1397.0, + 295.0, + 1453.0, + 295.0, + 1453.0, + 334.0, + 1397.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 327.0, + 252.0, + 327.0, + 252.0, + 366.0, + 248.0, + 366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 414.0, + 327.0, + 458.0, + 327.0, + 458.0, + 366.0, + 414.0, + 366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 534.0, + 327.0, + 768.0, + 327.0, + 768.0, + 366.0, + 534.0, + 366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 327.0, + 1301.0, + 327.0, + 1301.0, + 366.0, + 1078.0, + 366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1446.0, + 327.0, + 1449.0, + 327.0, + 1449.0, + 366.0, + 1446.0, + 366.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 355.0, + 1350.0, + 355.0, + 1350.0, + 397.0, + 246.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1446.0, + 355.0, + 1450.0, + 355.0, + 1450.0, + 397.0, + 1446.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 388.0, + 1164.0, + 388.0, + 1164.0, + 424.0, + 249.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1185.0, + 388.0, + 1450.0, + 388.0, + 1450.0, + 424.0, + 1185.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 418.0, + 805.0, + 418.0, + 805.0, + 455.0, + 249.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 685.0, + 1236.0, + 685.0, + 1236.0, + 725.0, + 249.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1287.0, + 685.0, + 1449.0, + 685.0, + 1449.0, + 725.0, + 1287.0, + 725.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 245.0, + 713.0, + 459.0, + 713.0, + 459.0, + 762.0, + 245.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 512.0, + 713.0, + 1163.0, + 713.0, + 1163.0, + 762.0, + 512.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 713.0, + 1454.0, + 713.0, + 1454.0, + 762.0, + 1194.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 243.0, + 743.0, + 596.0, + 743.0, + 596.0, + 806.0, + 243.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 779.0, + 743.0, + 1456.0, + 743.0, + 1456.0, + 806.0, + 779.0, + 806.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 784.0, + 512.0, + 784.0, + 512.0, + 837.0, + 244.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 541.0, + 784.0, + 1454.0, + 784.0, + 1454.0, + 837.0, + 541.0, + 837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 824.0, + 476.0, + 824.0, + 476.0, + 859.0, + 249.0, + 859.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1656.0, + 1452.0, + 1656.0, + 1452.0, + 1693.0, + 250.0, + 1693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1685.0, + 830.0, + 1685.0, + 830.0, + 1724.0, + 249.0, + 1724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 849.0, + 1685.0, + 1451.0, + 1685.0, + 1451.0, + 1724.0, + 849.0, + 1724.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1717.0, + 373.0, + 1717.0, + 373.0, + 1755.0, + 249.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 1717.0, + 1451.0, + 1717.0, + 1451.0, + 1755.0, + 393.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1747.0, + 1173.0, + 1747.0, + 1173.0, + 1787.0, + 248.0, + 1787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1194.0, + 1747.0, + 1265.0, + 1747.0, + 1265.0, + 1787.0, + 1194.0, + 1787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1439.0, + 1747.0, + 1451.0, + 1747.0, + 1451.0, + 1787.0, + 1439.0, + 1787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1779.0, + 446.0, + 1779.0, + 446.0, + 1816.0, + 249.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 466.0, + 1779.0, + 535.0, + 1779.0, + 535.0, + 1816.0, + 466.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 680.0, + 1779.0, + 867.0, + 1779.0, + 867.0, + 1816.0, + 680.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 866.0, + 746.0, + 866.0, + 746.0, + 908.0, + 249.0, + 908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 798.0, + 866.0, + 939.0, + 866.0, + 939.0, + 908.0, + 798.0, + 908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 991.0, + 866.0, + 1450.0, + 866.0, + 1450.0, + 908.0, + 991.0, + 908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 898.0, + 317.0, + 898.0, + 317.0, + 939.0, + 248.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 493.0, + 898.0, + 1130.0, + 898.0, + 1130.0, + 939.0, + 493.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 898.0, + 1250.0, + 898.0, + 1250.0, + 939.0, + 1241.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1385.0, + 898.0, + 1450.0, + 898.0, + 1450.0, + 939.0, + 1385.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 932.0, + 252.0, + 932.0, + 252.0, + 976.0, + 248.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 505.0, + 932.0, + 554.0, + 932.0, + 554.0, + 976.0, + 505.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 679.0, + 932.0, + 1450.0, + 932.0, + 1450.0, + 976.0, + 679.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 965.0, + 327.0, + 965.0, + 327.0, + 1006.0, + 249.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 625.0, + 965.0, + 636.0, + 965.0, + 636.0, + 1006.0, + 625.0, + 1006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 541.0, + 1449.0, + 541.0, + 1449.0, + 582.0, + 249.0, + 582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 572.0, + 504.0, + 572.0, + 504.0, + 613.0, + 248.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 572.0, + 1388.0, + 572.0, + 1388.0, + 613.0, + 556.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1439.0, + 572.0, + 1449.0, + 572.0, + 1449.0, + 613.0, + 1439.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 603.0, + 578.0, + 603.0, + 578.0, + 642.0, + 250.0, + 642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 630.0, + 603.0, + 910.0, + 603.0, + 910.0, + 642.0, + 630.0, + 642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 603.0, + 1232.0, + 603.0, + 1232.0, + 642.0, + 962.0, + 642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 603.0, + 1451.0, + 603.0, + 1451.0, + 642.0, + 1378.0, + 642.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 632.0, + 386.0, + 632.0, + 386.0, + 675.0, + 248.0, + 675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 632.0, + 817.0, + 632.0, + 817.0, + 675.0, + 437.0, + 675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 992.0, + 632.0, + 1350.0, + 632.0, + 1350.0, + 675.0, + 992.0, + 675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1579.0, + 1450.0, + 1579.0, + 1450.0, + 1615.0, + 250.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1610.0, + 486.0, + 1610.0, + 486.0, + 1644.0, + 247.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1010.0, + 993.0, + 1010.0, + 993.0, + 1053.0, + 247.0, + 1053.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 487.0, + 1148.0, + 487.0, + 1148.0, + 522.0, + 249.0, + 522.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 16, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 250, + 1378, + 1449, + 1378, + 1449, + 1568, + 250, + 1568 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 252, + 725, + 1450, + 725, + 1450, + 882, + 252, + 882 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 251, + 1270, + 1455, + 1270, + 1455, + 1366, + 251, + 1366 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 251, + 1577, + 1448, + 1577, + 1448, + 1693, + 251, + 1693 + ], + "score": 0.973 + }, + { + "category_id": 8, + "poly": [ + 473, + 627, + 1228, + 627, + 1228, + 715, + 473, + 715 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 252, + 894, + 1449, + 894, + 1449, + 958, + 252, + 958 + ], + "score": 0.951 + }, + { + "category_id": 8, + "poly": [ + 307, + 1715, + 1347, + 1715, + 1347, + 1824, + 307, + 1824 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 247, + 429, + 1456, + 429, + 1456, + 494, + 247, + 494 + ], + "score": 0.946 + }, + { + "category_id": 8, + "poly": [ + 537, + 514, + 1164, + 514, + 1164, + 602, + 537, + 602 + ], + "score": 0.944 + }, + { + "category_id": 8, + "poly": [ + 408, + 1181, + 1291, + 1181, + 1291, + 1257, + 408, + 1257 + ], + "score": 0.942 + }, + { + "category_id": 1, + "poly": [ + 249, + 1053, + 1449, + 1053, + 1449, + 1120, + 249, + 1120 + ], + "score": 0.941 + }, + { + "category_id": 2, + "poly": [ + 253, + 110, + 770, + 110, + 770, + 142, + 253, + 142 + ], + "score": 0.931 + }, + { + "category_id": 1, + "poly": [ + 252, + 266, + 809, + 266, + 809, + 300, + 252, + 300 + ], + "score": 0.926 + }, + { + "category_id": 1, + "poly": [ + 256, + 1131, + 1149, + 1131, + 1149, + 1166, + 256, + 1166 + ], + "score": 0.924 + }, + { + "category_id": 0, + "poly": [ + 255, + 996, + 587, + 996, + 587, + 1029, + 255, + 1029 + ], + "score": 0.922 + }, + { + "category_id": 9, + "poly": [ + 1396, + 655, + 1445, + 655, + 1445, + 687, + 1396, + 687 + ], + "score": 0.903 + }, + { + "category_id": 8, + "poly": [ + 532, + 317, + 1170, + 317, + 1170, + 356, + 532, + 356 + ], + "score": 0.901 + }, + { + "category_id": 9, + "poly": [ + 1397, + 1202, + 1445, + 1202, + 1445, + 1233, + 1397, + 1233 + ], + "score": 0.898 + }, + { + "category_id": 9, + "poly": [ + 1396, + 541, + 1445, + 541, + 1445, + 573, + 1396, + 573 + ], + "score": 0.885 + }, + { + "category_id": 9, + "poly": [ + 1396, + 320, + 1446, + 320, + 1446, + 351, + 1396, + 351 + ], + "score": 0.878 + }, + { + "category_id": 2, + "poly": [ + 790, + 1865, + 820, + 1865, + 820, + 1890, + 790, + 1890 + ], + "score": 0.862 + }, + { + "category_id": 9, + "poly": [ + 1397, + 364, + 1446, + 364, + 1446, + 394, + 1397, + 394 + ], + "score": 0.862 + }, + { + "category_id": 9, + "poly": [ + 1397, + 1750, + 1446, + 1750, + 1446, + 1780, + 1397, + 1780 + ], + "score": 0.854 + }, + { + "category_id": 8, + "poly": [ + 369, + 361, + 1240, + 361, + 1240, + 399, + 369, + 399 + ], + "score": 0.786 + }, + { + "category_id": 8, + "poly": [ + 451, + 361, + 1335, + 361, + 1335, + 398, + 451, + 398 + ], + "score": 0.243 + }, + { + "category_id": 8, + "poly": [ + 369, + 361, + 1265, + 361, + 1265, + 399, + 369, + 399 + ], + "score": 0.101 + }, + { + "category_id": 14, + "poly": [ + 476, + 622, + 1228, + 622, + 1228, + 718, + 476, + 718 + ], + "score": 0.94, + "latex": "\\log p _ { \\theta , \\psi } ( x | y ) \\approx \\log \\frac { 1 } { S } \\sum _ { i = 1 } ^ { S } \\frac { p _ { \\theta } ( x | z _ { i } , y ) p _ { \\psi } ( z _ { i } | y ) } { q _ { \\phi } ( z _ { i } | x , y ) } , ~ z _ { i } \\sim q _ { \\phi } ( z | x , y )" + }, + { + "category_id": 13, + "poly": [ + 253, + 1613, + 558, + 1613, + 558, + 1668, + 253, + 1668 + ], + "score": 0.94, + "latex": "\\begin{array} { r } { q ( z = i | x ) = \\frac { f ( x , v _ { i } ) } { \\sum _ { j = 1 } ^ { K } f ( x , v _ { j } ) } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1100, + 1577, + 1232, + 1577, + 1232, + 1616, + 1100, + 1616 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { p _ { \\psi } ( z ) = \\frac { 1 } { K } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 729, + 1577, + 861, + 1577, + 861, + 1616, + 729, + 1616 + ], + "score": 0.93, + "latex": "\\begin{array} { r } { p _ { \\psi } ( z ) = \\frac { 1 } { K } } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 304, + 1711, + 1348, + 1711, + 1348, + 1822, + 304, + 1822 + ], + "score": 0.93, + "latex": "\\operatorname* { m a x } _ { v _ { 1 } , v _ { 2 } , \\ldots , v _ { K } } \\frac { 1 } { K } \\sum _ { i = 1 } ^ { K } \\left[ \\alpha \\sum _ { j = 1 } ^ { K } \\frac { f ( x _ { i } , v _ { j } ) } { \\sum _ { k = 1 } ^ { K } f ( x _ { i } , v _ { k } ) } \\log \\frac { f ( x _ { i } , v _ { j } ) \\frac { 1 } { K } } { \\frac { f ( x _ { i } , v _ { j } ) } { \\sum _ { k = 1 } ^ { K } f ( x _ { i } , v _ { k } ) } } + ( 1 - \\alpha ) \\sum _ { j = 1 } ^ { K } \\frac { 1 } { K } \\log f ( x _ { i } , v _ { j } ) \\right]" + }, + { + "category_id": 14, + "poly": [ + 539, + 509, + 1164, + 509, + 1164, + 606, + 539, + 606 + ], + "score": 0.93, + "latex": "\\log p _ { \\theta , \\psi } ( x | y ) \\approx \\log \\frac { 1 } { S } \\sum _ { i = 1 } ^ { S } p _ { \\theta } ( x | z _ { i } , y ) , ~ z _ { i } \\sim p _ { \\psi } ( z | y )" + }, + { + "category_id": 13, + "poly": [ + 357, + 1502, + 544, + 1502, + 544, + 1535, + 357, + 1535 + ], + "score": 0.93, + "latex": "\\begin{array} { r l r } { v _ { z } } & { { } = } & { \\bar { N } N _ { \\theta } ( z ) } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 364, + 314, + 1335, + 314, + 1335, + 402, + 364, + 402 + ], + "score": 0.93, + "latex": "\\begin{array} { r l } & { L _ { G S N N } ( x , b ; \\theta , \\psi ) = \\mathbb { E } _ { p _ { \\psi } ( z | x _ { 1 - b } , b ) } \\log p _ { \\theta } ( x _ { b } | z , x _ { 1 - b } , b ) } \\\\ & { L ( x , b ; \\theta , \\psi , \\phi ) = \\alpha L _ { V A E A C } ( x , b ; \\theta , \\psi , \\phi ) + ( 1 - \\alpha ) L _ { G S N N } ( x , b ; \\theta , \\psi ) , \\quad \\alpha \\in [ 0 , 1 ] } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 407, + 1181, + 1292, + 1181, + 1292, + 1257, + 407, + 1257 + ], + "score": 0.92, + "latex": "L ( x ; \\phi , \\psi , \\theta ) = \\alpha \\mathbb { E } _ { z \\sim q _ { \\phi } ( z \\mid x ) } \\log \\frac { p _ { \\theta } ( x \\mid z ) p _ { \\psi } ( z ) } { q _ { \\phi } ( z \\mid x ) } + ( 1 - \\alpha ) \\mathbb { E } _ { z \\sim p _ { \\psi } ( z ) } \\log p _ { \\theta } ( x \\mid z )" + }, + { + "category_id": 13, + "poly": [ + 456, + 925, + 532, + 925, + 532, + 959, + 456, + 959 + ], + "score": 0.92, + "latex": "p ( x | y )" + }, + { + "category_id": 13, + "poly": [ + 292, + 1442, + 510, + 1442, + 510, + 1475, + 292, + 1475 + ], + "score": 0.92, + "latex": "i \\in \\{ 1 , 2 , \\ldots , K \\}" + }, + { + "category_id": 13, + "poly": [ + 494, + 1412, + 560, + 1412, + 560, + 1444, + 494, + 1444 + ], + "score": 0.92, + "latex": "p _ { d } ( x )" + }, + { + "category_id": 13, + "poly": [ + 1251, + 1409, + 1445, + 1409, + 1445, + 1447, + 1251, + 1447 + ], + "score": 0.91, + "latex": "\\begin{array} { r } { p _ { d } ( x = x _ { i } ) = \\frac { 1 } { K } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 252, + 1533, + 483, + 1533, + 483, + 1567, + 252, + 1567 + ], + "score": 0.91, + "latex": "p _ { \\theta } ( x | z ) \\ : = \\ : f ( x , v _ { z } )" + }, + { + "category_id": 13, + "poly": [ + 1006, + 1537, + 1168, + 1537, + 1168, + 1565, + 1006, + 1565 + ], + "score": 0.89, + "latex": "v _ { 1 } , v _ { 2 } , \\dotsc , v _ { K }" + }, + { + "category_id": 13, + "poly": [ + 432, + 1308, + 490, + 1308, + 490, + 1335, + 432, + 1335 + ], + "score": 0.89, + "latex": "x _ { 1 - b }" + }, + { + "category_id": 13, + "poly": [ + 766, + 1474, + 829, + 1474, + 829, + 1503, + 766, + 1503 + ], + "score": 0.88, + "latex": "N N _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 707, + 1447, + 877, + 1447, + 877, + 1473, + 707, + 1473 + ], + "score": 0.86, + "latex": "x _ { 1 } , x _ { 2 } , \\dotsc , x _ { K }" + }, + { + "category_id": 13, + "poly": [ + 1183, + 1277, + 1214, + 1277, + 1214, + 1304, + 1183, + 1304 + ], + "score": 0.84, + "latex": "x _ { b }" + }, + { + "category_id": 13, + "poly": [ + 1141, + 1625, + 1171, + 1625, + 1171, + 1653, + 1141, + 1653 + ], + "score": 0.84, + "latex": "p _ { \\theta }" + }, + { + "category_id": 13, + "poly": [ + 1012, + 1382, + 1042, + 1382, + 1042, + 1408, + 1012, + 1408 + ], + "score": 0.83, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 1119, + 789, + 1140, + 789, + 1140, + 816, + 1119, + 816 + ], + "score": 0.82, + "latex": "S" + }, + { + "category_id": 13, + "poly": [ + 1200, + 1443, + 1229, + 1443, + 1229, + 1469, + 1200, + 1469 + ], + "score": 0.82, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 789, + 1534, + 806, + 1534, + 806, + 1560, + 789, + 1560 + ], + "score": 0.8, + "latex": "\\theta" + }, + { + "category_id": 13, + "poly": [ + 813, + 436, + 831, + 436, + 831, + 458, + 813, + 458 + ], + "score": 0.78, + "latex": "z" + }, + { + "category_id": 13, + "poly": [ + 1172, + 1479, + 1191, + 1479, + 1191, + 1500, + 1172, + 1500 + ], + "score": 0.77, + "latex": "z" + }, + { + "category_id": 13, + "poly": [ + 1102, + 1279, + 1122, + 1279, + 1122, + 1300, + 1102, + 1300 + ], + "score": 0.76, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 732, + 1387, + 751, + 1387, + 751, + 1408, + 732, + 1408 + ], + "score": 0.76, + "latex": "z" + }, + { + "category_id": 13, + "poly": [ + 1181, + 1387, + 1201, + 1387, + 1201, + 1408, + 1181, + 1408 + ], + "score": 0.76, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 541, + 1305, + 556, + 1305, + 556, + 1331, + 541, + 1331 + ], + "score": 0.72, + "latex": "b" + }, + { + "category_id": 13, + "poly": [ + 1055, + 1627, + 1087, + 1627, + 1087, + 1655, + 1055, + 1655 + ], + "score": 0.64, + "latex": "p _ { \\psi }" + }, + { + "category_id": 13, + "poly": [ + 1013, + 1627, + 1041, + 1627, + 1041, + 1655, + 1013, + 1655 + ], + "score": 0.61, + "latex": "q _ { \\phi }" + }, + { + "category_id": 13, + "poly": [ + 1013, + 1625, + 1088, + 1625, + 1088, + 1655, + 1013, + 1655 + ], + "score": 0.3, + "latex": "q _ { \\phi } , p _ { \\psi }" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 110.0, + 773.0, + 110.0, + 773.0, + 145.0, + 251.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 994.0, + 590.0, + 994.0, + 590.0, + 1033.0, + 250.0, + 1033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 1862.0, + 825.0, + 1862.0, + 825.0, + 1899.0, + 787.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1378.0, + 731.0, + 1378.0, + 731.0, + 1411.0, + 251.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 752.0, + 1378.0, + 1011.0, + 1378.0, + 1011.0, + 1411.0, + 752.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1043.0, + 1378.0, + 1180.0, + 1378.0, + 1180.0, + 1411.0, + 1043.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1202.0, + 1378.0, + 1449.0, + 1378.0, + 1449.0, + 1411.0, + 1202.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1395.0, + 291.0, + 1395.0, + 291.0, + 1479.0, + 239.0, + 1479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 1395.0, + 706.0, + 1395.0, + 706.0, + 1479.0, + 561.0, + 1479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 878.0, + 1395.0, + 1199.0, + 1395.0, + 1199.0, + 1479.0, + 878.0, + 1479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1230.0, + 1395.0, + 1250.0, + 1395.0, + 1250.0, + 1479.0, + 1230.0, + 1479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1446.0, + 1395.0, + 1460.0, + 1395.0, + 1460.0, + 1479.0, + 1446.0, + 1479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1472.0, + 765.0, + 1472.0, + 765.0, + 1508.0, + 249.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 830.0, + 1472.0, + 1171.0, + 1472.0, + 1171.0, + 1508.0, + 830.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1192.0, + 1472.0, + 1448.0, + 1472.0, + 1448.0, + 1508.0, + 1192.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1500.0, + 356.0, + 1500.0, + 356.0, + 1539.0, + 246.0, + 1539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 545.0, + 1500.0, + 1452.0, + 1500.0, + 1452.0, + 1539.0, + 545.0, + 1539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1530.0, + 251.0, + 1530.0, + 251.0, + 1573.0, + 246.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 1530.0, + 788.0, + 1530.0, + 788.0, + 1573.0, + 484.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 1530.0, + 1005.0, + 1530.0, + 1005.0, + 1573.0, + 807.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 1530.0, + 1181.0, + 1530.0, + 1181.0, + 1573.0, + 1169.0, + 1573.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 724.0, + 1449.0, + 724.0, + 1449.0, + 761.0, + 251.0, + 761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 758.0, + 1450.0, + 758.0, + 1450.0, + 791.0, + 249.0, + 791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 789.0, + 1118.0, + 789.0, + 1118.0, + 820.0, + 248.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1141.0, + 789.0, + 1451.0, + 789.0, + 1451.0, + 820.0, + 1141.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 814.0, + 1451.0, + 814.0, + 1451.0, + 853.0, + 247.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 845.0, + 790.0, + 845.0, + 790.0, + 887.0, + 248.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 253.0, + 1272.0, + 1101.0, + 1272.0, + 1101.0, + 1306.0, + 253.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 1272.0, + 1182.0, + 1272.0, + 1182.0, + 1306.0, + 1123.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 1272.0, + 1448.0, + 1272.0, + 1448.0, + 1306.0, + 1215.0, + 1306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1300.0, + 431.0, + 1300.0, + 431.0, + 1342.0, + 249.0, + 1342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 491.0, + 1300.0, + 540.0, + 1300.0, + 540.0, + 1342.0, + 491.0, + 1342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 557.0, + 1300.0, + 1452.0, + 1300.0, + 1452.0, + 1342.0, + 557.0, + 1342.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1330.0, + 448.0, + 1330.0, + 448.0, + 1369.0, + 250.0, + 1369.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 1574.0, + 728.0, + 1574.0, + 728.0, + 1619.0, + 244.0, + 1619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 862.0, + 1574.0, + 1099.0, + 1574.0, + 1099.0, + 1619.0, + 862.0, + 1619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 1574.0, + 1452.0, + 1574.0, + 1452.0, + 1619.0, + 1233.0, + 1619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1616.0, + 252.0, + 1616.0, + 252.0, + 1660.0, + 249.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 567.0, + 1616.0, + 1012.0, + 1616.0, + 1012.0, + 1657.0, + 567.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 1616.0, + 1140.0, + 1616.0, + 1140.0, + 1657.0, + 1089.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 1616.0, + 1448.0, + 1616.0, + 1448.0, + 1657.0, + 1172.0, + 1657.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1663.0, + 508.0, + 1663.0, + 508.0, + 1697.0, + 251.0, + 1697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 412.25, + 1628.0, + 564.25, + 1628.0, + 564.25, + 1673.0, + 412.25, + 1673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 893.0, + 1448.0, + 893.0, + 1448.0, + 929.0, + 252.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 925.0, + 455.0, + 925.0, + 455.0, + 961.0, + 251.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 925.0, + 1005.0, + 925.0, + 1005.0, + 961.0, + 533.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 429.0, + 812.0, + 429.0, + 812.0, + 465.0, + 249.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 429.0, + 1450.0, + 429.0, + 1450.0, + 465.0, + 832.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 459.0, + 770.0, + 459.0, + 770.0, + 499.0, + 249.0, + 499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 245.0, + 1050.0, + 1454.0, + 1050.0, + 1454.0, + 1092.0, + 245.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1084.0, + 505.0, + 1084.0, + 505.0, + 1124.0, + 248.0, + 1124.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 265.0, + 809.0, + 265.0, + 809.0, + 302.0, + 252.0, + 302.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1130.0, + 1147.0, + 1130.0, + 1147.0, + 1171.0, + 248.0, + 1171.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 17, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 251, + 735, + 1449, + 735, + 1449, + 927, + 251, + 927 + ], + "score": 0.98 + }, + { + "category_id": 1, + "poly": [ + 253, + 941, + 1448, + 941, + 1448, + 1097, + 253, + 1097 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 253, + 1464, + 1449, + 1464, + 1449, + 1627, + 253, + 1627 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 253, + 1247, + 1447, + 1247, + 1447, + 1372, + 253, + 1372 + ], + "score": 0.976 + }, + { + "category_id": 1, + "poly": [ + 253, + 1109, + 1447, + 1109, + 1447, + 1235, + 253, + 1235 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 253, + 1641, + 1449, + 1641, + 1449, + 1736, + 253, + 1736 + ], + "score": 0.973 + }, + { + "category_id": 5, + "poly": [ + 641, + 377, + 1060, + 377, + 1060, + 507, + 641, + 507 + ], + "score": 0.96, + "html": "
VAEAC weightIS-10MC-10
a=10.2285
α = 0.990.3511
α = 0.90.621.7
" + }, + { + "category_id": 8, + "poly": [ + 452, + 602, + 1242, + 602, + 1242, + 711, + 452, + 711 + ], + "score": 0.959 + }, + { + "category_id": 1, + "poly": [ + 250, + 1748, + 1447, + 1748, + 1447, + 1812, + 250, + 1812 + ], + "score": 0.939 + }, + { + "category_id": 1, + "poly": [ + 248, + 558, + 1133, + 558, + 1133, + 593, + 248, + 593 + ], + "score": 0.931 + }, + { + "category_id": 2, + "poly": [ + 253, + 111, + 770, + 111, + 770, + 142, + 253, + 142 + ], + "score": 0.925 + }, + { + "category_id": 1, + "poly": [ + 251, + 259, + 1449, + 259, + 1449, + 354, + 251, + 354 + ], + "score": 0.916 + }, + { + "category_id": 0, + "poly": [ + 255, + 1408, + 537, + 1408, + 537, + 1441, + 255, + 1441 + ], + "score": 0.911 + }, + { + "category_id": 9, + "poly": [ + 1397, + 641, + 1445, + 641, + 1445, + 673, + 1397, + 673 + ], + "score": 0.903 + }, + { + "category_id": 2, + "poly": [ + 791, + 1865, + 820, + 1865, + 820, + 1890, + 791, + 1890 + ], + "score": 0.848 + }, + { + "category_id": 14, + "poly": [ + 452, + 601, + 1252, + 601, + 1252, + 713, + 452, + 713 + ], + "score": 0.93, + "latex": "\\operatorname* { m a x } _ { v _ { 1 } , v _ { 2 } , \\ldots , v _ { K } } \\sum _ { i = 1 } ^ { K } \\left[ \\alpha \\log \\frac { \\sum _ { j = 1 } ^ { K } f ( x _ { i } , v _ { j } ) } { K } + ( 1 - \\alpha ) \\sum _ { j = 1 } ^ { K } \\frac { 1 } { K } \\log f ( x _ { i } , v _ { j } ) \\right]" + }, + { + "category_id": 13, + "poly": [ + 635, + 835, + 718, + 835, + 718, + 867, + 635, + 867 + ], + "score": 0.93, + "latex": "f ( x , v )" + }, + { + "category_id": 13, + "poly": [ + 252, + 1671, + 389, + 1671, + 389, + 1706, + 252, + 1706 + ], + "score": 0.92, + "latex": "p _ { \\psi , \\theta } ( x _ { 1 } | x _ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 661, + 1703, + 726, + 1703, + 726, + 1736, + 661, + 1736 + ], + "score": 0.92, + "latex": "p ( x _ { 1 } )" + }, + { + "category_id": 13, + "poly": [ + 451, + 1671, + 588, + 1671, + 588, + 1705, + 451, + 1705 + ], + "score": 0.92, + "latex": "p _ { \\psi , \\theta } ( x _ { 2 } | x _ { 1 } )" + }, + { + "category_id": 13, + "poly": [ + 994, + 1533, + 1049, + 1533, + 1049, + 1567, + 994, + 1567 + ], + "score": 0.92, + "latex": "p ( x )" + }, + { + "category_id": 13, + "poly": [ + 1297, + 1641, + 1437, + 1641, + 1437, + 1675, + 1297, + 1675 + ], + "score": 0.91, + "latex": "p _ { \\psi , \\theta } ( x _ { 1 } , x _ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 586, + 1531, + 784, + 1531, + 784, + 1565, + 586, + 1565 + ], + "score": 0.91, + "latex": "\\mu _ { i } \\sim \\mathcal N ( \\mu _ { i } | 0 , I )" + }, + { + "category_id": 13, + "poly": [ + 543, + 1705, + 609, + 1705, + 609, + 1736, + 543, + 1736 + ], + "score": 0.91, + "latex": "p ( x _ { 2 } )" + }, + { + "category_id": 13, + "poly": [ + 816, + 1495, + 908, + 1495, + 908, + 1527, + 816, + 1527 + ], + "score": 0.91, + "latex": "x \\in \\mathbb { R } ^ { \\bar { 2 } }" + }, + { + "category_id": 13, + "poly": [ + 680, + 1565, + 747, + 1565, + 747, + 1597, + 680, + 1597 + ], + "score": 0.91, + "latex": "p _ { d } ( x )" + }, + { + "category_id": 13, + "poly": [ + 322, + 774, + 403, + 774, + 403, + 802, + 322, + 802 + ], + "score": 0.9, + "latex": "\\alpha = 0" + }, + { + "category_id": 13, + "poly": [ + 590, + 739, + 661, + 739, + 661, + 766, + 590, + 766 + ], + "score": 0.9, + "latex": "\\alpha = 1" + }, + { + "category_id": 13, + "poly": [ + 826, + 870, + 996, + 870, + 996, + 897, + 826, + 897 + ], + "score": 0.9, + "latex": "x _ { 1 } , x _ { 2 } , \\dotsc , x _ { K }" + }, + { + "category_id": 13, + "poly": [ + 964, + 1496, + 1328, + 1496, + 1328, + 1531, + 964, + 1531 + ], + "score": 0.89, + "latex": "p ( \\bar { b } _ { 1 } = 1 ) = p ( b _ { 2 } = 1 ) = 0 . 5" + }, + { + "category_id": 13, + "poly": [ + 253, + 1528, + 502, + 1528, + 502, + 1569, + 253, + 1569 + ], + "score": 0.89, + "latex": "\\begin{array} { r l } { \\frac { 1 } { 8 } \\sum _ { i = 1 } ^ { 8 } \\mathcal { N } ( x | \\mu _ { i } , \\frac { 1 } { 1 0 } I ) } & { { } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1055, + 768, + 1258, + 768, + 1258, + 808, + 1055, + 808 + ], + "score": 0.85, + "latex": "\\textstyle \\sum _ { i = 1 } ^ { K } \\log f ( x _ { i } , v )" + }, + { + "category_id": 13, + "poly": [ + 399, + 866, + 419, + 866, + 419, + 896, + 399, + 896 + ], + "score": 0.85, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 1277, + 974, + 1306, + 974, + 1306, + 1000, + 1277, + 1000 + ], + "score": 0.83, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 1344, + 1497, + 1447, + 1497, + 1447, + 1531, + 1344, + 1531 + ], + "score": 0.82, + "latex": "p _ { d } ( x ) =" + }, + { + "category_id": 13, + "poly": [ + 594, + 323, + 616, + 323, + 616, + 350, + 594, + 350 + ], + "score": 0.8, + "latex": "S" + }, + { + "category_id": 13, + "poly": [ + 1312, + 944, + 1342, + 944, + 1342, + 971, + 1312, + 971 + ], + "score": 0.8, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 1025, + 1256, + 1047, + 1256, + 1047, + 1277, + 1025, + 1277 + ], + "score": 0.78, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 771, + 293, + 794, + 293, + 794, + 319, + 771, + 319 + ], + "score": 0.77, + "latex": "S" + }, + { + "category_id": 13, + "poly": [ + 974, + 840, + 993, + 840, + 993, + 862, + 974, + 862 + ], + "score": 0.76, + "latex": "x" + }, + { + "category_id": 13, + "poly": [ + 641, + 768, + 1259, + 768, + 1259, + 809, + 641, + 809 + ], + "score": 0.76, + "latex": "v _ { 1 } = v _ { 2 } = \\cdots = v _ { K } = \\operatorname { a r g m a x } _ { v } \\sum _ { i = 1 } ^ { K } \\log f ( x _ { i } , v ) ." + }, + { + "category_id": 13, + "poly": [ + 649, + 1648, + 671, + 1648, + 671, + 1669, + 649, + 1669 + ], + "score": 0.75, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 763, + 811, + 782, + 811, + 782, + 832, + 763, + 832 + ], + "score": 0.75, + "latex": "z" + }, + { + "category_id": 13, + "poly": [ + 1272, + 1039, + 1291, + 1039, + 1291, + 1061, + 1272, + 1061 + ], + "score": 0.74, + "latex": "z" + }, + { + "category_id": 13, + "poly": [ + 1345, + 1286, + 1367, + 1286, + 1367, + 1309, + 1345, + 1309 + ], + "score": 0.74, + "latex": "\\alpha" + }, + { + "category_id": 13, + "poly": [ + 1177, + 292, + 1199, + 292, + 1199, + 320, + 1177, + 320 + ], + "score": 0.73, + "latex": "S" + }, + { + "category_id": 13, + "poly": [ + 252, + 1039, + 270, + 1039, + 270, + 1061, + 252, + 1061 + ], + "score": 0.73, + "latex": "z" + }, + { + "category_id": 13, + "poly": [ + 483, + 870, + 502, + 870, + 502, + 892, + 483, + 892 + ], + "score": 0.72, + "latex": "v" + }, + { + "category_id": 13, + "poly": [ + 1107, + 812, + 1126, + 812, + 1126, + 832, + 1107, + 832 + ], + "score": 0.71, + "latex": "v" + }, + { + "category_id": 13, + "poly": [ + 1240, + 261, + 1262, + 261, + 1262, + 289, + 1240, + 289 + ], + "score": 0.69, + "latex": "S" + }, + { + "category_id": 13, + "poly": [ + 1337, + 779, + 1357, + 779, + 1357, + 801, + 1337, + 801 + ], + "score": 0.68, + "latex": "z" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 109.0, + 772.0, + 109.0, + 772.0, + 145.0, + 251.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1406.0, + 541.0, + 1406.0, + 541.0, + 1445.0, + 250.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 1863.0, + 825.0, + 1863.0, + 825.0, + 1902.0, + 787.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 737.0, + 589.0, + 737.0, + 589.0, + 770.0, + 250.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 662.0, + 737.0, + 1448.0, + 737.0, + 1448.0, + 770.0, + 662.0, + 770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 753.0, + 321.0, + 753.0, + 321.0, + 840.0, + 238.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 753.0, + 640.0, + 753.0, + 640.0, + 840.0, + 404.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 753.0, + 1336.0, + 753.0, + 1336.0, + 840.0, + 1260.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1358.0, + 753.0, + 1464.0, + 753.0, + 1464.0, + 840.0, + 1358.0, + 840.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 834.0, + 634.0, + 834.0, + 634.0, + 870.0, + 250.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 719.0, + 834.0, + 973.0, + 834.0, + 973.0, + 870.0, + 719.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 834.0, + 1450.0, + 834.0, + 1450.0, + 870.0, + 994.0, + 870.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 862.0, + 398.0, + 862.0, + 398.0, + 903.0, + 248.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 420.0, + 862.0, + 482.0, + 862.0, + 482.0, + 903.0, + 420.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 862.0, + 825.0, + 862.0, + 825.0, + 903.0, + 503.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 997.0, + 862.0, + 1452.0, + 862.0, + 1452.0, + 903.0, + 997.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 895.0, + 712.0, + 895.0, + 712.0, + 931.0, + 250.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 944.0, + 1311.0, + 944.0, + 1311.0, + 974.0, + 252.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1343.0, + 944.0, + 1449.0, + 944.0, + 1449.0, + 974.0, + 1343.0, + 974.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 971.0, + 1276.0, + 971.0, + 1276.0, + 1007.0, + 249.0, + 1007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1307.0, + 971.0, + 1449.0, + 971.0, + 1449.0, + 1007.0, + 1307.0, + 1007.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1002.0, + 1449.0, + 1002.0, + 1449.0, + 1038.0, + 246.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1032.0, + 251.0, + 1032.0, + 251.0, + 1070.0, + 248.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 271.0, + 1032.0, + 1271.0, + 1032.0, + 1271.0, + 1070.0, + 271.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1292.0, + 1032.0, + 1451.0, + 1032.0, + 1451.0, + 1070.0, + 1292.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1064.0, + 814.0, + 1064.0, + 814.0, + 1098.0, + 249.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1466.0, + 1449.0, + 1466.0, + 1449.0, + 1500.0, + 249.0, + 1500.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1493.0, + 815.0, + 1493.0, + 815.0, + 1535.0, + 248.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 909.0, + 1493.0, + 963.0, + 1493.0, + 963.0, + 1535.0, + 909.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1329.0, + 1493.0, + 1343.0, + 1493.0, + 1343.0, + 1535.0, + 1329.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1448.0, + 1493.0, + 1452.0, + 1493.0, + 1452.0, + 1535.0, + 1448.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1513.0, + 252.0, + 1513.0, + 252.0, + 1598.0, + 238.0, + 1598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 1513.0, + 585.0, + 1513.0, + 585.0, + 1598.0, + 503.0, + 1598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 1513.0, + 993.0, + 1513.0, + 993.0, + 1598.0, + 785.0, + 1598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 1513.0, + 1461.0, + 1513.0, + 1461.0, + 1598.0, + 1050.0, + 1598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1592.0, + 914.0, + 1592.0, + 914.0, + 1628.0, + 249.0, + 1628.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1249.0, + 1024.0, + 1249.0, + 1024.0, + 1284.0, + 250.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1048.0, + 1249.0, + 1451.0, + 1249.0, + 1451.0, + 1284.0, + 1048.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1280.0, + 1344.0, + 1280.0, + 1344.0, + 1315.0, + 250.0, + 1315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1368.0, + 1280.0, + 1450.0, + 1280.0, + 1450.0, + 1315.0, + 1368.0, + 1315.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1308.0, + 1447.0, + 1308.0, + 1447.0, + 1346.0, + 249.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1339.0, + 680.0, + 1339.0, + 680.0, + 1376.0, + 249.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1110.0, + 1450.0, + 1110.0, + 1450.0, + 1145.0, + 250.0, + 1145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1141.0, + 1450.0, + 1141.0, + 1450.0, + 1176.0, + 249.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1169.0, + 1451.0, + 1169.0, + 1451.0, + 1207.0, + 249.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1200.0, + 479.0, + 1200.0, + 479.0, + 1240.0, + 248.0, + 1240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1632.0, + 648.0, + 1632.0, + 648.0, + 1682.0, + 246.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 1632.0, + 1296.0, + 1632.0, + 1296.0, + 1682.0, + 672.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1438.0, + 1632.0, + 1450.0, + 1632.0, + 1450.0, + 1682.0, + 1438.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1669.0, + 251.0, + 1669.0, + 251.0, + 1710.0, + 246.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 1669.0, + 450.0, + 1669.0, + 450.0, + 1710.0, + 390.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 589.0, + 1669.0, + 1450.0, + 1669.0, + 1450.0, + 1710.0, + 589.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1699.0, + 542.0, + 1699.0, + 542.0, + 1740.0, + 249.0, + 1740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 1699.0, + 660.0, + 1699.0, + 660.0, + 1740.0, + 610.0, + 1740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 727.0, + 1699.0, + 874.0, + 1699.0, + 874.0, + 1740.0, + 727.0, + 1740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1744.0, + 1450.0, + 1744.0, + 1450.0, + 1788.0, + 249.0, + 1788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1779.0, + 1448.0, + 1779.0, + 1448.0, + 1815.0, + 250.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 554.0, + 1136.0, + 554.0, + 1136.0, + 600.0, + 247.0, + 600.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 258.0, + 1239.0, + 258.0, + 1239.0, + 296.0, + 251.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1263.0, + 258.0, + 1448.0, + 258.0, + 1448.0, + 296.0, + 1263.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 291.0, + 770.0, + 291.0, + 770.0, + 325.0, + 251.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 291.0, + 1176.0, + 291.0, + 1176.0, + 325.0, + 795.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1200.0, + 291.0, + 1448.0, + 291.0, + 1448.0, + 325.0, + 1200.0, + 325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 320.0, + 593.0, + 320.0, + 593.0, + 354.0, + 251.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 617.0, + 320.0, + 943.0, + 320.0, + 943.0, + 354.0, + 617.0, + 354.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 18, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 297, + 945, + 1422, + 945, + 1422, + 1450, + 297, + 1450 + ], + "score": 0.954 + }, + { + "category_id": 4, + "poly": [ + 250, + 618, + 713, + 618, + 713, + 682, + 250, + 682 + ], + "score": 0.941 + }, + { + "category_id": 3, + "poly": [ + 918, + 264, + 1425, + 264, + 1425, + 766, + 918, + 766 + ], + "score": 0.94 + }, + { + "category_id": 1, + "poly": [ + 253, + 1670, + 1446, + 1670, + 1446, + 1734, + 253, + 1734 + ], + "score": 0.934 + }, + { + "category_id": 1, + "poly": [ + 251, + 1748, + 1440, + 1748, + 1440, + 1812, + 251, + 1812 + ], + "score": 0.924 + }, + { + "category_id": 2, + "poly": [ + 252, + 112, + 770, + 112, + 770, + 141, + 252, + 141 + ], + "score": 0.901 + }, + { + "category_id": 4, + "poly": [ + 967, + 792, + 1375, + 792, + 1375, + 826, + 967, + 826 + ], + "score": 0.897 + }, + { + "category_id": 2, + "poly": [ + 789, + 1865, + 821, + 1865, + 821, + 1890, + 789, + 1890 + ], + "score": 0.854 + }, + { + "category_id": 3, + "poly": [ + 372, + 411, + 592, + 411, + 592, + 589, + 372, + 589 + ], + "score": 0.479 + }, + { + "category_id": 4, + "poly": [ + 271, + 1492, + 1407, + 1492, + 1407, + 1557, + 271, + 1557 + ], + "score": 0.372 + }, + { + "category_id": 3, + "poly": [ + 918, + 263, + 1426, + 263, + 1426, + 765, + 918, + 765 + ], + "score": 0.093 + }, + { + "category_id": 13, + "poly": [ + 459, + 1750, + 536, + 1750, + 536, + 1781, + 459, + 1781 + ], + "score": 0.9, + "latex": "\\alpha \\neq 1" + }, + { + "category_id": 13, + "poly": [ + 252, + 1781, + 324, + 1781, + 324, + 1807, + 252, + 1807 + ], + "score": 0.89, + "latex": "\\alpha = 1" + }, + { + "category_id": 13, + "poly": [ + 252, + 1703, + 346, + 1703, + 346, + 1732, + 252, + 1732 + ], + "score": 0.88, + "latex": "\\alpha = 0 . 9" + }, + { + "category_id": 13, + "poly": [ + 442, + 1783, + 514, + 1783, + 514, + 1808, + 442, + 1808 + ], + "score": 0.88, + "latex": "\\alpha \\approx 1" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 957.0, + 814.0, + 957.0, + 814.0, + 1358.0, + 769.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 950.0, + 962.0, + 995.0, + 962.0, + 995.0, + 1341.0, + 950.0, + 1341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 516.0, + 963.0, + 561.0, + 963.0, + 561.0, + 1184.0, + 516.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 983.0, + 387.0, + 983.0, + 387.0, + 1228.0, + 347.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 995.0, + 983.0, + 1036.0, + 983.0, + 1036.0, + 1258.0, + 995.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 690.0, + 988.0, + 723.0, + 988.0, + 723.0, + 1258.0, + 690.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 732.0, + 991.0, + 764.0, + 991.0, + 764.0, + 1340.0, + 732.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 561.0, + 993.0, + 602.0, + 993.0, + 602.0, + 1162.0, + 561.0, + 1162.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 436.0, + 994.0, + 469.0, + 994.0, + 469.0, + 1184.0, + 436.0, + 1184.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 477.0, + 995.0, + 514.0, + 995.0, + 514.0, + 1206.0, + 477.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 603.0, + 995.0, + 640.0, + 995.0, + 640.0, + 1262.0, + 603.0, + 1262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1381.0, + 995.0, + 1418.0, + 995.0, + 1418.0, + 1339.0, + 1381.0, + 1339.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 649.0, + 996.0, + 681.0, + 996.0, + 681.0, + 1345.0, + 649.0, + 1345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 996.0, + 1070.0, + 996.0, + 1070.0, + 1206.0, + 1041.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 401.0, + 1002.0, + 414.0, + 1002.0, + 414.0, + 1025.0, + 401.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1351.0, + 1058.0, + 1364.0, + 1058.0, + 1364.0, + 1090.0, + 1351.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 1071.0, + 426.0, + 1071.0, + 426.0, + 1182.0, + 392.0, + 1182.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1097.0, + 1099.0, + 1106.0, + 1099.0, + 1106.0, + 1113.0, + 1097.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1351.0, + 1088.0, + 1364.0, + 1088.0, + 1364.0, + 1123.0, + 1351.0, + 1123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1092.0, + 1128.0, + 1107.0, + 1128.0, + 1107.0, + 1146.0, + 1092.0, + 1146.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1351.0, + 1123.0, + 1364.0, + 1123.0, + 1364.0, + 1146.0, + 1351.0, + 1146.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1214.0, + 373.0, + 1214.0, + 373.0, + 1242.0, + 361.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 401.0, + 1203.0, + 414.0, + 1203.0, + 414.0, + 1242.0, + 401.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 1180.0, + 459.0, + 1180.0, + 459.0, + 1233.0, + 440.0, + 1233.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 484.0, + 1190.0, + 510.0, + 1190.0, + 510.0, + 1242.0, + 484.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 1200.0, + 551.0, + 1200.0, + 551.0, + 1255.0, + 525.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1185.0, + 588.0, + 1185.0, + 588.0, + 1257.0, + 575.0, + 1257.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1353.0, + 1216.0, + 1360.0, + 1216.0, + 1360.0, + 1223.0, + 1353.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 612.0, + 1288.0, + 630.0, + 1288.0, + 630.0, + 1332.0, + 612.0, + 1332.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 1302.0, + 371.0, + 1302.0, + 371.0, + 1321.0, + 362.0, + 1321.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1302.0, + 548.0, + 1302.0, + 548.0, + 1358.0, + 526.0, + 1358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 482.0, + 1411.0, + 613.0, + 1411.0, + 613.0, + 1452.0, + 482.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1093.0, + 1411.0, + 1214.0, + 1411.0, + 1214.0, + 1449.0, + 1093.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 615.0, + 712.0, + 615.0, + 712.0, + 653.0, + 250.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 649.0, + 553.0, + 649.0, + 553.0, + 683.0, + 250.0, + 683.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1008.0, + 264.0, + 1106.0, + 264.0, + 1106.0, + 287.0, + 1008.0, + 287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1152.0, + 264.0, + 1249.0, + 264.0, + 1249.0, + 287.0, + 1152.0, + 287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1015.0, + 282.0, + 1097.0, + 282.0, + 1097.0, + 305.0, + 1015.0, + 305.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1161.0, + 280.0, + 1241.0, + 280.0, + 1241.0, + 306.0, + 1161.0, + 306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1281.0, + 280.0, + 1407.0, + 280.0, + 1407.0, + 304.0, + 1281.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 299.0, + 1114.0, + 299.0, + 1114.0, + 323.0, + 998.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1141.0, + 299.0, + 1259.0, + 299.0, + 1259.0, + 323.0, + 1141.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1272.0, + 299.0, + 1413.0, + 299.0, + 1413.0, + 323.0, + 1272.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 338.0, + 998.0, + 338.0, + 998.0, + 349.0, + 985.0, + 349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 338.0, + 1142.0, + 338.0, + 1142.0, + 350.0, + 1130.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 337.0, + 1286.0, + 337.0, + 1286.0, + 350.0, + 1273.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.0, + 364.0, + 970.0, + 364.0, + 970.0, + 383.0, + 927.0, + 383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 372.0, + 1000.0, + 372.0, + 1000.0, + 388.0, + 984.0, + 388.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 373.0, + 1143.0, + 373.0, + 1143.0, + 387.0, + 1130.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 374.0, + 1286.0, + 374.0, + 1286.0, + 387.0, + 1273.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1403.0, + 378.0, + 1427.0, + 378.0, + 1427.0, + 398.0, + 1403.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 444.0, + 1064.0, + 444.0, + 1064.0, + 457.0, + 1052.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 443.0, + 1207.0, + 443.0, + 1207.0, + 457.0, + 1195.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 447.0, + 1241.0, + 447.0, + 1241.0, + 455.0, + 1233.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1309.0, + 446.0, + 1317.0, + 446.0, + 1317.0, + 455.0, + 1309.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 443.0, + 1352.0, + 443.0, + 1352.0, + 457.0, + 1339.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 446.0, + 1385.0, + 446.0, + 1385.0, + 456.0, + 1377.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 988.0, + 483.0, + 993.0, + 483.0, + 993.0, + 488.0, + 988.0, + 488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1131.0, + 481.0, + 1139.0, + 481.0, + 1139.0, + 490.0, + 1131.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1275.0, + 481.0, + 1284.0, + 481.0, + 1284.0, + 490.0, + 1275.0, + 490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 503.0, + 983.0, + 503.0, + 983.0, + 526.0, + 915.0, + 526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 517.0, + 999.0, + 517.0, + 999.0, + 530.0, + 986.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 516.0, + 1143.0, + 516.0, + 1143.0, + 530.0, + 1130.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1274.0, + 517.0, + 1286.0, + 517.0, + 1286.0, + 530.0, + 1274.0, + 530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 521.0, + 1427.0, + 521.0, + 1427.0, + 541.0, + 1404.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 586.0, + 1063.0, + 586.0, + 1063.0, + 599.0, + 1052.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 585.0, + 1207.0, + 585.0, + 1207.0, + 599.0, + 1195.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1341.0, + 587.0, + 1350.0, + 587.0, + 1350.0, + 597.0, + 1341.0, + 597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 588.0, + 1385.0, + 588.0, + 1385.0, + 597.0, + 1377.0, + 597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 623.0, + 998.0, + 623.0, + 998.0, + 635.0, + 985.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1275.0, + 624.0, + 1283.0, + 624.0, + 1283.0, + 633.0, + 1275.0, + 633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 648.0, + 978.0, + 648.0, + 978.0, + 667.0, + 921.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 658.0, + 1000.0, + 658.0, + 1000.0, + 673.0, + 984.0, + 673.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 659.0, + 1143.0, + 659.0, + 1143.0, + 672.0, + 1130.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 659.0, + 1285.0, + 659.0, + 1285.0, + 672.0, + 1273.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 664.0, + 1426.0, + 664.0, + 1426.0, + 682.0, + 1404.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1015.0, + 727.0, + 1033.0, + 727.0, + 1033.0, + 742.0, + 1015.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 726.0, + 1066.0, + 726.0, + 1066.0, + 743.0, + 1050.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 729.0, + 1099.0, + 729.0, + 1099.0, + 742.0, + 1088.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1161.0, + 729.0, + 1174.0, + 729.0, + 1174.0, + 741.0, + 1161.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1193.0, + 726.0, + 1209.0, + 726.0, + 1209.0, + 743.0, + 1193.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1233.0, + 730.0, + 1241.0, + 730.0, + 1241.0, + 741.0, + 1233.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1303.0, + 727.0, + 1320.0, + 727.0, + 1320.0, + 742.0, + 1303.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 728.0, + 1352.0, + 728.0, + 1352.0, + 742.0, + 1339.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 730.0, + 1385.0, + 730.0, + 1385.0, + 740.0, + 1377.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 742.0, + 1070.0, + 742.0, + 1070.0, + 765.0, + 1044.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1187.0, + 742.0, + 1213.0, + 742.0, + 1213.0, + 765.0, + 1187.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1333.0, + 743.0, + 1356.0, + 743.0, + 1356.0, + 764.0, + 1333.0, + 764.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 109.0, + 773.0, + 109.0, + 773.0, + 145.0, + 251.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 787.0, + 1377.0, + 787.0, + 1377.0, + 832.0, + 962.0, + 832.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 1860.0, + 827.0, + 1860.0, + 827.0, + 1900.0, + 783.0, + 1900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 398.0, + 403.0, + 557.0, + 403.0, + 557.0, + 566.0, + 398.0, + 566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 552.0, + 413.0, + 588.0, + 413.0, + 588.0, + 429.0, + 552.0, + 429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 423.0, + 591.0, + 423.0, + 591.0, + 449.0, + 554.0, + 449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 439.0, + 591.0, + 439.0, + 591.0, + 463.0, + 554.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 458.0, + 590.0, + 458.0, + 590.0, + 478.0, + 555.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 473.0, + 590.0, + 473.0, + 590.0, + 494.0, + 554.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 488.0, + 590.0, + 488.0, + 590.0, + 510.0, + 554.0, + 510.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 555.0, + 506.0, + 588.0, + 506.0, + 588.0, + 523.0, + 555.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 520.0, + 588.0, + 520.0, + 588.0, + 541.0, + 554.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 556.0, + 537.0, + 589.0, + 537.0, + 589.0, + 554.0, + 556.0, + 554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 434.0, + 553.0, + 444.0, + 553.0, + 444.0, + 560.0, + 434.0, + 560.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 463.0, + 569.0, + 487.0, + 569.0, + 487.0, + 588.0, + 463.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 390.0, + 473.5, + 417.0, + 473.5, + 417.0, + 489.5, + 390.0, + 489.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 678.0, + 1490.0, + 1018.0, + 1490.0, + 1018.0, + 1526.0, + 678.0, + 1526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1519.0, + 1402.0, + 1519.0, + 1402.0, + 1559.0, + 296.0, + 1559.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1008.0, + 264.0, + 1105.0, + 264.0, + 1105.0, + 287.0, + 1008.0, + 287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1152.0, + 264.0, + 1248.0, + 264.0, + 1248.0, + 287.0, + 1152.0, + 287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1016.0, + 280.0, + 1096.0, + 280.0, + 1096.0, + 306.0, + 1016.0, + 306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1160.0, + 280.0, + 1240.0, + 280.0, + 1240.0, + 306.0, + 1160.0, + 306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1283.0, + 282.0, + 1406.0, + 282.0, + 1406.0, + 303.0, + 1283.0, + 303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 998.0, + 299.0, + 1114.0, + 299.0, + 1114.0, + 323.0, + 998.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1141.0, + 299.0, + 1260.0, + 299.0, + 1260.0, + 323.0, + 1141.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1272.0, + 299.0, + 1414.0, + 299.0, + 1414.0, + 323.0, + 1272.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 337.0, + 998.0, + 337.0, + 998.0, + 350.0, + 985.0, + 350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 337.0, + 1142.0, + 337.0, + 1142.0, + 351.0, + 1129.0, + 351.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1272.0, + 335.0, + 1287.0, + 335.0, + 1287.0, + 352.0, + 1272.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 925.0, + 363.0, + 969.0, + 363.0, + 969.0, + 381.0, + 925.0, + 381.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 372.0, + 1000.0, + 372.0, + 1000.0, + 389.0, + 984.0, + 389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 373.0, + 1142.0, + 373.0, + 1142.0, + 387.0, + 1130.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 374.0, + 1286.0, + 374.0, + 1286.0, + 387.0, + 1273.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1403.0, + 378.0, + 1427.0, + 378.0, + 1427.0, + 398.0, + 1403.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 414.0, + 994.0, + 414.0, + 994.0, + 422.0, + 986.0, + 422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 444.0, + 1063.0, + 444.0, + 1063.0, + 457.0, + 1051.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1090.0, + 447.0, + 1097.0, + 447.0, + 1097.0, + 455.0, + 1090.0, + 455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 444.0, + 1208.0, + 444.0, + 1208.0, + 457.0, + 1195.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 444.0, + 1352.0, + 444.0, + 1352.0, + 457.0, + 1339.0, + 457.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1378.0, + 446.0, + 1386.0, + 446.0, + 1386.0, + 456.0, + 1378.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 472.0, + 1362.0, + 472.0, + 1362.0, + 479.0, + 1354.0, + 479.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 479.0, + 998.0, + 479.0, + 998.0, + 493.0, + 985.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 479.0, + 1141.0, + 479.0, + 1141.0, + 492.0, + 1129.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 479.0, + 1285.0, + 479.0, + 1285.0, + 493.0, + 1273.0, + 493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 915.0, + 504.0, + 983.0, + 504.0, + 983.0, + 527.0, + 915.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 516.0, + 999.0, + 516.0, + 999.0, + 529.0, + 985.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 516.0, + 1142.0, + 516.0, + 1142.0, + 529.0, + 1130.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 517.0, + 1285.0, + 517.0, + 1285.0, + 529.0, + 1273.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 521.0, + 1427.0, + 521.0, + 1427.0, + 541.0, + 1404.0, + 541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1051.0, + 586.0, + 1063.0, + 586.0, + 1063.0, + 599.0, + 1051.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 585.0, + 1208.0, + 585.0, + 1208.0, + 599.0, + 1195.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1234.0, + 588.0, + 1242.0, + 588.0, + 1242.0, + 596.0, + 1234.0, + 596.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1305.0, + 586.0, + 1319.0, + 586.0, + 1319.0, + 597.0, + 1305.0, + 597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 585.0, + 1352.0, + 585.0, + 1352.0, + 599.0, + 1339.0, + 599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1375.0, + 586.0, + 1387.0, + 586.0, + 1387.0, + 598.0, + 1375.0, + 598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1368.0, + 614.0, + 1379.0, + 614.0, + 1379.0, + 622.0, + 1368.0, + 622.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 622.0, + 997.0, + 622.0, + 997.0, + 635.0, + 985.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 624.0, + 1138.0, + 624.0, + 1138.0, + 633.0, + 1130.0, + 633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 622.0, + 1285.0, + 622.0, + 1285.0, + 636.0, + 1273.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 919.0, + 646.0, + 978.0, + 646.0, + 978.0, + 669.0, + 919.0, + 669.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 659.0, + 1142.0, + 659.0, + 1142.0, + 671.0, + 1129.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1273.0, + 659.0, + 1285.0, + 659.0, + 1285.0, + 671.0, + 1273.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 664.0, + 1427.0, + 664.0, + 1427.0, + 682.0, + 1404.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 727.0, + 1032.0, + 727.0, + 1032.0, + 743.0, + 1012.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1049.0, + 726.0, + 1066.0, + 726.0, + 1066.0, + 744.0, + 1049.0, + 744.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1088.0, + 729.0, + 1099.0, + 729.0, + 1099.0, + 742.0, + 1088.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1156.0, + 728.0, + 1176.0, + 728.0, + 1176.0, + 742.0, + 1156.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1193.0, + 726.0, + 1210.0, + 726.0, + 1210.0, + 743.0, + 1193.0, + 743.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1232.0, + 729.0, + 1243.0, + 729.0, + 1243.0, + 742.0, + 1232.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1300.0, + 728.0, + 1320.0, + 728.0, + 1320.0, + 742.0, + 1300.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1339.0, + 723.0, + 1355.0, + 723.0, + 1355.0, + 745.0, + 1339.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1376.0, + 729.0, + 1387.0, + 729.0, + 1387.0, + 742.0, + 1376.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1044.0, + 742.0, + 1070.0, + 742.0, + 1070.0, + 765.0, + 1044.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1188.0, + 742.0, + 1214.0, + 742.0, + 1214.0, + 765.0, + 1188.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1332.0, + 742.0, + 1358.0, + 742.0, + 1358.0, + 765.0, + 1332.0, + 765.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 985.0, + 657.0, + 997.0, + 657.0, + 997.0, + 672.0, + 985.0, + 672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1670.0, + 1450.0, + 1670.0, + 1450.0, + 1706.0, + 249.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 1703.0, + 781.0, + 1703.0, + 781.0, + 1735.0, + 347.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1745.0, + 458.0, + 1745.0, + 458.0, + 1785.0, + 249.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 537.0, + 1745.0, + 1448.0, + 1745.0, + 1448.0, + 1785.0, + 537.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1780.0, + 251.0, + 1780.0, + 251.0, + 1809.0, + 248.0, + 1809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 325.0, + 1780.0, + 441.0, + 1780.0, + 441.0, + 1809.0, + 325.0, + 1809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 515.0, + 1780.0, + 527.0, + 1780.0, + 527.0, + 1809.0, + 515.0, + 1809.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 19, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 254, + 1717, + 1449, + 1717, + 1449, + 1811, + 254, + 1811 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 254, + 1388, + 1449, + 1388, + 1449, + 1482, + 254, + 1482 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 252, + 1311, + 1446, + 1311, + 1446, + 1373, + 252, + 1373 + ], + "score": 0.931 + }, + { + "category_id": 5, + "poly": [ + 492, + 408, + 1208, + 408, + 1208, + 636, + 492, + 636 + ], + "score": 0.929, + "html": "
MethodMNISTOmniglotCelebA
VAEAC IS-10261±1275±1734035 ± 1609
VAEAC MC-10494±41452 ± 10941513 ± 2163
VAEAC MC-102156 ±12203 ± 15053904 ± 3121
GSNN MC-104141 ±71199 ± 6253427 ± 2208
GSNN MC-10²141 ±11200 ± 6253486 ± 2210
Naive Bayes2052490269480
" + }, + { + "category_id": 4, + "poly": [ + 492, + 1120, + 1205, + 1120, + 1205, + 1154, + 492, + 1154 + ], + "score": 0.926 + }, + { + "category_id": 3, + "poly": [ + 578, + 726, + 1075, + 726, + 1075, + 1056, + 578, + 1056 + ], + "score": 0.914 + }, + { + "category_id": 0, + "poly": [ + 255, + 1646, + 597, + 1646, + 597, + 1680, + 255, + 1680 + ], + "score": 0.903 + }, + { + "category_id": 2, + "poly": [ + 253, + 112, + 770, + 112, + 770, + 141, + 253, + 141 + ], + "score": 0.901 + }, + { + "category_id": 0, + "poly": [ + 254, + 1559, + 712, + 1559, + 712, + 1595, + 254, + 1595 + ], + "score": 0.859 + }, + { + "category_id": 1, + "poly": [ + 252, + 1241, + 943, + 1241, + 943, + 1273, + 252, + 1273 + ], + "score": 0.85 + }, + { + "category_id": 2, + "poly": [ + 790, + 1865, + 818, + 1865, + 818, + 1889, + 790, + 1889 + ], + "score": 0.822 + }, + { + "category_id": 6, + "poly": [ + 252, + 259, + 1448, + 259, + 1448, + 383, + 252, + 383 + ], + "score": 0.605 + }, + { + "category_id": 1, + "poly": [ + 252, + 259, + 1448, + 259, + 1448, + 383, + 252, + 383 + ], + "score": 0.543 + }, + { + "category_id": 13, + "poly": [ + 430, + 323, + 452, + 323, + 452, + 350, + 430, + 350 + ], + "score": 0.79, + "latex": "S" + }, + { + "category_id": 13, + "poly": [ + 597, + 293, + 619, + 293, + 619, + 320, + 597, + 320 + ], + "score": 0.77, + "latex": "S" + }, + { + "category_id": 13, + "poly": [ + 1010, + 292, + 1032, + 292, + 1032, + 320, + 1010, + 320 + ], + "score": 0.69, + "latex": "S" + }, + { + "category_id": 13, + "poly": [ + 1086, + 261, + 1108, + 261, + 1108, + 289, + 1086, + 289 + ], + "score": 0.61, + "latex": "S" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 1117.0, + 1208.0, + 1117.0, + 1208.0, + 1159.0, + 490.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 725.0, + 648.0, + 725.0, + 648.0, + 750.0, + 610.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 751.0, + 606.0, + 751.0, + 606.0, + 992.0, + 575.0, + 992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 611.0, + 762.0, + 649.0, + 762.0, + 649.0, + 789.0, + 611.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 604.0, + 802.0, + 649.0, + 802.0, + 649.0, + 827.0, + 604.0, + 827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 606.0, + 840.0, + 648.0, + 840.0, + 648.0, + 866.0, + 606.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 604.0, + 880.0, + 649.0, + 880.0, + 649.0, + 904.0, + 604.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 603.0, + 916.0, + 649.0, + 916.0, + 649.0, + 944.0, + 603.0, + 944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 941.0, + 779.0, + 941.0, + 779.0, + 972.0, + 706.0, + 972.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 604.0, + 956.0, + 648.0, + 956.0, + 648.0, + 982.0, + 604.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 706.0, + 966.0, + 756.0, + 966.0, + 756.0, + 1000.0, + 706.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 604.0, + 994.0, + 647.0, + 994.0, + 647.0, + 1019.0, + 604.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 642.0, + 1013.0, + 657.0, + 1013.0, + 657.0, + 1031.0, + 642.0, + 1031.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 722.0, + 1007.0, + 741.0, + 1007.0, + 741.0, + 1033.0, + 722.0, + 1033.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 1009.0, + 828.0, + 1009.0, + 828.0, + 1034.0, + 801.0, + 1034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 885.0, + 1009.0, + 912.0, + 1009.0, + 912.0, + 1034.0, + 885.0, + 1034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 968.0, + 1009.0, + 995.0, + 1009.0, + 995.0, + 1034.0, + 968.0, + 1034.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1052.0, + 1012.0, + 1077.0, + 1012.0, + 1077.0, + 1032.0, + 1052.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 825.0, + 1025.0, + 890.0, + 1025.0, + 890.0, + 1056.0, + 825.0, + 1056.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1645.0, + 600.0, + 1645.0, + 600.0, + 1682.0, + 250.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 109.0, + 772.0, + 109.0, + 772.0, + 145.0, + 251.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1557.0, + 717.0, + 1557.0, + 717.0, + 1599.0, + 248.0, + 1599.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1861.0, + 824.0, + 1861.0, + 824.0, + 1902.0, + 784.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 257.0, + 1085.0, + 257.0, + 1085.0, + 299.0, + 247.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 257.0, + 1449.0, + 257.0, + 1449.0, + 299.0, + 1109.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 292.0, + 596.0, + 292.0, + 596.0, + 327.0, + 251.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 620.0, + 292.0, + 1009.0, + 292.0, + 1009.0, + 327.0, + 620.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 292.0, + 1448.0, + 292.0, + 1448.0, + 327.0, + 1033.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 321.0, + 429.0, + 321.0, + 429.0, + 356.0, + 249.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 321.0, + 1448.0, + 321.0, + 1448.0, + 356.0, + 453.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 350.0, + 492.0, + 350.0, + 492.0, + 390.0, + 248.0, + 390.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1716.0, + 1449.0, + 1716.0, + 1449.0, + 1752.0, + 249.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1747.0, + 1449.0, + 1747.0, + 1449.0, + 1784.0, + 250.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1776.0, + 1140.0, + 1776.0, + 1140.0, + 1816.0, + 249.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1382.0, + 1449.0, + 1382.0, + 1449.0, + 1429.0, + 247.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1421.0, + 1449.0, + 1421.0, + 1449.0, + 1455.0, + 251.0, + 1455.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1449.0, + 1370.0, + 1449.0, + 1370.0, + 1485.0, + 250.0, + 1485.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 245.0, + 1308.0, + 1452.0, + 1308.0, + 1452.0, + 1348.0, + 245.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1342.0, + 412.0, + 1342.0, + 412.0, + 1375.0, + 249.0, + 1375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1240.0, + 947.0, + 1240.0, + 947.0, + 1277.0, + 251.0, + 1277.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 257.0, + 1085.0, + 257.0, + 1085.0, + 299.0, + 247.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 257.0, + 1449.0, + 257.0, + 1449.0, + 299.0, + 1109.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 292.0, + 596.0, + 292.0, + 596.0, + 327.0, + 251.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 620.0, + 292.0, + 1009.0, + 292.0, + 1009.0, + 327.0, + 620.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1033.0, + 292.0, + 1448.0, + 292.0, + 1448.0, + 327.0, + 1033.0, + 327.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 321.0, + 429.0, + 321.0, + 429.0, + 356.0, + 249.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 321.0, + 1448.0, + 321.0, + 1448.0, + 356.0, + 453.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 350.0, + 492.0, + 350.0, + 492.0, + 390.0, + 248.0, + 390.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 20, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 260, + 873, + 1442, + 873, + 1442, + 1324, + 260, + 1324 + ], + "score": 0.98, + "html": "
DatasetMICEMissForestGAINVAEACGSNNNN
Boston0.57 ± 0.080.6 ± 0.10.50 ± 0.100.5 ± 0.10.5± 0.10.50±0.09
Breast0.96 ± 0.020.95 ± 0.020.94± 0.010.95 ± 0.020.96 ± 0.020.95 ± 0.02
Concrete0.35 ± 0.050.33 ± 0.040.28±0.060.30 ± 0.080.32 ± 0.050±1
Diabetes0.37 ± 0.060.34±0.060.34±0.030.34± 0.040.33 ±0.040.27±0.06
Digits0.86±0.020.887 ±0.0080.83 ±0.030.892 ± 0.0100.895 ± 0.0100.912 ± 0.010
Glass0.44 ±0.080.53 ±0.050.37±0.050.49 ±0.090.47 ±0.090.48 ±0.09
Iris0.81± 0.020.84±0.020.66 ±0.060.84±0.050.82±0.060.73±0.09
Mushroom0.92 ±0.010.972 ± 0.0030.969 ± 0.0050.987 ± 0.0010.986 ± 0.0020.989 ± 0.003
Orthopedic0.71 ± 0.020.72 ± 0.030.60±0.030.71 ±0.020.70± 0.040.61±0.04
Phishing0.75 ± 0.020.73±0.030.74± 0.030.75 ± 0.010.74±0.040.73±0.02
WallRobot0.55 ±0.010.697 ± 0.0050.56 ±0.010.62±0.020.62 ± 0.010.64±0.02
WhiteWine0.13 ± 0.020.17 ± 0.010.11± 0.010.18 ±0.020.17 ± 0.010.15 ± 0.03
Yeast0.42 ±0.020.41 ±0.020.39 ±0.060.42 ± 0.010.425 ± 0.0100.33±0.03
Zoo0.78 ± 0.060.71 ±0.080.67±0.060.77 ± 0.090.8±0.10.83 ± 0.08
" + }, + { + "category_id": 5, + "poly": [ + 265, + 312, + 1444, + 312, + 1444, + 757, + 265, + 757 + ], + "score": 0.979, + "html": "
DatasetMICEMissForestGAINVAEACGSNNNN
Boston0.69 ± 0.020.58±0.020.78 ± 0.030.71 ± 0.020.70 ± 0.010.69 ± 0.01
Breast0.58 ±0.020.515 ± 0.0080.67 ±0.010.55±0.020.55 ± 0.020.52 ±0.02
Concrete0.850 ± 0.0070.78 ± 0.010.96 ±0.010.84 ±0.020.85 ± 0.012±3
Diabetes0.80 ±0.010.84±0.020.911 ± 0.0090.90 ±0.030.91 ± 0.030.90 ±0.02
Digits0.69 ±0.020.61± 0.020.79±0.020.69 ±0.020.69 ± 0.020.67 ±0.02
Glass0.91 ±0.020.83± 0.041.06 ±0.050.91 ±0.040.91 ± 0.050.87 ± 0.04
Iris0.59 ± 0.020.62 ± 0.040.72 ± 0.040.64± 0.040.62 ± 0.040.61± 0.02
Mushroom0.334 ± 0.0020.249 ±0.0060.271 ±0.0030.241 ± 0.0020.2412 ± 0.00090.239 ± 0.001
Orthopedic0.76 ±0.020.79±0.030.91±0.030.80±0.030.81 ±0.030.81 ±0.02
Phishing0.422 ± 0.0060.422 ±0.0090.427 ± 0.0100.397 ± 0.0100.392 ±0.0090.41 ± 0.01
WallRobot0.885 ± 0.0030.640 ± 0.0030.907 ± 0.0050.78 ±0.010.776 ±0.0070.757± 0.005
WhiteWine0.964 ± 0.0070.878 ±0.0090.97 ± 0.020.850 ± 0.0050.848 ± 0.0070.85 ± 0.01
Yeast0.98±0.021.00 ± 0.020.99 ±0.030.95 ± 0.010.958 ± 0.0070.97 ± 0.03
Zoo0.19 ± 0.030.16 ±0.020.20 ±0.020.16±0.020.17 ±0.020.16 ±0.01
" + }, + { + "category_id": 1, + "poly": [ + 253, + 1440, + 1449, + 1440, + 1449, + 1596, + 253, + 1596 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 255, + 1687, + 1448, + 1687, + 1448, + 1812, + 255, + 1812 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 249, + 1609, + 1448, + 1609, + 1448, + 1673, + 249, + 1673 + ], + "score": 0.95 + }, + { + "category_id": 6, + "poly": [ + 247, + 788, + 1440, + 788, + 1440, + 851, + 247, + 851 + ], + "score": 0.92 + }, + { + "category_id": 2, + "poly": [ + 254, + 112, + 769, + 112, + 769, + 141, + 254, + 141 + ], + "score": 0.914 + }, + { + "category_id": 6, + "poly": [ + 270, + 259, + 1428, + 259, + 1428, + 294, + 270, + 294 + ], + "score": 0.877 + }, + { + "category_id": 0, + "poly": [ + 255, + 1382, + 721, + 1382, + 721, + 1415, + 255, + 1415 + ], + "score": 0.848 + }, + { + "category_id": 2, + "poly": [ + 790, + 1865, + 820, + 1865, + 820, + 1889, + 790, + 1889 + ], + "score": 0.839 + }, + { + "category_id": 13, + "poly": [ + 444, + 1502, + 596, + 1502, + 596, + 1536, + 444, + 1536 + ], + "score": 0.93, + "latex": "p _ { \\theta } ( z | x _ { 1 - b } , b )" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 786.0, + 1444.0, + 786.0, + 1444.0, + 825.0, + 250.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 821.0, + 627.0, + 821.0, + 627.0, + 853.0, + 251.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 109.0, + 772.0, + 109.0, + 772.0, + 145.0, + 251.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 267.0, + 255.0, + 1431.0, + 255.0, + 1431.0, + 298.0, + 267.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1381.0, + 724.0, + 1381.0, + 724.0, + 1418.0, + 249.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 1862.0, + 824.0, + 1862.0, + 824.0, + 1901.0, + 786.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 1442.0, + 1449.0, + 1442.0, + 1449.0, + 1475.0, + 252.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1473.0, + 1449.0, + 1473.0, + 1449.0, + 1506.0, + 250.0, + 1506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1501.0, + 443.0, + 1501.0, + 443.0, + 1537.0, + 246.0, + 1537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 597.0, + 1501.0, + 1450.0, + 1501.0, + 1450.0, + 1537.0, + 597.0, + 1537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1533.0, + 1449.0, + 1533.0, + 1449.0, + 1566.0, + 250.0, + 1566.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1563.0, + 472.0, + 1563.0, + 472.0, + 1598.0, + 250.0, + 1598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1686.0, + 1450.0, + 1686.0, + 1450.0, + 1721.0, + 251.0, + 1721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1716.0, + 1451.0, + 1716.0, + 1451.0, + 1752.0, + 248.0, + 1752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1747.0, + 1450.0, + 1747.0, + 1450.0, + 1783.0, + 248.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1779.0, + 451.0, + 1779.0, + 451.0, + 1814.0, + 249.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1608.0, + 1452.0, + 1608.0, + 1452.0, + 1645.0, + 249.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1641.0, + 1122.0, + 1641.0, + 1122.0, + 1673.0, + 250.0, + 1673.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 21, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 419, + 264, + 1292, + 264, + 1292, + 1010, + 419, + 1010 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 254, + 1654, + 1446, + 1654, + 1446, + 1748, + 254, + 1748 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 253, + 1485, + 1447, + 1485, + 1447, + 1640, + 253, + 1640 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 253, + 1378, + 1447, + 1378, + 1447, + 1471, + 253, + 1471 + ], + "score": 0.969 + }, + { + "category_id": 2, + "poly": [ + 288, + 1782, + 672, + 1782, + 672, + 1811, + 288, + 1811 + ], + "score": 0.912 + }, + { + "category_id": 0, + "poly": [ + 254, + 1185, + 552, + 1185, + 552, + 1217, + 254, + 1217 + ], + "score": 0.912 + }, + { + "category_id": 1, + "poly": [ + 256, + 1244, + 1389, + 1244, + 1389, + 1277, + 256, + 1277 + ], + "score": 0.891 + }, + { + "category_id": 2, + "poly": [ + 252, + 113, + 770, + 113, + 770, + 141, + 252, + 141 + ], + "score": 0.89 + }, + { + "category_id": 0, + "poly": [ + 254, + 1318, + 692, + 1318, + 692, + 1350, + 254, + 1350 + ], + "score": 0.856 + }, + { + "category_id": 4, + "poly": [ + 342, + 1048, + 1376, + 1048, + 1376, + 1111, + 342, + 1111 + ], + "score": 0.856 + }, + { + "category_id": 2, + "poly": [ + 790, + 1865, + 820, + 1865, + 820, + 1889, + 790, + 1889 + ], + "score": 0.839 + }, + { + "category_id": 13, + "poly": [ + 452, + 1608, + 501, + 1608, + 501, + 1642, + 452, + 1642 + ], + "score": 0.92, + "latex": "{ \\dot { p } } ( b )" + }, + { + "category_id": 13, + "poly": [ + 319, + 1717, + 369, + 1717, + 369, + 1750, + 319, + 1750 + ], + "score": 0.92, + "latex": "p ( b )" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 266.0, + 1293.0, + 266.0, + 1293.0, + 395.0, + 449.0, + 395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 479.0, + 448.0, + 494.0, + 448.0, + 494.0, + 467.0, + 479.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 376.0, + 1298.0, + 376.0, + 1298.0, + 527.0, + 657.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 442.0, + 508.0, + 1302.0, + 508.0, + 1302.0, + 654.0, + 442.0, + 654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 691.0, + 634.0, + 1298.0, + 634.0, + 1298.0, + 771.0, + 691.0, + 771.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 422.0, + 883.0, + 1291.0, + 883.0, + 1291.0, + 1016.0, + 422.0, + 1016.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1196.75, + 792.5, + 1245.75, + 792.5, + 1245.75, + 875.5, + 1196.75, + 875.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 1775.0, + 678.0, + 1775.0, + 678.0, + 1817.0, + 286.0, + 1817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1185.0, + 555.0, + 1185.0, + 555.0, + 1221.0, + 250.0, + 1221.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 110.0, + 772.0, + 110.0, + 772.0, + 145.0, + 251.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1318.0, + 695.0, + 1318.0, + 695.0, + 1354.0, + 250.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 487.0, + 1048.0, + 1211.0, + 1048.0, + 1211.0, + 1083.0, + 487.0, + 1083.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 336.0, + 1076.0, + 1381.0, + 1076.0, + 1381.0, + 1115.0, + 336.0, + 1115.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 785.0, + 1861.0, + 824.0, + 1861.0, + 824.0, + 1899.0, + 785.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1651.0, + 1446.0, + 1651.0, + 1446.0, + 1690.0, + 248.0, + 1690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1684.0, + 1449.0, + 1684.0, + 1449.0, + 1722.0, + 250.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 1715.0, + 318.0, + 1715.0, + 318.0, + 1753.0, + 251.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1715.0, + 1142.0, + 1715.0, + 1142.0, + 1753.0, + 370.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1484.0, + 1450.0, + 1484.0, + 1450.0, + 1520.0, + 250.0, + 1520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1513.0, + 1451.0, + 1513.0, + 1451.0, + 1552.0, + 248.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1547.0, + 1451.0, + 1547.0, + 1451.0, + 1582.0, + 247.0, + 1582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1576.0, + 1452.0, + 1576.0, + 1452.0, + 1616.0, + 246.0, + 1616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1605.0, + 451.0, + 1605.0, + 451.0, + 1645.0, + 248.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 502.0, + 1605.0, + 512.0, + 1605.0, + 512.0, + 1645.0, + 502.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1372.0, + 1451.0, + 1372.0, + 1451.0, + 1417.0, + 248.0, + 1417.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1408.0, + 1448.0, + 1408.0, + 1448.0, + 1446.0, + 250.0, + 1446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1437.0, + 1099.0, + 1437.0, + 1099.0, + 1476.0, + 248.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1241.0, + 1395.0, + 1241.0, + 1395.0, + 1283.0, + 248.0, + 1283.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 22, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 299, + 780, + 1421, + 780, + 1421, + 1278, + 299, + 1278 + ], + "score": 0.97 + }, + { + "category_id": 3, + "poly": [ + 592, + 272, + 1108, + 272, + 1108, + 639, + 592, + 639 + ], + "score": 0.966 + }, + { + "category_id": 2, + "poly": [ + 253, + 113, + 770, + 113, + 770, + 141, + 253, + 141 + ], + "score": 0.9 + }, + { + "category_id": 2, + "poly": [ + 789, + 1865, + 820, + 1865, + 820, + 1890, + 789, + 1890 + ], + "score": 0.85 + }, + { + "category_id": 1, + "poly": [ + 255, + 1496, + 1348, + 1496, + 1348, + 1531, + 255, + 1531 + ], + "score": 0.846 + }, + { + "category_id": 0, + "poly": [ + 251, + 1440, + 871, + 1440, + 871, + 1473, + 251, + 1473 + ], + "score": 0.835 + }, + { + "category_id": 1, + "poly": [ + 247, + 1544, + 1451, + 1544, + 1451, + 1610, + 247, + 1610 + ], + "score": 0.833 + }, + { + "category_id": 4, + "poly": [ + 609, + 676, + 1092, + 676, + 1092, + 708, + 609, + 708 + ], + "score": 0.705 + }, + { + "category_id": 4, + "poly": [ + 290, + 1321, + 1406, + 1321, + 1406, + 1387, + 290, + 1387 + ], + "score": 0.503 + }, + { + "category_id": 4, + "poly": [ + 675, + 1322, + 1021, + 1322, + 1021, + 1353, + 675, + 1353 + ], + "score": 0.381 + }, + { + "category_id": 1, + "poly": [ + 285, + 706, + 1400, + 706, + 1400, + 739, + 285, + 739 + ], + "score": 0.231 + }, + { + "category_id": 15, + "poly": [ + 1032.0, + 782.0, + 1088.0, + 782.0, + 1088.0, + 1170.0, + 1032.0, + 1170.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 765.0, + 784.0, + 821.0, + 784.0, + 821.0, + 1207.0, + 765.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1370.0, + 784.0, + 1426.0, + 784.0, + 1426.0, + 1207.0, + 1370.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 682.0, + 785.0, + 730.0, + 785.0, + 730.0, + 1206.0, + 682.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1334.0, + 786.0, + 1378.0, + 786.0, + 1378.0, + 1083.0, + 1334.0, + 1083.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 941.0, + 788.0, + 997.0, + 788.0, + 997.0, + 1207.0, + 941.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 789.0, + 394.0, + 789.0, + 394.0, + 1211.0, + 324.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 990.0, + 789.0, + 1042.0, + 789.0, + 1042.0, + 1175.0, + 990.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1244.0, + 789.0, + 1293.0, + 789.0, + 1293.0, + 1206.0, + 1244.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1285.0, + 789.0, + 1337.0, + 789.0, + 1337.0, + 1207.0, + 1285.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 725.0, + 790.0, + 773.0, + 790.0, + 773.0, + 1206.0, + 725.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 515.0, + 798.0, + 564.0, + 798.0, + 564.0, + 1136.0, + 515.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 639.0, + 800.0, + 687.0, + 800.0, + 687.0, + 1205.0, + 639.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1219.0, + 804.0, + 1250.0, + 804.0, + 1250.0, + 866.0, + 1219.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 421.0, + 806.0, + 486.0, + 806.0, + 486.0, + 1172.0, + 421.0, + 1172.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 806.0, + 525.0, + 806.0, + 525.0, + 1171.0, + 460.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1079.0, + 806.0, + 1128.0, + 806.0, + 1128.0, + 1133.0, + 1079.0, + 1133.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 593.0, + 810.0, + 645.0, + 810.0, + 645.0, + 1175.0, + 593.0, + 1175.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1172.0, + 811.0, + 1202.0, + 811.0, + 1202.0, + 867.0, + 1172.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1132.0, + 813.0, + 1158.0, + 813.0, + 1158.0, + 863.0, + 1132.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 816.0, + 423.0, + 816.0, + 423.0, + 883.0, + 394.0, + 883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 820.0, + 599.0, + 820.0, + 599.0, + 990.0, + 562.0, + 990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 372.0, + 857.0, + 446.0, + 857.0, + 446.0, + 1174.0, + 372.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1205.0, + 858.0, + 1253.0, + 858.0, + 1253.0, + 1206.0, + 1205.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 866.0, + 1206.0, + 866.0, + 1206.0, + 1004.0, + 1169.0, + 1004.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1136.0, + 881.0, + 1149.0, + 881.0, + 1149.0, + 903.0, + 1136.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 891.0, + 1168.0, + 891.0, + 1168.0, + 1171.0, + 1123.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1164.0, + 990.0, + 1213.0, + 990.0, + 1213.0, + 1206.0, + 1164.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 992.0, + 606.0, + 992.0, + 606.0, + 1169.0, + 553.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1344.0, + 1068.0, + 1369.0, + 1068.0, + 1369.0, + 1119.0, + 1344.0, + 1119.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 492.0, + 1107.0, + 501.0, + 1107.0, + 501.0, + 1126.0, + 492.0, + 1126.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 508.0, + 1108.0, + 558.0, + 1108.0, + 558.0, + 1161.0, + 508.0, + 1161.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1078.0, + 1109.0, + 1117.0, + 1109.0, + 1117.0, + 1159.0, + 1078.0, + 1159.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1335.0, + 1102.0, + 1382.0, + 1102.0, + 1382.0, + 1206.0, + 1335.0, + 1206.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 431.0, + 1147.0, + 515.0, + 1147.0, + 515.0, + 1203.0, + 431.0, + 1203.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 521.0, + 1155.0, + 555.0, + 1155.0, + 555.0, + 1200.0, + 521.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 1155.0, + 595.0, + 1155.0, + 595.0, + 1200.0, + 563.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 607.0, + 1161.0, + 635.0, + 1161.0, + 635.0, + 1198.0, + 607.0, + 1198.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 1154.0, + 1081.0, + 1154.0, + 1081.0, + 1200.0, + 1041.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1084.0, + 1164.0, + 1125.0, + 1164.0, + 1125.0, + 1200.0, + 1084.0, + 1200.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1129.0, + 1151.0, + 1167.0, + 1151.0, + 1167.0, + 1201.0, + 1129.0, + 1201.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 478.0, + 1239.0, + 615.0, + 1239.0, + 615.0, + 1282.0, + 478.0, + 1282.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1104.0, + 1239.0, + 1201.0, + 1239.0, + 1201.0, + 1281.0, + 1104.0, + 1281.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 369.0, + 1160.0, + 448.0, + 1160.0, + 448.0, + 1201.5, + 369.0, + 1201.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1054.0, + 582.0, + 1091.0, + 582.0, + 1091.0, + 629.0, + 1054.0, + 629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 110.0, + 772.0, + 110.0, + 772.0, + 145.0, + 251.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1861.0, + 826.0, + 1861.0, + 826.0, + 1902.0, + 784.0, + 1902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1440.0, + 876.0, + 1440.0, + 876.0, + 1475.0, + 250.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 603.0, + 673.0, + 1095.0, + 673.0, + 1095.0, + 716.0, + 603.0, + 716.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 1320.0, + 1028.0, + 1320.0, + 1028.0, + 1357.0, + 672.0, + 1357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1349.0, + 1402.0, + 1349.0, + 1402.0, + 1389.0, + 296.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 670.0, + 1317.0, + 1028.0, + 1317.0, + 1028.0, + 1362.0, + 670.0, + 1362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1494.0, + 1348.0, + 1494.0, + 1348.0, + 1536.0, + 248.0, + 1536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 253.0, + 1546.0, + 1448.0, + 1546.0, + 1448.0, + 1578.0, + 253.0, + 1578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1576.0, + 1237.0, + 1576.0, + 1237.0, + 1611.0, + 247.0, + 1611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 700.0, + 1402.0, + 700.0, + 1402.0, + 743.0, + 294.0, + 743.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 23, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 3, + "poly": [ + 271, + 826, + 1429, + 826, + 1429, + 1169, + 271, + 1169 + ], + "score": 0.953 + }, + { + "category_id": 4, + "poly": [ + 568, + 1212, + 1128, + 1212, + 1128, + 1247, + 568, + 1247 + ], + "score": 0.918 + }, + { + "category_id": 2, + "poly": [ + 253, + 112, + 770, + 112, + 770, + 142, + 253, + 142 + ], + "score": 0.893 + }, + { + "category_id": 2, + "poly": [ + 789, + 1865, + 820, + 1865, + 820, + 1890, + 789, + 1890 + ], + "score": 0.833 + }, + { + "category_id": 15, + "poly": [ + 558.0, + 826.0, + 620.0, + 826.0, + 620.0, + 858.0, + 558.0, + 858.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1372.0, + 844.0, + 1382.0, + 844.0, + 1382.0, + 854.0, + 1372.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 313.0, + 868.0, + 330.0, + 868.0, + 330.0, + 885.0, + 313.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 494.0, + 867.0, + 515.0, + 867.0, + 515.0, + 887.0, + 494.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 850.0, + 622.0, + 850.0, + 622.0, + 885.0, + 560.0, + 885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 773.0, + 863.0, + 795.0, + 863.0, + 795.0, + 887.0, + 773.0, + 887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1176.0, + 862.0, + 1203.0, + 862.0, + 1203.0, + 888.0, + 1176.0, + 888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 875.0, + 622.0, + 875.0, + 622.0, + 911.0, + 562.0, + 911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 902.0, + 621.0, + 902.0, + 621.0, + 933.0, + 563.0, + 933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1230.0, + 914.0, + 1241.0, + 914.0, + 1241.0, + 925.0, + 1230.0, + 925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 926.0, + 622.0, + 926.0, + 622.0, + 961.0, + 562.0, + 961.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 952.0, + 621.0, + 952.0, + 621.0, + 984.0, + 563.0, + 984.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 560.0, + 976.0, + 622.0, + 976.0, + 622.0, + 1011.0, + 560.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 996.0, + 335.0, + 996.0, + 335.0, + 1025.0, + 297.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 363.0, + 1000.0, + 382.0, + 1000.0, + 382.0, + 1021.0, + 363.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 999.0, + 621.0, + 999.0, + 621.0, + 1036.0, + 563.0, + 1036.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 762.0, + 999.0, + 798.0, + 999.0, + 798.0, + 1028.0, + 762.0, + 1028.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1167.0, + 998.0, + 1203.0, + 998.0, + 1203.0, + 1026.0, + 1167.0, + 1026.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 563.0, + 1026.0, + 621.0, + 1026.0, + 621.0, + 1061.0, + 563.0, + 1061.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1049.0, + 393.0, + 1049.0, + 393.0, + 1078.0, + 355.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 424.0, + 1050.0, + 452.0, + 1050.0, + 452.0, + 1078.0, + 424.0, + 1078.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 492.0, + 1050.0, + 513.0, + 1050.0, + 513.0, + 1077.0, + 492.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 817.0, + 1049.0, + 860.0, + 1049.0, + 860.0, + 1083.0, + 817.0, + 1083.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 895.0, + 1053.0, + 919.0, + 1053.0, + 919.0, + 1081.0, + 895.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 964.0, + 1053.0, + 989.0, + 1053.0, + 989.0, + 1081.0, + 964.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1226.0, + 1053.0, + 1263.0, + 1053.0, + 1263.0, + 1082.0, + 1226.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1299.0, + 1053.0, + 1326.0, + 1053.0, + 1326.0, + 1082.0, + 1299.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1369.0, + 1053.0, + 1394.0, + 1053.0, + 1394.0, + 1082.0, + 1369.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 414.0, + 1077.0, + 461.0, + 1077.0, + 461.0, + 1117.0, + 414.0, + 1117.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 885.0, + 1081.0, + 926.0, + 1081.0, + 926.0, + 1116.0, + 885.0, + 1116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1290.0, + 1081.0, + 1332.0, + 1081.0, + 1332.0, + 1116.0, + 1290.0, + 1116.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 337.0, + 1131.0, + 552.0, + 1131.0, + 552.0, + 1165.0, + 337.0, + 1165.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 714.0, + 1133.0, + 1046.0, + 1133.0, + 1046.0, + 1167.0, + 714.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1134.0, + 1430.0, + 1134.0, + 1430.0, + 1167.0, + 1137.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 273.25, + 916.0, + 335.25, + 916.0, + 335.25, + 955.0, + 273.25, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 736.0, + 917.5, + 797.0, + 917.5, + 797.0, + 956.5, + 736.0, + 956.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 917.5, + 1202.0, + 917.5, + 1202.0, + 956.5, + 1140.0, + 956.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 805.75, + 920.5, + 1015.75, + 920.5, + 1015.75, + 962.5, + 805.75, + 962.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 1211.0, + 1129.0, + 1211.0, + 1129.0, + 1251.0, + 570.0, + 1251.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 110.0, + 772.0, + 110.0, + 772.0, + 145.0, + 251.0, + 145.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 784.0, + 1860.0, + 826.0, + 1860.0, + 826.0, + 1901.0, + 784.0, + 1901.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 24, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/TScOKQW67Mj/images/07a3724823bcd540052ed15c8c366ca29e6a1843d56d7cce4940e24c70fd634b.jpg b/parse/train/TScOKQW67Mj/images/07a3724823bcd540052ed15c8c366ca29e6a1843d56d7cce4940e24c70fd634b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8a78b9abdaa6da79afa18ad37af1266a48132d70 --- /dev/null +++ b/parse/train/TScOKQW67Mj/images/07a3724823bcd540052ed15c8c366ca29e6a1843d56d7cce4940e24c70fd634b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:672614ef5db157e6d9a05804f63ad18a7ed32f175df758568b1cef3eaffb2fc0 +size 3222 diff --git a/parse/train/TScOKQW67Mj/images/18032ac4c37b6c2eb1a06638b0da073beea2c546e17d030f4ad490a25ab947a5.jpg b/parse/train/TScOKQW67Mj/images/18032ac4c37b6c2eb1a06638b0da073beea2c546e17d030f4ad490a25ab947a5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..daccfdb76c683c949a6562cb6ca1fab5ad648d24 --- /dev/null +++ b/parse/train/TScOKQW67Mj/images/18032ac4c37b6c2eb1a06638b0da073beea2c546e17d030f4ad490a25ab947a5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4f01bc64dec2a20b4d8c9d269f446b9a1b7c9b601c246cbf8183a745cbbbe233 +size 51391 diff --git a/parse/train/TScOKQW67Mj/images/2181d68a11352a1504787146df9a48b0cf9ff24ed9fa565eb281c5e574c5f139.jpg b/parse/train/TScOKQW67Mj/images/2181d68a11352a1504787146df9a48b0cf9ff24ed9fa565eb281c5e574c5f139.jpg new file mode 100644 index 0000000000000000000000000000000000000000..290303649a6bf03250102fa7a574ac819569c83a --- /dev/null +++ b/parse/train/TScOKQW67Mj/images/2181d68a11352a1504787146df9a48b0cf9ff24ed9fa565eb281c5e574c5f139.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4866b5afa4d9c494ef38dd36391d2ed09016b45bbd20f03eaad823d89c9497ce +size 7806 diff --git a/parse/train/TScOKQW67Mj/images/2450ef698b17c3ee11267e5b440cbba24aa7840aadfa0ceacde0176d198accc9.jpg b/parse/train/TScOKQW67Mj/images/2450ef698b17c3ee11267e5b440cbba24aa7840aadfa0ceacde0176d198accc9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ebea73b59db4be59d149543babfbb283b7a334ca --- /dev/null +++ b/parse/train/TScOKQW67Mj/images/2450ef698b17c3ee11267e5b440cbba24aa7840aadfa0ceacde0176d198accc9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d01a6d98279a1d3913e56418d21a5519941b1dad4e81701d6b7ba41ed5d2bc7c +size 53795 diff --git a/parse/train/TScOKQW67Mj/images/3e7f965619c868b1860965c878110bc5f1bab3dc23c1ef22b1ae8806227859a2.jpg b/parse/train/TScOKQW67Mj/images/3e7f965619c868b1860965c878110bc5f1bab3dc23c1ef22b1ae8806227859a2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7677f02f14eec457a32fade2ef0b04b0b897bc42 --- /dev/null +++ b/parse/train/TScOKQW67Mj/images/3e7f965619c868b1860965c878110bc5f1bab3dc23c1ef22b1ae8806227859a2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9271755663e317d22784d427c92dfe671fbd867b6ecfa81deaaf8b6b56e49c8c +size 5731 diff --git a/parse/train/TScOKQW67Mj/images/3eeb6aa7e389a3166d9eb880eac8e3c7c21890d6008716eb8a28ac7dc995876d.jpg b/parse/train/TScOKQW67Mj/images/3eeb6aa7e389a3166d9eb880eac8e3c7c21890d6008716eb8a28ac7dc995876d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ebcdeafa3b0fab7232d1409dd0f87a3001913802 --- /dev/null +++ b/parse/train/TScOKQW67Mj/images/3eeb6aa7e389a3166d9eb880eac8e3c7c21890d6008716eb8a28ac7dc995876d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:88a0a045328e95fde2234b8a5442765127e18f7ec3cd5b46ffdb7bcd2981e3cf +size 6093 diff --git a/parse/train/TScOKQW67Mj/images/3ef719ad199ed10bb386f869d510c72f28f844604126e4dc885e5655f56702ff.jpg b/parse/train/TScOKQW67Mj/images/3ef719ad199ed10bb386f869d510c72f28f844604126e4dc885e5655f56702ff.jpg new file mode 100644 index 0000000000000000000000000000000000000000..636011eaaccd554b1d08c11ec08f32af226e899e --- /dev/null +++ b/parse/train/TScOKQW67Mj/images/3ef719ad199ed10bb386f869d510c72f28f844604126e4dc885e5655f56702ff.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7e4494d6cbb0c11ab4356bdb25dce315d284f2769a80d06525c0eb6b2867b2c4 +size 5541 diff --git a/parse/train/TScOKQW67Mj/images/4cfd48b9db0746072b72af0de68b39ab4016c8d335b1958b155742ab3f026d99.jpg b/parse/train/TScOKQW67Mj/images/4cfd48b9db0746072b72af0de68b39ab4016c8d335b1958b155742ab3f026d99.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2b41099604dd062c6125611029c359183f0e7742 --- /dev/null +++ b/parse/train/TScOKQW67Mj/images/4cfd48b9db0746072b72af0de68b39ab4016c8d335b1958b155742ab3f026d99.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4dc9367c3844b5c5725aa82ba898c85425756ca4a4edbb4f8320e1a08d1e018f +size 34046 diff --git a/parse/train/TScOKQW67Mj/images/63b875463223072a756a88240fbb3c990bbfc1acb164be1af730b837f1c16e7d.jpg b/parse/train/TScOKQW67Mj/images/63b875463223072a756a88240fbb3c990bbfc1acb164be1af730b837f1c16e7d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ae133ca49f8522b04c784e8bf47def2602499d34 --- /dev/null +++ b/parse/train/TScOKQW67Mj/images/63b875463223072a756a88240fbb3c990bbfc1acb164be1af730b837f1c16e7d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5030350e55b6d0197827eb873fb26bb8a9ac4c9cf2871b67ba2e3472366e3696 +size 5647 diff --git a/parse/train/TScOKQW67Mj/images/67339367399355818951e439e4b03006b2efddb824fe5ef4573766f2b7054a4c.jpg b/parse/train/TScOKQW67Mj/images/67339367399355818951e439e4b03006b2efddb824fe5ef4573766f2b7054a4c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..599492f0ecdac1be6d2d43a1b564e68d2826ecf2 --- /dev/null +++ b/parse/train/TScOKQW67Mj/images/67339367399355818951e439e4b03006b2efddb824fe5ef4573766f2b7054a4c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6ca232f88ee0aa14e53fcfac43404b0d2afe00689d3bb182e50f13f283c4bb91 +size 6406 diff --git a/parse/train/TScOKQW67Mj/images/6bf2f1e1e7866cae718c8f6238665f93e6a3f031263848f929f44aedbeb5fb88.jpg b/parse/train/TScOKQW67Mj/images/6bf2f1e1e7866cae718c8f6238665f93e6a3f031263848f929f44aedbeb5fb88.jpg new file mode 100644 index 0000000000000000000000000000000000000000..38e5a22da4e5496e8e4120a626b2708eea19d74c --- /dev/null +++ b/parse/train/TScOKQW67Mj/images/6bf2f1e1e7866cae718c8f6238665f93e6a3f031263848f929f44aedbeb5fb88.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4928e0e973536f9adf5f559386bd1a1852f304b36285702d55c7164ee7fc02d1 +size 76403 diff --git a/parse/train/TScOKQW67Mj/images/780a766770dcd8458dab8aa7254aaca93fb9a45f81584481cc5375bc6ed8a884.jpg b/parse/train/TScOKQW67Mj/images/780a766770dcd8458dab8aa7254aaca93fb9a45f81584481cc5375bc6ed8a884.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0b087559eed9b87c103b18dca82269588139d053 --- /dev/null +++ b/parse/train/TScOKQW67Mj/images/780a766770dcd8458dab8aa7254aaca93fb9a45f81584481cc5375bc6ed8a884.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:061682e2bc0a1d9d2363c30107e9d18b7f7c4496fcaa08e7d909dbbd0e80469d +size 4301 diff --git a/parse/train/TScOKQW67Mj/images/a2d0098d330c6217878fa160f5e6520aff0c5f2421a8b03dd2beb0f03f404954.jpg b/parse/train/TScOKQW67Mj/images/a2d0098d330c6217878fa160f5e6520aff0c5f2421a8b03dd2beb0f03f404954.jpg new file mode 100644 index 0000000000000000000000000000000000000000..57ca0a51e2b93ec146d68c290bb351fa3cd26e3d --- /dev/null +++ b/parse/train/TScOKQW67Mj/images/a2d0098d330c6217878fa160f5e6520aff0c5f2421a8b03dd2beb0f03f404954.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0f836bb44fb54203e5230d1c06e3cfc728190a18ab09fae33af04940a81018e7 +size 50301 diff --git a/parse/train/TScOKQW67Mj/images/d68cc2e28b1b9e274d28ca88c036e9bccaecefb85e19b46d45c093abfa0373c1.jpg b/parse/train/TScOKQW67Mj/images/d68cc2e28b1b9e274d28ca88c036e9bccaecefb85e19b46d45c093abfa0373c1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..891582c41ab31ad3251e747e03538d18c09eb5f3 --- /dev/null +++ b/parse/train/TScOKQW67Mj/images/d68cc2e28b1b9e274d28ca88c036e9bccaecefb85e19b46d45c093abfa0373c1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5c1e34a6320065a7434ee6abb5518f4683efda93717e856292eae5335006df4a +size 7538 diff --git a/parse/train/TScOKQW67Mj/images/e3fbfc3ced46fcebe16d429cac63ce9d30373f76281f3704a14ef61faf0c4aae.jpg b/parse/train/TScOKQW67Mj/images/e3fbfc3ced46fcebe16d429cac63ce9d30373f76281f3704a14ef61faf0c4aae.jpg new file mode 100644 index 0000000000000000000000000000000000000000..02a6995508de5853d9fb0d43dc1e52dc350eb4c1 --- /dev/null +++ b/parse/train/TScOKQW67Mj/images/e3fbfc3ced46fcebe16d429cac63ce9d30373f76281f3704a14ef61faf0c4aae.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ca37b2ca1561c815cf192b2a21cf7926f032e24f5b9f5852dc142e8ec12cd189 +size 54743 diff --git a/parse/train/TScOKQW67Mj/images/fd92bf06428cd0ba128ceffac2f7b143f36c94a058964482dce4b6205b8af4d2.jpg b/parse/train/TScOKQW67Mj/images/fd92bf06428cd0ba128ceffac2f7b143f36c94a058964482dce4b6205b8af4d2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8f334a2ec308f32286ee84f597adf41b63a7334d --- /dev/null +++ b/parse/train/TScOKQW67Mj/images/fd92bf06428cd0ba128ceffac2f7b143f36c94a058964482dce4b6205b8af4d2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d559caccb527fb46b565dbfbf54605b6f0495e1d0bd7d9e52fcae313a2ed4f67 +size 4119 diff --git a/parse/train/Y4RJQO2jivm/images/0dc6f1c5170894fb1079ff0e126bbe04a06e3277531931379d08860f453c6496.jpg b/parse/train/Y4RJQO2jivm/images/0dc6f1c5170894fb1079ff0e126bbe04a06e3277531931379d08860f453c6496.jpg new file mode 100644 index 0000000000000000000000000000000000000000..13ff4159a4fb68a284eea45da663d1cf0d008827 --- /dev/null +++ b/parse/train/Y4RJQO2jivm/images/0dc6f1c5170894fb1079ff0e126bbe04a06e3277531931379d08860f453c6496.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f9944e1f1f1397fa7a9277b6e5d22742afb2ce7228018953f83e2bef084eb525 +size 16511 diff --git a/parse/train/Y4RJQO2jivm/images/3b78823e4a911b0c11023be6953766bcaaa5b74e18160e0e6f8e113a9888cbfc.jpg b/parse/train/Y4RJQO2jivm/images/3b78823e4a911b0c11023be6953766bcaaa5b74e18160e0e6f8e113a9888cbfc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..290af9700c53a4e53d24f6f042a4f7f098f02937 --- /dev/null +++ b/parse/train/Y4RJQO2jivm/images/3b78823e4a911b0c11023be6953766bcaaa5b74e18160e0e6f8e113a9888cbfc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:490e5ace106cdbe4e5559479152f63da539026d58c8a4760d97df2ca47fe39b8 +size 6142 diff --git a/parse/train/Y4RJQO2jivm/images/3e9e1eca0aaaff32fe77d54bb6b0b97c94fca8880f22c1f363f30ee98fc2e35f.jpg b/parse/train/Y4RJQO2jivm/images/3e9e1eca0aaaff32fe77d54bb6b0b97c94fca8880f22c1f363f30ee98fc2e35f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9207afd6b5608f8fd7bc2833176c134d7508803a --- /dev/null +++ b/parse/train/Y4RJQO2jivm/images/3e9e1eca0aaaff32fe77d54bb6b0b97c94fca8880f22c1f363f30ee98fc2e35f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6d79f64e70c025b62bd99299ff3f300259d718032b6229c7d2dc74d917d7d5fd +size 26813 diff --git a/parse/train/Y4RJQO2jivm/images/614f29dff36a6519e42ce212cbb5aa93a68a076bc3b562246f821453fecc3457.jpg b/parse/train/Y4RJQO2jivm/images/614f29dff36a6519e42ce212cbb5aa93a68a076bc3b562246f821453fecc3457.jpg new file mode 100644 index 0000000000000000000000000000000000000000..00a14bb20c5a86d7d47d8b2ace977d5d60e30b56 --- /dev/null +++ b/parse/train/Y4RJQO2jivm/images/614f29dff36a6519e42ce212cbb5aa93a68a076bc3b562246f821453fecc3457.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cf3a9fae83e6e170f30d7c5c83c127d18e6cbfcfc55b9720e5c5fe964deaf93d +size 31100 diff --git a/parse/train/Y4RJQO2jivm/images/8b83bd8c825c733d3485b5a8200dfc7d60f7b93a62f64b5df2595c9f6f9b2b4f.jpg b/parse/train/Y4RJQO2jivm/images/8b83bd8c825c733d3485b5a8200dfc7d60f7b93a62f64b5df2595c9f6f9b2b4f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..38da08ca9a22b285ea4db3e6d269ace77577f12a --- /dev/null +++ b/parse/train/Y4RJQO2jivm/images/8b83bd8c825c733d3485b5a8200dfc7d60f7b93a62f64b5df2595c9f6f9b2b4f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:42efa69cc1373ce1e40882b622a93b2fc0ff142531e5b4f66d8519ce9b33fef6 +size 37047 diff --git a/parse/train/Y4RJQO2jivm/images/8f1dd5b12cc123508d9556718f84534af03609fbb67172024262913ac4297253.jpg b/parse/train/Y4RJQO2jivm/images/8f1dd5b12cc123508d9556718f84534af03609fbb67172024262913ac4297253.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e6c29d8028e223c2b792afbf5d3c68fec43ebe99 --- /dev/null +++ b/parse/train/Y4RJQO2jivm/images/8f1dd5b12cc123508d9556718f84534af03609fbb67172024262913ac4297253.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e346fede70bfdfa96b0ea8ecc0a70904b8c49aa2d995794732567c444d725de1 +size 19452 diff --git a/parse/train/Y4RJQO2jivm/images/a1c3ceb95e91048ec8b4d3a49c34c3dbf66e29180b8005562419aa4c0258b35d.jpg b/parse/train/Y4RJQO2jivm/images/a1c3ceb95e91048ec8b4d3a49c34c3dbf66e29180b8005562419aa4c0258b35d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..892380be4c1ca02c93743f90ae471f485bae7d18 --- /dev/null +++ b/parse/train/Y4RJQO2jivm/images/a1c3ceb95e91048ec8b4d3a49c34c3dbf66e29180b8005562419aa4c0258b35d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1c9bb15d7aed55d4c2b39134b1a178f3f5065a8f5a798f90364a33fac14e567b +size 44861 diff --git a/parse/train/Y4RJQO2jivm/images/aef78747b1a5d74e877194462c516c6adf98b8e402ddcefa7c5a77926354d0d6.jpg b/parse/train/Y4RJQO2jivm/images/aef78747b1a5d74e877194462c516c6adf98b8e402ddcefa7c5a77926354d0d6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..93c4d7442bd3a9788c29b6b18374f5ad64849799 --- /dev/null +++ b/parse/train/Y4RJQO2jivm/images/aef78747b1a5d74e877194462c516c6adf98b8e402ddcefa7c5a77926354d0d6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:621f6376b0be41c61e417c05fa50d846590ac613c355d6b08b0f8f018fe873bb +size 69271 diff --git a/parse/train/Y4RJQO2jivm/images/edfa7e3a4d69c92e2755d8f053f6f4e75391f4b78a888fd91e3842c79a760a21.jpg b/parse/train/Y4RJQO2jivm/images/edfa7e3a4d69c92e2755d8f053f6f4e75391f4b78a888fd91e3842c79a760a21.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6e1e25b43aa1d5b35ba7eab5954677c44de43030 --- /dev/null +++ b/parse/train/Y4RJQO2jivm/images/edfa7e3a4d69c92e2755d8f053f6f4e75391f4b78a888fd91e3842c79a760a21.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:619f1683fd7cffbb5ba7e81026036359b7281321fdd7a2e0a5593f120155759a +size 108044 diff --git a/parse/train/YJL3amnDf2w/images/00f98f73f46f71d6df911c29c060f2e118accc18e86fd0811ff7c7e21448b17e.jpg b/parse/train/YJL3amnDf2w/images/00f98f73f46f71d6df911c29c060f2e118accc18e86fd0811ff7c7e21448b17e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3b05e217258f3b2d138bd2d01eabee04935eac94 --- /dev/null +++ b/parse/train/YJL3amnDf2w/images/00f98f73f46f71d6df911c29c060f2e118accc18e86fd0811ff7c7e21448b17e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7d4c87b75ca32d28d7f9d46705ef22ff712cf4dace242411c83042cc2e6868f1 +size 15248 diff --git a/parse/train/YJL3amnDf2w/images/085cd134b67ef651f05c215634d5b19d46cdc030e19d1f9fcafa7a1527a5e897.jpg b/parse/train/YJL3amnDf2w/images/085cd134b67ef651f05c215634d5b19d46cdc030e19d1f9fcafa7a1527a5e897.jpg new file mode 100644 index 0000000000000000000000000000000000000000..05c90697b62f586bd926fc8650bcd0643d36a0f8 --- /dev/null +++ b/parse/train/YJL3amnDf2w/images/085cd134b67ef651f05c215634d5b19d46cdc030e19d1f9fcafa7a1527a5e897.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cc50a00f02e08b7a7f29ae897365eb04d279e66d4a9ca6159991bed16be20ec2 +size 13773 diff --git a/parse/train/YJL3amnDf2w/images/281e9df0b0689f87146fe1b2cf41575e4d0bd6ba7ba76ea282d9fa40c4d26f24.jpg b/parse/train/YJL3amnDf2w/images/281e9df0b0689f87146fe1b2cf41575e4d0bd6ba7ba76ea282d9fa40c4d26f24.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9a9ab82ba3ab618d85fdc57a1bff9ae2349614cb --- /dev/null +++ b/parse/train/YJL3amnDf2w/images/281e9df0b0689f87146fe1b2cf41575e4d0bd6ba7ba76ea282d9fa40c4d26f24.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f11a0062f904ed74209ef35c2326b7bc05108f23fe1aab4076c8960f174ec8da +size 43932 diff --git a/parse/train/YJL3amnDf2w/images/4ae49d47ed5f3dfc42c4c4576a49ef5b9b79cfcf5d867d423af4cbb208ff394d.jpg b/parse/train/YJL3amnDf2w/images/4ae49d47ed5f3dfc42c4c4576a49ef5b9b79cfcf5d867d423af4cbb208ff394d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f37c2aa9ee575c942b6628b4f6a8bcaadfd4a4a2 --- /dev/null +++ b/parse/train/YJL3amnDf2w/images/4ae49d47ed5f3dfc42c4c4576a49ef5b9b79cfcf5d867d423af4cbb208ff394d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:57a5c2c14f6f8bd5fb3a4a7a472603c8c5a31537841d944e28ce38cc95a2b7c0 +size 30292 diff --git a/parse/train/YJL3amnDf2w/images/53d9e86c84294cf4c12deba92e0105ab0f84f8bccce5a7589a7cf2b53504f5a6.jpg b/parse/train/YJL3amnDf2w/images/53d9e86c84294cf4c12deba92e0105ab0f84f8bccce5a7589a7cf2b53504f5a6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..32ab1db256fc6747a82fc19784485ebd9731f3d2 --- /dev/null +++ b/parse/train/YJL3amnDf2w/images/53d9e86c84294cf4c12deba92e0105ab0f84f8bccce5a7589a7cf2b53504f5a6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:48549693bc3447f21a08ffbc5869d7b69cd17a0d769fbd6b157d090fa6cbef41 +size 10320 diff --git a/parse/train/YJL3amnDf2w/images/559a51ccfd5db65c11944cfa01d77ecba6b3c1db92ca256be94031cb550f4a14.jpg b/parse/train/YJL3amnDf2w/images/559a51ccfd5db65c11944cfa01d77ecba6b3c1db92ca256be94031cb550f4a14.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a7e89b9a92c1532d2c6e89c4eff7717dad3e9441 --- /dev/null +++ b/parse/train/YJL3amnDf2w/images/559a51ccfd5db65c11944cfa01d77ecba6b3c1db92ca256be94031cb550f4a14.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:43415eb9a9de531d779bb7f9f917ee78e9029b001be824b6aeb44645477eeaa3 +size 36068 diff --git a/parse/train/YJL3amnDf2w/images/6354166f46f5911900c5e12b2ffffa62366acc4c2bc055304d040883cc44a054.jpg b/parse/train/YJL3amnDf2w/images/6354166f46f5911900c5e12b2ffffa62366acc4c2bc055304d040883cc44a054.jpg new file mode 100644 index 0000000000000000000000000000000000000000..00539505e8dd257d582030959b0d5470dcc5817f --- /dev/null +++ b/parse/train/YJL3amnDf2w/images/6354166f46f5911900c5e12b2ffffa62366acc4c2bc055304d040883cc44a054.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ad6f5c36140637099a6fae24e937b78ba28b65533783bc99753c208b63691eb9 +size 14821 diff --git a/parse/train/YJL3amnDf2w/images/66305ef0f9e0887db98dc89187307f6f7bfba19b94af36b57fbc8ef6631e54c3.jpg b/parse/train/YJL3amnDf2w/images/66305ef0f9e0887db98dc89187307f6f7bfba19b94af36b57fbc8ef6631e54c3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6626d78614889fa3153a64091d6a5fe51fac64ce --- /dev/null +++ b/parse/train/YJL3amnDf2w/images/66305ef0f9e0887db98dc89187307f6f7bfba19b94af36b57fbc8ef6631e54c3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:602372a91603ff5478325f38287c502495b5b702447a14af40c748c835b6d047 +size 6068 diff --git a/parse/train/YJL3amnDf2w/images/6c3c68081622c8acff26b5f17769ffe7824c85830952a870eb50159ad90c49ae.jpg b/parse/train/YJL3amnDf2w/images/6c3c68081622c8acff26b5f17769ffe7824c85830952a870eb50159ad90c49ae.jpg new file mode 100644 index 0000000000000000000000000000000000000000..06abcd4838dde4aa0ebdf685c1a7c7b110b7a1d6 --- /dev/null +++ b/parse/train/YJL3amnDf2w/images/6c3c68081622c8acff26b5f17769ffe7824c85830952a870eb50159ad90c49ae.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b1bb4d6d4d7e74ba1bcf78000c1f85e3c7f421e0266a7a64067e3773ca74e4ef +size 42219 diff --git a/parse/train/YJL3amnDf2w/images/6e31eb74fc7cdf45d33b666c97857207f158e84fb8b3932f16e05bb37b351085.jpg b/parse/train/YJL3amnDf2w/images/6e31eb74fc7cdf45d33b666c97857207f158e84fb8b3932f16e05bb37b351085.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2df8e84bf7c980fe12e805ac923830a58f7d46e2 --- /dev/null +++ b/parse/train/YJL3amnDf2w/images/6e31eb74fc7cdf45d33b666c97857207f158e84fb8b3932f16e05bb37b351085.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4db57084a4a943141ac37df64d899628876363c4f0bcdcb0f655b3c3f05e926d +size 44674 diff --git a/parse/train/YJL3amnDf2w/images/73e39d3dc4753ca462ee1e95a548c430992d2bf235245ca45f32899cf34fe7f5.jpg b/parse/train/YJL3amnDf2w/images/73e39d3dc4753ca462ee1e95a548c430992d2bf235245ca45f32899cf34fe7f5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5db12e6ce00d08ae6d6c701f5b3e58ca8e7dfdf2 --- /dev/null +++ b/parse/train/YJL3amnDf2w/images/73e39d3dc4753ca462ee1e95a548c430992d2bf235245ca45f32899cf34fe7f5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0a356fbfa3c9e4749ca37504134f9b052054ff8c6fa036a56536dc6576307ff8 +size 29970 diff --git a/parse/train/YJL3amnDf2w/images/8559c917c0acc8c7d69786cf6f5de571611c840fdc3d7095265d913c78fafd15.jpg b/parse/train/YJL3amnDf2w/images/8559c917c0acc8c7d69786cf6f5de571611c840fdc3d7095265d913c78fafd15.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fe9170e371e7f19fc9e664e3d419579fe25636bc --- /dev/null +++ b/parse/train/YJL3amnDf2w/images/8559c917c0acc8c7d69786cf6f5de571611c840fdc3d7095265d913c78fafd15.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0f1b80e86679bdd5e9a2b3420b0908a5e21264c6158966c69d12b1252f9d4dde +size 38107 diff --git a/parse/train/YJL3amnDf2w/images/8d0660d82604d5fc291491a839afd4989f3459f34be1679ba839e7b3ce2c5f41.jpg b/parse/train/YJL3amnDf2w/images/8d0660d82604d5fc291491a839afd4989f3459f34be1679ba839e7b3ce2c5f41.jpg new file mode 100644 index 0000000000000000000000000000000000000000..420e390b5a6b9ed647881df2f6428080a8708b36 --- /dev/null +++ b/parse/train/YJL3amnDf2w/images/8d0660d82604d5fc291491a839afd4989f3459f34be1679ba839e7b3ce2c5f41.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c9771107d636772c14a45eea078ed1afb74ee09cad51b9b75943c13e600a7e52 +size 21691 diff --git a/parse/train/YJL3amnDf2w/images/b015cfad36b9dee8a12d50a50a085eefe7f8c5ef7794e608689df09756bccb7a.jpg b/parse/train/YJL3amnDf2w/images/b015cfad36b9dee8a12d50a50a085eefe7f8c5ef7794e608689df09756bccb7a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..99dd4ab4944739fc908f600928411276fbc94fa3 --- /dev/null +++ b/parse/train/YJL3amnDf2w/images/b015cfad36b9dee8a12d50a50a085eefe7f8c5ef7794e608689df09756bccb7a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7bed03a28d0a93402893150fed42c36324eadf3b1af8816e86a5c7fb8108be2c +size 14035 diff --git a/parse/train/YJL3amnDf2w/images/c72d2cbf791099982c447bc978948aa767a9f483be3aba1bb03557373ece3783.jpg b/parse/train/YJL3amnDf2w/images/c72d2cbf791099982c447bc978948aa767a9f483be3aba1bb03557373ece3783.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5fa531b70e875d8e0a38b08407b8f038d03a006a --- /dev/null +++ b/parse/train/YJL3amnDf2w/images/c72d2cbf791099982c447bc978948aa767a9f483be3aba1bb03557373ece3783.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6d402780e5412eb880a5bf96de9f16d89852368b574911a1b6a2a28e76d214f2 +size 13059 diff --git a/parse/train/YJL3amnDf2w/images/cb5993891431f609fb6744917a66f2dbcbf6927945c5dec26084f7a46780a3e0.jpg b/parse/train/YJL3amnDf2w/images/cb5993891431f609fb6744917a66f2dbcbf6927945c5dec26084f7a46780a3e0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ac93fce41cb078d4fc9d717b30c316bf9d21cf64 --- /dev/null +++ b/parse/train/YJL3amnDf2w/images/cb5993891431f609fb6744917a66f2dbcbf6927945c5dec26084f7a46780a3e0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0a6ef3492ebfe8b27230576e049821272feaad5fd8903c4319bd0d0b19dc0d95 +size 8679 diff --git a/parse/train/YJL3amnDf2w/images/dc07914ebe395995851df33d9261d32b164361fa45423f942ca3fe37e5fe7f27.jpg b/parse/train/YJL3amnDf2w/images/dc07914ebe395995851df33d9261d32b164361fa45423f942ca3fe37e5fe7f27.jpg new file mode 100644 index 0000000000000000000000000000000000000000..fb4f863f89d7e793916b63a7712f2b1b88122e03 --- /dev/null +++ b/parse/train/YJL3amnDf2w/images/dc07914ebe395995851df33d9261d32b164361fa45423f942ca3fe37e5fe7f27.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a4e7dabe9300edca68a2b26bae82de2124946069890a2351b3897131dc93bf87 +size 20480 diff --git a/parse/train/_X_4Akcd8Re/images/06bb11307f2db06390766e78d948efd63e255df6301abb6d361df3262f392f60.jpg b/parse/train/_X_4Akcd8Re/images/06bb11307f2db06390766e78d948efd63e255df6301abb6d361df3262f392f60.jpg new file mode 100644 index 0000000000000000000000000000000000000000..48c7f6312b4357442d31966f4131a637da26df9f --- /dev/null +++ b/parse/train/_X_4Akcd8Re/images/06bb11307f2db06390766e78d948efd63e255df6301abb6d361df3262f392f60.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4b6ad3bfdaf30c6ba5c35b03e42d1e608da1045503ea39ae2eae40bc88bc56aa +size 8845 diff --git a/parse/train/_X_4Akcd8Re/images/0b41d50c27cc9b6937b35a3119bc78bc8cf402bfc6746959eea4c38dcc388422.jpg b/parse/train/_X_4Akcd8Re/images/0b41d50c27cc9b6937b35a3119bc78bc8cf402bfc6746959eea4c38dcc388422.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d68b36fe39803b57d533cdd89b17e76c5b06ced8 --- /dev/null +++ b/parse/train/_X_4Akcd8Re/images/0b41d50c27cc9b6937b35a3119bc78bc8cf402bfc6746959eea4c38dcc388422.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4aabeeab4f30b66adb87a54fd69f8f8aad3780f9ab3f2891a90e495d22bb589a +size 60264 diff --git a/parse/train/_X_4Akcd8Re/images/19d566559141436db945d2174916ecbf1ddaefb8284bcb860135b0361e2e770c.jpg b/parse/train/_X_4Akcd8Re/images/19d566559141436db945d2174916ecbf1ddaefb8284bcb860135b0361e2e770c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..edc5c0edd09fb3bc735d9e375a7d070434e3502e --- /dev/null +++ b/parse/train/_X_4Akcd8Re/images/19d566559141436db945d2174916ecbf1ddaefb8284bcb860135b0361e2e770c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:04c3ec9562798d13759b29250d304cc5fcb02ddaa5d8c6f1febc97eaf29d0e51 +size 66704 diff --git a/parse/train/_X_4Akcd8Re/images/1ae42b21e1b94441e8e221df5f318c0df61a2557097a77564e791cc14f84c7aa.jpg b/parse/train/_X_4Akcd8Re/images/1ae42b21e1b94441e8e221df5f318c0df61a2557097a77564e791cc14f84c7aa.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a9351f2f87c9ba4db3e1bb41680d5623dd944445 --- /dev/null +++ b/parse/train/_X_4Akcd8Re/images/1ae42b21e1b94441e8e221df5f318c0df61a2557097a77564e791cc14f84c7aa.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e01f33ea163748db6fd26ff8ea4ca8d084e2f0bfa2ba996f0faf8fe819120d74 +size 18828 diff --git a/parse/train/_X_4Akcd8Re/images/20621f3512e9e9f0213f9a8b079f53dbfbc18c9b7fa48b914ba280017c3cbf6e.jpg b/parse/train/_X_4Akcd8Re/images/20621f3512e9e9f0213f9a8b079f53dbfbc18c9b7fa48b914ba280017c3cbf6e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7b3271cd5587d0d34b863c96ddd4978d40027083 --- /dev/null +++ b/parse/train/_X_4Akcd8Re/images/20621f3512e9e9f0213f9a8b079f53dbfbc18c9b7fa48b914ba280017c3cbf6e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9511e090a9a21908951f9cd96ef8b92fdefb6cf806c6ce92ac6e1b7c8b4734dc +size 16865 diff --git a/parse/train/_X_4Akcd8Re/images/23b9b79138180f34629dc569b6a4984dc827218b3c64a7a4c7b93886a4ed09ba.jpg b/parse/train/_X_4Akcd8Re/images/23b9b79138180f34629dc569b6a4984dc827218b3c64a7a4c7b93886a4ed09ba.jpg new file mode 100644 index 0000000000000000000000000000000000000000..144688492990231291f88f584951b27b088725d9 --- /dev/null +++ b/parse/train/_X_4Akcd8Re/images/23b9b79138180f34629dc569b6a4984dc827218b3c64a7a4c7b93886a4ed09ba.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c59d367ca8433e05817a51130e6278c45e41d8b08e89a72cade5105e51d4e3a6 +size 4741 diff --git a/parse/train/_X_4Akcd8Re/images/24a1d1d978ce8402f6243581c8fb72ffc7215b4eee77da1487bce904debebb13.jpg b/parse/train/_X_4Akcd8Re/images/24a1d1d978ce8402f6243581c8fb72ffc7215b4eee77da1487bce904debebb13.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ff49ef6bdf584cecdd41c31c33cd3d50585a1ddc --- /dev/null +++ b/parse/train/_X_4Akcd8Re/images/24a1d1d978ce8402f6243581c8fb72ffc7215b4eee77da1487bce904debebb13.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4e36cc405060c95bed595dbc27d4fae08268f61d3719dc036f31bf5f34ed05dd +size 30720 diff --git a/parse/train/_X_4Akcd8Re/images/3b41d5e72e65b3c83e8cbf9dcf6ded21e5ae03eefb7e5fd08412600d6e70ad1a.jpg b/parse/train/_X_4Akcd8Re/images/3b41d5e72e65b3c83e8cbf9dcf6ded21e5ae03eefb7e5fd08412600d6e70ad1a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..662b9c90d7b8a0029c89d3f73b36ff136a758899 --- /dev/null +++ b/parse/train/_X_4Akcd8Re/images/3b41d5e72e65b3c83e8cbf9dcf6ded21e5ae03eefb7e5fd08412600d6e70ad1a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c8853c8ac085f539b1a68a7b59e66216b8f7e5aaea5a13f1d81e0fabb012667c +size 63783 diff --git a/parse/train/_X_4Akcd8Re/images/5625a4aae9c47f60c1232de28cfa80e4a0cfdaaf5b122056b2355f22deb483f0.jpg b/parse/train/_X_4Akcd8Re/images/5625a4aae9c47f60c1232de28cfa80e4a0cfdaaf5b122056b2355f22deb483f0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..eb467f01b95d5f7e39ce39cc7831740ce37c861e --- /dev/null +++ b/parse/train/_X_4Akcd8Re/images/5625a4aae9c47f60c1232de28cfa80e4a0cfdaaf5b122056b2355f22deb483f0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6363b6a39bd15d80e3f25a6bbf333d3f1bd0eb7e5264ee76f69116d427f4b40b +size 50968 diff --git a/parse/train/_X_4Akcd8Re/images/5d9e390619b5491c2cb7f6e579355ed31cdbf899381189db3e63cd3983b125e3.jpg b/parse/train/_X_4Akcd8Re/images/5d9e390619b5491c2cb7f6e579355ed31cdbf899381189db3e63cd3983b125e3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b43db206df34619c1965f8507a72db034b8e6459 --- /dev/null +++ b/parse/train/_X_4Akcd8Re/images/5d9e390619b5491c2cb7f6e579355ed31cdbf899381189db3e63cd3983b125e3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d9582c114c9a08d60ff4bc40e753404deec5fd02af92048a8240c82613e7ef0d +size 5935 diff --git a/parse/train/_X_4Akcd8Re/images/7beace613bc020608801052c89f18747fd9be818952f8c7a379d6c9c7420a4fe.jpg b/parse/train/_X_4Akcd8Re/images/7beace613bc020608801052c89f18747fd9be818952f8c7a379d6c9c7420a4fe.jpg new file mode 100644 index 0000000000000000000000000000000000000000..89e4a2ea433badc53f876d9096722bdbda064b93 --- /dev/null +++ b/parse/train/_X_4Akcd8Re/images/7beace613bc020608801052c89f18747fd9be818952f8c7a379d6c9c7420a4fe.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:14c268980324df61af33dd9496f88d18a1a511fee1669e2a86088e1cb740824f +size 11269 diff --git a/parse/train/_X_4Akcd8Re/images/acb2f9c6263005e18846f3081fe5543ab0b5272ebc1e0dc0e6beae6d4ca494a6.jpg b/parse/train/_X_4Akcd8Re/images/acb2f9c6263005e18846f3081fe5543ab0b5272ebc1e0dc0e6beae6d4ca494a6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c283359628ec3f1acf081bcd6937954b13111ced --- /dev/null +++ b/parse/train/_X_4Akcd8Re/images/acb2f9c6263005e18846f3081fe5543ab0b5272ebc1e0dc0e6beae6d4ca494a6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c5fb496ffd2588c77e6a195a59be29e339bb9291a015bdc1e2d7f81491ec43ca +size 10544 diff --git a/parse/train/_X_4Akcd8Re/images/d2f7e576d8d7fe742bce84281066ea6d0abae4d18ceb29d317b44eab945243b3.jpg b/parse/train/_X_4Akcd8Re/images/d2f7e576d8d7fe742bce84281066ea6d0abae4d18ceb29d317b44eab945243b3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a8a8de5e2a76d83946bb5e7e4c44959bbd329879 --- /dev/null +++ b/parse/train/_X_4Akcd8Re/images/d2f7e576d8d7fe742bce84281066ea6d0abae4d18ceb29d317b44eab945243b3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b944cf11a6afc782bb895e3847ae2915dd4f25ddce174624fbc56d20d82dd89b +size 75035 diff --git a/parse/train/_X_4Akcd8Re/images/e8978241f8e12ee992b116aed5b9805c5b794ef308b29af359b9de918009a725.jpg b/parse/train/_X_4Akcd8Re/images/e8978241f8e12ee992b116aed5b9805c5b794ef308b29af359b9de918009a725.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7c1f16c1d700da22d52ef6f84966013b5cb7d759 --- /dev/null +++ b/parse/train/_X_4Akcd8Re/images/e8978241f8e12ee992b116aed5b9805c5b794ef308b29af359b9de918009a725.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ba211960634ac4b10b651be1fe8c62c621f928ff694dddac6154c7f508f907db +size 52030 diff --git a/parse/train/jCxDyge46t2/jCxDyge46t2.md b/parse/train/jCxDyge46t2/jCxDyge46t2.md new file mode 100644 index 0000000000000000000000000000000000000000..0725d3ecc8088f7304df17d018df62bc49bc8861 --- /dev/null +++ b/parse/train/jCxDyge46t2/jCxDyge46t2.md @@ -0,0 +1,528 @@ +# Neural Contextual Bandits with Deep Representation and Shallow Exploration + +Anonymous Author(s) +Affiliation +Address +email + +# Abstract + +1 We study neural contextual bandits, a general class of contextual bandits, where +2 each context-action pair is associated with a raw feature vector, but the specific +3 reward generating function is unknown. We propose a novel learning algorithm +4 that transforms the raw feature vector using the last hidden layer of a deep ReLU +5 neural network (deep representation learning), and uses an upper confidence bound +6 (UCB) approach to explore in the last linear layer (shallow exploration). We prove +7 that under standard assumptions, our proposed algorithm achieves $\widetilde { O } ( \sqrt { T } )$ finite +8 time regret, where $T$ is the learning time horizon. Compared with existing neural +9 contextual bandit algorithms, our approach is computationally much more efficient +10 since it only needs to explore in the last layer of the deep neural network. + +# 11 1 Introduction + +12 Multi-armed bandits (MAB) [9, 8, 30] are a class of online decision-making problems where an +13 agent needs to learn to maximize its expected cumulative reward while repeatedly interacting with a +14 partially known environment. Based on a bandit algorithm (also called a strategy or policy), in each +15 round, the agent adaptively chooses an arm, and then observes and receives a reward associated with +16 that arm. Since only the reward of the chosen arm will be observed (bandit information feedback), +17 a good bandit algorithm has to deal with the exploration-exploitation dilemma: trade-off between +18 pulling the best arm based on existing knowledge/history data (exploitation) and trying the arms that +19 have not been fully explored (exploration). +20 In many real-world applications, the agent will also be able to access detailed contexts associated +21 with the arms. For example, when a company wants to choose an advertisement to present to a user, +22 the recommendation will be much more accurate if the company takes into consideration the contents, +23 specifications, and other features of the advertisements in the arm set as well as the profile of the user. +24 To encode the contextual information, contextual bandit models and algorithms have been developed, +25 and widely studied both in theory and in practice [19, 39, 34, 16, 1]. Most existing contextual bandit +26 algorithms assume that the expected reward of an arm at a context is a linear function in a known +27 context-action feature vector, which leads to many useful algorithms such as LinUCB [16], OFUL [1], +28 etc. The representation power of the linear model can be limited in applications such as marketing, +29 social networking, clinical studies, etc., where the rewards are usually counts or binary variables. The +30 linear contextual bandit problem has also been extended to richer classes of parametric bandits such +31 as the generalized linear bandits [24, 35] and kernelised bandits [44, 15]. +32 With the prevalence of deep neural networks (DNNs) and their phenomenal performances in many +33 machine learning tasks [32, 25], there has emerged a line of work that employs DNNs to increase the +34 representation power of contextual bandit algorithms [5, 38, 17, 49, 52, 20, 51]. The problems they +35 solve are usually referred to as neural contextual bandits. For example, Zhou et al. [52] developed +36 the NeuralUCB algorithm, which can be viewed as a natural extension of LinUCB [16, 1], where they +37 use the output of a deep neural network with the feature vector as input to approximate the reward. +38 Zhang et al. [51] adapted neural networks in Thompson Sampling [43, 14, 40] for both exploration +39 and exploitation and proposed NeuralTS . For a fixed time horizon $T$ , it has been proved that both +40 NeuralUCB and NeuralTS achieve a $O ( \widetilde { d } \sqrt { T } )$ regret bound, where $\hat { d }$ is the effective dimension of a +41 neural tangent kernel matrix which can potentially scale with $O ( T K )$ for $K$ -armed bandits. This +42 high complexity is mainly due to that the exploration is performed over the entire huge neural network +43 parameter space, which is inefficient and even infeasible when the number of neurons is large. A more +44 realistic and efficient way of learning neural contextual bandits may be to just explore different arms +45 using the last layer as the exploration parameter. More specifically, Riquelme et al. [38] provided +46 an extensive empirical study of benchmark algorithms for contextual-bandits through the lens of +47 Thompson Sampling, which suggests decoupling representation learning and uncertainty estimation +48 improves performance. +49 In this paper, we show that the decoupling of representation learning and the exploration can be +50 theoretically validated. We study a new neural contextual bandit algorithm, which learns a mapping +51 to transform the raw features associated with each context-action pair using a deep neural network +52 (deep representation), and then performs an upper confidence bound (UCB)-type exploration over the +53 linear output layer of the network (shallow exploration). We prove a sublinear regret of the proposed +54 algorithm by exploiting the UCB exploration techniques in linear contextual bandits [1] and the +55 analysis of deep overparameterized neural networks using neural tangent kernels [27]. Our theory +56 confirms the empirically observed effectiveness of decoupling the deep representation learning and +57 the UCB exploration in contextual bandits [38, 49]. + +58 Contributions we summarize the main contributions of this paper as follows. + +• We propose a contextual bandit algorithm, Neural-LinUCB, for solving a general class of contextual bandit problems without knowing the specific reward generating function. The proposed algorithm learns a deep representation to transform the raw feature vectors and performs UCB-type exploration in the last layer of the neural network, which we refer to as deep representation and shallow exploration. Compared with LinUCB [34, 16] and neural bandits such as NeuralUCB [52] and NeuralTS [51], our algorithm enjoys the best of two worlds: strong expressiveness due to the deep representation and computational efficiency due to the shallow exploration. + +66 • Despite the usage of a DNN as the feature mapping, we prove a $\widetilde { O } ( \sqrt { T } )$ regret for the proposed +67 Neural-LinUCB algorithm, which matches the regret bound of linear contextual bandits [16, 1]. +68 To the best of our knowledge, this is the first work that theoretically shows the convergence of +69 bandits algorithms under the scheme of deep representation and shallow exploration. It is notable +70 that a similar scheme called Neural-Linear was proposed by Riquelme et al. [38] for Thompson +71 sampling algorithms, and they empirically showed that decoupling representation learning and +72 uncertainty estimation improves the performance. Our work confirms this observation from a +73 theoretical perspective. + +• We conduct experiments on contextual bandit problems based on real-world datasets, demonstrating a better performance and computational efficiency of Neural-LinUCB over LinUCB and NeuralUCB, which well aligns with our theory. + +# 77 1.1 Additional related work + +78 There is a line of related work to ours on the recent advance in the optimization and generalization +79 analysis of deep neural networks. In particular, Jacot et al. [27] first introduced the neural tangent +80 kernel (NTK) to characterize the training dynamics of network outputs in the infinite width limit. +81 From the notion of NTK, a fruitful line of research emerged and showed that loss functions of +82 deep neural networks trained by (stochastic) gradient descent can converge to the global minimum +83 [22, 4, 21, 54, 53]. The generalization bounds for overparameterized deep neural networks are also +84 established in Arora et al. [6, 7], Allen-Zhu et al. [3], Cao and Gu [12, 13]. Recently, the NTK based +85 analysis is also extended to the study of sequential decision problems including bandits [52, 51], and +86 reinforcement learning algorithms [11, 36, 45, 47]. +87 Our algorithm is also different from Langford and Zhang [29], Agarwal et al. [2] which reduce the +88 bandit problem to supervised learning. Moreover, their algorithms need to access an oracle that +89 returns the optimal policy in a policy class given a sequence of context and reward vectors, whose +90 regret depends on the VC-dimension of the policy class. +91 Notation We use $[ k ]$ to denote a set $\{ 1 , \ldots , k \}$ , $k \in \mathbb { N } ^ { + }$ . $\| \mathbf { x } \| _ { 2 } = \sqrt { \mathbf { x } ^ { \top } \mathbf { x } }$ is the Euclidean norm of +92 a vector $\mathbf { x } \in \mathbb { R } ^ { d }$ . For a matrix $\mathbf { W } \in \mathbb { R } ^ { m \times n }$ , we denote by $\lVert \mathbf { W } \rVert _ { 2 }$ and $\| \mathbf { W } \| _ { F }$ its operator norm +93 and Frobenius norm respectively. For a semi-definite matrix $\mathbf { A } \in \mathbb { R } ^ { d \times d }$ and a vector $\mathbf { x } \in \mathbb { R } ^ { d }$ , we +94 denote the Mahalanobis norm as $\| \mathbf { x } \| _ { \mathbf { A } } = \sqrt { \mathbf { x } ^ { \top } \mathbf { A } \mathbf { x } }$ . Throughout this paper, we reserve the notations +95 $\{ C _ { i } \} _ { i = 0 , 1 , \ldots }$ to represent absolute positive constants that are independent of problem parameters such +96 as dimension, sample size, iteration number, step size, network length and so on. The specific values +97 of $\{ C _ { i } \} _ { i = 0 , 1 , \ldots }$ . can be different in different context. For a parameter of interest $T$ and a function +98 $f ( T )$ , we use notations such as $O ( f ( T ) )$ and $\Omega ( f ( T ) )$ to hide constant factors and ${ \widetilde { O } } ( f ( T ) )$ to hide +99 constant and logarithmic dependence of $T$ . + +# 2 Preliminaries + +In this section, we provide the background of contextual bandits and deep neural networks. + +# 2.1 Linear contextual bandits + +103 A contextual bandit is characterized by a tuple $( S , A , r )$ , where $s$ is the context (state) space, $\mathcal { A }$ is the +104 arm (action) space, and $r$ encodes the unknown reward generating function at all context-arm pairs. +105 A learning agent, who knows $s$ and $\mathcal { A }$ but does not know the true reward $r$ (values bounded in $( 0 , 1 )$ +106 for simplicity), needs to interact with the contextual bandit for $T$ rounds. At each round $t = 1 , \dots , T$ , +107 the agent first observes a context $s _ { t } \in S$ chosen by the environment; then it needs to adaptively select +108 an arm $a _ { t } \in \mathcal A$ based on its past observations; finally it receives a reward $\widehat { r } _ { t } ( \mathbf { x } _ { s , a _ { t } } ) = r ( \mathbf { x } _ { s , a _ { t } } ) + \xi _ { t }$ , +109 where $\mathbf { x } _ { s , a } \in \mathbb { R } ^ { d }$ is a known feature vector for context-arm pair $( s , a ) \in S \times A$ , and $\xi _ { t }$ is a random +110 noise with zero mean. The agent’s objective is to maximize its expected total reward over these $T$ +111 rounds, which is equivalent to minimizing the pseudo regret [8]: + +$$ +R _ { T } = \mathbb { E } \bigg [ \sum _ { t = 1 } ^ { T } \big ( \widehat { r } ( \mathbf { x } _ { s _ { t } , a _ { t } ^ { * } } ) - \widehat { r } ( \mathbf { x } _ { s _ { t } , a _ { t } } ) \big ) \bigg ] , +$$ + +where 112 $a _ { t } ^ { * } \in \operatorname { a r g m a x } _ { a \in \mathcal { A } } \{ r ( \mathbf { x } _ { s _ { t } , a } ) = \mathbb { E } [ \widehat { r } ( \mathbf { x } _ { s _ { t } , a } ) ] \}$ . To simplify the exposition, we use $\mathbf { x } _ { t , a }$ to denote 113 $\mathbf { x } _ { s _ { t } , a }$ bsince it only depends on the round index $t$ in most bandit problems, and we assume $A = [ K ]$ . + +114 In some practical problems, the agent has a prior knowledge that the reward-generating function +115 $r$ has some specific parametric form. For instance, in linear contextual bandits, the agent knows +116 that $r ( \mathbf { x } _ { s , a } ) \stackrel { * } { = } \mathbf { x } _ { s , a } ^ { \top } \pmb { \theta } ^ { * }$ for some unknown weight vector $\pmb { \theta } ^ { * } \in \mathbb { R } ^ { d }$ . One provably sample efficient +117 algorithm for linear contextual bandits is Linear Upper Confidence Bound (LinUCB) [1]. Specifically, +118 at each round $t$ , LinUCB chooses action by the following strategy + +$$ +a _ { t } = \underset { a \in [ K ] } { \operatorname { a r g m a x } } \left. \mathbf { x } _ { t , a } ^ { \top } \pmb { \theta } _ { t } + \alpha _ { t } \Vert \mathbf { x } _ { t , a } \Vert _ { \mathbf { A } _ { t } ^ { - 1 } } \right. , +$$ + +119 where θt is a point estimate of θ∗, At = λI + Pti=1 xi,ai x> i,ai with some λ > 0 is a matrix +120 defined based on the historical context-arm pairs, and $\alpha _ { t } > 0$ is a tuning parameter that controls the +121 exploration rate in LinUCB. + +# 2.2 Deep neural networks + +In this paper, we use 123 $f ( \mathbf { x } )$ to denote a neural network with input data $\mathbf { x } \in \mathbb { R } ^ { d }$ . Let $L$ be the number 124 of hidden layers and $\mathbf { W } _ { l } \in \mathbb { R } ^ { m _ { l } \times m _ { l - 1 } }$ be the weight matrices in the $l$ -th layer, where $l = 1 , \ldots , L$ 125 $m _ { 1 } = . . . = m _ { L - 1 } = m$ and $m _ { 0 } = m _ { L } = d$ . Then a $L$ -hidden layer neural network is defined as + +$$ +f ( \mathbf { x } ) = \sqrt { m } \pmb { \theta } ^ { \ast \top } \sigma _ { L } ( \mathbf { W } _ { L } \sigma _ { L - 1 } ( \mathbf { W } _ { L - 1 } \cdot \cdot \cdot \sigma _ { 1 } ( \mathbf { W } _ { 1 } \mathbf { x } ) \cdot \cdot \cdot ) ) , +$$ + +126 where $\sigma _ { l }$ is an activation function and $\pmb { \theta } ^ { * } \in \mathbb { R } ^ { d }$ is the weight of the output layer. To simplify the +127 presentation, we will assume $\sigma _ { 1 } = \sigma _ { 2 } = . . . = \sigma _ { L } = \sigma$ is the ReLU activation function, i.e., +128 ${ \bar { \sigma } } ( x ) = \operatorname* { m a x } \{ 0 , x \}$ for $x \in \mathbb { R }$ . We denote $\mathbf { w } = ( \mathrm { v e c } ( \mathbf { W } _ { 1 } ) ^ { \top } , \ldots , \mathrm { v e c } ( \mathbf { W } _ { L } ) ^ { \top } ) ^ { \top }$ , which is the +129 concatenation of the vectorized weight parameters of all hidden layers of the neural network. We also +130 write $f ( \mathbf { x } ; \pmb { \theta } ^ { * } , \mathbf { w } ) = f ( \mathbf { x } )$ in order to explicitly specify the weight parameters of neural network $f$ . It +131 is easy to show that the dimension $p$ of vector w satisfies $p = ( L - 2 ) m ^ { 2 } + 2 m d .$ . To simplify the +132 notation, we define $\phi ( \mathbf { x } ; \mathbf { w } )$ as the output of the $L$ -th hidden layer of neural network $f$ . + +$$ +\boldsymbol { \phi } ( \mathbf { x } ; \mathbf { w } ) = \sqrt { m } \sigma ( \mathbf { W } _ { L } \sigma ( \mathbf { W } _ { L - 1 } \cdot \cdot \cdot \sigma ( \mathbf { W } _ { 1 } \mathbf { x } ) \cdot \cdot \cdot \cdot ) ) . +$$ + +33 Note that $\phi ( \mathbf { x } ; \mathbf { w } )$ itself can also be viewed as a neural network with vector-valued outputs. + +# 134 3 Deep Representation and Shallow Exploration + +135 The linear parametric form in linear contextual bandits might produce biased estimates of the reward +136 due to the lack of representation power [42, 38]. In contrast, it is well known that deep neural networks +137 are powerful enough to approximate an arbitrary function [18]. Therefore, a natural extension of +138 linear contextual bandits is to use a deep neural network to approximate the reward generating +139 function $r ( \cdot )$ . Nonetheless, DNNs usually have a prohibitively large dimension for weight parameters, +140 which makes the exploration in neural networks based UCB algorithm inefficient [28, 52]. +141 In this work, we study a neural contextual bandit algorithm, where the hidden layers of a deep neural +142 network are used to represent the features and the exploration is only performed in the last layer of the +143 neural network. In particular, we assume that the reward generating function $r ( \cdot )$ can be expressed as +144 the inner product between a deep represented feature vector and an exploration weight parameter, +145 namely, $\bar { r ( \cdot ) } = \langle \theta ^ { * } , \psi ( \cdot ) \rangle$ , where $\pmb { \theta } ^ { * } \in \mathbb { R } ^ { d }$ is some weight parameter and $\psi ( \cdot )$ is an unknown feature +146 mapping. This decoupling of the representation and the exploration will achieve the best of both +147 worlds: efficient exploration in shallow (linear) models and high expressive power of deep models. +148 To learn the unknown feature mapping, we propose to use a neural network to approximate it. In +149 what follows, we will describe a neural contextual bandit algorithm that uses the output of the last +150 hidden layer of a neural network to transform the raw feature vectors (deep representation) and +151 performs UCB-type exploration in the last layer of the neural network (shallow exploration). Since +152 the exploration is performed only in the last linear layer, we call this procedure Neural-LinUCB, +153 which is displayed in Algorithm 1. +154 Specifically, in round $t$ , the agent receives an action set with raw features $\mathcal { X } _ { t } = \{ \mathbf { x } _ { t , 1 } , . . . , \mathbf { x } _ { t , K } \}$ . +155 Then the agent chooses an arm $a _ { t }$ that maximizes the following upper confidence bound: + +$$ +a _ { t } = \underset { k \in [ K ] } { \operatorname { a r g m a x } } \Big \{ \langle \phi ( \mathbf { x } _ { t , k } ; \mathbf { w } _ { t - 1 } ) , \theta _ { t - 1 } \rangle + \alpha _ { t } \| \phi ( \mathbf { x } _ { t , k } ; \mathbf { w } _ { t - 1 } ) \| _ { \mathbf { A } _ { t - 1 } ^ { - 1 } } \Big \} , +$$ + +156 where $\pmb { \theta } _ { t - 1 }$ is a point estimate of the unknown weight in the last layer, $\phi ( \mathbf { x } ; \mathbf { w } )$ is defined as in (2.3), +157 $\mathbf { w } _ { t - 1 }$ is an estimate of all the weight parameters in the hidden layers of the neural network, $\alpha _ { t } > 0$ is +158 the algorithmic parameter controlling the exploration, and ${ \bf A } _ { t }$ is a matrix defined based on historical +159 transformed features: + +$$ +\mathbf { A } _ { t } = \lambda \mathbf { I } + \sum _ { i = 1 } ^ { t } \phi ( \mathbf { x } _ { i , a _ { i } } ; \mathbf { w } _ { i - 1 } ) \phi ( \mathbf { x } _ { i , a _ { i } } ; \mathbf { w } _ { i - 1 } ) ^ { \top } , +$$ + +and 160 $\lambda > 0$ . After pulling arm $a _ { t }$ , the agent will observe a noisy reward $\widehat { r } _ { t } : = \widehat { r } ( \mathbf { x } _ { t , a _ { t } } )$ defined as + +$$ +\widehat { r } ( \mathbf { x } _ { t , k } ) = r ( \mathbf { x } _ { t , k } ) + \xi _ { t } , +$$ + +161 where $\xi _ { t }$ is an independent $\nu$ -subGaussian random noise for some $\nu > 0$ and $r ( \cdot )$ is an unknown +162 reward function. In this paper, we will interchangeably use notation $\widehat { r _ { t } }$ to denote the reward received +163 at the $t$ -th step and an equivalent notation $\widehat { r } ( \mathbf { x } )$ b to express its dependence on the feature vector $\mathbf { x }$ . + +164 Upon receiving the reward $\widehat { r _ { t } }$ , the agent updates its estimate $\theta _ { t }$ of the output layer weight by using the same 165 $\ell ^ { 2 }$ b-regularized least-squares estimate in linear contextual bandits [1]. In particular, we have + +$$ +\pmb { \theta } _ { t } = \mathbf { A } _ { t } ^ { - 1 } \mathbf { b } _ { t } , +$$ + +where 166 $\begin{array} { r } { \mathbf { b } _ { t } = \sum _ { i = 1 } ^ { t } \widehat { r } _ { i } \phi ( \mathbf { x } _ { i , a _ { i } } ; \mathbf { w } _ { i - 1 } ) } \end{array}$ . + +167 To save the computation, the neural network $\phi ( \cdot ; { \mathbf w } _ { t } )$ will be updated once every $H$ steps. Therefore, +168 we have $\mathbf { w } _ { ( q - 1 ) H + 1 } = . . . = \mathbf { w } _ { q H }$ for $q = 1 , 2 , \ldots$ We call the time steps $\{ ( q - 1 ) H + 1 , \ldots , q H \}$ +169 an epoch with length $H$ . At time step $t = H q$ , for any $q = 1 , 2 , \ldots$ , Algorithm 1 will retrain the +170 neural network based on all the historical data. In Algorithm 2, our goal is to minimize the following +171 empirical loss function: + +$$ +\mathcal { L } _ { q } ( \mathbf { w } ) = \sum _ { i = 1 } ^ { q H } \big ( \pmb { \theta } _ { i } ^ { \top } \phi ( \mathbf { x } _ { i , a _ { i } } ; \mathbf { w } ) - \widehat { r } _ { i } \big ) ^ { 2 } . +$$ + +In practi172 i=1e, one can further save computational cost by onl eding data $\{ \mathbf { x } _ { i , a _ { i } } , \widehat { r } _ { i } , \pmb { \theta } _ { i } \} _ { i = ( q - 1 ) H + 1 } ^ { q H }$ $q$ $\mathbf { w } _ { t }$ b, which does not hurt the performance 174 since the historical information has been encoded into the estimate of $\theta _ { i }$ . In this paper, we will 175 perform the following gradient descent step + +$$ +\mathbf { w } _ { q } ^ { ( s ) } = \mathbf { w } _ { q } ^ { ( s - 1 ) } - \eta _ { q } \nabla _ { \mathbf { w } } \mathcal { L } _ { q } \big ( \mathbf { w } ^ { ( s - 1 ) } \big ) . +$$ + +176 for $s = 1 , \ldots , n$ , where $\mathbf { w } _ { q } ^ { ( 0 ) } = \mathbf { w } ^ { ( 0 ) }$ is chosen as the same random initialization point. We will +177 discuss more about the initial point $\mathbf { w } ^ { ( 0 ) }$ in the next paragraph. Then Algorithm 2 outputs $\mathbf { w } _ { q } ^ { ( n ) }$ and +178 we set it as the updated weight parameter $\mathbf { w } _ { H q + 1 }$ in Algorithm 1. In the next round, the agent will +179 receive another action set $\mathcal { X } _ { t + 1 }$ with raw feature vectors and repeat the above steps to choose the +180 sub-optimal arm and update estimation for contextual parameters. +181 Initialization: Recall that w is the collection of all hidden layer weight parameters of the neural +182 network. We will follow the same initialization scheme as used in Zhou et al. [52], where each entry +183 of the weight matrices follows some Gaussian distribution. Specifically, for any $l \in \{ 1 , \ldots , L - 1 \}$ , +184 we set $\mathbf { W } _ { l } = \left[ \begin{array} { c c } { \mathbf { W } } & { \mathbf { 0 } } \\ { \mathbf { 0 } } & { \mathbf { W } } \end{array} \right]$ , where each entry of $\mathbf { W }$ follows distribution $N ( 0 , 4 / m )$ independently; for +185 $\mathbf { W } _ { L }$ , we set it as $\begin{array} { r l } { [ \mathbf { V } } & { { } - \mathbf { V } ] } \end{array}$ , where each entry of $\mathbf { V }$ follows distribution $N ( 0 , 2 / m )$ independently. + +Comparison with LinUCB and NeuralUCB: Compared with linear contextual bandits in Section 2.1, Algorithm 1 has a distinct feature that it learns a deep neural network to obtain a deep representation of the raw data vectors and then performs UCB exploration. This deep representation allows our algorithm to characterize more intrinsic and latent information about the raw data $\left\{ \mathbf { x } _ { t , k } \right\} _ { t \in [ T ] , k \in [ K ] } \subset \mathbb { R } ^ { d }$ . However, the increased complexity of the feature mapping $\phi ( \cdot ; { \mathbf { w } } )$ also introduces great hardness in training. For instance, a recent work by Zhou et al. [52] also studied the neural contextual bandit problem, but different from (3.1), their algorithm (NeuralUCB) performs the UCB exploration on the entire network parameter space, which is $\dot { \mathbb { R } } ^ { \widetilde { p } + d }$ , where $\ddot { \tilde { p } } = m + m d + ( L \dot { \bar { \mathbf { \alpha } } } 1 ) m ^ { 2 }$ . Note that in Zhou et al. [52], they need to compute the inverse eof a matrix $\mathbf { Z } _ { t } \in \mathbb { R } ^ { ( \widetilde { p } + d ) \times ( \widetilde { p } + d ) }$ , which is defined in a similar way to the matrix ${ \bf A } _ { t }$ in our paper except that $\mathbf { Z } _ { t }$ is defined based on the gradient of the network instead of the output of the last hidden layer as in (3.2). In sharp contrast, ${ \bf A } _ { t }$ in our paper is only of size $d \times d$ and thus is much more efficient and practical in implementation, which will be seen from our experiments in later sections. + +199 We note that there is also a similar algorithm to our Neural-LinUCB presented in Deshmukh et al. +200 [20], where they studied the self-supervised learning loss in contextual bandits with neural network +201 representation for computer vision problems. However, no regret analysis has been provided. When +202 the feature mapping $\phi ( \cdot ; { \mathbf { w } } )$ is an identity function, the problem reduces to linear contextual bandits +203 where we directly use $\mathbf { x } _ { t }$ as the feature vector. In this case, it is easy to see that Algorithm 1 reduces +204 to LinUCB [16] since we do not need to learn the representation parameter w anymore. +205 Comparison with Neural-Linear: The high-level idea of decoupling the representation and explo +206 ration in our algorithm is also similar to that of the Neural-Linear algorithm [38, 49], which trains a +207 deep neural network to learn a representation of the raw feature vectors, and then uses a Bayesian +208 linear regression to estimate the uncertainty in the bandit problem. However, these two algorithms +209 are significantly different since Neural-Linear [38] is a Thompson sampling based algorithm that +210 uses posterior sampling to estimate the weight parameter $\pmb { \theta } ^ { * }$ via Bayesian linear regression, whereas +211 Neural-LinUCB adopts upper confidence bound based techniques to estimate the weight $\pmb { \theta } ^ { * }$ . Never +212 theless, both algorithms share the same idea of deep representation and shallow exploration, and we +213 view our Neural-LinUCB algorithm as one instantiation of the Neural-Linear scheme. + +# 214 4 Main Results + +215 To analyze the regret bound of Algorithm 1, we first lay down some important assumptions on the +216 neural contextual bandit model. + +# Algorithm 1 Deep Representation and Shallow Exploration (Neural-LinUCB) + +1: Input: regularization parameter $\lambda > 0$ , number of total steps $T$ , episode length $H$ , exploration parameters $\{ \alpha _ { t } > 0 \} _ { t \in [ T ] }$ +2: Initialization: $\mathbf { A } _ { 0 } = \lambda \mathbf { I }$ , $\mathbf { b } _ { 0 } = \mathbf { 0 }$ ; entries of $\pmb { \theta } _ { 0 }$ follow $N ( 0 , 1 / d )$ , and $\mathbf { w } ^ { ( 0 ) }$ is initialized as described in Section 3; $q = 1$ ; $\mathbf { w } _ { 0 } = \mathbf { w } ^ { ( 0 ) }$ +3: for $t = 1 , \dots , T$ do +4: receive feature vectors $\left\{ \mathbf { x } _ { t , 1 } , \ldots , \mathbf { x } _ { t , K } \right\}$ +5: choose arm $\begin{array} { r } { a _ { t } \ = \ \mathrm { a r g m a x } _ { k \in [ K ] } \pmb { \theta } _ { t - 1 } ^ { \top } \phi ( \mathbf { x } _ { t , k } ; \mathbf { w } _ { t - 1 } ) \ + \alpha _ { t } \| \phi ( \mathbf { x } _ { t , k } ; \mathbf { w } _ { t - 1 } ) \| _ { \mathbf { A } _ { t - 1 } ^ { - 1 } } } \end{array}$ , and obtain reward $\widehat { r } _ { t }$ +6: update ${ \bf A } _ { t }$ and $\mathbf { b } _ { t }$ as follows: $\begin{array} { r l } & { \mathbf { \Phi } ^ { \mathbf { A } } t = \tilde { \mathbf { A } } _ { t - 1 } + \tilde { \phi } ( \mathbf { x } _ { t , a _ { t } } ; \mathbf { w } _ { t - 1 } ) \phi ( \mathbf { x } _ { t , a _ { t } } ; \mathbf { w } _ { t - 1 } ) ^ { \top } } \\ & { \mathbf { b } _ { t } = \mathbf { b } _ { t - 1 } + \hat { r } _ { t } \phi ( \mathbf { x } _ { t , a _ { t } } ; \mathbf { w } _ { t - 1 } ) , } \end{array}$ , +7: update $\pmb { \theta } _ { t } = \mathbf { A } _ { t } ^ { - 1 } \mathbf { b } _ { t }$ +8: if $\mathrm { n o d } ( t , H ) = 0$ then +9: $\mathbf { w } _ { t } \gets$ output of Algorithm 2 +10: $q = q + 1$ +11: else +12: $\mathbf { w } _ { t } = \mathbf { w } _ { t - 1 }$ +13: end if +14: end for +15: Output $\mathbf { w } _ { T }$ + +# Algorithm 2 Update Weight Parameters with Gradient Descent + +1: Input: initial point $\mathbf { w } _ { q } ^ { ( 0 ) } = \mathbf { w } ^ { ( 0 ) }$ , maximum iteration number $n$ , step size $\eta _ { q }$ , and loss function +defined in (3.5). +2: for $s = 1 , \ldots , n$ do +3: $\begin{array} { r } { \mathbf { w } _ { q } ^ { ( s ) } = \mathbf { w } _ { q } ^ { ( s - 1 ) } - \eta _ { q } \nabla _ { \mathbf { w } } \mathcal { L } _ { q } ( \mathbf { w } _ { q } ^ { ( s - 1 ) } ) . } \end{array}$ +4: end for +5: Output w(n)q +7 Assumption 4.1. For all $i \geq 1$ and $k \in [ K ]$ , we assume that $\| \mathbf { x } _ { i , k } \| _ { 2 } = 1$ and its entries satisfy +8 $[ { \bf { x } } _ { i , k } ] _ { j } \stackrel { - } { = } [ { \bf { x } } _ { j , k } ] _ { j + d / 2 }$ . + +19 The assumption that $\| \mathbf { x } _ { i , k } \| _ { 2 } = 1$ is not essential and is only imposed for simplicity, which is also 20 used in Zou and $\mathrm { G u }$ [53], Zhou et al. [52]. Finally, the condition on the entries of $\mathbf { x } _ { i , k }$ is also mild since otherwise we could always construct 21 $\mathbf { x } _ { i , k } ^ { \prime } = [ \mathbf { x } _ { i , k } ^ { \top } , \mathbf { x } _ { i , k } ^ { \top } ] ^ { \top } / \sqrt { 2 }$ to replace it. An implication of + +Assumption 4.1 is that the initialization scheme in Algorithm 1 results in $\phi ( \mathbf { x } _ { i , k } ; \mathbf { w } ^ { ( 0 ) } ) = \mathbf { 0 }$ for all $i \in [ T ]$ and $k \in [ K ]$ . + +24 We assume the following stability condition on the spectral norm of the neural network gradient: + +225 Assumption 4.2. There is a constant $\ell _ { \mathrm { L i p } } > 0$ such that it holds + +$$ +\left\| \frac { \partial \phi } { \partial \mathbf { w } } ( \mathbf { x } ; \mathbf { w } _ { 0 } ) - \frac { \partial \phi } { \partial \mathbf { w } } ( \mathbf { x } ^ { \prime } ; \mathbf { w } _ { 0 } ) \right\| _ { 2 } \leq \ell _ { \mathrm { L i p } } \| \mathbf { x } - \mathbf { x } ^ { \prime } \| _ { 2 } , +$$ + +for all 226 $\mathbf { x } , \mathbf { x } ^ { \prime } \in \{ \mathbf { x } _ { i , k } \} _ { i \in [ T ] , k \in [ K ] }$ + +227 The inequality in Assumption 4.2 resembles the Lipschitz condition on the gradient of the neural +228 network. However, it is essentially different from the smoothness condition since here the gradient +229 is taken with respect to the neural network weights while the Lipschitz condition is imposed on the +230 feature parameter x. Similar conditions are widely made in nonconvex optimization [46, 10, 48], in +231 the name of first-order stability, which is essential to derive the convergence of alternating optimization +232 algorithms. Furthermore, Assumption 4.2 is only required on the $T K$ training data points and a +233 specific weight parameter $\mathbf { w } _ { 0 }$ . Therefore, the condition will hold if the raw feature data lie in a +234 certain subspace of $\mathbb { R } ^ { d }$ . We provided some further discussions in the supplementary material about +235 this assumption for interested readers. +236 In order to analyze the regret bound of Algorithm 1, we need to characterize the properties of the +237 deep neural network in (2.2) that is used to represent the feature vectors. Following a recent line of +238 research [27, 12, 7, 52], we define the covariance between two data point $\mathbf { x } , \mathbf { y } \in \mathbb { R } ^ { \bar { d } }$ as follows. + +$$ +\begin{array} { r l } & { \widetilde { \pmb { \Sigma } } ^ { ( 0 ) } ( \mathbf x , \mathbf y ) = \pmb { \Sigma } ^ { ( 0 ) } ( \mathbf x , \mathbf y ) = \mathbf x ^ { \top } \mathbf y , } \\ & { \pmb { \Lambda } ^ { ( l ) } ( \mathbf x , \mathbf y ) = \left[ \pmb { \Sigma } ^ { l - 1 } ( \mathbf x , \mathbf x ) \quad \pmb { \Sigma } ^ { l - 1 } ( \mathbf x , \mathbf y ) \right] , } \\ & { \pmb { \Sigma } ^ { ( l ) } ( \mathbf x , \mathbf y ) = 2 \mathbb { E } _ { ( u , v ) \sim N ( \mathbf 0 , \mathbf { A } ^ { ( l - 1 ) } ( \mathbf x , \mathbf y ) ) } [ \sigma ( u ) \sigma ( v ) ] , } \\ & { \widetilde { \pmb { \Sigma } } ^ { ( l ) } ( \mathbf x , \mathbf y ) = 2 \widetilde { \pmb { \Sigma } } ^ { ( l - 1 ) } ( \mathbf x , \mathbf y ) \mathbb { E } _ { u , v } [ \dot { \sigma } ( u ) \dot { \sigma } ( v ) ] + \pmb { \Sigma } ^ { ( l ) } ( \mathbf x , \mathbf y ) , } \end{array} +$$ + +where 239 the ne240 $( u , v ) \sim N ( \mathbf { 0 } , \mathbf { \Lambda } \Lambda ^ { ( l - 1 ) } ( \mathbf { x } , \mathbf { y } ) )$ , arix $\dot { \sigma } ( \cdot )$ rivative of activation functiobased on all feature vectors $\sigma ( \cdot )$ $\mathbf { H } \in \mathbb { R } ^ { T K \times T K }$ $\{ \mathbf { x } _ { t , k } \} _ { t \in [ T ] , k \in [ K ] }$ 241 Renumbering $\{ \mathbf { x } _ { t , k } \} _ { t \in [ T ] , k \in [ K ] }$ as $\{ \mathbf { x } _ { i } \} _ { i = 1 , \dots , T K }$ , then each entry $\mathbf { H } _ { i j }$ is defined as + +$$ +\mathbf { H } _ { i j } = \frac { 1 } { 2 } \big ( \widetilde { \boldsymbol { \Sigma } } ^ { ( L ) } ( \mathbf { x } _ { i } , \mathbf { x } _ { j } ) + \boldsymbol { \Sigma } ^ { ( L ) } ( \mathbf { x } _ { i } , \mathbf { x } _ { j } ) \big ) , +$$ + +42 for all $i , j \in [ T K ]$ . Based on the above definition, we impose the following assumption on $\mathbf { H }$ + +43 Assumption 4.3. The neural tangent kernel defined in (4.2) is positive definite, i.e., $\lambda _ { \operatorname* { m i n } } ( \mathbf { H } ) \geq \lambda _ { 0 }$ +244 for some constant $\lambda _ { 0 } > 0$ . +245 Assumption 4.3 essentially requires the neural tangent kernel matrix $\mathbf { H }$ to be non-singular, which is +246 a mild condition and also imposed in other related work [21, 7, 12, 52]. Moreover, it is shown that +247 Assumption 4.3 can be easily derived from Assumption 4.1 for two-layer ReLU networks [37, 53]. +248 Therefore, Assumption 4.3 is mild or even negligible given the non-degeneration assumption on the +249 feature vectors. Also note that matrix $\mathbf { H }$ is only defined based on layers $l = 1 , \ldots , L$ of the neural +250 network, and does not depend on the output layer $\pmb \theta$ . It is easy to extend the definition of $\mathbf { H }$ to the +251 NTK matrix defined on all layers including the output layer $\pmb \theta$ , which would also be positive definite +252 by Assumption 4.3 and the recursion in (4.2). +253 Before we present the regret analysis of the neural contextual bandit, we need to modify the regret +254 defined in (2.1) to account for the randomness of the neural network initialization. For a fixed time +255 horizon $T$ , we define the regret of Algorithm 1 as follows. + +$$ +R _ { T } = \mathbb { E } \bigg [ \sum _ { t = 1 } ^ { T } \big ( \widehat { r } ( \mathbf { x } _ { t , a _ { t } ^ { * } } ) - \widehat { r } ( \mathbf { x } _ { t , a _ { t } } ) \big ) \big | \mathbf { w } ^ { ( 0 ) } \bigg ] , +$$ + +256 where the expectation is taken over the randomness of the reward noise. Note that $R _ { T }$ defined in (4.3) +257 is still a random variable since the initialization of Algorithm 2 is randomly generated. + +258 Now we are going to present the regret bound of the proposed algorithm. + +59 Theorem 4.4. Suppose Assumptions 4.1, 4.2 and 4.3 hold. Assume that $\lVert \pmb { \theta } ^ { * } \rVert _ { 2 } \leq M$ for some +0 positive constant $M > 0$ . For any $\delta \in ( 0 , 1 )$ , let us choose $\alpha _ { t }$ in Neural-LinUCB as + +$$ +\alpha _ { t } = \nu \sqrt { 2 \big ( d \log ( 1 + t \log ( H K ) / \lambda ) + \log ( 1 / \delta ) \big ) } + \lambda ^ { 1 / 2 } M . +$$ + +261 We choose the step size $\eta _ { q }$ of Algorithm 2 as + +$$ +\eta _ { q } \leq C _ { 0 } \big ( d ^ { 2 } m n T ^ { 5 . 5 } L ^ { 6 } \log ( T K / \delta ) \big ) ^ { - 1 } , +$$ + +262 and the width of the neural network satisfies $m = \mathrm { p o l y } ( L , d , 1 / \delta , H , \log ( T K / \delta ) )$ . With probability +263 at least $1 - \delta$ over the randomness of the initialization of the neural network, it holds that + +$$ +R _ { T } \leq C _ { 1 } \alpha _ { T } \sqrt { T d \log \left( 1 + \frac { T G ^ { 2 } } { \lambda d } \right) } + \frac { C _ { 2 } \ell _ { \mathrm { L i p } } L ^ { 3 } d ^ { 5 / 2 } T \sqrt { \log m \log ( \frac { 1 } { \delta } ) \log ( \frac { T K } { \delta } ) } \| \mathbf { r } - \widetilde { \mathbf { r } } \| _ { \mathbf { H } ^ { - 1 } } } { m ^ { 1 / 6 } } , +$$ + +264 where $\{ C _ { i } \} _ { i = 0 , 1 , 2 }$ are absolute constants independent of the problem parameters, $\begin{array} { r l } { \mathbf { r } } & { { } = } \end{array}$ +265 $( r ( \mathbf { x } _ { 1 } ) , r ( \mathbf { x } _ { 2 } ) , \ldots , r ( \mathbf { x } _ { T K } ) ) ^ { \top } \ \in \ \mathbb { R } ^ { T K }$ and $\widetilde { \textbf { r } } = ( f ( \mathbf { x } _ { 1 } ; \pmb { \theta } _ { 0 } , \mathbf { w } _ { 0 } ) , \dots , f ( \mathbf { x } _ { T K } ; \pmb { \theta } _ { T - 1 } , \mathbf { w } _ { T - 1 } ) ) ^ { \top } \ \in$ +266 $\mathbb { R } ^ { T K }$ , and $\| \mathbf { r } \| _ { \mathbf { A } } = \sqrt { \mathbf { r } ^ { \top } \mathbf { A } \mathbf { r } }$ . +267 Remark 4.5. Theorem 4.4 shows that the regret of Algorithm 1 can be bounded by two parts: the +268 first part is of order $\widetilde { O } ( \sqrt { T } )$ , which resembles the regret bound of linear contextual bandits [1]; the +269 second part is of order $\widetilde { O } ( m ^ { - 1 / 6 } T \sqrt { ( \mathbf { r } - \widetilde { \mathbf { r } } ) ^ { \top } \mathbf { H } ^ { - 1 } ( \mathbf { r } - \widetilde { \mathbf { r } } ) } )$ , which depends on the estimation error +270 of the neural network $f$ e efor the reward generating function $r$ and the neural tangent kernel $\mathbf { H }$ . + +It is worth noting that our theoretical analysis depends on the reward structure assumption that $r ( \cdot ) = \langle \theta \ast , \psi ( \cdot ) \rangle$ . However, the linear structure between $\pmb { \theta } \ast$ and $\psi ( \cdot )$ is not essential. As long as the deep representation of the feature vector and the uncertainty weight parameter can be decoupled, Algorithm 1 can be easily extended to settings with milder assumptions on the reward structure such as generalized linear models [41, 24, 35, 28]. For more general bandit models where no assumption is imposed to the reward generating function, it is still unclear whether the decoupled deep representation and shallow exploration would work especially in cases a thorough exploration may be needed. + +Based on the result in Theorem 4.4, we can easily verify the following conclusion: + +Corollary 4.6. Under the same conditions of Theorem 4.4, if we choose a sufficiently overparameterized neural network mapping $\phi ( \cdot )$ such that $m \geq T ^ { 3 }$ , then the regret of Algorithm 1 is $R _ { T } = { \widetilde { O } } ( { \sqrt { T } } { \sqrt { ( \mathbf { r } - { \widetilde { \mathbf { r } } } ) ^ { \top } \mathbf { H } ^ { - 1 } ( \mathbf { r } - { \widetilde { \mathbf { r } } } ) } } )$ . + +Remark 4.7. For the ease of presentation, let us denote $\mathcal { E } : = \| \mathbf { r } - \widetilde { \mathbf { r } } \| _ { \mathbf { H } ^ { - 1 } }$ . If we have $\mathcal { E } = O ( 1 )$ , the total regret in Theorem 4.4 becomes $\widetilde { O } ( \sqrt { T } )$ which matches the regret of linear contextual bandits [1]. We remark that there is a similar assumption in [52] where they assume that $\mathbf { r } ^ { \top } \mathbf { H } ^ { - 1 } \mathbf { r }$ can be upper bounded by a constant. They show that this term can be bounded by the RKHS norm of $\mathbf { r }$ if it belongs to the RKHS induced by the neural tangent kernel [6, 7, 33]. In addition, $\mathcal { E }$ here is the difference between the true reward function and the neural network function, which can also be small if the deep neural network function well approximates the reward generating function $r ( \cdot )$ . + +# 290 5 Experiments + +291 +292 +293 +294 +295 +296 +297 +298 +299 +300 +301 +302 +303 +304 +305 + +In this section, we provide empirical evaluations of Neural-LinUCB on real-world datasets. As we have discussed in Section 3, Neural-LinUCB could be viewed as an instantiation of the NeuralLinear scheme studied in Riquelme et al. [38] except that we use the UCB exploration instead of the posterior sampling exploration therein. Note that there has been an extensive comparison [38] of the Neural-Linear methods with many other baselines such as greedy algorithms, Variational Inference, Expectation-Propagation, Bayesian Non-parametrics and so on. Therefore, we do not seek a thorough empirical comparison of Neural-LinUCB with all existing bandits algorithms. We refer readers who are interested in the performance of Neural-Linear methods with deep representation and shallow exploration compared with a vast of baselines in the literature to the benchmark study by Riquelme et al. [38]. In this experiment, we only aim to show the advantages of our algorithm over the following baselines: (1) Neural-Linear [38]; (2) LinUCB [16], which does not have a deep representation of the feature vectors; and (3) NeuralUCB [52], which performs UCB exploration on all the parameters of the neural network instead of the shallow exploration used in our paper. All numerical experiments were run on a workstation with Intel(R) Xeon(R) CPU E5-2637 v4 $@$ 3.50GHz. + +Datasets: we evaluate the performances of all algorithms on bandit problems created from real-world data. Specifically, following the experimental setting in Zhou et al. [52],we use datasets (Shuttle) Statlog, Magic and Covertype from UCI machine learning repository [23], and the MINST dataset from LeCun et al. [31]. The details of these datasets are presented in Table 1. In Table 1, each instance represents a feature vector $\mathbf { x } \in \mathbb { R } ^ { d }$ that is associated with one of the $K$ arms, and dimension $d$ is the number of attributes in each instance. + +Table 1: Specifications of datasets from the UCI machine learning repository used in this paper. + +
StatlogMagicCovertypeMNIST
Number of attributes91154784
Number of arms72710
Number of instances58,00019,020581,01260,000
+ +![](images/943bd96058f4439657901f76b8c00684d91994e2afa724f14d33961644a23aaf.jpg) +Figure 1: The cumulative regrets of LinUCB, NeuralUCB, Neural-Linear and Neural-LinUCB over 15, 000 rounds. Experiments are averaged over 10 repetitions. + +311 Implementations: for LinUCB, we follow the setting in Li et al. [34] to use disjoint models +312 for different arms. For neural network based algorithms such as NeuralUCB, Neural-Linear and +313 Neural-LinUCB, we use a ReLU neural network defined as in (2.2) with $L = 2$ and 2000 for the +314 UCI datasets (Statlog, Magic, Covertype). Thus the neural network weights are $\mathbf { W } _ { 1 } \in \mathbb { R } ^ { m \times d }$ +315 $\mathbf { W } _ { 2 } \in \mathbb { R } ^ { k \times m }$ , and $\pmb { \theta } \in \mathbb { R } ^ { \widetilde { k } }$ respectively, where $k = 1 0 0$ , $m = 2 0 0 0$ , and $d$ is the dimension of +316 features in the corresponding task. Since the problem size of the MNIST dataset is larger, inspired +317 by Hinton and Salakhutdinov [26], we use a deeper NN and set $L = 3$ , $k = 1 0 0$ and $m = 1 0 0$ , +318 with weights $\mathbf { W } _ { 1 } \in \mathbb { R } ^ { m \times d }$ , $\mathbf { W } _ { 2 } \in \mathbb { R } ^ { m \times m }$ , $\mathbf { W _ { 3 } } \in \mathbb { R } ^ { k \times m }$ , and $\pmb \theta \in \mathbb { R } ^ { k }$ . We set the time horizon +319 $T = 1 5 , 0 0 0$ , which is the total number of rounds for each algorithm on each dataset. We use +320 gradient decent to optimize the network weights, with a step size $\eta _ { q } = 1 \mathrm { e } { - 5 }$ and maximum iteration +321 number $n = 1 , 0 0 0$ . To speed up the training process, the network parameter w is updated every +322 $H = 1 0 0$ rounds starting from round 2000. We also apply early stopping when the loss difference +323 of two consecutive iterations is smaller than a threshold of 1e-6. We set $\lambda = 1$ and $\alpha _ { t } = 0 . 0 2$ +324 for all algorithms, $t \in [ T ]$ . Following the setting in Riquelme et al. [38], we use round-robin to +325 independently select each arm for 3 times at the beginning of each algorithm. For NeuralUCB, since +326 it is computationally unaffordable to perform the original UCB exploration as displayed in Zhou et al. +327 [52], we follow their experimental setting to replace the matrix $\mathbf { Z } _ { t } \in \mathbb { R } ^ { ( d + \widetilde { p } ) \times \widetilde { ( } d + \widetilde { p } ) }$ in Zhou et al. +328 [52] with its diagonal matrix. + +Results: we plot the cumulative regret of all algorithms versus round in Figures 1(a), 1(b) and 1(c) for UCI datasets and in Figure 1(d) for MNIST. The results are reported based on the average of 10 repetitions over different random shuffles of the datasets. It can be seen that algorithms based on neural network representations (NeuralUCB, Neural-Linear and Neural-LinUCB) consistently outperform the linear contextual bandit method LinUCB, which shows that linear models may lack representation power and find biased estimates for the underlying reward generating function. Furthermore, our proposed Neural-LinUCB achieves a comparable regret with NeuralUCB in all experiments despite the fact that our algorithm only explores in the output layer of the neural network, which is more computationally efficient as we will show in the sequel.The results in our experiment are well aligned with our theory that deep representation and shallow exploration are sufficient to guarantee a good performance of neural contextual bandit algorithms, which is also consistent with the findings in existing literature [38] that decoupling the representation learning and uncertainty estimation improves the performance. + +We also conducted experiments to study the effects of different widths of deep neural networks on the regret performance and to show the computational efficiency of Neural-LinUCB compared with existing neural bandit algorithms. Due to the space limit, we defer the results to Appendix A. + +# 6 Conclusions + +In this paper, we propose a new neural contextual bandit algorithm called Neural-LinUCB, which uses the hidden layers of a ReLU neural network as a deep representation of the raw feature vectors and performs UCB type exploration on the last layer of the neural network. By incorporating techniques in liner contextual bandits and neural tangent kernels, we prove that the proposed algorithm achieves a sublinear regret when the width of the network is sufficiently large. This is the first regret analysis of neural contextual bandit algorithms with deep representation and shallow exploration, which have been observed in practice to work well on many benchmark bandit problems [38]. We also conducted experiments on real-world datasets to demonstrate the advantage of the proposed algorithm over LinUCB and existing neural contextual bandit algorithms. + +355 References [1] Yasin Abbasi-Yadkori, Dávid Pál, and Csaba Szepesvári. Improved algorithms for linear stochastic bandits. In Advances in Neural Information Processing Systems, pages 2312–2320, 2011. [2] Alekh Agarwal, Daniel Hsu, Satyen Kale, John Langford, Lihong Li, and Robert Schapire. Taming the monster: A fast and simple algorithm for contextual bandits. In International Conference on Machine Learning, pages 1638–1646, 2014. [3] Zeyuan Allen-Zhu, Yuanzhi Li, and Yingyu Liang. Learning and generalization in overparameterized neural networks, going beyond two layers. In Advances in neural information processing systems, pages 6155–6166, 2019. [4] Zeyuan Allen-Zhu, Yuanzhi Li, and Zhao Song. A convergence theory for deep learning via over-parameterization. In International Conference on Machine Learning, pages 242–252, 2019. [5] Robin Allesiardo, Raphaël Féraud, and Djallel Bouneffouf. A neural networks committee for the contextual bandit problem. In International Conference on Neural Information Processing, pages 374–381. Springer, 2014. [6] Sanjeev Arora, Simon Du, Wei Hu, Zhiyuan Li, and Ruosong Wang. Fine-grained analysis of optimization and generalization for overparameterized two-layer neural networks. In International Conference on Machine Learning, pages 322–332, 2019. [7] Sanjeev Arora, Simon S Du, Wei Hu, Zhiyuan Li, Russ R Salakhutdinov, and Ruosong Wang. On exact computation with an infinitely wide neural net. In Advances in Neural Information Processing Systems, pages 8139–8148, 2019. [8] Jean-Yves Audibert, Rémi Munos, and Csaba Szepesvári. Exploration–exploitation tradeoff using variance estimates in multi-armed bandits. Theoretical Computer Science, 410(19): 1876–1902, 2009. [9] Peter Auer, Nicolo Cesa-Bianchi, and Paul Fischer. Finite-time analysis of the multiarmed bandit problem. Machine learning, 47(2-3):235–256, 2002. +382 [10] Sivaraman Balakrishnan, Martin J Wainwright, Bin Yu, et al. Statistical guarantees for the em algorithm: From population to sample-based analysis. The Annals of Statistics, 45(1):77–120, 2017. [11] Qi Cai, Zhuoran Yang, Jason D Lee, and Zhaoran Wang. Neural temporal-difference learning converges to global optima. In Advances in Neural Information Processing Systems, 2019. +387 [12] Yuan Cao and Quanquan Gu. A generalization theory of gradient descent for learning overparameterized deep relu networks. arXiv preprint arXiv:1902.01384, 2019. [13] Yuan Cao and Quanquan Gu. Generalization bounds of stochastic gradient descent for wide and deep neural networks. In Advances in Neural Information Processing Systems, pages 10835–10845, 2019. [14] Olivier Chapelle and Lihong Li. An empirical evaluation of thompson sampling. In Advances in neural information processing systems, pages 2249–2257, 2011. [15] Sayak Ray Chowdhury and Aditya Gopalan. On kernelized multi-armed bandits. In International Conference on Machine Learning, pages 844–853, 2017. [16] Wei Chu, Lihong Li, Lev Reyzin, and Robert Schapire. Contextual bandits with linear payoff functions. In Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics, pages 208–214, 2011. [17] Mark Collier and Hector Urdiales Llorens. Deep contextual multi-armed bandits. arXiv preprint arXiv:1807.09809, 2018. +[18] George Cybenko. Approximation by superpositions of a sigmoidal function. Mathematics of control, signals and systems, 2(4):303–314, 1989. +[19] Varsha Dani, Thomas P Hayes, and Sham M Kakade. Stochastic linear optimization under bandit feedback. In Conference on Learning Theory, 2008. +[20] Aniket Anand Deshmukh, Abhimanu Kumar, Levi Boyles, Denis Charles, Eren Manavoglu, and Urun Dogan. Self-supervised contextual bandits in computer vision. arXiv preprint arXiv:2003.08485, 2020. +[21] Simon Du, Jason Lee, Haochuan Li, Liwei Wang, and Xiyu Zhai. Gradient descent finds global minima of deep neural networks. In International Conference on Machine Learning, pages 1675–1685, 2019. +[22] Simon S. Du, Xiyu Zhai, Barnabas Poczos, and Aarti Singh. Gradient descent provably optimizes over-parameterized neural networks. In International Conference on Learning Representations, 2019. URL https://openreview.net/forum?id=S1eK3i09YQ. +[23] Dheeru Dua and Casey Graff. UCI machine learning repository, 2017. URL http://archive. ics.uci.edu/ml. +[24] Sarah Filippi, Olivier Cappe, Aurélien Garivier, and Csaba Szepesvári. Parametric bandits: The generalized linear case. In Advances in Neural Information Processing Systems, pages 586–594, 2010. +[25] Ian Goodfellow, Yoshua Bengio, and Aaron Courville. Deep learning. MIT press, 2016. +[26] Geoffrey E Hinton and Ruslan R Salakhutdinov. Reducing the dimensionality of data with neural networks. science, 313(5786):504–507, 2006. +[27] Arthur Jacot, Franck Gabriel, and Clément Hongler. Neural tangent kernel: Convergence and generalization in neural networks. In Advances in neural information processing systems, pages 8571–8580, 2018. +[28] Branislav Kveton, Manzil Zaheer, Csaba Szepesvari, Lihong Li, Mohammad Ghavamzadeh, and Craig Boutilier. Randomized exploration in generalized linear bandits. In International Conference on Artificial Intelligence and Statistics, pages 2066–2076, 2020. +[29] John Langford and Tong Zhang. The epoch-greedy algorithm for multi-armed bandits with side information. In Advances in neural information processing systems, pages 817–824, 2008. +[30] Tor Lattimore and Csaba Szepesvári. Bandit algorithms. Cambridge University Press, 2020. +[31] Yann LeCun, Léon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998. +[32] Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning. nature, 521(7553):436–444, 2015. +[33] Jaehoon Lee, Lechao Xiao, Samuel Schoenholz, Yasaman Bahri, Roman Novak, Jascha SohlDickstein, and Jeffrey Pennington. Wide neural networks of any depth evolve as linear models under gradient descent. In Advances in neural information processing systems, pages 8570–8581, 2019. +[34] Lihong Li, Wei Chu, John Langford, and Robert E Schapire. A contextual-bandit approach to personalized news article recommendation. In Proceedings of the 19th international conference on World wide web, pages 661–670, 2010. +[35] Lihong Li, Yu Lu, and Dengyong Zhou. Provably optimal algorithms for generalized linear contextual bandits. In International Conference on Machine Learning, pages 2071–2080, 2017. +[36] Boyi Liu, Qi Cai, Zhuoran Yang, and Zhaoran Wang. Neural trust region/proximal policy optimization attains globally optimal policy. In Advances in Neural Information Processing Systems, pages 10564–10575, 2019. +[37] Samet Oymak and Mahdi Soltanolkotabi. Towards moderate overparameterization: global convergence guarantees for training shallow neural networks. IEEE Journal on Selected Areas in Information Theory, 2020. +[38] Carlos Riquelme, George Tucker, and Jasper Snoek. Deep bayesian bandits showdown: An empirical comparison of bayesian deep networks for thompson sampling. In International Conference on Learning Representations, 2018. URL https://openreview.net/forum? id=SyYe6k-CW. +[39] Paat Rusmevichientong and John N Tsitsiklis. Linearly parameterized bandits. Mathematics of Operations Research, 35(2):395–411, 2010. +[40] Daniel J. Russo, Benjamin Van Roy, Abbas Kazerouni, Ian Osband, and Zheng Wen. A tutorial on thompson sampling. Foundations and Trends $\textsuperscript { \textregistered }$ in Machine Learning, 11(1):1–96, 2018. ISSN 1935-8237. +[41] Jyotirmoy Sarkar. One-armed bandit problems with covariates. The Annals of Statistics, pages 1978–2002, 1991. +[42] Jasper Snoek, Oren Rippel, Kevin Swersky, Ryan Kiros, Nadathur Satish, Narayanan Sundaram, Mostofa Patwary, Mr Prabhat, and Ryan Adams. Scalable bayesian optimization using deep neural networks. In International conference on machine learning, pages 2171–2180, 2015. +[43] William R Thompson. On the likelihood that one unknown probability exceeds another in view of the evidence of two samples. Biometrika, 25(3/4):285–294, 1933. +[44] Michal Valko, Nathan Korda, Rémi Munos, Ilias Flaounas, and Nello Cristianini. Finite-time analysis of kernelised contextual bandits. In Proceedings of the Twenty-Ninth Conference on Uncertainty in Artificial Intelligence, pages 654–663, 2013. +[45] Lingxiao Wang, Qi Cai, Zhuoran Yang, and Zhaoran Wang. Neural policy gradient methods: Global optimality and rates of convergence. In International Conference on Learning Representations, 2020. URL https://openreview.net/forum?id=BJgQfkSYDS. +[46] Zhaoran Wang, Han Liu, and Tong Zhang. Optimal computational and statistical rates of convergence for sparse nonconvex learning problems. Annals of statistics, 42(6):2164, 2014. +[47] Pan Xu and Quanquan Gu. A finite-time analysis of q-learning with neural network function approximation. In International Conference on Machine Learning, 2020. +[48] Pan Xu, Jian Ma, and Quanquan Gu. Speeding up latent variable gaussian graphical model estimation via nonconvex optimization. In Advances in Neural Information Processing Systems, pages 1933–1944, 2017. +[49] Tom Zahavy and Shie Mannor. Deep neural linear bandits: Overcoming catastrophic forgetting through likelihood matching. arXiv preprint arXiv:1901.08612, 2019. +[50] Chiyuan Zhang, Samy Bengio, Moritz Hardt, Benjamin Recht, and Oriol Vinyals. Understanding deep learning requires rethinking generalization. In International Conference on Learning Representations, 2017. URL https://openreview.net/forum?id=Sy8gdB9xx. +[51] Weitong Zhang, Dongruo Zhou, Lihong Li, and Quanquan Gu. Neural thompson sampling. arXiv preprint arXiv:2010.00827, 2020. +[52] Dongruo Zhou, Lihong Li, and Quanquan Gu. Neural contextual bandits with ucb-based exploration. In International Conference on Machine Learning, 2020. +[53] Difan Zou and Quanquan Gu. An improved analysis of training over-parameterized deep neural networks. In Advances in Neural Information Processing Systems, pages 2053–2062, 2019. +[54] Difan Zou, Yuan Cao, Dongruo Zhou, and Quanquan Gu. Stochastic gradient descent optimizes over-parameterized deep relu networks. arXiv preprint arXiv:1811.08888, 2018. + +1. For all authors... + +(a) Do the main claims made in the abstract and introduction accurately reflect the paper’s contributions and scope? [Yes] +(b) Did you describe the limitations of your work? [Yes] We discussed the limitation of the assumptions made in this paper. We also admit in the experiment that the theory maybe conservative since our experiment does not require a very wide neural network to achieve good performance. +(c) Did you discuss any potential negative societal impacts of your work? [N/A] This work focuses on a general methodology in bandit problems and its theoretical analysis. It does not cause any negative social impact. +(d) Have you read the ethics review guidelines and ensured that your paper conforms to them? [Yes] + +2. If you are including theoretical results... + +(a) Did you state the full set of assumptions of all theoretical results? [Yes] See the assumptions listed in Section 4 +(b) Did you include complete proofs of all theoretical results? [Yes] Proofs are provided in the appendix. + +3. If you ran experiments... + +(a) Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes] We provide them in the supplementary material. +(b) Did you specify all the training details (e.g., data splits, hyperparameters, how they were chosen)? [Yes] We specify all the details in the Implementations paragraph of Section 5. +(c) Did you report error bars (e.g., with respect to the random seed after running experiments multiple times)? [Yes] All the figures are plotted with the standard error with respect to random repetitions. +(d) Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [Yes] We stated the type of workstation at the end of the first paragraph of Section 5. + +4. If you are using existing assets (e.g., code, data, models) or curating/releasing new assets... + +(a) If your work uses existing assets, did you cite the creators? [Yes] As we mentioned in Section 5, we used codes from baseline algorithms and public available datasets. All the assets were properly cited. +(b) Did you mention the license of the assets? [N/A] All the codes and datasets are open-source. +(c) Did you include any new assets either in the supplemental material or as a URL? [Yes] We include our code in the supplementary for reproduction. +(d) Did you discuss whether and how consent was obtained from people whose data you’re using/curating? [N/A] +(e) Did you discuss whether the data you are using/curating contains personally identifiable information or offensive content? [N/A] The data does not contain any personally identifiable information or offensive content. + +5. If you used crowdsourcing or conducted research with human subjects... + +(a) Did you include the full text of instructions given to participants and screenshots, if applicable? [N/A] +(b) Did you describe any potential participant risks, with links to Institutional Review Board (IRB) approvals, if applicable? [N/A] +(c) Did you include the estimated hourly wage paid to participants and the total amount spent on participant compensation? [N/A] \ No newline at end of file diff --git a/parse/train/jCxDyge46t2/jCxDyge46t2_content_list.json b/parse/train/jCxDyge46t2/jCxDyge46t2_content_list.json new file mode 100644 index 0000000000000000000000000000000000000000..f48b4cf3239a10dac4f495dc997c38d26202aab5 --- /dev/null +++ b/parse/train/jCxDyge46t2/jCxDyge46t2_content_list.json @@ -0,0 +1,1356 @@ +[ + { + "type": "text", + "text": "Neural Contextual Bandits with Deep Representation and Shallow Exploration ", + "text_level": 1, + "bbox": [ + 178, + 122, + 821, + 172 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Anonymous Author(s) \nAffiliation \nAddress \nemail ", + "bbox": [ + 423, + 226, + 580, + 281 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "Abstract ", + "text_level": 1, + "bbox": [ + 462, + 318, + 535, + 334 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "1 We study neural contextual bandits, a general class of contextual bandits, where \n2 each context-action pair is associated with a raw feature vector, but the specific \n3 reward generating function is unknown. We propose a novel learning algorithm \n4 that transforms the raw feature vector using the last hidden layer of a deep ReLU \n5 neural network (deep representation learning), and uses an upper confidence bound \n6 (UCB) approach to explore in the last linear layer (shallow exploration). We prove \n7 that under standard assumptions, our proposed algorithm achieves $\\widetilde { O } ( \\sqrt { T } )$ finite \n8 time regret, where $T$ is the learning time horizon. Compared with existing neural \n9 contextual bandit algorithms, our approach is computationally much more efficient \n10 since it only needs to explore in the last layer of the deep neural network. ", + "bbox": [ + 150, + 348, + 766, + 489 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "11 1 Introduction ", + "text_level": 1, + "bbox": [ + 148, + 512, + 312, + 530 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "12 Multi-armed bandits (MAB) [9, 8, 30] are a class of online decision-making problems where an \n13 agent needs to learn to maximize its expected cumulative reward while repeatedly interacting with a \n14 partially known environment. Based on a bandit algorithm (also called a strategy or policy), in each \n15 round, the agent adaptively chooses an arm, and then observes and receives a reward associated with \n16 that arm. Since only the reward of the chosen arm will be observed (bandit information feedback), \n17 a good bandit algorithm has to deal with the exploration-exploitation dilemma: trade-off between \n18 pulling the best arm based on existing knowledge/history data (exploitation) and trying the arms that \n19 have not been fully explored (exploration). \n20 In many real-world applications, the agent will also be able to access detailed contexts associated \n21 with the arms. For example, when a company wants to choose an advertisement to present to a user, \n22 the recommendation will be much more accurate if the company takes into consideration the contents, \n23 specifications, and other features of the advertisements in the arm set as well as the profile of the user. \n24 To encode the contextual information, contextual bandit models and algorithms have been developed, \n25 and widely studied both in theory and in practice [19, 39, 34, 16, 1]. Most existing contextual bandit \n26 algorithms assume that the expected reward of an arm at a context is a linear function in a known \n27 context-action feature vector, which leads to many useful algorithms such as LinUCB [16], OFUL [1], \n28 etc. The representation power of the linear model can be limited in applications such as marketing, \n29 social networking, clinical studies, etc., where the rewards are usually counts or binary variables. The \n30 linear contextual bandit problem has also been extended to richer classes of parametric bandits such \n31 as the generalized linear bandits [24, 35] and kernelised bandits [44, 15]. \n32 With the prevalence of deep neural networks (DNNs) and their phenomenal performances in many \n33 machine learning tasks [32, 25], there has emerged a line of work that employs DNNs to increase the \n34 representation power of contextual bandit algorithms [5, 38, 17, 49, 52, 20, 51]. The problems they \n35 solve are usually referred to as neural contextual bandits. For example, Zhou et al. [52] developed \n36 the NeuralUCB algorithm, which can be viewed as a natural extension of LinUCB [16, 1], where they \n37 use the output of a deep neural network with the feature vector as input to approximate the reward. \n38 Zhang et al. [51] adapted neural networks in Thompson Sampling [43, 14, 40] for both exploration \n39 and exploitation and proposed NeuralTS . For a fixed time horizon $T$ , it has been proved that both \n40 NeuralUCB and NeuralTS achieve a $O ( \\widetilde { d } \\sqrt { T } )$ regret bound, where $\\hat { d }$ is the effective dimension of a \n41 neural tangent kernel matrix which can potentially scale with $O ( T K )$ for $K$ -armed bandits. This \n42 high complexity is mainly due to that the exploration is performed over the entire huge neural network \n43 parameter space, which is inefficient and even infeasible when the number of neurons is large. A more \n44 realistic and efficient way of learning neural contextual bandits may be to just explore different arms \n45 using the last layer as the exploration parameter. More specifically, Riquelme et al. [38] provided \n46 an extensive empirical study of benchmark algorithms for contextual-bandits through the lens of \n47 Thompson Sampling, which suggests decoupling representation learning and uncertainty estimation \n48 improves performance. \n49 In this paper, we show that the decoupling of representation learning and the exploration can be \n50 theoretically validated. We study a new neural contextual bandit algorithm, which learns a mapping \n51 to transform the raw features associated with each context-action pair using a deep neural network \n52 (deep representation), and then performs an upper confidence bound (UCB)-type exploration over the \n53 linear output layer of the network (shallow exploration). We prove a sublinear regret of the proposed \n54 algorithm by exploiting the UCB exploration techniques in linear contextual bandits [1] and the \n55 analysis of deep overparameterized neural networks using neural tangent kernels [27]. Our theory \n56 confirms the empirically observed effectiveness of decoupling the deep representation learning and \n57 the UCB exploration in contextual bandits [38, 49]. ", + "bbox": [ + 147, + 542, + 825, + 655 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "", + "bbox": [ + 147, + 661, + 826, + 827 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "", + "bbox": [ + 147, + 833, + 823, + 902 + ], + "page_idx": 0 + }, + { + "type": "text", + "text": "", + "bbox": [ + 145, + 92, + 825, + 260 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 145, + 266, + 825, + 391 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "58 Contributions we summarize the main contributions of this paper as follows. ", + "bbox": [ + 147, + 397, + 681, + 411 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "• We propose a contextual bandit algorithm, Neural-LinUCB, for solving a general class of contextual bandit problems without knowing the specific reward generating function. The proposed algorithm learns a deep representation to transform the raw feature vectors and performs UCB-type exploration in the last layer of the neural network, which we refer to as deep representation and shallow exploration. Compared with LinUCB [34, 16] and neural bandits such as NeuralUCB [52] and NeuralTS [51], our algorithm enjoys the best of two worlds: strong expressiveness due to the deep representation and computational efficiency due to the shallow exploration. ", + "bbox": [ + 158, + 422, + 825, + 520 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "66 • Despite the usage of a DNN as the feature mapping, we prove a $\\widetilde { O } ( \\sqrt { T } )$ regret for the proposed \n67 Neural-LinUCB algorithm, which matches the regret bound of linear contextual bandits [16, 1]. \n68 To the best of our knowledge, this is the first work that theoretically shows the convergence of \n69 bandits algorithms under the scheme of deep representation and shallow exploration. It is notable \n70 that a similar scheme called Neural-Linear was proposed by Riquelme et al. [38] for Thompson \n71 sampling algorithms, and they empirically showed that decoupling representation learning and \n72 uncertainty estimation improves the performance. Our work confirms this observation from a \n73 theoretical perspective. ", + "bbox": [ + 148, + 525, + 825, + 637 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "• We conduct experiments on contextual bandit problems based on real-world datasets, demonstrating a better performance and computational efficiency of Neural-LinUCB over LinUCB and NeuralUCB, which well aligns with our theory. ", + "bbox": [ + 161, + 641, + 823, + 683 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "77 1.1 Additional related work ", + "text_level": 1, + "bbox": [ + 150, + 699, + 379, + 713 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "78 There is a line of related work to ours on the recent advance in the optimization and generalization \n79 analysis of deep neural networks. In particular, Jacot et al. [27] first introduced the neural tangent \n80 kernel (NTK) to characterize the training dynamics of network outputs in the infinite width limit. \n81 From the notion of NTK, a fruitful line of research emerged and showed that loss functions of \n82 deep neural networks trained by (stochastic) gradient descent can converge to the global minimum \n83 [22, 4, 21, 54, 53]. The generalization bounds for overparameterized deep neural networks are also \n84 established in Arora et al. [6, 7], Allen-Zhu et al. [3], Cao and Gu [12, 13]. Recently, the NTK based \n85 analysis is also extended to the study of sequential decision problems including bandits [52, 51], and \n86 reinforcement learning algorithms [11, 36, 45, 47]. \n87 Our algorithm is also different from Langford and Zhang [29], Agarwal et al. [2] which reduce the \n88 bandit problem to supervised learning. Moreover, their algorithms need to access an oracle that \n89 returns the optimal policy in a policy class given a sequence of context and reward vectors, whose \n90 regret depends on the VC-dimension of the policy class. \n91 Notation We use $[ k ]$ to denote a set $\\{ 1 , \\ldots , k \\}$ , $k \\in \\mathbb { N } ^ { + }$ . $\\| \\mathbf { x } \\| _ { 2 } = \\sqrt { \\mathbf { x } ^ { \\top } \\mathbf { x } }$ is the Euclidean norm of \n92 a vector $\\mathbf { x } \\in \\mathbb { R } ^ { d }$ . For a matrix $\\mathbf { W } \\in \\mathbb { R } ^ { m \\times n }$ , we denote by $\\lVert \\mathbf { W } \\rVert _ { 2 }$ and $\\| \\mathbf { W } \\| _ { F }$ its operator norm \n93 and Frobenius norm respectively. For a semi-definite matrix $\\mathbf { A } \\in \\mathbb { R } ^ { d \\times d }$ and a vector $\\mathbf { x } \\in \\mathbb { R } ^ { d }$ , we \n94 denote the Mahalanobis norm as $\\| \\mathbf { x } \\| _ { \\mathbf { A } } = \\sqrt { \\mathbf { x } ^ { \\top } \\mathbf { A } \\mathbf { x } }$ . Throughout this paper, we reserve the notations \n95 $\\{ C _ { i } \\} _ { i = 0 , 1 , \\ldots }$ to represent absolute positive constants that are independent of problem parameters such \n96 as dimension, sample size, iteration number, step size, network length and so on. The specific values \n97 of $\\{ C _ { i } \\} _ { i = 0 , 1 , \\ldots }$ . can be different in different context. For a parameter of interest $T$ and a function \n98 $f ( T )$ , we use notations such as $O ( f ( T ) )$ and $\\Omega ( f ( T ) )$ to hide constant factors and ${ \\widetilde { O } } ( f ( T ) )$ to hide \n99 constant and logarithmic dependence of $T$ . ", + "bbox": [ + 147, + 724, + 825, + 849 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 147, + 856, + 823, + 911 + ], + "page_idx": 1 + }, + { + "type": "text", + "text": "", + "bbox": [ + 145, + 89, + 826, + 224 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "2 Preliminaries ", + "text_level": 1, + "bbox": [ + 168, + 244, + 318, + 262 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "In this section, we provide the background of contextual bandits and deep neural networks. ", + "bbox": [ + 173, + 276, + 767, + 292 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "2.1 Linear contextual bandits ", + "text_level": 1, + "bbox": [ + 171, + 309, + 392, + 324 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "103 A contextual bandit is characterized by a tuple $( S , A , r )$ , where $s$ is the context (state) space, $\\mathcal { A }$ is the \n104 arm (action) space, and $r$ encodes the unknown reward generating function at all context-arm pairs. \n105 A learning agent, who knows $s$ and $\\mathcal { A }$ but does not know the true reward $r$ (values bounded in $( 0 , 1 )$ \n106 for simplicity), needs to interact with the contextual bandit for $T$ rounds. At each round $t = 1 , \\dots , T$ , \n107 the agent first observes a context $s _ { t } \\in S$ chosen by the environment; then it needs to adaptively select \n108 an arm $a _ { t } \\in \\mathcal A$ based on its past observations; finally it receives a reward $\\widehat { r } _ { t } ( \\mathbf { x } _ { s , a _ { t } } ) = r ( \\mathbf { x } _ { s , a _ { t } } ) + \\xi _ { t }$ , \n109 where $\\mathbf { x } _ { s , a } \\in \\mathbb { R } ^ { d }$ is a known feature vector for context-arm pair $( s , a ) \\in S \\times A$ , and $\\xi _ { t }$ is a random \n110 noise with zero mean. The agent’s objective is to maximize its expected total reward over these $T$ \n111 rounds, which is equivalent to minimizing the pseudo regret [8]: ", + "bbox": [ + 140, + 334, + 826, + 463 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/4ac3360b2010939a09fbbebb9b156b63a1082607a5e40bddbe053ae9cd752970.jpg", + "text": "$$\nR _ { T } = \\mathbb { E } \\bigg [ \\sum _ { t = 1 } ^ { T } \\big ( \\widehat { r } ( \\mathbf { x } _ { s _ { t } , a _ { t } ^ { * } } ) - \\widehat { r } ( \\mathbf { x } _ { s _ { t } , a _ { t } } ) \\big ) \\bigg ] ,\n$$", + "text_format": "latex", + "bbox": [ + 366, + 469, + 630, + 512 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "where 112 $a _ { t } ^ { * } \\in \\operatorname { a r g m a x } _ { a \\in \\mathcal { A } } \\{ r ( \\mathbf { x } _ { s _ { t } , a } ) = \\mathbb { E } [ \\widehat { r } ( \\mathbf { x } _ { s _ { t } , a } ) ] \\}$ . To simplify the exposition, we use $\\mathbf { x } _ { t , a }$ to denote 113 $\\mathbf { x } _ { s _ { t } , a }$ bsince it only depends on the round index $t$ in most bandit problems, and we assume $A = [ K ]$ . ", + "bbox": [ + 143, + 520, + 823, + 550 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "114 In some practical problems, the agent has a prior knowledge that the reward-generating function \n115 $r$ has some specific parametric form. For instance, in linear contextual bandits, the agent knows \n116 that $r ( \\mathbf { x } _ { s , a } ) \\stackrel { * } { = } \\mathbf { x } _ { s , a } ^ { \\top } \\pmb { \\theta } ^ { * }$ for some unknown weight vector $\\pmb { \\theta } ^ { * } \\in \\mathbb { R } ^ { d }$ . One provably sample efficient \n117 algorithm for linear contextual bandits is Linear Upper Confidence Bound (LinUCB) [1]. Specifically, \n118 at each round $t$ , LinUCB chooses action by the following strategy ", + "bbox": [ + 140, + 555, + 825, + 626 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/7a82c47bf9fdeb914a8ed73004df11005f1391e8ddacd3616c91fada54d74cfa.jpg", + "text": "$$\na _ { t } = \\underset { a \\in [ K ] } { \\operatorname { a r g m a x } } \\left. \\mathbf { x } _ { t , a } ^ { \\top } \\pmb { \\theta } _ { t } + \\alpha _ { t } \\Vert \\mathbf { x } _ { t , a } \\Vert _ { \\mathbf { A } _ { t } ^ { - 1 } } \\right. ,\n$$", + "text_format": "latex", + "bbox": [ + 362, + 632, + 633, + 665 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "119 where θt is a point estimate of θ∗, At = λI + Pti=1 xi,ai x> i,ai with some λ > 0 is a matrix \n120 defined based on the historical context-arm pairs, and $\\alpha _ { t } > 0$ is a tuning parameter that controls the \n121 exploration rate in LinUCB. ", + "bbox": [ + 142, + 675, + 825, + 719 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "2.2 Deep neural networks ", + "text_level": 1, + "bbox": [ + 161, + 736, + 366, + 752 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "In this paper, we use 123 $f ( \\mathbf { x } )$ to denote a neural network with input data $\\mathbf { x } \\in \\mathbb { R } ^ { d }$ . Let $L$ be the number 124 of hidden layers and $\\mathbf { W } _ { l } \\in \\mathbb { R } ^ { m _ { l } \\times m _ { l - 1 } }$ be the weight matrices in the $l$ -th layer, where $l = 1 , \\ldots , L$ 125 $m _ { 1 } = . . . = m _ { L - 1 } = m$ and $m _ { 0 } = m _ { L } = d$ . Then a $L$ -hidden layer neural network is defined as ", + "bbox": [ + 140, + 761, + 825, + 805 + ], + "page_idx": 2 + }, + { + "type": "equation", + "img_path": "images/8e1edc132f0151005d0207b777a8d3f22e9f803748e11419052ffd4036ea6c87.jpg", + "text": "$$\nf ( \\mathbf { x } ) = \\sqrt { m } \\pmb { \\theta } ^ { \\ast \\top } \\sigma _ { L } ( \\mathbf { W } _ { L } \\sigma _ { L - 1 } ( \\mathbf { W } _ { L - 1 } \\cdot \\cdot \\cdot \\sigma _ { 1 } ( \\mathbf { W } _ { 1 } \\mathbf { x } ) \\cdot \\cdot \\cdot ) ) ,\n$$", + "text_format": "latex", + "bbox": [ + 305, + 813, + 692, + 832 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "126 where $\\sigma _ { l }$ is an activation function and $\\pmb { \\theta } ^ { * } \\in \\mathbb { R } ^ { d }$ is the weight of the output layer. To simplify the \n127 presentation, we will assume $\\sigma _ { 1 } = \\sigma _ { 2 } = . . . = \\sigma _ { L } = \\sigma$ is the ReLU activation function, i.e., \n128 ${ \\bar { \\sigma } } ( x ) = \\operatorname* { m a x } \\{ 0 , x \\}$ for $x \\in \\mathbb { R }$ . We denote $\\mathbf { w } = ( \\mathrm { v e c } ( \\mathbf { W } _ { 1 } ) ^ { \\top } , \\ldots , \\mathrm { v e c } ( \\mathbf { W } _ { L } ) ^ { \\top } ) ^ { \\top }$ , which is the \n129 concatenation of the vectorized weight parameters of all hidden layers of the neural network. We also \n130 write $f ( \\mathbf { x } ; \\pmb { \\theta } ^ { * } , \\mathbf { w } ) = f ( \\mathbf { x } )$ in order to explicitly specify the weight parameters of neural network $f$ . It \n131 is easy to show that the dimension $p$ of vector w satisfies $p = ( L - 2 ) m ^ { 2 } + 2 m d .$ . To simplify the \n132 notation, we define $\\phi ( \\mathbf { x } ; \\mathbf { w } )$ as the output of the $L$ -th hidden layer of neural network $f$ . ", + "bbox": [ + 142, + 840, + 825, + 912 + ], + "page_idx": 2 + }, + { + "type": "text", + "text": "", + "bbox": [ + 142, + 90, + 825, + 121 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/adb847137db9f7bc640f969b5c2162050750e1e37bd0bca12621f7b419713ad6.jpg", + "text": "$$\n\\boldsymbol { \\phi } ( \\mathbf { x } ; \\mathbf { w } ) = \\sqrt { m } \\sigma ( \\mathbf { W } _ { L } \\sigma ( \\mathbf { W } _ { L - 1 } \\cdot \\cdot \\cdot \\sigma ( \\mathbf { W } _ { 1 } \\mathbf { x } ) \\cdot \\cdot \\cdot \\cdot ) ) .\n$$", + "text_format": "latex", + "bbox": [ + 328, + 127, + 669, + 146 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "33 Note that $\\phi ( \\mathbf { x } ; \\mathbf { w } )$ itself can also be viewed as a neural network with vector-valued outputs. ", + "bbox": [ + 153, + 154, + 774, + 170 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "134 3 Deep Representation and Shallow Exploration ", + "text_level": 1, + "bbox": [ + 148, + 188, + 591, + 207 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "135 The linear parametric form in linear contextual bandits might produce biased estimates of the reward \n136 due to the lack of representation power [42, 38]. In contrast, it is well known that deep neural networks \n137 are powerful enough to approximate an arbitrary function [18]. Therefore, a natural extension of \n138 linear contextual bandits is to use a deep neural network to approximate the reward generating \n139 function $r ( \\cdot )$ . Nonetheless, DNNs usually have a prohibitively large dimension for weight parameters, \n140 which makes the exploration in neural networks based UCB algorithm inefficient [28, 52]. \n141 In this work, we study a neural contextual bandit algorithm, where the hidden layers of a deep neural \n142 network are used to represent the features and the exploration is only performed in the last layer of the \n143 neural network. In particular, we assume that the reward generating function $r ( \\cdot )$ can be expressed as \n144 the inner product between a deep represented feature vector and an exploration weight parameter, \n145 namely, $\\bar { r ( \\cdot ) } = \\langle \\theta ^ { * } , \\psi ( \\cdot ) \\rangle$ , where $\\pmb { \\theta } ^ { * } \\in \\mathbb { R } ^ { d }$ is some weight parameter and $\\psi ( \\cdot )$ is an unknown feature \n146 mapping. This decoupling of the representation and the exploration will achieve the best of both \n147 worlds: efficient exploration in shallow (linear) models and high expressive power of deep models. \n148 To learn the unknown feature mapping, we propose to use a neural network to approximate it. In \n149 what follows, we will describe a neural contextual bandit algorithm that uses the output of the last \n150 hidden layer of a neural network to transform the raw feature vectors (deep representation) and \n151 performs UCB-type exploration in the last layer of the neural network (shallow exploration). Since \n152 the exploration is performed only in the last linear layer, we call this procedure Neural-LinUCB, \n153 which is displayed in Algorithm 1. \n154 Specifically, in round $t$ , the agent receives an action set with raw features $\\mathcal { X } _ { t } = \\{ \\mathbf { x } _ { t , 1 } , . . . , \\mathbf { x } _ { t , K } \\}$ . \n155 Then the agent chooses an arm $a _ { t }$ that maximizes the following upper confidence bound: ", + "bbox": [ + 140, + 220, + 826, + 304 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "", + "bbox": [ + 138, + 310, + 826, + 491 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "", + "bbox": [ + 140, + 496, + 828, + 525 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/4e4e26795f0bc8675499c5a7e793d1a20254225213eaf68061c88a5bda924326.jpg", + "text": "$$\na _ { t } = \\underset { k \\in [ K ] } { \\operatorname { a r g m a x } } \\Big \\{ \\langle \\phi ( \\mathbf { x } _ { t , k } ; \\mathbf { w } _ { t - 1 } ) , \\theta _ { t - 1 } \\rangle + \\alpha _ { t } \\| \\phi ( \\mathbf { x } _ { t , k } ; \\mathbf { w } _ { t - 1 } ) \\| _ { \\mathbf { A } _ { t - 1 } ^ { - 1 } } \\Big \\} ,\n$$", + "text_format": "latex", + "bbox": [ + 276, + 532, + 720, + 565 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "156 where $\\pmb { \\theta } _ { t - 1 }$ is a point estimate of the unknown weight in the last layer, $\\phi ( \\mathbf { x } ; \\mathbf { w } )$ is defined as in (2.3), \n157 $\\mathbf { w } _ { t - 1 }$ is an estimate of all the weight parameters in the hidden layers of the neural network, $\\alpha _ { t } > 0$ is \n158 the algorithmic parameter controlling the exploration, and ${ \\bf A } _ { t }$ is a matrix defined based on historical \n159 transformed features: ", + "bbox": [ + 140, + 574, + 825, + 631 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/bb0921b282182ea1867f207b10a77c9e44b5aa08b66915662ab83d66b11081cc.jpg", + "text": "$$\n\\mathbf { A } _ { t } = \\lambda \\mathbf { I } + \\sum _ { i = 1 } ^ { t } \\phi ( \\mathbf { x } _ { i , a _ { i } } ; \\mathbf { w } _ { i - 1 } ) \\phi ( \\mathbf { x } _ { i , a _ { i } } ; \\mathbf { w } _ { i - 1 } ) ^ { \\top } ,\n$$", + "text_format": "latex", + "bbox": [ + 338, + 636, + 658, + 679 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "and 160 $\\lambda > 0$ . After pulling arm $a _ { t }$ , the agent will observe a noisy reward $\\widehat { r } _ { t } : = \\widehat { r } ( \\mathbf { x } _ { t , a _ { t } } )$ defined as ", + "bbox": [ + 140, + 685, + 800, + 702 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/8b6a8c034c1ffd16f9b08b84c4acebf1cf345d72ed096ceeb6def98731c11d51.jpg", + "text": "$$\n\\widehat { r } ( \\mathbf { x } _ { t , k } ) = r ( \\mathbf { x } _ { t , k } ) + \\xi _ { t } ,\n$$", + "text_format": "latex", + "bbox": [ + 419, + 709, + 576, + 727 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "161 where $\\xi _ { t }$ is an independent $\\nu$ -subGaussian random noise for some $\\nu > 0$ and $r ( \\cdot )$ is an unknown \n162 reward function. In this paper, we will interchangeably use notation $\\widehat { r _ { t } }$ to denote the reward received \n163 at the $t$ -th step and an equivalent notation $\\widehat { r } ( \\mathbf { x } )$ b to express its dependence on the feature vector $\\mathbf { x }$ . ", + "bbox": [ + 142, + 733, + 825, + 777 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "164 Upon receiving the reward $\\widehat { r _ { t } }$ , the agent updates its estimate $\\theta _ { t }$ of the output layer weight by using the same 165 $\\ell ^ { 2 }$ b-regularized least-squares estimate in linear contextual bandits [1]. In particular, we have ", + "bbox": [ + 142, + 781, + 825, + 811 + ], + "page_idx": 3 + }, + { + "type": "equation", + "img_path": "images/4298c842934c21749769fa88aac2870581f38ab0de80fecd1721b4d74a21a85c.jpg", + "text": "$$\n\\pmb { \\theta } _ { t } = \\mathbf { A } _ { t } ^ { - 1 } \\mathbf { b } _ { t } ,\n$$", + "text_format": "latex", + "bbox": [ + 452, + 818, + 544, + 837 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "where 166 $\\begin{array} { r } { \\mathbf { b } _ { t } = \\sum _ { i = 1 } ^ { t } \\widehat { r } _ { i } \\phi ( \\mathbf { x } _ { i , a _ { i } } ; \\mathbf { w } _ { i - 1 } ) } \\end{array}$ . ", + "bbox": [ + 150, + 847, + 413, + 866 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "167 To save the computation, the neural network $\\phi ( \\cdot ; { \\mathbf w } _ { t } )$ will be updated once every $H$ steps. Therefore, \n168 we have $\\mathbf { w } _ { ( q - 1 ) H + 1 } = . . . = \\mathbf { w } _ { q H }$ for $q = 1 , 2 , \\ldots$ We call the time steps $\\{ ( q - 1 ) H + 1 , \\ldots , q H \\}$ \n169 an epoch with length $H$ . At time step $t = H q$ , for any $q = 1 , 2 , \\ldots$ , Algorithm 1 will retrain the \n170 neural network based on all the historical data. In Algorithm 2, our goal is to minimize the following \n171 empirical loss function: ", + "bbox": [ + 142, + 868, + 825, + 912 + ], + "page_idx": 3 + }, + { + "type": "text", + "text": "", + "bbox": [ + 143, + 90, + 825, + 119 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/44b174f71bcc58f2a5d6084eeb0f52dce165c4ca6d6d1b01659c83c8da9c4d91.jpg", + "text": "$$\n\\mathcal { L } _ { q } ( \\mathbf { w } ) = \\sum _ { i = 1 } ^ { q H } \\big ( \\pmb { \\theta } _ { i } ^ { \\top } \\phi ( \\mathbf { x } _ { i , a _ { i } } ; \\mathbf { w } ) - \\widehat { r } _ { i } \\big ) ^ { 2 } .\n$$", + "text_format": "latex", + "bbox": [ + 372, + 121, + 625, + 166 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "In practi172 i=1e, one can further save computational cost by onl eding data $\\{ \\mathbf { x } _ { i , a _ { i } } , \\widehat { r } _ { i } , \\pmb { \\theta } _ { i } \\} _ { i = ( q - 1 ) H + 1 } ^ { q H }$ $q$ $\\mathbf { w } _ { t }$ b, which does not hurt the performance 174 since the historical information has been encoded into the estimate of $\\theta _ { i }$ . In this paper, we will 175 perform the following gradient descent step ", + "bbox": [ + 140, + 170, + 825, + 229 + ], + "page_idx": 4 + }, + { + "type": "equation", + "img_path": "images/9b9d794581546b55e07caaecc79341427bc37d80fd6b763c6db383977553ad99.jpg", + "text": "$$\n\\mathbf { w } _ { q } ^ { ( s ) } = \\mathbf { w } _ { q } ^ { ( s - 1 ) } - \\eta _ { q } \\nabla _ { \\mathbf { w } } \\mathcal { L } _ { q } \\big ( \\mathbf { w } ^ { ( s - 1 ) } \\big ) .\n$$", + "text_format": "latex", + "bbox": [ + 374, + 231, + 622, + 252 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "176 for $s = 1 , \\ldots , n$ , where $\\mathbf { w } _ { q } ^ { ( 0 ) } = \\mathbf { w } ^ { ( 0 ) }$ is chosen as the same random initialization point. We will \n177 discuss more about the initial point $\\mathbf { w } ^ { ( 0 ) }$ in the next paragraph. Then Algorithm 2 outputs $\\mathbf { w } _ { q } ^ { ( n ) }$ and \n178 we set it as the updated weight parameter $\\mathbf { w } _ { H q + 1 }$ in Algorithm 1. In the next round, the agent will \n179 receive another action set $\\mathcal { X } _ { t + 1 }$ with raw feature vectors and repeat the above steps to choose the \n180 sub-optimal arm and update estimation for contextual parameters. \n181 Initialization: Recall that w is the collection of all hidden layer weight parameters of the neural \n182 network. We will follow the same initialization scheme as used in Zhou et al. [52], where each entry \n183 of the weight matrices follows some Gaussian distribution. Specifically, for any $l \\in \\{ 1 , \\ldots , L - 1 \\}$ , \n184 we set $\\mathbf { W } _ { l } = \\left[ \\begin{array} { c c } { \\mathbf { W } } & { \\mathbf { 0 } } \\\\ { \\mathbf { 0 } } & { \\mathbf { W } } \\end{array} \\right]$ , where each entry of $\\mathbf { W }$ follows distribution $N ( 0 , 4 / m )$ independently; for \n185 $\\mathbf { W } _ { L }$ , we set it as $\\begin{array} { r l } { [ \\mathbf { V } } & { { } - \\mathbf { V } ] } \\end{array}$ , where each entry of $\\mathbf { V }$ follows distribution $N ( 0 , 2 / m )$ independently. ", + "bbox": [ + 140, + 255, + 825, + 333 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "", + "bbox": [ + 140, + 337, + 825, + 428 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Comparison with LinUCB and NeuralUCB: Compared with linear contextual bandits in Section 2.1, Algorithm 1 has a distinct feature that it learns a deep neural network to obtain a deep representation of the raw data vectors and then performs UCB exploration. This deep representation allows our algorithm to characterize more intrinsic and latent information about the raw data $\\left\\{ \\mathbf { x } _ { t , k } \\right\\} _ { t \\in [ T ] , k \\in [ K ] } \\subset \\mathbb { R } ^ { d }$ . However, the increased complexity of the feature mapping $\\phi ( \\cdot ; { \\mathbf { w } } )$ also introduces great hardness in training. For instance, a recent work by Zhou et al. [52] also studied the neural contextual bandit problem, but different from (3.1), their algorithm (NeuralUCB) performs the UCB exploration on the entire network parameter space, which is $\\dot { \\mathbb { R } } ^ { \\widetilde { p } + d }$ , where $\\ddot { \\tilde { p } } = m + m d + ( L \\dot { \\bar { \\mathbf { \\alpha } } } 1 ) m ^ { 2 }$ . Note that in Zhou et al. [52], they need to compute the inverse eof a matrix $\\mathbf { Z } _ { t } \\in \\mathbb { R } ^ { ( \\widetilde { p } + d ) \\times ( \\widetilde { p } + d ) }$ , which is defined in a similar way to the matrix ${ \\bf A } _ { t }$ in our paper except that $\\mathbf { Z } _ { t }$ is defined based on the gradient of the network instead of the output of the last hidden layer as in (3.2). In sharp contrast, ${ \\bf A } _ { t }$ in our paper is only of size $d \\times d$ and thus is much more efficient and practical in implementation, which will be seen from our experiments in later sections. ", + "bbox": [ + 166, + 431, + 825, + 614 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "199 We note that there is also a similar algorithm to our Neural-LinUCB presented in Deshmukh et al. \n200 [20], where they studied the self-supervised learning loss in contextual bandits with neural network \n201 representation for computer vision problems. However, no regret analysis has been provided. When \n202 the feature mapping $\\phi ( \\cdot ; { \\mathbf { w } } )$ is an identity function, the problem reduces to linear contextual bandits \n203 where we directly use $\\mathbf { x } _ { t }$ as the feature vector. In this case, it is easy to see that Algorithm 1 reduces \n204 to LinUCB [16] since we do not need to learn the representation parameter w anymore. \n205 Comparison with Neural-Linear: The high-level idea of decoupling the representation and explo \n206 ration in our algorithm is also similar to that of the Neural-Linear algorithm [38, 49], which trains a \n207 deep neural network to learn a representation of the raw feature vectors, and then uses a Bayesian \n208 linear regression to estimate the uncertainty in the bandit problem. However, these two algorithms \n209 are significantly different since Neural-Linear [38] is a Thompson sampling based algorithm that \n210 uses posterior sampling to estimate the weight parameter $\\pmb { \\theta } ^ { * }$ via Bayesian linear regression, whereas \n211 Neural-LinUCB adopts upper confidence bound based techniques to estimate the weight $\\pmb { \\theta } ^ { * }$ . Never \n212 theless, both algorithms share the same idea of deep representation and shallow exploration, and we \n213 view our Neural-LinUCB algorithm as one instantiation of the Neural-Linear scheme. ", + "bbox": [ + 142, + 619, + 825, + 703 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "", + "bbox": [ + 142, + 709, + 825, + 834 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "214 4 Main Results ", + "text_level": 1, + "bbox": [ + 151, + 852, + 315, + 869 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "215 To analyze the regret bound of Algorithm 1, we first lay down some important assumptions on the \n216 neural contextual bandit model. ", + "bbox": [ + 147, + 882, + 825, + 911 + ], + "page_idx": 4 + }, + { + "type": "text", + "text": "Algorithm 1 Deep Representation and Shallow Exploration (Neural-LinUCB) ", + "text_level": 1, + "bbox": [ + 174, + 90, + 687, + 106 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "1: Input: regularization parameter $\\lambda > 0$ , number of total steps $T$ , episode length $H$ , exploration parameters $\\{ \\alpha _ { t } > 0 \\} _ { t \\in [ T ] }$ \n2: Initialization: $\\mathbf { A } _ { 0 } = \\lambda \\mathbf { I }$ , $\\mathbf { b } _ { 0 } = \\mathbf { 0 }$ ; entries of $\\pmb { \\theta } _ { 0 }$ follow $N ( 0 , 1 / d )$ , and $\\mathbf { w } ^ { ( 0 ) }$ is initialized as described in Section 3; $q = 1$ ; $\\mathbf { w } _ { 0 } = \\mathbf { w } ^ { ( 0 ) }$ \n3: for $t = 1 , \\dots , T$ do \n4: receive feature vectors $\\left\\{ \\mathbf { x } _ { t , 1 } , \\ldots , \\mathbf { x } _ { t , K } \\right\\}$ \n5: choose arm $\\begin{array} { r } { a _ { t } \\ = \\ \\mathrm { a r g m a x } _ { k \\in [ K ] } \\pmb { \\theta } _ { t - 1 } ^ { \\top } \\phi ( \\mathbf { x } _ { t , k } ; \\mathbf { w } _ { t - 1 } ) \\ + \\alpha _ { t } \\| \\phi ( \\mathbf { x } _ { t , k } ; \\mathbf { w } _ { t - 1 } ) \\| _ { \\mathbf { A } _ { t - 1 } ^ { - 1 } } } \\end{array}$ , and obtain reward $\\widehat { r } _ { t }$ \n6: update ${ \\bf A } _ { t }$ and $\\mathbf { b } _ { t }$ as follows: $\\begin{array} { r l } & { \\mathbf { \\Phi } ^ { \\mathbf { A } } t = \\tilde { \\mathbf { A } } _ { t - 1 } + \\tilde { \\phi } ( \\mathbf { x } _ { t , a _ { t } } ; \\mathbf { w } _ { t - 1 } ) \\phi ( \\mathbf { x } _ { t , a _ { t } } ; \\mathbf { w } _ { t - 1 } ) ^ { \\top } } \\\\ & { \\mathbf { b } _ { t } = \\mathbf { b } _ { t - 1 } + \\hat { r } _ { t } \\phi ( \\mathbf { x } _ { t , a _ { t } } ; \\mathbf { w } _ { t - 1 } ) , } \\end{array}$ , \n7: update $\\pmb { \\theta } _ { t } = \\mathbf { A } _ { t } ^ { - 1 } \\mathbf { b } _ { t }$ \n8: if $\\mathrm { n o d } ( t , H ) = 0$ then \n9: $\\mathbf { w } _ { t } \\gets$ output of Algorithm 2 \n10: $q = q + 1$ \n11: else \n12: $\\mathbf { w } _ { t } = \\mathbf { w } _ { t - 1 }$ \n13: end if \n14: end for \n15: Output $\\mathbf { w } _ { T }$ ", + "bbox": [ + 179, + 107, + 826, + 400 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Algorithm 2 Update Weight Parameters with Gradient Descent ", + "text_level": 1, + "bbox": [ + 173, + 417, + 589, + 433 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "1: Input: initial point $\\mathbf { w } _ { q } ^ { ( 0 ) } = \\mathbf { w } ^ { ( 0 ) }$ , maximum iteration number $n$ , step size $\\eta _ { q }$ , and loss function \ndefined in (3.5). \n2: for $s = 1 , \\ldots , n$ do \n3: $\\begin{array} { r } { \\mathbf { w } _ { q } ^ { ( s ) } = \\mathbf { w } _ { q } ^ { ( s - 1 ) } - \\eta _ { q } \\nabla _ { \\mathbf { w } } \\mathcal { L } _ { q } ( \\mathbf { w } _ { q } ^ { ( s - 1 ) } ) . } \\end{array}$ \n4: end for \n5: Output w(n)q \n7 Assumption 4.1. For all $i \\geq 1$ and $k \\in [ K ]$ , we assume that $\\| \\mathbf { x } _ { i , k } \\| _ { 2 } = 1$ and its entries satisfy \n8 $[ { \\bf { x } } _ { i , k } ] _ { j } \\stackrel { - } { = } [ { \\bf { x } } _ { j , k } ] _ { j + d / 2 }$ . ", + "bbox": [ + 179, + 436, + 825, + 530 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "", + "bbox": [ + 156, + 556, + 823, + 588 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "19 The assumption that $\\| \\mathbf { x } _ { i , k } \\| _ { 2 } = 1$ is not essential and is only imposed for simplicity, which is also 20 used in Zou and $\\mathrm { G u }$ [53], Zhou et al. [52]. Finally, the condition on the entries of $\\mathbf { x } _ { i , k }$ is also mild since otherwise we could always construct 21 $\\mathbf { x } _ { i , k } ^ { \\prime } = [ \\mathbf { x } _ { i , k } ^ { \\top } , \\mathbf { x } _ { i , k } ^ { \\top } ] ^ { \\top } / \\sqrt { 2 }$ to replace it. An implication of ", + "bbox": [ + 155, + 597, + 825, + 643 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "Assumption 4.1 is that the initialization scheme in Algorithm 1 results in $\\phi ( \\mathbf { x } _ { i , k } ; \\mathbf { w } ^ { ( 0 ) } ) = \\mathbf { 0 }$ for all $i \\in [ T ]$ and $k \\in [ K ]$ . ", + "bbox": [ + 158, + 645, + 818, + 675 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "24 We assume the following stability condition on the spectral norm of the neural network gradient: ", + "bbox": [ + 153, + 679, + 807, + 695 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "225 Assumption 4.2. There is a constant $\\ell _ { \\mathrm { L i p } } > 0$ such that it holds ", + "bbox": [ + 142, + 698, + 601, + 714 + ], + "page_idx": 5 + }, + { + "type": "equation", + "img_path": "images/35b5d1a533a2cbb936460df839bd22f4a249c0863f54259893aff798251c8f90.jpg", + "text": "$$\n\\left\\| \\frac { \\partial \\phi } { \\partial \\mathbf { w } } ( \\mathbf { x } ; \\mathbf { w } _ { 0 } ) - \\frac { \\partial \\phi } { \\partial \\mathbf { w } } ( \\mathbf { x } ^ { \\prime } ; \\mathbf { w } _ { 0 } ) \\right\\| _ { 2 } \\leq \\ell _ { \\mathrm { L i p } } \\| \\mathbf { x } - \\mathbf { x } ^ { \\prime } \\| _ { 2 } ,\n$$", + "text_format": "latex", + "bbox": [ + 333, + 719, + 661, + 756 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "for all 226 $\\mathbf { x } , \\mathbf { x } ^ { \\prime } \\in \\{ \\mathbf { x } _ { i , k } \\} _ { i \\in [ T ] , k \\in [ K ] }$ ", + "bbox": [ + 142, + 761, + 387, + 779 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "227 The inequality in Assumption 4.2 resembles the Lipschitz condition on the gradient of the neural \n228 network. However, it is essentially different from the smoothness condition since here the gradient \n229 is taken with respect to the neural network weights while the Lipschitz condition is imposed on the \n230 feature parameter x. Similar conditions are widely made in nonconvex optimization [46, 10, 48], in \n231 the name of first-order stability, which is essential to derive the convergence of alternating optimization \n232 algorithms. Furthermore, Assumption 4.2 is only required on the $T K$ training data points and a \n233 specific weight parameter $\\mathbf { w } _ { 0 }$ . Therefore, the condition will hold if the raw feature data lie in a \n234 certain subspace of $\\mathbb { R } ^ { d }$ . We provided some further discussions in the supplementary material about \n235 this assumption for interested readers. \n236 In order to analyze the regret bound of Algorithm 1, we need to characterize the properties of the \n237 deep neural network in (2.2) that is used to represent the feature vectors. Following a recent line of \n238 research [27, 12, 7, 52], we define the covariance between two data point $\\mathbf { x } , \\mathbf { y } \\in \\mathbb { R } ^ { \\bar { d } }$ as follows. ", + "bbox": [ + 138, + 786, + 825, + 912 + ], + "page_idx": 5 + }, + { + "type": "text", + "text": "", + "bbox": [ + 142, + 90, + 823, + 133 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/1ca433195a3a22df3540af3ebe2aaaac2e2cd8d29ba8df51dd10b38683d4f874.jpg", + "text": "$$\n\\begin{array} { r l } & { \\widetilde { \\pmb { \\Sigma } } ^ { ( 0 ) } ( \\mathbf x , \\mathbf y ) = \\pmb { \\Sigma } ^ { ( 0 ) } ( \\mathbf x , \\mathbf y ) = \\mathbf x ^ { \\top } \\mathbf y , } \\\\ & { \\pmb { \\Lambda } ^ { ( l ) } ( \\mathbf x , \\mathbf y ) = \\left[ \\pmb { \\Sigma } ^ { l - 1 } ( \\mathbf x , \\mathbf x ) \\quad \\pmb { \\Sigma } ^ { l - 1 } ( \\mathbf x , \\mathbf y ) \\right] , } \\\\ & { \\pmb { \\Sigma } ^ { ( l ) } ( \\mathbf x , \\mathbf y ) = 2 \\mathbb { E } _ { ( u , v ) \\sim N ( \\mathbf 0 , \\mathbf { A } ^ { ( l - 1 ) } ( \\mathbf x , \\mathbf y ) ) } [ \\sigma ( u ) \\sigma ( v ) ] , } \\\\ & { \\widetilde { \\pmb { \\Sigma } } ^ { ( l ) } ( \\mathbf x , \\mathbf y ) = 2 \\widetilde { \\pmb { \\Sigma } } ^ { ( l - 1 ) } ( \\mathbf x , \\mathbf y ) \\mathbb { E } _ { u , v } [ \\dot { \\sigma } ( u ) \\dot { \\sigma } ( v ) ] + \\pmb { \\Sigma } ^ { ( l ) } ( \\mathbf x , \\mathbf y ) , } \\end{array}\n$$", + "text_format": "latex", + "bbox": [ + 305, + 138, + 691, + 242 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "where 239 the ne240 $( u , v ) \\sim N ( \\mathbf { 0 } , \\mathbf { \\Lambda } \\Lambda ^ { ( l - 1 ) } ( \\mathbf { x } , \\mathbf { y } ) )$ , arix $\\dot { \\sigma } ( \\cdot )$ rivative of activation functiobased on all feature vectors $\\sigma ( \\cdot )$ $\\mathbf { H } \\in \\mathbb { R } ^ { T K \\times T K }$ $\\{ \\mathbf { x } _ { t , k } \\} _ { t \\in [ T ] , k \\in [ K ] }$ 241 Renumbering $\\{ \\mathbf { x } _ { t , k } \\} _ { t \\in [ T ] , k \\in [ K ] }$ as $\\{ \\mathbf { x } _ { i } \\} _ { i = 1 , \\dots , T K }$ , then each entry $\\mathbf { H } _ { i j }$ is defined as ", + "bbox": [ + 138, + 247, + 826, + 295 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/2fb231f3d5356d34d8d8c49952e6f3a715cd685b08edbbe90de4ddb6668893cc.jpg", + "text": "$$\n\\mathbf { H } _ { i j } = \\frac { 1 } { 2 } \\big ( \\widetilde { \\boldsymbol { \\Sigma } } ^ { ( L ) } ( \\mathbf { x } _ { i } , \\mathbf { x } _ { j } ) + \\boldsymbol { \\Sigma } ^ { ( L ) } ( \\mathbf { x } _ { i } , \\mathbf { x } _ { j } ) \\big ) ,\n$$", + "text_format": "latex", + "bbox": [ + 361, + 301, + 635, + 332 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "42 for all $i , j \\in [ T K ]$ . Based on the above definition, we impose the following assumption on $\\mathbf { H }$ ", + "bbox": [ + 155, + 338, + 784, + 353 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "43 Assumption 4.3. The neural tangent kernel defined in (4.2) is positive definite, i.e., $\\lambda _ { \\operatorname* { m i n } } ( \\mathbf { H } ) \\geq \\lambda _ { 0 }$ \n244 for some constant $\\lambda _ { 0 } > 0$ . \n245 Assumption 4.3 essentially requires the neural tangent kernel matrix $\\mathbf { H }$ to be non-singular, which is \n246 a mild condition and also imposed in other related work [21, 7, 12, 52]. Moreover, it is shown that \n247 Assumption 4.3 can be easily derived from Assumption 4.1 for two-layer ReLU networks [37, 53]. \n248 Therefore, Assumption 4.3 is mild or even negligible given the non-degeneration assumption on the \n249 feature vectors. Also note that matrix $\\mathbf { H }$ is only defined based on layers $l = 1 , \\ldots , L$ of the neural \n250 network, and does not depend on the output layer $\\pmb \\theta$ . It is easy to extend the definition of $\\mathbf { H }$ to the \n251 NTK matrix defined on all layers including the output layer $\\pmb \\theta$ , which would also be positive definite \n252 by Assumption 4.3 and the recursion in (4.2). \n253 Before we present the regret analysis of the neural contextual bandit, we need to modify the regret \n254 defined in (2.1) to account for the randomness of the neural network initialization. For a fixed time \n255 horizon $T$ , we define the regret of Algorithm 1 as follows. ", + "bbox": [ + 150, + 357, + 820, + 386 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "", + "bbox": [ + 140, + 396, + 825, + 508 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "", + "bbox": [ + 140, + 513, + 825, + 556 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/e138cc6ba15066bbaecd1cf6411285fc128bb09ba373e687791499a7b87b9e82.jpg", + "text": "$$\nR _ { T } = \\mathbb { E } \\bigg [ \\sum _ { t = 1 } ^ { T } \\big ( \\widehat { r } ( \\mathbf { x } _ { t , a _ { t } ^ { * } } ) - \\widehat { r } ( \\mathbf { x } _ { t , a _ { t } } ) \\big ) \\big | \\mathbf { w } ^ { ( 0 ) } \\bigg ] ,\n$$", + "text_format": "latex", + "bbox": [ + 356, + 564, + 642, + 606 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "256 where the expectation is taken over the randomness of the reward noise. Note that $R _ { T }$ defined in (4.3) \n257 is still a random variable since the initialization of Algorithm 2 is randomly generated. ", + "bbox": [ + 145, + 613, + 826, + 642 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "258 Now we are going to present the regret bound of the proposed algorithm. ", + "bbox": [ + 150, + 647, + 651, + 662 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "59 Theorem 4.4. Suppose Assumptions 4.1, 4.2 and 4.3 hold. Assume that $\\lVert \\pmb { \\theta } ^ { * } \\rVert _ { 2 } \\leq M$ for some \n0 positive constant $M > 0$ . For any $\\delta \\in ( 0 , 1 )$ , let us choose $\\alpha _ { t }$ in Neural-LinUCB as ", + "bbox": [ + 155, + 666, + 823, + 696 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/1ccf3a022ad345f8b9e1e117135312146f7c078729730e4c25e7b626c7abe5a8.jpg", + "text": "$$\n\\alpha _ { t } = \\nu \\sqrt { 2 \\big ( d \\log ( 1 + t \\log ( H K ) / \\lambda ) + \\log ( 1 / \\delta ) \\big ) } + \\lambda ^ { 1 / 2 } M .\n$$", + "text_format": "latex", + "bbox": [ + 297, + 703, + 700, + 731 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "261 We choose the step size $\\eta _ { q }$ of Algorithm 2 as ", + "bbox": [ + 140, + 736, + 470, + 751 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/1a0c80ba7aa5ba80ea8ffcf392b283294329a6a569c484aefbd52aab86c2c2b9.jpg", + "text": "$$\n\\eta _ { q } \\leq C _ { 0 } \\big ( d ^ { 2 } m n T ^ { 5 . 5 } L ^ { 6 } \\log ( T K / \\delta ) \\big ) ^ { - 1 } ,\n$$", + "text_format": "latex", + "bbox": [ + 367, + 758, + 627, + 781 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "262 and the width of the neural network satisfies $m = \\mathrm { p o l y } ( L , d , 1 / \\delta , H , \\log ( T K / \\delta ) )$ . With probability \n263 at least $1 - \\delta$ over the randomness of the initialization of the neural network, it holds that ", + "bbox": [ + 137, + 787, + 826, + 815 + ], + "page_idx": 6 + }, + { + "type": "equation", + "img_path": "images/12ee80d826958d4079dc3a4c3a914f1252140256056fcf795f509a38f52cb3e7.jpg", + "text": "$$\nR _ { T } \\leq C _ { 1 } \\alpha _ { T } \\sqrt { T d \\log \\left( 1 + \\frac { T G ^ { 2 } } { \\lambda d } \\right) } + \\frac { C _ { 2 } \\ell _ { \\mathrm { L i p } } L ^ { 3 } d ^ { 5 / 2 } T \\sqrt { \\log m \\log ( \\frac { 1 } { \\delta } ) \\log ( \\frac { T K } { \\delta } ) } \\| \\mathbf { r } - \\widetilde { \\mathbf { r } } \\| _ { \\mathbf { H } ^ { - 1 } } } { m ^ { 1 / 6 } } ,\n$$", + "text_format": "latex", + "bbox": [ + 217, + 820, + 779, + 859 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "264 where $\\{ C _ { i } \\} _ { i = 0 , 1 , 2 }$ are absolute constants independent of the problem parameters, $\\begin{array} { r l } { \\mathbf { r } } & { { } = } \\end{array}$ \n265 $( r ( \\mathbf { x } _ { 1 } ) , r ( \\mathbf { x } _ { 2 } ) , \\ldots , r ( \\mathbf { x } _ { T K } ) ) ^ { \\top } \\ \\in \\ \\mathbb { R } ^ { T K }$ and $\\widetilde { \\textbf { r } } = ( f ( \\mathbf { x } _ { 1 } ; \\pmb { \\theta } _ { 0 } , \\mathbf { w } _ { 0 } ) , \\dots , f ( \\mathbf { x } _ { T K } ; \\pmb { \\theta } _ { T - 1 } , \\mathbf { w } _ { T - 1 } ) ) ^ { \\top } \\ \\in$ \n266 $\\mathbb { R } ^ { T K }$ , and $\\| \\mathbf { r } \\| _ { \\mathbf { A } } = \\sqrt { \\mathbf { r } ^ { \\top } \\mathbf { A } \\mathbf { r } }$ . \n267 Remark 4.5. Theorem 4.4 shows that the regret of Algorithm 1 can be bounded by two parts: the \n268 first part is of order $\\widetilde { O } ( \\sqrt { T } )$ , which resembles the regret bound of linear contextual bandits [1]; the \n269 second part is of order $\\widetilde { O } ( m ^ { - 1 / 6 } T \\sqrt { ( \\mathbf { r } - \\widetilde { \\mathbf { r } } ) ^ { \\top } \\mathbf { H } ^ { - 1 } ( \\mathbf { r } - \\widetilde { \\mathbf { r } } ) } )$ , which depends on the estimation error \n270 of the neural network $f$ e efor the reward generating function $r$ and the neural tangent kernel $\\mathbf { H }$ . ", + "bbox": [ + 140, + 864, + 826, + 912 + ], + "page_idx": 6 + }, + { + "type": "text", + "text": "", + "bbox": [ + 140, + 90, + 825, + 152 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "It is worth noting that our theoretical analysis depends on the reward structure assumption that $r ( \\cdot ) = \\langle \\theta \\ast , \\psi ( \\cdot ) \\rangle$ . However, the linear structure between $\\pmb { \\theta } \\ast$ and $\\psi ( \\cdot )$ is not essential. As long as the deep representation of the feature vector and the uncertainty weight parameter can be decoupled, Algorithm 1 can be easily extended to settings with milder assumptions on the reward structure such as generalized linear models [41, 24, 35, 28]. For more general bandit models where no assumption is imposed to the reward generating function, it is still unclear whether the decoupled deep representation and shallow exploration would work especially in cases a thorough exploration may be needed. ", + "bbox": [ + 166, + 165, + 825, + 276 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Based on the result in Theorem 4.4, we can easily verify the following conclusion: ", + "bbox": [ + 171, + 281, + 714, + 297 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Corollary 4.6. Under the same conditions of Theorem 4.4, if we choose a sufficiently overparameterized neural network mapping $\\phi ( \\cdot )$ such that $m \\geq T ^ { 3 }$ , then the regret of Algorithm 1 is $R _ { T } = { \\widetilde { O } } ( { \\sqrt { T } } { \\sqrt { ( \\mathbf { r } - { \\widetilde { \\mathbf { r } } } ) ^ { \\top } \\mathbf { H } ^ { - 1 } ( \\mathbf { r } - { \\widetilde { \\mathbf { r } } } ) } } )$ . ", + "bbox": [ + 173, + 303, + 825, + 349 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Remark 4.7. For the ease of presentation, let us denote $\\mathcal { E } : = \\| \\mathbf { r } - \\widetilde { \\mathbf { r } } \\| _ { \\mathbf { H } ^ { - 1 } }$ . If we have $\\mathcal { E } = O ( 1 )$ , the total regret in Theorem 4.4 becomes $\\widetilde { O } ( \\sqrt { T } )$ which matches the regret of linear contextual bandits [1]. We remark that there is a similar assumption in [52] where they assume that $\\mathbf { r } ^ { \\top } \\mathbf { H } ^ { - 1 } \\mathbf { r }$ can be upper bounded by a constant. They show that this term can be bounded by the RKHS norm of $\\mathbf { r }$ if it belongs to the RKHS induced by the neural tangent kernel [6, 7, 33]. In addition, $\\mathcal { E }$ here is the difference between the true reward function and the neural network function, which can also be small if the deep neural network function well approximates the reward generating function $r ( \\cdot )$ . ", + "bbox": [ + 173, + 353, + 825, + 455 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "290 5 Experiments ", + "text_level": 1, + "bbox": [ + 143, + 478, + 312, + 496 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "291 \n292 \n293 \n294 \n295 \n296 \n297 \n298 \n299 \n300 \n301 \n302 \n303 \n304 \n305 ", + "bbox": [ + 140, + 511, + 163, + 733 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "In this section, we provide empirical evaluations of Neural-LinUCB on real-world datasets. As we have discussed in Section 3, Neural-LinUCB could be viewed as an instantiation of the NeuralLinear scheme studied in Riquelme et al. [38] except that we use the UCB exploration instead of the posterior sampling exploration therein. Note that there has been an extensive comparison [38] of the Neural-Linear methods with many other baselines such as greedy algorithms, Variational Inference, Expectation-Propagation, Bayesian Non-parametrics and so on. Therefore, we do not seek a thorough empirical comparison of Neural-LinUCB with all existing bandits algorithms. We refer readers who are interested in the performance of Neural-Linear methods with deep representation and shallow exploration compared with a vast of baselines in the literature to the benchmark study by Riquelme et al. [38]. In this experiment, we only aim to show the advantages of our algorithm over the following baselines: (1) Neural-Linear [38]; (2) LinUCB [16], which does not have a deep representation of the feature vectors; and (3) NeuralUCB [52], which performs UCB exploration on all the parameters of the neural network instead of the shallow exploration used in our paper. All numerical experiments were run on a workstation with Intel(R) Xeon(R) CPU E5-2637 v4 $@$ 3.50GHz. ", + "bbox": [ + 166, + 512, + 825, + 705 + ], + "page_idx": 7 + }, + { + "type": "text", + "text": "Datasets: we evaluate the performances of all algorithms on bandit problems created from real-world data. Specifically, following the experimental setting in Zhou et al. [52],we use datasets (Shuttle) Statlog, Magic and Covertype from UCI machine learning repository [23], and the MINST dataset from LeCun et al. [31]. The details of these datasets are presented in Table 1. In Table 1, each instance represents a feature vector $\\mathbf { x } \\in \\mathbb { R } ^ { d }$ that is associated with one of the $K$ arms, and dimension $d$ is the number of attributes in each instance. ", + "bbox": [ + 173, + 712, + 825, + 795 + ], + "page_idx": 7 + }, + { + "type": "table", + "img_path": "images/17a76c88452a1df4dd62f466cc96b5f8f0025af5a5dc5f886cbdbb807d4353c0.jpg", + "table_caption": [ + "Table 1: Specifications of datasets from the UCI machine learning repository used in this paper. " + ], + "table_footnote": [], + "table_body": "
StatlogMagicCovertypeMNIST
Number of attributes91154784
Number of arms72710
Number of instances58,00019,020581,01260,000
", + "bbox": [ + 290, + 830, + 704, + 897 + ], + "page_idx": 7 + }, + { + "type": "image", + "img_path": "images/943bd96058f4439657901f76b8c00684d91994e2afa724f14d33961644a23aaf.jpg", + "image_caption": [ + "Figure 1: The cumulative regrets of LinUCB, NeuralUCB, Neural-Linear and Neural-LinUCB over 15, 000 rounds. Experiments are averaged over 10 repetitions. " + ], + "image_footnote": [], + "bbox": [ + 173, + 93, + 823, + 200 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "311 Implementations: for LinUCB, we follow the setting in Li et al. [34] to use disjoint models \n312 for different arms. For neural network based algorithms such as NeuralUCB, Neural-Linear and \n313 Neural-LinUCB, we use a ReLU neural network defined as in (2.2) with $L = 2$ and 2000 for the \n314 UCI datasets (Statlog, Magic, Covertype). Thus the neural network weights are $\\mathbf { W } _ { 1 } \\in \\mathbb { R } ^ { m \\times d }$ \n315 $\\mathbf { W } _ { 2 } \\in \\mathbb { R } ^ { k \\times m }$ , and $\\pmb { \\theta } \\in \\mathbb { R } ^ { \\widetilde { k } }$ respectively, where $k = 1 0 0$ , $m = 2 0 0 0$ , and $d$ is the dimension of \n316 features in the corresponding task. Since the problem size of the MNIST dataset is larger, inspired \n317 by Hinton and Salakhutdinov [26], we use a deeper NN and set $L = 3$ , $k = 1 0 0$ and $m = 1 0 0$ , \n318 with weights $\\mathbf { W } _ { 1 } \\in \\mathbb { R } ^ { m \\times d }$ , $\\mathbf { W } _ { 2 } \\in \\mathbb { R } ^ { m \\times m }$ , $\\mathbf { W _ { 3 } } \\in \\mathbb { R } ^ { k \\times m }$ , and $\\pmb \\theta \\in \\mathbb { R } ^ { k }$ . We set the time horizon \n319 $T = 1 5 , 0 0 0$ , which is the total number of rounds for each algorithm on each dataset. We use \n320 gradient decent to optimize the network weights, with a step size $\\eta _ { q } = 1 \\mathrm { e } { - 5 }$ and maximum iteration \n321 number $n = 1 , 0 0 0$ . To speed up the training process, the network parameter w is updated every \n322 $H = 1 0 0$ rounds starting from round 2000. We also apply early stopping when the loss difference \n323 of two consecutive iterations is smaller than a threshold of 1e-6. We set $\\lambda = 1$ and $\\alpha _ { t } = 0 . 0 2$ \n324 for all algorithms, $t \\in [ T ]$ . Following the setting in Riquelme et al. [38], we use round-robin to \n325 independently select each arm for 3 times at the beginning of each algorithm. For NeuralUCB, since \n326 it is computationally unaffordable to perform the original UCB exploration as displayed in Zhou et al. \n327 [52], we follow their experimental setting to replace the matrix $\\mathbf { Z } _ { t } \\in \\mathbb { R } ^ { ( d + \\widetilde { p } ) \\times \\widetilde { ( } d + \\widetilde { p } ) }$ in Zhou et al. \n328 [52] with its diagonal matrix. ", + "bbox": [ + 143, + 257, + 825, + 507 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "Results: we plot the cumulative regret of all algorithms versus round in Figures 1(a), 1(b) and 1(c) for UCI datasets and in Figure 1(d) for MNIST. The results are reported based on the average of 10 repetitions over different random shuffles of the datasets. It can be seen that algorithms based on neural network representations (NeuralUCB, Neural-Linear and Neural-LinUCB) consistently outperform the linear contextual bandit method LinUCB, which shows that linear models may lack representation power and find biased estimates for the underlying reward generating function. Furthermore, our proposed Neural-LinUCB achieves a comparable regret with NeuralUCB in all experiments despite the fact that our algorithm only explores in the output layer of the neural network, which is more computationally efficient as we will show in the sequel.The results in our experiment are well aligned with our theory that deep representation and shallow exploration are sufficient to guarantee a good performance of neural contextual bandit algorithms, which is also consistent with the findings in existing literature [38] that decoupling the representation learning and uncertainty estimation improves the performance. ", + "bbox": [ + 169, + 513, + 825, + 693 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "We also conducted experiments to study the effects of different widths of deep neural networks on the regret performance and to show the computational efficiency of Neural-LinUCB compared with existing neural bandit algorithms. Due to the space limit, we defer the results to Appendix A. ", + "bbox": [ + 173, + 700, + 825, + 741 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "6 Conclusions ", + "text_level": 1, + "bbox": [ + 165, + 753, + 305, + 770 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "In this paper, we propose a new neural contextual bandit algorithm called Neural-LinUCB, which uses the hidden layers of a ReLU neural network as a deep representation of the raw feature vectors and performs UCB type exploration on the last layer of the neural network. By incorporating techniques in liner contextual bandits and neural tangent kernels, we prove that the proposed algorithm achieves a sublinear regret when the width of the network is sufficiently large. This is the first regret analysis of neural contextual bandit algorithms with deep representation and shallow exploration, which have been observed in practice to work well on many benchmark bandit problems [38]. We also conducted experiments on real-world datasets to demonstrate the advantage of the proposed algorithm over LinUCB and existing neural contextual bandit algorithms. ", + "bbox": [ + 171, + 786, + 825, + 911 + ], + "page_idx": 8 + }, + { + "type": "text", + "text": "355 References [1] Yasin Abbasi-Yadkori, Dávid Pál, and Csaba Szepesvári. Improved algorithms for linear stochastic bandits. In Advances in Neural Information Processing Systems, pages 2312–2320, 2011. [2] Alekh Agarwal, Daniel Hsu, Satyen Kale, John Langford, Lihong Li, and Robert Schapire. Taming the monster: A fast and simple algorithm for contextual bandits. In International Conference on Machine Learning, pages 1638–1646, 2014. [3] Zeyuan Allen-Zhu, Yuanzhi Li, and Yingyu Liang. Learning and generalization in overparameterized neural networks, going beyond two layers. In Advances in neural information processing systems, pages 6155–6166, 2019. [4] Zeyuan Allen-Zhu, Yuanzhi Li, and Zhao Song. A convergence theory for deep learning via over-parameterization. In International Conference on Machine Learning, pages 242–252, 2019. [5] Robin Allesiardo, Raphaël Féraud, and Djallel Bouneffouf. A neural networks committee for the contextual bandit problem. In International Conference on Neural Information Processing, pages 374–381. Springer, 2014. [6] Sanjeev Arora, Simon Du, Wei Hu, Zhiyuan Li, and Ruosong Wang. Fine-grained analysis of optimization and generalization for overparameterized two-layer neural networks. In International Conference on Machine Learning, pages 322–332, 2019. [7] Sanjeev Arora, Simon S Du, Wei Hu, Zhiyuan Li, Russ R Salakhutdinov, and Ruosong Wang. On exact computation with an infinitely wide neural net. In Advances in Neural Information Processing Systems, pages 8139–8148, 2019. [8] Jean-Yves Audibert, Rémi Munos, and Csaba Szepesvári. Exploration–exploitation tradeoff using variance estimates in multi-armed bandits. Theoretical Computer Science, 410(19): 1876–1902, 2009. [9] Peter Auer, Nicolo Cesa-Bianchi, and Paul Fischer. Finite-time analysis of the multiarmed bandit problem. Machine learning, 47(2-3):235–256, 2002. \n382 [10] Sivaraman Balakrishnan, Martin J Wainwright, Bin Yu, et al. Statistical guarantees for the em algorithm: From population to sample-based analysis. The Annals of Statistics, 45(1):77–120, 2017. [11] Qi Cai, Zhuoran Yang, Jason D Lee, and Zhaoran Wang. Neural temporal-difference learning converges to global optima. In Advances in Neural Information Processing Systems, 2019. \n387 [12] Yuan Cao and Quanquan Gu. A generalization theory of gradient descent for learning overparameterized deep relu networks. arXiv preprint arXiv:1902.01384, 2019. [13] Yuan Cao and Quanquan Gu. Generalization bounds of stochastic gradient descent for wide and deep neural networks. In Advances in Neural Information Processing Systems, pages 10835–10845, 2019. [14] Olivier Chapelle and Lihong Li. An empirical evaluation of thompson sampling. In Advances in neural information processing systems, pages 2249–2257, 2011. [15] Sayak Ray Chowdhury and Aditya Gopalan. On kernelized multi-armed bandits. In International Conference on Machine Learning, pages 844–853, 2017. [16] Wei Chu, Lihong Li, Lev Reyzin, and Robert Schapire. Contextual bandits with linear payoff functions. In Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics, pages 208–214, 2011. [17] Mark Collier and Hector Urdiales Llorens. Deep contextual multi-armed bandits. arXiv preprint arXiv:1807.09809, 2018. \n[18] George Cybenko. Approximation by superpositions of a sigmoidal function. Mathematics of control, signals and systems, 2(4):303–314, 1989. \n[19] Varsha Dani, Thomas P Hayes, and Sham M Kakade. Stochastic linear optimization under bandit feedback. In Conference on Learning Theory, 2008. \n[20] Aniket Anand Deshmukh, Abhimanu Kumar, Levi Boyles, Denis Charles, Eren Manavoglu, and Urun Dogan. Self-supervised contextual bandits in computer vision. arXiv preprint arXiv:2003.08485, 2020. \n[21] Simon Du, Jason Lee, Haochuan Li, Liwei Wang, and Xiyu Zhai. Gradient descent finds global minima of deep neural networks. In International Conference on Machine Learning, pages 1675–1685, 2019. \n[22] Simon S. Du, Xiyu Zhai, Barnabas Poczos, and Aarti Singh. Gradient descent provably optimizes over-parameterized neural networks. In International Conference on Learning Representations, 2019. URL https://openreview.net/forum?id=S1eK3i09YQ. \n[23] Dheeru Dua and Casey Graff. UCI machine learning repository, 2017. URL http://archive. ics.uci.edu/ml. \n[24] Sarah Filippi, Olivier Cappe, Aurélien Garivier, and Csaba Szepesvári. Parametric bandits: The generalized linear case. In Advances in Neural Information Processing Systems, pages 586–594, 2010. \n[25] Ian Goodfellow, Yoshua Bengio, and Aaron Courville. Deep learning. MIT press, 2016. \n[26] Geoffrey E Hinton and Ruslan R Salakhutdinov. Reducing the dimensionality of data with neural networks. science, 313(5786):504–507, 2006. \n[27] Arthur Jacot, Franck Gabriel, and Clément Hongler. Neural tangent kernel: Convergence and generalization in neural networks. In Advances in neural information processing systems, pages 8571–8580, 2018. \n[28] Branislav Kveton, Manzil Zaheer, Csaba Szepesvari, Lihong Li, Mohammad Ghavamzadeh, and Craig Boutilier. Randomized exploration in generalized linear bandits. In International Conference on Artificial Intelligence and Statistics, pages 2066–2076, 2020. \n[29] John Langford and Tong Zhang. The epoch-greedy algorithm for multi-armed bandits with side information. In Advances in neural information processing systems, pages 817–824, 2008. \n[30] Tor Lattimore and Csaba Szepesvári. Bandit algorithms. Cambridge University Press, 2020. \n[31] Yann LeCun, Léon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998. \n[32] Yann LeCun, Yoshua Bengio, and Geoffrey Hinton. Deep learning. nature, 521(7553):436–444, 2015. \n[33] Jaehoon Lee, Lechao Xiao, Samuel Schoenholz, Yasaman Bahri, Roman Novak, Jascha SohlDickstein, and Jeffrey Pennington. Wide neural networks of any depth evolve as linear models under gradient descent. In Advances in neural information processing systems, pages 8570–8581, 2019. \n[34] Lihong Li, Wei Chu, John Langford, and Robert E Schapire. A contextual-bandit approach to personalized news article recommendation. In Proceedings of the 19th international conference on World wide web, pages 661–670, 2010. \n[35] Lihong Li, Yu Lu, and Dengyong Zhou. Provably optimal algorithms for generalized linear contextual bandits. In International Conference on Machine Learning, pages 2071–2080, 2017. \n[36] Boyi Liu, Qi Cai, Zhuoran Yang, and Zhaoran Wang. Neural trust region/proximal policy optimization attains globally optimal policy. In Advances in Neural Information Processing Systems, pages 10564–10575, 2019. \n[37] Samet Oymak and Mahdi Soltanolkotabi. Towards moderate overparameterization: global convergence guarantees for training shallow neural networks. IEEE Journal on Selected Areas in Information Theory, 2020. \n[38] Carlos Riquelme, George Tucker, and Jasper Snoek. Deep bayesian bandits showdown: An empirical comparison of bayesian deep networks for thompson sampling. In International Conference on Learning Representations, 2018. URL https://openreview.net/forum? id=SyYe6k-CW. \n[39] Paat Rusmevichientong and John N Tsitsiklis. Linearly parameterized bandits. Mathematics of Operations Research, 35(2):395–411, 2010. \n[40] Daniel J. Russo, Benjamin Van Roy, Abbas Kazerouni, Ian Osband, and Zheng Wen. A tutorial on thompson sampling. Foundations and Trends $\\textsuperscript { \\textregistered }$ in Machine Learning, 11(1):1–96, 2018. ISSN 1935-8237. \n[41] Jyotirmoy Sarkar. One-armed bandit problems with covariates. The Annals of Statistics, pages 1978–2002, 1991. \n[42] Jasper Snoek, Oren Rippel, Kevin Swersky, Ryan Kiros, Nadathur Satish, Narayanan Sundaram, Mostofa Patwary, Mr Prabhat, and Ryan Adams. Scalable bayesian optimization using deep neural networks. In International conference on machine learning, pages 2171–2180, 2015. \n[43] William R Thompson. On the likelihood that one unknown probability exceeds another in view of the evidence of two samples. Biometrika, 25(3/4):285–294, 1933. \n[44] Michal Valko, Nathan Korda, Rémi Munos, Ilias Flaounas, and Nello Cristianini. Finite-time analysis of kernelised contextual bandits. In Proceedings of the Twenty-Ninth Conference on Uncertainty in Artificial Intelligence, pages 654–663, 2013. \n[45] Lingxiao Wang, Qi Cai, Zhuoran Yang, and Zhaoran Wang. Neural policy gradient methods: Global optimality and rates of convergence. In International Conference on Learning Representations, 2020. URL https://openreview.net/forum?id=BJgQfkSYDS. \n[46] Zhaoran Wang, Han Liu, and Tong Zhang. Optimal computational and statistical rates of convergence for sparse nonconvex learning problems. Annals of statistics, 42(6):2164, 2014. \n[47] Pan Xu and Quanquan Gu. A finite-time analysis of q-learning with neural network function approximation. In International Conference on Machine Learning, 2020. \n[48] Pan Xu, Jian Ma, and Quanquan Gu. Speeding up latent variable gaussian graphical model estimation via nonconvex optimization. In Advances in Neural Information Processing Systems, pages 1933–1944, 2017. \n[49] Tom Zahavy and Shie Mannor. Deep neural linear bandits: Overcoming catastrophic forgetting through likelihood matching. arXiv preprint arXiv:1901.08612, 2019. \n[50] Chiyuan Zhang, Samy Bengio, Moritz Hardt, Benjamin Recht, and Oriol Vinyals. Understanding deep learning requires rethinking generalization. In International Conference on Learning Representations, 2017. URL https://openreview.net/forum?id=Sy8gdB9xx. \n[51] Weitong Zhang, Dongruo Zhou, Lihong Li, and Quanquan Gu. Neural thompson sampling. arXiv preprint arXiv:2010.00827, 2020. \n[52] Dongruo Zhou, Lihong Li, and Quanquan Gu. Neural contextual bandits with ucb-based exploration. In International Conference on Machine Learning, 2020. \n[53] Difan Zou and Quanquan Gu. An improved analysis of training over-parameterized deep neural networks. In Advances in Neural Information Processing Systems, pages 2053–2062, 2019. \n[54] Difan Zou, Yuan Cao, Dongruo Zhou, and Quanquan Gu. Stochastic gradient descent optimizes over-parameterized deep relu networks. arXiv preprint arXiv:1811.08888, 2018. ", + "bbox": [ + 148, + 71, + 828, + 920 + ], + "page_idx": 9 + }, + { + "type": "text", + "text": "", + "bbox": [ + 160, + 50, + 828, + 920 + ], + "page_idx": 10 + }, + { + "type": "text", + "text": "", + "bbox": [ + 161, + 70, + 828, + 914 + ], + "page_idx": 11 + }, + { + "type": "text", + "text": "1. For all authors... ", + "bbox": [ + 214, + 116, + 339, + 130 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "(a) Do the main claims made in the abstract and introduction accurately reflect the paper’s contributions and scope? [Yes] \n(b) Did you describe the limitations of your work? [Yes] We discussed the limitation of the assumptions made in this paper. We also admit in the experiment that the theory maybe conservative since our experiment does not require a very wide neural network to achieve good performance. \n(c) Did you discuss any potential negative societal impacts of your work? [N/A] This work focuses on a general methodology in bandit problems and its theoretical analysis. It does not cause any negative social impact. \n(d) Have you read the ethics review guidelines and ensured that your paper conforms to them? [Yes] ", + "bbox": [ + 238, + 136, + 825, + 295 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "2. If you are including theoretical results... ", + "bbox": [ + 214, + 299, + 493, + 314 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "(a) Did you state the full set of assumptions of all theoretical results? [Yes] See the assumptions listed in Section 4 \n(b) Did you include complete proofs of all theoretical results? [Yes] Proofs are provided in the appendix. ", + "bbox": [ + 238, + 318, + 825, + 376 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "3. If you ran experiments... ", + "bbox": [ + 214, + 381, + 393, + 395 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "(a) Did you include the code, data, and instructions needed to reproduce the main experimental results (either in the supplemental material or as a URL)? [Yes] We provide them in the supplementary material. \n(b) Did you specify all the training details (e.g., data splits, hyperparameters, how they were chosen)? [Yes] We specify all the details in the Implementations paragraph of Section 5. \n(c) Did you report error bars (e.g., with respect to the random seed after running experiments multiple times)? [Yes] All the figures are plotted with the standard error with respect to random repetitions. \n(d) Did you include the total amount of compute and the type of resources used (e.g., type of GPUs, internal cluster, or cloud provider)? [Yes] We stated the type of workstation at the end of the first paragraph of Section 5. ", + "bbox": [ + 238, + 398, + 825, + 573 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "4. If you are using existing assets (e.g., code, data, models) or curating/releasing new assets... ", + "bbox": [ + 214, + 577, + 823, + 592 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "(a) If your work uses existing assets, did you cite the creators? [Yes] As we mentioned in Section 5, we used codes from baseline algorithms and public available datasets. All the assets were properly cited. \n(b) Did you mention the license of the assets? [N/A] All the codes and datasets are open-source. \n(c) Did you include any new assets either in the supplemental material or as a URL? [Yes] We include our code in the supplementary for reproduction. \n(d) Did you discuss whether and how consent was obtained from people whose data you’re using/curating? [N/A] \n(e) Did you discuss whether the data you are using/curating contains personally identifiable information or offensive content? [N/A] The data does not contain any personally identifiable information or offensive content. ", + "bbox": [ + 238, + 595, + 825, + 771 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "5. If you used crowdsourcing or conducted research with human subjects... ", + "bbox": [ + 214, + 776, + 705, + 791 + ], + "page_idx": 12 + }, + { + "type": "text", + "text": "(a) Did you include the full text of instructions given to participants and screenshots, if applicable? [N/A] \n(b) Did you describe any potential participant risks, with links to Institutional Review Board (IRB) approvals, if applicable? [N/A] \n(c) Did you include the estimated hourly wage paid to participants and the total amount spent on participant compensation? [N/A] ", + "bbox": [ + 238, + 795, + 825, + 885 + ], + "page_idx": 12 + } +] \ No newline at end of file diff --git a/parse/train/jCxDyge46t2/jCxDyge46t2_model.json b/parse/train/jCxDyge46t2/jCxDyge46t2_model.json new file mode 100644 index 0000000000000000000000000000000000000000..5a0d1884e0fabe8fa4b7da17c2a9133afad31a42 --- /dev/null +++ b/parse/train/jCxDyge46t2/jCxDyge46t2_model.json @@ -0,0 +1,28836 @@ +[ + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 250, + 1456, + 1406, + 1456, + 1406, + 1821, + 250, + 1821 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 256, + 767, + 1304, + 767, + 1304, + 1079, + 256, + 1079 + ], + "score": 0.973 + }, + { + "category_id": 0, + "poly": [ + 303, + 271, + 1398, + 271, + 1398, + 381, + 303, + 381 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 251, + 1197, + 1405, + 1197, + 1405, + 1442, + 251, + 1442 + ], + "score": 0.962 + }, + { + "category_id": 1, + "poly": [ + 252, + 1835, + 1401, + 1835, + 1401, + 1988, + 252, + 1988 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 720, + 499, + 987, + 499, + 987, + 620, + 720, + 620 + ], + "score": 0.898 + }, + { + "category_id": 0, + "poly": [ + 253, + 1130, + 531, + 1130, + 531, + 1168, + 253, + 1168 + ], + "score": 0.891 + }, + { + "category_id": 0, + "poly": [ + 787, + 701, + 913, + 701, + 913, + 737, + 787, + 737 + ], + "score": 0.891 + }, + { + "category_id": 2, + "poly": [ + 288, + 2032, + 1387, + 2032, + 1387, + 2062, + 288, + 2062 + ], + "score": 0.598 + }, + { + "category_id": 1, + "poly": [ + 288, + 2032, + 1387, + 2032, + 1387, + 2062, + 288, + 2062 + ], + "score": 0.276 + }, + { + "category_id": 13, + "poly": [ + 1143, + 950, + 1231, + 950, + 1231, + 989, + 1143, + 989 + ], + "score": 0.93, + "latex": "\\widetilde { O } ( \\sqrt { T } )" + }, + { + "category_id": 13, + "poly": [ + 606, + 987, + 630, + 987, + 630, + 1013, + 606, + 1013 + ], + "score": 0.81, + "latex": "T" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 267.0, + 1398.0, + 267.0, + 1398.0, + 334.0, + 298.0, + 334.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 588.0, + 326.0, + 1110.0, + 326.0, + 1110.0, + 386.0, + 588.0, + 386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 1128.0, + 535.0, + 1128.0, + 535.0, + 1174.0, + 244.0, + 1174.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 782.0, + 698.0, + 920.0, + 698.0, + 920.0, + 742.0, + 782.0, + 742.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2027.0, + 1391.0, + 2027.0, + 1391.0, + 2066.0, + 294.0, + 2066.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1461.0, + 278.0, + 1461.0, + 278.0, + 1488.0, + 248.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1455.0, + 1406.0, + 1455.0, + 1406.0, + 1491.0, + 292.0, + 1491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1492.0, + 277.0, + 1492.0, + 277.0, + 1518.0, + 247.0, + 1518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1485.0, + 1408.0, + 1485.0, + 1408.0, + 1521.0, + 293.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1522.0, + 278.0, + 1522.0, + 278.0, + 1549.0, + 247.0, + 1549.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1514.0, + 1410.0, + 1514.0, + 1410.0, + 1554.0, + 293.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1553.0, + 278.0, + 1553.0, + 278.0, + 1580.0, + 248.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1549.0, + 1409.0, + 1549.0, + 1409.0, + 1582.0, + 294.0, + 1582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1583.0, + 278.0, + 1583.0, + 278.0, + 1609.0, + 248.0, + 1609.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1576.0, + 1406.0, + 1576.0, + 1406.0, + 1611.0, + 293.0, + 1611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1613.0, + 278.0, + 1613.0, + 278.0, + 1640.0, + 247.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1607.0, + 1405.0, + 1607.0, + 1405.0, + 1640.0, + 294.0, + 1640.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1644.0, + 278.0, + 1644.0, + 278.0, + 1670.0, + 247.0, + 1670.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1638.0, + 1405.0, + 1638.0, + 1405.0, + 1672.0, + 296.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1674.0, + 277.0, + 1674.0, + 277.0, + 1701.0, + 248.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1669.0, + 1406.0, + 1669.0, + 1406.0, + 1702.0, + 294.0, + 1702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1705.0, + 278.0, + 1705.0, + 278.0, + 1731.0, + 247.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1697.0, + 1408.0, + 1697.0, + 1408.0, + 1735.0, + 292.0, + 1735.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1736.0, + 278.0, + 1736.0, + 278.0, + 1761.0, + 247.0, + 1761.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1730.0, + 1405.0, + 1730.0, + 1405.0, + 1763.0, + 293.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1765.0, + 278.0, + 1765.0, + 278.0, + 1792.0, + 247.0, + 1792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1759.0, + 1405.0, + 1759.0, + 1405.0, + 1792.0, + 294.0, + 1792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1796.0, + 277.0, + 1796.0, + 277.0, + 1821.0, + 248.0, + 1821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1788.0, + 1107.0, + 1788.0, + 1107.0, + 1823.0, + 290.0, + 1823.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 261.0, + 778.0, + 275.0, + 778.0, + 275.0, + 797.0, + 261.0, + 797.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 769.0, + 1305.0, + 769.0, + 1305.0, + 801.0, + 396.0, + 801.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 806.0, + 277.0, + 806.0, + 277.0, + 829.0, + 259.0, + 829.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 801.0, + 1304.0, + 801.0, + 1304.0, + 831.0, + 396.0, + 831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 838.0, + 277.0, + 838.0, + 277.0, + 860.0, + 259.0, + 860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 393.0, + 829.0, + 1305.0, + 829.0, + 1305.0, + 861.0, + 393.0, + 861.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 868.0, + 277.0, + 868.0, + 277.0, + 890.0, + 259.0, + 890.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 860.0, + 1305.0, + 860.0, + 1305.0, + 893.0, + 395.0, + 893.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 897.0, + 277.0, + 897.0, + 277.0, + 921.0, + 259.0, + 921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 889.0, + 1306.0, + 889.0, + 1306.0, + 924.0, + 391.0, + 924.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 928.0, + 277.0, + 928.0, + 277.0, + 951.0, + 259.0, + 951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 919.0, + 1305.0, + 919.0, + 1305.0, + 955.0, + 394.0, + 955.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 963.0, + 277.0, + 963.0, + 277.0, + 986.0, + 259.0, + 986.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 953.0, + 1142.0, + 953.0, + 1142.0, + 990.0, + 391.0, + 990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1232.0, + 953.0, + 1308.0, + 953.0, + 1308.0, + 990.0, + 1232.0, + 990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 993.0, + 277.0, + 993.0, + 277.0, + 1017.0, + 259.0, + 1017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 986.0, + 605.0, + 986.0, + 605.0, + 1019.0, + 395.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 631.0, + 986.0, + 1306.0, + 986.0, + 1306.0, + 1019.0, + 631.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 1024.0, + 277.0, + 1024.0, + 277.0, + 1046.0, + 259.0, + 1046.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 1015.0, + 1306.0, + 1015.0, + 1306.0, + 1052.0, + 394.0, + 1052.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 250.0, + 1054.0, + 278.0, + 1054.0, + 278.0, + 1077.0, + 250.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 395.0, + 1048.0, + 1212.0, + 1048.0, + 1212.0, + 1081.0, + 395.0, + 1081.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 1205.0, + 277.0, + 1205.0, + 277.0, + 1228.0, + 252.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1195.0, + 1405.0, + 1195.0, + 1405.0, + 1235.0, + 291.0, + 1235.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1235.0, + 278.0, + 1235.0, + 278.0, + 1263.0, + 249.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1228.0, + 1407.0, + 1228.0, + 1407.0, + 1266.0, + 292.0, + 1266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1266.0, + 278.0, + 1266.0, + 278.0, + 1293.0, + 249.0, + 1293.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1259.0, + 1407.0, + 1259.0, + 1407.0, + 1297.0, + 294.0, + 1297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1294.0, + 281.0, + 1294.0, + 281.0, + 1325.0, + 247.0, + 1325.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1290.0, + 1404.0, + 1290.0, + 1404.0, + 1324.0, + 294.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1326.0, + 279.0, + 1326.0, + 279.0, + 1353.0, + 249.0, + 1353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1318.0, + 1407.0, + 1318.0, + 1407.0, + 1356.0, + 294.0, + 1356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1356.0, + 278.0, + 1356.0, + 278.0, + 1383.0, + 249.0, + 1383.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1349.0, + 1407.0, + 1349.0, + 1407.0, + 1386.0, + 291.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1387.0, + 279.0, + 1387.0, + 279.0, + 1414.0, + 249.0, + 1414.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1381.0, + 1405.0, + 1381.0, + 1405.0, + 1415.0, + 295.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1417.0, + 279.0, + 1417.0, + 279.0, + 1442.0, + 249.0, + 1442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1410.0, + 776.0, + 1410.0, + 776.0, + 1446.0, + 292.0, + 1446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1840.0, + 279.0, + 1840.0, + 279.0, + 1865.0, + 249.0, + 1865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1832.0, + 1404.0, + 1832.0, + 1404.0, + 1871.0, + 294.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1871.0, + 279.0, + 1871.0, + 279.0, + 1897.0, + 249.0, + 1897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1864.0, + 1402.0, + 1864.0, + 1402.0, + 1897.0, + 296.0, + 1897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1902.0, + 279.0, + 1902.0, + 279.0, + 1927.0, + 249.0, + 1927.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1892.0, + 1405.0, + 1892.0, + 1405.0, + 1933.0, + 292.0, + 1933.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1933.0, + 279.0, + 1933.0, + 279.0, + 1959.0, + 249.0, + 1959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1926.0, + 1402.0, + 1926.0, + 1402.0, + 1959.0, + 294.0, + 1959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1962.0, + 280.0, + 1962.0, + 280.0, + 1988.0, + 249.0, + 1988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1954.0, + 1404.0, + 1954.0, + 1404.0, + 1992.0, + 292.0, + 1992.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 718.0, + 499.0, + 988.0, + 499.0, + 988.0, + 531.0, + 718.0, + 531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 787.0, + 527.0, + 913.0, + 527.0, + 913.0, + 563.0, + 787.0, + 563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 797.0, + 556.0, + 902.0, + 556.0, + 902.0, + 593.0, + 797.0, + 593.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 807.0, + 593.0, + 891.0, + 593.0, + 891.0, + 619.0, + 807.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 2027.0, + 1391.0, + 2027.0, + 1391.0, + 2066.0, + 294.0, + 2066.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 0, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 251, + 1596, + 1405, + 1596, + 1405, + 1872, + 251, + 1872 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 249, + 587, + 1404, + 587, + 1404, + 863, + 249, + 863 + ], + "score": 0.956 + }, + { + "category_id": 1, + "poly": [ + 249, + 203, + 1405, + 203, + 1405, + 573, + 249, + 573 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 252, + 1884, + 1402, + 1884, + 1402, + 2008, + 252, + 2008 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 276, + 1412, + 1401, + 1412, + 1401, + 1505, + 276, + 1505 + ], + "score": 0.899 + }, + { + "category_id": 0, + "poly": [ + 256, + 1541, + 646, + 1541, + 646, + 1572, + 256, + 1572 + ], + "score": 0.891 + }, + { + "category_id": 1, + "poly": [ + 252, + 875, + 1160, + 875, + 1160, + 907, + 252, + 907 + ], + "score": 0.881 + }, + { + "category_id": 1, + "poly": [ + 270, + 931, + 1404, + 931, + 1404, + 1145, + 270, + 1145 + ], + "score": 0.783 + }, + { + "category_id": 1, + "poly": [ + 253, + 1156, + 1404, + 1156, + 1404, + 1403, + 253, + 1403 + ], + "score": 0.772 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 858, + 2061, + 858, + 2085, + 841, + 2085 + ], + "score": 0.716 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.137 + }, + { + "category_id": 13, + "poly": [ + 1047, + 1153, + 1136, + 1153, + 1136, + 1193, + 1047, + 1193 + ], + "score": 0.93, + "latex": "\\widetilde { O } ( \\sqrt { T } )" + }, + { + "category_id": 13, + "poly": [ + 708, + 295, + 811, + 295, + 811, + 333, + 708, + 333 + ], + "score": 0.93, + "latex": "O ( \\widetilde { d } \\sqrt { T } )" + }, + { + "category_id": 13, + "poly": [ + 999, + 331, + 1090, + 331, + 1090, + 364, + 999, + 364 + ], + "score": 0.92, + "latex": "O ( T K )" + }, + { + "category_id": 13, + "poly": [ + 1044, + 297, + 1061, + 297, + 1061, + 326, + 1044, + 326 + ], + "score": 0.86, + "latex": "\\hat { d }" + }, + { + "category_id": 13, + "poly": [ + 1136, + 332, + 1164, + 332, + 1164, + 358, + 1136, + 358 + ], + "score": 0.85, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 1053, + 267, + 1075, + 267, + 1075, + 291, + 1053, + 291 + ], + "score": 0.83, + "latex": "T" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1538.0, + 650.0, + 1538.0, + 650.0, + 1575.0, + 246.0, + 1575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2093.0, + 839.0, + 2093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2093.0, + 838.0, + 2093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1602.0, + 278.0, + 1602.0, + 278.0, + 1627.0, + 249.0, + 1627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1597.0, + 1404.0, + 1597.0, + 1404.0, + 1630.0, + 295.0, + 1630.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1632.0, + 278.0, + 1632.0, + 278.0, + 1658.0, + 248.0, + 1658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1627.0, + 1405.0, + 1627.0, + 1405.0, + 1660.0, + 295.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1664.0, + 278.0, + 1664.0, + 278.0, + 1688.0, + 248.0, + 1688.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1656.0, + 1408.0, + 1656.0, + 1408.0, + 1692.0, + 294.0, + 1692.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1692.0, + 278.0, + 1692.0, + 278.0, + 1720.0, + 248.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1687.0, + 1405.0, + 1687.0, + 1405.0, + 1720.0, + 295.0, + 1720.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1721.0, + 281.0, + 1721.0, + 281.0, + 1751.0, + 247.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1718.0, + 1405.0, + 1718.0, + 1405.0, + 1751.0, + 295.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1752.0, + 281.0, + 1752.0, + 281.0, + 1782.0, + 247.0, + 1782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1747.0, + 1405.0, + 1747.0, + 1405.0, + 1784.0, + 298.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1785.0, + 279.0, + 1785.0, + 279.0, + 1811.0, + 248.0, + 1811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1777.0, + 1407.0, + 1777.0, + 1407.0, + 1814.0, + 294.0, + 1814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1814.0, + 278.0, + 1814.0, + 278.0, + 1841.0, + 248.0, + 1841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1808.0, + 1405.0, + 1808.0, + 1405.0, + 1841.0, + 294.0, + 1841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1845.0, + 279.0, + 1845.0, + 279.0, + 1871.0, + 248.0, + 1871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1840.0, + 863.0, + 1840.0, + 863.0, + 1872.0, + 294.0, + 1872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 593.0, + 277.0, + 593.0, + 277.0, + 619.0, + 249.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 587.0, + 1406.0, + 587.0, + 1406.0, + 623.0, + 292.0, + 623.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 623.0, + 277.0, + 623.0, + 277.0, + 651.0, + 247.0, + 651.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 611.0, + 1407.0, + 611.0, + 1407.0, + 658.0, + 291.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 655.0, + 277.0, + 655.0, + 277.0, + 681.0, + 247.0, + 681.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 649.0, + 1404.0, + 649.0, + 1404.0, + 682.0, + 294.0, + 682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 683.0, + 277.0, + 683.0, + 277.0, + 711.0, + 247.0, + 711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 679.0, + 1404.0, + 679.0, + 1404.0, + 712.0, + 296.0, + 712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 713.0, + 277.0, + 713.0, + 277.0, + 741.0, + 247.0, + 741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 708.0, + 1407.0, + 708.0, + 1407.0, + 745.0, + 294.0, + 745.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 743.0, + 280.0, + 743.0, + 280.0, + 773.0, + 246.0, + 773.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 739.0, + 1404.0, + 739.0, + 1404.0, + 772.0, + 294.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 776.0, + 277.0, + 776.0, + 277.0, + 803.0, + 249.0, + 803.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 768.0, + 1404.0, + 768.0, + 1404.0, + 805.0, + 294.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 806.0, + 277.0, + 806.0, + 277.0, + 833.0, + 249.0, + 833.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 799.0, + 1406.0, + 799.0, + 1406.0, + 836.0, + 293.0, + 836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 837.0, + 277.0, + 837.0, + 277.0, + 863.0, + 249.0, + 863.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 829.0, + 872.0, + 829.0, + 872.0, + 866.0, + 293.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 209.0, + 277.0, + 209.0, + 277.0, + 236.0, + 249.0, + 236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 204.0, + 1408.0, + 204.0, + 1408.0, + 237.0, + 295.0, + 237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 237.0, + 280.0, + 237.0, + 280.0, + 268.0, + 246.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 232.0, + 1407.0, + 232.0, + 1407.0, + 269.0, + 293.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 269.0, + 277.0, + 269.0, + 277.0, + 297.0, + 247.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 263.0, + 1052.0, + 263.0, + 1052.0, + 299.0, + 293.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 263.0, + 1407.0, + 263.0, + 1407.0, + 299.0, + 1076.0, + 299.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 305.0, + 277.0, + 305.0, + 277.0, + 333.0, + 249.0, + 333.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 295.0, + 707.0, + 295.0, + 707.0, + 335.0, + 293.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 812.0, + 295.0, + 1043.0, + 295.0, + 1043.0, + 335.0, + 812.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1062.0, + 295.0, + 1408.0, + 295.0, + 1408.0, + 335.0, + 1062.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 336.0, + 277.0, + 336.0, + 277.0, + 363.0, + 247.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 329.0, + 998.0, + 329.0, + 998.0, + 365.0, + 291.0, + 365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 329.0, + 1135.0, + 329.0, + 1135.0, + 365.0, + 1091.0, + 365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1165.0, + 329.0, + 1407.0, + 329.0, + 1407.0, + 365.0, + 1165.0, + 365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 367.0, + 279.0, + 367.0, + 279.0, + 393.0, + 247.0, + 393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 361.0, + 1405.0, + 361.0, + 1405.0, + 395.0, + 295.0, + 395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 396.0, + 277.0, + 396.0, + 277.0, + 424.0, + 247.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 392.0, + 1405.0, + 392.0, + 1405.0, + 424.0, + 293.0, + 424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 427.0, + 279.0, + 427.0, + 279.0, + 453.0, + 247.0, + 453.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 420.0, + 1407.0, + 420.0, + 1407.0, + 456.0, + 291.0, + 456.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 457.0, + 279.0, + 457.0, + 279.0, + 484.0, + 247.0, + 484.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 451.0, + 1407.0, + 451.0, + 1407.0, + 487.0, + 292.0, + 487.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 488.0, + 279.0, + 488.0, + 279.0, + 514.0, + 247.0, + 514.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 483.0, + 1405.0, + 483.0, + 1405.0, + 515.0, + 292.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 518.0, + 279.0, + 518.0, + 279.0, + 544.0, + 247.0, + 544.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 511.0, + 1407.0, + 511.0, + 1407.0, + 548.0, + 293.0, + 548.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 548.0, + 279.0, + 548.0, + 279.0, + 575.0, + 247.0, + 575.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 543.0, + 560.0, + 543.0, + 560.0, + 580.0, + 292.0, + 580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1890.0, + 279.0, + 1890.0, + 279.0, + 1917.0, + 249.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1884.0, + 1404.0, + 1884.0, + 1404.0, + 1919.0, + 296.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1919.0, + 279.0, + 1919.0, + 279.0, + 1947.0, + 248.0, + 1947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1912.0, + 1405.0, + 1912.0, + 1405.0, + 1949.0, + 293.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1950.0, + 279.0, + 1950.0, + 279.0, + 1978.0, + 249.0, + 1978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1945.0, + 1404.0, + 1945.0, + 1404.0, + 1980.0, + 294.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1981.0, + 279.0, + 1981.0, + 279.0, + 2008.0, + 249.0, + 2008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1975.0, + 925.0, + 1975.0, + 925.0, + 2010.0, + 293.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1413.0, + 1405.0, + 1413.0, + 1405.0, + 1446.0, + 296.0, + 1446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1444.0, + 1405.0, + 1444.0, + 1405.0, + 1478.0, + 323.0, + 1478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1469.0, + 855.0, + 1469.0, + 855.0, + 1512.0, + 322.0, + 1512.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 243.0, + 872.0, + 1165.0, + 872.0, + 1165.0, + 912.0, + 243.0, + 912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 931.0, + 1406.0, + 931.0, + 1406.0, + 966.0, + 293.0, + 966.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 962.0, + 1405.0, + 962.0, + 1405.0, + 996.0, + 323.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 990.0, + 1405.0, + 990.0, + 1405.0, + 1029.0, + 322.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1023.0, + 1405.0, + 1023.0, + 1405.0, + 1055.0, + 322.0, + 1055.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 320.0, + 1049.0, + 1404.0, + 1049.0, + 1404.0, + 1088.0, + 320.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1083.0, + 1405.0, + 1083.0, + 1405.0, + 1118.0, + 323.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 324.0, + 1114.0, + 1221.0, + 1114.0, + 1221.0, + 1148.0, + 324.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1166.0, + 280.0, + 1166.0, + 280.0, + 1191.0, + 249.0, + 1191.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 1156.0, + 1046.0, + 1156.0, + 1046.0, + 1194.0, + 301.0, + 1194.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1137.0, + 1156.0, + 1406.0, + 1156.0, + 1406.0, + 1194.0, + 1137.0, + 1194.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1195.0, + 277.0, + 1195.0, + 277.0, + 1223.0, + 249.0, + 1223.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1190.0, + 1407.0, + 1190.0, + 1407.0, + 1224.0, + 323.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1227.0, + 277.0, + 1227.0, + 277.0, + 1252.0, + 249.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1220.0, + 1406.0, + 1220.0, + 1406.0, + 1254.0, + 323.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1256.0, + 279.0, + 1256.0, + 279.0, + 1284.0, + 247.0, + 1284.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1248.0, + 1407.0, + 1248.0, + 1407.0, + 1286.0, + 322.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1288.0, + 277.0, + 1288.0, + 277.0, + 1313.0, + 249.0, + 1313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1280.0, + 1407.0, + 1280.0, + 1407.0, + 1317.0, + 322.0, + 1317.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1316.0, + 277.0, + 1316.0, + 277.0, + 1345.0, + 247.0, + 1345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1312.0, + 1406.0, + 1312.0, + 1406.0, + 1346.0, + 323.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1347.0, + 277.0, + 1347.0, + 277.0, + 1374.0, + 249.0, + 1374.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 323.0, + 1342.0, + 1407.0, + 1342.0, + 1407.0, + 1376.0, + 323.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 1377.0, + 279.0, + 1377.0, + 279.0, + 1404.0, + 249.0, + 1404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 322.0, + 1370.0, + 586.0, + 1370.0, + 586.0, + 1408.0, + 322.0, + 1408.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 1, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 248, + 198, + 1406, + 198, + 1406, + 497, + 248, + 497 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 240, + 738, + 1406, + 738, + 1406, + 1020, + 240, + 1020 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 242, + 1851, + 1405, + 1851, + 1405, + 2010, + 242, + 2010 + ], + "score": 0.975 + }, + { + "category_id": 1, + "poly": [ + 241, + 1223, + 1405, + 1223, + 1405, + 1378, + 241, + 1378 + ], + "score": 0.973 + }, + { + "category_id": 1, + "poly": [ + 241, + 1677, + 1405, + 1677, + 1405, + 1773, + 241, + 1773 + ], + "score": 0.971 + }, + { + "category_id": 1, + "poly": [ + 243, + 1487, + 1404, + 1487, + 1404, + 1584, + 243, + 1584 + ], + "score": 0.971 + }, + { + "category_id": 8, + "poly": [ + 626, + 1040, + 1074, + 1040, + 1074, + 1130, + 626, + 1130 + ], + "score": 0.959 + }, + { + "category_id": 8, + "poly": [ + 618, + 1395, + 1079, + 1395, + 1079, + 1465, + 618, + 1465 + ], + "score": 0.954 + }, + { + "category_id": 1, + "poly": [ + 245, + 1146, + 1402, + 1146, + 1402, + 1212, + 245, + 1212 + ], + "score": 0.942 + }, + { + "category_id": 8, + "poly": [ + 518, + 1791, + 1180, + 1791, + 1180, + 1833, + 518, + 1833 + ], + "score": 0.939 + }, + { + "category_id": 1, + "poly": [ + 296, + 611, + 1308, + 611, + 1308, + 645, + 296, + 645 + ], + "score": 0.917 + }, + { + "category_id": 9, + "poly": [ + 1345, + 1067, + 1400, + 1067, + 1400, + 1099, + 1345, + 1099 + ], + "score": 0.906 + }, + { + "category_id": 0, + "poly": [ + 275, + 1622, + 623, + 1622, + 623, + 1656, + 275, + 1656 + ], + "score": 0.905 + }, + { + "category_id": 9, + "poly": [ + 1345, + 1796, + 1400, + 1796, + 1400, + 1827, + 1345, + 1827 + ], + "score": 0.901 + }, + { + "category_id": 0, + "poly": [ + 293, + 682, + 669, + 682, + 669, + 715, + 293, + 715 + ], + "score": 0.895 + }, + { + "category_id": 0, + "poly": [ + 288, + 541, + 543, + 541, + 543, + 579, + 288, + 579 + ], + "score": 0.889 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.644 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.494 + }, + { + "category_id": 14, + "poly": [ + 624, + 1036, + 1073, + 1036, + 1073, + 1130, + 624, + 1130 + ], + "score": 0.95, + "latex": "R _ { T } = \\mathbb { E } \\bigg [ \\sum _ { t = 1 } ^ { T } \\big ( \\widehat { r } ( \\mathbf { x } _ { s _ { t } , a _ { t } ^ { * } } ) - \\widehat { r } ( \\mathbf { x } _ { s _ { t } , a _ { t } } ) \\big ) \\bigg ] ," + }, + { + "category_id": 13, + "poly": [ + 299, + 335, + 436, + 335, + 436, + 369, + 299, + 369 + ], + "score": 0.94, + "latex": "\\{ C _ { i } \\} _ { i = 0 , 1 , \\ldots }" + }, + { + "category_id": 13, + "poly": [ + 1214, + 427, + 1318, + 427, + 1318, + 467, + 1214, + 467 + ], + "score": 0.94, + "latex": "{ \\widetilde { O } } ( f ( T ) )" + }, + { + "category_id": 14, + "poly": [ + 618, + 1393, + 1078, + 1393, + 1078, + 1466, + 618, + 1466 + ], + "score": 0.94, + "latex": "a _ { t } = \\underset { a \\in [ K ] } { \\operatorname { a r g m a x } } \\left. \\mathbf { x } _ { t , a } ^ { \\top } \\pmb { \\theta } _ { t } + \\alpha _ { t } \\Vert \\mathbf { x } _ { t , a } \\Vert _ { \\mathbf { A } _ { t } ^ { - 1 } } \\right. ," + }, + { + "category_id": 13, + "poly": [ + 662, + 300, + 864, + 300, + 864, + 339, + 662, + 339 + ], + "score": 0.93, + "latex": "\\| \\mathbf { x } \\| _ { \\mathbf { A } } = \\sqrt { \\mathbf { x } ^ { \\top } \\mathbf { A } \\mathbf { x } }" + }, + { + "category_id": 13, + "poly": [ + 360, + 1975, + 585, + 1975, + 585, + 2009, + 360, + 2009 + ], + "score": 0.93, + "latex": "f ( \\mathbf { x } ; \\pmb { \\theta } ^ { * } , \\mathbf { w } ) = f ( \\mathbf { x } )" + }, + { + "category_id": 13, + "poly": [ + 348, + 1283, + 549, + 1283, + 549, + 1320, + 348, + 1320 + ], + "score": 0.93, + "latex": "r ( \\mathbf { x } _ { s , a } ) \\stackrel { * } { = } \\mathbf { x } _ { s , a } ^ { \\top } \\pmb { \\theta } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 646, + 432, + 749, + 432, + 749, + 467, + 646, + 467 + ], + "score": 0.93, + "latex": "O ( f ( T ) )" + }, + { + "category_id": 13, + "poly": [ + 1289, + 1179, + 1390, + 1179, + 1390, + 1212, + 1289, + 1212 + ], + "score": 0.92, + "latex": "A = [ K ]" + }, + { + "category_id": 13, + "poly": [ + 805, + 740, + 905, + 740, + 905, + 774, + 805, + 774 + ], + "score": 0.92, + "latex": "( S , A , r )" + }, + { + "category_id": 13, + "poly": [ + 943, + 198, + 1116, + 198, + 1116, + 236, + 943, + 236 + ], + "score": 0.92, + "latex": "\\| \\mathbf { x } \\| _ { 2 } = \\sqrt { \\mathbf { x } ^ { \\top } \\mathbf { x } }" + }, + { + "category_id": 13, + "poly": [ + 800, + 432, + 901, + 432, + 901, + 467, + 800, + 467 + ], + "score": 0.92, + "latex": "\\Omega ( f ( T ) )" + }, + { + "category_id": 13, + "poly": [ + 370, + 1147, + 850, + 1147, + 850, + 1183, + 370, + 1183 + ], + "score": 0.92, + "latex": "a _ { t } ^ { * } \\in \\operatorname { a r g m a x } _ { a \\in \\mathcal { A } } \\{ r ( \\mathbf { x } _ { s _ { t } , a } ) = \\mathbb { E } [ \\widehat { r } ( \\mathbf { x } _ { s _ { t } , a } ) ] \\}" + }, + { + "category_id": 13, + "poly": [ + 1113, + 236, + 1194, + 236, + 1194, + 269, + 1113, + 269 + ], + "score": 0.92, + "latex": "\\| \\mathbf { W } \\| _ { F }" + }, + { + "category_id": 13, + "poly": [ + 736, + 1853, + 838, + 1853, + 838, + 1883, + 736, + 1883 + ], + "score": 0.92, + "latex": "\\pmb { \\theta } ^ { * } \\in \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 1010, + 925, + 1181, + 925, + 1181, + 959, + 1010, + 959 + ], + "score": 0.92, + "latex": "( s , a ) \\in S \\times A" + }, + { + "category_id": 13, + "poly": [ + 658, + 863, + 736, + 863, + 736, + 892, + 658, + 892 + ], + "score": 0.92, + "latex": "s _ { t } \\in S" + }, + { + "category_id": 13, + "poly": [ + 298, + 1914, + 524, + 1914, + 524, + 1948, + 298, + 1948 + ], + "score": 0.91, + "latex": "{ \\bar { \\sigma } } ( x ) = \\operatorname* { m a x } \\{ 0 , x \\}" + }, + { + "category_id": 13, + "poly": [ + 938, + 1283, + 1040, + 1283, + 1040, + 1314, + 938, + 1314 + ], + "score": 0.91, + "latex": "\\pmb { \\theta } ^ { * } \\in \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 1253, + 832, + 1397, + 832, + 1397, + 862, + 1253, + 862 + ], + "score": 0.91, + "latex": "t = 1 , \\dots , T" + }, + { + "category_id": 13, + "poly": [ + 371, + 923, + 488, + 923, + 488, + 959, + 371, + 959 + ], + "score": 0.91, + "latex": "\\mathbf { x } _ { s , a } \\in \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 1263, + 267, + 1352, + 267, + 1352, + 299, + 1263, + 299 + ], + "score": 0.91, + "latex": "\\mathbf { x } \\in \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 1111, + 892, + 1397, + 892, + 1397, + 925, + 1111, + 925 + ], + "score": 0.91, + "latex": "\\widehat { r } _ { t } ( \\mathbf { x } _ { s , a _ { t } } ) = r ( \\mathbf { x } _ { s , a _ { t } } ) + \\xi _ { t }" + }, + { + "category_id": 13, + "poly": [ + 533, + 1680, + 589, + 1680, + 589, + 1712, + 533, + 1712 + ], + "score": 0.91, + "latex": "f ( \\mathbf { x } )" + }, + { + "category_id": 13, + "poly": [ + 397, + 234, + 489, + 234, + 489, + 266, + 397, + 266 + ], + "score": 0.91, + "latex": "\\mathbf { x } \\in \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 1070, + 1677, + 1157, + 1677, + 1157, + 1708, + 1070, + 1708 + ], + "score": 0.91, + "latex": "\\mathbf { x } \\in \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 298, + 433, + 358, + 433, + 358, + 467, + 298, + 467 + ], + "score": 0.91, + "latex": "f ( T )" + }, + { + "category_id": 13, + "poly": [ + 625, + 1742, + 794, + 1742, + 794, + 1772, + 625, + 1772 + ], + "score": 0.91, + "latex": "m _ { 0 } = m _ { L } = d" + }, + { + "category_id": 13, + "poly": [ + 330, + 396, + 465, + 396, + 465, + 431, + 330, + 431 + ], + "score": 0.91, + "latex": "\\{ C _ { i } \\} _ { i = 0 , 1 , \\ldots }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1745, + 573, + 1745, + 573, + 1773, + 297, + 1773 + ], + "score": 0.91, + "latex": "m _ { 1 } = . . . = m _ { L - 1 } = m" + }, + { + "category_id": 13, + "poly": [ + 1255, + 1711, + 1397, + 1711, + 1397, + 1742, + 1255, + 1742 + ], + "score": 0.91, + "latex": "l = 1 , \\ldots , L" + }, + { + "category_id": 13, + "poly": [ + 988, + 268, + 1113, + 268, + 1113, + 299, + 988, + 299 + ], + "score": 0.91, + "latex": "\\mathbf { A } \\in \\mathbb { R } ^ { d \\times d }" + }, + { + "category_id": 13, + "poly": [ + 574, + 1916, + 658, + 1916, + 658, + 1943, + 574, + 1943 + ], + "score": 0.9, + "latex": "x \\in \\mathbb { R }" + }, + { + "category_id": 13, + "poly": [ + 534, + 1710, + 727, + 1710, + 727, + 1741, + 534, + 1741 + ], + "score": 0.9, + "latex": "\\mathbf { W } _ { l } \\in \\mathbb { R } ^ { m _ { l } \\times m _ { l - 1 } }" + }, + { + "category_id": 13, + "poly": [ + 982, + 237, + 1057, + 237, + 1057, + 269, + 982, + 269 + ], + "score": 0.9, + "latex": "\\lVert \\mathbf { W } \\rVert _ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 1183, + 1492, + 1266, + 1492, + 1266, + 1521, + 1183, + 1521 + ], + "score": 0.9, + "latex": "\\lambda > 0" + }, + { + "category_id": 13, + "poly": [ + 658, + 237, + 805, + 237, + 805, + 266, + 658, + 266 + ], + "score": 0.9, + "latex": "\\mathbf { W } \\in \\mathbb { R } ^ { m \\times n }" + }, + { + "category_id": 13, + "poly": [ + 813, + 1913, + 1240, + 1913, + 1240, + 1948, + 813, + 1948 + ], + "score": 0.9, + "latex": "\\mathbf { w } = ( \\mathrm { v e c } ( \\mathbf { W } _ { 1 } ) ^ { \\top } , \\ldots , \\mathrm { v e c } ( \\mathbf { W } _ { L } ) ^ { \\top } ) ^ { \\top }" + }, + { + "category_id": 13, + "poly": [ + 1341, + 801, + 1403, + 801, + 1403, + 833, + 1341, + 833 + ], + "score": 0.9, + "latex": "( 0 , 1 )" + }, + { + "category_id": 13, + "poly": [ + 296, + 1184, + 356, + 1184, + 356, + 1213, + 296, + 1213 + ], + "score": 0.9, + "latex": "\\mathbf { x } _ { s _ { t } , a }" + }, + { + "category_id": 13, + "poly": [ + 381, + 892, + 464, + 892, + 464, + 922, + 381, + 922 + ], + "score": 0.9, + "latex": "a _ { t } \\in \\mathcal A" + }, + { + "category_id": 13, + "poly": [ + 894, + 1526, + 976, + 1526, + 976, + 1552, + 894, + 1552 + ], + "score": 0.88, + "latex": "\\alpha _ { t } > 0" + }, + { + "category_id": 13, + "poly": [ + 718, + 1486, + 1046, + 1486, + 1046, + 1527, + 718, + 1527 + ], + "score": 0.88, + "latex": "\\begin{array} { r } { \\mathbf { A } _ { t } \\ = \\ \\lambda \\mathbf { I } + \\sum _ { i = 1 } ^ { t } \\mathbf { x } _ { i , a _ { i } } \\mathbf { x } _ { i , a _ { i } } ^ { \\top } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1244, + 1155, + 1292, + 1155, + 1292, + 1182, + 1244, + 1182 + ], + "score": 0.88, + "latex": "\\mathbf { x } _ { t , a }" + }, + { + "category_id": 13, + "poly": [ + 497, + 203, + 530, + 203, + 530, + 236, + 497, + 236 + ], + "score": 0.88, + "latex": "[ k ]" + }, + { + "category_id": 13, + "poly": [ + 1240, + 928, + 1263, + 928, + 1263, + 958, + 1240, + 958 + ], + "score": 0.87, + "latex": "\\xi _ { t }" + }, + { + "category_id": 14, + "poly": [ + 520, + 1789, + 1179, + 1789, + 1179, + 1832, + 520, + 1832 + ], + "score": 0.87, + "latex": "f ( \\mathbf { x } ) = \\sqrt { m } \\pmb { \\theta } ^ { \\ast \\top } \\sigma _ { L } ( \\mathbf { W } _ { L } \\sigma _ { L - 1 } ( \\mathbf { W } _ { L - 1 } \\cdot \\cdot \\cdot \\sigma _ { 1 } ( \\mathbf { W } _ { 1 } \\mathbf { x } ) \\cdot \\cdot \\cdot ) ) ," + }, + { + "category_id": 13, + "poly": [ + 375, + 1493, + 404, + 1493, + 404, + 1522, + 375, + 1522 + ], + "score": 0.87, + "latex": "\\theta _ { t }" + }, + { + "category_id": 13, + "poly": [ + 642, + 1889, + 971, + 1889, + 971, + 1915, + 642, + 1915 + ], + "score": 0.86, + "latex": "\\sigma _ { 1 } = \\sigma _ { 2 } = . . . = \\sigma _ { L } = \\sigma" + }, + { + "category_id": 13, + "poly": [ + 373, + 1861, + 401, + 1861, + 401, + 1885, + 373, + 1885 + ], + "score": 0.84, + "latex": "\\sigma _ { l }" + }, + { + "category_id": 13, + "poly": [ + 1349, + 1977, + 1368, + 1977, + 1368, + 2008, + 1349, + 2008 + ], + "score": 0.84, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 746, + 467, + 768, + 467, + 768, + 492, + 746, + 492 + ], + "score": 0.84, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 989, + 742, + 1011, + 742, + 1011, + 768, + 989, + 768 + ], + "score": 0.83, + "latex": "s" + }, + { + "category_id": 13, + "poly": [ + 976, + 832, + 1000, + 832, + 1000, + 858, + 976, + 858 + ], + "score": 0.82, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 889, + 1742, + 910, + 1742, + 910, + 1768, + 889, + 1768 + ], + "score": 0.82, + "latex": "L" + }, + { + "category_id": 13, + "poly": [ + 697, + 802, + 722, + 802, + 722, + 829, + 697, + 829 + ], + "score": 0.82, + "latex": "\\mathcal { A }" + }, + { + "category_id": 13, + "poly": [ + 1376, + 958, + 1401, + 958, + 1401, + 984, + 1376, + 984 + ], + "score": 0.82, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 1213, + 1681, + 1235, + 1681, + 1235, + 1707, + 1213, + 1707 + ], + "score": 0.81, + "latex": "L" + }, + { + "category_id": 13, + "poly": [ + 1312, + 742, + 1337, + 742, + 1337, + 768, + 1312, + 768 + ], + "score": 0.81, + "latex": "\\mathcal { A }" + }, + { + "category_id": 13, + "poly": [ + 624, + 802, + 647, + 802, + 647, + 829, + 624, + 829 + ], + "score": 0.81, + "latex": "s" + }, + { + "category_id": 13, + "poly": [ + 1206, + 398, + 1230, + 398, + 1230, + 424, + 1206, + 424 + ], + "score": 0.8, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 705, + 202, + 825, + 202, + 825, + 237, + 705, + 237 + ], + "score": 0.8, + "latex": "\\{ 1 , \\ldots , k \\}" + }, + { + "category_id": 13, + "poly": [ + 815, + 1184, + 828, + 1184, + 828, + 1205, + 815, + 1205 + ], + "score": 0.78, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 836, + 201, + 928, + 201, + 928, + 235, + 836, + 235 + ], + "score": 0.75, + "latex": "k \\in \\mathbb { N } ^ { + }" + }, + { + "category_id": 13, + "poly": [ + 1063, + 1712, + 1075, + 1712, + 1075, + 1738, + 1063, + 1738 + ], + "score": 0.75, + "latex": "l" + }, + { + "category_id": 13, + "poly": [ + 1104, + 807, + 1122, + 807, + 1122, + 828, + 1104, + 828 + ], + "score": 0.74, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 453, + 1349, + 467, + 1349, + 467, + 1373, + 453, + 1373 + ], + "score": 0.74, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 563, + 776, + 580, + 776, + 580, + 798, + 563, + 798 + ], + "score": 0.73, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 297, + 1261, + 314, + 1261, + 314, + 1282, + 297, + 1282 + ], + "score": 0.7, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 669, + 1492, + 700, + 1492, + 700, + 1520, + 669, + 1520 + ], + "score": 0.7, + "latex": "\\pmb { \\theta } ^ { * }" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1619.0, + 627.0, + 1619.0, + 627.0, + 1660.0, + 294.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 680.0, + 671.0, + 680.0, + 671.0, + 717.0, + 294.0, + 717.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 538.0, + 546.0, + 538.0, + 546.0, + 584.0, + 292.0, + 584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 209.0, + 277.0, + 209.0, + 277.0, + 234.0, + 249.0, + 234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 199.0, + 496.0, + 199.0, + 496.0, + 238.0, + 292.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 531.0, + 199.0, + 704.0, + 199.0, + 704.0, + 238.0, + 531.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 826.0, + 199.0, + 835.0, + 199.0, + 835.0, + 238.0, + 826.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 929.0, + 199.0, + 942.0, + 199.0, + 942.0, + 238.0, + 929.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1117.0, + 199.0, + 1408.0, + 199.0, + 1408.0, + 238.0, + 1117.0, + 238.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 242.0, + 278.0, + 242.0, + 278.0, + 268.0, + 248.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 229.0, + 396.0, + 229.0, + 396.0, + 274.0, + 287.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 229.0, + 657.0, + 229.0, + 657.0, + 274.0, + 490.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 806.0, + 229.0, + 981.0, + 229.0, + 981.0, + 274.0, + 806.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1058.0, + 229.0, + 1112.0, + 229.0, + 1112.0, + 274.0, + 1058.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 229.0, + 1409.0, + 229.0, + 1409.0, + 274.0, + 1195.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 275.0, + 278.0, + 275.0, + 278.0, + 301.0, + 248.0, + 301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 264.0, + 987.0, + 264.0, + 987.0, + 304.0, + 290.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1114.0, + 264.0, + 1262.0, + 264.0, + 1262.0, + 304.0, + 1114.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1353.0, + 264.0, + 1407.0, + 264.0, + 1407.0, + 304.0, + 1353.0, + 304.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 310.0, + 278.0, + 310.0, + 278.0, + 336.0, + 248.0, + 336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 303.0, + 661.0, + 303.0, + 661.0, + 340.0, + 292.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 865.0, + 303.0, + 1407.0, + 303.0, + 1407.0, + 340.0, + 865.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 341.0, + 278.0, + 341.0, + 278.0, + 367.0, + 248.0, + 367.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 333.0, + 298.0, + 333.0, + 298.0, + 371.0, + 294.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 437.0, + 333.0, + 1406.0, + 333.0, + 1406.0, + 371.0, + 437.0, + 371.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 371.0, + 278.0, + 371.0, + 278.0, + 397.0, + 248.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 364.0, + 1405.0, + 364.0, + 1405.0, + 399.0, + 292.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 400.0, + 279.0, + 400.0, + 279.0, + 430.0, + 246.0, + 430.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 395.0, + 329.0, + 395.0, + 329.0, + 434.0, + 291.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 466.0, + 395.0, + 1205.0, + 395.0, + 1205.0, + 434.0, + 466.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1231.0, + 395.0, + 1408.0, + 395.0, + 1408.0, + 434.0, + 1231.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 436.0, + 279.0, + 436.0, + 279.0, + 466.0, + 246.0, + 466.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 432.0, + 297.0, + 432.0, + 297.0, + 467.0, + 294.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 359.0, + 432.0, + 645.0, + 432.0, + 645.0, + 467.0, + 359.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 750.0, + 432.0, + 799.0, + 432.0, + 799.0, + 467.0, + 750.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 902.0, + 432.0, + 1213.0, + 432.0, + 1213.0, + 467.0, + 902.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1319.0, + 432.0, + 1404.0, + 432.0, + 1404.0, + 467.0, + 1319.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 471.0, + 278.0, + 471.0, + 278.0, + 496.0, + 248.0, + 496.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 465.0, + 745.0, + 465.0, + 745.0, + 497.0, + 291.0, + 497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 769.0, + 465.0, + 780.0, + 465.0, + 780.0, + 497.0, + 769.0, + 497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 746.0, + 277.0, + 746.0, + 277.0, + 772.0, + 241.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 737.0, + 804.0, + 737.0, + 804.0, + 776.0, + 290.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 906.0, + 737.0, + 988.0, + 737.0, + 988.0, + 776.0, + 906.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1012.0, + 737.0, + 1311.0, + 737.0, + 1311.0, + 776.0, + 1012.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1338.0, + 737.0, + 1406.0, + 737.0, + 1406.0, + 776.0, + 1338.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 777.0, + 277.0, + 777.0, + 277.0, + 802.0, + 240.0, + 802.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 772.0, + 562.0, + 772.0, + 562.0, + 805.0, + 294.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 581.0, + 772.0, + 1409.0, + 772.0, + 1409.0, + 805.0, + 581.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 805.0, + 278.0, + 805.0, + 278.0, + 834.0, + 239.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 799.0, + 623.0, + 799.0, + 623.0, + 835.0, + 292.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 648.0, + 799.0, + 696.0, + 799.0, + 696.0, + 835.0, + 648.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 723.0, + 799.0, + 1103.0, + 799.0, + 1103.0, + 835.0, + 723.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1123.0, + 799.0, + 1340.0, + 799.0, + 1340.0, + 835.0, + 1123.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1404.0, + 799.0, + 1408.0, + 799.0, + 1408.0, + 835.0, + 1404.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 835.0, + 280.0, + 835.0, + 280.0, + 864.0, + 239.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 826.0, + 975.0, + 826.0, + 975.0, + 867.0, + 292.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1001.0, + 826.0, + 1252.0, + 826.0, + 1252.0, + 867.0, + 1001.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 826.0, + 1409.0, + 826.0, + 1409.0, + 867.0, + 1398.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 866.0, + 278.0, + 866.0, + 278.0, + 895.0, + 239.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 862.0, + 657.0, + 862.0, + 657.0, + 895.0, + 294.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 737.0, + 862.0, + 1406.0, + 862.0, + 1406.0, + 895.0, + 737.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 896.0, + 280.0, + 896.0, + 280.0, + 925.0, + 239.0, + 925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 887.0, + 380.0, + 887.0, + 380.0, + 931.0, + 290.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 465.0, + 887.0, + 1110.0, + 887.0, + 1110.0, + 931.0, + 465.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 887.0, + 1410.0, + 887.0, + 1410.0, + 931.0, + 1398.0, + 931.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 932.0, + 277.0, + 932.0, + 277.0, + 957.0, + 240.0, + 957.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 921.0, + 370.0, + 921.0, + 370.0, + 963.0, + 293.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 489.0, + 921.0, + 1009.0, + 921.0, + 1009.0, + 963.0, + 489.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1182.0, + 921.0, + 1239.0, + 921.0, + 1239.0, + 963.0, + 1182.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1264.0, + 921.0, + 1407.0, + 921.0, + 1407.0, + 963.0, + 1264.0, + 963.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 963.0, + 278.0, + 963.0, + 278.0, + 988.0, + 240.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 955.0, + 1375.0, + 955.0, + 1375.0, + 990.0, + 292.0, + 990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1402.0, + 955.0, + 1406.0, + 955.0, + 1406.0, + 990.0, + 1402.0, + 990.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 993.0, + 277.0, + 993.0, + 277.0, + 1018.0, + 240.0, + 1018.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 988.0, + 1013.0, + 988.0, + 1013.0, + 1021.0, + 294.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1858.0, + 280.0, + 1858.0, + 280.0, + 1887.0, + 239.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1854.0, + 372.0, + 1854.0, + 372.0, + 1887.0, + 293.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 1854.0, + 735.0, + 1854.0, + 735.0, + 1887.0, + 402.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 1854.0, + 1405.0, + 1854.0, + 1405.0, + 1887.0, + 839.0, + 1887.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1890.0, + 278.0, + 1890.0, + 278.0, + 1916.0, + 241.0, + 1916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1881.0, + 641.0, + 1881.0, + 641.0, + 1921.0, + 291.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 972.0, + 1881.0, + 1409.0, + 1881.0, + 1409.0, + 1921.0, + 972.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1920.0, + 278.0, + 1920.0, + 278.0, + 1945.0, + 241.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1910.0, + 297.0, + 1910.0, + 297.0, + 1951.0, + 291.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 525.0, + 1910.0, + 573.0, + 1910.0, + 573.0, + 1951.0, + 525.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 659.0, + 1910.0, + 812.0, + 1910.0, + 812.0, + 1951.0, + 659.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1241.0, + 1910.0, + 1408.0, + 1910.0, + 1408.0, + 1951.0, + 1241.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1949.0, + 280.0, + 1949.0, + 280.0, + 1979.0, + 239.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1943.0, + 1406.0, + 1943.0, + 1406.0, + 1979.0, + 292.0, + 1979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1982.0, + 279.0, + 1982.0, + 279.0, + 2006.0, + 241.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1974.0, + 359.0, + 1974.0, + 359.0, + 2011.0, + 292.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 586.0, + 1974.0, + 1348.0, + 1974.0, + 1348.0, + 2011.0, + 586.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1369.0, + 1974.0, + 1406.0, + 1974.0, + 1406.0, + 2011.0, + 1369.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1230.0, + 278.0, + 1230.0, + 278.0, + 1255.0, + 241.0, + 1255.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1225.0, + 1404.0, + 1225.0, + 1404.0, + 1258.0, + 294.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1261.0, + 278.0, + 1261.0, + 278.0, + 1286.0, + 241.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1255.0, + 296.0, + 1255.0, + 296.0, + 1289.0, + 292.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 315.0, + 1255.0, + 1405.0, + 1255.0, + 1405.0, + 1289.0, + 315.0, + 1289.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1291.0, + 278.0, + 1291.0, + 278.0, + 1316.0, + 241.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1278.0, + 347.0, + 1278.0, + 347.0, + 1328.0, + 290.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 550.0, + 1278.0, + 937.0, + 1278.0, + 937.0, + 1328.0, + 550.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1041.0, + 1278.0, + 1410.0, + 1278.0, + 1410.0, + 1328.0, + 1041.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1319.0, + 279.0, + 1319.0, + 279.0, + 1348.0, + 238.0, + 1348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1311.0, + 1409.0, + 1311.0, + 1409.0, + 1354.0, + 291.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1351.0, + 279.0, + 1351.0, + 279.0, + 1376.0, + 241.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1340.0, + 452.0, + 1340.0, + 452.0, + 1384.0, + 290.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 468.0, + 1340.0, + 1030.0, + 1340.0, + 1030.0, + 1384.0, + 468.0, + 1384.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1677.0, + 532.0, + 1677.0, + 532.0, + 1713.0, + 234.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 590.0, + 1677.0, + 1069.0, + 1677.0, + 1069.0, + 1713.0, + 590.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1158.0, + 1677.0, + 1212.0, + 1677.0, + 1212.0, + 1713.0, + 1158.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1236.0, + 1677.0, + 1408.0, + 1677.0, + 1408.0, + 1713.0, + 1236.0, + 1713.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 242.0, + 1718.0, + 279.0, + 1718.0, + 279.0, + 1740.0, + 242.0, + 1740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 1705.0, + 533.0, + 1705.0, + 533.0, + 1746.0, + 286.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 728.0, + 1705.0, + 1062.0, + 1705.0, + 1062.0, + 1746.0, + 728.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1076.0, + 1705.0, + 1254.0, + 1705.0, + 1254.0, + 1746.0, + 1076.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 1705.0, + 1409.0, + 1705.0, + 1409.0, + 1746.0, + 1398.0, + 1746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 242.0, + 1748.0, + 279.0, + 1748.0, + 279.0, + 1772.0, + 242.0, + 1772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1740.0, + 296.0, + 1740.0, + 296.0, + 1775.0, + 287.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 574.0, + 1740.0, + 624.0, + 1740.0, + 624.0, + 1775.0, + 574.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 795.0, + 1740.0, + 888.0, + 1740.0, + 888.0, + 1775.0, + 795.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 911.0, + 1740.0, + 1381.0, + 1740.0, + 1381.0, + 1775.0, + 911.0, + 1775.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 243.0, + 1497.0, + 280.0, + 1497.0, + 280.0, + 1521.0, + 243.0, + 1521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 242.0, + 1528.0, + 278.0, + 1528.0, + 278.0, + 1550.0, + 242.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1521.0, + 893.0, + 1521.0, + 893.0, + 1555.0, + 290.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 977.0, + 1521.0, + 1405.0, + 1521.0, + 1405.0, + 1555.0, + 977.0, + 1555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 242.0, + 1558.0, + 281.0, + 1558.0, + 281.0, + 1585.0, + 242.0, + 1585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1552.0, + 617.0, + 1552.0, + 617.0, + 1586.0, + 290.0, + 1586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 280.0, + 1478.0, + 1413.0, + 1478.0, + 1413.0, + 1536.0, + 280.0, + 1536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1144.0, + 369.0, + 1144.0, + 369.0, + 1185.0, + 234.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 851.0, + 1144.0, + 1243.0, + 1144.0, + 1243.0, + 1185.0, + 851.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1293.0, + 1144.0, + 1406.0, + 1144.0, + 1406.0, + 1185.0, + 1293.0, + 1185.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1176.0, + 295.0, + 1176.0, + 295.0, + 1215.0, + 234.0, + 1215.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1176.0, + 814.0, + 1176.0, + 814.0, + 1215.0, + 357.0, + 1215.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 829.0, + 1176.0, + 1288.0, + 1176.0, + 1288.0, + 1215.0, + 829.0, + 1215.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1391.0, + 1176.0, + 1404.0, + 1176.0, + 1404.0, + 1215.0, + 1391.0, + 1215.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 608.0, + 1311.0, + 608.0, + 1311.0, + 649.0, + 292.0, + 649.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 2, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 238, + 685, + 1406, + 685, + 1406, + 1083, + 238, + 1083 + ], + "score": 0.979 + }, + { + "category_id": 1, + "poly": [ + 240, + 488, + 1406, + 488, + 1406, + 672, + 240, + 672 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 242, + 1913, + 1405, + 1913, + 1405, + 2010, + 242, + 2010 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 241, + 1265, + 1405, + 1265, + 1405, + 1389, + 241, + 1389 + ], + "score": 0.961 + }, + { + "category_id": 8, + "poly": [ + 576, + 1406, + 1123, + 1406, + 1123, + 1493, + 576, + 1493 + ], + "score": 0.948 + }, + { + "category_id": 1, + "poly": [ + 243, + 200, + 1405, + 200, + 1405, + 267, + 243, + 267 + ], + "score": 0.943 + }, + { + "category_id": 0, + "poly": [ + 254, + 416, + 1008, + 416, + 1008, + 458, + 254, + 458 + ], + "score": 0.941 + }, + { + "category_id": 1, + "poly": [ + 244, + 1722, + 1403, + 1722, + 1403, + 1788, + 244, + 1788 + ], + "score": 0.938 + }, + { + "category_id": 8, + "poly": [ + 467, + 1174, + 1227, + 1174, + 1227, + 1244, + 467, + 1244 + ], + "score": 0.938 + }, + { + "category_id": 8, + "poly": [ + 553, + 285, + 1139, + 285, + 1139, + 324, + 553, + 324 + ], + "score": 0.937 + }, + { + "category_id": 8, + "poly": [ + 714, + 1562, + 982, + 1562, + 982, + 1602, + 714, + 1602 + ], + "score": 0.934 + }, + { + "category_id": 8, + "poly": [ + 770, + 1805, + 926, + 1805, + 926, + 1844, + 770, + 1844 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 241, + 1511, + 1363, + 1511, + 1363, + 1546, + 241, + 1546 + ], + "score": 0.926 + }, + { + "category_id": 1, + "poly": [ + 240, + 1094, + 1410, + 1094, + 1410, + 1158, + 240, + 1158 + ], + "score": 0.923 + }, + { + "category_id": 9, + "poly": [ + 1345, + 1565, + 1400, + 1565, + 1400, + 1596, + 1345, + 1596 + ], + "score": 0.907 + }, + { + "category_id": 9, + "poly": [ + 1344, + 1431, + 1400, + 1431, + 1400, + 1462, + 1344, + 1462 + ], + "score": 0.904 + }, + { + "category_id": 1, + "poly": [ + 262, + 340, + 1319, + 340, + 1319, + 375, + 262, + 375 + ], + "score": 0.903 + }, + { + "category_id": 9, + "poly": [ + 1345, + 1810, + 1400, + 1810, + 1400, + 1840, + 1345, + 1840 + ], + "score": 0.899 + }, + { + "category_id": 9, + "poly": [ + 1345, + 288, + 1400, + 288, + 1400, + 319, + 1345, + 319 + ], + "score": 0.895 + }, + { + "category_id": 9, + "poly": [ + 1345, + 1195, + 1400, + 1195, + 1400, + 1227, + 1345, + 1227 + ], + "score": 0.884 + }, + { + "category_id": 1, + "poly": [ + 243, + 1616, + 1403, + 1616, + 1403, + 1712, + 243, + 1712 + ], + "score": 0.86 + }, + { + "category_id": 1, + "poly": [ + 256, + 1866, + 704, + 1866, + 704, + 1906, + 256, + 1906 + ], + "score": 0.84 + }, + { + "category_id": 2, + "poly": [ + 840, + 2061, + 859, + 2061, + 859, + 2085, + 840, + 2085 + ], + "score": 0.762 + }, + { + "category_id": 13, + "poly": [ + 389, + 807, + 585, + 807, + 585, + 841, + 389, + 841 + ], + "score": 0.95, + "latex": "\\bar { r ( \\cdot ) } = \\langle \\theta ^ { * } , \\psi ( \\cdot ) \\rangle" + }, + { + "category_id": 14, + "poly": [ + 576, + 1400, + 1120, + 1400, + 1120, + 1495, + 576, + 1495 + ], + "score": 0.95, + "latex": "\\mathbf { A } _ { t } = \\lambda \\mathbf { I } + \\sum _ { i = 1 } ^ { t } \\phi ( \\mathbf { x } _ { i , a _ { i } } ; \\mathbf { w } _ { i - 1 } ) \\phi ( \\mathbf { x } _ { i , a _ { i } } ; \\mathbf { w } _ { i - 1 } ) ^ { \\top } ," + }, + { + "category_id": 13, + "poly": [ + 1085, + 1512, + 1243, + 1512, + 1243, + 1547, + 1085, + 1547 + ], + "score": 0.94, + "latex": "\\widehat { r } _ { t } : = \\widehat { r } ( \\mathbf { x } _ { t , a _ { t } } )" + }, + { + "category_id": 13, + "poly": [ + 1072, + 1267, + 1167, + 1267, + 1167, + 1299, + 1072, + 1299 + ], + "score": 0.94, + "latex": "\\phi ( \\mathbf { x } ; \\mathbf { w } )" + }, + { + "category_id": 14, + "poly": [ + 470, + 1173, + 1226, + 1173, + 1226, + 1247, + 470, + 1247 + ], + "score": 0.94, + "latex": "a _ { t } = \\underset { k \\in [ K ] } { \\operatorname { a r g m a x } } \\Big \\{ \\langle \\phi ( \\mathbf { x } _ { t , k } ; \\mathbf { w } _ { t - 1 } ) , \\theta _ { t - 1 } \\rangle + \\alpha _ { t } \\| \\phi ( \\mathbf { x } _ { t , k } ; \\mathbf { w } _ { t - 1 } ) \\| _ { \\mathbf { A } _ { t - 1 } ^ { - 1 } } \\Big \\} ," + }, + { + "category_id": 13, + "poly": [ + 516, + 234, + 612, + 234, + 612, + 267, + 516, + 267 + ], + "score": 0.93, + "latex": "\\phi ( \\mathbf { x } ; \\mathbf { w } )" + }, + { + "category_id": 13, + "poly": [ + 407, + 341, + 502, + 341, + 502, + 375, + 407, + 375 + ], + "score": 0.93, + "latex": "\\phi ( \\mathbf { x } ; \\mathbf { w } )" + }, + { + "category_id": 13, + "poly": [ + 947, + 202, + 1215, + 202, + 1215, + 237, + 947, + 237 + ], + "score": 0.92, + "latex": "p = ( L - 2 ) m ^ { 2 } + 2 m d ." + }, + { + "category_id": 14, + "poly": [ + 716, + 1562, + 982, + 1562, + 982, + 1601, + 716, + 1601 + ], + "score": 0.92, + "latex": "\\widehat { r } ( \\mathbf { x } _ { t , k } ) = r ( \\mathbf { x } _ { t , k } ) + \\xi _ { t } ," + }, + { + "category_id": 14, + "poly": [ + 770, + 1802, + 927, + 1802, + 927, + 1844, + 770, + 1844 + ], + "score": 0.92, + "latex": "\\pmb { \\theta } _ { t } = \\mathbf { A } _ { t } ^ { - 1 } \\mathbf { b } _ { t } ," + }, + { + "category_id": 13, + "poly": [ + 763, + 1678, + 817, + 1678, + 817, + 1712, + 763, + 1712 + ], + "score": 0.92, + "latex": "\\widehat { r } ( \\mathbf { x } )" + }, + { + "category_id": 13, + "poly": [ + 669, + 806, + 767, + 806, + 767, + 836, + 669, + 836 + ], + "score": 0.92, + "latex": "\\pmb { \\theta } ^ { * } \\in \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 1101, + 808, + 1155, + 808, + 1155, + 840, + 1101, + 840 + ], + "score": 0.91, + "latex": "\\psi ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 1139, + 1095, + 1397, + 1095, + 1397, + 1129, + 1139, + 1129 + ], + "score": 0.91, + "latex": "\\mathcal { X } _ { t } = \\{ \\mathbf { x } _ { t , 1 } , . . . , \\mathbf { x } _ { t , K } \\}" + }, + { + "category_id": 13, + "poly": [ + 786, + 1914, + 882, + 1914, + 882, + 1947, + 786, + 1947 + ], + "score": 0.91, + "latex": "\\phi ( \\cdot ; { \\mathbf w } _ { t } )" + }, + { + "category_id": 13, + "poly": [ + 1295, + 1299, + 1376, + 1299, + 1376, + 1327, + 1295, + 1327 + ], + "score": 0.91, + "latex": "\\alpha _ { t } > 0" + }, + { + "category_id": 13, + "poly": [ + 1182, + 1617, + 1227, + 1617, + 1227, + 1651, + 1182, + 1651 + ], + "score": 0.91, + "latex": "r ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 370, + 1268, + 426, + 1268, + 426, + 1299, + 370, + 1299 + ], + "score": 0.91, + "latex": "\\pmb { \\theta } _ { t - 1 }" + }, + { + "category_id": 13, + "poly": [ + 1135, + 747, + 1180, + 747, + 1180, + 781, + 1135, + 781 + ], + "score": 0.91, + "latex": "r ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 1120, + 1945, + 1401, + 1945, + 1401, + 1979, + 1120, + 1979 + ], + "score": 0.91, + "latex": "\\{ ( q - 1 ) H + 1 , \\ldots , q H \\}" + }, + { + "category_id": 13, + "poly": [ + 735, + 1978, + 827, + 1978, + 827, + 2008, + 735, + 2008 + ], + "score": 0.91, + "latex": "t = H q" + }, + { + "category_id": 14, + "poly": [ + 559, + 282, + 1139, + 282, + 1139, + 324, + 559, + 324 + ], + "score": 0.91, + "latex": "\\boldsymbol { \\phi } ( \\mathbf { x } ; \\mathbf { w } ) = \\sqrt { m } \\sigma ( \\mathbf { W } _ { L } \\sigma ( \\mathbf { W } _ { L - 1 } \\cdot \\cdot \\cdot \\sigma ( \\mathbf { W } _ { 1 } \\mathbf { x } ) \\cdot \\cdot \\cdot \\cdot ) ) ." + }, + { + "category_id": 13, + "poly": [ + 371, + 1864, + 696, + 1864, + 696, + 1905, + 371, + 1905 + ], + "score": 0.9, + "latex": "\\begin{array} { r } { \\mathbf { b } _ { t } = \\sum _ { i = 1 } ^ { t } \\widehat { r } _ { i } \\phi ( \\mathbf { x } _ { i , a _ { i } } ; \\mathbf { w } _ { i - 1 } ) } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1055, + 1619, + 1129, + 1619, + 1129, + 1646, + 1055, + 1646 + ], + "score": 0.9, + "latex": "\\nu > 0" + }, + { + "category_id": 13, + "poly": [ + 392, + 1950, + 683, + 1950, + 683, + 1981, + 392, + 1981 + ], + "score": 0.9, + "latex": "\\mathbf { w } _ { ( q - 1 ) H + 1 } = . . . = \\mathbf { w } _ { q H }" + }, + { + "category_id": 13, + "poly": [ + 941, + 1329, + 977, + 1329, + 977, + 1358, + 941, + 1358 + ], + "score": 0.89, + "latex": "{ \\bf A } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 930, + 1977, + 1072, + 1977, + 1072, + 2008, + 930, + 2008 + ], + "score": 0.89, + "latex": "q = 1 , 2 , \\ldots" + }, + { + "category_id": 13, + "poly": [ + 394, + 610, + 439, + 610, + 439, + 644, + 394, + 644 + ], + "score": 0.89, + "latex": "r ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 344, + 1513, + 414, + 1513, + 414, + 1541, + 344, + 1541 + ], + "score": 0.88, + "latex": "\\lambda > 0" + }, + { + "category_id": 13, + "poly": [ + 973, + 1724, + 1002, + 1724, + 1002, + 1755, + 973, + 1755 + ], + "score": 0.88, + "latex": "\\theta _ { t }" + }, + { + "category_id": 13, + "poly": [ + 373, + 1619, + 398, + 1619, + 398, + 1650, + 373, + 1650 + ], + "score": 0.88, + "latex": "\\xi _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1045, + 1649, + 1071, + 1649, + 1071, + 1679, + 1045, + 1679 + ], + "score": 0.87, + "latex": "\\widehat { r _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 296, + 1302, + 360, + 1302, + 360, + 1329, + 296, + 1329 + ], + "score": 0.87, + "latex": "\\mathbf { w } _ { t - 1 }" + }, + { + "category_id": 13, + "poly": [ + 601, + 1724, + 627, + 1724, + 627, + 1755, + 601, + 1755 + ], + "score": 0.86, + "latex": "\\widehat { r _ { t } }" + }, + { + "category_id": 13, + "poly": [ + 401, + 1753, + 427, + 1753, + 427, + 1782, + 401, + 1782 + ], + "score": 0.86, + "latex": "\\ell ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 645, + 1130, + 673, + 1130, + 673, + 1157, + 645, + 1157 + ], + "score": 0.85, + "latex": "a _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1241, + 236, + 1259, + 236, + 1259, + 266, + 1241, + 266 + ], + "score": 0.85, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 1186, + 1916, + 1214, + 1916, + 1214, + 1942, + 1186, + 1942 + ], + "score": 0.83, + "latex": "H" + }, + { + "category_id": 13, + "poly": [ + 836, + 235, + 857, + 235, + 857, + 261, + 836, + 261 + ], + "score": 0.83, + "latex": "L" + }, + { + "category_id": 13, + "poly": [ + 628, + 1518, + 654, + 1518, + 654, + 1544, + 628, + 1544 + ], + "score": 0.83, + "latex": "a _ { t }" + }, + { + "category_id": 13, + "poly": [ + 691, + 211, + 708, + 211, + 708, + 235, + 691, + 235 + ], + "score": 0.81, + "latex": "p" + }, + { + "category_id": 13, + "poly": [ + 727, + 1948, + 868, + 1948, + 868, + 1977, + 727, + 1977 + ], + "score": 0.8, + "latex": "q = 1 , 2 , \\ldots" + }, + { + "category_id": 13, + "poly": [ + 610, + 1623, + 629, + 1623, + 629, + 1645, + 610, + 1645 + ], + "score": 0.78, + "latex": "\\nu" + }, + { + "category_id": 13, + "poly": [ + 544, + 1979, + 572, + 1979, + 572, + 2003, + 544, + 2003 + ], + "score": 0.76, + "latex": "H" + }, + { + "category_id": 13, + "poly": [ + 365, + 1681, + 378, + 1681, + 378, + 1706, + 365, + 1706 + ], + "score": 0.75, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 547, + 1099, + 561, + 1099, + 561, + 1123, + 547, + 1123 + ], + "score": 0.74, + "latex": "t" + }, + { + "category_id": 13, + "poly": [ + 1344, + 1684, + 1364, + 1684, + 1364, + 1706, + 1344, + 1706 + ], + "score": 0.67, + "latex": "\\mathbf { x }" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 413.0, + 1012.0, + 413.0, + 1012.0, + 469.0, + 240.0, + 469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 863.0, + 2058.0, + 863.0, + 2091.0, + 838.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 691.0, + 278.0, + 691.0, + 278.0, + 719.0, + 240.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 687.0, + 1404.0, + 687.0, + 1404.0, + 719.0, + 294.0, + 719.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 723.0, + 276.0, + 723.0, + 276.0, + 748.0, + 240.0, + 748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 718.0, + 1404.0, + 718.0, + 1404.0, + 750.0, + 295.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 754.0, + 276.0, + 754.0, + 276.0, + 779.0, + 240.0, + 779.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 747.0, + 1134.0, + 747.0, + 1134.0, + 784.0, + 292.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1181.0, + 747.0, + 1406.0, + 747.0, + 1406.0, + 784.0, + 1181.0, + 784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 784.0, + 278.0, + 784.0, + 278.0, + 809.0, + 240.0, + 809.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 775.0, + 1408.0, + 775.0, + 1408.0, + 814.0, + 294.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 814.0, + 278.0, + 814.0, + 278.0, + 839.0, + 240.0, + 839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 804.0, + 388.0, + 804.0, + 388.0, + 844.0, + 292.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 586.0, + 804.0, + 668.0, + 804.0, + 668.0, + 844.0, + 586.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 768.0, + 804.0, + 1100.0, + 804.0, + 1100.0, + 844.0, + 768.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1156.0, + 804.0, + 1408.0, + 804.0, + 1408.0, + 844.0, + 1156.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 842.0, + 279.0, + 842.0, + 279.0, + 871.0, + 238.0, + 871.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 836.0, + 1407.0, + 836.0, + 1407.0, + 872.0, + 291.0, + 872.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 875.0, + 278.0, + 875.0, + 278.0, + 900.0, + 240.0, + 900.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 870.0, + 1407.0, + 870.0, + 1407.0, + 902.0, + 294.0, + 902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 903.0, + 279.0, + 903.0, + 279.0, + 932.0, + 238.0, + 932.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 896.0, + 1406.0, + 896.0, + 1406.0, + 935.0, + 292.0, + 935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 933.0, + 278.0, + 933.0, + 278.0, + 962.0, + 238.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 929.0, + 1406.0, + 929.0, + 1406.0, + 962.0, + 294.0, + 962.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 964.0, + 278.0, + 964.0, + 278.0, + 993.0, + 238.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 958.0, + 1405.0, + 958.0, + 1405.0, + 993.0, + 292.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 994.0, + 278.0, + 994.0, + 278.0, + 1023.0, + 238.0, + 1023.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 989.0, + 1406.0, + 989.0, + 1406.0, + 1024.0, + 292.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1024.0, + 278.0, + 1024.0, + 278.0, + 1053.0, + 240.0, + 1053.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1019.0, + 1408.0, + 1019.0, + 1408.0, + 1055.0, + 294.0, + 1055.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1058.0, + 278.0, + 1058.0, + 278.0, + 1082.0, + 241.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1051.0, + 687.0, + 1051.0, + 687.0, + 1084.0, + 294.0, + 1084.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 494.0, + 278.0, + 494.0, + 278.0, + 522.0, + 240.0, + 522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 490.0, + 1405.0, + 490.0, + 1405.0, + 522.0, + 294.0, + 522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 524.0, + 280.0, + 524.0, + 280.0, + 552.0, + 240.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 520.0, + 1405.0, + 520.0, + 1405.0, + 552.0, + 296.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 555.0, + 278.0, + 555.0, + 278.0, + 583.0, + 240.0, + 583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 551.0, + 1405.0, + 551.0, + 1405.0, + 583.0, + 294.0, + 583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 585.0, + 280.0, + 585.0, + 280.0, + 613.0, + 240.0, + 613.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 579.0, + 1406.0, + 579.0, + 1406.0, + 617.0, + 293.0, + 617.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 616.0, + 280.0, + 616.0, + 280.0, + 643.0, + 239.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 608.0, + 393.0, + 608.0, + 393.0, + 646.0, + 292.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 440.0, + 608.0, + 1409.0, + 608.0, + 1409.0, + 646.0, + 440.0, + 646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 647.0, + 278.0, + 647.0, + 278.0, + 671.0, + 240.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 638.0, + 1299.0, + 638.0, + 1299.0, + 674.0, + 293.0, + 674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1918.0, + 280.0, + 1918.0, + 280.0, + 1945.0, + 241.0, + 1945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 285.0, + 1912.0, + 785.0, + 1912.0, + 785.0, + 1950.0, + 285.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 1912.0, + 1185.0, + 1912.0, + 1185.0, + 1950.0, + 883.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1215.0, + 1912.0, + 1408.0, + 1912.0, + 1408.0, + 1950.0, + 1215.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 242.0, + 1953.0, + 279.0, + 1953.0, + 279.0, + 1975.0, + 242.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 1944.0, + 391.0, + 1944.0, + 391.0, + 1981.0, + 287.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 684.0, + 1944.0, + 726.0, + 1944.0, + 726.0, + 1981.0, + 684.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 869.0, + 1944.0, + 1119.0, + 1944.0, + 1119.0, + 1981.0, + 869.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1974.0, + 543.0, + 1974.0, + 543.0, + 2012.0, + 235.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 573.0, + 1974.0, + 734.0, + 1974.0, + 734.0, + 2012.0, + 573.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 828.0, + 1974.0, + 929.0, + 1974.0, + 929.0, + 2012.0, + 828.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 1974.0, + 1406.0, + 1974.0, + 1406.0, + 2012.0, + 1073.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1272.0, + 278.0, + 1272.0, + 278.0, + 1298.0, + 240.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1265.0, + 369.0, + 1265.0, + 369.0, + 1300.0, + 294.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 427.0, + 1265.0, + 1071.0, + 1265.0, + 1071.0, + 1300.0, + 427.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1168.0, + 1265.0, + 1406.0, + 1265.0, + 1406.0, + 1300.0, + 1168.0, + 1300.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1302.0, + 278.0, + 1302.0, + 278.0, + 1328.0, + 240.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1296.0, + 295.0, + 1296.0, + 295.0, + 1330.0, + 292.0, + 1330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 1296.0, + 1294.0, + 1296.0, + 1294.0, + 1330.0, + 361.0, + 1330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1377.0, + 1296.0, + 1406.0, + 1296.0, + 1406.0, + 1330.0, + 1377.0, + 1330.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1333.0, + 278.0, + 1333.0, + 278.0, + 1359.0, + 240.0, + 1359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1327.0, + 940.0, + 1327.0, + 940.0, + 1362.0, + 295.0, + 1362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 978.0, + 1327.0, + 1404.0, + 1327.0, + 1404.0, + 1362.0, + 978.0, + 1362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1362.0, + 278.0, + 1362.0, + 278.0, + 1389.0, + 240.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1353.0, + 541.0, + 1353.0, + 541.0, + 1393.0, + 293.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 201.0, + 690.0, + 201.0, + 690.0, + 237.0, + 235.0, + 237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 709.0, + 201.0, + 946.0, + 201.0, + 946.0, + 237.0, + 709.0, + 237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1216.0, + 201.0, + 1405.0, + 201.0, + 1405.0, + 237.0, + 1216.0, + 237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 231.0, + 515.0, + 231.0, + 515.0, + 270.0, + 234.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 613.0, + 231.0, + 835.0, + 231.0, + 835.0, + 270.0, + 613.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 858.0, + 231.0, + 1240.0, + 231.0, + 1240.0, + 270.0, + 858.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 231.0, + 1271.0, + 231.0, + 1271.0, + 270.0, + 1260.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1719.0, + 600.0, + 1719.0, + 600.0, + 1763.0, + 235.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 628.0, + 1719.0, + 972.0, + 1719.0, + 972.0, + 1763.0, + 628.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1003.0, + 1719.0, + 1406.0, + 1719.0, + 1406.0, + 1763.0, + 1003.0, + 1763.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1751.0, + 400.0, + 1751.0, + 400.0, + 1791.0, + 236.0, + 1791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 428.0, + 1751.0, + 1406.0, + 1751.0, + 1406.0, + 1791.0, + 428.0, + 1791.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1507.0, + 343.0, + 1507.0, + 343.0, + 1551.0, + 237.0, + 1551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 415.0, + 1507.0, + 627.0, + 1507.0, + 627.0, + 1551.0, + 415.0, + 1551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 655.0, + 1507.0, + 1084.0, + 1507.0, + 1084.0, + 1551.0, + 655.0, + 1551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1244.0, + 1507.0, + 1367.0, + 1507.0, + 1367.0, + 1551.0, + 1244.0, + 1551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1101.0, + 279.0, + 1101.0, + 279.0, + 1125.0, + 238.0, + 1125.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1090.0, + 546.0, + 1090.0, + 546.0, + 1132.0, + 289.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 1090.0, + 1138.0, + 1090.0, + 1138.0, + 1132.0, + 562.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 1090.0, + 1410.0, + 1090.0, + 1410.0, + 1132.0, + 1398.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1125.0, + 644.0, + 1125.0, + 644.0, + 1158.0, + 235.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 1125.0, + 1289.0, + 1125.0, + 1289.0, + 1158.0, + 674.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 253.0, + 339.0, + 406.0, + 339.0, + 406.0, + 378.0, + 253.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 503.0, + 339.0, + 1318.0, + 339.0, + 1318.0, + 378.0, + 503.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 242.0, + 1622.0, + 278.0, + 1622.0, + 278.0, + 1649.0, + 242.0, + 1649.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1615.0, + 372.0, + 1615.0, + 372.0, + 1653.0, + 291.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 399.0, + 1615.0, + 609.0, + 1615.0, + 609.0, + 1653.0, + 399.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 630.0, + 1615.0, + 1054.0, + 1615.0, + 1054.0, + 1653.0, + 630.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1130.0, + 1615.0, + 1181.0, + 1615.0, + 1181.0, + 1653.0, + 1130.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1228.0, + 1615.0, + 1406.0, + 1615.0, + 1406.0, + 1653.0, + 1228.0, + 1653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 243.0, + 1657.0, + 277.0, + 1657.0, + 277.0, + 1679.0, + 243.0, + 1679.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1648.0, + 1044.0, + 1648.0, + 1044.0, + 1682.0, + 290.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1072.0, + 1648.0, + 1404.0, + 1648.0, + 1404.0, + 1682.0, + 1072.0, + 1682.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 242.0, + 1685.0, + 280.0, + 1685.0, + 280.0, + 1712.0, + 242.0, + 1712.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1680.0, + 364.0, + 1680.0, + 364.0, + 1714.0, + 291.0, + 1714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 379.0, + 1680.0, + 762.0, + 1680.0, + 762.0, + 1714.0, + 379.0, + 1714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 818.0, + 1680.0, + 1343.0, + 1680.0, + 1343.0, + 1714.0, + 818.0, + 1714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1365.0, + 1680.0, + 1376.0, + 1680.0, + 1376.0, + 1714.0, + 1365.0, + 1714.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 243.0, + 1857.0, + 370.0, + 1857.0, + 370.0, + 1912.0, + 243.0, + 1912.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 697.0, + 1857.0, + 708.0, + 1857.0, + 708.0, + 1912.0, + 697.0, + 1912.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 3, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 285, + 952, + 1405, + 952, + 1405, + 1353, + 285, + 1353 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 241, + 376, + 1403, + 376, + 1403, + 507, + 241, + 507 + ], + "score": 0.966 + }, + { + "category_id": 1, + "poly": [ + 241, + 563, + 1404, + 563, + 1404, + 734, + 241, + 734 + ], + "score": 0.965 + }, + { + "category_id": 1, + "poly": [ + 243, + 1562, + 1405, + 1562, + 1405, + 1838, + 243, + 1838 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 243, + 1365, + 1405, + 1365, + 1405, + 1549, + 243, + 1549 + ], + "score": 0.958 + }, + { + "category_id": 1, + "poly": [ + 241, + 744, + 1405, + 744, + 1405, + 942, + 241, + 942 + ], + "score": 0.957 + }, + { + "category_id": 8, + "poly": [ + 635, + 274, + 1063, + 274, + 1063, + 365, + 635, + 365 + ], + "score": 0.955 + }, + { + "category_id": 1, + "poly": [ + 245, + 202, + 1404, + 202, + 1404, + 266, + 245, + 266 + ], + "score": 0.937 + }, + { + "category_id": 1, + "poly": [ + 250, + 1944, + 1404, + 1944, + 1404, + 2008, + 250, + 2008 + ], + "score": 0.937 + }, + { + "category_id": 8, + "poly": [ + 638, + 511, + 1058, + 511, + 1058, + 556, + 638, + 556 + ], + "score": 0.936 + }, + { + "category_id": 9, + "poly": [ + 1344, + 303, + 1400, + 303, + 1400, + 334, + 1344, + 334 + ], + "score": 0.895 + }, + { + "category_id": 0, + "poly": [ + 259, + 1877, + 538, + 1877, + 538, + 1915, + 259, + 1915 + ], + "score": 0.882 + }, + { + "category_id": 2, + "poly": [ + 841, + 2061, + 859, + 2061, + 859, + 2085, + 841, + 2085 + ], + "score": 0.726 + }, + { + "category_id": 14, + "poly": [ + 634, + 269, + 1064, + 269, + 1064, + 367, + 634, + 367 + ], + "score": 0.95, + "latex": "\\mathcal { L } _ { q } ( \\mathbf { w } ) = \\sum _ { i = 1 } ^ { q H } \\big ( \\pmb { \\theta } _ { i } ^ { \\top } \\phi ( \\mathbf { x } _ { i , a _ { i } } ; \\mathbf { w } ) - \\widehat { r } _ { i } \\big ) ^ { 2 } ." + }, + { + "category_id": 14, + "poly": [ + 638, + 511, + 1059, + 511, + 1059, + 557, + 638, + 557 + ], + "score": 0.93, + "latex": "\\mathbf { w } _ { q } ^ { ( s ) } = \\mathbf { w } _ { q } ^ { ( s - 1 ) } - \\eta _ { q } \\nabla _ { \\mathbf { w } } \\mathcal { L } _ { q } \\big ( \\mathbf { w } ^ { ( s - 1 ) } \\big ) ." + }, + { + "category_id": 13, + "poly": [ + 1097, + 907, + 1223, + 907, + 1223, + 941, + 1097, + 941 + ], + "score": 0.93, + "latex": "N ( 0 , 2 / m )" + }, + { + "category_id": 13, + "poly": [ + 1067, + 856, + 1194, + 856, + 1194, + 890, + 1067, + 890 + ], + "score": 0.93, + "latex": "N ( 0 , 4 / m )" + }, + { + "category_id": 13, + "poly": [ + 298, + 1073, + 567, + 1073, + 567, + 1111, + 298, + 1111 + ], + "score": 0.92, + "latex": "\\left\\{ \\mathbf { x } _ { t , k } \\right\\} _ { t \\in [ T ] , k \\in [ K ] } \\subset \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 1259, + 1074, + 1346, + 1074, + 1346, + 1107, + 1259, + 1107 + ], + "score": 0.92, + "latex": "\\phi ( \\cdot ; { \\mathbf { w } } )" + }, + { + "category_id": 13, + "poly": [ + 1183, + 806, + 1397, + 806, + 1397, + 840, + 1183, + 840 + ], + "score": 0.92, + "latex": "l \\in \\{ 1 , \\ldots , L - 1 \\}" + }, + { + "category_id": 13, + "poly": [ + 372, + 837, + 575, + 837, + 575, + 910, + 372, + 910 + ], + "score": 0.92, + "latex": "\\mathbf { W } _ { l } = \\left[ \\begin{array} { c c } { \\mathbf { W } } & { \\mathbf { 0 } } \\\\ { \\mathbf { 0 } } & { \\mathbf { W } } \\end{array} \\right]" + }, + { + "category_id": 13, + "poly": [ + 1295, + 602, + 1353, + 602, + 1353, + 643, + 1295, + 643 + ], + "score": 0.92, + "latex": "\\mathbf { w } _ { q } ^ { ( n ) }" + }, + { + "category_id": 13, + "poly": [ + 524, + 1458, + 611, + 1458, + 611, + 1490, + 524, + 1490 + ], + "score": 0.92, + "latex": "\\phi ( \\cdot ; { \\mathbf { w } } )" + }, + { + "category_id": 13, + "poly": [ + 577, + 563, + 727, + 563, + 727, + 605, + 577, + 605 + ], + "score": 0.92, + "latex": "\\mathbf { w } _ { q } ^ { ( 0 ) } = \\mathbf { w } ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 1116, + 372, + 1400, + 372, + 1400, + 419, + 1116, + 419 + ], + "score": 0.92, + "latex": "\\{ \\mathbf { x } _ { i , a _ { i } } , \\widehat { r } _ { i } , \\pmb { \\theta } _ { i } \\} _ { i = ( q - 1 ) H + 1 } ^ { q H }" + }, + { + "category_id": 13, + "poly": [ + 594, + 672, + 654, + 672, + 654, + 703, + 594, + 703 + ], + "score": 0.91, + "latex": "\\mathcal { X } _ { t + 1 }" + }, + { + "category_id": 13, + "poly": [ + 1247, + 1163, + 1312, + 1163, + 1312, + 1194, + 1247, + 1194 + ], + "score": 0.91, + "latex": "\\dot { \\mathbb { R } } ^ { \\widetilde { p } + d }" + }, + { + "category_id": 13, + "poly": [ + 1063, + 1291, + 1132, + 1291, + 1132, + 1320, + 1063, + 1320 + ], + "score": 0.91, + "latex": "d \\times d" + }, + { + "category_id": 13, + "poly": [ + 339, + 572, + 487, + 572, + 487, + 603, + 339, + 603 + ], + "score": 0.91, + "latex": "s = 1 , \\ldots , n" + }, + { + "category_id": 13, + "poly": [ + 767, + 644, + 852, + 644, + 852, + 674, + 767, + 674 + ], + "score": 0.9, + "latex": "\\mathbf { w } _ { H q + 1 }" + }, + { + "category_id": 13, + "poly": [ + 434, + 1227, + 656, + 1227, + 656, + 1260, + 434, + 1260 + ], + "score": 0.89, + "latex": "\\mathbf { Z } _ { t } \\in \\mathbb { R } ^ { ( \\widetilde { p } + d ) \\times ( \\widetilde { p } + d ) }" + }, + { + "category_id": 13, + "poly": [ + 423, + 1262, + 455, + 1262, + 455, + 1291, + 423, + 1291 + ], + "score": 0.88, + "latex": "\\mathbf { Z } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 1216, + 1230, + 1254, + 1230, + 1254, + 1260, + 1216, + 1260 + ], + "score": 0.88, + "latex": "{ \\bf A } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 704, + 1292, + 741, + 1292, + 741, + 1321, + 704, + 1321 + ], + "score": 0.88, + "latex": "{ \\bf A } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1194, + 633, + 1194, + 633, + 1228, + 297, + 1228 + ], + "score": 0.87, + "latex": "\\ddot { \\tilde { p } } = m + m d + ( L \\dot { \\bar { \\mathbf { \\alpha } } } 1 ) m ^ { 2 }" + }, + { + "category_id": 13, + "poly": [ + 295, + 908, + 347, + 908, + 347, + 938, + 295, + 938 + ], + "score": 0.87, + "latex": "\\mathbf { W } _ { L }" + }, + { + "category_id": 13, + "poly": [ + 1106, + 445, + 1133, + 445, + 1133, + 475, + 1106, + 475 + ], + "score": 0.87, + "latex": "\\theta _ { i }" + }, + { + "category_id": 13, + "poly": [ + 959, + 417, + 993, + 417, + 993, + 445, + 959, + 445 + ], + "score": 0.86, + "latex": "\\mathbf { w } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 928, + 1715, + 961, + 1715, + 961, + 1743, + 928, + 1743 + ], + "score": 0.86, + "latex": "\\pmb { \\theta } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 543, + 1493, + 574, + 1493, + 574, + 1518, + 543, + 1518 + ], + "score": 0.86, + "latex": "\\mathbf { x } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 489, + 907, + 604, + 907, + 604, + 941, + 489, + 941 + ], + "score": 0.86, + "latex": "\\begin{array} { r l } { [ \\mathbf { V } } & { { } - \\mathbf { V } ] } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 1282, + 1746, + 1313, + 1746, + 1313, + 1774, + 1282, + 1774 + ], + "score": 0.84, + "latex": "\\pmb { \\theta } ^ { * }" + }, + { + "category_id": 13, + "poly": [ + 692, + 606, + 746, + 606, + 746, + 638, + 692, + 638 + ], + "score": 0.82, + "latex": "\\mathbf { w } ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 395, + 418, + 412, + 418, + 412, + 445, + 395, + 445 + ], + "score": 0.8, + "latex": "q" + }, + { + "category_id": 13, + "poly": [ + 842, + 908, + 869, + 908, + 869, + 936, + 842, + 936 + ], + "score": 0.49, + "latex": "\\mathbf { V }" + }, + { + "category_id": 13, + "poly": [ + 809, + 857, + 844, + 857, + 844, + 884, + 809, + 884 + ], + "score": 0.25, + "latex": "\\mathbf { W }" + }, + { + "category_id": 15, + "poly": [ + 248.0, + 1873.0, + 543.0, + 1873.0, + 543.0, + 1922.0, + 248.0, + 1922.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 838.0, + 2058.0, + 862.0, + 2058.0, + 862.0, + 2092.0, + 838.0, + 2092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 950.0, + 1409.0, + 950.0, + 1409.0, + 988.0, + 295.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 980.0, + 1405.0, + 980.0, + 1405.0, + 1019.0, + 292.0, + 1019.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1012.0, + 1407.0, + 1012.0, + 1407.0, + 1048.0, + 293.0, + 1048.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1044.0, + 1405.0, + 1044.0, + 1405.0, + 1077.0, + 295.0, + 1077.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1068.0, + 297.0, + 1068.0, + 297.0, + 1113.0, + 291.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 568.0, + 1068.0, + 1258.0, + 1068.0, + 1258.0, + 1113.0, + 568.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1347.0, + 1068.0, + 1410.0, + 1068.0, + 1410.0, + 1113.0, + 1347.0, + 1113.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1103.0, + 1408.0, + 1103.0, + 1408.0, + 1138.0, + 292.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1133.0, + 1408.0, + 1133.0, + 1408.0, + 1169.0, + 293.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1160.0, + 1246.0, + 1160.0, + 1246.0, + 1202.0, + 290.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1313.0, + 1160.0, + 1405.0, + 1160.0, + 1405.0, + 1202.0, + 1313.0, + 1202.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1193.0, + 296.0, + 1193.0, + 296.0, + 1232.0, + 290.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 634.0, + 1193.0, + 1407.0, + 1193.0, + 1407.0, + 1232.0, + 634.0, + 1232.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1218.0, + 433.0, + 1218.0, + 433.0, + 1271.0, + 290.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 1218.0, + 1215.0, + 1218.0, + 1215.0, + 1271.0, + 657.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1255.0, + 1218.0, + 1409.0, + 1218.0, + 1409.0, + 1271.0, + 1255.0, + 1271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1259.0, + 422.0, + 1259.0, + 422.0, + 1295.0, + 292.0, + 1295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 456.0, + 1259.0, + 1405.0, + 1259.0, + 1405.0, + 1295.0, + 456.0, + 1295.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1289.0, + 703.0, + 1289.0, + 703.0, + 1326.0, + 295.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 742.0, + 1289.0, + 1062.0, + 1289.0, + 1062.0, + 1326.0, + 742.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1133.0, + 1289.0, + 1405.0, + 1289.0, + 1405.0, + 1326.0, + 1133.0, + 1326.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1319.0, + 1407.0, + 1319.0, + 1407.0, + 1355.0, + 295.0, + 1355.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 337.0, + 394.0, + 337.0, + 394.0, + 442.0, + 240.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 413.0, + 337.0, + 958.0, + 337.0, + 958.0, + 442.0, + 413.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 337.0, + 1115.0, + 337.0, + 1115.0, + 442.0, + 994.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1401.0, + 337.0, + 1419.0, + 337.0, + 1419.0, + 442.0, + 1401.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 416.0, + 279.0, + 416.0, + 279.0, + 447.0, + 240.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 408.0, + 394.0, + 408.0, + 394.0, + 452.0, + 292.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 413.0, + 408.0, + 958.0, + 408.0, + 958.0, + 452.0, + 413.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 994.0, + 408.0, + 1406.0, + 408.0, + 1406.0, + 452.0, + 994.0, + 452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 447.0, + 280.0, + 447.0, + 280.0, + 478.0, + 238.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 444.0, + 1105.0, + 444.0, + 1105.0, + 480.0, + 295.0, + 480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1134.0, + 444.0, + 1403.0, + 444.0, + 1403.0, + 480.0, + 1134.0, + 480.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 479.0, + 278.0, + 479.0, + 278.0, + 506.0, + 241.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 472.0, + 789.0, + 472.0, + 789.0, + 515.0, + 291.0, + 515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 572.0, + 279.0, + 572.0, + 279.0, + 603.0, + 240.0, + 603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 552.0, + 338.0, + 552.0, + 338.0, + 616.0, + 286.0, + 616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 488.0, + 552.0, + 576.0, + 552.0, + 576.0, + 616.0, + 488.0, + 616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 728.0, + 552.0, + 1407.0, + 552.0, + 1407.0, + 616.0, + 728.0, + 616.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 612.0, + 279.0, + 612.0, + 279.0, + 643.0, + 238.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 595.0, + 691.0, + 595.0, + 691.0, + 653.0, + 288.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 747.0, + 595.0, + 1294.0, + 595.0, + 1294.0, + 653.0, + 747.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 595.0, + 1412.0, + 595.0, + 1412.0, + 653.0, + 1354.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 644.0, + 278.0, + 644.0, + 278.0, + 671.0, + 240.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 637.0, + 766.0, + 637.0, + 766.0, + 677.0, + 292.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 853.0, + 637.0, + 1405.0, + 637.0, + 1405.0, + 677.0, + 853.0, + 677.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 674.0, + 279.0, + 674.0, + 279.0, + 705.0, + 238.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 670.0, + 593.0, + 670.0, + 593.0, + 705.0, + 294.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 655.0, + 670.0, + 1404.0, + 670.0, + 1404.0, + 705.0, + 655.0, + 705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 706.0, + 279.0, + 706.0, + 279.0, + 732.0, + 240.0, + 732.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 698.0, + 1033.0, + 698.0, + 1033.0, + 736.0, + 292.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1569.0, + 277.0, + 1569.0, + 277.0, + 1594.0, + 239.0, + 1594.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1564.0, + 1405.0, + 1564.0, + 1405.0, + 1597.0, + 297.0, + 1597.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1598.0, + 278.0, + 1598.0, + 278.0, + 1627.0, + 238.0, + 1627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1594.0, + 1406.0, + 1594.0, + 1406.0, + 1627.0, + 293.0, + 1627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1630.0, + 276.0, + 1630.0, + 276.0, + 1654.0, + 239.0, + 1654.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1622.0, + 1406.0, + 1622.0, + 1406.0, + 1660.0, + 292.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1661.0, + 276.0, + 1661.0, + 276.0, + 1686.0, + 239.0, + 1686.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1653.0, + 1405.0, + 1653.0, + 1405.0, + 1690.0, + 294.0, + 1690.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1690.0, + 278.0, + 1690.0, + 278.0, + 1718.0, + 238.0, + 1718.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1684.0, + 1406.0, + 1684.0, + 1406.0, + 1721.0, + 294.0, + 1721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1720.0, + 278.0, + 1720.0, + 278.0, + 1748.0, + 238.0, + 1748.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1714.0, + 927.0, + 1714.0, + 927.0, + 1751.0, + 294.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 962.0, + 1714.0, + 1406.0, + 1714.0, + 1406.0, + 1751.0, + 962.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1750.0, + 277.0, + 1750.0, + 277.0, + 1778.0, + 238.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1746.0, + 1281.0, + 1746.0, + 1281.0, + 1778.0, + 293.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1314.0, + 1746.0, + 1406.0, + 1746.0, + 1406.0, + 1778.0, + 1314.0, + 1778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1780.0, + 278.0, + 1780.0, + 278.0, + 1808.0, + 238.0, + 1808.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1773.0, + 1406.0, + 1773.0, + 1406.0, + 1811.0, + 292.0, + 1811.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1812.0, + 277.0, + 1812.0, + 277.0, + 1837.0, + 239.0, + 1837.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1804.0, + 1254.0, + 1804.0, + 1254.0, + 1841.0, + 293.0, + 1841.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1371.0, + 278.0, + 1371.0, + 278.0, + 1399.0, + 239.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1367.0, + 1408.0, + 1367.0, + 1408.0, + 1399.0, + 294.0, + 1399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1401.0, + 277.0, + 1401.0, + 277.0, + 1429.0, + 238.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1397.0, + 1406.0, + 1397.0, + 1406.0, + 1429.0, + 297.0, + 1429.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1432.0, + 277.0, + 1432.0, + 277.0, + 1460.0, + 238.0, + 1460.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1425.0, + 1406.0, + 1425.0, + 1406.0, + 1461.0, + 290.0, + 1461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1462.0, + 278.0, + 1462.0, + 278.0, + 1490.0, + 238.0, + 1490.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1457.0, + 523.0, + 1457.0, + 523.0, + 1493.0, + 294.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 612.0, + 1457.0, + 1405.0, + 1457.0, + 1405.0, + 1493.0, + 612.0, + 1493.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1493.0, + 278.0, + 1493.0, + 278.0, + 1520.0, + 238.0, + 1520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1489.0, + 542.0, + 1489.0, + 542.0, + 1520.0, + 294.0, + 1520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1489.0, + 1405.0, + 1489.0, + 1405.0, + 1520.0, + 575.0, + 1520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1523.0, + 280.0, + 1523.0, + 280.0, + 1550.0, + 238.0, + 1550.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1517.0, + 1272.0, + 1517.0, + 1272.0, + 1552.0, + 290.0, + 1552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 751.0, + 277.0, + 751.0, + 277.0, + 776.0, + 240.0, + 776.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 743.0, + 1405.0, + 743.0, + 1405.0, + 778.0, + 294.0, + 778.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 782.0, + 278.0, + 782.0, + 278.0, + 807.0, + 240.0, + 807.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 774.0, + 1406.0, + 774.0, + 1406.0, + 812.0, + 291.0, + 812.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 812.0, + 278.0, + 812.0, + 278.0, + 838.0, + 240.0, + 838.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 803.0, + 1182.0, + 803.0, + 1182.0, + 843.0, + 292.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 803.0, + 1408.0, + 803.0, + 1408.0, + 843.0, + 1398.0, + 843.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 860.0, + 278.0, + 860.0, + 278.0, + 891.0, + 240.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 846.0, + 371.0, + 846.0, + 371.0, + 899.0, + 291.0, + 899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 577.0, + 854.0, + 808.0, + 854.0, + 808.0, + 892.0, + 577.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 845.0, + 854.0, + 1066.0, + 854.0, + 1066.0, + 892.0, + 845.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1195.0, + 854.0, + 1406.0, + 854.0, + 1406.0, + 892.0, + 1195.0, + 892.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 912.0, + 278.0, + 912.0, + 278.0, + 941.0, + 240.0, + 941.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 903.0, + 294.0, + 903.0, + 294.0, + 944.0, + 291.0, + 944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 903.0, + 488.0, + 903.0, + 488.0, + 944.0, + 348.0, + 944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 605.0, + 903.0, + 841.0, + 903.0, + 841.0, + 944.0, + 605.0, + 944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 870.0, + 903.0, + 1096.0, + 903.0, + 1096.0, + 944.0, + 870.0, + 944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1224.0, + 903.0, + 1398.0, + 903.0, + 1398.0, + 944.0, + 1224.0, + 944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 198.0, + 1407.0, + 198.0, + 1407.0, + 242.0, + 233.0, + 242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 234.0, + 566.0, + 234.0, + 566.0, + 266.0, + 236.0, + 266.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 1950.0, + 281.0, + 1950.0, + 281.0, + 1975.0, + 244.0, + 1975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1944.0, + 1404.0, + 1944.0, + 1404.0, + 1980.0, + 290.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1976.0, + 652.0, + 1976.0, + 652.0, + 2007.0, + 239.0, + 2007.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 4, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 238, + 1731, + 1405, + 1731, + 1405, + 2009, + 238, + 2009 + ], + "score": 0.968 + }, + { + "category_id": 8, + "poly": [ + 571, + 1586, + 1126, + 1586, + 1126, + 1663, + 571, + 1663 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 306, + 238, + 1406, + 238, + 1406, + 881, + 306, + 881 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 269, + 1225, + 1402, + 1225, + 1402, + 1296, + 269, + 1296 + ], + "score": 0.933 + }, + { + "category_id": 1, + "poly": [ + 306, + 963, + 1404, + 963, + 1404, + 1169, + 306, + 1169 + ], + "score": 0.925 + }, + { + "category_id": 1, + "poly": [ + 242, + 1677, + 661, + 1677, + 661, + 1715, + 242, + 1715 + ], + "score": 0.885 + }, + { + "category_id": 1, + "poly": [ + 243, + 1538, + 1024, + 1538, + 1024, + 1573, + 243, + 1573 + ], + "score": 0.883 + }, + { + "category_id": 1, + "poly": [ + 263, + 1497, + 1374, + 1497, + 1374, + 1533, + 263, + 1533 + ], + "score": 0.873 + }, + { + "category_id": 0, + "poly": [ + 297, + 922, + 1005, + 922, + 1005, + 954, + 297, + 954 + ], + "score": 0.846 + }, + { + "category_id": 2, + "poly": [ + 839, + 2061, + 859, + 2061, + 859, + 2086, + 839, + 2086 + ], + "score": 0.668 + }, + { + "category_id": 0, + "poly": [ + 298, + 201, + 1170, + 201, + 1170, + 235, + 298, + 235 + ], + "score": 0.529 + }, + { + "category_id": 1, + "poly": [ + 272, + 1422, + 1393, + 1422, + 1393, + 1487, + 272, + 1487 + ], + "score": 0.516 + }, + { + "category_id": 1, + "poly": [ + 264, + 1315, + 1405, + 1315, + 1405, + 1418, + 264, + 1418 + ], + "score": 0.355 + }, + { + "category_id": 2, + "poly": [ + 839, + 2062, + 859, + 2062, + 859, + 2086, + 839, + 2086 + ], + "score": 0.222 + }, + { + "category_id": 13, + "poly": [ + 1119, + 1418, + 1326, + 1418, + 1326, + 1457, + 1119, + 1457 + ], + "score": 0.93, + "latex": "\\phi ( \\mathbf { x } _ { i , k } ; \\mathbf { w } ^ { ( 0 ) } ) = \\mathbf { 0 }" + }, + { + "category_id": 13, + "poly": [ + 777, + 1379, + 1049, + 1379, + 1049, + 1422, + 777, + 1422 + ], + "score": 0.93, + "latex": "\\mathbf { x } _ { i , k } ^ { \\prime } = [ \\mathbf { x } _ { i , k } ^ { \\top } , \\mathbf { x } _ { i , k } ^ { \\top } ] ^ { \\top } / \\sqrt { 2 }" + }, + { + "category_id": 13, + "poly": [ + 983, + 306, + 1100, + 306, + 1100, + 341, + 983, + 341 + ], + "score": 0.93, + "latex": "N ( 0 , 1 / d )" + }, + { + "category_id": 13, + "poly": [ + 1007, + 1228, + 1151, + 1228, + 1151, + 1263, + 1007, + 1263 + ], + "score": 0.93, + "latex": "\\| \\mathbf { x } _ { i , k } \\| _ { 2 } = 1" + }, + { + "category_id": 14, + "poly": [ + 567, + 1584, + 1125, + 1584, + 1125, + 1664, + 567, + 1664 + ], + "score": 0.92, + "latex": "\\left\\| \\frac { \\partial \\phi } { \\partial \\mathbf { w } } ( \\mathbf { x } ; \\mathbf { w } _ { 0 } ) - \\frac { \\partial \\phi } { \\partial \\mathbf { w } } ( \\mathbf { x } ^ { \\prime } ; \\mathbf { w } _ { 0 } ) \\right\\| _ { 2 } \\leq \\ell _ { \\mathrm { L i p } } \\| \\mathbf { x } - \\mathbf { x } ^ { \\prime } \\| _ { 2 } ," + }, + { + "category_id": 13, + "poly": [ + 710, + 1228, + 809, + 1228, + 809, + 1262, + 710, + 1262 + ], + "score": 0.92, + "latex": "k \\in [ K ]" + }, + { + "category_id": 13, + "poly": [ + 715, + 1540, + 810, + 1540, + 810, + 1574, + 715, + 1574 + ], + "score": 0.92, + "latex": "\\ell _ { \\mathrm { L i p } } > 0" + }, + { + "category_id": 13, + "poly": [ + 430, + 1452, + 522, + 1452, + 522, + 1486, + 430, + 1486 + ], + "score": 0.92, + "latex": "k \\in [ K ]" + }, + { + "category_id": 13, + "poly": [ + 534, + 1317, + 673, + 1317, + 673, + 1351, + 534, + 1351 + ], + "score": 0.92, + "latex": "\\| \\mathbf { x } _ { i , k } \\| _ { 2 } = 1" + }, + { + "category_id": 13, + "poly": [ + 299, + 1258, + 532, + 1258, + 532, + 1296, + 299, + 1296 + ], + "score": 0.91, + "latex": "[ { \\bf { x } } _ { i , k } ] _ { j } \\stackrel { - } { = } [ { \\bf { x } } _ { j , k } ] _ { j + d / 2 }" + }, + { + "category_id": 13, + "poly": [ + 297, + 1453, + 379, + 1453, + 379, + 1486, + 297, + 1486 + ], + "score": 0.91, + "latex": "i \\in [ T ]" + }, + { + "category_id": 13, + "poly": [ + 386, + 372, + 532, + 372, + 532, + 403, + 386, + 403 + ], + "score": 0.91, + "latex": "t = 1 , \\dots , T" + }, + { + "category_id": 13, + "poly": [ + 474, + 270, + 634, + 270, + 634, + 306, + 474, + 306 + ], + "score": 0.91, + "latex": "\\{ \\alpha _ { t } > 0 \\} _ { t \\in [ T ] }" + }, + { + "category_id": 13, + "poly": [ + 565, + 960, + 710, + 960, + 710, + 1002, + 565, + 1002 + ], + "score": 0.9, + "latex": "\\mathbf { w } _ { q } ^ { ( 0 ) } = \\mathbf { w } ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 1215, + 1351, + 1263, + 1351, + 1263, + 1381, + 1215, + 1381 + ], + "score": 0.9, + "latex": "\\mathbf { x } _ { i , k }" + }, + { + "category_id": 13, + "poly": [ + 707, + 241, + 776, + 241, + 776, + 268, + 707, + 268 + ], + "score": 0.9, + "latex": "\\lambda > 0" + }, + { + "category_id": 13, + "poly": [ + 586, + 1229, + 656, + 1229, + 656, + 1260, + 586, + 1260 + ], + "score": 0.89, + "latex": "i \\geq 1" + }, + { + "category_id": 13, + "poly": [ + 629, + 402, + 818, + 402, + 818, + 436, + 629, + 436 + ], + "score": 0.88, + "latex": "\\left\\{ \\mathbf { x } _ { t , 1 } , \\ldots , \\mathbf { x } _ { t , K } \\right\\}" + }, + { + "category_id": 13, + "poly": [ + 682, + 338, + 810, + 338, + 810, + 372, + 682, + 372 + ], + "score": 0.88, + "latex": "\\mathbf { w } _ { 0 } = \\mathbf { w } ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 518, + 1944, + 553, + 1944, + 553, + 1973, + 518, + 1973 + ], + "score": 0.88, + "latex": "\\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 865, + 308, + 895, + 308, + 895, + 338, + 865, + 338 + ], + "score": 0.87, + "latex": "\\pmb { \\theta } _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 604, + 343, + 671, + 343, + 671, + 373, + 604, + 373 + ], + "score": 0.87, + "latex": "q = 1" + }, + { + "category_id": 13, + "poly": [ + 453, + 509, + 489, + 509, + 489, + 537, + 453, + 537 + ], + "score": 0.87, + "latex": "{ \\bf A } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 641, + 309, + 734, + 309, + 734, + 338, + 641, + 338 + ], + "score": 0.87, + "latex": "\\mathbf { b } _ { 0 } = \\mathbf { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1163, + 303, + 1219, + 303, + 1219, + 336, + 1163, + 336 + ], + "score": 0.87, + "latex": "\\mathbf { w } ^ { ( 0 ) }" + }, + { + "category_id": 13, + "poly": [ + 599, + 1918, + 637, + 1918, + 637, + 1946, + 599, + 1946 + ], + "score": 0.87, + "latex": "\\mathbf { w } _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 540, + 508, + 571, + 508, + 571, + 537, + 540, + 537 + ], + "score": 0.86, + "latex": "\\mathbf { b } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 453, + 601, + 599, + 601, + 599, + 634, + 453, + 634 + ], + "score": 0.86, + "latex": "\\pmb { \\theta } _ { t } = \\mathbf { A } _ { t } ^ { - 1 } \\mathbf { b } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 371, + 1676, + 653, + 1676, + 653, + 1714, + 371, + 1714 + ], + "score": 0.85, + "latex": "\\mathbf { x } , \\mathbf { x } ^ { \\prime } \\in \\{ \\mathbf { x } _ { i , k } \\} _ { i \\in [ T ] , k \\in [ K ] }" + }, + { + "category_id": 13, + "poly": [ + 1166, + 974, + 1196, + 974, + 1196, + 1004, + 1166, + 1004 + ], + "score": 0.84, + "latex": "\\eta _ { q }" + }, + { + "category_id": 13, + "poly": [ + 517, + 309, + 629, + 309, + 629, + 339, + 517, + 339 + ], + "score": 0.84, + "latex": "\\mathbf { A } _ { 0 } = \\lambda \\mathbf { I }" + }, + { + "category_id": 13, + "poly": [ + 454, + 476, + 480, + 476, + 480, + 507, + 454, + 507 + ], + "score": 0.83, + "latex": "\\widehat { r } _ { t }" + }, + { + "category_id": 13, + "poly": [ + 409, + 634, + 569, + 634, + 569, + 665, + 409, + 665 + ], + "score": 0.83, + "latex": "\\mathrm { n o d } ( t , H ) = 0" + }, + { + "category_id": 13, + "poly": [ + 370, + 1059, + 785, + 1059, + 785, + 1101, + 370, + 1101 + ], + "score": 0.83, + "latex": "\\begin{array} { r } { \\mathbf { w } _ { q } ^ { ( s ) } = \\mathbf { w } _ { q } ^ { ( s - 1 ) } - \\eta _ { q } \\nabla _ { \\mathbf { w } } \\mathcal { L } _ { q } ( \\mathbf { w } _ { q } ^ { ( s - 1 ) } ) . } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 514, + 436, + 1259, + 436, + 1259, + 477, + 514, + 477 + ], + "score": 0.8, + "latex": "\\begin{array} { r } { a _ { t } \\ = \\ \\mathrm { a r g m a x } _ { k \\in [ K ] } \\pmb { \\theta } _ { t - 1 } ^ { \\top } \\phi ( \\mathbf { x } _ { t , k } ; \\mathbf { w } _ { t - 1 } ) \\ + \\alpha _ { t } \\| \\phi ( \\mathbf { x } _ { t , k } ; \\mathbf { w } _ { t - 1 } ) \\| _ { \\mathbf { A } _ { t - 1 } ^ { - 1 } } } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 386, + 1031, + 532, + 1031, + 532, + 1059, + 386, + 1059 + ], + "score": 0.8, + "latex": "s = 1 , \\ldots , n" + }, + { + "category_id": 13, + "poly": [ + 1052, + 1885, + 1102, + 1885, + 1102, + 1912, + 1052, + 1912 + ], + "score": 0.79, + "latex": "T K" + }, + { + "category_id": 13, + "poly": [ + 1029, + 241, + 1052, + 241, + 1052, + 267, + 1029, + 267 + ], + "score": 0.78, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 1232, + 242, + 1259, + 242, + 1259, + 267, + 1232, + 267 + ], + "score": 0.71, + "latex": "H" + }, + { + "category_id": 13, + "poly": [ + 1034, + 976, + 1054, + 976, + 1054, + 997, + 1034, + 997 + ], + "score": 0.7, + "latex": "n" + }, + { + "category_id": 13, + "poly": [ + 392, + 569, + 740, + 569, + 740, + 601, + 392, + 601 + ], + "score": 0.67, + "latex": "\\mathbf { b } _ { t } = \\mathbf { b } _ { t - 1 } + \\widehat { r } _ { t } \\phi ( \\mathbf { x } _ { t , a _ { t } } ; \\mathbf { w } _ { t - 1 } )" + }, + { + "category_id": 13, + "poly": [ + 394, + 535, + 920, + 535, + 920, + 570, + 394, + 570 + ], + "score": 0.64, + "latex": "\\mathbf { A } _ { t } = \\mathbf { A } _ { t - 1 } + \\phi ( \\mathbf { x } _ { t , a _ { t } } ; \\mathbf { w } _ { t - 1 } ) \\phi ( \\mathbf { x } _ { t , a _ { t } } ; \\mathbf { w } _ { t - 1 } ) ^ { \\top }" + }, + { + "category_id": 13, + "poly": [ + 400, + 667, + 474, + 667, + 474, + 693, + 400, + 693 + ], + "score": 0.63, + "latex": "\\mathbf { w } _ { t } \\gets" + }, + { + "category_id": 14, + "poly": [ + 399, + 758, + 534, + 758, + 534, + 786, + 399, + 786 + ], + "score": 0.46, + "latex": "\\mathbf { w } _ { t } = \\mathbf { w } _ { t - 1 }" + }, + { + "category_id": 13, + "poly": [ + 437, + 849, + 482, + 849, + 482, + 876, + 437, + 876 + ], + "score": 0.45, + "latex": "\\mathbf { w } _ { T }" + }, + { + "category_id": 13, + "poly": [ + 485, + 1349, + 525, + 1349, + 525, + 1378, + 485, + 1378 + ], + "score": 0.32, + "latex": "\\mathrm { G u }" + }, + { + "category_id": 14, + "poly": [ + 387, + 533, + 920, + 533, + 920, + 600, + 387, + 600 + ], + "score": 0.29, + "latex": "\\begin{array} { r l } & { \\mathbf { \\Phi } ^ { \\mathbf { A } } t = \\tilde { \\mathbf { A } } _ { t - 1 } + \\tilde { \\phi } ( \\mathbf { x } _ { t , a _ { t } } ; \\mathbf { w } _ { t - 1 } ) \\phi ( \\mathbf { x } _ { t , a _ { t } } ; \\mathbf { w } _ { t - 1 } ) ^ { \\top } } \\\\ & { \\mathbf { b } _ { t } = \\mathbf { b } _ { t - 1 } + \\hat { r } _ { t } \\phi ( \\mathbf { x } _ { t , a _ { t } } ; \\mathbf { w } _ { t - 1 } ) , } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 399, + 695, + 516, + 695, + 516, + 727, + 399, + 727 + ], + "score": 0.29, + "latex": "q = q + 1" + }, + { + "category_id": 14, + "poly": [ + 399, + 695, + 516, + 695, + 516, + 727, + 399, + 727 + ], + "score": 0.25, + "latex": "q = q + 1" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 919.0, + 1007.0, + 919.0, + 1007.0, + 959.0, + 295.0, + 959.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2060.0, + 861.0, + 2060.0, + 861.0, + 2090.0, + 840.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 196.0, + 1173.0, + 196.0, + 1173.0, + 243.0, + 293.0, + 243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 840.0, + 2059.0, + 861.0, + 2059.0, + 861.0, + 2090.0, + 840.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1737.0, + 278.0, + 1737.0, + 278.0, + 1766.0, + 238.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1734.0, + 1405.0, + 1734.0, + 1405.0, + 1766.0, + 294.0, + 1766.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1768.0, + 278.0, + 1768.0, + 278.0, + 1796.0, + 238.0, + 1796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1764.0, + 1405.0, + 1764.0, + 1405.0, + 1796.0, + 295.0, + 1796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1799.0, + 276.0, + 1799.0, + 276.0, + 1824.0, + 239.0, + 1824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1794.0, + 1405.0, + 1794.0, + 1405.0, + 1827.0, + 294.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1828.0, + 278.0, + 1828.0, + 278.0, + 1857.0, + 238.0, + 1857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1824.0, + 1405.0, + 1824.0, + 1405.0, + 1857.0, + 295.0, + 1857.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1861.0, + 276.0, + 1861.0, + 276.0, + 1886.0, + 239.0, + 1886.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1856.0, + 1405.0, + 1856.0, + 1405.0, + 1888.0, + 295.0, + 1888.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1888.0, + 278.0, + 1888.0, + 278.0, + 1917.0, + 238.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1886.0, + 1051.0, + 1886.0, + 1051.0, + 1919.0, + 296.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1103.0, + 1886.0, + 1406.0, + 1886.0, + 1406.0, + 1919.0, + 1103.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1921.0, + 276.0, + 1921.0, + 276.0, + 1946.0, + 239.0, + 1946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1913.0, + 598.0, + 1913.0, + 598.0, + 1950.0, + 294.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 638.0, + 1913.0, + 1409.0, + 1913.0, + 1409.0, + 1950.0, + 638.0, + 1950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1953.0, + 276.0, + 1953.0, + 276.0, + 1976.0, + 239.0, + 1976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1943.0, + 517.0, + 1943.0, + 517.0, + 1980.0, + 294.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 1943.0, + 1406.0, + 1943.0, + 1406.0, + 1980.0, + 554.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1982.0, + 276.0, + 1982.0, + 276.0, + 2006.0, + 239.0, + 2006.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1976.0, + 723.0, + 1976.0, + 723.0, + 2009.0, + 295.0, + 2009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 238.0, + 706.0, + 238.0, + 706.0, + 272.0, + 307.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 777.0, + 238.0, + 1028.0, + 238.0, + 1028.0, + 272.0, + 777.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1053.0, + 238.0, + 1231.0, + 238.0, + 1231.0, + 272.0, + 1053.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 238.0, + 1404.0, + 238.0, + 1404.0, + 272.0, + 1260.0, + 272.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 339.0, + 263.0, + 473.0, + 263.0, + 473.0, + 313.0, + 339.0, + 313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 635.0, + 263.0, + 640.0, + 263.0, + 640.0, + 313.0, + 635.0, + 313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 303.0, + 516.0, + 303.0, + 516.0, + 341.0, + 303.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 630.0, + 303.0, + 640.0, + 303.0, + 640.0, + 341.0, + 630.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 303.0, + 864.0, + 303.0, + 864.0, + 341.0, + 735.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 896.0, + 303.0, + 982.0, + 303.0, + 982.0, + 341.0, + 896.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1101.0, + 303.0, + 1162.0, + 303.0, + 1162.0, + 341.0, + 1101.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1220.0, + 303.0, + 1406.0, + 303.0, + 1406.0, + 341.0, + 1220.0, + 341.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 338.0, + 337.0, + 603.0, + 337.0, + 603.0, + 375.0, + 338.0, + 375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 672.0, + 337.0, + 681.0, + 337.0, + 681.0, + 375.0, + 672.0, + 375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 337.0, + 816.0, + 337.0, + 816.0, + 375.0, + 811.0, + 375.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 369.0, + 385.0, + 369.0, + 385.0, + 403.0, + 305.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 369.0, + 573.0, + 369.0, + 573.0, + 403.0, + 533.0, + 403.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 403.0, + 341.0, + 403.0, + 341.0, + 434.0, + 307.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 362.0, + 401.0, + 628.0, + 401.0, + 628.0, + 439.0, + 362.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 819.0, + 401.0, + 822.0, + 401.0, + 822.0, + 439.0, + 819.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 437.0, + 342.0, + 437.0, + 342.0, + 468.0, + 307.0, + 468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 432.0, + 513.0, + 432.0, + 513.0, + 478.0, + 360.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1260.0, + 432.0, + 1406.0, + 432.0, + 1406.0, + 478.0, + 1260.0, + 478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 474.0, + 453.0, + 474.0, + 453.0, + 509.0, + 368.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 481.0, + 474.0, + 485.0, + 474.0, + 485.0, + 509.0, + 481.0, + 509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 310.0, + 509.0, + 337.0, + 509.0, + 337.0, + 536.0, + 310.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 507.0, + 452.0, + 507.0, + 452.0, + 537.0, + 371.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 490.0, + 507.0, + 539.0, + 507.0, + 539.0, + 537.0, + 490.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 572.0, + 507.0, + 701.0, + 507.0, + 701.0, + 537.0, + 572.0, + 537.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 921.0, + 531.0, + 927.0, + 531.0, + 927.0, + 572.0, + 921.0, + 572.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 308.0, + 603.0, + 337.0, + 603.0, + 337.0, + 635.0, + 308.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 371.0, + 601.0, + 452.0, + 601.0, + 452.0, + 636.0, + 371.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 307.0, + 630.0, + 339.0, + 630.0, + 339.0, + 663.0, + 307.0, + 663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 366.0, + 629.0, + 408.0, + 629.0, + 408.0, + 667.0, + 366.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 570.0, + 629.0, + 632.0, + 629.0, + 632.0, + 667.0, + 570.0, + 667.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 662.0, + 339.0, + 662.0, + 339.0, + 694.0, + 306.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 396.0, + 664.0, + 399.0, + 664.0, + 399.0, + 694.0, + 396.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 475.0, + 664.0, + 726.0, + 664.0, + 726.0, + 694.0, + 475.0, + 694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 693.0, + 340.0, + 693.0, + 340.0, + 726.0, + 297.0, + 726.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 391.0, + 695.0, + 398.0, + 695.0, + 398.0, + 728.0, + 391.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 517.0, + 695.0, + 522.0, + 695.0, + 522.0, + 728.0, + 517.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 723.0, + 341.0, + 723.0, + 341.0, + 756.0, + 297.0, + 756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 368.0, + 724.0, + 425.0, + 724.0, + 425.0, + 755.0, + 368.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 754.0, + 341.0, + 754.0, + 341.0, + 786.0, + 297.0, + 786.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 394.0, + 753.0, + 398.0, + 753.0, + 398.0, + 790.0, + 394.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 535.0, + 753.0, + 538.0, + 753.0, + 538.0, + 790.0, + 535.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 784.0, + 341.0, + 784.0, + 341.0, + 817.0, + 296.0, + 817.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 366.0, + 784.0, + 450.0, + 784.0, + 450.0, + 815.0, + 366.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 812.0, + 440.0, + 812.0, + 440.0, + 847.0, + 297.0, + 847.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 840.0, + 436.0, + 840.0, + 436.0, + 882.0, + 298.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 483.0, + 840.0, + 486.0, + 840.0, + 486.0, + 882.0, + 483.0, + 882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 263.0, + 1223.0, + 585.0, + 1223.0, + 585.0, + 1265.0, + 263.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 1223.0, + 709.0, + 1223.0, + 709.0, + 1265.0, + 657.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 810.0, + 1223.0, + 1006.0, + 1223.0, + 1006.0, + 1265.0, + 810.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1152.0, + 1223.0, + 1406.0, + 1223.0, + 1406.0, + 1265.0, + 1152.0, + 1265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 265.0, + 1252.0, + 298.0, + 1252.0, + 298.0, + 1301.0, + 265.0, + 1301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 1252.0, + 547.0, + 1252.0, + 547.0, + 1301.0, + 533.0, + 1301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 954.0, + 564.0, + 954.0, + 564.0, + 1009.0, + 302.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 711.0, + 954.0, + 1033.0, + 954.0, + 1033.0, + 1009.0, + 711.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1055.0, + 954.0, + 1165.0, + 954.0, + 1165.0, + 1009.0, + 1055.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1197.0, + 954.0, + 1409.0, + 954.0, + 1409.0, + 1009.0, + 1197.0, + 1009.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 341.0, + 998.0, + 526.0, + 998.0, + 526.0, + 1029.0, + 341.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 305.0, + 1028.0, + 385.0, + 1028.0, + 385.0, + 1062.0, + 305.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 533.0, + 1028.0, + 573.0, + 1028.0, + 573.0, + 1062.0, + 533.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 1065.0, + 342.0, + 1065.0, + 342.0, + 1098.0, + 306.0, + 1098.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 303.0, + 1093.0, + 439.0, + 1093.0, + 439.0, + 1132.0, + 303.0, + 1132.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 457.0, + 1116.0, + 498.0, + 1116.0, + 498.0, + 1147.0, + 457.0, + 1147.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 301.0, + 1117.0, + 489.0, + 1117.0, + 489.0, + 1171.0, + 301.0, + 1171.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 370.0, + 1070.0, + 415.0, + 1070.0, + 415.0, + 1100.5, + 370.0, + 1100.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 430.0, + 1065.0, + 505.0, + 1065.0, + 505.0, + 1103.0, + 430.0, + 1103.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 553.0, + 1053.5, + 799.0, + 1053.5, + 799.0, + 1112.0, + 553.0, + 1112.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1668.0, + 370.0, + 1668.0, + 370.0, + 1722.0, + 235.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 654.0, + 1668.0, + 666.0, + 1668.0, + 666.0, + 1722.0, + 654.0, + 1722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1535.0, + 714.0, + 1535.0, + 714.0, + 1577.0, + 235.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 811.0, + 1535.0, + 1009.0, + 1535.0, + 1009.0, + 1577.0, + 811.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 255.0, + 1495.0, + 1371.0, + 1495.0, + 1371.0, + 1535.0, + 255.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1416.0, + 1118.0, + 1416.0, + 1118.0, + 1458.0, + 291.0, + 1458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1327.0, + 1416.0, + 1400.0, + 1416.0, + 1400.0, + 1458.0, + 1327.0, + 1458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1451.0, + 296.0, + 1451.0, + 296.0, + 1488.0, + 293.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 380.0, + 1451.0, + 429.0, + 1451.0, + 429.0, + 1488.0, + 380.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 523.0, + 1451.0, + 535.0, + 1451.0, + 535.0, + 1488.0, + 523.0, + 1488.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 259.0, + 1314.0, + 533.0, + 1314.0, + 533.0, + 1353.0, + 259.0, + 1353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 674.0, + 1314.0, + 1406.0, + 1314.0, + 1406.0, + 1353.0, + 674.0, + 1353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 255.0, + 1346.0, + 484.0, + 1346.0, + 484.0, + 1382.0, + 255.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 526.0, + 1346.0, + 1214.0, + 1346.0, + 1214.0, + 1382.0, + 526.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1264.0, + 1346.0, + 1403.0, + 1346.0, + 1403.0, + 1382.0, + 1264.0, + 1382.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 253.0, + 1378.0, + 776.0, + 1378.0, + 776.0, + 1425.0, + 253.0, + 1425.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1050.0, + 1378.0, + 1408.0, + 1378.0, + 1408.0, + 1425.0, + 1050.0, + 1425.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 5, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 239, + 1903, + 1406, + 1903, + 1406, + 2010, + 239, + 2010 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 239, + 873, + 1405, + 873, + 1405, + 1122, + 239, + 1122 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 242, + 201, + 1402, + 201, + 1402, + 297, + 242, + 297 + ], + "score": 0.966 + }, + { + "category_id": 8, + "poly": [ + 522, + 311, + 1179, + 311, + 1179, + 532, + 522, + 532 + ], + "score": 0.959 + }, + { + "category_id": 1, + "poly": [ + 240, + 1132, + 1404, + 1132, + 1404, + 1227, + 240, + 1227 + ], + "score": 0.957 + }, + { + "category_id": 8, + "poly": [ + 606, + 1246, + 1093, + 1246, + 1093, + 1335, + 606, + 1335 + ], + "score": 0.957 + }, + { + "category_id": 1, + "poly": [ + 238, + 547, + 1407, + 547, + 1407, + 650, + 238, + 650 + ], + "score": 0.949 + }, + { + "category_id": 1, + "poly": [ + 264, + 1468, + 1400, + 1468, + 1400, + 1534, + 264, + 1534 + ], + "score": 0.946 + }, + { + "category_id": 8, + "poly": [ + 616, + 667, + 1083, + 667, + 1083, + 729, + 616, + 729 + ], + "score": 0.944 + }, + { + "category_id": 1, + "poly": [ + 249, + 1350, + 1408, + 1350, + 1408, + 1414, + 249, + 1414 + ], + "score": 0.94 + }, + { + "category_id": 1, + "poly": [ + 256, + 787, + 1396, + 787, + 1396, + 851, + 256, + 851 + ], + "score": 0.935 + }, + { + "category_id": 1, + "poly": [ + 266, + 745, + 1336, + 745, + 1336, + 779, + 266, + 779 + ], + "score": 0.931 + }, + { + "category_id": 1, + "poly": [ + 240, + 1621, + 802, + 1621, + 802, + 1655, + 240, + 1655 + ], + "score": 0.93 + }, + { + "category_id": 8, + "poly": [ + 627, + 1674, + 1071, + 1674, + 1071, + 1721, + 627, + 1721 + ], + "score": 0.927 + }, + { + "category_id": 1, + "poly": [ + 235, + 1735, + 1407, + 1735, + 1407, + 1797, + 235, + 1797 + ], + "score": 0.927 + }, + { + "category_id": 8, + "poly": [ + 504, + 1553, + 1190, + 1553, + 1190, + 1605, + 504, + 1605 + ], + "score": 0.921 + }, + { + "category_id": 8, + "poly": [ + 368, + 1810, + 1323, + 1810, + 1323, + 1892, + 368, + 1892 + ], + "score": 0.919 + }, + { + "category_id": 1, + "poly": [ + 258, + 1426, + 1111, + 1426, + 1111, + 1461, + 258, + 1461 + ], + "score": 0.909 + }, + { + "category_id": 9, + "poly": [ + 1344, + 683, + 1400, + 683, + 1400, + 715, + 1344, + 715 + ], + "score": 0.908 + }, + { + "category_id": 9, + "poly": [ + 1344, + 493, + 1400, + 493, + 1400, + 525, + 1344, + 525 + ], + "score": 0.904 + }, + { + "category_id": 9, + "poly": [ + 1344, + 1274, + 1400, + 1274, + 1400, + 1306, + 1344, + 1306 + ], + "score": 0.9 + }, + { + "category_id": 2, + "poly": [ + 841, + 2060, + 859, + 2060, + 859, + 2084, + 841, + 2084 + ], + "score": 0.73 + }, + { + "category_id": 2, + "poly": [ + 841, + 2060, + 859, + 2060, + 859, + 2084, + 841, + 2084 + ], + "score": 0.138 + }, + { + "category_id": 13, + "poly": [ + 1144, + 1469, + 1290, + 1469, + 1290, + 1503, + 1144, + 1503 + ], + "score": 0.95, + "latex": "\\lVert \\pmb { \\theta } ^ { * } \\rVert _ { 2 } \\leq M" + }, + { + "category_id": 14, + "poly": [ + 606, + 1242, + 1093, + 1242, + 1093, + 1336, + 606, + 1336 + ], + "score": 0.95, + "latex": "R _ { T } = \\mathbb { E } \\bigg [ \\sum _ { t = 1 } ^ { T } \\big ( \\widehat { r } ( \\mathbf { x } _ { t , a _ { t } ^ { * } } ) - \\widehat { r } ( \\mathbf { x } _ { t , a _ { t } } ) \\big ) \\big | \\mathbf { w } ^ { ( 0 ) } \\bigg ] ," + }, + { + "category_id": 14, + "poly": [ + 522, + 308, + 1177, + 308, + 1177, + 534, + 522, + 534 + ], + "score": 0.94, + "latex": "\\begin{array} { r l } & { \\widetilde { \\pmb { \\Sigma } } ^ { ( 0 ) } ( \\mathbf x , \\mathbf y ) = \\pmb { \\Sigma } ^ { ( 0 ) } ( \\mathbf x , \\mathbf y ) = \\mathbf x ^ { \\top } \\mathbf y , } \\\\ & { \\pmb { \\Lambda } ^ { ( l ) } ( \\mathbf x , \\mathbf y ) = \\left[ \\pmb { \\Sigma } ^ { l - 1 } ( \\mathbf x , \\mathbf x ) \\quad \\pmb { \\Sigma } ^ { l - 1 } ( \\mathbf x , \\mathbf y ) \\right] , } \\\\ & { \\pmb { \\Sigma } ^ { ( l ) } ( \\mathbf x , \\mathbf y ) = 2 \\mathbb { E } _ { ( u , v ) \\sim N ( \\mathbf 0 , \\mathbf { A } ^ { ( l - 1 ) } ( \\mathbf x , \\mathbf y ) ) } [ \\sigma ( u ) \\sigma ( v ) ] , } \\\\ & { \\widetilde { \\pmb { \\Sigma } } ^ { ( l ) } ( \\mathbf x , \\mathbf y ) = 2 \\widetilde { \\pmb { \\Sigma } } ^ { ( l - 1 ) } ( \\mathbf x , \\mathbf y ) \\mathbb { E } _ { u , v } [ \\dot { \\sigma } ( u ) \\dot { \\sigma } ( v ) ] + \\pmb { \\Sigma } ^ { ( l ) } ( \\mathbf x , \\mathbf y ) , } \\end{array}" + }, + { + "category_id": 14, + "poly": [ + 616, + 664, + 1081, + 664, + 1081, + 731, + 616, + 731 + ], + "score": 0.94, + "latex": "\\mathbf { H } _ { i j } = \\frac { 1 } { 2 } \\big ( \\widetilde { \\boldsymbol { \\Sigma } } ^ { ( L ) } ( \\mathbf { x } _ { i } , \\mathbf { x } _ { j } ) + \\boldsymbol { \\Sigma } ^ { ( L ) } ( \\mathbf { x } _ { i } , \\mathbf { x } _ { j } ) \\big ) ," + }, + { + "category_id": 13, + "poly": [ + 678, + 1500, + 789, + 1500, + 789, + 1533, + 678, + 1533 + ], + "score": 0.94, + "latex": "\\delta \\in ( 0 , 1 )" + }, + { + "category_id": 14, + "poly": [ + 370, + 1806, + 1326, + 1806, + 1326, + 1892, + 370, + 1892 + ], + "score": 0.93, + "latex": "R _ { T } \\leq C _ { 1 } \\alpha _ { T } \\sqrt { T d \\log \\left( 1 + \\frac { T G ^ { 2 } } { \\lambda d } \\right) } + \\frac { C _ { 2 } \\ell _ { \\mathrm { L i p } } L ^ { 3 } d ^ { 5 / 2 } T \\sqrt { \\log m \\log ( \\frac { 1 } { \\delta } ) \\log ( \\frac { T K } { \\delta } ) } \\| \\mathbf { r } - \\widetilde { \\mathbf { r } } \\| _ { \\mathbf { H } ^ { - 1 } } } { m ^ { 1 / 6 } } ," + }, + { + "category_id": 13, + "poly": [ + 456, + 613, + 650, + 613, + 650, + 650, + 456, + 650 + ], + "score": 0.93, + "latex": "\\{ \\mathbf { x } _ { t , k } \\} _ { t \\in [ T ] , k \\in [ K ] }" + }, + { + "category_id": 14, + "poly": [ + 508, + 1548, + 1192, + 1548, + 1192, + 1609, + 508, + 1609 + ], + "score": 0.92, + "latex": "\\alpha _ { t } = \\nu \\sqrt { 2 \\big ( d \\log ( 1 + t \\log ( H K ) / \\lambda ) + \\log ( 1 / \\delta ) \\big ) } + \\lambda ^ { 1 / 2 } M ." + }, + { + "category_id": 13, + "poly": [ + 372, + 746, + 503, + 746, + 503, + 781, + 372, + 781 + ], + "score": 0.92, + "latex": "i , j \\in [ T K ]" + }, + { + "category_id": 13, + "poly": [ + 371, + 547, + 692, + 547, + 692, + 584, + 371, + 584 + ], + "score": 0.92, + "latex": "( u , v ) \\sim N ( \\mathbf { 0 } , \\mathbf { \\Lambda } \\Lambda ^ { ( l - 1 ) } ( \\mathbf { x } , \\mathbf { y } ) )" + }, + { + "category_id": 13, + "poly": [ + 1232, + 787, + 1401, + 787, + 1401, + 822, + 1232, + 822 + ], + "score": 0.92, + "latex": "\\lambda _ { \\operatorname* { m i n } } ( \\mathbf { H } ) \\geq \\lambda _ { 0 }" + }, + { + "category_id": 13, + "poly": [ + 1109, + 998, + 1251, + 998, + 1251, + 1029, + 1109, + 1029 + ], + "score": 0.91, + "latex": "l = 1 , \\ldots , L" + }, + { + "category_id": 13, + "poly": [ + 501, + 819, + 583, + 819, + 583, + 849, + 501, + 849 + ], + "score": 0.91, + "latex": "\\lambda _ { 0 } > 0" + }, + { + "category_id": 13, + "poly": [ + 1034, + 614, + 1081, + 614, + 1081, + 648, + 1034, + 648 + ], + "score": 0.91, + "latex": "\\mathbf { H } _ { i j }" + }, + { + "category_id": 13, + "poly": [ + 685, + 613, + 846, + 613, + 846, + 648, + 685, + 648 + ], + "score": 0.91, + "latex": "\\{ \\mathbf { x } _ { i } \\} _ { i = 1 , \\dots , T K }" + }, + { + "category_id": 13, + "poly": [ + 1204, + 582, + 1398, + 582, + 1398, + 617, + 1204, + 617 + ], + "score": 0.91, + "latex": "\\{ \\mathbf { x } _ { t , k } \\} _ { t \\in [ T ] , k \\in [ K ] }" + }, + { + "category_id": 13, + "poly": [ + 383, + 1766, + 446, + 1766, + 446, + 1793, + 383, + 1793 + ], + "score": 0.9, + "latex": "1 - \\delta" + }, + { + "category_id": 13, + "poly": [ + 384, + 1903, + 517, + 1903, + 517, + 1938, + 384, + 1938 + ], + "score": 0.9, + "latex": "\\{ C _ { i } \\} _ { i = 0 , 1 , 2 }" + }, + { + "category_id": 13, + "poly": [ + 490, + 1501, + 574, + 1501, + 574, + 1528, + 490, + 1528 + ], + "score": 0.9, + "latex": "M > 0" + }, + { + "category_id": 13, + "poly": [ + 728, + 579, + 898, + 579, + 898, + 608, + 728, + 608 + ], + "score": 0.9, + "latex": "\\mathbf { H } \\in \\mathbb { R } ^ { T K \\times T K }" + }, + { + "category_id": 13, + "poly": [ + 417, + 1972, + 609, + 1972, + 609, + 2009, + 417, + 2009 + ], + "score": 0.9, + "latex": "\\| \\mathbf { r } \\| _ { \\mathbf { A } } = \\sqrt { \\mathbf { r } ^ { \\top } \\mathbf { A } \\mathbf { r } }" + }, + { + "category_id": 14, + "poly": [ + 627, + 1672, + 1069, + 1672, + 1069, + 1720, + 627, + 1720 + ], + "score": 0.9, + "latex": "\\eta _ { q } \\leq C _ { 0 } \\big ( d ^ { 2 } m n T ^ { 5 . 5 } L ^ { 6 } \\log ( T K / \\delta ) \\big ) ^ { - 1 } ," + }, + { + "category_id": 13, + "poly": [ + 1188, + 1353, + 1228, + 1353, + 1228, + 1382, + 1188, + 1382 + ], + "score": 0.89, + "latex": "R _ { T }" + }, + { + "category_id": 13, + "poly": [ + 1222, + 550, + 1269, + 550, + 1269, + 583, + 1222, + 583 + ], + "score": 0.89, + "latex": "\\sigma ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 796, + 1737, + 1203, + 1737, + 1203, + 1770, + 796, + 1770 + ], + "score": 0.89, + "latex": "m = \\mathrm { p o l y } ( L , d , 1 / \\delta , H , \\log ( T K / \\delta ) )" + }, + { + "category_id": 13, + "poly": [ + 750, + 550, + 798, + 550, + 798, + 582, + 750, + 582 + ], + "score": 0.89, + "latex": "\\dot { \\sigma } ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 297, + 1974, + 357, + 1974, + 357, + 2004, + 297, + 2004 + ], + "score": 0.89, + "latex": "\\mathbb { R } ^ { T K }" + }, + { + "category_id": 13, + "poly": [ + 1112, + 263, + 1227, + 263, + 1227, + 295, + 1112, + 295 + ], + "score": 0.88, + "latex": "\\mathbf { x } , \\mathbf { y } \\in \\mathbb { R } ^ { \\bar { d } }" + }, + { + "category_id": 13, + "poly": [ + 565, + 1628, + 594, + 1628, + 594, + 1656, + 565, + 1656 + ], + "score": 0.86, + "latex": "\\eta _ { q }" + }, + { + "category_id": 13, + "poly": [ + 950, + 1504, + 980, + 1504, + 980, + 1530, + 950, + 1530 + ], + "score": 0.86, + "latex": "\\alpha _ { t }" + }, + { + "category_id": 13, + "poly": [ + 793, + 1936, + 1404, + 1936, + 1404, + 1973, + 793, + 1973 + ], + "score": 0.85, + "latex": "\\widetilde { \\textbf { r } } = ( f ( \\mathbf { x } _ { 1 } ; \\pmb { \\theta } _ { 0 } , \\mathbf { w } _ { 0 } ) , \\dots , f ( \\mathbf { x } _ { T K } ; \\pmb { \\theta } _ { T - 1 } , \\mathbf { w } _ { T - 1 } ) ) ^ { \\top } \\ \\in" + }, + { + "category_id": 13, + "poly": [ + 296, + 1936, + 734, + 1936, + 734, + 1972, + 296, + 1972 + ], + "score": 0.85, + "latex": "( r ( \\mathbf { x } _ { 1 } ) , r ( \\mathbf { x } _ { 2 } ) , \\ldots , r ( \\mathbf { x } _ { T K } ) ) ^ { \\top } \\ \\in \\ \\mathbb { R } ^ { T K }" + }, + { + "category_id": 13, + "poly": [ + 1333, + 1907, + 1404, + 1907, + 1404, + 1935, + 1333, + 1935 + ], + "score": 0.81, + "latex": "\\begin{array} { r l } { \\mathbf { r } } & { { } = } \\end{array}" + }, + { + "category_id": 13, + "poly": [ + 389, + 1196, + 412, + 1196, + 412, + 1222, + 389, + 1222 + ], + "score": 0.8, + "latex": "T" + }, + { + "category_id": 13, + "poly": [ + 959, + 1059, + 978, + 1059, + 978, + 1085, + 959, + 1085 + ], + "score": 0.79, + "latex": "\\pmb \\theta" + }, + { + "category_id": 13, + "poly": [ + 1307, + 747, + 1335, + 747, + 1335, + 774, + 1307, + 774 + ], + "score": 0.68, + "latex": "\\mathbf { H }" + }, + { + "category_id": 13, + "poly": [ + 1059, + 879, + 1086, + 879, + 1086, + 904, + 1059, + 904 + ], + "score": 0.63, + "latex": "\\mathbf { H }" + }, + { + "category_id": 13, + "poly": [ + 1302, + 1029, + 1331, + 1029, + 1331, + 1055, + 1302, + 1055 + ], + "score": 0.6, + "latex": "\\mathbf { H }" + }, + { + "category_id": 13, + "poly": [ + 863, + 1029, + 882, + 1029, + 882, + 1055, + 863, + 1055 + ], + "score": 0.57, + "latex": "\\pmb \\theta" + }, + { + "category_id": 13, + "poly": [ + 724, + 998, + 753, + 998, + 753, + 1026, + 724, + 1026 + ], + "score": 0.34, + "latex": "\\mathbf { H }" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2059.0, + 860.0, + 2059.0, + 860.0, + 2091.0, + 839.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2059.0, + 860.0, + 2059.0, + 860.0, + 2091.0, + 839.0, + 2091.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 231.0, + 1897.0, + 383.0, + 1897.0, + 383.0, + 1944.0, + 231.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 518.0, + 1897.0, + 1332.0, + 1897.0, + 1332.0, + 1944.0, + 518.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1405.0, + 1897.0, + 1409.0, + 1897.0, + 1409.0, + 1944.0, + 1405.0, + 1944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 230.0, + 1925.0, + 295.0, + 1925.0, + 295.0, + 1980.0, + 230.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 735.0, + 1925.0, + 792.0, + 1925.0, + 792.0, + 1980.0, + 735.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1405.0, + 1925.0, + 1411.0, + 1925.0, + 1411.0, + 1980.0, + 1405.0, + 1980.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 228.0, + 1963.0, + 296.0, + 1963.0, + 296.0, + 2013.0, + 228.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 358.0, + 1963.0, + 416.0, + 1963.0, + 416.0, + 2013.0, + 358.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 610.0, + 1963.0, + 624.0, + 1963.0, + 624.0, + 2013.0, + 610.0, + 2013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 882.0, + 277.0, + 882.0, + 277.0, + 908.0, + 239.0, + 908.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 875.0, + 1058.0, + 875.0, + 1058.0, + 909.0, + 293.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1087.0, + 875.0, + 1406.0, + 875.0, + 1406.0, + 909.0, + 1087.0, + 909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 910.0, + 279.0, + 910.0, + 279.0, + 939.0, + 236.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 905.0, + 1406.0, + 905.0, + 1406.0, + 939.0, + 293.0, + 939.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 940.0, + 279.0, + 940.0, + 279.0, + 969.0, + 236.0, + 969.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 936.0, + 1406.0, + 936.0, + 1406.0, + 970.0, + 295.0, + 970.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 970.0, + 279.0, + 970.0, + 279.0, + 999.0, + 236.0, + 999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 966.0, + 1404.0, + 966.0, + 1404.0, + 1000.0, + 295.0, + 1000.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1003.0, + 277.0, + 1003.0, + 277.0, + 1029.0, + 239.0, + 1029.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 997.0, + 723.0, + 997.0, + 723.0, + 1032.0, + 296.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 754.0, + 997.0, + 1108.0, + 997.0, + 1108.0, + 1032.0, + 754.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1252.0, + 997.0, + 1405.0, + 997.0, + 1405.0, + 1032.0, + 1252.0, + 1032.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1032.0, + 279.0, + 1032.0, + 279.0, + 1060.0, + 236.0, + 1060.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1029.0, + 862.0, + 1029.0, + 862.0, + 1059.0, + 296.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 883.0, + 1029.0, + 1301.0, + 1029.0, + 1301.0, + 1059.0, + 883.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1332.0, + 1029.0, + 1404.0, + 1029.0, + 1404.0, + 1059.0, + 1332.0, + 1059.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1062.0, + 279.0, + 1062.0, + 279.0, + 1090.0, + 236.0, + 1090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1055.0, + 958.0, + 1055.0, + 958.0, + 1092.0, + 293.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 979.0, + 1055.0, + 1405.0, + 1055.0, + 1405.0, + 1092.0, + 979.0, + 1092.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1094.0, + 277.0, + 1094.0, + 277.0, + 1120.0, + 239.0, + 1120.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1089.0, + 806.0, + 1089.0, + 806.0, + 1123.0, + 295.0, + 1123.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 207.0, + 280.0, + 207.0, + 280.0, + 234.0, + 239.0, + 234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 202.0, + 1405.0, + 202.0, + 1405.0, + 237.0, + 289.0, + 237.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 238.0, + 279.0, + 238.0, + 279.0, + 265.0, + 239.0, + 265.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 233.0, + 1407.0, + 233.0, + 1407.0, + 267.0, + 293.0, + 267.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 269.0, + 281.0, + 269.0, + 281.0, + 296.0, + 239.0, + 296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 287.0, + 261.0, + 1111.0, + 261.0, + 1111.0, + 298.0, + 287.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1228.0, + 261.0, + 1356.0, + 261.0, + 1356.0, + 298.0, + 1228.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1138.0, + 280.0, + 1138.0, + 280.0, + 1165.0, + 240.0, + 1165.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1130.0, + 1407.0, + 1130.0, + 1407.0, + 1170.0, + 290.0, + 1170.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1170.0, + 278.0, + 1170.0, + 278.0, + 1193.0, + 240.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1165.0, + 1404.0, + 1165.0, + 1404.0, + 1195.0, + 294.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1199.0, + 280.0, + 1199.0, + 280.0, + 1226.0, + 240.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1194.0, + 388.0, + 1194.0, + 388.0, + 1228.0, + 291.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 413.0, + 1194.0, + 947.0, + 1194.0, + 947.0, + 1228.0, + 413.0, + 1228.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 544.0, + 370.0, + 544.0, + 370.0, + 627.0, + 237.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 693.0, + 544.0, + 727.0, + 544.0, + 727.0, + 627.0, + 693.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 544.0, + 1203.0, + 544.0, + 1203.0, + 627.0, + 899.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1399.0, + 544.0, + 1415.0, + 544.0, + 1415.0, + 627.0, + 1399.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 619.0, + 278.0, + 619.0, + 278.0, + 643.0, + 240.0, + 643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 610.0, + 455.0, + 610.0, + 455.0, + 653.0, + 288.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 651.0, + 610.0, + 684.0, + 610.0, + 684.0, + 653.0, + 651.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 847.0, + 610.0, + 1033.0, + 610.0, + 1033.0, + 653.0, + 847.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1082.0, + 610.0, + 1232.0, + 610.0, + 1232.0, + 653.0, + 1082.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 258.0, + 1476.0, + 280.0, + 1476.0, + 280.0, + 1498.0, + 258.0, + 1498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1466.0, + 1143.0, + 1466.0, + 1143.0, + 1505.0, + 290.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1291.0, + 1466.0, + 1404.0, + 1466.0, + 1404.0, + 1505.0, + 1291.0, + 1505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 262.0, + 1507.0, + 281.0, + 1507.0, + 281.0, + 1530.0, + 262.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 1498.0, + 489.0, + 1498.0, + 489.0, + 1535.0, + 289.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 575.0, + 1498.0, + 677.0, + 1498.0, + 677.0, + 1535.0, + 575.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 790.0, + 1498.0, + 949.0, + 1498.0, + 949.0, + 1535.0, + 790.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 981.0, + 1498.0, + 1232.0, + 1498.0, + 1232.0, + 1535.0, + 981.0, + 1535.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1350.0, + 1187.0, + 1350.0, + 1187.0, + 1386.0, + 237.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1229.0, + 1350.0, + 1407.0, + 1350.0, + 1407.0, + 1386.0, + 1229.0, + 1386.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1381.0, + 1261.0, + 1381.0, + 1261.0, + 1415.0, + 237.0, + 1415.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 251.0, + 783.0, + 1231.0, + 783.0, + 1231.0, + 825.0, + 251.0, + 825.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 245.0, + 818.0, + 500.0, + 818.0, + 500.0, + 854.0, + 245.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 584.0, + 818.0, + 595.0, + 818.0, + 595.0, + 854.0, + 584.0, + 854.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 257.0, + 744.0, + 371.0, + 744.0, + 371.0, + 783.0, + 257.0, + 783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 504.0, + 744.0, + 1306.0, + 744.0, + 1306.0, + 783.0, + 504.0, + 783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1619.0, + 564.0, + 1619.0, + 564.0, + 1659.0, + 237.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 595.0, + 1619.0, + 802.0, + 1619.0, + 802.0, + 1659.0, + 595.0, + 1659.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1734.0, + 795.0, + 1734.0, + 795.0, + 1774.0, + 234.0, + 1774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1204.0, + 1734.0, + 1405.0, + 1734.0, + 1405.0, + 1774.0, + 1204.0, + 1774.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1763.0, + 382.0, + 1763.0, + 382.0, + 1798.0, + 234.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 447.0, + 1763.0, + 1294.0, + 1763.0, + 1294.0, + 1798.0, + 447.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 247.0, + 1422.0, + 1109.0, + 1422.0, + 1109.0, + 1468.0, + 247.0, + 1468.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 6, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 5, + "poly": [ + 495, + 1828, + 1198, + 1828, + 1198, + 1977, + 495, + 1977 + ], + "score": 0.977, + "html": "
StatlogMagicCovertypeMNIST
Number of attributes91154784
Number of arms72710
Number of instances58,00019,020581,01260,000
" + }, + { + "category_id": 1, + "poly": [ + 286, + 1129, + 1405, + 1129, + 1405, + 1554, + 286, + 1554 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 296, + 1568, + 1403, + 1568, + 1403, + 1751, + 296, + 1751 + ], + "score": 0.968 + }, + { + "category_id": 1, + "poly": [ + 296, + 780, + 1404, + 780, + 1404, + 1005, + 296, + 1005 + ], + "score": 0.967 + }, + { + "category_id": 1, + "poly": [ + 241, + 202, + 1403, + 202, + 1403, + 338, + 241, + 338 + ], + "score": 0.963 + }, + { + "category_id": 1, + "poly": [ + 284, + 366, + 1404, + 366, + 1404, + 610, + 284, + 610 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 297, + 668, + 1405, + 668, + 1405, + 771, + 297, + 771 + ], + "score": 0.953 + }, + { + "category_id": 1, + "poly": [ + 292, + 622, + 1215, + 622, + 1215, + 656, + 292, + 656 + ], + "score": 0.892 + }, + { + "category_id": 0, + "poly": [ + 246, + 1054, + 531, + 1054, + 531, + 1093, + 246, + 1093 + ], + "score": 0.89 + }, + { + "category_id": 6, + "poly": [ + 302, + 1779, + 1371, + 1779, + 1371, + 1814, + 302, + 1814 + ], + "score": 0.874 + }, + { + "category_id": 2, + "poly": [ + 840, + 2061, + 859, + 2061, + 859, + 2085, + 840, + 2085 + ], + "score": 0.817 + }, + { + "category_id": 8, + "poly": [ + 296, + 732, + 733, + 732, + 733, + 771, + 296, + 771 + ], + "score": 0.216 + }, + { + "category_id": 1, + "poly": [ + 240, + 1125, + 278, + 1125, + 278, + 1616, + 240, + 1616 + ], + "score": 0.166 + }, + { + "category_id": 13, + "poly": [ + 711, + 812, + 800, + 812, + 800, + 852, + 711, + 852 + ], + "score": 0.94, + "latex": "\\widetilde { O } ( \\sqrt { T } )" + }, + { + "category_id": 13, + "poly": [ + 554, + 270, + 959, + 270, + 959, + 310, + 554, + 310 + ], + "score": 0.94, + "latex": "\\widetilde { O } ( m ^ { - 1 / 6 } T \\sqrt { ( \\mathbf { r } - \\widetilde { \\mathbf { r } } ) ^ { \\top } \\mathbf { H } ^ { - 1 } ( \\mathbf { r } - \\widetilde { \\mathbf { r } } ) } )" + }, + { + "category_id": 13, + "poly": [ + 912, + 781, + 1109, + 781, + 1109, + 815, + 912, + 815 + ], + "score": 0.93, + "latex": "\\mathcal { E } : = \\| \\mathbf { r } - \\widetilde { \\mathbf { r } } \\| _ { \\mathbf { H } ^ { - 1 } }" + }, + { + "category_id": 13, + "poly": [ + 521, + 233, + 608, + 233, + 608, + 271, + 521, + 271 + ], + "score": 0.93, + "latex": "\\widetilde { O } ( \\sqrt { T } )" + }, + { + "category_id": 13, + "poly": [ + 1240, + 781, + 1353, + 781, + 1353, + 815, + 1240, + 815 + ], + "score": 0.92, + "latex": "\\mathcal { E } = O ( 1 )" + }, + { + "category_id": 13, + "poly": [ + 1251, + 972, + 1295, + 972, + 1295, + 1006, + 1251, + 1006 + ], + "score": 0.92, + "latex": "r ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 900, + 699, + 1004, + 699, + 1004, + 730, + 900, + 730 + ], + "score": 0.92, + "latex": "m \\geq T ^ { 3 }" + }, + { + "category_id": 13, + "poly": [ + 298, + 396, + 497, + 396, + 497, + 430, + 298, + 430 + ], + "score": 0.92, + "latex": "r ( \\cdot ) = \\langle \\theta \\ast , \\psi ( \\cdot ) \\rangle" + }, + { + "category_id": 13, + "poly": [ + 686, + 1688, + 773, + 1688, + 773, + 1718, + 686, + 1718 + ], + "score": 0.92, + "latex": "\\mathbf { x } \\in \\mathbb { R } ^ { d }" + }, + { + "category_id": 13, + "poly": [ + 1034, + 396, + 1088, + 396, + 1088, + 430, + 1034, + 430 + ], + "score": 0.92, + "latex": "\\psi ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 1217, + 849, + 1322, + 849, + 1322, + 880, + 1217, + 880 + ], + "score": 0.91, + "latex": "\\mathbf { r } ^ { \\top } \\mathbf { H } ^ { - 1 } \\mathbf { r }" + }, + { + "category_id": 13, + "poly": [ + 297, + 730, + 725, + 730, + 725, + 772, + 297, + 772 + ], + "score": 0.91, + "latex": "R _ { T } = { \\widetilde { O } } ( { \\sqrt { T } } { \\sqrt { ( \\mathbf { r } - { \\widetilde { \\mathbf { r } } } ) ^ { \\top } \\mathbf { H } ^ { - 1 } ( \\mathbf { r } - { \\widetilde { \\mathbf { r } } } ) } } )" + }, + { + "category_id": 13, + "poly": [ + 729, + 699, + 782, + 699, + 782, + 733, + 729, + 733 + ], + "score": 0.9, + "latex": "\\phi ( \\cdot )" + }, + { + "category_id": 13, + "poly": [ + 948, + 397, + 983, + 397, + 983, + 425, + 948, + 425 + ], + "score": 0.85, + "latex": "\\pmb { \\theta } \\ast" + }, + { + "category_id": 13, + "poly": [ + 542, + 308, + 561, + 308, + 561, + 337, + 542, + 337 + ], + "score": 0.84, + "latex": "f" + }, + { + "category_id": 13, + "poly": [ + 1139, + 1691, + 1168, + 1691, + 1168, + 1717, + 1139, + 1717 + ], + "score": 0.79, + "latex": "K" + }, + { + "category_id": 13, + "poly": [ + 951, + 312, + 967, + 312, + 967, + 333, + 951, + 333 + ], + "score": 0.76, + "latex": "r" + }, + { + "category_id": 13, + "poly": [ + 1255, + 913, + 1276, + 913, + 1276, + 940, + 1255, + 940 + ], + "score": 0.75, + "latex": "\\mathcal { E }" + }, + { + "category_id": 13, + "poly": [ + 298, + 1722, + 316, + 1722, + 316, + 1748, + 298, + 1748 + ], + "score": 0.72, + "latex": "d" + }, + { + "category_id": 13, + "poly": [ + 1045, + 1525, + 1072, + 1525, + 1072, + 1552, + 1045, + 1552 + ], + "score": 0.53, + "latex": "@" + }, + { + "category_id": 13, + "poly": [ + 1360, + 886, + 1378, + 886, + 1378, + 909, + 1360, + 909 + ], + "score": 0.45, + "latex": "\\mathbf { r }" + }, + { + "category_id": 13, + "poly": [ + 1299, + 306, + 1327, + 306, + 1327, + 333, + 1299, + 333 + ], + "score": 0.26, + "latex": "\\mathbf { H }" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1052.0, + 536.0, + 1052.0, + 536.0, + 1101.0, + 234.0, + 1101.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 309.0, + 1771.0, + 1377.0, + 1771.0, + 1377.0, + 1820.0, + 309.0, + 1820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2058.0, + 861.0, + 2058.0, + 861.0, + 2089.0, + 839.0, + 2089.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1127.0, + 1404.0, + 1127.0, + 1404.0, + 1162.0, + 294.0, + 1162.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1159.0, + 1408.0, + 1159.0, + 1408.0, + 1193.0, + 292.0, + 1193.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1188.0, + 1405.0, + 1188.0, + 1405.0, + 1224.0, + 293.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1220.0, + 1404.0, + 1220.0, + 1404.0, + 1254.0, + 293.0, + 1254.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1251.0, + 1408.0, + 1251.0, + 1408.0, + 1286.0, + 294.0, + 1286.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1281.0, + 1404.0, + 1281.0, + 1404.0, + 1316.0, + 294.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1311.0, + 1405.0, + 1311.0, + 1405.0, + 1345.0, + 293.0, + 1345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1342.0, + 1404.0, + 1342.0, + 1404.0, + 1376.0, + 292.0, + 1376.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1372.0, + 1405.0, + 1372.0, + 1405.0, + 1406.0, + 293.0, + 1406.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1401.0, + 1405.0, + 1401.0, + 1405.0, + 1439.0, + 292.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1430.0, + 1406.0, + 1430.0, + 1406.0, + 1468.0, + 293.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1464.0, + 1405.0, + 1464.0, + 1405.0, + 1495.0, + 294.0, + 1495.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1492.0, + 1404.0, + 1492.0, + 1404.0, + 1527.0, + 294.0, + 1527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1523.0, + 1044.0, + 1523.0, + 1044.0, + 1557.0, + 292.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1073.0, + 1523.0, + 1192.0, + 1523.0, + 1192.0, + 1557.0, + 1073.0, + 1557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1567.0, + 1405.0, + 1567.0, + 1405.0, + 1603.0, + 294.0, + 1603.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1599.0, + 1406.0, + 1599.0, + 1406.0, + 1634.0, + 295.0, + 1634.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1629.0, + 1406.0, + 1629.0, + 1406.0, + 1665.0, + 294.0, + 1665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1658.0, + 1404.0, + 1658.0, + 1404.0, + 1694.0, + 294.0, + 1694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1690.0, + 685.0, + 1690.0, + 685.0, + 1721.0, + 295.0, + 1721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 774.0, + 1690.0, + 1138.0, + 1690.0, + 1138.0, + 1721.0, + 774.0, + 1721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1169.0, + 1690.0, + 1405.0, + 1690.0, + 1405.0, + 1721.0, + 1169.0, + 1721.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1720.0, + 297.0, + 1720.0, + 297.0, + 1753.0, + 294.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 317.0, + 1720.0, + 806.0, + 1720.0, + 806.0, + 1753.0, + 317.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 780.0, + 911.0, + 780.0, + 911.0, + 815.0, + 294.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1110.0, + 780.0, + 1239.0, + 780.0, + 1239.0, + 815.0, + 1110.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1354.0, + 780.0, + 1405.0, + 780.0, + 1405.0, + 815.0, + 1354.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 814.0, + 710.0, + 814.0, + 710.0, + 853.0, + 293.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 801.0, + 814.0, + 1406.0, + 814.0, + 1406.0, + 853.0, + 801.0, + 853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 849.0, + 1216.0, + 849.0, + 1216.0, + 884.0, + 295.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1323.0, + 849.0, + 1405.0, + 849.0, + 1405.0, + 884.0, + 1323.0, + 884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 881.0, + 1359.0, + 881.0, + 1359.0, + 914.0, + 294.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1379.0, + 881.0, + 1406.0, + 881.0, + 1406.0, + 914.0, + 1379.0, + 914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 910.0, + 1254.0, + 910.0, + 1254.0, + 946.0, + 294.0, + 946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1277.0, + 910.0, + 1406.0, + 910.0, + 1406.0, + 946.0, + 1277.0, + 946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 939.0, + 1406.0, + 939.0, + 1406.0, + 976.0, + 293.0, + 976.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 970.0, + 1250.0, + 970.0, + 1250.0, + 1008.0, + 293.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1296.0, + 970.0, + 1307.0, + 970.0, + 1307.0, + 1008.0, + 1296.0, + 1008.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 207.0, + 279.0, + 207.0, + 279.0, + 234.0, + 238.0, + 234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 199.0, + 1406.0, + 199.0, + 1406.0, + 239.0, + 290.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 243.0, + 279.0, + 243.0, + 279.0, + 270.0, + 238.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 236.0, + 520.0, + 236.0, + 520.0, + 274.0, + 292.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 609.0, + 236.0, + 1406.0, + 236.0, + 1406.0, + 274.0, + 609.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 282.0, + 276.0, + 282.0, + 276.0, + 306.0, + 240.0, + 306.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 270.0, + 553.0, + 270.0, + 553.0, + 312.0, + 291.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 960.0, + 270.0, + 1408.0, + 270.0, + 1408.0, + 312.0, + 960.0, + 312.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 311.0, + 279.0, + 311.0, + 279.0, + 338.0, + 238.0, + 338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 306.0, + 541.0, + 306.0, + 541.0, + 340.0, + 292.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 562.0, + 306.0, + 950.0, + 306.0, + 950.0, + 340.0, + 562.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 968.0, + 306.0, + 1298.0, + 306.0, + 1298.0, + 340.0, + 968.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1328.0, + 306.0, + 1336.0, + 306.0, + 1336.0, + 340.0, + 1328.0, + 340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 363.0, + 1407.0, + 363.0, + 1407.0, + 402.0, + 292.0, + 402.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 394.0, + 297.0, + 394.0, + 297.0, + 433.0, + 294.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 498.0, + 394.0, + 947.0, + 394.0, + 947.0, + 433.0, + 498.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 984.0, + 394.0, + 1033.0, + 394.0, + 1033.0, + 433.0, + 984.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1089.0, + 394.0, + 1407.0, + 394.0, + 1407.0, + 433.0, + 1089.0, + 433.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 423.0, + 1407.0, + 423.0, + 1407.0, + 462.0, + 292.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 457.0, + 1406.0, + 457.0, + 1406.0, + 491.0, + 296.0, + 491.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 485.0, + 1407.0, + 485.0, + 1407.0, + 523.0, + 295.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 516.0, + 1404.0, + 516.0, + 1404.0, + 552.0, + 292.0, + 552.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 547.0, + 1406.0, + 547.0, + 1406.0, + 585.0, + 295.0, + 585.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 579.0, + 475.0, + 579.0, + 475.0, + 614.0, + 295.0, + 614.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 662.0, + 1408.0, + 662.0, + 1408.0, + 707.0, + 295.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 697.0, + 728.0, + 697.0, + 728.0, + 736.0, + 292.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 783.0, + 697.0, + 899.0, + 697.0, + 899.0, + 736.0, + 783.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1005.0, + 697.0, + 1406.0, + 697.0, + 1406.0, + 736.0, + 1005.0, + 736.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 726.0, + 732.0, + 735.0, + 732.0, + 735.0, + 772.0, + 726.0, + 772.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 619.0, + 1218.0, + 619.0, + 1218.0, + 660.0, + 293.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1136.0, + 276.0, + 1136.0, + 276.0, + 1164.0, + 240.0, + 1164.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1164.0, + 278.0, + 1164.0, + 278.0, + 1196.0, + 239.0, + 1196.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1195.0, + 278.0, + 1195.0, + 278.0, + 1226.0, + 239.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1226.0, + 278.0, + 1226.0, + 278.0, + 1258.0, + 239.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1257.0, + 278.0, + 1257.0, + 278.0, + 1288.0, + 239.0, + 1288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1287.0, + 276.0, + 1287.0, + 276.0, + 1316.0, + 240.0, + 1316.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1318.0, + 276.0, + 1318.0, + 276.0, + 1347.0, + 240.0, + 1347.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1347.0, + 278.0, + 1347.0, + 278.0, + 1378.0, + 239.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1377.0, + 278.0, + 1377.0, + 278.0, + 1407.0, + 239.0, + 1407.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1408.0, + 278.0, + 1408.0, + 278.0, + 1439.0, + 239.0, + 1439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1439.0, + 276.0, + 1439.0, + 276.0, + 1468.0, + 240.0, + 1468.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1466.0, + 280.0, + 1466.0, + 280.0, + 1501.0, + 239.0, + 1501.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1499.0, + 278.0, + 1499.0, + 278.0, + 1530.0, + 237.0, + 1530.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1530.0, + 278.0, + 1530.0, + 278.0, + 1563.0, + 239.0, + 1563.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1574.0, + 278.0, + 1574.0, + 278.0, + 1605.0, + 239.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 241.0, + 1604.0, + 275.0, + 1604.0, + 275.0, + 1627.0, + 241.0, + 1627.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 7, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 245, + 568, + 1404, + 568, + 1404, + 1118, + 245, + 1118 + ], + "score": 0.978 + }, + { + "category_id": 1, + "poly": [ + 290, + 1131, + 1405, + 1131, + 1405, + 1526, + 290, + 1526 + ], + "score": 0.977 + }, + { + "category_id": 1, + "poly": [ + 292, + 1733, + 1404, + 1733, + 1404, + 2008, + 292, + 2008 + ], + "score": 0.975 + }, + { + "category_id": 3, + "poly": [ + 296, + 207, + 1402, + 207, + 1402, + 444, + 296, + 444 + ], + "score": 0.961 + }, + { + "category_id": 1, + "poly": [ + 296, + 1542, + 1404, + 1542, + 1404, + 1633, + 296, + 1633 + ], + "score": 0.959 + }, + { + "category_id": 4, + "poly": [ + 297, + 469, + 1399, + 469, + 1399, + 532, + 297, + 532 + ], + "score": 0.945 + }, + { + "category_id": 0, + "poly": [ + 282, + 1661, + 522, + 1661, + 522, + 1697, + 282, + 1697 + ], + "score": 0.877 + }, + { + "category_id": 2, + "poly": [ + 840, + 2061, + 858, + 2061, + 858, + 2084, + 840, + 2084 + ], + "score": 0.791 + }, + { + "category_id": 13, + "poly": [ + 296, + 687, + 455, + 687, + 455, + 720, + 296, + 720 + ], + "score": 0.93, + "latex": "\\mathbf { W } _ { 2 } \\in \\mathbb { R } ^ { k \\times m }" + }, + { + "category_id": 13, + "poly": [ + 1235, + 657, + 1397, + 657, + 1397, + 690, + 1235, + 690 + ], + "score": 0.92, + "latex": "\\mathbf { W } _ { 1 } \\in \\mathbb { R } ^ { m \\times d }" + }, + { + "category_id": 13, + "poly": [ + 1022, + 1052, + 1244, + 1052, + 1244, + 1087, + 1022, + 1087 + ], + "score": 0.92, + "latex": "\\mathbf { Z } _ { t } \\in \\mathbb { R } ^ { ( d + \\widetilde { p } ) \\times \\widetilde { ( } d + \\widetilde { p } ) }" + }, + { + "category_id": 13, + "poly": [ + 797, + 779, + 955, + 779, + 955, + 811, + 797, + 811 + ], + "score": 0.92, + "latex": "\\mathbf { W _ { 3 } } \\in \\mathbb { R } ^ { k \\times m }" + }, + { + "category_id": 13, + "poly": [ + 297, + 902, + 405, + 902, + 405, + 931, + 297, + 931 + ], + "score": 0.91, + "latex": "H = 1 0 0" + }, + { + "category_id": 13, + "poly": [ + 451, + 779, + 607, + 779, + 607, + 811, + 451, + 811 + ], + "score": 0.91, + "latex": "\\mathbf { W } _ { 1 } \\in \\mathbb { R } ^ { m \\times d }" + }, + { + "category_id": 13, + "poly": [ + 512, + 962, + 600, + 962, + 600, + 995, + 512, + 995 + ], + "score": 0.91, + "latex": "t \\in [ T ]" + }, + { + "category_id": 13, + "poly": [ + 1017, + 780, + 1108, + 780, + 1108, + 809, + 1017, + 809 + ], + "score": 0.91, + "latex": "\\pmb \\theta \\in \\mathbb { R } ^ { k }" + }, + { + "category_id": 13, + "poly": [ + 519, + 688, + 612, + 688, + 612, + 719, + 519, + 719 + ], + "score": 0.91, + "latex": "\\pmb { \\theta } \\in \\mathbb { R } ^ { \\widetilde { k } }" + }, + { + "category_id": 13, + "poly": [ + 1142, + 932, + 1221, + 932, + 1221, + 960, + 1142, + 960 + ], + "score": 0.9, + "latex": "\\lambda = 1" + }, + { + "category_id": 13, + "poly": [ + 620, + 780, + 785, + 780, + 785, + 811, + 620, + 811 + ], + "score": 0.9, + "latex": "\\mathbf { W } _ { 2 } \\in \\mathbb { R } ^ { m \\times m }" + }, + { + "category_id": 13, + "poly": [ + 1276, + 932, + 1402, + 932, + 1402, + 963, + 1276, + 963 + ], + "score": 0.9, + "latex": "\\alpha _ { t } = 0 . 0 2" + }, + { + "category_id": 13, + "poly": [ + 1128, + 629, + 1206, + 629, + 1206, + 657, + 1128, + 657 + ], + "score": 0.9, + "latex": "L = 2" + }, + { + "category_id": 13, + "poly": [ + 1025, + 842, + 1134, + 842, + 1134, + 875, + 1025, + 875 + ], + "score": 0.9, + "latex": "\\eta _ { q } = 1 \\mathrm { e } { - 5 }" + }, + { + "category_id": 13, + "poly": [ + 1285, + 751, + 1397, + 751, + 1397, + 780, + 1285, + 780 + ], + "score": 0.88, + "latex": "m = 1 0 0" + }, + { + "category_id": 13, + "poly": [ + 297, + 811, + 449, + 811, + 449, + 842, + 297, + 842 + ], + "score": 0.87, + "latex": "T = 1 5 , 0 0 0" + }, + { + "category_id": 13, + "poly": [ + 1035, + 751, + 1114, + 751, + 1114, + 778, + 1035, + 778 + ], + "score": 0.86, + "latex": "L = 3" + }, + { + "category_id": 13, + "poly": [ + 963, + 690, + 1090, + 690, + 1090, + 718, + 963, + 718 + ], + "score": 0.86, + "latex": "m = 2 0 0 0" + }, + { + "category_id": 13, + "poly": [ + 1127, + 751, + 1231, + 751, + 1231, + 779, + 1127, + 779 + ], + "score": 0.84, + "latex": "k = 1 0 0" + }, + { + "category_id": 13, + "poly": [ + 846, + 690, + 950, + 690, + 950, + 718, + 846, + 718 + ], + "score": 0.84, + "latex": "k = 1 0 0" + }, + { + "category_id": 13, + "poly": [ + 391, + 872, + 520, + 872, + 520, + 902, + 391, + 902 + ], + "score": 0.8, + "latex": "n = 1 , 0 0 0" + }, + { + "category_id": 13, + "poly": [ + 1153, + 691, + 1172, + 691, + 1172, + 717, + 1153, + 717 + ], + "score": 0.79, + "latex": "d" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 200.0, + 345.0, + 200.0, + 345.0, + 345.0, + 294.0, + 345.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 460.0, + 253.0, + 471.0, + 253.0, + 471.0, + 262.0, + 460.0, + 262.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 472.0, + 216.0, + 548.0, + 216.0, + 548.0, + 269.0, + 472.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 554.0, + 225.0, + 639.0, + 225.0, + 639.0, + 362.0, + 554.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 656.0, + 216.0, + 731.0, + 216.0, + 731.0, + 271.0, + 656.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 841.0, + 210.0, + 908.0, + 210.0, + 908.0, + 298.0, + 841.0, + 298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 911.0, + 216.0, + 1006.0, + 216.0, + 1006.0, + 268.0, + 911.0, + 268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1110.0, + 216.0, + 1285.0, + 216.0, + 1285.0, + 364.0, + 1110.0, + 364.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 266.0, + 345.0, + 266.0, + 345.0, + 290.0, + 314.0, + 290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 472.0, + 264.0, + 559.0, + 264.0, + 559.0, + 285.0, + 472.0, + 285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 657.0, + 264.0, + 741.0, + 264.0, + 741.0, + 285.0, + 657.0, + 285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 823.0, + 262.0, + 829.0, + 262.0, + 829.0, + 269.0, + 823.0, + 269.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 836.0, + 265.0, + 912.0, + 265.0, + 912.0, + 357.0, + 836.0, + 357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 933.0, + 264.0, + 1019.0, + 264.0, + 1019.0, + 285.0, + 933.0, + 285.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 272.0, + 1109.0, + 272.0, + 1109.0, + 287.0, + 1091.0, + 287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 262.0, + 1296.0, + 262.0, + 1296.0, + 287.0, + 1207.0, + 287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1140.0, + 287.0, + 1174.0, + 287.0, + 1174.0, + 311.0, + 1140.0, + 311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 298.0, + 345.0, + 298.0, + 345.0, + 323.0, + 314.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1139.0, + 310.0, + 1174.0, + 310.0, + 1174.0, + 335.0, + 1139.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 314.0, + 332.0, + 345.0, + 332.0, + 345.0, + 356.0, + 314.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 585.0, + 324.0, + 623.0, + 324.0, + 623.0, + 348.0, + 585.0, + 348.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 862.0, + 339.0, + 899.0, + 339.0, + 899.0, + 362.0, + 862.0, + 362.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1143.0, + 333.0, + 1176.0, + 333.0, + 1176.0, + 358.0, + 1143.0, + 358.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 605.0, + 361.0, + 620.0, + 361.0, + 620.0, + 377.0, + 605.0, + 377.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 880.0, + 360.0, + 897.0, + 360.0, + 897.0, + 378.0, + 880.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1157.0, + 360.0, + 1173.0, + 360.0, + 1173.0, + 378.0, + 1157.0, + 378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 328.0, + 369.0, + 343.0, + 369.0, + 343.0, + 385.0, + 328.0, + 385.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 347.0, + 378.0, + 362.0, + 378.0, + 362.0, + 395.0, + 347.0, + 395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 375.0, + 374.0, + 414.0, + 374.0, + 414.0, + 398.0, + 375.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 417.0, + 376.0, + 576.0, + 376.0, + 576.0, + 397.0, + 417.0, + 397.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 624.0, + 378.0, + 640.0, + 378.0, + 640.0, + 395.0, + 624.0, + 395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 652.0, + 376.0, + 853.0, + 376.0, + 853.0, + 398.0, + 652.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 899.0, + 378.0, + 917.0, + 378.0, + 917.0, + 395.0, + 899.0, + 395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 927.0, + 376.0, + 1129.0, + 376.0, + 1129.0, + 398.0, + 927.0, + 398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1176.0, + 378.0, + 1193.0, + 378.0, + 1193.0, + 395.0, + 1176.0, + 395.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1205.0, + 374.0, + 1407.0, + 374.0, + 1407.0, + 399.0, + 1205.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 432.0, + 389.0, + 476.0, + 389.0, + 476.0, + 411.0, + 432.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 710.0, + 389.0, + 752.0, + 389.0, + 752.0, + 411.0, + 710.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 986.0, + 389.0, + 1029.0, + 389.0, + 1029.0, + 411.0, + 986.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1262.0, + 389.0, + 1305.0, + 389.0, + 1305.0, + 411.0, + 1262.0, + 411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 374.0, + 405.0, + 498.0, + 405.0, + 498.0, + 450.0, + 374.0, + 450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 653.0, + 407.0, + 771.0, + 407.0, + 771.0, + 447.0, + 653.0, + 447.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 910.0, + 407.0, + 1067.0, + 407.0, + 1067.0, + 448.0, + 910.0, + 448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1198.0, + 407.0, + 1331.0, + 407.0, + 1331.0, + 442.0, + 1198.0, + 442.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 392.0, + 248.0, + 440.0, + 248.0, + 440.0, + 259.0, + 392.0, + 259.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 649.0, + 311.5, + 733.0, + 311.5, + 733.0, + 337.5, + 649.0, + 337.5 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 469.0, + 1404.0, + 469.0, + 1404.0, + 506.0, + 293.0, + 506.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 501.0, + 988.0, + 501.0, + 988.0, + 536.0, + 295.0, + 536.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1657.0, + 528.0, + 1657.0, + 528.0, + 1705.0, + 293.0, + 1705.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 839.0, + 2060.0, + 860.0, + 2060.0, + 860.0, + 2090.0, + 839.0, + 2090.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 572.0, + 278.0, + 572.0, + 278.0, + 602.0, + 238.0, + 602.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 567.0, + 1405.0, + 567.0, + 1405.0, + 605.0, + 295.0, + 605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 602.0, + 279.0, + 602.0, + 279.0, + 632.0, + 237.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 598.0, + 1404.0, + 598.0, + 1404.0, + 632.0, + 295.0, + 632.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 633.0, + 279.0, + 633.0, + 279.0, + 663.0, + 237.0, + 663.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 628.0, + 1127.0, + 628.0, + 1127.0, + 662.0, + 295.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1207.0, + 628.0, + 1405.0, + 628.0, + 1405.0, + 662.0, + 1207.0, + 662.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 664.0, + 279.0, + 664.0, + 279.0, + 693.0, + 237.0, + 693.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 652.0, + 1234.0, + 652.0, + 1234.0, + 697.0, + 288.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 652.0, + 1409.0, + 652.0, + 1409.0, + 697.0, + 1398.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 694.0, + 279.0, + 694.0, + 279.0, + 722.0, + 237.0, + 722.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 682.0, + 295.0, + 682.0, + 295.0, + 728.0, + 291.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 456.0, + 682.0, + 518.0, + 682.0, + 518.0, + 728.0, + 456.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 613.0, + 682.0, + 845.0, + 682.0, + 845.0, + 728.0, + 613.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 951.0, + 682.0, + 962.0, + 682.0, + 962.0, + 728.0, + 951.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1091.0, + 682.0, + 1152.0, + 682.0, + 1152.0, + 728.0, + 1091.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1173.0, + 682.0, + 1408.0, + 682.0, + 1408.0, + 728.0, + 1173.0, + 728.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 724.0, + 279.0, + 724.0, + 279.0, + 754.0, + 237.0, + 754.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 718.0, + 1404.0, + 718.0, + 1404.0, + 752.0, + 293.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 755.0, + 279.0, + 755.0, + 279.0, + 785.0, + 237.0, + 785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 748.0, + 1034.0, + 748.0, + 1034.0, + 785.0, + 292.0, + 785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1115.0, + 748.0, + 1126.0, + 748.0, + 1126.0, + 785.0, + 1115.0, + 785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1232.0, + 748.0, + 1284.0, + 748.0, + 1284.0, + 785.0, + 1232.0, + 785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1398.0, + 748.0, + 1408.0, + 748.0, + 1408.0, + 785.0, + 1398.0, + 785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 785.0, + 279.0, + 785.0, + 279.0, + 815.0, + 237.0, + 815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 773.0, + 450.0, + 773.0, + 450.0, + 819.0, + 289.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 608.0, + 773.0, + 619.0, + 773.0, + 619.0, + 819.0, + 608.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 786.0, + 773.0, + 796.0, + 773.0, + 796.0, + 819.0, + 786.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 956.0, + 773.0, + 1016.0, + 773.0, + 1016.0, + 819.0, + 956.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1109.0, + 773.0, + 1411.0, + 773.0, + 1411.0, + 819.0, + 1109.0, + 819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 815.0, + 279.0, + 815.0, + 279.0, + 844.0, + 237.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 810.0, + 296.0, + 810.0, + 296.0, + 844.0, + 293.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 810.0, + 1405.0, + 810.0, + 1405.0, + 844.0, + 450.0, + 844.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 847.0, + 278.0, + 847.0, + 278.0, + 873.0, + 240.0, + 873.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 839.0, + 1024.0, + 839.0, + 1024.0, + 877.0, + 293.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1135.0, + 839.0, + 1405.0, + 839.0, + 1405.0, + 877.0, + 1135.0, + 877.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 877.0, + 276.0, + 877.0, + 276.0, + 903.0, + 238.0, + 903.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 867.0, + 390.0, + 867.0, + 390.0, + 911.0, + 291.0, + 911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 521.0, + 867.0, + 1405.0, + 867.0, + 1405.0, + 911.0, + 521.0, + 911.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 905.0, + 279.0, + 905.0, + 279.0, + 935.0, + 238.0, + 935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 900.0, + 296.0, + 900.0, + 296.0, + 938.0, + 293.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 900.0, + 1407.0, + 900.0, + 1407.0, + 938.0, + 406.0, + 938.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 938.0, + 278.0, + 938.0, + 278.0, + 964.0, + 240.0, + 964.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 931.0, + 1141.0, + 931.0, + 1141.0, + 965.0, + 293.0, + 965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1222.0, + 931.0, + 1275.0, + 931.0, + 1275.0, + 965.0, + 1222.0, + 965.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 966.0, + 278.0, + 966.0, + 278.0, + 996.0, + 238.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 959.0, + 511.0, + 959.0, + 511.0, + 999.0, + 292.0, + 999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 601.0, + 959.0, + 1407.0, + 959.0, + 1407.0, + 999.0, + 601.0, + 999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 999.0, + 276.0, + 999.0, + 276.0, + 1024.0, + 240.0, + 1024.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 993.0, + 1405.0, + 993.0, + 1405.0, + 1027.0, + 295.0, + 1027.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1030.0, + 276.0, + 1030.0, + 276.0, + 1054.0, + 240.0, + 1054.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1023.0, + 1408.0, + 1023.0, + 1408.0, + 1057.0, + 293.0, + 1057.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1064.0, + 276.0, + 1064.0, + 276.0, + 1088.0, + 240.0, + 1088.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1049.0, + 1021.0, + 1049.0, + 1021.0, + 1093.0, + 292.0, + 1093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 1245.0, + 1049.0, + 1413.0, + 1049.0, + 1413.0, + 1093.0, + 1245.0, + 1093.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 240.0, + 1092.0, + 278.0, + 1092.0, + 278.0, + 1118.0, + 240.0, + 1118.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1087.0, + 628.0, + 1087.0, + 628.0, + 1121.0, + 296.0, + 1121.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1131.0, + 1407.0, + 1131.0, + 1407.0, + 1167.0, + 296.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1160.0, + 1407.0, + 1160.0, + 1407.0, + 1197.0, + 293.0, + 1197.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1192.0, + 1403.0, + 1192.0, + 1403.0, + 1224.0, + 296.0, + 1224.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1222.0, + 1404.0, + 1222.0, + 1404.0, + 1258.0, + 293.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1252.0, + 1406.0, + 1252.0, + 1406.0, + 1290.0, + 292.0, + 1290.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1281.0, + 1407.0, + 1281.0, + 1407.0, + 1320.0, + 293.0, + 1320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1313.0, + 1407.0, + 1313.0, + 1407.0, + 1349.0, + 293.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1343.0, + 1408.0, + 1343.0, + 1408.0, + 1380.0, + 292.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1376.0, + 1406.0, + 1376.0, + 1406.0, + 1408.0, + 297.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1405.0, + 1407.0, + 1405.0, + 1407.0, + 1441.0, + 293.0, + 1441.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1436.0, + 1404.0, + 1436.0, + 1404.0, + 1469.0, + 293.0, + 1469.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1464.0, + 1404.0, + 1464.0, + 1404.0, + 1503.0, + 293.0, + 1503.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 1495.0, + 722.0, + 1495.0, + 722.0, + 1532.0, + 296.0, + 1532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1734.0, + 1406.0, + 1734.0, + 1406.0, + 1770.0, + 294.0, + 1770.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1761.0, + 1406.0, + 1761.0, + 1406.0, + 1799.0, + 293.0, + 1799.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 1792.0, + 1407.0, + 1792.0, + 1407.0, + 1831.0, + 291.0, + 1831.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1824.0, + 1406.0, + 1824.0, + 1406.0, + 1860.0, + 293.0, + 1860.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1856.0, + 1405.0, + 1856.0, + 1405.0, + 1889.0, + 294.0, + 1889.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1885.0, + 1403.0, + 1885.0, + 1403.0, + 1917.0, + 295.0, + 1917.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1913.0, + 1406.0, + 1913.0, + 1406.0, + 1949.0, + 292.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1946.0, + 1406.0, + 1946.0, + 1406.0, + 1982.0, + 294.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1973.0, + 947.0, + 1973.0, + 947.0, + 2011.0, + 293.0, + 2011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1540.0, + 1405.0, + 1540.0, + 1405.0, + 1577.0, + 294.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1572.0, + 1404.0, + 1572.0, + 1404.0, + 1605.0, + 295.0, + 1605.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1602.0, + 1334.0, + 1602.0, + 1334.0, + 1635.0, + 295.0, + 1635.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 8, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 836, + 2061, + 866, + 2061, + 866, + 2086, + 836, + 2086 + ], + "score": 0.829 + }, + { + "category_id": 0, + "poly": [ + 257, + 200, + 456, + 200, + 456, + 235, + 257, + 235 + ], + "score": 0.779 + }, + { + "category_id": 1, + "poly": [ + 253, + 161, + 1411, + 161, + 1411, + 2025, + 253, + 2025 + ], + "score": 0.721 + }, + { + "category_id": 2, + "poly": [ + 238, + 213, + 286, + 213, + 286, + 2042, + 238, + 2042 + ], + "score": 0.161 + }, + { + "category_id": 2, + "poly": [ + 238, + 235, + 287, + 235, + 287, + 2043, + 238, + 2043 + ], + "score": 0.103 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2058.0, + 870.0, + 2058.0, + 870.0, + 2096.0, + 832.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 252.0, + 208.0, + 279.0, + 208.0, + 279.0, + 234.0, + 252.0, + 234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 196.0, + 459.0, + 196.0, + 459.0, + 240.0, + 291.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 209.0, + 280.0, + 209.0, + 280.0, + 239.0, + 237.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 255.0, + 280.0, + 255.0, + 280.0, + 288.0, + 237.0, + 288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 288.0, + 280.0, + 288.0, + 280.0, + 320.0, + 237.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 318.0, + 280.0, + 318.0, + 280.0, + 352.0, + 237.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 372.0, + 280.0, + 372.0, + 280.0, + 404.0, + 237.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 402.0, + 280.0, + 402.0, + 280.0, + 436.0, + 237.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 432.0, + 280.0, + 432.0, + 280.0, + 470.0, + 237.0, + 470.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 489.0, + 280.0, + 489.0, + 280.0, + 521.0, + 237.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 517.0, + 280.0, + 517.0, + 280.0, + 551.0, + 237.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 547.0, + 283.0, + 547.0, + 283.0, + 589.0, + 234.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 601.0, + 280.0, + 601.0, + 280.0, + 635.0, + 237.0, + 635.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 633.0, + 280.0, + 633.0, + 280.0, + 665.0, + 237.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 665.0, + 280.0, + 665.0, + 280.0, + 697.0, + 237.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 718.0, + 280.0, + 718.0, + 280.0, + 752.0, + 237.0, + 752.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 750.0, + 280.0, + 750.0, + 280.0, + 782.0, + 237.0, + 782.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 780.0, + 280.0, + 780.0, + 280.0, + 814.0, + 237.0, + 814.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 834.0, + 280.0, + 834.0, + 280.0, + 866.0, + 237.0, + 866.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 864.0, + 280.0, + 864.0, + 280.0, + 896.0, + 237.0, + 896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 892.0, + 283.0, + 892.0, + 283.0, + 935.0, + 234.0, + 935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 947.0, + 280.0, + 947.0, + 280.0, + 981.0, + 237.0, + 981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 979.0, + 280.0, + 979.0, + 280.0, + 1011.0, + 237.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1009.0, + 283.0, + 1009.0, + 283.0, + 1049.0, + 234.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1063.0, + 280.0, + 1063.0, + 280.0, + 1097.0, + 237.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1093.0, + 280.0, + 1093.0, + 280.0, + 1127.0, + 237.0, + 1127.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1123.0, + 283.0, + 1123.0, + 283.0, + 1164.0, + 234.0, + 1164.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1178.0, + 280.0, + 1178.0, + 280.0, + 1212.0, + 237.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1206.0, + 280.0, + 1206.0, + 280.0, + 1249.0, + 238.0, + 1249.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1264.0, + 280.0, + 1264.0, + 280.0, + 1296.0, + 237.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1294.0, + 280.0, + 1294.0, + 280.0, + 1328.0, + 237.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1322.0, + 283.0, + 1322.0, + 283.0, + 1365.0, + 234.0, + 1365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1377.0, + 280.0, + 1377.0, + 280.0, + 1411.0, + 237.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1409.0, + 283.0, + 1409.0, + 283.0, + 1449.0, + 234.0, + 1449.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1459.0, + 283.0, + 1459.0, + 283.0, + 1499.0, + 234.0, + 1499.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1493.0, + 283.0, + 1493.0, + 283.0, + 1539.0, + 234.0, + 1539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1549.0, + 280.0, + 1549.0, + 280.0, + 1584.0, + 237.0, + 1584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1580.0, + 280.0, + 1580.0, + 280.0, + 1612.0, + 237.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1608.0, + 280.0, + 1608.0, + 280.0, + 1650.0, + 234.0, + 1650.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1664.0, + 280.0, + 1664.0, + 280.0, + 1698.0, + 237.0, + 1698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1696.0, + 280.0, + 1696.0, + 280.0, + 1730.0, + 237.0, + 1730.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1750.0, + 280.0, + 1750.0, + 280.0, + 1783.0, + 237.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1779.0, + 283.0, + 1779.0, + 283.0, + 1821.0, + 234.0, + 1821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1833.0, + 280.0, + 1833.0, + 280.0, + 1867.0, + 237.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1865.0, + 280.0, + 1865.0, + 280.0, + 1897.0, + 237.0, + 1897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1895.0, + 280.0, + 1895.0, + 280.0, + 1929.0, + 237.0, + 1929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1949.0, + 280.0, + 1949.0, + 280.0, + 1981.0, + 237.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1979.0, + 280.0, + 1979.0, + 280.0, + 2014.0, + 237.0, + 2014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 257.0, + 281.0, + 257.0, + 281.0, + 288.0, + 235.0, + 288.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 288.0, + 281.0, + 288.0, + 281.0, + 320.0, + 237.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 316.0, + 283.0, + 316.0, + 283.0, + 356.0, + 235.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 372.0, + 281.0, + 372.0, + 281.0, + 404.0, + 235.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 404.0, + 281.0, + 404.0, + 281.0, + 435.0, + 235.0, + 435.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 433.0, + 279.0, + 433.0, + 279.0, + 467.0, + 237.0, + 467.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 487.0, + 281.0, + 487.0, + 281.0, + 521.0, + 237.0, + 521.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 517.0, + 281.0, + 517.0, + 281.0, + 551.0, + 237.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 547.0, + 281.0, + 547.0, + 281.0, + 590.0, + 235.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 602.0, + 281.0, + 602.0, + 281.0, + 636.0, + 237.0, + 636.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 632.0, + 281.0, + 632.0, + 281.0, + 666.0, + 237.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 662.0, + 283.0, + 662.0, + 283.0, + 702.0, + 235.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 718.0, + 281.0, + 718.0, + 281.0, + 749.0, + 237.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 749.0, + 281.0, + 749.0, + 281.0, + 781.0, + 237.0, + 781.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 781.0, + 281.0, + 781.0, + 281.0, + 813.0, + 237.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 833.0, + 281.0, + 833.0, + 281.0, + 867.0, + 237.0, + 867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 865.0, + 281.0, + 865.0, + 281.0, + 897.0, + 237.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 893.0, + 283.0, + 893.0, + 283.0, + 934.0, + 235.0, + 934.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 948.0, + 281.0, + 948.0, + 281.0, + 982.0, + 237.0, + 982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 978.0, + 281.0, + 978.0, + 281.0, + 1012.0, + 237.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1010.0, + 281.0, + 1010.0, + 281.0, + 1044.0, + 237.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1063.0, + 281.0, + 1063.0, + 281.0, + 1095.0, + 237.0, + 1095.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1093.0, + 281.0, + 1093.0, + 281.0, + 1127.0, + 237.0, + 1127.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1123.0, + 283.0, + 1123.0, + 283.0, + 1163.0, + 235.0, + 1163.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1179.0, + 281.0, + 1179.0, + 281.0, + 1211.0, + 237.0, + 1211.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1209.0, + 281.0, + 1209.0, + 281.0, + 1252.0, + 235.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1262.0, + 281.0, + 1262.0, + 281.0, + 1296.0, + 237.0, + 1296.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1294.0, + 281.0, + 1294.0, + 281.0, + 1328.0, + 237.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1326.0, + 279.0, + 1326.0, + 279.0, + 1360.0, + 237.0, + 1360.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1377.0, + 281.0, + 1377.0, + 281.0, + 1411.0, + 237.0, + 1411.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1407.0, + 283.0, + 1407.0, + 283.0, + 1451.0, + 235.0, + 1451.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1463.0, + 279.0, + 1463.0, + 279.0, + 1497.0, + 237.0, + 1497.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1495.0, + 283.0, + 1495.0, + 283.0, + 1538.0, + 235.0, + 1538.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1548.0, + 281.0, + 1548.0, + 281.0, + 1580.0, + 237.0, + 1580.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1580.0, + 281.0, + 1580.0, + 281.0, + 1612.0, + 235.0, + 1612.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1610.0, + 281.0, + 1610.0, + 281.0, + 1648.0, + 237.0, + 1648.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1664.0, + 281.0, + 1664.0, + 281.0, + 1697.0, + 237.0, + 1697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1694.0, + 283.0, + 1694.0, + 283.0, + 1737.0, + 235.0, + 1737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1749.0, + 281.0, + 1749.0, + 281.0, + 1783.0, + 237.0, + 1783.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1779.0, + 283.0, + 1779.0, + 283.0, + 1821.0, + 235.0, + 1821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 232.0, + 1831.0, + 283.0, + 1831.0, + 283.0, + 1868.0, + 232.0, + 1868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1864.0, + 281.0, + 1864.0, + 281.0, + 1896.0, + 237.0, + 1896.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1896.0, + 281.0, + 1896.0, + 281.0, + 1930.0, + 237.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1950.0, + 281.0, + 1950.0, + 281.0, + 1982.0, + 237.0, + 1982.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 232.0, + 1978.0, + 283.0, + 1978.0, + 283.0, + 2015.0, + 232.0, + 2015.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 249.0, + 197.0, + 462.0, + 197.0, + 462.0, + 240.0, + 249.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 254.0, + 282.0, + 254.0, + 282.0, + 287.0, + 244.0, + 287.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 252.0, + 1407.0, + 252.0, + 1407.0, + 291.0, + 306.0, + 291.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 285.0, + 280.0, + 285.0, + 280.0, + 320.0, + 244.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 281.0, + 1407.0, + 281.0, + 1407.0, + 320.0, + 352.0, + 320.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 318.0, + 280.0, + 318.0, + 280.0, + 352.0, + 244.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 311.0, + 431.0, + 311.0, + 431.0, + 353.0, + 354.0, + 353.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 373.0, + 282.0, + 373.0, + 282.0, + 404.0, + 244.0, + 404.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 369.0, + 1409.0, + 369.0, + 1409.0, + 408.0, + 306.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 402.0, + 280.0, + 402.0, + 280.0, + 434.0, + 244.0, + 434.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 397.0, + 1407.0, + 397.0, + 1407.0, + 436.0, + 354.0, + 436.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 432.0, + 280.0, + 432.0, + 280.0, + 465.0, + 244.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 428.0, + 1022.0, + 428.0, + 1022.0, + 465.0, + 354.0, + 465.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 483.0, + 282.0, + 483.0, + 282.0, + 518.0, + 246.0, + 518.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 481.0, + 1407.0, + 481.0, + 1407.0, + 520.0, + 306.0, + 520.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 516.0, + 280.0, + 516.0, + 280.0, + 551.0, + 244.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 510.0, + 1407.0, + 510.0, + 1407.0, + 557.0, + 350.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 549.0, + 280.0, + 549.0, + 280.0, + 584.0, + 246.0, + 584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 541.0, + 735.0, + 541.0, + 735.0, + 584.0, + 350.0, + 584.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 602.0, + 282.0, + 602.0, + 282.0, + 633.0, + 246.0, + 633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 598.0, + 1407.0, + 598.0, + 1407.0, + 637.0, + 306.0, + 637.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 631.0, + 280.0, + 631.0, + 280.0, + 665.0, + 244.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 627.0, + 1405.0, + 627.0, + 1405.0, + 665.0, + 352.0, + 665.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 663.0, + 280.0, + 663.0, + 280.0, + 696.0, + 244.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 657.0, + 431.0, + 657.0, + 431.0, + 698.0, + 352.0, + 698.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 717.0, + 282.0, + 717.0, + 282.0, + 749.0, + 246.0, + 749.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 710.0, + 1409.0, + 710.0, + 1409.0, + 755.0, + 304.0, + 755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 749.0, + 282.0, + 749.0, + 282.0, + 780.0, + 246.0, + 780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 739.0, + 1411.0, + 739.0, + 1411.0, + 788.0, + 350.0, + 788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 778.0, + 280.0, + 778.0, + 280.0, + 813.0, + 246.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 776.0, + 720.0, + 776.0, + 720.0, + 813.0, + 352.0, + 813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 831.0, + 280.0, + 831.0, + 280.0, + 864.0, + 246.0, + 864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 829.0, + 1409.0, + 829.0, + 1409.0, + 868.0, + 306.0, + 868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 864.0, + 282.0, + 864.0, + 282.0, + 897.0, + 246.0, + 897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 862.0, + 1405.0, + 862.0, + 1405.0, + 895.0, + 354.0, + 895.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 893.0, + 280.0, + 893.0, + 280.0, + 925.0, + 246.0, + 925.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 886.0, + 997.0, + 886.0, + 997.0, + 929.0, + 350.0, + 929.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 946.0, + 282.0, + 946.0, + 282.0, + 978.0, + 246.0, + 978.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 302.0, + 942.0, + 1409.0, + 942.0, + 1409.0, + 985.0, + 302.0, + 985.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 978.0, + 282.0, + 978.0, + 282.0, + 1011.0, + 246.0, + 1011.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 974.0, + 1407.0, + 974.0, + 1407.0, + 1013.0, + 352.0, + 1013.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1009.0, + 280.0, + 1009.0, + 280.0, + 1044.0, + 246.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1005.0, + 869.0, + 1005.0, + 869.0, + 1044.0, + 352.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1062.0, + 280.0, + 1062.0, + 280.0, + 1095.0, + 246.0, + 1095.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 306.0, + 1058.0, + 1407.0, + 1058.0, + 1407.0, + 1097.0, + 306.0, + 1097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1093.0, + 282.0, + 1093.0, + 282.0, + 1126.0, + 246.0, + 1126.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1089.0, + 1409.0, + 1089.0, + 1409.0, + 1128.0, + 352.0, + 1128.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1126.0, + 280.0, + 1126.0, + 280.0, + 1158.0, + 246.0, + 1158.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 1122.0, + 563.0, + 1122.0, + 563.0, + 1154.0, + 356.0, + 1154.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 1179.0, + 282.0, + 1179.0, + 282.0, + 1212.0, + 244.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 304.0, + 1175.0, + 1405.0, + 1175.0, + 1405.0, + 1212.0, + 304.0, + 1212.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 1210.0, + 280.0, + 1210.0, + 280.0, + 1242.0, + 244.0, + 1242.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1206.0, + 1024.0, + 1206.0, + 1024.0, + 1244.0, + 352.0, + 1244.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1259.0, + 1407.0, + 1259.0, + 1407.0, + 1298.0, + 246.0, + 1298.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 1293.0, + 282.0, + 1293.0, + 282.0, + 1328.0, + 244.0, + 1328.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1291.0, + 1405.0, + 1291.0, + 1405.0, + 1324.0, + 354.0, + 1324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 1324.0, + 282.0, + 1324.0, + 282.0, + 1357.0, + 244.0, + 1357.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1318.0, + 431.0, + 1318.0, + 431.0, + 1359.0, + 352.0, + 1359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1379.0, + 282.0, + 1379.0, + 282.0, + 1410.0, + 246.0, + 1410.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 1369.0, + 1409.0, + 1369.0, + 1409.0, + 1418.0, + 286.0, + 1418.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 1410.0, + 280.0, + 1410.0, + 280.0, + 1443.0, + 244.0, + 1443.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1406.0, + 1366.0, + 1406.0, + 1366.0, + 1445.0, + 354.0, + 1445.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1459.0, + 1409.0, + 1459.0, + 1409.0, + 1498.0, + 246.0, + 1498.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 1494.0, + 280.0, + 1494.0, + 280.0, + 1527.0, + 244.0, + 1527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1488.0, + 1200.0, + 1488.0, + 1200.0, + 1531.0, + 350.0, + 1531.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1547.0, + 284.0, + 1547.0, + 284.0, + 1578.0, + 246.0, + 1578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1543.0, + 1407.0, + 1543.0, + 1407.0, + 1582.0, + 290.0, + 1582.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 350.0, + 1570.0, + 1409.0, + 1570.0, + 1409.0, + 1619.0, + 350.0, + 1619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 1611.0, + 280.0, + 1611.0, + 280.0, + 1641.0, + 244.0, + 1641.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1600.0, + 594.0, + 1600.0, + 594.0, + 1644.0, + 352.0, + 1644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1662.0, + 284.0, + 1662.0, + 284.0, + 1694.0, + 246.0, + 1694.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1660.0, + 1405.0, + 1660.0, + 1405.0, + 1699.0, + 292.0, + 1699.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1694.0, + 280.0, + 1694.0, + 280.0, + 1729.0, + 246.0, + 1729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 1688.0, + 1111.0, + 1688.0, + 1111.0, + 1731.0, + 348.0, + 1731.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1750.0, + 288.0, + 1750.0, + 288.0, + 1780.0, + 246.0, + 1780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1746.0, + 1407.0, + 1746.0, + 1407.0, + 1784.0, + 290.0, + 1784.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 1780.0, + 280.0, + 1780.0, + 280.0, + 1813.0, + 244.0, + 1813.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1776.0, + 995.0, + 1776.0, + 995.0, + 1815.0, + 354.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1834.0, + 284.0, + 1834.0, + 284.0, + 1864.0, + 246.0, + 1864.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1829.0, + 1407.0, + 1829.0, + 1407.0, + 1868.0, + 290.0, + 1868.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 1864.0, + 282.0, + 1864.0, + 282.0, + 1897.0, + 244.0, + 1897.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1860.0, + 1407.0, + 1860.0, + 1407.0, + 1899.0, + 354.0, + 1899.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1895.0, + 282.0, + 1895.0, + 282.0, + 1928.0, + 246.0, + 1928.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 354.0, + 1891.0, + 774.0, + 1891.0, + 774.0, + 1930.0, + 354.0, + 1930.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 246.0, + 1950.0, + 284.0, + 1950.0, + 284.0, + 1981.0, + 246.0, + 1981.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 286.0, + 1940.0, + 1409.0, + 1940.0, + 1409.0, + 1987.0, + 286.0, + 1987.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.0, + 1979.0, + 280.0, + 1979.0, + 280.0, + 2012.0, + 244.0, + 2012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 1977.0, + 640.0, + 1977.0, + 640.0, + 2010.0, + 356.0, + 2010.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 244.75, + 1577.5, + 276.75, + 1577.5, + 276.75, + 1611.0, + 244.75, + 1611.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 9, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 835, + 2061, + 864, + 2061, + 864, + 2087, + 835, + 2087 + ], + "score": 0.827 + }, + { + "category_id": 1, + "poly": [ + 274, + 112, + 1410, + 112, + 1410, + 2027, + 274, + 2027 + ], + "score": 0.76 + }, + { + "category_id": 2, + "poly": [ + 238, + 205, + 285, + 205, + 285, + 2027, + 238, + 2027 + ], + "score": 0.188 + }, + { + "category_id": 15, + "poly": [ + 831.0, + 2058.0, + 869.0, + 2058.0, + 869.0, + 2097.0, + 831.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 207.0, + 278.0, + 207.0, + 278.0, + 239.0, + 236.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 239.0, + 280.0, + 239.0, + 280.0, + 273.0, + 236.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 291.0, + 280.0, + 291.0, + 280.0, + 323.0, + 236.0, + 323.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 319.0, + 282.0, + 319.0, + 282.0, + 363.0, + 234.0, + 363.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 373.0, + 282.0, + 373.0, + 282.0, + 405.0, + 236.0, + 405.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 401.0, + 284.0, + 401.0, + 284.0, + 439.0, + 234.0, + 439.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 433.0, + 282.0, + 433.0, + 282.0, + 473.0, + 234.0, + 473.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 483.0, + 284.0, + 483.0, + 284.0, + 523.0, + 234.0, + 523.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 517.0, + 280.0, + 517.0, + 280.0, + 551.0, + 236.0, + 551.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 547.0, + 282.0, + 547.0, + 282.0, + 589.0, + 234.0, + 589.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 599.0, + 280.0, + 599.0, + 280.0, + 633.0, + 236.0, + 633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 627.0, + 284.0, + 627.0, + 284.0, + 666.0, + 234.0, + 666.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 660.0, + 282.0, + 660.0, + 282.0, + 702.0, + 234.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 712.0, + 284.0, + 712.0, + 284.0, + 750.0, + 234.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 744.0, + 282.0, + 744.0, + 282.0, + 790.0, + 234.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 794.0, + 284.0, + 794.0, + 284.0, + 834.0, + 234.0, + 834.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 830.0, + 280.0, + 830.0, + 280.0, + 862.0, + 236.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 858.0, + 282.0, + 858.0, + 282.0, + 902.0, + 234.0, + 902.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 912.0, + 282.0, + 912.0, + 282.0, + 950.0, + 239.0, + 950.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 964.0, + 280.0, + 964.0, + 280.0, + 996.0, + 236.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 994.0, + 280.0, + 994.0, + 280.0, + 1038.0, + 234.0, + 1038.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1044.0, + 284.0, + 1044.0, + 284.0, + 1082.0, + 234.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1078.0, + 280.0, + 1078.0, + 280.0, + 1112.0, + 236.0, + 1112.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1108.0, + 282.0, + 1108.0, + 282.0, + 1148.0, + 234.0, + 1148.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1160.0, + 280.0, + 1160.0, + 280.0, + 1194.0, + 236.0, + 1194.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1188.0, + 284.0, + 1188.0, + 284.0, + 1226.0, + 234.0, + 1226.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1222.0, + 282.0, + 1222.0, + 282.0, + 1264.0, + 236.0, + 1264.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1276.0, + 280.0, + 1276.0, + 280.0, + 1308.0, + 236.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1304.0, + 282.0, + 1304.0, + 282.0, + 1346.0, + 234.0, + 1346.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1356.0, + 282.0, + 1356.0, + 282.0, + 1398.0, + 236.0, + 1398.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1412.0, + 278.0, + 1412.0, + 278.0, + 1444.0, + 236.0, + 1444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1440.0, + 282.0, + 1440.0, + 282.0, + 1486.0, + 234.0, + 1486.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1492.0, + 280.0, + 1492.0, + 280.0, + 1526.0, + 236.0, + 1526.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1524.0, + 284.0, + 1524.0, + 284.0, + 1562.0, + 234.0, + 1562.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1574.0, + 284.0, + 1574.0, + 284.0, + 1615.0, + 234.0, + 1615.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1605.0, + 284.0, + 1605.0, + 284.0, + 1643.0, + 234.0, + 1643.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1637.0, + 280.0, + 1637.0, + 280.0, + 1671.0, + 236.0, + 1671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1667.0, + 282.0, + 1667.0, + 282.0, + 1711.0, + 234.0, + 1711.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1721.0, + 280.0, + 1721.0, + 280.0, + 1755.0, + 236.0, + 1755.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1753.0, + 280.0, + 1753.0, + 280.0, + 1785.0, + 236.0, + 1785.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1785.0, + 278.0, + 1785.0, + 278.0, + 1819.0, + 236.0, + 1819.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1835.0, + 280.0, + 1835.0, + 280.0, + 1869.0, + 236.0, + 1869.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1865.0, + 282.0, + 1865.0, + 282.0, + 1909.0, + 234.0, + 1909.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1915.0, + 284.0, + 1915.0, + 284.0, + 1953.0, + 234.0, + 1953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1949.0, + 280.0, + 1949.0, + 280.0, + 1983.0, + 236.0, + 1983.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1977.0, + 282.0, + 1977.0, + 282.0, + 2017.0, + 234.0, + 2017.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 291.0, + 201.0, + 1411.0, + 201.0, + 1411.0, + 243.0, + 291.0, + 243.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 348.0, + 230.0, + 915.0, + 230.0, + 915.0, + 270.0, + 348.0, + 270.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 284.0, + 1408.0, + 284.0, + 1408.0, + 324.0, + 293.0, + 324.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 318.0, + 1016.0, + 318.0, + 1016.0, + 352.0, + 355.0, + 352.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 368.0, + 1408.0, + 368.0, + 1408.0, + 408.0, + 293.0, + 408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 398.0, + 1406.0, + 398.0, + 1406.0, + 440.0, + 351.0, + 440.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 429.0, + 640.0, + 429.0, + 640.0, + 463.0, + 353.0, + 463.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 482.0, + 1404.0, + 482.0, + 1404.0, + 522.0, + 295.0, + 522.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 511.0, + 1406.0, + 511.0, + 1406.0, + 555.0, + 351.0, + 555.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 538.0, + 568.0, + 538.0, + 568.0, + 583.0, + 351.0, + 583.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 599.0, + 1404.0, + 599.0, + 1404.0, + 633.0, + 297.0, + 633.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 622.0, + 1411.0, + 622.0, + 1411.0, + 671.0, + 349.0, + 671.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 656.0, + 1306.0, + 656.0, + 1306.0, + 696.0, + 353.0, + 696.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 706.0, + 1411.0, + 706.0, + 1411.0, + 750.0, + 290.0, + 750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 735.0, + 576.0, + 735.0, + 576.0, + 780.0, + 351.0, + 780.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 792.0, + 1408.0, + 792.0, + 1408.0, + 830.0, + 293.0, + 830.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 822.0, + 1406.0, + 822.0, + 1406.0, + 862.0, + 353.0, + 862.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 853.0, + 430.0, + 853.0, + 430.0, + 891.0, + 355.0, + 891.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 906.0, + 1344.0, + 906.0, + 1344.0, + 946.0, + 295.0, + 946.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 962.0, + 1404.0, + 962.0, + 1404.0, + 996.0, + 295.0, + 996.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 992.0, + 949.0, + 992.0, + 949.0, + 1025.0, + 353.0, + 1025.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1042.0, + 1408.0, + 1042.0, + 1408.0, + 1082.0, + 293.0, + 1082.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1074.0, + 1406.0, + 1074.0, + 1406.0, + 1114.0, + 353.0, + 1114.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1099.0, + 568.0, + 1099.0, + 568.0, + 1143.0, + 351.0, + 1143.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1156.0, + 1408.0, + 1156.0, + 1408.0, + 1195.0, + 295.0, + 1195.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1187.0, + 1406.0, + 1187.0, + 1406.0, + 1227.0, + 353.0, + 1227.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1219.0, + 1207.0, + 1219.0, + 1207.0, + 1258.0, + 355.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1273.0, + 1404.0, + 1273.0, + 1404.0, + 1307.0, + 295.0, + 1307.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1300.0, + 1368.0, + 1300.0, + 1368.0, + 1340.0, + 353.0, + 1340.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 297.0, + 1355.0, + 1385.0, + 1355.0, + 1385.0, + 1389.0, + 297.0, + 1389.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 288.0, + 1401.0, + 1409.0, + 1401.0, + 1409.0, + 1452.0, + 288.0, + 1452.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1439.0, + 1308.0, + 1439.0, + 1308.0, + 1475.0, + 351.0, + 1475.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1489.0, + 1411.0, + 1489.0, + 1411.0, + 1529.0, + 293.0, + 1529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1519.0, + 432.0, + 1519.0, + 432.0, + 1561.0, + 351.0, + 1561.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1573.0, + 1408.0, + 1573.0, + 1408.0, + 1611.0, + 290.0, + 1611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1605.0, + 1404.0, + 1605.0, + 1404.0, + 1638.0, + 355.0, + 1638.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1632.0, + 1408.0, + 1632.0, + 1408.0, + 1672.0, + 353.0, + 1672.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1661.0, + 429.0, + 1661.0, + 429.0, + 1704.0, + 353.0, + 1704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1716.0, + 1406.0, + 1716.0, + 1406.0, + 1756.0, + 293.0, + 1756.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1747.0, + 1404.0, + 1747.0, + 1404.0, + 1787.0, + 353.0, + 1787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1777.0, + 835.0, + 1777.0, + 835.0, + 1815.0, + 351.0, + 1815.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 295.0, + 1834.0, + 1406.0, + 1834.0, + 1406.0, + 1867.0, + 295.0, + 1867.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1861.0, + 1411.0, + 1861.0, + 1411.0, + 1901.0, + 353.0, + 1901.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 293.0, + 1913.0, + 1404.0, + 1913.0, + 1404.0, + 1953.0, + 293.0, + 1953.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1941.0, + 1409.0, + 1941.0, + 1409.0, + 1989.0, + 349.0, + 1989.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1974.0, + 765.0, + 1974.0, + 765.0, + 2014.0, + 353.0, + 2014.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 10, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 2, + "poly": [ + 836, + 2061, + 866, + 2061, + 866, + 2087, + 836, + 2087 + ], + "score": 0.826 + }, + { + "category_id": 1, + "poly": [ + 277, + 158, + 1410, + 158, + 1410, + 2013, + 277, + 2013 + ], + "score": 0.739 + }, + { + "category_id": 2, + "poly": [ + 239, + 166, + 284, + 166, + 284, + 2035, + 239, + 2035 + ], + "score": 0.12 + }, + { + "category_id": 2, + "poly": [ + 238, + 203, + 285, + 203, + 285, + 2019, + 238, + 2019 + ], + "score": 0.12 + }, + { + "category_id": 13, + "poly": [ + 911, + 582, + 940, + 582, + 940, + 613, + 911, + 613 + ], + "score": 0.52, + "latex": "\\textsuperscript { \\textregistered }" + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2057.0, + 869.0, + 2057.0, + 869.0, + 2097.0, + 832.0, + 2097.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 206.0, + 282.0, + 206.0, + 282.0, + 239.0, + 237.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 235.0, + 284.0, + 235.0, + 284.0, + 274.0, + 234.0, + 274.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 268.0, + 280.0, + 268.0, + 280.0, + 301.0, + 237.0, + 301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 321.0, + 280.0, + 321.0, + 280.0, + 356.0, + 237.0, + 356.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 354.0, + 277.0, + 354.0, + 277.0, + 387.0, + 237.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 381.0, + 284.0, + 381.0, + 284.0, + 420.0, + 234.0, + 420.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 411.0, + 282.0, + 411.0, + 282.0, + 454.0, + 234.0, + 454.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 469.0, + 280.0, + 469.0, + 280.0, + 504.0, + 237.0, + 504.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 497.0, + 282.0, + 497.0, + 282.0, + 539.0, + 234.0, + 539.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 553.0, + 282.0, + 553.0, + 282.0, + 588.0, + 239.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 584.0, + 280.0, + 584.0, + 280.0, + 619.0, + 237.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 614.0, + 282.0, + 614.0, + 282.0, + 660.0, + 234.0, + 660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 670.0, + 282.0, + 670.0, + 282.0, + 703.0, + 237.0, + 703.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 698.0, + 282.0, + 698.0, + 282.0, + 740.0, + 234.0, + 740.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 756.0, + 280.0, + 756.0, + 280.0, + 789.0, + 237.0, + 789.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 783.0, + 282.0, + 783.0, + 282.0, + 824.0, + 234.0, + 824.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 815.0, + 282.0, + 815.0, + 282.0, + 865.0, + 237.0, + 865.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 871.0, + 280.0, + 871.0, + 280.0, + 906.0, + 237.0, + 906.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 902.0, + 282.0, + 902.0, + 282.0, + 947.0, + 234.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 955.0, + 284.0, + 955.0, + 284.0, + 994.0, + 234.0, + 994.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 988.0, + 280.0, + 988.0, + 280.0, + 1021.0, + 237.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1016.0, + 282.0, + 1016.0, + 282.0, + 1062.0, + 237.0, + 1062.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1070.0, + 282.0, + 1070.0, + 282.0, + 1111.0, + 234.0, + 1111.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1103.0, + 280.0, + 1103.0, + 280.0, + 1137.0, + 237.0, + 1137.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1133.0, + 280.0, + 1133.0, + 280.0, + 1176.0, + 234.0, + 1176.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1189.0, + 282.0, + 1189.0, + 282.0, + 1222.0, + 237.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1219.0, + 282.0, + 1219.0, + 282.0, + 1258.0, + 239.0, + 1258.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1273.0, + 280.0, + 1273.0, + 280.0, + 1308.0, + 237.0, + 1308.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1304.0, + 282.0, + 1304.0, + 282.0, + 1349.0, + 234.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1357.0, + 284.0, + 1357.0, + 284.0, + 1396.0, + 234.0, + 1396.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1390.0, + 280.0, + 1390.0, + 280.0, + 1423.0, + 237.0, + 1423.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1418.0, + 282.0, + 1418.0, + 282.0, + 1464.0, + 237.0, + 1464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1474.0, + 280.0, + 1474.0, + 280.0, + 1509.0, + 237.0, + 1509.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1505.0, + 282.0, + 1505.0, + 282.0, + 1554.0, + 237.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1560.0, + 277.0, + 1560.0, + 277.0, + 1595.0, + 237.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1593.0, + 280.0, + 1593.0, + 280.0, + 1626.0, + 237.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1619.0, + 282.0, + 1619.0, + 282.0, + 1660.0, + 234.0, + 1660.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1677.0, + 282.0, + 1677.0, + 282.0, + 1710.0, + 237.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1706.0, + 282.0, + 1706.0, + 282.0, + 1751.0, + 234.0, + 1751.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1763.0, + 280.0, + 1763.0, + 280.0, + 1798.0, + 237.0, + 1798.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1789.0, + 283.0, + 1789.0, + 283.0, + 1836.0, + 236.0, + 1836.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1845.0, + 284.0, + 1845.0, + 284.0, + 1884.0, + 234.0, + 1884.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1878.0, + 282.0, + 1878.0, + 282.0, + 1921.0, + 234.0, + 1921.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1933.0, + 280.0, + 1933.0, + 280.0, + 1968.0, + 237.0, + 1968.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 237.0, + 1964.0, + 280.0, + 1964.0, + 280.0, + 2003.0, + 237.0, + 2003.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 207.0, + 280.0, + 207.0, + 280.0, + 239.0, + 236.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 235.0, + 284.0, + 235.0, + 284.0, + 273.0, + 234.0, + 273.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 269.0, + 280.0, + 269.0, + 280.0, + 301.0, + 236.0, + 301.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 321.0, + 284.0, + 321.0, + 284.0, + 359.0, + 234.0, + 359.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 353.0, + 278.0, + 353.0, + 278.0, + 387.0, + 236.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 385.0, + 280.0, + 385.0, + 280.0, + 416.0, + 236.0, + 416.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 412.0, + 282.0, + 412.0, + 282.0, + 458.0, + 234.0, + 458.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 468.0, + 280.0, + 468.0, + 280.0, + 502.0, + 236.0, + 502.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 498.0, + 282.0, + 498.0, + 282.0, + 540.0, + 236.0, + 540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 554.0, + 280.0, + 554.0, + 280.0, + 588.0, + 236.0, + 588.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 584.0, + 280.0, + 584.0, + 280.0, + 618.0, + 236.0, + 618.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 614.0, + 282.0, + 614.0, + 282.0, + 658.0, + 236.0, + 658.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 670.0, + 280.0, + 670.0, + 280.0, + 704.0, + 236.0, + 704.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 700.0, + 282.0, + 700.0, + 282.0, + 746.0, + 234.0, + 746.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 756.0, + 278.0, + 756.0, + 278.0, + 790.0, + 236.0, + 790.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 786.0, + 280.0, + 786.0, + 280.0, + 820.0, + 236.0, + 820.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 816.0, + 282.0, + 816.0, + 282.0, + 856.0, + 234.0, + 856.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 870.0, + 284.0, + 870.0, + 284.0, + 907.0, + 234.0, + 907.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 901.0, + 282.0, + 901.0, + 282.0, + 945.0, + 234.0, + 945.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 953.0, + 282.0, + 953.0, + 282.0, + 993.0, + 234.0, + 993.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 987.0, + 280.0, + 987.0, + 280.0, + 1021.0, + 236.0, + 1021.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1017.0, + 282.0, + 1017.0, + 282.0, + 1063.0, + 234.0, + 1063.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1073.0, + 280.0, + 1073.0, + 280.0, + 1107.0, + 236.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1103.0, + 280.0, + 1103.0, + 280.0, + 1135.0, + 236.0, + 1135.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 1135.0, + 278.0, + 1135.0, + 278.0, + 1173.0, + 239.0, + 1173.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1185.0, + 282.0, + 1185.0, + 282.0, + 1225.0, + 234.0, + 1225.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1217.0, + 282.0, + 1217.0, + 282.0, + 1261.0, + 234.0, + 1261.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1273.0, + 284.0, + 1273.0, + 284.0, + 1311.0, + 234.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 232.0, + 1303.0, + 284.0, + 1303.0, + 284.0, + 1354.0, + 232.0, + 1354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1356.0, + 284.0, + 1356.0, + 284.0, + 1394.0, + 234.0, + 1394.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1390.0, + 280.0, + 1390.0, + 280.0, + 1422.0, + 236.0, + 1422.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1420.0, + 282.0, + 1420.0, + 282.0, + 1464.0, + 234.0, + 1464.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1474.0, + 280.0, + 1474.0, + 280.0, + 1508.0, + 236.0, + 1508.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 232.0, + 1502.0, + 284.0, + 1502.0, + 284.0, + 1554.0, + 232.0, + 1554.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1558.0, + 282.0, + 1558.0, + 282.0, + 1598.0, + 234.0, + 1598.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1592.0, + 280.0, + 1592.0, + 280.0, + 1626.0, + 236.0, + 1626.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1620.0, + 282.0, + 1620.0, + 282.0, + 1664.0, + 234.0, + 1664.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1678.0, + 280.0, + 1678.0, + 280.0, + 1710.0, + 236.0, + 1710.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1708.0, + 282.0, + 1708.0, + 282.0, + 1750.0, + 236.0, + 1750.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1762.0, + 280.0, + 1762.0, + 280.0, + 1796.0, + 236.0, + 1796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1792.0, + 282.0, + 1792.0, + 282.0, + 1839.0, + 236.0, + 1839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1849.0, + 280.0, + 1849.0, + 280.0, + 1881.0, + 236.0, + 1881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 234.0, + 1879.0, + 282.0, + 1879.0, + 282.0, + 1919.0, + 234.0, + 1919.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1933.0, + 280.0, + 1933.0, + 280.0, + 1967.0, + 236.0, + 1967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 236.0, + 1965.0, + 278.0, + 1965.0, + 278.0, + 1999.0, + 236.0, + 1999.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 204.0, + 1405.0, + 204.0, + 1405.0, + 236.0, + 298.0, + 236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 232.0, + 1405.0, + 232.0, + 1405.0, + 271.0, + 353.0, + 271.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 265.0, + 685.0, + 265.0, + 685.0, + 297.0, + 355.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 296.0, + 316.0, + 1407.0, + 316.0, + 1407.0, + 354.0, + 296.0, + 354.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 348.0, + 1403.0, + 348.0, + 1403.0, + 387.0, + 353.0, + 387.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 381.0, + 1403.0, + 381.0, + 1403.0, + 413.0, + 357.0, + 413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 405.0, + 548.0, + 405.0, + 548.0, + 448.0, + 353.0, + 448.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 458.0, + 1411.0, + 458.0, + 1411.0, + 505.0, + 292.0, + 505.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 493.0, + 852.0, + 493.0, + 852.0, + 532.0, + 353.0, + 532.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 548.0, + 1405.0, + 548.0, + 1405.0, + 587.0, + 294.0, + 587.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 580.0, + 910.0, + 580.0, + 910.0, + 619.0, + 353.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 941.0, + 580.0, + 1409.0, + 580.0, + 1409.0, + 619.0, + 941.0, + 619.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 611.0, + 560.0, + 611.0, + 560.0, + 644.0, + 357.0, + 644.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 660.0, + 1407.0, + 660.0, + 1407.0, + 707.0, + 290.0, + 707.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 697.0, + 562.0, + 697.0, + 562.0, + 729.0, + 355.0, + 729.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 745.0, + 1411.0, + 745.0, + 1411.0, + 792.0, + 290.0, + 792.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 778.0, + 1405.0, + 778.0, + 1405.0, + 821.0, + 353.0, + 821.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 811.0, + 1385.0, + 811.0, + 1385.0, + 849.0, + 353.0, + 849.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 866.0, + 1405.0, + 866.0, + 1405.0, + 904.0, + 294.0, + 904.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 896.0, + 1125.0, + 896.0, + 1125.0, + 935.0, + 353.0, + 935.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 951.0, + 1407.0, + 951.0, + 1407.0, + 988.0, + 292.0, + 988.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 982.0, + 1405.0, + 982.0, + 1405.0, + 1020.0, + 353.0, + 1020.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1012.0, + 1024.0, + 1012.0, + 1024.0, + 1049.0, + 353.0, + 1049.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1065.0, + 1409.0, + 1065.0, + 1409.0, + 1110.0, + 294.0, + 1110.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1094.0, + 1409.0, + 1094.0, + 1409.0, + 1141.0, + 349.0, + 1141.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1128.0, + 1304.0, + 1128.0, + 1304.0, + 1167.0, + 353.0, + 1167.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1183.0, + 1407.0, + 1183.0, + 1407.0, + 1222.0, + 294.0, + 1222.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1214.0, + 1391.0, + 1214.0, + 1391.0, + 1252.0, + 353.0, + 1252.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 298.0, + 1271.0, + 1403.0, + 1271.0, + 1403.0, + 1303.0, + 298.0, + 1303.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 351.0, + 1301.0, + 1174.0, + 1301.0, + 1174.0, + 1338.0, + 351.0, + 1338.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1354.0, + 1405.0, + 1354.0, + 1405.0, + 1393.0, + 294.0, + 1393.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1385.0, + 1409.0, + 1385.0, + 1409.0, + 1424.0, + 353.0, + 1424.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 352.0, + 1416.0, + 638.0, + 1416.0, + 638.0, + 1450.0, + 352.0, + 1450.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 292.0, + 1464.0, + 1407.0, + 1464.0, + 1407.0, + 1513.0, + 292.0, + 1513.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1501.0, + 1140.0, + 1501.0, + 1140.0, + 1540.0, + 353.0, + 1540.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1554.0, + 1405.0, + 1554.0, + 1405.0, + 1595.0, + 294.0, + 1595.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1582.0, + 1407.0, + 1582.0, + 1407.0, + 1629.0, + 349.0, + 1629.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1617.0, + 1290.0, + 1617.0, + 1290.0, + 1656.0, + 353.0, + 1656.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 290.0, + 1668.0, + 1407.0, + 1668.0, + 1407.0, + 1715.0, + 290.0, + 1715.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1703.0, + 809.0, + 1703.0, + 809.0, + 1741.0, + 353.0, + 1741.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1758.0, + 1405.0, + 1758.0, + 1405.0, + 1796.0, + 294.0, + 1796.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 355.0, + 1788.0, + 1140.0, + 1788.0, + 1140.0, + 1827.0, + 355.0, + 1827.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1843.0, + 1405.0, + 1843.0, + 1405.0, + 1882.0, + 294.0, + 1882.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 349.0, + 1872.0, + 1381.0, + 1872.0, + 1381.0, + 1914.0, + 349.0, + 1914.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 294.0, + 1929.0, + 1407.0, + 1929.0, + 1407.0, + 1967.0, + 294.0, + 1967.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 353.0, + 1959.0, + 1255.0, + 1959.0, + 1255.0, + 1998.0, + 353.0, + 1998.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 11, + "width": 1700, + "height": 2200 + } + }, + { + "layout_dets": [ + { + "category_id": 1, + "poly": [ + 407, + 1313, + 1403, + 1313, + 1403, + 1699, + 407, + 1699 + ], + "score": 0.972 + }, + { + "category_id": 1, + "poly": [ + 406, + 300, + 1404, + 300, + 1404, + 651, + 406, + 651 + ], + "score": 0.97 + }, + { + "category_id": 1, + "poly": [ + 407, + 880, + 1404, + 880, + 1404, + 1262, + 407, + 1262 + ], + "score": 0.969 + }, + { + "category_id": 1, + "poly": [ + 407, + 1752, + 1404, + 1752, + 1404, + 1949, + 407, + 1949 + ], + "score": 0.962 + }, + { + "category_id": 1, + "poly": [ + 407, + 701, + 1404, + 701, + 1404, + 830, + 407, + 830 + ], + "score": 0.951 + }, + { + "category_id": 1, + "poly": [ + 366, + 660, + 840, + 660, + 840, + 693, + 366, + 693 + ], + "score": 0.922 + }, + { + "category_id": 1, + "poly": [ + 364, + 840, + 672, + 840, + 672, + 872, + 364, + 872 + ], + "score": 0.91 + }, + { + "category_id": 1, + "poly": [ + 365, + 258, + 580, + 258, + 580, + 288, + 365, + 288 + ], + "score": 0.897 + }, + { + "category_id": 1, + "poly": [ + 366, + 1272, + 1401, + 1272, + 1401, + 1305, + 366, + 1305 + ], + "score": 0.842 + }, + { + "category_id": 2, + "poly": [ + 836, + 2061, + 865, + 2061, + 865, + 2085, + 836, + 2085 + ], + "score": 0.841 + }, + { + "category_id": 1, + "poly": [ + 366, + 1710, + 1200, + 1710, + 1200, + 1743, + 366, + 1743 + ], + "score": 0.838 + }, + { + "category_id": 0, + "poly": [ + 242, + 199, + 435, + 199, + 435, + 235, + 242, + 235 + ], + "score": 0.591 + }, + { + "category_id": 2, + "poly": [ + 242, + 199, + 435, + 199, + 435, + 235, + 242, + 235 + ], + "score": 0.316 + }, + { + "category_id": 2, + "poly": [ + 236, + 190, + 286, + 190, + 286, + 1976, + 236, + 1976 + ], + "score": 0.308 + }, + { + "category_id": 15, + "poly": [ + 832.0, + 2057.0, + 870.0, + 2057.0, + 870.0, + 2096.0, + 832.0, + 2096.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 206.0, + 284.0, + 206.0, + 284.0, + 236.0, + 239.0, + 236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 196.0, + 440.0, + 196.0, + 440.0, + 239.0, + 289.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 239.0, + 206.0, + 284.0, + 206.0, + 284.0, + 236.0, + 239.0, + 236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 289.0, + 196.0, + 440.0, + 196.0, + 440.0, + 239.0, + 289.0, + 239.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 209.0, + 282.0, + 209.0, + 282.0, + 240.0, + 238.0, + 240.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 260.0, + 284.0, + 260.0, + 284.0, + 297.0, + 233.0, + 297.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 305.0, + 280.0, + 305.0, + 280.0, + 336.0, + 238.0, + 336.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 334.0, + 282.0, + 334.0, + 282.0, + 368.0, + 238.0, + 368.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 370.0, + 280.0, + 370.0, + 280.0, + 401.0, + 238.0, + 401.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 399.0, + 280.0, + 399.0, + 280.0, + 431.0, + 238.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 431.0, + 280.0, + 431.0, + 280.0, + 462.0, + 238.0, + 462.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 462.0, + 280.0, + 462.0, + 280.0, + 494.0, + 235.0, + 494.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 496.0, + 280.0, + 496.0, + 280.0, + 529.0, + 235.0, + 529.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 527.0, + 280.0, + 527.0, + 280.0, + 558.0, + 235.0, + 558.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 556.0, + 280.0, + 556.0, + 280.0, + 590.0, + 235.0, + 590.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 590.0, + 284.0, + 590.0, + 284.0, + 627.0, + 233.0, + 627.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 621.0, + 284.0, + 621.0, + 284.0, + 661.0, + 233.0, + 661.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 665.0, + 282.0, + 665.0, + 282.0, + 702.0, + 238.0, + 702.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 706.0, + 280.0, + 706.0, + 280.0, + 739.0, + 235.0, + 739.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 737.0, + 280.0, + 737.0, + 280.0, + 769.0, + 235.0, + 769.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 773.0, + 282.0, + 773.0, + 282.0, + 804.0, + 235.0, + 804.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 802.0, + 282.0, + 802.0, + 282.0, + 839.0, + 238.0, + 839.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 843.0, + 284.0, + 843.0, + 284.0, + 881.0, + 233.0, + 881.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 885.0, + 280.0, + 885.0, + 280.0, + 916.0, + 235.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 916.0, + 282.0, + 916.0, + 282.0, + 947.0, + 235.0, + 947.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 947.0, + 282.0, + 947.0, + 282.0, + 979.0, + 235.0, + 979.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 983.0, + 282.0, + 983.0, + 282.0, + 1012.0, + 235.0, + 1012.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1012.0, + 282.0, + 1012.0, + 282.0, + 1044.0, + 235.0, + 1044.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1044.0, + 282.0, + 1044.0, + 282.0, + 1075.0, + 235.0, + 1075.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1079.0, + 280.0, + 1079.0, + 280.0, + 1111.0, + 235.0, + 1111.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1109.0, + 282.0, + 1109.0, + 282.0, + 1138.0, + 235.0, + 1138.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1140.0, + 280.0, + 1140.0, + 280.0, + 1169.0, + 235.0, + 1169.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1173.0, + 280.0, + 1173.0, + 280.0, + 1205.0, + 235.0, + 1205.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1205.0, + 280.0, + 1205.0, + 280.0, + 1236.0, + 235.0, + 1236.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1236.0, + 280.0, + 1236.0, + 280.0, + 1268.0, + 235.0, + 1268.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 1276.0, + 284.0, + 1276.0, + 284.0, + 1313.0, + 233.0, + 1313.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1319.0, + 282.0, + 1319.0, + 282.0, + 1350.0, + 235.0, + 1350.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1348.0, + 282.0, + 1348.0, + 282.0, + 1380.0, + 235.0, + 1380.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 1376.0, + 284.0, + 1376.0, + 284.0, + 1413.0, + 233.0, + 1413.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 238.0, + 1413.0, + 280.0, + 1413.0, + 280.0, + 1446.0, + 238.0, + 1446.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1444.0, + 280.0, + 1444.0, + 280.0, + 1478.0, + 235.0, + 1478.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 1478.0, + 284.0, + 1478.0, + 284.0, + 1515.0, + 233.0, + 1515.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1511.0, + 280.0, + 1511.0, + 280.0, + 1545.0, + 235.0, + 1545.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1547.0, + 280.0, + 1547.0, + 280.0, + 1578.0, + 235.0, + 1578.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1578.0, + 280.0, + 1578.0, + 280.0, + 1611.0, + 235.0, + 1611.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1613.0, + 282.0, + 1613.0, + 282.0, + 1645.0, + 235.0, + 1645.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1643.0, + 280.0, + 1643.0, + 280.0, + 1674.0, + 235.0, + 1674.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1674.0, + 280.0, + 1674.0, + 280.0, + 1706.0, + 235.0, + 1706.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 1716.0, + 284.0, + 1716.0, + 284.0, + 1753.0, + 233.0, + 1753.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1757.0, + 280.0, + 1757.0, + 280.0, + 1788.0, + 235.0, + 1788.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 233.0, + 1784.0, + 284.0, + 1784.0, + 284.0, + 1822.0, + 233.0, + 1822.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1822.0, + 280.0, + 1822.0, + 280.0, + 1855.0, + 235.0, + 1855.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1853.0, + 280.0, + 1853.0, + 280.0, + 1885.0, + 235.0, + 1885.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1888.0, + 280.0, + 1888.0, + 280.0, + 1920.0, + 235.0, + 1920.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 235.0, + 1920.0, + 282.0, + 1920.0, + 282.0, + 1951.0, + 235.0, + 1951.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1313.0, + 1405.0, + 1313.0, + 1405.0, + 1349.0, + 406.0, + 1349.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1342.0, + 1404.0, + 1342.0, + 1404.0, + 1378.0, + 450.0, + 1378.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1373.0, + 792.0, + 1373.0, + 792.0, + 1408.0, + 450.0, + 1408.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1408.0, + 1405.0, + 1408.0, + 1405.0, + 1444.0, + 406.0, + 1444.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 453.0, + 1445.0, + 599.0, + 1445.0, + 599.0, + 1476.0, + 453.0, + 1476.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 1474.0, + 1404.0, + 1474.0, + 1404.0, + 1511.0, + 407.0, + 1511.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1506.0, + 1117.0, + 1506.0, + 1117.0, + 1541.0, + 450.0, + 1541.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 1541.0, + 1403.0, + 1541.0, + 1403.0, + 1577.0, + 406.0, + 1577.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 452.0, + 1574.0, + 705.0, + 1574.0, + 705.0, + 1608.0, + 452.0, + 1608.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 1606.0, + 1406.0, + 1606.0, + 1406.0, + 1646.0, + 407.0, + 1646.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 1636.0, + 1404.0, + 1636.0, + 1404.0, + 1675.0, + 448.0, + 1675.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1667.0, + 954.0, + 1667.0, + 954.0, + 1701.0, + 450.0, + 1701.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 298.0, + 1404.0, + 298.0, + 1404.0, + 335.0, + 405.0, + 335.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 329.0, + 802.0, + 329.0, + 802.0, + 365.0, + 449.0, + 365.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 366.0, + 1406.0, + 366.0, + 1406.0, + 399.0, + 406.0, + 399.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 394.0, + 1406.0, + 394.0, + 1406.0, + 431.0, + 450.0, + 431.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 447.0, + 425.0, + 1406.0, + 425.0, + 1406.0, + 461.0, + 447.0, + 461.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 455.0, + 786.0, + 455.0, + 786.0, + 492.0, + 449.0, + 492.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 409.0, + 491.0, + 1406.0, + 491.0, + 1406.0, + 527.0, + 409.0, + 527.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 524.0, + 1406.0, + 524.0, + 1406.0, + 557.0, + 451.0, + 557.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 553.0, + 926.0, + 553.0, + 926.0, + 586.0, + 450.0, + 586.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 588.0, + 1405.0, + 588.0, + 1405.0, + 625.0, + 406.0, + 625.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 614.0, + 595.0, + 614.0, + 595.0, + 653.0, + 449.0, + 653.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 880.0, + 1407.0, + 880.0, + 1407.0, + 916.0, + 406.0, + 916.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 911.0, + 1404.0, + 911.0, + 1404.0, + 944.0, + 451.0, + 944.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 942.0, + 855.0, + 942.0, + 855.0, + 975.0, + 450.0, + 975.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 406.0, + 976.0, + 1406.0, + 976.0, + 1406.0, + 1014.0, + 406.0, + 1014.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 448.0, + 1007.0, + 1406.0, + 1007.0, + 1406.0, + 1043.0, + 448.0, + 1043.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 1037.0, + 571.0, + 1037.0, + 571.0, + 1070.0, + 451.0, + 1070.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 408.0, + 1074.0, + 1407.0, + 1074.0, + 1407.0, + 1107.0, + 408.0, + 1107.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1103.0, + 1404.0, + 1103.0, + 1404.0, + 1136.0, + 450.0, + 1136.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 1136.0, + 788.0, + 1136.0, + 788.0, + 1168.0, + 451.0, + 1168.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 407.0, + 1167.0, + 1404.0, + 1167.0, + 1404.0, + 1207.0, + 407.0, + 1207.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 1202.0, + 1403.0, + 1202.0, + 1403.0, + 1234.0, + 451.0, + 1234.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1231.0, + 952.0, + 1231.0, + 952.0, + 1263.0, + 450.0, + 1263.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 404.0, + 1748.0, + 1408.0, + 1748.0, + 1408.0, + 1787.0, + 404.0, + 1787.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1781.0, + 661.0, + 1781.0, + 661.0, + 1816.0, + 449.0, + 1816.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 405.0, + 1817.0, + 1404.0, + 1817.0, + 1404.0, + 1853.0, + 405.0, + 1853.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 450.0, + 1847.0, + 951.0, + 1847.0, + 951.0, + 1883.0, + 450.0, + 1883.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 410.0, + 1882.0, + 1404.0, + 1882.0, + 1404.0, + 1918.0, + 410.0, + 1918.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 1917.0, + 926.0, + 1917.0, + 926.0, + 1949.0, + 449.0, + 1949.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 403.0, + 698.0, + 1406.0, + 698.0, + 1406.0, + 737.0, + 403.0, + 737.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 451.0, + 732.0, + 803.0, + 732.0, + 803.0, + 762.0, + 451.0, + 762.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 402.0, + 764.0, + 1406.0, + 764.0, + 1406.0, + 805.0, + 402.0, + 805.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 449.0, + 794.0, + 609.0, + 794.0, + 609.0, + 835.0, + 449.0, + 835.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 360.0, + 658.0, + 846.0, + 658.0, + 846.0, + 697.0, + 360.0, + 697.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 834.0, + 675.0, + 834.0, + 675.0, + 879.0, + 357.0, + 879.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 361.0, + 253.0, + 586.0, + 253.0, + 586.0, + 292.0, + 361.0, + 292.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 356.0, + 1267.0, + 1408.0, + 1267.0, + 1408.0, + 1311.0, + 356.0, + 1311.0 + ], + "score": 1.0, + "text": "" + }, + { + "category_id": 15, + "poly": [ + 357.0, + 1706.0, + 1205.0, + 1706.0, + 1205.0, + 1749.0, + 357.0, + 1749.0 + ], + "score": 1.0, + "text": "" + } + ], + "page_info": { + "page_no": 12, + "width": 1700, + "height": 2200 + } + } +] \ No newline at end of file diff --git a/parse/train/lQdXeXDoWtI/images/0120a81fef26784e00e1df4268bd3bbdc5eddad2b4cb84b865458f75af293dde.jpg b/parse/train/lQdXeXDoWtI/images/0120a81fef26784e00e1df4268bd3bbdc5eddad2b4cb84b865458f75af293dde.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ccf35101aa034a2716cbb0c96de3140c68da83d2 --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/0120a81fef26784e00e1df4268bd3bbdc5eddad2b4cb84b865458f75af293dde.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1c0ffa1bdfaa2146252a855091bff1572a495ee2f30d458327826b196f0cbc65 +size 125619 diff --git a/parse/train/lQdXeXDoWtI/images/0b3f38ed22131c2c84f068f99124fb6b96ed3a2deb910e8285ba4fb5105924d1.jpg b/parse/train/lQdXeXDoWtI/images/0b3f38ed22131c2c84f068f99124fb6b96ed3a2deb910e8285ba4fb5105924d1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e929fc0fcc4b79f3dfb3049098da5f6f8df2d9ec --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/0b3f38ed22131c2c84f068f99124fb6b96ed3a2deb910e8285ba4fb5105924d1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fd0e48479a2a5a4665b585c69f1c9fb40074d129b9e99ad56eed39429621d1fd +size 97560 diff --git a/parse/train/lQdXeXDoWtI/images/0f8472c0204af51b88c38b0dbf2cb08deeb15fdd93def2900905e3ae51335355.jpg b/parse/train/lQdXeXDoWtI/images/0f8472c0204af51b88c38b0dbf2cb08deeb15fdd93def2900905e3ae51335355.jpg new file mode 100644 index 0000000000000000000000000000000000000000..837d8f1bb0b22744f0d6cf675ac7c0b69e164732 --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/0f8472c0204af51b88c38b0dbf2cb08deeb15fdd93def2900905e3ae51335355.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5018bceaf4a76261857b47dfc442973269cf2c314bc1b40bee580a7eb53d17ec +size 117124 diff --git a/parse/train/lQdXeXDoWtI/images/1bb53236240e1cc5485a5dd1191f549477c33bac0872e8141388c636c162cbfc.jpg b/parse/train/lQdXeXDoWtI/images/1bb53236240e1cc5485a5dd1191f549477c33bac0872e8141388c636c162cbfc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a22cfd688366ecc92ae3e0b273d1742f330ed05b --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/1bb53236240e1cc5485a5dd1191f549477c33bac0872e8141388c636c162cbfc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1a41a390b97e5e9384da17e47c85cb0529b18d4361cf565c2cd69e870e600bdc +size 125816 diff --git a/parse/train/lQdXeXDoWtI/images/2b88a1364198777f3b8bfa6b2b07d9aac25b2e33fbf5125f2782452508a0e818.jpg b/parse/train/lQdXeXDoWtI/images/2b88a1364198777f3b8bfa6b2b07d9aac25b2e33fbf5125f2782452508a0e818.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e312b1e6ce724657d8c627a9463a1d4c3ecc4a6d --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/2b88a1364198777f3b8bfa6b2b07d9aac25b2e33fbf5125f2782452508a0e818.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e3ebe8a896fa048e234fee056301fefaac9933b7db00ce10976e76d757bf3d41 +size 92053 diff --git a/parse/train/lQdXeXDoWtI/images/2d74690da04e9674d1a499b85fcad2b40c5f6c26d28131cb40f2f73951bde193.jpg b/parse/train/lQdXeXDoWtI/images/2d74690da04e9674d1a499b85fcad2b40c5f6c26d28131cb40f2f73951bde193.jpg new file mode 100644 index 0000000000000000000000000000000000000000..99b0200c53735e38734ecc27d26bc9a10b619729 --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/2d74690da04e9674d1a499b85fcad2b40c5f6c26d28131cb40f2f73951bde193.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:51d0afcf5dc22ced75b10658f05192ed2fa2eec35b0a34370afd945d73231613 +size 38592 diff --git a/parse/train/lQdXeXDoWtI/images/36e50de8e3e8a779ea8c1b16f08a783ca37292925f539cf0cbaa0f2ed07a3c6b.jpg b/parse/train/lQdXeXDoWtI/images/36e50de8e3e8a779ea8c1b16f08a783ca37292925f539cf0cbaa0f2ed07a3c6b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..965f96f7eed6c71010cc9296c1ef810ad4964a29 --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/36e50de8e3e8a779ea8c1b16f08a783ca37292925f539cf0cbaa0f2ed07a3c6b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8e8dccbe7148c56e93ebb3dc2763634786d69ef8d9bd175fd643a839a4a8d88e +size 119032 diff --git a/parse/train/lQdXeXDoWtI/images/393f4239d51de146347ecef49abbca02af4949e307f6ed74bc1d486ff8d4e201.jpg b/parse/train/lQdXeXDoWtI/images/393f4239d51de146347ecef49abbca02af4949e307f6ed74bc1d486ff8d4e201.jpg new file mode 100644 index 0000000000000000000000000000000000000000..18c86dff098d0da6bff3776a6f23ff695e16885b --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/393f4239d51de146347ecef49abbca02af4949e307f6ed74bc1d486ff8d4e201.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7f2ee50664228fd6ad1f4c22a081b624f7a746f52538995632e24ebf17ea10ca +size 96525 diff --git a/parse/train/lQdXeXDoWtI/images/49c452879bd5790b748dbeffcc79b23ef8c1b37b58978f645686665cf347bbbe.jpg b/parse/train/lQdXeXDoWtI/images/49c452879bd5790b748dbeffcc79b23ef8c1b37b58978f645686665cf347bbbe.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2403fb3b1d083b4bd89795247971d23438a27091 --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/49c452879bd5790b748dbeffcc79b23ef8c1b37b58978f645686665cf347bbbe.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7f6c0db539060b84655a91944e2edd0b608096adcf48bcee23d2605978cdb635 +size 121154 diff --git a/parse/train/lQdXeXDoWtI/images/51f8ca840d9f95411fdcbd81687b11c282f50fec3778f05df62df655dbc8e09a.jpg b/parse/train/lQdXeXDoWtI/images/51f8ca840d9f95411fdcbd81687b11c282f50fec3778f05df62df655dbc8e09a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1916d1deafde9e1d3c27fe5978efa67baba260ff --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/51f8ca840d9f95411fdcbd81687b11c282f50fec3778f05df62df655dbc8e09a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3aa4004cc1afdce451415527921129024dc5a8478ee1fa31f5a795723f1fcc2d +size 94892 diff --git a/parse/train/lQdXeXDoWtI/images/5b165549e593254b181604eb39468c1f6a9b999469ef30c0a4133bf877440869.jpg b/parse/train/lQdXeXDoWtI/images/5b165549e593254b181604eb39468c1f6a9b999469ef30c0a4133bf877440869.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4a1c8ec8cffdf3a8f0acc3061440caccf99d0342 --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/5b165549e593254b181604eb39468c1f6a9b999469ef30c0a4133bf877440869.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9068eb206a576543b45915a4d2b8d1ba2545b572e741333b39f4d7a6c37de0d7 +size 22157 diff --git a/parse/train/lQdXeXDoWtI/images/5db0112704627a186f621562de7ef29ba142fb829968975a3f8d82c39f02b891.jpg b/parse/train/lQdXeXDoWtI/images/5db0112704627a186f621562de7ef29ba142fb829968975a3f8d82c39f02b891.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8b25cfd7a58bd7cdf7fac54813af8dc1d622c62a --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/5db0112704627a186f621562de7ef29ba142fb829968975a3f8d82c39f02b891.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e7aa2e9b577312eb9214a2f23e858ada3ee73d89496e739ad01c472c7fa31a29 +size 53124 diff --git a/parse/train/lQdXeXDoWtI/images/683e0e70fcaf68a6555a6bc63671dabe65b336f939a4076e7fe8112daecb246d.jpg b/parse/train/lQdXeXDoWtI/images/683e0e70fcaf68a6555a6bc63671dabe65b336f939a4076e7fe8112daecb246d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f861bc9ab38c4504ff921c800d31e29b737d9745 --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/683e0e70fcaf68a6555a6bc63671dabe65b336f939a4076e7fe8112daecb246d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3baf4701fb92a76e09d28b4ce88d3b625f1af87aca4e7fcf5dd283d3f3b46103 +size 81304 diff --git a/parse/train/lQdXeXDoWtI/images/773d0eb4db746b90448f3503d7016ea8afef74eeed5e561b387f9b866b42a07b.jpg b/parse/train/lQdXeXDoWtI/images/773d0eb4db746b90448f3503d7016ea8afef74eeed5e561b387f9b866b42a07b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0b90d5b31672297e26e00dac528afca653a9b6e3 --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/773d0eb4db746b90448f3503d7016ea8afef74eeed5e561b387f9b866b42a07b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1a57eeff3703490cb079bc06afe640bbfd74e7b2fd49d3d82b32378fa795a566 +size 212094 diff --git a/parse/train/lQdXeXDoWtI/images/77eaeaddb0bb3b9496d55dd75803c2f8b65be4eadb418ecbd5f3b48aedcb8dbe.jpg b/parse/train/lQdXeXDoWtI/images/77eaeaddb0bb3b9496d55dd75803c2f8b65be4eadb418ecbd5f3b48aedcb8dbe.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f9f155e1efad5ee6b2bdaa769713f43d849585e3 --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/77eaeaddb0bb3b9496d55dd75803c2f8b65be4eadb418ecbd5f3b48aedcb8dbe.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:81687a6a0ba8c10815789fba73ed3d77760dee976f47f9fa03f9fcb3e9167306 +size 104128 diff --git a/parse/train/lQdXeXDoWtI/images/795dcdcf8333803ae9a01cdc83a378ccb7ca36645261c108f2c49b9f119e2d78.jpg b/parse/train/lQdXeXDoWtI/images/795dcdcf8333803ae9a01cdc83a378ccb7ca36645261c108f2c49b9f119e2d78.jpg new file mode 100644 index 0000000000000000000000000000000000000000..096d5741f5ad632834d442f118e76f1adf47720b --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/795dcdcf8333803ae9a01cdc83a378ccb7ca36645261c108f2c49b9f119e2d78.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fa0ee883d52a4f239ba7dbd6a46aa2c1aa6f47a487aeea9bc8f704e261aa2226 +size 77541 diff --git a/parse/train/lQdXeXDoWtI/images/7d3457ca1e32c8f76aa3e2a1a099eda49f55304dabe1dbd5f4289c125aa81df1.jpg b/parse/train/lQdXeXDoWtI/images/7d3457ca1e32c8f76aa3e2a1a099eda49f55304dabe1dbd5f4289c125aa81df1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..83ef6e30a390b7e6f3b00d2c9f8e412053728830 --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/7d3457ca1e32c8f76aa3e2a1a099eda49f55304dabe1dbd5f4289c125aa81df1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:94a299412fa72a733274f0c7d73c5f6de3e04b3b584022e432e9877ea2009fad +size 107135 diff --git a/parse/train/lQdXeXDoWtI/images/7d8344916306ee01fa4d0c7fcfef1c03b182d8146281a51e280ff5aac7c68f9a.jpg b/parse/train/lQdXeXDoWtI/images/7d8344916306ee01fa4d0c7fcfef1c03b182d8146281a51e280ff5aac7c68f9a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..02a14ac6efca209946bfa89c8a74f67805a265ee --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/7d8344916306ee01fa4d0c7fcfef1c03b182d8146281a51e280ff5aac7c68f9a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:18fcbe37233c6da45e0831e2f9586074e86ee0416f61065bc51fece5dfe4767d +size 119534 diff --git a/parse/train/lQdXeXDoWtI/images/8aa9c3829c5936bb50b6038f2621223677e132bf28288b6b142b7fd1d59742cd.jpg b/parse/train/lQdXeXDoWtI/images/8aa9c3829c5936bb50b6038f2621223677e132bf28288b6b142b7fd1d59742cd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d631324a83ec25a9f6f300b4146cae073b66130a --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/8aa9c3829c5936bb50b6038f2621223677e132bf28288b6b142b7fd1d59742cd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ec38da5805b203aeb8c24482525897bcaea7e1545405988d9f7d2524a6303536 +size 97665 diff --git a/parse/train/lQdXeXDoWtI/images/8b932c06e524be086905dce2ce843484018ae33ac2d617171072e42976aba107.jpg b/parse/train/lQdXeXDoWtI/images/8b932c06e524be086905dce2ce843484018ae33ac2d617171072e42976aba107.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2b1f6261d93bf39e5824d2997ebfdb1fd851779a --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/8b932c06e524be086905dce2ce843484018ae33ac2d617171072e42976aba107.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ef63a08d1cc373e4daa9471145466cecdc178c4f35fe9542c82d7581539395d1 +size 121403 diff --git a/parse/train/lQdXeXDoWtI/images/8e5c9a1918061bd97c079cbdf8ac9decdce53739516f27fc383dfc7fd8266da7.jpg b/parse/train/lQdXeXDoWtI/images/8e5c9a1918061bd97c079cbdf8ac9decdce53739516f27fc383dfc7fd8266da7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2b8907a0e47800f7099ff61569d1640af534303c --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/8e5c9a1918061bd97c079cbdf8ac9decdce53739516f27fc383dfc7fd8266da7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9135a7df553f1698805b97178b18ca07c7ea38592a49a6bb649eb65f9a86d72f +size 114527 diff --git a/parse/train/lQdXeXDoWtI/images/922a0607fde24e03492675346a1e9084e75b1a41ab10fa265fad9a0cd87b4ba0.jpg b/parse/train/lQdXeXDoWtI/images/922a0607fde24e03492675346a1e9084e75b1a41ab10fa265fad9a0cd87b4ba0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..83b2ffbf4f3b1b1a463498ae17e817ef1d7b3b86 --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/922a0607fde24e03492675346a1e9084e75b1a41ab10fa265fad9a0cd87b4ba0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8b8021aebc90f611d68b1e179bdbf3727ba87c3cfcaa6d3cfc4704e070011362 +size 92988 diff --git a/parse/train/lQdXeXDoWtI/images/95c1c298ff6893ebcbf1dec48e4922dde6316823fff2f5782056a00dc5aa994b.jpg b/parse/train/lQdXeXDoWtI/images/95c1c298ff6893ebcbf1dec48e4922dde6316823fff2f5782056a00dc5aa994b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1b6bb12a6f6b90e14d6f3048da5fc845ac36bcbc --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/95c1c298ff6893ebcbf1dec48e4922dde6316823fff2f5782056a00dc5aa994b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ab328378437e4be59d9714731188dd561a2d2a6a9bf7ea8922362820f35a567f +size 43077 diff --git a/parse/train/lQdXeXDoWtI/images/9ba7746fffce965e4bc35feae1b7eb5d2689f91d8d3c777c598bfa2a16f700a6.jpg b/parse/train/lQdXeXDoWtI/images/9ba7746fffce965e4bc35feae1b7eb5d2689f91d8d3c777c598bfa2a16f700a6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..55f241df676d6e77bc3a52cc8a8cd28db4c1f6eb --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/9ba7746fffce965e4bc35feae1b7eb5d2689f91d8d3c777c598bfa2a16f700a6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7ab4a3eb920a89b8d62eb7053eb75502ce0dbf4289738cbd7081fcfef089c2fb +size 23574 diff --git a/parse/train/lQdXeXDoWtI/images/a0c4d3464d159f7ed64a2c91c3eb5058f42226bf23fbf07c2d0b6733fe21d9e1.jpg b/parse/train/lQdXeXDoWtI/images/a0c4d3464d159f7ed64a2c91c3eb5058f42226bf23fbf07c2d0b6733fe21d9e1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9294decbcd0a97acf62f79983f5dac82a2f4e603 --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/a0c4d3464d159f7ed64a2c91c3eb5058f42226bf23fbf07c2d0b6733fe21d9e1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6cc2ada472307e7e17af1f470bd1ff0949da7230dfe80f2f421906175bd7ed87 +size 93586 diff --git a/parse/train/lQdXeXDoWtI/images/a614f6a4ee55a6be51ddcbab232396166fc604c205fedd94a63a6317f616b0a8.jpg b/parse/train/lQdXeXDoWtI/images/a614f6a4ee55a6be51ddcbab232396166fc604c205fedd94a63a6317f616b0a8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d34b82b2c3b656691ea97e6644689d13dd856bd0 --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/a614f6a4ee55a6be51ddcbab232396166fc604c205fedd94a63a6317f616b0a8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:306e9ac57c361b06d673596968cf4b4bdf8d3a90cb28577c58684d14ab708d78 +size 94615 diff --git a/parse/train/lQdXeXDoWtI/images/a745ddf06308e85653ef51637a0831d42abe8385193a3a5d7b11d1c3c797b7cc.jpg b/parse/train/lQdXeXDoWtI/images/a745ddf06308e85653ef51637a0831d42abe8385193a3a5d7b11d1c3c797b7cc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f1e263c20bd5332a35b8e29ed9bb5ec109ba0b28 --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/a745ddf06308e85653ef51637a0831d42abe8385193a3a5d7b11d1c3c797b7cc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:84bff8e4fea053320077dea4726bff9b7d3c546f0883cd6dab624705311244f7 +size 93071 diff --git a/parse/train/lQdXeXDoWtI/images/a83f4c071ed20c188f609c07faefb1f31a82a78ee40a65d697b1fe36cb58e6cb.jpg b/parse/train/lQdXeXDoWtI/images/a83f4c071ed20c188f609c07faefb1f31a82a78ee40a65d697b1fe36cb58e6cb.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7d88c55fd5319b3a2d5a7752b5d3640e927f9c9a --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/a83f4c071ed20c188f609c07faefb1f31a82a78ee40a65d697b1fe36cb58e6cb.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:996cfbdc3c79ba38bd73c777edd737deab1561b0f4a402f35be51be9e67abdbc +size 326259 diff --git a/parse/train/lQdXeXDoWtI/images/ad6bde8098fd77c826962a8b466f5aa4ab7f749d4824faf3091cf001933b785c.jpg b/parse/train/lQdXeXDoWtI/images/ad6bde8098fd77c826962a8b466f5aa4ab7f749d4824faf3091cf001933b785c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..475482543b54fe0aa7af804cf78578e9d6c796a0 --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/ad6bde8098fd77c826962a8b466f5aa4ab7f749d4824faf3091cf001933b785c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:240f2ec1b23ae69d1c322c838675210cd51cee0d86339d9fbe8d5a10c0165c9b +size 75278 diff --git a/parse/train/lQdXeXDoWtI/images/b55d1fd4d81a302f4b4acb319702a34dc996eec6a09064cf816cd47bf8400ff1.jpg b/parse/train/lQdXeXDoWtI/images/b55d1fd4d81a302f4b4acb319702a34dc996eec6a09064cf816cd47bf8400ff1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..264b85b01a2ff5ab9706f865a4c1545592a6477f --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/b55d1fd4d81a302f4b4acb319702a34dc996eec6a09064cf816cd47bf8400ff1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1ba8611633360ac97343e5f4a388e2455bea55a9daa6d0074d198f56dce7d4f +size 41764 diff --git a/parse/train/lQdXeXDoWtI/images/c1e2c84cf16bd15118a7bc65114442035ee17cf70d4f0e2ecf4314aef226dec0.jpg b/parse/train/lQdXeXDoWtI/images/c1e2c84cf16bd15118a7bc65114442035ee17cf70d4f0e2ecf4314aef226dec0.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4d9e9542531a5bbfe88cc6c506e287f39fb7fb95 --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/c1e2c84cf16bd15118a7bc65114442035ee17cf70d4f0e2ecf4314aef226dec0.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4ca020786838307bce358b336829c816df01ca80790c9432e6e3144fe7781f6b +size 96434 diff --git a/parse/train/lQdXeXDoWtI/images/c54abd4d3fa730f9c90560253c19b8c0c7614b8921fd026b719b15140cdd58df.jpg b/parse/train/lQdXeXDoWtI/images/c54abd4d3fa730f9c90560253c19b8c0c7614b8921fd026b719b15140cdd58df.jpg new file mode 100644 index 0000000000000000000000000000000000000000..66fa19d9bb439fccf601cfa6409ad824b267f94a --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/c54abd4d3fa730f9c90560253c19b8c0c7614b8921fd026b719b15140cdd58df.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4fc2dc877de415d0be6ea4bc3c98c4d7e3dd8e873697d263fd5a7729524827a6 +size 102806 diff --git a/parse/train/lQdXeXDoWtI/images/cc17db6e45c55cebfe78ed7099ae50d3c1e2fc05ce5320a896417b9a16fa2716.jpg b/parse/train/lQdXeXDoWtI/images/cc17db6e45c55cebfe78ed7099ae50d3c1e2fc05ce5320a896417b9a16fa2716.jpg new file mode 100644 index 0000000000000000000000000000000000000000..26eeb9c8e84cbee7aa412f56ebc389790ac52b99 --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/cc17db6e45c55cebfe78ed7099ae50d3c1e2fc05ce5320a896417b9a16fa2716.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f726f4ecc3837439d7c63d933e65af577284d2944c6d5c887f62a164bc5acefe +size 95284 diff --git a/parse/train/lQdXeXDoWtI/images/d687ebe242f336585a5eb79fd18b836bc728b90114f888f021f92dc526c40e6e.jpg b/parse/train/lQdXeXDoWtI/images/d687ebe242f336585a5eb79fd18b836bc728b90114f888f021f92dc526c40e6e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..61615cc276aa65106d5772aaa61957a3413372fd --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/d687ebe242f336585a5eb79fd18b836bc728b90114f888f021f92dc526c40e6e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7fa6291cb7365ab48bb2a410292b918d15c00a899edcfc767d50df4257a36099 +size 95527 diff --git a/parse/train/lQdXeXDoWtI/images/e09a23fc758c95ed3e528b354b69902c988fc290fdd3421b469ffd37c3176c73.jpg b/parse/train/lQdXeXDoWtI/images/e09a23fc758c95ed3e528b354b69902c988fc290fdd3421b469ffd37c3176c73.jpg new file mode 100644 index 0000000000000000000000000000000000000000..21d58852b691dbcc96a945f17a067e1a09be2c6c --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/e09a23fc758c95ed3e528b354b69902c988fc290fdd3421b469ffd37c3176c73.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e1d29f0e2a64d9bc485c17f66eed4d19294e83cde68cdcb6bdcb3f95f9b9bb63 +size 94420 diff --git a/parse/train/lQdXeXDoWtI/images/f07a1ef753111a6720546b65ba84b34a4240db479bd6d1304bbc1df482fa438a.jpg b/parse/train/lQdXeXDoWtI/images/f07a1ef753111a6720546b65ba84b34a4240db479bd6d1304bbc1df482fa438a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9f84f33a01083976e31fb66942dc846d4b382946 --- /dev/null +++ b/parse/train/lQdXeXDoWtI/images/f07a1ef753111a6720546b65ba84b34a4240db479bd6d1304bbc1df482fa438a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:12afa71d84e882d5ec7d07c21192c395359a270de32580942f9153365fff5900 +size 103244 diff --git a/parse/train/nzpLWnVAyah/images/17e75685fd0e4826b400d7908b5e71f88d0b2ad10037586d5cf0ebde73a2ccb4.jpg b/parse/train/nzpLWnVAyah/images/17e75685fd0e4826b400d7908b5e71f88d0b2ad10037586d5cf0ebde73a2ccb4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ff0eededdaadd6b813319be2963f8d9a8945b0e5 --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/17e75685fd0e4826b400d7908b5e71f88d0b2ad10037586d5cf0ebde73a2ccb4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:48ea5e6715f7f5246acc1bacf0d95813fcd3291c9356c0e6886482ae7e842329 +size 53097 diff --git a/parse/train/nzpLWnVAyah/images/2ca1eb5ee16c11f85ba9ece76c980550e0f43eb5a07b91cc6fbb1fed6786cf92.jpg b/parse/train/nzpLWnVAyah/images/2ca1eb5ee16c11f85ba9ece76c980550e0f43eb5a07b91cc6fbb1fed6786cf92.jpg new file mode 100644 index 0000000000000000000000000000000000000000..26ac320698fb8ef8cb24735cad030e9631ddcefa --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/2ca1eb5ee16c11f85ba9ece76c980550e0f43eb5a07b91cc6fbb1fed6786cf92.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:80f6c735f6212488f3bfd4d3b474e9c347905821fe625ff08f0ce31c71bbb67f +size 43632 diff --git a/parse/train/nzpLWnVAyah/images/3191efa17953497d2e9ffc0c93f807f8b51d41f4b8d3de7bc3e0485cd1eece89.jpg b/parse/train/nzpLWnVAyah/images/3191efa17953497d2e9ffc0c93f807f8b51d41f4b8d3de7bc3e0485cd1eece89.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a65e2544d779b25e21444f08ba814624b0650c75 --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/3191efa17953497d2e9ffc0c93f807f8b51d41f4b8d3de7bc3e0485cd1eece89.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a80665688a24d387de57bd3591892a48aa7863ee47c10e93893c3af43bad0e41 +size 57020 diff --git a/parse/train/nzpLWnVAyah/images/36e44997ce4bc84b6bc75f869fe0272ab87f8a8612e07707a008598221a4ff40.jpg b/parse/train/nzpLWnVAyah/images/36e44997ce4bc84b6bc75f869fe0272ab87f8a8612e07707a008598221a4ff40.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2c06a240f2c12796279276e42650559f70569694 --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/36e44997ce4bc84b6bc75f869fe0272ab87f8a8612e07707a008598221a4ff40.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bf86650f2a1846a746acc6047cd53857924065e8ac724480d3aac87d9cff00e8 +size 29723 diff --git a/parse/train/nzpLWnVAyah/images/3c47d595d071fb108f72fd94bef1fccf858d1b6f1f998d6e516fba81f4a80708.jpg b/parse/train/nzpLWnVAyah/images/3c47d595d071fb108f72fd94bef1fccf858d1b6f1f998d6e516fba81f4a80708.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bf2ce9a78949dd983ae2e327073141cf86a2866a --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/3c47d595d071fb108f72fd94bef1fccf858d1b6f1f998d6e516fba81f4a80708.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bc886cac589bc734d523979461b145ed621ef05138baefd23c68c257998de5f9 +size 71514 diff --git a/parse/train/nzpLWnVAyah/images/3d348f706c864cb62ba315139c71ac8b5ecb6c4825a4126a3fb1b566bf718030.jpg b/parse/train/nzpLWnVAyah/images/3d348f706c864cb62ba315139c71ac8b5ecb6c4825a4126a3fb1b566bf718030.jpg new file mode 100644 index 0000000000000000000000000000000000000000..3a2542a2e1fb3076f292340a36ba99614a58da96 --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/3d348f706c864cb62ba315139c71ac8b5ecb6c4825a4126a3fb1b566bf718030.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fb96e74948a626fb8e1fe0f5b6d1182bfbdd10c856616698ead5abe6c3eddd65 +size 60630 diff --git a/parse/train/nzpLWnVAyah/images/4a893f8dabcba29556295ef549e7aaf7e62b1445705b134780d68aa339017a43.jpg b/parse/train/nzpLWnVAyah/images/4a893f8dabcba29556295ef549e7aaf7e62b1445705b134780d68aa339017a43.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6c3cd24319d2ece56a4b57c2302d8c9518eb8e38 --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/4a893f8dabcba29556295ef549e7aaf7e62b1445705b134780d68aa339017a43.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:27c983cc7973656f5e678c9c687f852b92667571e6a5cd2b6c45173119ab5150 +size 6578 diff --git a/parse/train/nzpLWnVAyah/images/57e7a65c978622e72bb1473321aaeda3de93a912bcd3ee37e09d63cc6b5f65c8.jpg b/parse/train/nzpLWnVAyah/images/57e7a65c978622e72bb1473321aaeda3de93a912bcd3ee37e09d63cc6b5f65c8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..325cdf7e2162596d87fa8d01beb1f39eac92e190 --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/57e7a65c978622e72bb1473321aaeda3de93a912bcd3ee37e09d63cc6b5f65c8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5d5f6dd36831e3ff7c129435b731fb10c08498673264994a45545424219dbfab +size 36281 diff --git a/parse/train/nzpLWnVAyah/images/8e6041c913757b68ade8bd2ce9cc243f8155476e8baa77a8c80c2cef0ac9edf6.jpg b/parse/train/nzpLWnVAyah/images/8e6041c913757b68ade8bd2ce9cc243f8155476e8baa77a8c80c2cef0ac9edf6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2b9262196292031b757419a5329afed9b625e232 --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/8e6041c913757b68ade8bd2ce9cc243f8155476e8baa77a8c80c2cef0ac9edf6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:295785063acef12cb7ce0f74271ef276b424fdfbc3d3589cb3d5de269ce60999 +size 152550 diff --git a/parse/train/nzpLWnVAyah/images/90f258b1a316c75b6c3fd83719d5d45da6c0f0e8634399099a0b6c0069373136.jpg b/parse/train/nzpLWnVAyah/images/90f258b1a316c75b6c3fd83719d5d45da6c0f0e8634399099a0b6c0069373136.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4413163cb3c38ff84391b1f1bae8a42b58eada97 --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/90f258b1a316c75b6c3fd83719d5d45da6c0f0e8634399099a0b6c0069373136.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1a7f472b10a51716a2fe44458537c1fffacc7ca2ecfaab21e20b64d522959dd7 +size 46584 diff --git a/parse/train/nzpLWnVAyah/images/9a66213890a3d625ea59c5e23e6ef48bb4a27bf9977d9ae208e9e6e199e42c24.jpg b/parse/train/nzpLWnVAyah/images/9a66213890a3d625ea59c5e23e6ef48bb4a27bf9977d9ae208e9e6e199e42c24.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dfd42434f5ff4544a648356a7d3d2358872d0c9a --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/9a66213890a3d625ea59c5e23e6ef48bb4a27bf9977d9ae208e9e6e199e42c24.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fb4a95467a1e9e63a7510f228279b98b6f9d55271177804388884b11eed632d4 +size 10364 diff --git a/parse/train/nzpLWnVAyah/images/9ab580f48b2fa5f55b9a68c1a50f9ee6baf599d3dcdf80a319ab2c6f3e99dd50.jpg b/parse/train/nzpLWnVAyah/images/9ab580f48b2fa5f55b9a68c1a50f9ee6baf599d3dcdf80a319ab2c6f3e99dd50.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a853a16f4b0b92c2d1a368b6d674a530315dc21e --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/9ab580f48b2fa5f55b9a68c1a50f9ee6baf599d3dcdf80a319ab2c6f3e99dd50.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:137b319260f6de7f0446b45702b8084e06ae2fbbd543c1e8cf77a407ffcba015 +size 79276 diff --git a/parse/train/nzpLWnVAyah/images/9b2b8cb13be63d20270ba2830c5391dab93cd991e475c515d86a0a0478d13ed1.jpg b/parse/train/nzpLWnVAyah/images/9b2b8cb13be63d20270ba2830c5391dab93cd991e475c515d86a0a0478d13ed1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..89c802933847d3a08f760b6aee9807351ef7ea39 --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/9b2b8cb13be63d20270ba2830c5391dab93cd991e475c515d86a0a0478d13ed1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4acea1fca127959636db88f32ef2795e64e81196aa20dee21e29e7ba0fda32c0 +size 7069 diff --git a/parse/train/nzpLWnVAyah/images/a48e471441bf46291293b3e5d857c3ac31e96d8e550e638819994d2dc05d8be2.jpg b/parse/train/nzpLWnVAyah/images/a48e471441bf46291293b3e5d857c3ac31e96d8e550e638819994d2dc05d8be2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a5e2b93481fbe0c272bc5bcad08bb43b834b21de --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/a48e471441bf46291293b3e5d857c3ac31e96d8e550e638819994d2dc05d8be2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dadc5e7d4318c9db3caf79f69a62c5d2d129f0897d729fd07012179e0f335850 +size 53677 diff --git a/parse/train/nzpLWnVAyah/images/aa8a8b9659df6d688662b4bea2049adae33425a7f0f112cf24a5af9c2ebe29c9.jpg b/parse/train/nzpLWnVAyah/images/aa8a8b9659df6d688662b4bea2049adae33425a7f0f112cf24a5af9c2ebe29c9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..88b72059f7644dc08c861cd23ccbf252090566e1 --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/aa8a8b9659df6d688662b4bea2049adae33425a7f0f112cf24a5af9c2ebe29c9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ca1fc4fc925fcc09bead01619fb658848fd83cbc908179f0631b70b50ab16059 +size 134871 diff --git a/parse/train/nzpLWnVAyah/images/b8e84593c9e859f37803c22973a47e3e0af11e360b1685bcbc1c6724b38ee6e8.jpg b/parse/train/nzpLWnVAyah/images/b8e84593c9e859f37803c22973a47e3e0af11e360b1685bcbc1c6724b38ee6e8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2ad1e3daeae5af345ca580da04aeaec66a706f9f --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/b8e84593c9e859f37803c22973a47e3e0af11e360b1685bcbc1c6724b38ee6e8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b7dd72c374d21103750e739586d88cf992efc01777f0df6adbecf8371ac45c6f +size 40783 diff --git a/parse/train/nzpLWnVAyah/images/c22373173b2d3e4c52148f931644cc42972a88544fe122a835913b32b836d35e.jpg b/parse/train/nzpLWnVAyah/images/c22373173b2d3e4c52148f931644cc42972a88544fe122a835913b32b836d35e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b2fded81b6c00d007f3de5c2382cddcea7be6e81 --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/c22373173b2d3e4c52148f931644cc42972a88544fe122a835913b32b836d35e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9071dffeab2fc445647da5f552030dc5f71761564f41381e63816f17f386593f +size 157065 diff --git a/parse/train/nzpLWnVAyah/images/ca95a024a64a8b0ba9fc93b72ee20f9b1017d95083887ebc0519890dcc0a0e16.jpg b/parse/train/nzpLWnVAyah/images/ca95a024a64a8b0ba9fc93b72ee20f9b1017d95083887ebc0519890dcc0a0e16.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bd458a4b329188f09e3cf9d9665113c65245f377 --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/ca95a024a64a8b0ba9fc93b72ee20f9b1017d95083887ebc0519890dcc0a0e16.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4db83540bd74a3082f20c08362698b6f5d4f8c1c835fa459b9ba82be53a63cac +size 55000 diff --git a/parse/train/nzpLWnVAyah/images/e666631860a1280ac5795253fc6d64c705542da5faa8e31d107fe066b7fd8042.jpg b/parse/train/nzpLWnVAyah/images/e666631860a1280ac5795253fc6d64c705542da5faa8e31d107fe066b7fd8042.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4d73eb82de5db6cb4d342f20b643ac27f73efaf7 --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/e666631860a1280ac5795253fc6d64c705542da5faa8e31d107fe066b7fd8042.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:74035b9f4d145c5c6aaace2d51ef4fb46862659b66853cbb957ff5dc80045c99 +size 32253 diff --git a/parse/train/nzpLWnVAyah/images/e8af2a647d12834bd5ffbce865436fae63c57e76348d1aebd96e277cf5707c12.jpg b/parse/train/nzpLWnVAyah/images/e8af2a647d12834bd5ffbce865436fae63c57e76348d1aebd96e277cf5707c12.jpg new file mode 100644 index 0000000000000000000000000000000000000000..297e6b321096c7b266c5bb33f1e969be75592350 --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/e8af2a647d12834bd5ffbce865436fae63c57e76348d1aebd96e277cf5707c12.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2475b88ee1ee7fd67acae2d43ed84abf2f227d29ceb891d2733a64da5756636c +size 4787 diff --git a/parse/train/nzpLWnVAyah/images/f80e109368e2f0c95feae8d7bd688ef690c92a1c69f2044b53235fdd49e4a206.jpg b/parse/train/nzpLWnVAyah/images/f80e109368e2f0c95feae8d7bd688ef690c92a1c69f2044b53235fdd49e4a206.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9bcfb71bd2a1bc0d6007ba4647355b93e98f91ec --- /dev/null +++ b/parse/train/nzpLWnVAyah/images/f80e109368e2f0c95feae8d7bd688ef690c92a1c69f2044b53235fdd49e4a206.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6d3759f4d686b2a8d0f8898fb0cea468bba64b2632185aa72b14c7ff135608cb +size 136630 diff --git a/parse/train/rJMcdsA5FX/images/0df3c385897ed64919fb9e3f39d2ca80b219fa7a079faf7cae307ac228a97b55.jpg b/parse/train/rJMcdsA5FX/images/0df3c385897ed64919fb9e3f39d2ca80b219fa7a079faf7cae307ac228a97b55.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0e1fab9406f8c9bdf26401ba9d0b61838ebd9605 --- /dev/null +++ b/parse/train/rJMcdsA5FX/images/0df3c385897ed64919fb9e3f39d2ca80b219fa7a079faf7cae307ac228a97b55.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0ab5ff2d91e4fa9df6aba2f2598023e671b1bba5df911a24b195597a5486492f +size 40718 diff --git a/parse/train/rJMcdsA5FX/images/1aa609605e995a6e9074f63b0c576953a10eb3fd26ec5d445183f4bf7214f311.jpg b/parse/train/rJMcdsA5FX/images/1aa609605e995a6e9074f63b0c576953a10eb3fd26ec5d445183f4bf7214f311.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d37bec60d8da0c1907e45da7696668abf3b049e4 --- /dev/null +++ b/parse/train/rJMcdsA5FX/images/1aa609605e995a6e9074f63b0c576953a10eb3fd26ec5d445183f4bf7214f311.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6dbf3b4d5b296adb0b987cda5901854a0d88efa0c2cee73eca00e4da2531e67f +size 43046 diff --git a/parse/train/rJMcdsA5FX/images/20c86c4fc1cfc11ca2cd3b647a5abbbbd546585021c3c938f00f015952633895.jpg b/parse/train/rJMcdsA5FX/images/20c86c4fc1cfc11ca2cd3b647a5abbbbd546585021c3c938f00f015952633895.jpg new file mode 100644 index 0000000000000000000000000000000000000000..51ae71a051d995e91549a5ee93ac1fabed622cce --- /dev/null +++ b/parse/train/rJMcdsA5FX/images/20c86c4fc1cfc11ca2cd3b647a5abbbbd546585021c3c938f00f015952633895.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7ecc3bc953828528f659ee459ebb3107ffeb1de665c9413b6022f7f1cf6eabf9 +size 21525 diff --git a/parse/train/rJMcdsA5FX/images/227d6f4004d00727d8bf35132de1faa2ce5c73e0719438e56a468545e2537968.jpg b/parse/train/rJMcdsA5FX/images/227d6f4004d00727d8bf35132de1faa2ce5c73e0719438e56a468545e2537968.jpg new file mode 100644 index 0000000000000000000000000000000000000000..24063e30600d2736adc7ed476b9ebc4df7fffcb7 --- /dev/null +++ b/parse/train/rJMcdsA5FX/images/227d6f4004d00727d8bf35132de1faa2ce5c73e0719438e56a468545e2537968.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ea48509604bb5c7ea96698744ba77be1349cb888e9b72f93848e5c1208be6bc3 +size 70416 diff --git a/parse/train/rJMcdsA5FX/images/2d66806817441a4e7cd4b5ffcccb5a8f34ea10f6fe1f4fe596085100d972243d.jpg b/parse/train/rJMcdsA5FX/images/2d66806817441a4e7cd4b5ffcccb5a8f34ea10f6fe1f4fe596085100d972243d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..42b0c26ac2b27d343a5e58bc56e65aabb2bfd3a9 --- /dev/null +++ b/parse/train/rJMcdsA5FX/images/2d66806817441a4e7cd4b5ffcccb5a8f34ea10f6fe1f4fe596085100d972243d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c7f9c53a98f106cc29c6455736771d0efb999d124fad4deb6099967c0c5a06ec +size 54779 diff --git a/parse/train/rJMcdsA5FX/images/422a3abb9625417983ec2390a4d1bcfab21050655d4379154a223d91af23d5d1.jpg b/parse/train/rJMcdsA5FX/images/422a3abb9625417983ec2390a4d1bcfab21050655d4379154a223d91af23d5d1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..8fe2ffd0a00da904bf3dfeffaee5a33f60512825 --- /dev/null +++ b/parse/train/rJMcdsA5FX/images/422a3abb9625417983ec2390a4d1bcfab21050655d4379154a223d91af23d5d1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c62388907463288aa5bf09f8d4642e6d182d6f0fece92d0ac3b83e6306433772 +size 31857 diff --git a/parse/train/rJMcdsA5FX/images/8e0ad235d8c4512ba3eb632c3d44519952b7a10f6c4dd71df3d8ef3d55a7d4b5.jpg b/parse/train/rJMcdsA5FX/images/8e0ad235d8c4512ba3eb632c3d44519952b7a10f6c4dd71df3d8ef3d55a7d4b5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..090eada7957dc81481e7812299bef104371a923f --- /dev/null +++ b/parse/train/rJMcdsA5FX/images/8e0ad235d8c4512ba3eb632c3d44519952b7a10f6c4dd71df3d8ef3d55a7d4b5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ef5917456f5d797f6e87086df63cedae546bed54fd1eb786f383f31b3b4ef9ec +size 10204 diff --git a/parse/train/rJMcdsA5FX/images/9e503f8dbad11444d251950e1bee5dc03336300e0bdab930a5d7dba48de97b17.jpg b/parse/train/rJMcdsA5FX/images/9e503f8dbad11444d251950e1bee5dc03336300e0bdab930a5d7dba48de97b17.jpg new file mode 100644 index 0000000000000000000000000000000000000000..207d6cb5fda76203c39ce38539da5b4b673a8509 --- /dev/null +++ b/parse/train/rJMcdsA5FX/images/9e503f8dbad11444d251950e1bee5dc03336300e0bdab930a5d7dba48de97b17.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:404b051b8dc274a74d4bd92275959085dd381d1aeb78b713d0ed5f2a40c7583a +size 159480 diff --git a/parse/train/ryUlhzWCZ/images/0a60fc25dd766e79cb183aadac82633c14d2c1ceb105e6f8c393a19d9cb3e718.jpg b/parse/train/ryUlhzWCZ/images/0a60fc25dd766e79cb183aadac82633c14d2c1ceb105e6f8c393a19d9cb3e718.jpg new file mode 100644 index 0000000000000000000000000000000000000000..1f048c4f7e76730636ece3d6f5d3eb0a1f8bbf05 --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/0a60fc25dd766e79cb183aadac82633c14d2c1ceb105e6f8c393a19d9cb3e718.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b91d89a618bd1ebfdd01484ac3ee5a524d61b945127f78b0317807dc6190fcf1 +size 119104 diff --git a/parse/train/ryUlhzWCZ/images/0c62ed8c17a2404928864988ea594fcdb1fa14001a5a9833fc8e6fd30ad1fa01.jpg b/parse/train/ryUlhzWCZ/images/0c62ed8c17a2404928864988ea594fcdb1fa14001a5a9833fc8e6fd30ad1fa01.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c671efe58c104efb6eedf4ed76aedfdee1036099 --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/0c62ed8c17a2404928864988ea594fcdb1fa14001a5a9833fc8e6fd30ad1fa01.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:95698fa4ca669527c6acef29b32c09f8c7807e517d14bf5905810b2d62f3eb31 +size 6836 diff --git a/parse/train/ryUlhzWCZ/images/10b4958a755369c971b6faa6833f7e329f47fec23a64c101e7215c0b06062cb6.jpg b/parse/train/ryUlhzWCZ/images/10b4958a755369c971b6faa6833f7e329f47fec23a64c101e7215c0b06062cb6.jpg new file mode 100644 index 0000000000000000000000000000000000000000..bf44299ea37fa8f0938ad4b9060174492dd0a1b9 --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/10b4958a755369c971b6faa6833f7e329f47fec23a64c101e7215c0b06062cb6.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d395b567da93b19e874fe0f3cb8c2d87f18535444260b6076356e7b5785cadcb +size 15604 diff --git a/parse/train/ryUlhzWCZ/images/295b509937dcbf68cc1cd4cda5591b7c5d36d809fa772174a33d31a34978ef68.jpg b/parse/train/ryUlhzWCZ/images/295b509937dcbf68cc1cd4cda5591b7c5d36d809fa772174a33d31a34978ef68.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f200e68a6c9f34b14ff0a310424d7c5ba144ad59 --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/295b509937dcbf68cc1cd4cda5591b7c5d36d809fa772174a33d31a34978ef68.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bd86a2e4d3988d0cdc5147583d5958d7ab09eb613c1cf3da6e3823c4a8d14cd9 +size 19472 diff --git a/parse/train/ryUlhzWCZ/images/4d73247096383f71927694671b41d4f4a200910e3625be0b760d4e293fc8936e.jpg b/parse/train/ryUlhzWCZ/images/4d73247096383f71927694671b41d4f4a200910e3625be0b760d4e293fc8936e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..988b0b3c9137d8c635ff1760bc1c665fff81f6e7 --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/4d73247096383f71927694671b41d4f4a200910e3625be0b760d4e293fc8936e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5c417c4db771e6484aae0284ffc9d95333eee6fcfb34bfd73d11cbdba3060f88 +size 6449 diff --git a/parse/train/ryUlhzWCZ/images/565f0feca138756b6e8195f96957546a08f1d5dfedf6e40b3b99903e70b71efd.jpg b/parse/train/ryUlhzWCZ/images/565f0feca138756b6e8195f96957546a08f1d5dfedf6e40b3b99903e70b71efd.jpg new file mode 100644 index 0000000000000000000000000000000000000000..110c0d682c2655c09cdf6d800c8f3274476bdf4a --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/565f0feca138756b6e8195f96957546a08f1d5dfedf6e40b3b99903e70b71efd.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f36882f45f7f01b85c258838aedfb68bb0f86d5715ca0688bdb362ac64bd1e3e +size 4663 diff --git a/parse/train/ryUlhzWCZ/images/609a929f379d237d60d4984ff8ab90798a726d06ad088233ddce9c5ae2512fda.jpg b/parse/train/ryUlhzWCZ/images/609a929f379d237d60d4984ff8ab90798a726d06ad088233ddce9c5ae2512fda.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a225ab22f5fb95325a73cf6ec5b88aa282f47f93 --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/609a929f379d237d60d4984ff8ab90798a726d06ad088233ddce9c5ae2512fda.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d9a87fed88809a96f8dfd96363c2ebf3907db9a70bb5bcf8e19e09fe5c8fca94 +size 8804 diff --git a/parse/train/ryUlhzWCZ/images/618cef080b7db9a9ac2eceafa3462d4095d5c0079c8dbaf8a00be6d51632f671.jpg b/parse/train/ryUlhzWCZ/images/618cef080b7db9a9ac2eceafa3462d4095d5c0079c8dbaf8a00be6d51632f671.jpg new file mode 100644 index 0000000000000000000000000000000000000000..93302189c17170590fdb0b543ca68999d41ea027 --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/618cef080b7db9a9ac2eceafa3462d4095d5c0079c8dbaf8a00be6d51632f671.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cc59cf129c46c565f820604566479d51159ba22198c63eabcf0fc9766099fd4a +size 23164 diff --git a/parse/train/ryUlhzWCZ/images/6d6900d2126ea36c2e5ff17408f2e38d32ea900734bd973331775adcf1588471.jpg b/parse/train/ryUlhzWCZ/images/6d6900d2126ea36c2e5ff17408f2e38d32ea900734bd973331775adcf1588471.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4ee7114df9c05394b0835994861ed22d98b6603a --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/6d6900d2126ea36c2e5ff17408f2e38d32ea900734bd973331775adcf1588471.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:50fd35985d1fa70719e591f8706ccc75a4245520eedc3e0ff2a6fbaca38a26e3 +size 24424 diff --git a/parse/train/ryUlhzWCZ/images/80284d922486510af6d136e0fd21e498f9a1d1fc21c4950a523676fa0f7afdd7.jpg b/parse/train/ryUlhzWCZ/images/80284d922486510af6d136e0fd21e498f9a1d1fc21c4950a523676fa0f7afdd7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ce61b26df24b3103527a68d66125b53b7ff22f03 --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/80284d922486510af6d136e0fd21e498f9a1d1fc21c4950a523676fa0f7afdd7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f332522c0fc7c373cd3eeed6846885ebab3796c834d3e0a649837f2728b8358a +size 50478 diff --git a/parse/train/ryUlhzWCZ/images/9dd5bee0a660b49f2853e3c3224939397ea9ec85ae2177e1015eb63cd45b64e2.jpg b/parse/train/ryUlhzWCZ/images/9dd5bee0a660b49f2853e3c3224939397ea9ec85ae2177e1015eb63cd45b64e2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c4bb7704b39aa4dc7a1259526c3c666c05181fbf --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/9dd5bee0a660b49f2853e3c3224939397ea9ec85ae2177e1015eb63cd45b64e2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5714bc5dfc5505672fd7c0213c643af3d805dda529bf6e158e37e6d3f5b44ab8 +size 11941 diff --git a/parse/train/ryUlhzWCZ/images/9eeeab7e50421a167ecbb633d84fb8ef142c374ed98765d8096f78a25c91b62d.jpg b/parse/train/ryUlhzWCZ/images/9eeeab7e50421a167ecbb633d84fb8ef142c374ed98765d8096f78a25c91b62d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f16a7a91c778b44e88a73d537c5f2b8e9f73181b --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/9eeeab7e50421a167ecbb633d84fb8ef142c374ed98765d8096f78a25c91b62d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2402d0bd36fd79aa646efb6668ca15524a0d7da935f53df6938e759d67f5df72 +size 6706 diff --git a/parse/train/ryUlhzWCZ/images/a0594bc31da2ab8bbaec0336684d89314745224541705d890fbbc61d5f28327b.jpg b/parse/train/ryUlhzWCZ/images/a0594bc31da2ab8bbaec0336684d89314745224541705d890fbbc61d5f28327b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cc4df3cd6a02d70eb5530b1e2e91847bd5b15bde --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/a0594bc31da2ab8bbaec0336684d89314745224541705d890fbbc61d5f28327b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8f0d2c17571a5e8ff08edb7d65e43d7eb141698ab73e7d1e55bf4e3b52e6bfeb +size 8674 diff --git a/parse/train/ryUlhzWCZ/images/a7356ecb2f2a9f5b2dcdc25bf44f13f253b43e9c2308ab5be079eab9b73cd49d.jpg b/parse/train/ryUlhzWCZ/images/a7356ecb2f2a9f5b2dcdc25bf44f13f253b43e9c2308ab5be079eab9b73cd49d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..f7a0b68ac7338861f29c02ad930ce0a2283867ca --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/a7356ecb2f2a9f5b2dcdc25bf44f13f253b43e9c2308ab5be079eab9b73cd49d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7368fb284c17aaf46f28aacf75b959c48f7554e524d8b06d8baef4addf52e39c +size 38421 diff --git a/parse/train/ryUlhzWCZ/images/abc9d244711254b2baf3dfe3303685c997a82425b0ddb8e784d78475440af170.jpg b/parse/train/ryUlhzWCZ/images/abc9d244711254b2baf3dfe3303685c997a82425b0ddb8e784d78475440af170.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a8a5eeb39783cdac79db526c9a8dbf4f4a2f8f6f --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/abc9d244711254b2baf3dfe3303685c997a82425b0ddb8e784d78475440af170.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1e53181be0c03cc22e28eb4292941e777babcb8189816d9ab2607941a4900c32 +size 5141 diff --git a/parse/train/ryUlhzWCZ/images/b65f759e13c3bfec893e23678eadf2206468017f9d2ceb6c81ab620d6c642b7d.jpg b/parse/train/ryUlhzWCZ/images/b65f759e13c3bfec893e23678eadf2206468017f9d2ceb6c81ab620d6c642b7d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..37a3b8acd8c771311e1701a8ae4905e14c0530ea --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/b65f759e13c3bfec893e23678eadf2206468017f9d2ceb6c81ab620d6c642b7d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bdd6e8738f67f2f7aba5f3f20aba9e1189d6dc3728605a6c6d73d9698bcf25b5 +size 6562 diff --git a/parse/train/ryUlhzWCZ/images/ba6421da7be698d6baace476cad143acaf089c7de27ddcbf16cea74dd708fbc5.jpg b/parse/train/ryUlhzWCZ/images/ba6421da7be698d6baace476cad143acaf089c7de27ddcbf16cea74dd708fbc5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..695af60662220932fb4bd53aef2b4c97cd0ce415 --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/ba6421da7be698d6baace476cad143acaf089c7de27ddcbf16cea74dd708fbc5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ebb2c2afc29d03ff86a86c5f60319465383e5372e10eeee89daff9f5fa865579 +size 4606 diff --git a/parse/train/ryUlhzWCZ/images/bd3928672cc4b735dc03d3fb2c9ed432af6e6e76fa64eecccc0191ac26689f85.jpg b/parse/train/ryUlhzWCZ/images/bd3928672cc4b735dc03d3fb2c9ed432af6e6e76fa64eecccc0191ac26689f85.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2d40b150a6a989a5382cc5a75e66fc0b46b99230 --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/bd3928672cc4b735dc03d3fb2c9ed432af6e6e76fa64eecccc0191ac26689f85.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:008c59b622596647971121f195e97ece6254575ebbae7687f85f787dab3fe500 +size 17962 diff --git a/parse/train/ryUlhzWCZ/images/c0dd8c50bf04b425719647bf1d59345149e20ebb18bb0691da4da5e755d46154.jpg b/parse/train/ryUlhzWCZ/images/c0dd8c50bf04b425719647bf1d59345149e20ebb18bb0691da4da5e755d46154.jpg new file mode 100644 index 0000000000000000000000000000000000000000..349cd7ed31e865b1ba5f92cbf0f45c1510306b36 --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/c0dd8c50bf04b425719647bf1d59345149e20ebb18bb0691da4da5e755d46154.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:84b43e7479fa40ba772106db2c52955828bed5b922c747a6dc3cc7d539eee5aa +size 118054 diff --git a/parse/train/ryUlhzWCZ/images/c63faf26a3cfaa525a8154d21c312c4d8a5e620a82d5e88f156179828b2a7d30.jpg b/parse/train/ryUlhzWCZ/images/c63faf26a3cfaa525a8154d21c312c4d8a5e620a82d5e88f156179828b2a7d30.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cf4e84083122ac9b5bdfb4a47bf03a973eeaa4ca --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/c63faf26a3cfaa525a8154d21c312c4d8a5e620a82d5e88f156179828b2a7d30.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7db8242556368155aa866882b9833379fbef15f3360f07028354ab4f2045cc69 +size 6524 diff --git a/parse/train/ryUlhzWCZ/images/cf1bceb0e6c2cb9f812301b70a8c77a35fa7fd5e4afe66e08c868fa77f7877af.jpg b/parse/train/ryUlhzWCZ/images/cf1bceb0e6c2cb9f812301b70a8c77a35fa7fd5e4afe66e08c868fa77f7877af.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9fa6b04b46e42944cce09e3c6c467a0d06cb29ae --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/cf1bceb0e6c2cb9f812301b70a8c77a35fa7fd5e4afe66e08c868fa77f7877af.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c957832542fa7f7661b6752250a82529c27222846cc24d141c122e84ecbcd2d1 +size 6639 diff --git a/parse/train/ryUlhzWCZ/images/dcba1130782b5b067ab41d949d07adec8498cba0f3b77cc86058d74972d02453.jpg b/parse/train/ryUlhzWCZ/images/dcba1130782b5b067ab41d949d07adec8498cba0f3b77cc86058d74972d02453.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7de8d1cd3019c5685cc976b8a6a82ed8fb35b086 --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/dcba1130782b5b067ab41d949d07adec8498cba0f3b77cc86058d74972d02453.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f30c5c759b6d8bf904aed8ca4bb1a54561d0004f93e937181a45a424be7b7eb1 +size 15160 diff --git a/parse/train/ryUlhzWCZ/images/dfaff605ea378a79cac5754dabdf9e9e7ea742c1d70e71fc43491097dde6bba1.jpg b/parse/train/ryUlhzWCZ/images/dfaff605ea378a79cac5754dabdf9e9e7ea742c1d70e71fc43491097dde6bba1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d7b58c0c0c4dd2c098b4a97a588071da4dd38d13 --- /dev/null +++ b/parse/train/ryUlhzWCZ/images/dfaff605ea378a79cac5754dabdf9e9e7ea742c1d70e71fc43491097dde6bba1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:806c86d732aba1acf977d8245ef106933fc4bf990e255c99ec0c63a9dec190b9 +size 5214 diff --git a/parse/train/rygeHgSFDH/images/06bc29372b002fdb1babf2b606fbd141ceb2aa4f6db98963f9ae706843b53ace.jpg b/parse/train/rygeHgSFDH/images/06bc29372b002fdb1babf2b606fbd141ceb2aa4f6db98963f9ae706843b53ace.jpg new file mode 100644 index 0000000000000000000000000000000000000000..89126cc7252806122eeb98c8506aabd249219d58 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/06bc29372b002fdb1babf2b606fbd141ceb2aa4f6db98963f9ae706843b53ace.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6c9b349fcec4300b03290abbceaeb2dc03aa16ef62b0de55adc7b0bb99ace815 +size 14910 diff --git a/parse/train/rygeHgSFDH/images/08a23af0aa91e36b9530e8e754e35342ea717c63742d33852564748746fc98df.jpg b/parse/train/rygeHgSFDH/images/08a23af0aa91e36b9530e8e754e35342ea717c63742d33852564748746fc98df.jpg new file mode 100644 index 0000000000000000000000000000000000000000..74ad31158b643b7c72468e09e7617b8f0719b32e --- /dev/null +++ b/parse/train/rygeHgSFDH/images/08a23af0aa91e36b9530e8e754e35342ea717c63742d33852564748746fc98df.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:914b35a39b03e9fd18538e2b60b27682a307d1cb66f8fc99b65215f7bdbcfb88 +size 42552 diff --git a/parse/train/rygeHgSFDH/images/08aacec8b565dd5ece128e23d2291a98644dc2d6cbc03465e11bc02d07d1b915.jpg b/parse/train/rygeHgSFDH/images/08aacec8b565dd5ece128e23d2291a98644dc2d6cbc03465e11bc02d07d1b915.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0c95498e2ff0690acc0bcf03bf6d3e50f7147d5e --- /dev/null +++ b/parse/train/rygeHgSFDH/images/08aacec8b565dd5ece128e23d2291a98644dc2d6cbc03465e11bc02d07d1b915.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4257d952740429470883c6d07785ceae99d434f306e7a3b03d88c4faf48461e1 +size 14300 diff --git a/parse/train/rygeHgSFDH/images/090ec6671c64890af51a975547a8637f1d6d9ce3cd648fbdcb0ed8f3388747f4.jpg b/parse/train/rygeHgSFDH/images/090ec6671c64890af51a975547a8637f1d6d9ce3cd648fbdcb0ed8f3388747f4.jpg new file mode 100644 index 0000000000000000000000000000000000000000..056f703f2b8721617a389f7d6ff5ebab51b0d7ed --- /dev/null +++ b/parse/train/rygeHgSFDH/images/090ec6671c64890af51a975547a8637f1d6d9ce3cd648fbdcb0ed8f3388747f4.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bfc4118642ed2beedf78512a08ac9b6c13e8fac5754f39e16e20aab44f7493e3 +size 10294 diff --git a/parse/train/rygeHgSFDH/images/0bc0b27ca77cb6e38600939b4d79c85ff44e5db51c709d0d6f9336359b17f7c2.jpg b/parse/train/rygeHgSFDH/images/0bc0b27ca77cb6e38600939b4d79c85ff44e5db51c709d0d6f9336359b17f7c2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9d34a7e4d7255b08622ff04cbf834cb325c7ed8a --- /dev/null +++ b/parse/train/rygeHgSFDH/images/0bc0b27ca77cb6e38600939b4d79c85ff44e5db51c709d0d6f9336359b17f7c2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:dadfa655cced1a521b5f723a347f73b4cce55bfa57c5ec5e6faf317f420707d5 +size 4108 diff --git a/parse/train/rygeHgSFDH/images/0c0d02761112e2eefe700c362c247b7fec05421fc9fbdd228b1603378ecea080.jpg b/parse/train/rygeHgSFDH/images/0c0d02761112e2eefe700c362c247b7fec05421fc9fbdd228b1603378ecea080.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9ca0d1dd88147ba6d6d9807e740c9ee340da424b --- /dev/null +++ b/parse/train/rygeHgSFDH/images/0c0d02761112e2eefe700c362c247b7fec05421fc9fbdd228b1603378ecea080.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5aa063c85740ae520ce633c228c4c14ecbb8cebb25d6a77581e4ae7fcf8923c8 +size 9630 diff --git a/parse/train/rygeHgSFDH/images/0c2f53f89a24f1b96d86d6eb2a963f85c0fc5cba08bd9b7712fd0ded79c611e9.jpg b/parse/train/rygeHgSFDH/images/0c2f53f89a24f1b96d86d6eb2a963f85c0fc5cba08bd9b7712fd0ded79c611e9.jpg new file mode 100644 index 0000000000000000000000000000000000000000..be4c9c1a8668b2ff1dada994071895eb3c91e155 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/0c2f53f89a24f1b96d86d6eb2a963f85c0fc5cba08bd9b7712fd0ded79c611e9.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3c03b1e620f3ecba911cc6928922ad260b328f7caef3bbc16893effde764ea75 +size 198511 diff --git a/parse/train/rygeHgSFDH/images/0fbd7e4feaa6d11cc8234fc810961884b9a40edc1b9e8d6129554527fccdd34f.jpg b/parse/train/rygeHgSFDH/images/0fbd7e4feaa6d11cc8234fc810961884b9a40edc1b9e8d6129554527fccdd34f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ada400227f2ec510b74725fc07d595735614bd73 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/0fbd7e4feaa6d11cc8234fc810961884b9a40edc1b9e8d6129554527fccdd34f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:56ca948590b8033eccb5cc86c84304f1263932d52153b22f7bbd05a0faeaebda +size 192283 diff --git a/parse/train/rygeHgSFDH/images/1b3b34f82c0e12568863ab8047d704de0454df1fcb4f95161fc125019ef66612.jpg b/parse/train/rygeHgSFDH/images/1b3b34f82c0e12568863ab8047d704de0454df1fcb4f95161fc125019ef66612.jpg new file mode 100644 index 0000000000000000000000000000000000000000..89dd1658cc187654f3f6d3a57d5bd12d39250713 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/1b3b34f82c0e12568863ab8047d704de0454df1fcb4f95161fc125019ef66612.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6df4fc3dbff052b4dff2d27cf60dfbdcac067215c44b7ef0e1d5af8a8c5e5e82 +size 35638 diff --git a/parse/train/rygeHgSFDH/images/27968c3da2b642219dea2b4d0333ff044bf0bacfb0570f66da0de6347fa050a5.jpg b/parse/train/rygeHgSFDH/images/27968c3da2b642219dea2b4d0333ff044bf0bacfb0570f66da0de6347fa050a5.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b5adf7b6327a950bc95258b3fb4ea2cb5a0b52cd --- /dev/null +++ b/parse/train/rygeHgSFDH/images/27968c3da2b642219dea2b4d0333ff044bf0bacfb0570f66da0de6347fa050a5.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fa1c2f38be39bcd7566dc729eb1414be3a2a9122e60a93b62aec208b87934d97 +size 15886 diff --git a/parse/train/rygeHgSFDH/images/2a0060de14a884c380094b312fdd8d0282a24a9575efff9a6ef5bbb3fba7f33a.jpg b/parse/train/rygeHgSFDH/images/2a0060de14a884c380094b312fdd8d0282a24a9575efff9a6ef5bbb3fba7f33a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..5274f6c79e228fb05bd3bdd0884cda3fe2f0c0ef --- /dev/null +++ b/parse/train/rygeHgSFDH/images/2a0060de14a884c380094b312fdd8d0282a24a9575efff9a6ef5bbb3fba7f33a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e0c3ebaefd2281d1e6c82f3861e5642f006d75482ae519554a5b95ffa93efa89 +size 43571 diff --git a/parse/train/rygeHgSFDH/images/2a5169d5222b3683fbb02c1b6a2847965e01a4b40bcefa4d09a2599aab885e2d.jpg b/parse/train/rygeHgSFDH/images/2a5169d5222b3683fbb02c1b6a2847965e01a4b40bcefa4d09a2599aab885e2d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..540c7d586fe38c6272af35dd7684f28bfc2f7743 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/2a5169d5222b3683fbb02c1b6a2847965e01a4b40bcefa4d09a2599aab885e2d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c7b98de286845f2543e29434765a2d2eb2489dc94ce372c1cd55f4f1df800f3c +size 1829 diff --git a/parse/train/rygeHgSFDH/images/30b9d6cc307ab0d2e6358fe02ab461d744fc5e2291c31f24d38b5c985a082469.jpg b/parse/train/rygeHgSFDH/images/30b9d6cc307ab0d2e6358fe02ab461d744fc5e2291c31f24d38b5c985a082469.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a7075edce86c98a3bbc79cf473482bfe1bb9d6b7 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/30b9d6cc307ab0d2e6358fe02ab461d744fc5e2291c31f24d38b5c985a082469.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6684094b41b90bff44e17ea5c0fe2684032f019405dbcfbf226a2f110e4a63eb +size 46620 diff --git a/parse/train/rygeHgSFDH/images/3451edba3e04d905054ca8f1e6a2a51b6bc7b26f470090e4da4b5428f59c222b.jpg b/parse/train/rygeHgSFDH/images/3451edba3e04d905054ca8f1e6a2a51b6bc7b26f470090e4da4b5428f59c222b.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e5f21cfc766408daa4ec7d3a15e524edf8dccfea --- /dev/null +++ b/parse/train/rygeHgSFDH/images/3451edba3e04d905054ca8f1e6a2a51b6bc7b26f470090e4da4b5428f59c222b.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:20e56762d01a601d00fb033d56b857bae93b5b0818c66fca42cd22597c9359c8 +size 6272 diff --git a/parse/train/rygeHgSFDH/images/48b41c9f05d1da4c03c00d4461f2a9041607da8348e6f0def91536b4da033499.jpg b/parse/train/rygeHgSFDH/images/48b41c9f05d1da4c03c00d4461f2a9041607da8348e6f0def91536b4da033499.jpg new file mode 100644 index 0000000000000000000000000000000000000000..19347148b19ffd7e8696fbe9c4b7afc0229b572a --- /dev/null +++ b/parse/train/rygeHgSFDH/images/48b41c9f05d1da4c03c00d4461f2a9041607da8348e6f0def91536b4da033499.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:931f50646090ba55baacc548b7663f814270b402b0e9d731f2b089256ec108b4 +size 9650 diff --git a/parse/train/rygeHgSFDH/images/4c25749db65abf80480239b66996929072a3d5a06be3c48a9928f914df8e1512.jpg b/parse/train/rygeHgSFDH/images/4c25749db65abf80480239b66996929072a3d5a06be3c48a9928f914df8e1512.jpg new file mode 100644 index 0000000000000000000000000000000000000000..cf97998d9cddcc6f0cef3e1ec3d4899418d2d0e9 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/4c25749db65abf80480239b66996929072a3d5a06be3c48a9928f914df8e1512.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:13b8eb496c6dfb08cb557c36a9eff69ed96a04978525b56ee095c32eeb9d8caa +size 2751 diff --git a/parse/train/rygeHgSFDH/images/5788390e5eec6a322fe0450ba75daf121c3f0a0a8ce5bfcc22b1d14f2f31a334.jpg b/parse/train/rygeHgSFDH/images/5788390e5eec6a322fe0450ba75daf121c3f0a0a8ce5bfcc22b1d14f2f31a334.jpg new file mode 100644 index 0000000000000000000000000000000000000000..38bc71776b9e9e3206bd425b095f475e704b5244 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/5788390e5eec6a322fe0450ba75daf121c3f0a0a8ce5bfcc22b1d14f2f31a334.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:559b3187ac7e0124e7e1652c612a6260ccc2b1b2337594f306770d8254763d31 +size 36694 diff --git a/parse/train/rygeHgSFDH/images/5bb79ae7e6e76fa0a40cb194f806f4dc632c698b4aa9323e9b0a57e773e259b7.jpg b/parse/train/rygeHgSFDH/images/5bb79ae7e6e76fa0a40cb194f806f4dc632c698b4aa9323e9b0a57e773e259b7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..aae02019bdeec128f00ad55db012321f5d93c56a --- /dev/null +++ b/parse/train/rygeHgSFDH/images/5bb79ae7e6e76fa0a40cb194f806f4dc632c698b4aa9323e9b0a57e773e259b7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:84de2b04433f1773e18bde039fa58ebd2d705454a8b96f79a99c45d35be68b92 +size 32618 diff --git a/parse/train/rygeHgSFDH/images/5c2ef98ccecb74ed5a98234b677f8585c202da4a0cdf8391a573ac03a8f0a53a.jpg b/parse/train/rygeHgSFDH/images/5c2ef98ccecb74ed5a98234b677f8585c202da4a0cdf8391a573ac03a8f0a53a.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a68ca01d5d5b832607adfcaf1eef27f23ed3cc05 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/5c2ef98ccecb74ed5a98234b677f8585c202da4a0cdf8391a573ac03a8f0a53a.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:04cfe42bd3d6169ef97cef2d70edbc5c3498ea470f98bea11186ae63ad539ef0 +size 43950 diff --git a/parse/train/rygeHgSFDH/images/5d11f7e16c1935c8c48ed32f4d2eebdb728fd36d0fbaf8569da0d838bfb1fda3.jpg b/parse/train/rygeHgSFDH/images/5d11f7e16c1935c8c48ed32f4d2eebdb728fd36d0fbaf8569da0d838bfb1fda3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..64783843c01ed89d8b5fa7b9783770f02f5a2746 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/5d11f7e16c1935c8c48ed32f4d2eebdb728fd36d0fbaf8569da0d838bfb1fda3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:478901448447e48038385d44147ea6857225c7ca60b56829b72fb1015e23310e +size 27444 diff --git a/parse/train/rygeHgSFDH/images/5df6c7c4dd9186097a25e7b2aa367a2ac923678665440fa879e063d7767fd980.jpg b/parse/train/rygeHgSFDH/images/5df6c7c4dd9186097a25e7b2aa367a2ac923678665440fa879e063d7767fd980.jpg new file mode 100644 index 0000000000000000000000000000000000000000..34b7880e4225c8606c33ee3dc99402d66ee9db8a --- /dev/null +++ b/parse/train/rygeHgSFDH/images/5df6c7c4dd9186097a25e7b2aa367a2ac923678665440fa879e063d7767fd980.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6b343e2c8713d4805e21a69267ecf57c49498cb709c95ee81f5f17a007081b16 +size 9479 diff --git a/parse/train/rygeHgSFDH/images/6118d93d5a11ab491286f1599a7cb72a0359f509fdb0be03b0f488427eb5c40e.jpg b/parse/train/rygeHgSFDH/images/6118d93d5a11ab491286f1599a7cb72a0359f509fdb0be03b0f488427eb5c40e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..07ff41bd88e4f69278d4ab5d8cb5d3409a8f109f --- /dev/null +++ b/parse/train/rygeHgSFDH/images/6118d93d5a11ab491286f1599a7cb72a0359f509fdb0be03b0f488427eb5c40e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:94bdaa9fcfbdfaabd05a618869bdf26c113895d97eefad5d18f3b2ce2a94b171 +size 47254 diff --git a/parse/train/rygeHgSFDH/images/64133e7c3fb7e4d18667cdd6b160749f5f8e73779a05adc633473c7ed03d921e.jpg b/parse/train/rygeHgSFDH/images/64133e7c3fb7e4d18667cdd6b160749f5f8e73779a05adc633473c7ed03d921e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..29ccbdb1f4c7bf9bfca90496fafd23f6af4ccbb2 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/64133e7c3fb7e4d18667cdd6b160749f5f8e73779a05adc633473c7ed03d921e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:478e4cd6076241601b52ce9348a730baf5ed29dd16d6ee40f548e38dff558026 +size 8150 diff --git a/parse/train/rygeHgSFDH/images/6c6475f52de8d079aa5d60435d14fd6c856ab91fa9c2ac9777906b4361b695da.jpg b/parse/train/rygeHgSFDH/images/6c6475f52de8d079aa5d60435d14fd6c856ab91fa9c2ac9777906b4361b695da.jpg new file mode 100644 index 0000000000000000000000000000000000000000..e4995ff5497b6a04483c9c0cf4a5389935a94a4f --- /dev/null +++ b/parse/train/rygeHgSFDH/images/6c6475f52de8d079aa5d60435d14fd6c856ab91fa9c2ac9777906b4361b695da.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:693278eaee37db2f3b54a3999f95b8d02364232e494d6129902facd5808f9bf4 +size 34353 diff --git a/parse/train/rygeHgSFDH/images/7138fef8ebc7f8a4dd915b1411baaad9be3499f7e2eb1ba57bcbc30b82891668.jpg b/parse/train/rygeHgSFDH/images/7138fef8ebc7f8a4dd915b1411baaad9be3499f7e2eb1ba57bcbc30b82891668.jpg new file mode 100644 index 0000000000000000000000000000000000000000..30972cdfc1ce3dd7698afbef9e5eab76a8327dd9 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/7138fef8ebc7f8a4dd915b1411baaad9be3499f7e2eb1ba57bcbc30b82891668.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:faad50b539fed0df776833de466995ab954f80c755b4ea4f9e69b521117e34d4 +size 43569 diff --git a/parse/train/rygeHgSFDH/images/72b26b353c3f1287f1d1caed7a024cdfc46621e5cd5569ecef37650dbe737847.jpg b/parse/train/rygeHgSFDH/images/72b26b353c3f1287f1d1caed7a024cdfc46621e5cd5569ecef37650dbe737847.jpg new file mode 100644 index 0000000000000000000000000000000000000000..a2de634594dab8d126e374cb1e5c6e46960fa44d --- /dev/null +++ b/parse/train/rygeHgSFDH/images/72b26b353c3f1287f1d1caed7a024cdfc46621e5cd5569ecef37650dbe737847.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b6fe78dca24f3f7ba638af71b54eedd1a3bfe510b5094b5386dc3b4f876f1870 +size 4815 diff --git a/parse/train/rygeHgSFDH/images/73ddbc5353b3c5a5c3190e91eaf13e05ab48b101920c0d6bbbc8e278d79e7941.jpg b/parse/train/rygeHgSFDH/images/73ddbc5353b3c5a5c3190e91eaf13e05ab48b101920c0d6bbbc8e278d79e7941.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4cd48fddffc22a694346953583c96f5ec24d1f1d --- /dev/null +++ b/parse/train/rygeHgSFDH/images/73ddbc5353b3c5a5c3190e91eaf13e05ab48b101920c0d6bbbc8e278d79e7941.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9a9de0d9e6380e9f014e3446a8ff6aa20bcf064eaa41ff6cd7ec3a215cd3e7d9 +size 3869 diff --git a/parse/train/rygeHgSFDH/images/78595ca87b2fab439957444744b0bb1195a1ee74e2fbb8a7e2bf5990807554f1.jpg b/parse/train/rygeHgSFDH/images/78595ca87b2fab439957444744b0bb1195a1ee74e2fbb8a7e2bf5990807554f1.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4a65192db1800582158d036429fb75ac7d2c5bca --- /dev/null +++ b/parse/train/rygeHgSFDH/images/78595ca87b2fab439957444744b0bb1195a1ee74e2fbb8a7e2bf5990807554f1.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a2e69e12f43ceb4f9ae6c6bff24dc03c3bf7c5c6da7c50201f8fb1f630f72618 +size 11993 diff --git a/parse/train/rygeHgSFDH/images/7ac2a5e2f50394e486a897a5e5323ce04e443faf7d1e4ab730d54487d8951186.jpg b/parse/train/rygeHgSFDH/images/7ac2a5e2f50394e486a897a5e5323ce04e443faf7d1e4ab730d54487d8951186.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c5784b49f2fbbcc920ede54433508240001c1d02 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/7ac2a5e2f50394e486a897a5e5323ce04e443faf7d1e4ab730d54487d8951186.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7473156d5b751dd2d8226be02662be7436389cd37a018130b50396020e652d6a +size 6584 diff --git a/parse/train/rygeHgSFDH/images/7bb49e5c5191d19a972b97df447824406a105767c46c4333c88ad13bb883be64.jpg b/parse/train/rygeHgSFDH/images/7bb49e5c5191d19a972b97df447824406a105767c46c4333c88ad13bb883be64.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0c0ce3728d90c423a48281a3ce1c95dd4a7d4b8a --- /dev/null +++ b/parse/train/rygeHgSFDH/images/7bb49e5c5191d19a972b97df447824406a105767c46c4333c88ad13bb883be64.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e0c11d35fa7f2c0d584fd92665a6c2fdcd3119a0ceb2e90b41c44bc7d13b9609 +size 44868 diff --git a/parse/train/rygeHgSFDH/images/7f0e1be6ea5bf5e1817b98cb95c51a28e5ef23fbfaf888dc658965d5f7077008.jpg b/parse/train/rygeHgSFDH/images/7f0e1be6ea5bf5e1817b98cb95c51a28e5ef23fbfaf888dc658965d5f7077008.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7f8d1e8d7fcc81fea60b9f63dec0e5c69bf9773b --- /dev/null +++ b/parse/train/rygeHgSFDH/images/7f0e1be6ea5bf5e1817b98cb95c51a28e5ef23fbfaf888dc658965d5f7077008.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:32bd3c5fea87188a42c3ae354764f65a7b80602165df406e8e5cfe87fb3583d9 +size 2755 diff --git a/parse/train/rygeHgSFDH/images/801e8ca2a9e93b48a1d3ad734a43b1c099c51d20a7008219cbb905eb44297bc3.jpg b/parse/train/rygeHgSFDH/images/801e8ca2a9e93b48a1d3ad734a43b1c099c51d20a7008219cbb905eb44297bc3.jpg new file mode 100644 index 0000000000000000000000000000000000000000..dd64e24d3e969b8494b5e77a288aa2ed1342a488 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/801e8ca2a9e93b48a1d3ad734a43b1c099c51d20a7008219cbb905eb44297bc3.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:974413ca8c9a8e3fb9c958fdaed116bc8b8ae5794dd91c9cd048f9038bfc159c +size 3532 diff --git a/parse/train/rygeHgSFDH/images/810439ba15e783f176985da25adab05b44701671430f6886188d94264a088270.jpg b/parse/train/rygeHgSFDH/images/810439ba15e783f176985da25adab05b44701671430f6886188d94264a088270.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b22ed4b5a97e2a97e4322b3e7d32e47c810d6766 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/810439ba15e783f176985da25adab05b44701671430f6886188d94264a088270.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:08b5a3920988e1168fe0eda550a555a66397258d68903fba9cf44edf938306f3 +size 31701 diff --git a/parse/train/rygeHgSFDH/images/8525b23d0cfd21ed79bf021c6c5874436bdef10c3682142666fc99804ebb660e.jpg b/parse/train/rygeHgSFDH/images/8525b23d0cfd21ed79bf021c6c5874436bdef10c3682142666fc99804ebb660e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0d438b39a7b883420fb7bed3c788ab199b8e3d9c --- /dev/null +++ b/parse/train/rygeHgSFDH/images/8525b23d0cfd21ed79bf021c6c5874436bdef10c3682142666fc99804ebb660e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2c64e0d2929387d9ab39b7e445495647b39aac4be4ffbe9f9ee00cd9da5de01b +size 201587 diff --git a/parse/train/rygeHgSFDH/images/9218efc7063acf811d8e360cb7cbbb24be9f86bfbe93cb1572343d7762372d8d.jpg b/parse/train/rygeHgSFDH/images/9218efc7063acf811d8e360cb7cbbb24be9f86bfbe93cb1572343d7762372d8d.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7eea0f7520cd209b8350d617fc56763138a67c4e --- /dev/null +++ b/parse/train/rygeHgSFDH/images/9218efc7063acf811d8e360cb7cbbb24be9f86bfbe93cb1572343d7762372d8d.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cb54acf40ee9a9bca3ae6d1d16cdff57836872aae5f0032d31a28e9315ff47c2 +size 3865 diff --git a/parse/train/rygeHgSFDH/images/93d8ea80b932c6686e2ba85a5d8f61ca958fbd6cddbd52fc5ce67b75438c988c.jpg b/parse/train/rygeHgSFDH/images/93d8ea80b932c6686e2ba85a5d8f61ca958fbd6cddbd52fc5ce67b75438c988c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..37f13e943cc0f83f90770bd005a3519ec07a3923 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/93d8ea80b932c6686e2ba85a5d8f61ca958fbd6cddbd52fc5ce67b75438c988c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:93333ddfa0078f997ff6a80d50df6129537a83211bc1347952507cd2c1c9c2b4 +size 2664 diff --git a/parse/train/rygeHgSFDH/images/a09187cf2a23fa418eec8054278e4c49b2fa5c076c92780168a46026e7d11f0f.jpg b/parse/train/rygeHgSFDH/images/a09187cf2a23fa418eec8054278e4c49b2fa5c076c92780168a46026e7d11f0f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..6a86a54e4abc49c6b5938a122c4b48472d3dccca --- /dev/null +++ b/parse/train/rygeHgSFDH/images/a09187cf2a23fa418eec8054278e4c49b2fa5c076c92780168a46026e7d11f0f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bd966e8947bbc87f150e10899f3b5b06bb576ce06ecd7cbdd82d82f60234dbb3 +size 44193 diff --git a/parse/train/rygeHgSFDH/images/a72c7099256731351e339b3198b52582a87acffc940acfbb9eab6bf15c77877c.jpg b/parse/train/rygeHgSFDH/images/a72c7099256731351e339b3198b52582a87acffc940acfbb9eab6bf15c77877c.jpg new file mode 100644 index 0000000000000000000000000000000000000000..08fb4c2d4b18054c989bd391a75cc4040de49fd6 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/a72c7099256731351e339b3198b52582a87acffc940acfbb9eab6bf15c77877c.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b723bdf32ddab095ed38a4152b06d4352d32e0a29bb3e2ccf437ee5d7ca6de33 +size 202702 diff --git a/parse/train/rygeHgSFDH/images/ab156c443c0161d8ff7fac80791c928c45a687f7d242f75008d320622df6a21f.jpg b/parse/train/rygeHgSFDH/images/ab156c443c0161d8ff7fac80791c928c45a687f7d242f75008d320622df6a21f.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d4e05c882c785e020ef8e6a840f4106f5822aeae --- /dev/null +++ b/parse/train/rygeHgSFDH/images/ab156c443c0161d8ff7fac80791c928c45a687f7d242f75008d320622df6a21f.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:be034f72647b871e18384705ce14419c91cfe39f9faad6cc62b8366f91f7bfa0 +size 3474 diff --git a/parse/train/rygeHgSFDH/images/b246b5f86658fd7ed9cab885863a705e14577be35108155e062ecf6f00c7de23.jpg b/parse/train/rygeHgSFDH/images/b246b5f86658fd7ed9cab885863a705e14577be35108155e062ecf6f00c7de23.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b639f51901a7187922743363813b6dbc593eef87 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/b246b5f86658fd7ed9cab885863a705e14577be35108155e062ecf6f00c7de23.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3cd8a061e2a9a579375b80d542f527e798379bb441b3c1d06ed5a94dd3733946 +size 6987 diff --git a/parse/train/rygeHgSFDH/images/b26ed75bd83dd631622d9bb24da7267a86f1a1a1fa526d15b8eaf0e724a8b990.jpg b/parse/train/rygeHgSFDH/images/b26ed75bd83dd631622d9bb24da7267a86f1a1a1fa526d15b8eaf0e724a8b990.jpg new file mode 100644 index 0000000000000000000000000000000000000000..b9205f130e1ec6dd9e895c66e1cd51e7ce67dd71 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/b26ed75bd83dd631622d9bb24da7267a86f1a1a1fa526d15b8eaf0e724a8b990.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0473ec758bd8e16c21e1fa5b039cf7acad496d310c8fd8d2964e9d3d383ea61f +size 45542 diff --git a/parse/train/rygeHgSFDH/images/b5853e51f49960896d834bb181f2ede36d5f7c6d1d1b4c8cf1acff0526b597dc.jpg b/parse/train/rygeHgSFDH/images/b5853e51f49960896d834bb181f2ede36d5f7c6d1d1b4c8cf1acff0526b597dc.jpg new file mode 100644 index 0000000000000000000000000000000000000000..4ed6046b4201a494cee6e0781764517202f196f9 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/b5853e51f49960896d834bb181f2ede36d5f7c6d1d1b4c8cf1acff0526b597dc.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:974e5af55eeab6f968c03cc82cd7b4c7d94c0860ca80ed0a18e9d7ac842b4939 +size 14920 diff --git a/parse/train/rygeHgSFDH/images/b6662812ca5d7373599ff8e25589a6ec8c69c33688561ba2f3e9019cdba8eed8.jpg b/parse/train/rygeHgSFDH/images/b6662812ca5d7373599ff8e25589a6ec8c69c33688561ba2f3e9019cdba8eed8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..647f10c2bd24d706e6d02493c908d6472cd26232 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/b6662812ca5d7373599ff8e25589a6ec8c69c33688561ba2f3e9019cdba8eed8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a87fd9b2f0f82410f7a42de2fc6beff7627742b7b6388a4feb45f2dd462589b7 +size 1956 diff --git a/parse/train/rygeHgSFDH/images/b819d3a62f6c08088f40e4bcbfd193b42e8e5695287c6a8646f71cd0b1a8d3c8.jpg b/parse/train/rygeHgSFDH/images/b819d3a62f6c08088f40e4bcbfd193b42e8e5695287c6a8646f71cd0b1a8d3c8.jpg new file mode 100644 index 0000000000000000000000000000000000000000..9ebdc40f4591093b06ec1452b2d1fa9eae7ca057 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/b819d3a62f6c08088f40e4bcbfd193b42e8e5695287c6a8646f71cd0b1a8d3c8.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:0d697f9201b202104165cb99b9766332a091bd99ccbcf41e44dcf963b28a11cd +size 6104 diff --git a/parse/train/rygeHgSFDH/images/c754051019f7ef94697f88e9bb5b97e21ea7f979150c38019863512738d13100.jpg b/parse/train/rygeHgSFDH/images/c754051019f7ef94697f88e9bb5b97e21ea7f979150c38019863512738d13100.jpg new file mode 100644 index 0000000000000000000000000000000000000000..0620058b0bff35d002eba3fe8ae62373772049bd --- /dev/null +++ b/parse/train/rygeHgSFDH/images/c754051019f7ef94697f88e9bb5b97e21ea7f979150c38019863512738d13100.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d014a26160ac74b2deaa7d16c930df0bc16db2f7db2dd87249a45daf81c5d3e0 +size 8972 diff --git a/parse/train/rygeHgSFDH/images/cc26a5aaa8c50d9aea1698cf7128e7afef5d8bddc3981016438a677ee1523970.jpg b/parse/train/rygeHgSFDH/images/cc26a5aaa8c50d9aea1698cf7128e7afef5d8bddc3981016438a677ee1523970.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d3c238f9e5bf4f04622133867872befb2891070a --- /dev/null +++ b/parse/train/rygeHgSFDH/images/cc26a5aaa8c50d9aea1698cf7128e7afef5d8bddc3981016438a677ee1523970.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe8e1e86b0b59bf8d0d35a7144e4a703b9d4295dba7de65a423b3fcfa94d2d97 +size 19926 diff --git a/parse/train/rygeHgSFDH/images/cc362557d3293b0173e5f188953749a2cd4fbb9269e52876ba764e4ecfbe4a42.jpg b/parse/train/rygeHgSFDH/images/cc362557d3293b0173e5f188953749a2cd4fbb9269e52876ba764e4ecfbe4a42.jpg new file mode 100644 index 0000000000000000000000000000000000000000..d940549ab89ef362f99590f1719821faf5016531 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/cc362557d3293b0173e5f188953749a2cd4fbb9269e52876ba764e4ecfbe4a42.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:27cfc36ff050d7da37040a9e047a86aecd50319d1750316b412a7a05cef0e840 +size 43532 diff --git a/parse/train/rygeHgSFDH/images/d1f44f347464b1f00eb4f5f545d2037bc7a88086df3e6b13e394c0378415eab7.jpg b/parse/train/rygeHgSFDH/images/d1f44f347464b1f00eb4f5f545d2037bc7a88086df3e6b13e394c0378415eab7.jpg new file mode 100644 index 0000000000000000000000000000000000000000..c9a2fab2bbaac4ac3e00c8e471c3719830458a9c --- /dev/null +++ b/parse/train/rygeHgSFDH/images/d1f44f347464b1f00eb4f5f545d2037bc7a88086df3e6b13e394c0378415eab7.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1fd59a62a6ac0d1cb26f540a87be9a2b46a856dbd042c908e8a133096df4c397 +size 20345 diff --git a/parse/train/rygeHgSFDH/images/d6e7186fab3b4f7ac9bfa55896f1a0003072f6cdcd2372ac0cfba861e6438bc2.jpg b/parse/train/rygeHgSFDH/images/d6e7186fab3b4f7ac9bfa55896f1a0003072f6cdcd2372ac0cfba861e6438bc2.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ada84ac2cbc46ab59212fab2d262f704d45718ba --- /dev/null +++ b/parse/train/rygeHgSFDH/images/d6e7186fab3b4f7ac9bfa55896f1a0003072f6cdcd2372ac0cfba861e6438bc2.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:90a1cb95a8d1997bcb53a526dd2527a3e3d3a1e94513b349787a66c6fcea22bf +size 10917 diff --git a/parse/train/rygeHgSFDH/images/e162750ec52ba32c0393e1e16f453e749319dad8b9ef4bbfb0a26a57bb90b572.jpg b/parse/train/rygeHgSFDH/images/e162750ec52ba32c0393e1e16f453e749319dad8b9ef4bbfb0a26a57bb90b572.jpg new file mode 100644 index 0000000000000000000000000000000000000000..086ce86c0132cccafa0613668b0a842cc6a64d94 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/e162750ec52ba32c0393e1e16f453e749319dad8b9ef4bbfb0a26a57bb90b572.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1a6a11b9069cad7593ab84a16fc5ed7f5e6ff5344f9bb7515d5196e735099b1d +size 2835 diff --git a/parse/train/rygeHgSFDH/images/e48b2f43552953adee83e87b42c8753c87a6c18b6018bb861bac720f6a13f78e.jpg b/parse/train/rygeHgSFDH/images/e48b2f43552953adee83e87b42c8753c87a6c18b6018bb861bac720f6a13f78e.jpg new file mode 100644 index 0000000000000000000000000000000000000000..2f92aa0891ee689f9808dedcbb0d1b8c43352767 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/e48b2f43552953adee83e87b42c8753c87a6c18b6018bb861bac720f6a13f78e.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7a214bde83e0df79aa74acbfbd283d0e5e6950df9e57e4c97371cd6adf540058 +size 43155 diff --git a/parse/train/rygeHgSFDH/images/eb512e024e2dfc5db3691ee44246e8dbd9fd47343dc3840347f28630a90f9821.jpg b/parse/train/rygeHgSFDH/images/eb512e024e2dfc5db3691ee44246e8dbd9fd47343dc3840347f28630a90f9821.jpg new file mode 100644 index 0000000000000000000000000000000000000000..ada8d0a413c212a491ad7d580e440d6faf765c1c --- /dev/null +++ b/parse/train/rygeHgSFDH/images/eb512e024e2dfc5db3691ee44246e8dbd9fd47343dc3840347f28630a90f9821.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:b933ef0b91503cadc8b2a4eb5a6c9b3becc76b0547acd3873c200a90b7e39978 +size 52638 diff --git a/parse/train/rygeHgSFDH/images/ef2220de76173c538c6872d1dd8ef4213ef9e1b9c7e7af17502a05fd438c8c92.jpg b/parse/train/rygeHgSFDH/images/ef2220de76173c538c6872d1dd8ef4213ef9e1b9c7e7af17502a05fd438c8c92.jpg new file mode 100644 index 0000000000000000000000000000000000000000..52ef60634a81cefd63376b92b2f8fd23d695bd5a --- /dev/null +++ b/parse/train/rygeHgSFDH/images/ef2220de76173c538c6872d1dd8ef4213ef9e1b9c7e7af17502a05fd438c8c92.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:578a99e705b667b9f557346538995d52cecaedbcd4a20aa2c856875b54473663 +size 188908 diff --git a/parse/train/rygeHgSFDH/images/feff4a56d468155dc0a7b91024f148ed67781f6bbf7d7cc70ca6d0af14c6b992.jpg b/parse/train/rygeHgSFDH/images/feff4a56d468155dc0a7b91024f148ed67781f6bbf7d7cc70ca6d0af14c6b992.jpg new file mode 100644 index 0000000000000000000000000000000000000000..7a12d1d8951bd37546113e8c0ae6e9918ddd7cc8 --- /dev/null +++ b/parse/train/rygeHgSFDH/images/feff4a56d468155dc0a7b91024f148ed67781f6bbf7d7cc70ca6d0af14c6b992.jpg @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:09b742d3b00bd4cb0ba86aa1dc4c7d46b7d2e781b77b73528d4ee3987ea4ea16 +size 10784