renderDpi
int64 200
200
| name
stringlengths 1
7
| page
int64 0
485
| figType
stringclasses 2
values | regionBoundary
dict | caption
stringlengths 6
3.74k
| imageText
listlengths 0
74.4k
| renderURL
stringlengths 171
178
| captionBoundary
dict |
|---|---|---|---|---|---|---|---|---|
200
|
1
| 0
|
Figure
|
{
"x1": 304.92,
"x2": 505.08,
"y1": 509.4,
"y2": 656.28
}
|
Figure 1: Propagation of cars on a road using an advection process.
|
[
"u",
"∼",
"N",
"(",
"0,",
"(2ν",
"κ2",
"+",
"∆",
")−ν)",
",",
"(1)",
"Gaussian",
"processes",
"(GPs)",
"[19]",
"can",
"learn",
"unknown",
"functions",
"that",
"allow",
"use",
"of",
"prior",
"information",
"about",
"their",
"properties",
"and",
"for",
"uncertainty",
"modeling.",
"Küper",
"and",
"Waldherr",
"[10]",
"propose",
"the",
"Gaussian",
"Process",
"Kalman",
"Filter",
"(GPKF)",
"method",
"to",
"simu-",
"late",
"spatiotemporal",
"models,",
"and",
"test",
"on",
"the",
"ad-",
"vection",
"equation.",
"Raissi",
"et",
"al.",
"[17]",
"train",
"GPs",
"on",
"data",
"to",
"learn",
"the",
"underlying",
"physics",
"of",
"non-linear",
"advection-diffusion",
"equations.",
"Additional",
"physics-",
"based",
"machine",
"learning",
"models",
"[2]",
"use",
"the",
"Matérn",
"covariance",
"function",
"given",
"below:"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00001_1762755909/figures/2201.00001-Figure1-1.png
|
{
"x1": 108.73400115966797,
"x2": 302.77392578125,
"y1": 643.8775634765625,
"y2": 660.7890014648438
}
|
200
|
2
| 2
|
Figure
|
{
"x1": 132.84,
"x2": 496.08,
"y1": 235.79999999999998,
"y2": 293.03999999999996
}
|
Figure 2: Balanced graphs on which Ladv corresponds to finite difference stencils of linear advection.
|
[
"(b)",
"second",
"order",
"central",
"scheme",
"−v/2∆x",
"v/2∆x",
"ui−1",
"ui",
"ui+1",
"v/2∆x",
"−v/2∆x",
"(a)",
"first",
"order",
"upwind",
"scheme",
"ui−1",
"ui",
"ui+1",
"v/∆x",
"v/∆x"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00001_1762755909/figures/2201.00001-Figure2-1.png
|
{
"x1": 108,
"x2": 505.744140625,
"y1": 307.04254150390625,
"y2": 314.53900146484375
}
|
200
|
5
| 7
|
Figure
|
{
"x1": 145.79999999999998,
"x2": 466.2,
"y1": 208.79999999999998,
"y2": 339.12
}
|
Figure 5: Prior results using DGAMGP obtained using various graphs, and plotting tools from [2].
|
[
"(a)",
"complete",
"graph",
"prior.",
"(b)",
"star",
"graph",
"prior."
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00001_1762755909/figures/2201.00001-Figure5-1.png
|
{
"x1": 109.90199279785156,
"x2": 502.09954833984375,
"y1": 352.8645324707031,
"y2": 358.86700439453125
}
|
200
|
6
| 7
|
Figure
|
{
"x1": 109.8,
"x2": 502.2,
"y1": 479.88,
"y2": 625.3199999999999
}
|
Figure 6: Upwinding solution with RK5 to the linear advection equation over time and corresponding convergence study.
|
[
"(a)",
"Solution",
"of",
"the",
"linear",
"advection",
"equation",
"using",
"(2)",
"(b)",
"Convergence",
"study",
"in",
"a",
"log-log",
"plot"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00001_1762755909/figures/2201.00001-Figure6-1.png
|
{
"x1": 108,
"x2": 504.0015869140625,
"y1": 637.7215576171875,
"y2": 654.6329956054688
}
|
200
|
1
| 3
|
Table
|
{
"x1": 144.72,
"x2": 465.12,
"y1": 451.8,
"y2": 545.04
}
|
Table 1: Comparison of l2 test error on synthetic directed graphs with n nodes and the learned hyperparameters.
|
[
"Model",
"Graph",
"type",
"n",
"=",
"280",
"n",
"=",
"325",
"n",
"=",
"400",
"ν",
"κ",
"σ",
"Advection",
"Upwind",
"0.52",
"0.45",
"0.0005",
"0.65",
"8.09",
"7.75Consensus",
"0.51",
"0.44",
"0.0005",
"0.65",
"8.29",
"7.77",
"Advection",
"Central",
"1.31",
"0.85",
"8.41e-05",
"0.67",
"9.00",
"8.03Consensus",
"0.97",
"0.8",
"8.02e-05",
"0.67",
"9.45",
"8.11",
"Advection",
"Intersection",
"0.96",
"0.45",
"0.0005",
"0.65",
"8.19",
"7.75Consensus",
"0.52",
"0.46",
"0.0005",
"0.64",
"8.28",
"7.77",
"Advection",
"Loop",
"0.47",
"0.41",
"0.00045",
"0.65",
"8.49",
"7.76Consensus",
"0.47",
"0.41",
"0.00045",
"0.65",
"8.49",
"7.76"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00001_1762755909/figures/2201.00001-Table1-1.png
|
{
"x1": 107.69100189208984,
"x2": 503.99896240234375,
"y1": 553.8075561523438,
"y2": 570.718994140625
}
|
200
|
3
| 3
|
Figure
|
{
"x1": 110.88,
"x2": 504,
"y1": 599.76,
"y2": 698.04
}
|
Figure 3: Graphs representing two lanes merging into one (left) and a loop (right).
|
[
"(b)",
"loop",
"graph",
"v/∆x",
"u1",
"ui−1",
"ui",
"un",
"v/∆x",
"v/∆x",
"v/∆x",
"(a)",
"intersection",
"graph",
"2v/∆x",
"v/∆x",
"v/∆x",
"ui+1",
"v/∆x",
"v/∆x",
"ui−4",
"ui−3",
"ui−2",
"ui−1",
"ui"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00001_1762755909/figures/2201.00001-Figure3-1.png
|
{
"x1": 142.13101196289062,
"x2": 469.8692321777344,
"y1": 711.840576171875,
"y2": 717.843017578125
}
|
200
|
4
| 4
|
Figure
|
{
"x1": 108,
"x2": 504,
"y1": 173.88,
"y2": 452.15999999999997
}
|
Figure 4: Traffic speed interpolation over a graph of San Jose highways using our DGAMGP method with ν = 0.35, κ = 1002.8, σ = 1.14 and plotting tools from [2].
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00001_1762755909/figures/2201.00001-Figure4-1.png
|
{
"x1": 107.64099884033203,
"x2": 503.9951477050781,
"y1": 463.2015380859375,
"y2": 480.1130065917969
}
|
200
|
1
| 5
|
Figure
|
{
"x1": 54.72,
"x2": 296.28,
"y1": 136.79999999999998,
"y2": 568.0799999999999
}
|
Fig. 1. SORA-based RRBTO flowchart [57].
|
[
"Calculate",
"robust",
"objective",
"and",
"its",
"derivarives",
"using",
"the",
"SCM",
"MPP",
"Update",
"Young’s",
"modulus",
"using",
"Yes",
"End",
"No",
"Yes",
"SRSM",
"Inverse",
"Reliability",
"Analysis",
"Converge",
"Find",
"MPP",
"Construct",
"limit",
"state",
"function",
"Finite",
"Element",
"Analysis",
"SIMP",
"DTO",
"ConvergeNo",
"Update",
"densities",
"Sensitivity",
"and",
"filtering",
"Finite",
"Element",
"Analysis",
"Initialization:",
"-",
"Initial",
"values",
"-",
"KL",
"expansion",
"-",
"Collocation",
"and",
"sample",
"points",
"Start"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00004_1762755917/figures/2201.00004-Figure1-1.png
|
{
"x1": 95.70899963378906,
"x2": 254.60092163085938,
"y1": 581.8135375976562,
"y2": 586.7470092773438
}
|
200
|
4
| 10
|
Table
|
{
"x1": 63.72,
"x2": 539.28,
"y1": 58.68,
"y2": 362.15999999999997
}
|
Table 4. The L-shaped beam: RRBTO results
|
[
"β",
"=",
"3",
"β",
"=",
"1",
"(ε,1−",
"ε)",
"(1,0)",
"(0.9,0.1)",
"(0.8,0.2)"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00004_1762755917/figures/2201.00004-Table4-1.png
|
{
"x1": 218.91700744628906,
"x2": 384.1132507324219,
"y1": 48.38053512573242,
"y2": 53.31402587890625
}
|
200
|
5
| 10
|
Table
|
{
"x1": 63.72,
"x2": 548.28,
"y1": 397.8,
"y2": 756
}
|
Table 5. The L-shaped beam: RRBTO results
|
[
"11",
"Copyright",
"©",
"by",
"ASME",
"β",
"=",
"3",
"β",
"=",
"1",
"(ε,1−",
"ε)",
"(0.5,0.5)",
"(0.2,0.8)",
"(0,1)"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00004_1762755917/figures/2201.00004-Table5-1.png
|
{
"x1": 218.91700744628906,
"x2": 384.1132507324219,
"y1": 387.8055114746094,
"y2": 392.739013671875
}
|
200
|
3
| 9
|
Table
|
{
"x1": 75.96,
"x2": 527.04,
"y1": 47.519999999999996,
"y2": 504
}
|
Table 3. The cantilever beam: Numerical results
|
[
"(0,1)",
"206.2320",
"0.8559",
"-278.9110",
"0.5527",
"0.000000",
"-278.9110",
"0.5527",
"(0.2,0.8)",
"177.8977",
"0.8585",
"-240.8960",
"0.5570",
"0.000000",
"-240.8960",
"0.5570",
"(0.5,0.5)",
"165.0556",
"0.8938",
"-223.2560",
"0.5855",
"0.000000",
"-223.2560",
"0.5855",
"(0.8,0.2)",
"164.0458",
"0.9015",
"-221.7520",
"0.5907",
"0.001320",
"-221.7520",
"0.5907",
"(0.9,0.1)",
"164.0019",
"0.9078",
"-221.7630",
"0.5944",
"0.001340",
"-221.7630",
"0.5944",
"(1,0)",
"163.7490",
"0.9094",
"-221.7690",
"0.5962",
"0.001340",
"-221.7690",
"0.5962",
"3",
"0.001349",
"(0,1)",
"206.2320",
"0.8559",
"-278.9110",
"0.5527",
"0.00000",
"-278.9110",
"0.5527",
"(0.2,0.8)",
"178.0176",
"0.8586",
"-241.0670",
"0.5570",
"0.00000",
"-241.0670",
"0.5570",
"(0.5,0.5)",
"164.7008",
"0.8953",
"-222.9560",
"0.5866",
"0.00000",
"-222.9560",
"0.5866",
"(0.8,0.2)",
"163.7847",
"0.9118",
"-221.1890",
"0.5967",
"0.02266",
"-221.1890",
"0.5967",
"(0.9,0.1)",
"163.6040",
"0.9106",
"-221.1880",
"0.5961",
"0.02250",
"-221.1880",
"0.5961",
"(1,0)",
"163.4230",
"0.9150",
"-221.1960",
"0.5995",
"0.02252",
"-221.1960",
"0.5995",
"2",
"0.02275",
"(0,1)",
"206.2317",
"0.8559",
"-278.9100",
"0.5527",
"0.00000",
"-278.9100",
"0.5527",
"(0.2,0.8)",
"177.7632",
"0.8588",
"-240.6990",
"0.5572",
"0.00000",
"-240.6990",
"0.5572",
"(0.5,0.5)",
"166.5160",
"0.9032",
"-225.0110",
"0.5906",
"0.00000",
"-225.0110",
"0.5906",
"(0.8,0.2)",
"163.2204",
"0.8953",
"-220.5980",
"0.5853",
"0.15326",
"-220.5980",
"0.5853",
"(0.9,0.1)",
"162.9992",
"0.9188",
"-220.6070",
"0.6017",
"0.15642",
"-220.6070",
"0.6017",
"(1,0)",
"162.9505",
"0.9263",
"-220.6120",
"0.6071",
"0.15674",
"-220.6120",
"0.6071",
"1",
"0.15865",
"β",
"Expected",
"Pf",
"(ε,1−",
"ε)",
"µ[C]",
"σ[C]",
"µB",
"σB",
"Pf",
"µB",
"σB",
"MCS",
"SRSM"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00004_1762755917/figures/2201.00004-Table3-1.png
|
{
"x1": 214.49400329589844,
"x2": 388.5369567871094,
"y1": 38.76853561401367,
"y2": 43.7020263671875
}
|
200
|
6
| 11
|
Table
|
{
"x1": 60.12,
"x2": 548.28,
"y1": 48.96,
"y2": 756
}
|
Table 6. The L-shaped beam: Numerical results
|
[
"12",
"Copyright",
"©",
"by",
"ASME",
"rapid",
"in",
"the",
"remaining",
"cases",
"(i.e.,",
"the",
"sixth",
"column",
"of",
"Tables",
"3",
"and",
"6).",
"The",
"bigger",
"the",
"displacement",
"of",
"point",
"B",
"becomes,",
"the",
"smaller",
"the",
"MCS-based",
"Pf",
"is.",
"It",
"is",
"ob-",
"vious",
"that",
"under",
"a",
"specific",
"set",
"of",
"inputs",
"(i.e.,",
"loading",
"and",
"boundary",
"conditions,",
"material",
"property)",
"mechani-",
"cal",
"capabilities",
"of",
"a",
"structure",
"(i.e.,",
"stress,",
"strain,",
"dis-",
"placement),",
"even",
"if",
"heavily",
"optimized,",
"are",
"always",
"finite.",
"Thus,",
"the",
"MCS-based",
"Pf",
"approaches",
"0",
"when",
"the",
"dis-",
"placement",
"of",
"point",
"B",
"continues",
"to",
"increase.",
"4.",
"The",
"Smolyak-type",
"sparse",
"grid",
"and",
"the",
"SRSM",
"are",
"both",
"very",
"good",
"methods",
"for",
"their",
"respective",
"approximating",
"targets.",
"As",
"shown",
"in",
"Tables",
"3",
"and",
"6,",
"approximately",
"six",
"significant",
"digits",
"are",
"required",
"to",
"see",
"the",
"differences",
"between",
"the",
"MCS-based",
"and",
"the",
"SRSM",
"results.",
"The",
"cumulative",
"distribution",
"functions",
"of",
"point",
"B",
"displace-",
"ment",
"from",
"the",
"two",
"methods",
"are",
"also",
"almost",
"identi-",
"cal,",
"so",
"they",
"are",
"not",
"included",
"in",
"the",
"paper.",
"This",
"also",
"makes",
"us",
"confident",
"in",
"choosing",
"only",
"two",
"terms",
"in",
"the",
"KL",
"expansion−adding",
"more",
"terms",
"would",
"only",
"increase",
"computational",
"cost",
"without",
"clear",
"benefits.",
"The",
"same",
"ob-",
"servation",
"is",
"applied",
"to",
"the",
"Smolyak-type",
"sparse",
"grid",
"and",
"MCS-based",
"results",
"of",
"the",
"mean",
"and",
"standard",
"deviation",
"of",
"the",
"compliance.",
"Because",
"of",
"such",
"agreement,",
"we",
"de-",
"cide",
"to",
"use",
"only",
"one",
"level",
"of",
"the",
"sparse",
"grid.",
"Multiple",
"levels",
"were",
"tested",
"in",
"[17,",
"27].",
"lutions",
"are",
"away",
"from",
"the",
"constraint",
"boundary",
"but",
"still",
"inside",
"the",
"feasible",
"region:",
"the",
"decreasing",
"rate",
"acceler-",
"ates",
"and",
"the",
"MCS-based",
"Pf",
"eventually",
"becomes",
"0",
"(i.e.,",
"ε≥",
"0.5",
"in",
"Table",
"3,",
"and",
"ε≥",
"0.8",
"in",
"Table",
"6).",
"3.",
"From",
"our",
"understanding",
"of",
"robust",
"optimization,",
"increas-",
"ing",
"the",
"weight",
"on",
"standard",
"deviation",
"in",
"a",
"minimization",
"problem",
"will",
"decrease",
"its",
"value",
"and",
"have",
"the",
"opposite",
"ef-",
"fect",
"on",
"the",
"mean,",
"which",
"is",
"actually",
"the",
"case",
"as",
"observed",
"in",
"Tables",
"3",
"and",
"6.",
"However,",
"there",
"are",
"a",
"couple",
"of",
"out-",
"liers",
"to",
"this",
"trend",
"(i.e.,",
"(ε,β)",
"=",
"(0.8,2)",
"in",
"Table",
"3,",
"and",
"ε",
"=",
"0.9",
"in",
"Table",
"6).",
"An",
"explanation",
"can",
"be",
"made",
"by",
"in-",
"vestigating",
"where",
"the",
"solution",
"converges",
"in",
"the",
"design",
"space.",
"When",
"the",
"solutions",
"are",
"away",
"from",
"the",
"constraint",
"boundary,",
"the",
"robust",
"objective",
"is",
"more",
"dominant",
"in",
"the",
"solution:",
"a",
"marked",
"increase",
"and",
"decrease",
"of",
"the",
"mean",
"and",
"standard",
"deviation,",
"respectively,",
"is",
"observed.",
"In",
"so-",
"lutions",
"at",
"the",
"constraint",
"limit,",
"the",
"robust",
"objective",
"has",
"less",
"in",
"the",
"solution.",
"This",
"trend",
"explains",
"the",
"decreasing",
"tendency",
"of",
"the",
"MCS-",
"based",
"Pf",
".",
"Increasing",
"the",
"weight",
"on",
"standard",
"deviation",
"leads",
"to",
"decreased",
"influence",
"of",
"the",
"mean",
"compliance,",
"whose",
"major",
"proportion",
"is",
"contributed",
"by",
"the",
"displace-",
"ment",
"of",
"point",
"B",
"(Fig.",
"2",
"and",
"3)",
"which",
"is",
"constrained",
"prob-",
"abilistically",
"in",
"(22)",
"to",
"be",
"larger",
"than",
"the",
"minimum",
"al-",
"lowable",
"value",
"u0.",
"This",
"results",
"in",
"increasing",
"the",
"mean",
"of",
"point",
"B",
"displacement,",
"whose",
"changing",
"rate",
"is",
"slow",
"for",
"solutions",
"close",
"to",
"constraint",
"boundary",
"and",
"much",
"more",
"(0,1)",
"133.9731",
"0.4755",
"-184.5450",
"0.3229",
"0.00000",
"-184.5450",
"0.3229",
"(0.2,0.8)",
"108.7883",
"0.4916",
"-149.3880",
"0.3331",
"0.00000",
"-149.3880",
"0.3331",
"(0.5,0.5)",
"99.6087",
"0.5049",
"-135.9450",
"0.3399",
"0.00000",
"-135.9450",
"0.3399",
"(0.8,0.2)",
"96.8852",
"0.5240",
"-131.0560",
"0.3497",
"0.00106",
"-131.0560",
"0.3497",
"(0.9,0.1)",
"96.7519",
"0.5298",
"-131.0510",
"0.3545",
"0.00126",
"-131.0510",
"0.3545",
"(1,0)",
"96.8866",
"0.5300",
"-131.0510",
"0.3544",
"0.00130",
"-131.0510",
"0.3544",
"3",
"0.001349",
"(0,1)",
"133.3052",
"0.4762",
"-183.8160",
"0.3238",
"0.00000",
"-183.8160",
"0.3238",
"(0.2,0.8)",
"108.7352",
"0.4917",
"-149.3090",
"0.3332",
"0.00000",
"-149.3090",
"0.3332",
"(0.5,0.5)",
"99.5821",
"0.5045",
"-135.7740",
"0.3396",
"0.00000",
"-135.7740",
"0.3396",
"(0.8,0.2)",
"96.8678",
"0.5241",
"-131.0290",
"0.3498",
"0.00144",
"-131.0290",
"0.3498",
"(0.9,0.1)",
"96.3536",
"0.5321",
"-130.3560",
"0.3553",
"0.15790",
"-130.3560",
"0.3553",
"(1,0)",
"96.4893",
"0.5352",
"-130.3580",
"0.3577",
"0.15808",
"-130.3580",
"0.3577",
"1",
"0.15865",
"β",
"Expected",
"Pf",
"(ε,1−",
"ε)",
"µ[C]",
"σ[C]",
"µB",
"σB",
"Pf",
"µB",
"σB",
"MCS",
"SRSM"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00004_1762755917/figures/2201.00004-Table6-1.png
|
{
"x1": 214.71299743652344,
"x2": 388.3175964355469,
"y1": 38.76853561401367,
"y2": 43.7020263671875
}
|
200
|
1
| 10
|
Table
|
{
"x1": 158.76,
"x2": 456.12,
"y1": 263.88,
"y2": 401.03999999999996
}
|
Table 1. Overall performance. For average travel time, the smaller the better.
|
[
"M-QL",
"268.87",
"240.02",
"238.51",
"284.32",
"325.44",
"1197.59",
"1551.46",
"AttentionLight",
"254.82",
"239.68",
"236.62",
"283.64",
"316.38",
"1013.78",
"1401.32",
"FRAP",
"299.56",
"268.57",
"269.20",
"308.73",
"355.80",
"1192.23",
"1470.51",
"MPLight",
"297.68",
"274.32",
"268.00",
"313.16",
"355.35",
"1321.40",
"1642.05",
"PRGLight",
"291.27",
"257.52",
"261.74",
"301.06",
"369.98",
"1283.37",
"1472.73",
"CoLight",
"271.17",
"251.22",
"248.87",
"300.07",
"339.76",
"1065.64",
"1367.54",
"FixedTime",
"429.27",
"370.34",
"384.89",
"497.87",
"408.31",
"1507.12",
"1733.30",
"MP",
"274.99",
"246.41",
"244.63",
"289.54",
"349.85",
"1179.55",
"1536.17",
"1",
"2",
"3",
"1",
"2",
"1",
"2",
"Method",
"JiNan",
"HangZhou",
"New",
"York"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00006_1762755935/figures/2201.00006-Table1-1.png
|
{
"x1": 149.0396728515625,
"x2": 466.364501953125,
"y1": 248.32069396972656,
"y2": 252.36004638671875
}
|
200
|
5
| 13
|
Figure
|
{
"x1": 214.92,
"x2": 398.15999999999997,
"y1": 117.72,
"y2": 196.2
}
|
Fig. 5. The average travel time of transfer divided by the average travel time of direct training. The error bars represent the 95% confidence interval for the average travel time ratio.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00006_1762755935/figures/2201.00006-Figure5-1.png
|
{
"x1": 134.7598876953125,
"x2": 480.6470947265625,
"y1": 213.52064514160156,
"y2": 239.5201416015625
}
|
200
|
4
| 12
|
Figure
|
{
"x1": 160.92,
"x2": 451.08,
"y1": 335.88,
"y2": 417.24
}
|
Fig. 4. Model performance under different action duration.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00006_1762755935/figures/2201.00006-Figure4-1.png
|
{
"x1": 188.27999877929688,
"x2": 427.1462707519531,
"y1": 433.96063232421875,
"y2": 438
}
|
200
|
3
| 12
|
Figure
|
{
"x1": 157.68,
"x2": 457.2,
"y1": 114.83999999999999,
"y2": 200.16
}
|
Fig. 3. Model performance under different rewards w.r.t average travel time, the smaller the better.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00006_1762755935/figures/2201.00006-Figure3-1.png
|
{
"x1": 134.75999450683594,
"x2": 480.6089172363281,
"y1": 216.40065002441406,
"y2": 231.480224609375
}
|
200
|
1
| 3
|
Figure
|
{
"x1": 204.84,
"x2": 409.32,
"y1": 486,
"y2": 596.16
}
|
Fig. 1. The illustration of an intersection. In case (a), phase #2 is activated.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00006_1762755935/figures/2201.00006-Figure1-1.png
|
{
"x1": 152.52000427246094,
"x2": 462.7645568847656,
"y1": 613.6006469726562,
"y2": 617.6400146484375
}
|
200
|
2
| 11
|
Figure
|
{
"x1": 168.84,
"x2": 446.03999999999996,
"y1": 115.92,
"y2": 204.12
}
|
Fig. 2. Model performance comparison.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00006_1762755935/figures/2201.00006-Figure2-1.png
|
{
"x1": 227.63999938964844,
"x2": 387.5296936035156,
"y1": 220.00062561035156,
"y2": 224.03997802734375
}
|
200
|
1
| 0
|
Figure
|
{
"x1": 344.88,
"x2": 529.1999999999999,
"y1": 204.84,
"y2": 273.24
}
|
Fig. 1. Comparison of the previous average direction (green line) and our proposed confidence-aware direction (red line).
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00007_1762755947/figures/2201.00007-Figure1-1.png
|
{
"x1": 315.2130126953125,
"x2": 558.987548828125,
"y1": 287.61932373046875,
"y2": 306.05499267578125
}
|
200
|
6
| 5
|
Table
|
{
"x1": 334.8,
"x2": 539.28,
"y1": 335.52,
"y2": 493.2
}
|
Table 6. Top-1 test accuracy of CA-MKD compared to single-teacher knowledge distillation methods.
|
[
"CA-MKD",
"65.19±0.23",
"49.55±0.12",
"VID",
"64.45±0.23",
"47.76±0.08",
"CRD",
"64.61±0.17",
"48.11±0.07",
"FitNet",
"62.45±0.61",
"47.24±0.28",
"AT",
"63.48±0.60",
"45.73±0.05",
"Student",
"ShuffleNetV2x0.5",
"VGG859.36±0.73",
"44.40±0.15",
"KD",
"63.90±0.08",
"47.42±0.07",
"Teacher",
"ResNet34",
"ResNet32x465.97",
"53.45",
"Dataset",
"Dogs",
"Tinyimagenet"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00007_1762755947/figures/2201.00007-Table6-1.png
|
{
"x1": 315.2130126953125,
"x2": 558.9940185546875,
"y1": 314.07232666015625,
"y2": 332.50799560546875
}
|
200
|
5
| 5
|
Figure
|
{
"x1": 54.72,
"x2": 297,
"y1": 462.96,
"y2": 567
}
|
Fig. 5. The comparison of teacher-1 and teacher-2 classifiers.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00007_1762755947/figures/2201.00007-Figure5-1.png
|
{
"x1": 54.42499923706055,
"x2": 298.20343017578125,
"y1": 582.2213745117188,
"y2": 588.7020263671875
}
|
200
|
5
| 5
|
Table
|
{
"x1": 328.68,
"x2": 545.04,
"y1": 142.92,
"y2": 289.08
}
|
Table 5. Top-1 test accuracy of CA-MKD compared to multiple-teacher knowledge distillation methods.
|
[
"CA-MKD",
"65.19±0.23",
"49.55±0.12",
"Student",
"ShuffleNetV2x0.5",
"VGG859.36±0.73",
"44.40±0.15",
"AVER",
"64.49±0.16",
"47.82±0.15",
"FitNet-MKD",
"64.11±0.80",
"47.82±0.05",
"EBKD",
"64.32±0.23",
"47.20±0.10",
"AEKD",
"64.19±0.34",
"47.62±0.38",
"Teacher",
"ResNet34",
"ResNet32x464.76±1.06",
"53.38±0.11",
"Dataset",
"Dogs",
"Tinyimagenet"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00007_1762755947/figures/2201.00007-Table5-1.png
|
{
"x1": 315.2130126953125,
"x2": 558.9940185546875,
"y1": 121.31034851074219,
"y2": 139.74603271484375
}
|
200
|
2
| 1
|
Figure
|
{
"x1": 322.92,
"x2": 551.16,
"y1": 70.92,
"y2": 220.32
}
|
Fig. 2. An overview of our CA-MKD. The weight calculation of teacher predictions and intermediate teacher features are depicted as the red lines and green lines, respectively.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00007_1762755947/figures/2201.00007-Figure2-1.png
|
{
"x1": 315.2130126953125,
"x2": 558.991455078125,
"y1": 234.74534606933594,
"y2": 265.13702392578125
}
|
200
|
1
| 2
|
Table
|
{
"x1": 60.839999999999996,
"x2": 554.04,
"y1": 91.8,
"y2": 218.16
}
|
Table 1. Top-1 test accuracy of MKD methods by distilling the knowledge on multiple teachers with the same architectures.
|
[
"Student",
"ShuffleNetV1",
"MobileNetV2",
"VGG8",
"MobileNetV2",
"ResNet8x4",
"ShuffleNetV2",
"VGG871.70±0.43",
"65.64±0.19",
"70.74±0.40",
"65.64±0.19",
"72.79±0.14",
"72.94±0.24",
"70.74±0.40",
"AVER",
"[8]",
"76.30±0.25",
"70.21±0.10",
"74.07±0.23",
"68.91±0.35",
"74.99±0.24",
"75.87±0.19",
"73.26±0.39",
"FitNet-MKD",
"[5]",
"76.59±0.17",
"70.69±0.56",
"73.97±0.22",
"68.48±0.07",
"74.86±0.21",
"76.09±0.13",
"73.27±0.19",
"EBKD",
"[12]",
"76.61±0.14",
"70.91±0.22",
"74.10±0.27",
"68.24±0.82",
"75.59±0.15",
"76.41±0.12",
"73.60±0.22",
"AEKD",
"[11]",
"76.34±0.24",
"70.47±0.15",
"73.78±0.03",
"68.39±0.50",
"74.75±0.28",
"75.95±0.20",
"73.11±0.27",
"CA-MKD",
"77.94±0.31",
"71.38±0.02",
"74.30±0.16",
"69.41±0.20",
"75.90±0.13",
"77.41±0.14",
"75.26±0.32",
"Teacher",
"WRN40-2",
"ResNet56",
"VGG13",
"VGG13",
"ResNet32x4",
"ResNet32x4",
"ResNet32x476.62±0.26",
"73.28±0.30",
"75.17±0.18",
"75.17±0.18",
"79.31±0.14",
"79.31±0.14",
"79.31±0.14",
"Ensemble",
"79.62",
"76.00",
"77.07",
"77.07",
"81.16",
"81.16",
"81.16"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00007_1762755947/figures/2201.00007-Table1-1.png
|
{
"x1": 59.263999938964844,
"x2": 554.1561889648438,
"y1": 82.34635925292969,
"y2": 88.8270263671875
}
|
200
|
2
| 2
|
Table
|
{
"x1": 58.68,
"x2": 296.28,
"y1": 270,
"y2": 404.28
}
|
Table 2. Top-1 test accuracy of CA-MKD compared to single-teacher knowledge distillation methods.
|
[
"FitNet",
"[5]",
"76.22±0.21",
"72.55±0.66",
"69.02±0.28",
"AT",
"[6]",
"76.44±0.38",
"72.16±0.12",
"69.79±0.26",
"VID",
"[14]",
"76.32±0.08",
"73.09±0.29",
"69.45±0.17",
"CRD",
"[15]",
"76.58±0.23",
"73.57±0.25",
"71.15±0.44",
"CA-MKD",
"77.94±0.31",
"75.26±0.13",
"71.38±0.02",
"Student",
"ShuffleNetV1",
"VGG8",
"MobileNetV271.70±0.19",
"70.74±0.40",
"65.64±0.43",
"KD",
"[4]",
"75.77±0.14",
"72.90±0.34",
"69.96±0.14",
"Teacher",
"WRN40-2",
"ResNet32x4",
"ResNet5676.62±0.26",
"79.31±0.14",
"73.28±0.30"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00007_1762755947/figures/2201.00007-Table2-1.png
|
{
"x1": 54.42499923706055,
"x2": 298.20684814453125,
"y1": 248.5183563232422,
"y2": 266.95404052734375
}
|
200
|
4
| 3
|
Table
|
{
"x1": 319.68,
"x2": 557.28,
"y1": 292.68,
"y2": 328.32
}
|
Table 4. Ablation study with VGG13 & MobileNetV2.
|
[
"avg",
"weight",
"w/o",
"Linter",
"w/o",
"wkinter",
"CA-MKD",
"67.74±0.87",
"68.11±0.02",
"68.82±0.63",
"69.41±0.20"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00007_1762755947/figures/2201.00007-Table4-1.png
|
{
"x1": 327.0559997558594,
"x2": 547.1497802734375,
"y1": 283.0913391113281,
"y2": 289.572021484375
}
|
200
|
11
| 10
|
Figure
|
{
"x1": 54.36,
"x2": 299.88,
"y1": 43.92,
"y2": 143.28
}
|
Fig. 11. Impact of layer number on the NYC-Bike dataset.
|
[
"(b)",
"MAPE.",
"inflow",
"outflow",
"α",
"P",
"E",
"M",
"A",
"1",
"2",
"3",
"4",
"5",
"6",
"21%",
"20%",
"19%",
"18%",
"(a)",
"RMSE.",
"outflow",
"inflow",
"α",
"S",
"E",
"R",
"M",
"1",
"2",
"3",
"4",
"5",
"6",
"8.5",
"8",
"7.5",
"7"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00008_1762755955/figures/2201.00008-Figure11-1.png
|
{
"x1": 48,
"x2": 250.5599365234375,
"y1": 158.22897338867188,
"y2": 163.18096923828125
}
|
200
|
12
| 10
|
Figure
|
{
"x1": 54.36,
"x2": 299.88,
"y1": 178.92,
"y2": 278.64
}
|
Fig. 12. Impact of head number on the NYC-Taxi dataset.
|
[
"(b)",
"MAPE.",
"outflow",
"inflow",
"M",
"P",
"E",
"M",
"A",
"1",
"2",
"3",
"4",
"5",
"6",
"7",
"8",
"19%",
"18%",
"17%",
"16%",
"15%",
"14%",
"(a)",
"RMSE.",
"outflow",
"inflow",
"M",
"S",
"E",
"R",
"M",
"1",
"2",
"3",
"4",
"5",
"6",
"7",
"8",
"25",
"23",
"21",
"19",
"17"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00008_1762755955/figures/2201.00008-Figure12-1.png
|
{
"x1": 48,
"x2": 250.16793823242188,
"y1": 293.2610168457031,
"y2": 298.2130126953125
}
|
200
|
13
| 10
|
Figure
|
{
"x1": 318.24,
"x2": 563.76,
"y1": 43.92,
"y2": 143.28
}
|
Fig. 13. Impact of head number on the NYC-Bike dataset.
|
[
"(b)",
"MAPE.",
"outflow",
"inflow",
"M",
"P",
"E",
"M",
"A",
"1",
"2",
"3",
"4",
"5",
"6",
"7",
"8",
"22%",
"21%",
"20%",
"19%",
"18%",
"(a)",
"RMSE.",
"outflow",
"inflow",
"M",
"S",
"E",
"R",
"M",
"1",
"2",
"3",
"4",
"5",
"6",
"7",
"8",
"9",
"8.5",
"8",
"7.5"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00008_1762755955/figures/2201.00008-Figure13-1.png
|
{
"x1": 312,
"x2": 515.4159545898438,
"y1": 158.22897338867188,
"y2": 163.18096923828125
}
|
200
|
1
| 1
|
Figure
|
{
"x1": 78.84,
"x2": 269.28,
"y1": 42.839999999999996,
"y2": 172.07999999999998
}
|
Fig. 1. Visualization of attention scores between a target region (6,4) and other regions. The color of a cell (xi, yi) indicates the dependency of (6, 4) on (xi, yi), where a darker color indicates stronger dependency.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00008_1762755955/figures/2201.00008-Figure1-1.png
|
{
"x1": 48,
"x2": 300.001953125,
"y1": 186.32400512695312,
"y2": 210.49102783203125
}
|
200
|
8
| 9
|
Figure
|
{
"x1": 54.36,
"x2": 299.88,
"y1": 178.2,
"y2": 277.92
}
|
Fig. 8. Impact of surrounding observations on the NYC-Taxi dataset.
|
[
"(b)",
"MAPE.",
"inflow",
"outflow",
"w",
"P",
"E",
"M",
"A",
"1",
"2",
"3",
"4",
"5",
"6",
"7",
"8",
"17%",
"16%",
"15%",
"14%",
"(a)",
"RMSE.",
"outflow",
"inflow",
"w",
"S",
"E",
"R",
"M",
"1",
"2",
"3",
"4",
"5",
"6",
"7",
"8",
"25",
"23",
"21",
"19",
"17"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00008_1762755955/figures/2201.00008-Figure8-1.png
|
{
"x1": 48,
"x2": 288.51995849609375,
"y1": 292.5310363769531,
"y2": 297.4830322265625
}
|
200
|
10
| 9
|
Figure
|
{
"x1": 318.24,
"x2": 563.76,
"y1": 179.28,
"y2": 278.64
}
|
Fig. 10. Impact of layer number on the NYC-Taxi dataset.
|
[
"(b)",
"MAPE.",
"outflow",
"inflow",
"α",
"P",
"E",
"M",
"A",
"1",
"2",
"3",
"4",
"5",
"6",
"17%",
"16%",
"15%",
"14%",
"(a)",
"RMSE.",
"outflow",
"inflow",
"α",
"S",
"E",
"R",
"M",
"1",
"2",
"3",
"4",
"5",
"6",
"23",
"21",
"19",
"17"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00008_1762755955/figures/2201.00008-Figure10-1.png
|
{
"x1": 312,
"x2": 513.31201171875,
"y1": 293.3929748535156,
"y2": 298.344970703125
}
|
200
|
7
| 9
|
Figure
|
{
"x1": 55.08,
"x2": 298.08,
"y1": 48.96,
"y2": 143.28
}
|
Fig. 7. Performance of variants on the NYC-Bike dataset.
|
[
"(b)",
"MAPE.",
"NoIFM",
"NoDLM",
"NoRSM",
"ST−TIS",
"P",
"E",
"M",
"A",
"Inflow",
"Outflow",
"25%",
"24%",
"23%",
"22%",
"21%",
"20%",
"19%",
"18%",
"17%",
"(a)",
"RMSE.",
"NoIMF",
"NoDLM",
"NoRSM",
"ST−TIS",
"S",
"E",
"R",
"M",
"Inflow",
"Outflow",
"9.5",
"9",
"8.5",
"8",
"7.5"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00008_1762755955/figures/2201.00008-Figure7-1.png
|
{
"x1": 48,
"x2": 249.18394470214844,
"y1": 158.22897338867188,
"y2": 163.18096923828125
}
|
200
|
9
| 9
|
Figure
|
{
"x1": 318.24,
"x2": 563.76,
"y1": 43.92,
"y2": 143.28
}
|
Fig. 9. Impact of surrounding observations on the NYC-Bike dataset.
|
[
"(b)",
"MAPE.",
"outflow",
"inflow",
"w",
"P",
"E",
"M",
"A",
"1",
"2",
"3",
"4",
"5",
"6",
"7",
"8",
"22%",
"22%",
"21%",
"20%",
"20%",
"20%",
"19%",
"(a)",
"RMSE.",
"outflow",
"inflow",
"w",
"S",
"E",
"R",
"M",
"1",
"2",
"3",
"4",
"5",
"6",
"7",
"8",
"9",
"8.5",
"8",
"7.5",
"7"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00008_1762755955/figures/2201.00008-Figure9-1.png
|
{
"x1": 312,
"x2": 553.7679443359375,
"y1": 158.22897338867188,
"y2": 163.18096923828125
}
|
200
|
1
| 7
|
Table
|
{
"x1": 61.919999999999995,
"x2": 547.56,
"y1": 74.88,
"y2": 372.96
}
|
TABLE 1 Comparison with the state-of-the-art methods.
|
[
"tween",
"regions",
"are",
"used",
"for",
"correlation",
"modeling.",
"•",
"Ridge",
"Regression:",
"A",
"regression",
"approach",
"for",
"time",
"se-",
"ries",
"data",
"forecasting.",
"STDN",
"8.15±0.15",
"20.87±0.39%",
"8.85±0.11",
"21.84±0.36%",
"ASTGCN",
"9.05±0.31",
"22.25±0.36%",
"9.34±0.24",
"23.13±0.30%",
"STGODE",
"8.58±0.38",
"23.33±0.26%",
"9.23±0.31",
"23.99±0.23%",
"STSAN",
"8.20±0.45",
"20.42±1.33%",
"9.87±0.23",
"23.87±0.71%",
"DSAN",
"7.97±0.25",
"20.23±0.18%",
"10.07±0.58",
"23.92±0.39%",
"ST-TIS",
"7.57±0.04",
"18.64±0.23%",
"7.73±0.10",
"18.58±0.19%",
"HA",
"11.93",
"27.06%",
"12.49",
"27.82%",
"ARIMA",
"11.25",
"25.79%",
"11.53",
"26.35%",
"Ridge",
"10.33",
"24.58%",
"10.92",
"25.29%",
"XGBoost",
"8.94",
"22.54%",
"9.57",
"23.52%",
"MLP",
"9.12±0.24",
"22.40±0.40%",
"9.83±0.19",
"23.12±0.24%",
"ConvLSTM",
"9.22±0.19",
"23.20±0.47%",
"10.40±0.17",
"25.10±0.45%",
"ST-ResNet",
"8.85±0.13",
"22.98±0.53%",
"9.80±0.12",
"25.06±0.36%",
"NYC-Bike",
"STDN",
"19.05±0.31",
"16.25±0.26%",
"24.10±0.25",
"16.30±0.23%",
"ASTGCN",
"22.05±0.37",
"20.25±0.26%",
"26.10±0.25",
"20.30±0.31%",
"STGODE",
"21.46±0.42",
"19.22±0.36%",
"27.24±0.46",
"19.30±0.34%",
"STSAN",
"23.07±0.64",
"22.24±1.91%",
"27.83±0.30",
"25.90±1.67%",
"DSAN",
"18.32±0.39",
"16.07±0.31%",
"24.27±0.30",
"17.70±0.35%",
"ST-TIS",
"17.73±0.23",
"14.65±0.32%",
"21.96±0.13",
"14.83±0.76%",
"HA",
"33.83",
"21.14%",
"43.82",
"23.18%",
"ARIMA",
"27.25",
"20.91%",
"36.53",
"22.21%",
"Ridge",
"24.38",
"20.07%",
"28.51",
"19.94%",
"XGBoost",
"21.72",
"18.70%",
"26.07",
"19.35%",
"MLP",
"22.08±0.50",
"18.31±0.83%",
"26.67±0.56",
"18.43±0.62%",
"ConvLSTM",
"23.67±0.20",
"20.70±0.20%",
"28.13±0.25",
"20.50±0.10%",
"ST-ResNet",
"21.63±0.25",
"21.09±0.51%",
"26.23±0.33",
"21.13±0.63%",
"NYC-Taxi",
"Dataset",
"Method",
"Inflow",
"OutflowRMSE",
"MAPE",
"RMSE",
"MAPE"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00008_1762755955/figures/2201.00008-Table1-1.png
|
{
"x1": 224.47698974609375,
"x2": 387.52496337890625,
"y1": 46.82400131225586,
"y2": 60.7760009765625
}
|
200
|
2
| 3
|
Figure
|
{
"x1": 78.84,
"x2": 269.28,
"y1": 42.839999999999996,
"y2": 277.2
}
|
Fig. 2. ST-TIS overview.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00008_1762755955/figures/2201.00008-Figure2-1.png
|
{
"x1": 48,
"x2": 131.95997619628906,
"y1": 292.20599365234375,
"y2": 297.1579895019531
}
|
200
|
2
| 8
|
Table
|
{
"x1": 63.72,
"x2": 284.03999999999996,
"y1": 74.88,
"y2": 222.12
}
|
TABLE 2 Comparison of training time.
|
[
"ST-ResNet",
"7.25",
"2921.75",
"STDN",
"480.21",
"23066.43",
"ASTGCN",
"25.68",
"5084.64",
"STGODE",
"18.76",
"3752.89",
"STSAN",
"426.28",
"31216.28",
"DSAN",
"434.03",
"26476.20",
"ST-TIS",
"10.37",
"1556.8",
"NYC-Bike",
"ST-ResNet",
"7.31",
"3077.51",
"STDN",
"445.47",
"34746.66",
"ASTGCN",
"25.31",
"6272.88",
"STGODE",
"18.53",
"3423.48",
"STSAN",
"426.75",
"33769.32",
"DSAN",
"386.17",
"29390.75",
"ST-TIS",
"10.21",
"1231.5",
"NYC-Taxi",
"Dataset",
"Method",
"Average",
"timeper",
"epoch",
"(s)",
"Total",
"time",
"(s)"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00008_1762755955/figures/2201.00008-Table2-1.png
|
{
"x1": 124.02498626708984,
"x2": 223.97695922851562,
"y1": 46.82400131225586,
"y2": 60.7760009765625
}
|
200
|
3
| 8
|
Table
|
{
"x1": 365.76,
"x2": 510.12,
"y1": 74.88,
"y2": 151.2
}
|
TABLE 3 Comparison of the number of parameters.
|
[
"DSAN",
"1,621,634",
"STSAN",
"34,327,298",
"ST-TIS",
"139,506",
"STDN",
"9,446,274",
"ASTGCN",
"450,031",
"STGODE",
"433,073",
"Method",
"Number",
"of",
"parameters",
"ST-ResNet",
"4,917,041"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00008_1762755955/figures/2201.00008-Table3-1.png
|
{
"x1": 363.8289794921875,
"x2": 512.1729736328125,
"y1": 46.82406234741211,
"y2": 60.77606201171875
}
|
200
|
6
| 8
|
Figure
|
{
"x1": 318.59999999999997,
"x2": 562.3199999999999,
"y1": 169.56,
"y2": 264.24
}
|
Fig. 6. Performance of variants on the NYC-Taxi dataset.
|
[
"(b)",
"MAPE.",
"NoIMF",
"NoDLM",
"NoRSM",
"ST−TIS",
"P",
"E",
"M",
"A",
"Inflow",
"Outflow",
"19%",
"18%",
"17%",
"16%",
"15%",
"14%",
"(a)",
"RMSE.",
"NoIMF",
"NoDLM",
"NoRSM",
"ST−TIS",
"S",
"E",
"R",
"M",
"Inflow",
"Outflow",
"26",
"24",
"22",
"20",
"18",
"16"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00008_1762755955/figures/2201.00008-Figure6-1.png
|
{
"x1": 312,
"x2": 511.93597412109375,
"y1": 279.0370178222656,
"y2": 283.989013671875
}
|
200
|
4
| 4
|
Figure
|
{
"x1": 78.84,
"x2": 269.28,
"y1": 186.84,
"y2": 393.12
}
|
Fig. 4. The process of connected graph generation.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00008_1762755955/figures/2201.00008-Figure4-1.png
|
{
"x1": 48,
"x2": 229.1919403076172,
"y1": 408.0920104980469,
"y2": 413.04400634765625
}
|
200
|
3
| 4
|
Figure
|
{
"x1": 78.84,
"x2": 269.28,
"y1": 42.839999999999996,
"y2": 152.28
}
|
Fig. 3. Information propagation in a graph.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00008_1762755955/figures/2201.00008-Figure3-1.png
|
{
"x1": 48,
"x2": 196.76795959472656,
"y1": 166.76797485351562,
"y2": 171.719970703125
}
|
200
|
4
| 5
|
Figure
|
{
"x1": 74.88,
"x2": 265.32,
"y1": 449.64,
"y2": 588.9599999999999
}
|
Figure 4: CO score optimization through GAX x + tanℎ(w ∗ x) on ImageNet data using pre-trained Resnet34 (blue) and Alexnet (red), where each curve corresponds to a single image. The target sco is set to 48, exceeding most CO scores of other methods.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00009_1762755964/figures/2201.00009-Figure4-1.png
|
{
"x1": 51.30699157714844,
"x2": 288.6725769042969,
"y1": 595.8010864257812,
"y2": 645.2669677734375
}
|
200
|
3
| 5
|
Figure
|
{
"x1": 47.879999999999995,
"x2": 547.1999999999999,
"y1": 68.39999999999999,
"y2": 383.76
}
|
Figure 3: Similar to Figure 2, but the results are obtained from (A) Resnet34 architecture on Pneumonia dataset, but with less fine-tuning (Resnet34_sub). (B) Resnet34 on ImageNet.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00009_1762755964/figures/2201.00009-Figure3-1.png
|
{
"x1": 51.30699920654297,
"x2": 543.96484375,
"y1": 391.4971008300781,
"y2": 408.08697509765625
}
|
200
|
6
| 10
|
Figure
|
{
"x1": 173.88,
"x2": 421.2,
"y1": 54.36,
"y2": 144.72
}
|
Figure 6: Solid red (blue) lines are x1(x2) components of sample data x. Dotted red (blue) lines are ℎ1(ℎ2) components of heatmaps ℎ with k = 1.2. Heatmap values or attribute importances are assigned large values when either (1) the true components a1, a2 differ significantly (2) the W transforms the data heterogenously i.e. not ≈ (2k + 1) 4 .
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00009_1762755964/figures/2201.00009-Figure6-1.png
|
{
"x1": 51.30694580078125,
"x2": 543.9690551757812,
"y1": 152.2591094970703,
"y2": 184.11199951171875
}
|
200
|
7
| 10
|
Figure
|
{
"x1": 50.76,
"x2": 544.3199999999999,
"y1": 201.6,
"y2": 489.96
}
|
Figure 7: Boxplots of CO scores for existing XAI methods, including another GAX implementation x ∗ ℎ = x ∗ (w ∗ x)
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00009_1762755964/figures/2201.00009-Figure7-1.png
|
{
"x1": 70.26699829101562,
"x2": 525.0088500976562,
"y1": 501.3721008300781,
"y2": 507.00299072265625
}
|
200
|
11
| 14
|
Figure
|
{
"x1": 50.76,
"x2": 544.3199999999999,
"y1": 68.39999999999999,
"y2": 344.88
}
|
Figure 11: Boxplots of CO scores and histograms for chest X-Ray dataset for pneumonia classification with Resnet34, rerun using pytorch version 2.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00009_1762755964/figures/2201.00009-Figure11-1.png
|
{
"x1": 51.3069953918457,
"x2": 543.9720458984375,
"y1": 352.3140869140625,
"y2": 368.90399169921875
}
|
200
|
12
| 14
|
Figure
|
{
"x1": 50.76,
"x2": 544.3199999999999,
"y1": 412.56,
"y2": 696.9599999999999
}
|
Figure 12: Boxplots of CO scores and histograms for chest X-Ray dataset for pneumonia classification with Resnet34, trained sub-optimally, rerun using pytorch version 2.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00009_1762755964/figures/2201.00009-Figure12-1.png
|
{
"x1": 51.30699920654297,
"x2": 543.971435546875,
"y1": 703.6150512695312,
"y2": 720.2049560546875
}
|
200
|
1
| 1
|
Figure
|
{
"x1": 50.76,
"x2": 289.08,
"y1": 247.67999999999998,
"y2": 342.71999999999997
}
|
Figure 1: Solid red (blue) lines are x1(x2) components of sample data x. Dotted red (blue) lines are ℎ1(ℎ2) components of heatmaps ℎ with k = 1.2. Heatmap values or attribute importances are assigned large values when either (1) the true components a1, a2 differ significantly (2) the W transforms the data heterogenously i.e. not ≈ (2k+ 1) 4 . See interpretations in the main text for more details.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00009_1762755964/figures/2201.00009-Figure1-1.png
|
{
"x1": 51.30699920654297,
"x2": 288.6759338378906,
"y1": 349.7781066894531,
"y2": 421.1619873046875
}
|
200
|
5
| 6
|
Figure
|
{
"x1": 59.76,
"x2": 536.04,
"y1": 68.39999999999999,
"y2": 333.71999999999997
}
|
Figure 5: (A) GAX dynamic heatmaps displayed with a slider for users to observe the evolution of heatmaps through time steps. (B-E) GAX heatmaps generated on Resnet34 for (B) healthy chest X-Ray and (C) chest X-Ray of a patient with bacterial pneumonia; and for ImageNet images (D) a sheep dog image and (E) a planetarium image. (F) An instance of bacterial pneumonia chest X-Ray showing an irregular posture with heavy noise (top right). The empty space might have been used as a false distinct feature for pneumonia classification. Three heatmaps for each image correspond to the attribution values assigned to R, G and B color channels respectively. “Abs max" specifies the maximum absolute value attained by the heatmap throughout all three channels (max is 1, due to Tanh activation). Positive/negative heatmap or attribution values (red/blue) indicate pixels to be increased/reduced in intensity to attain higher prediction confidence. At higher CO scores, negative values emerge.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00009_1762755964/figures/2201.00009-Figure5-1.png
|
{
"x1": 51.30699920654297,
"x2": 543.9728393554688,
"y1": 340.7731018066406,
"y2": 423.11590576171875
}
|
200
|
10
| 13
|
Figure
|
{
"x1": 50.76,
"x2": 544.3199999999999,
"y1": 54.36,
"y2": 329.03999999999996
}
|
Figure 10: Boxplots of CO scores and histograms for ImageNet with Alexnet, rerun using pytorch version 2.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00009_1762755964/figures/2201.00009-Figure10-1.png
|
{
"x1": 90.36499786376953,
"x2": 504.9140319824219,
"y1": 340.4740905761719,
"y2": 346.10498046875
}
|
200
|
2
| 12
|
Table
|
{
"x1": 50.76,
"x2": 544.3199999999999,
"y1": 429.47999999999996,
"y2": 540
}
|
Table 2 Comparison of validation or testing accuracies before and after augmentative explanation process x + ℎ. IMN: ImageNet Pneu: chest x-ray for pneumonia detection. COVID: COVID-19 Radiography Database. CCF: credit card fraud detection . DB: dry bean classification. IXG: InputXGradient. LGC: LayerGradCam. GBP: GuidedBackprop.
|
[
"DB,",
"drybeanFPA",
"0.877",
"0.868",
"0.708",
"0.646",
"0.817",
"0.777",
"0.712",
"IMN,",
"Resnet34",
"0.722",
"0.715",
"0.526",
"0.703",
"0.659",
"0.706",
"0.610",
"IMG,",
"Alexnet",
"0.551",
"0.53",
"0",
"0.347",
"0.542",
"0.364",
"0.484",
"0.407",
"Pneu,",
"Resnet34",
"0.691",
"0.691",
"0.601",
"0.671",
"0.691",
"0.689",
"0.619",
"Pneu,",
"Resnet34_sub",
"0.708",
"0.708",
"0.643",
"0.713",
"0.700",
"0.712",
"0.668",
"Pneu,",
"Alexnet",
"0.628",
"0.628",
"0.51",
"0.606",
"0.636",
"0.635",
"0.546",
"COVID,",
"CXCMultiSPA",
"0.865",
"0.821",
"0.624",
"0.784",
"0.245",
"0.578",
"0.562",
"CCF,",
"ccfFPA",
"0.895",
"0.872",
"0.813",
"0.692",
"0.282",
"0.188",
"0.785",
"Baseline",
"Saliency",
"IXG",
"LGC",
"Deconvolution",
"GBP",
"DeepLift"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00009_1762755964/figures/2201.00009-Table2-1.png
|
{
"x1": 51.30699920654297,
"x2": 543.9696044921875,
"y1": 384.72607421875,
"y2": 423.23394775390625
}
|
200
|
9
| 12
|
Figure
|
{
"x1": 50.76,
"x2": 544.3199999999999,
"y1": 54.36,
"y2": 333.71999999999997
}
|
Figure 9: Boxplots of CO scores and histograms for ImageNet with Resnet34, rerun using pytorch version 2. 1/0 denote correct/wrong respectively.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00009_1762755964/figures/2201.00009-Figure9-1.png
|
{
"x1": 51.30699920654297,
"x2": 543.9710083007812,
"y1": 341.1430969238281,
"y2": 357.73297119140625
}
|
200
|
1
| 3
|
Table
|
{
"x1": 51.839999999999996,
"x2": 288,
"y1": 370.44,
"y2": 437.03999999999996
}
|
Table 1 Fine-tuning results for pre-trained models on Chest X-Ray Pneumonia test dataset. The architectures marked with _sub are deliberately trained to achieve lower validation accuracy for comparison.
|
[
"accuracy",
"0.800",
"0.636",
"0.745",
"precision",
"0.757",
"0.632",
"0.726",
"recall",
"1.000",
"1.000",
"0.951",
"val.",
"acc.",
"0.99",
"0.8",
"0.8",
"Resnet34_1",
"Resnet34_sub",
"Alexnet",
"_sub"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00009_1762755964/figures/2201.00009-Table1-1.png
|
{
"x1": 51.30699157714844,
"x2": 288.67364501953125,
"y1": 303.4270935058594,
"y2": 352.8929443359375
}
|
200
|
2
| 3
|
Figure
|
{
"x1": 47.879999999999995,
"x2": 547.1999999999999,
"y1": 68.39999999999999,
"y2": 225.72
}
|
Figure 2: Distribution of CO scores obtained through AX process on existing XAI methods. Classification probability is improved if the score is positive. All distributions show gaps between CO scores of data whose classes are correctly and wrongly predicted (e.g. red arrows); correct prediction tends to yield higher CO scores. The result is obtained using Resnet34_1 on Pneumonia dataset. [sum] denotes AX process with x + ℎ.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00009_1762755964/figures/2201.00009-Figure2-1.png
|
{
"x1": 51.30699920654297,
"x2": 543.9718017578125,
"y1": 233.39710998535156,
"y2": 271.90496826171875
}
|
200
|
16
| 16
|
Figure
|
{
"x1": 50.76,
"x2": 544.3199999999999,
"y1": 415.44,
"y2": 696.9599999999999
}
|
Figure 16: Boxplots of CO scores and histograms for dry bean dataset with drybeanFPA model. 1/0 denote correct/wrong respectively.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00009_1762755964/figures/2201.00009-Figure16-1.png
|
{
"x1": 51.30699920654297,
"x2": 543.9720458984375,
"y1": 704.1810913085938,
"y2": 720.77099609375
}
|
200
|
15
| 16
|
Figure
|
{
"x1": 50.76,
"x2": 544.3199999999999,
"y1": 68.39999999999999,
"y2": 349.91999999999996
}
|
Figure 15: Boxplots of CO scores and histograms for creditcard fraud dataset with ccfFPA model. 1/0 denote correct/wrong respectively.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00009_1762755964/figures/2201.00009-Figure15-1.png
|
{
"x1": 51.30699920654297,
"x2": 543.972412109375,
"y1": 356.77911376953125,
"y2": 373.3690185546875
}
|
200
|
8
| 11
|
Figure
|
{
"x1": 50.76,
"x2": 544.3199999999999,
"y1": 54.36,
"y2": 298.8
}
|
Figure 8: Boxplots of CO scores for heatmaps from Layer GradCAM for ResNet34 and ImageNet dataset. CO scores of heatmaps generated from different layers (and resized accordingly) are shown.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00009_1762755964/figures/2201.00009-Figure8-1.png
|
{
"x1": 51.30699920654297,
"x2": 543.9703979492188,
"y1": 305.860107421875,
"y2": 322.45001220703125
}
|
200
|
14
| 15
|
Figure
|
{
"x1": 50.76,
"x2": 544.3199999999999,
"y1": 408.59999999999997,
"y2": 696.9599999999999
}
|
Figure 14: Boxplots of CO scores and histograms for COVID-19 Radiography Database with CXCMultiSPA. 1/0 denote correct/wrong respectively.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00009_1762755964/figures/2201.00009-Figure14-1.png
|
{
"x1": 51.30699920654297,
"x2": 543.96630859375,
"y1": 704.7200927734375,
"y2": 721.3099975585938
}
|
200
|
13
| 15
|
Figure
|
{
"x1": 50.76,
"x2": 544.3199999999999,
"y1": 67.67999999999999,
"y2": 344.88
}
|
Figure 13: Boxplots of CO scores and histograms for chest X-Ray dataset for pneumonia classification with Alexnet rerun using pytorch version 2.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00009_1762755964/figures/2201.00009-Figure13-1.png
|
{
"x1": 51.3069953918457,
"x2": 543.9678344726562,
"y1": 352.0531005859375,
"y2": 368.64300537109375
}
|
200
|
1
| 5
|
Table
|
{
"x1": 123.83999999999999,
"x2": 488.15999999999997,
"y1": 73.8,
"y2": 523.0799999999999
}
|
TABLE 1 Details of 44 selected datasets from the UCR 2018.
|
[
"ShakeGestureW.Z",
"50",
"50",
"10",
"Vary",
"Sensor",
"GestureMidAirD1",
"208",
"130",
"26",
"Vary",
"Trajectory",
"GestureMidAirD2",
"208",
"130",
"26",
"Vary",
"Trajectory",
"GestureMidAirD3",
"208",
"130",
"26",
"Vary",
"Trajectory",
"GesturePebbleZ1",
"132",
"172",
"6",
"Vary",
"Sensor",
"GesturePebbleZ2",
"146",
"158",
"6",
"Vary",
"Sensor",
"PickupGestureW.Z",
"50",
"50",
"10",
"Vary",
"Sensor",
"PLAID",
"537",
"537",
"11",
"Vary",
"Device",
"AllGestureWiimoteX",
"300",
"700",
"10",
"Vary",
"Sensor",
"AllGestureWiimoteY",
"300",
"700",
"10",
"Vary",
"Sensor",
"AllGestureWiimoteZ",
"300",
"700",
"10",
"Vary",
"Sensor",
"Vary",
"SemgHandG.Ch2",
"300",
"600",
"2",
"1500",
"Spectrum",
"MixedShapesSmallT.",
"100",
"2425",
"5",
"1024",
"Image",
"ACSF1",
"100",
"100",
"10",
"1460",
"Device",
"StarLightCurves",
"1000",
"8236",
"3",
"1024",
"Sensor",
"MixedShapesRegularT.",
"500",
"2425",
"5",
"1024",
"Image",
"Mallat",
"55",
"2345",
"8",
"1024",
"Simulated",
"Phoneme",
"214",
"1896",
"39",
"1024",
"Sensor",
"OliveOil",
"30",
"30",
"4",
"570",
"Spectro",
"Car",
"60",
"60",
"4",
"577",
"Sensor",
"Lightning2",
"60",
"61",
"2",
"637",
"Sensor",
"Computers",
"250",
"250",
"2",
"720",
"Device",
"Long",
"Ham",
"109",
"105",
"2",
"431",
"Spectro",
"Meat",
"60",
"60",
"3",
"448",
"Spectro",
"Fish",
"175",
"175",
"7",
"463",
"Image",
"Beef",
"30",
"30",
"5",
"470",
"Spectro",
"DodgerLoopDay",
"78",
"80",
"7",
"288",
"Sensor",
"DodgerLoopGame",
"20",
"138",
"2",
"288",
"Sensor",
"DodgerLoopWeekend",
"20",
"138",
"2",
"288",
"Sensor",
"CricketX",
"390",
"390",
"12",
"300",
"Motion",
"CricketY",
"390",
"390",
"12",
"300",
"Motion",
"CricketZ",
"390",
"390",
"12",
"300",
"Motion",
"FaceFour",
"24",
"88",
"4",
"350",
"Image",
"Medium",
"ECG200",
"100",
"100",
"2",
"96",
"ECG",
"CBF",
"30",
"900",
"3",
"128",
"Simulated",
"TwoLeadECG",
"23",
"1139",
"2",
"82",
"ECG",
"MoteStrain",
"20",
"1252",
"2",
"84",
"Sensor",
"Chinatown",
"20",
"345",
"2",
"24",
"Traffic",
"MelbournePedestrian",
"1194",
"2439",
"10",
"24",
"Traffic",
"SonyAIBORobotSur.2",
"27",
"953",
"2",
"65",
"Sensor",
"SonyAIBORobotSur.1",
"20",
"601",
"2",
"70",
"Sensor",
"DistalPhalanxO.A.G",
"400",
"139",
"3",
"80",
"Image",
"DistalPhalanxO.C.",
"600",
"276",
"2",
"80",
"Image",
"DistalPhalanxTW",
"400",
"139",
"6",
"80",
"Image",
"Short",
"Scale",
"Dataset",
"Train",
"Test",
"Class",
"SeriesLength",
"Type"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00011_1762755975/figures/2201.00011-Table1-1.png
|
{
"x1": 214.84898376464844,
"x2": 397.1529541015625,
"y1": 46.82400131225586,
"y2": 60.7760009765625
}
|
200
|
2
| 6
|
Table
|
{
"x1": 94.67999999999999,
"x2": 517.3199999999999,
"y1": 74.88,
"y2": 603
}
|
TABLE 2 Experimental results of different algorithms on 44 datasets when Nconn = 44 and Ntot = 44.
|
[
"MeanACC",
"0.6622",
"0.2377",
"0.2557",
"0.4445",
"0.6604",
"0.6743",
"0.6878",
"0.7014",
"AVG",
"rank",
"3.5455",
"7.5",
"7.3409",
"6.0113",
"3.9204",
"2.8977",
"2.6364",
"2.1478",
"Lose",
"39",
"44",
"44",
"44",
"39",
"36",
"33",
"24",
"Best",
"5",
"0",
"0",
"0",
"5",
"8",
"11",
"20",
"GestureMidAirD1",
"0.5538",
"0.0384",
"0.0384",
"0.0923",
"0.5462",
"0.5538",
"0.4615",
"0.5769",
"GestureMidAirD2",
"0.4231",
"0.0384",
"0.0384",
"0.0923",
"0.4154",
"0.4462",
"0.4692",
"0.5308",
"GestureMidAirD3",
"0.3",
"0.0384",
"0.0384",
"0.0923",
"0.2693",
"0.2615",
"0.2231",
"0.2769",
"GesturePebbleZ1",
"0.4419",
"0.1628",
"0.1628",
"0.2558",
"0.4767",
"0.4826",
"0.5",
"0.4883",
"GesturePebbleZ2",
"0.4241",
"0.1519",
"0.1519",
"0.2722",
"0.5126",
"0.557",
"0.6013",
"0.5886",
"PickupGestureW.Z",
"0.56",
"0.1",
"0.1",
"0.24",
"0.62",
"0.6",
"0.7",
"0.74",
"PLAID",
"0.203",
"0.0615",
"0.0615",
"0.0615",
"0.2198",
"0.2253",
"0.2924",
"0.2589",
"ShakeGestureW.Z",
"0.92",
"0.1",
"0.1",
"0.1",
"0.96",
"0.92",
"0.96",
"0.96",
"Win",
"4",
"0",
"0",
"0",
"3",
"7",
"10",
"18",
"Tie",
"1",
"0",
"0",
"0",
"2",
"1",
"1",
"2",
"SemgHandG.Ch2",
"0.7067",
"0.65",
"0.65",
"0.555",
"0.72",
"0.7383",
"0.6867",
"0.72",
"AllGestureWiimoteX",
"0.2643",
"0.1",
"0.1",
"0.1371",
"0.2729",
"0.3043",
"0.2929",
"0.2914",
"AllGestureWiimoteY",
"0.2585",
"0.1",
"0.1",
"0.1357",
"0.3186",
"0.3029",
"0.2529",
"0.2829",
"AllGestureWiimoteZ",
"0.2886",
"0.1",
"0.1",
"0.1343",
"0.2671",
"0.29",
"0.4014",
"0.3786",
"MixedShapesSmallT.",
"0.8029",
"0.1889",
"0.1889",
"0.2421",
"0.7942",
"0.8062",
"0.8318",
"0.8388",
"ACSF1",
"0.77",
"0.1",
"0.19",
"0.19",
"0.82",
"0.89",
"0.87",
"0.88",
"Mallat",
"0.7446",
"0.1254",
"0.1254",
"0.4141",
"0.7638",
"0.7539",
"0.7906",
"0.8299",
"Phoneme",
"0.2231",
"0.02",
"0.02",
"0.1108",
"0.2147",
"0.2247",
"0.2859",
"0.2954",
"StarLightCurves",
"0.9534",
"0.1429",
"0.1429",
"0.5062",
"0.9519",
"0.9584",
"0.9571",
"0.9582",
"MixedShapesRegularT.",
"0.8586",
"0.1889",
"0.1889",
"0.2223",
"0.8384",
"0.8598",
"0.8643",
"0.8907",
"Ham",
"0.7143",
"0.4857",
"0.4857",
"0.6762",
"0.7048",
"0.7143",
"0.7048",
"0.6952",
"Meat",
"0.8667",
"0.3333",
"0.3333",
"0.7333",
"0.8333",
"0.8333",
"0.9",
"0.917",
"Fish",
"0.5657",
"0.1371",
"0.1371",
"0.2857",
"0.5771",
"0.6",
"0.6",
"0.6229",
"Beef",
"0.7667",
"0.2",
"0.2",
"0.5667",
"0.7",
"0.7",
"0.7",
"0.7667",
"OliveOil",
"0.8333",
"0.167",
"0.167",
"0.7",
"0.8667",
"0.8667",
"0.8333",
"0.8333",
"Car",
"0.5833",
"0.233",
"0.233",
"0.5",
"0.5667",
"0.5833",
"0.5667",
"0.6333",
"Lightning2",
"0.7869",
"0.459",
"0.459",
"0.7705",
"0.7869",
"0.8033",
"0.7541",
"0.7869",
"Computers",
"0.78",
"0.5",
"0.5",
"0.584",
"0.688",
"0.748",
"0.788",
"0.804",
"ECG200",
"0.86",
"0.36",
"0.36",
"0.8",
"0.84",
"0.85",
"0.87",
"0.85",
"CBF",
"0.987",
"0.3333",
"0.5911",
"0.5911",
"0.973",
"0.9922",
"0.9922",
"0.9956",
"DodgerLoopDay",
"0.575",
"0.15",
"0.15",
"0.3875",
"0.55",
"0.525",
"0.5125",
"0.5375",
"DodgerLoopGame",
"0.6884",
"0.5217",
"0.5217",
"0.6232",
"0.7826",
"0.7609",
"0.7609",
"0.7464",
"DodgerLoopWeekend",
"0.8261",
"0.7391",
"0.7391",
"0.7319",
"0.8841",
"0.8913",
"0.913",
"0.9203",
"CricketX",
"0.5897",
"0.0692",
"0.1371",
"0.2256",
"0.5667",
"0.6128",
"0.659",
"0.6718",
"CricketY",
"0.5051",
"0.0949",
"0.1357",
"0.1949",
"0.5",
"0.4949",
"0.5538",
"0.5974",
"CricketZ",
"0.6205",
"0.0846",
"0.0846",
"0.2256",
"0.5692",
"0.6",
"0.6692",
"0.7256",
"FaceFour",
"0.6477",
"0.1591",
"0.1591",
"0.4659",
"0.6591",
"0.6932",
"0.6932",
"0.6818",
"TwoLeadECG",
"0.7463",
"0.4996",
"0.4996",
"0.7305",
"0.7287",
"0.7278",
"0.8112",
"0.7665",
"MoteStrain",
"0.7788",
"0.5391",
"0.5391",
"0.6933",
"0.7923",
"0.8283",
"0.8163",
"0.8203",
"Dataset",
"Baseline",
"FedAvg",
"FedAvgM",
"FedGrad",
"FTL",
"FTLS",
"FKD",
"EFDLS",
"Chinatown",
"0.9623",
"0.2754",
"0.2754",
"0.9623",
"0.9665",
"0.9537",
"0.9275",
"0.9478",
"MelbournePedestrian",
"0.9139",
"0.1",
"0.1",
"0.7784",
"0.8486",
"0.8922",
"0.9379",
"0.9453",
"SonyAIBORobotSur.2",
"0.8961",
"0.383",
"0.383",
"0.8363",
"0.8688",
"0.9035",
"0.915",
"0.8961",
"SonyAIBORobotSur.1",
"0.8652",
"0.5707",
"0.6619",
"0.7887",
"0.8236",
"0.8702",
"0.8369",
"0.8819",
"DistalPhalanxO.A.G",
"0.6763",
"0.1079",
"0.1079",
"0.6187",
"0.6259",
"0.6475",
"0.6691",
"0.6475",
"DistalPhalanxO.C.",
"0.75",
"0.417",
"0.6619",
"0.6776",
"0.7464",
"0.7465",
"0.7536",
"0.7428",
"DistalPhalanxTW",
"0.6547",
"0.1295",
"0.1295",
"0.554",
"0.6259",
"0.6547",
"0.6835",
"0.6403"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00011_1762755975/figures/2201.00011-Table2-1.png
|
{
"x1": 143.0349884033203,
"x2": 468.9649658203125,
"y1": 46.82400131225586,
"y2": 61.77197265625
}
|
200
|
5
| 9
|
Figure
|
{
"x1": 47.879999999999995,
"x2": 564.12,
"y1": 188.64,
"y2": 746.28
}
|
Fig. 5. Critical difference diagram of the average ranks of various FL algorithms on 44 datasets.
|
[
"[10]",
"X.",
"Zhang,",
"Y.",
"Gao,",
"J.",
"Lin,",
"and",
"C.-T.",
"Lu,",
"“Tapnet:",
"Multivariate",
"time",
"series",
"classification",
"with",
"attentional",
"prototypical",
"network,”",
"In",
"Proc.",
"AAAI",
"2020,",
"pp.",
"6845–6852,",
"2020.",
"[11]",
"F.",
"Karim,",
"S.",
"Majumdar,",
"H.",
"Darabi,",
"and",
"S.",
"Harford,",
"“Multivariate",
"lstm-fcns",
"for",
"time",
"series",
"classification,”",
"Neural",
"Networks,",
"vol.",
"116,",
"pp.",
"237–245,",
"2019.",
"[12]",
"Z.",
"Xiao,",
"X.",
"Xu,",
"H.",
"Xing,",
"S.",
"Luo,",
"P.",
"Dai,",
"and",
"D.",
"Zhan,",
"“Rtfn:",
"A",
"robust",
"temporal",
"feature",
"network",
"for",
"time",
"series",
"classification.”",
"Inform.",
"Sciences,",
"vol.",
"571,",
"pp.",
"65–86,",
"2021.",
"[13]",
"G.",
"Li,",
"B.",
"Choi,",
"J.",
"Xu,",
"S.",
"Bhowmick,",
"K.-P.",
"Chun,",
"and",
"G.",
"Wong,",
"“Shapenet:",
"A",
"shapelet-neural",
"network",
"approach",
"for",
"multivariate",
"time",
"series",
"classification,”",
"In",
"Proc.",
"AAAI",
"2021,",
"vol.",
"35,",
"no.",
"9,",
"pp.",
"8375–8383,",
"2021.",
"[14]",
"D.",
"Lee,",
"S.",
"Lee,",
"and",
"H.",
"Yu,",
"“Learnable",
"dynamic",
"temporal",
"pooling",
"for",
"time",
"series",
"classification,”",
"In",
"Proc.",
"AAAI",
"2021,",
"vol.",
"35,",
"no.",
"9,",
"pp.",
"8288–8296,",
"2021.",
"[15]",
"B.",
"Arcas,",
"G.",
"Bacon,",
"K.",
"Bonawitz,",
"and",
"et",
"al,",
"“Federated",
"learning:",
"Collaborative",
"machine",
"learning",
"without",
"centralized",
"training",
"data,”",
"https://ai.googleblog.com/2017/04/federated-learning-",
"collaborative.html,",
"2017.",
"[16]",
"Q.",
"Yang,",
"Y.",
"Liu,",
"T.",
"Chen,",
"and",
"Y.",
"Tong,",
"“Federated",
"machine",
"learn-",
"ing:",
"Concept",
"and",
"applications,”",
"ACM",
"Trans.",
"Intell.",
"Syst.",
"Technol.,",
"vol.",
"10,",
"no.",
"2,",
"pp.",
"1–19,",
"2019.",
"[17]",
"Q.",
"Li,",
"Z.",
"Wen,",
"Z.",
"Wu,",
"S.",
"Hu,",
"N.",
"Wang,",
"Y.",
"Li,",
"X.",
"Liu,",
"and",
"B.",
"He,",
"“A",
"survey",
"on",
"federated",
"learning",
"systems:",
"Vision,",
"hype",
"and",
"reality",
"for",
"data",
"privacy",
"and",
"protection,”",
"IEEE",
"Trans.",
"Knowl.",
"Data",
"Eng.,",
"pp.",
"1–1,",
"2021.",
"[18]",
"M.",
"McMahan,",
"E.",
"Moore,",
"D.",
"Ramage,",
"S.",
"Hampson,",
"and",
"B.",
"Arcas,",
"“Communication-efficient",
"learning",
"of",
"deep",
"networks",
"from",
"decen-",
"tralized",
"data,”",
"In",
"Proc.",
"AISTATS",
"2017,",
"pp.",
"1–11,",
"2017.",
"[19]",
"J.",
"Ma,",
"Q.",
"Zhang,",
"J.",
"Lou,",
"L.",
"Xiong,",
"and",
"J.",
"Ho,",
"“Communication",
"ef-",
"ficient",
"federated",
"generalized",
"tensor",
"factorization",
"for",
"collaborative",
"health",
"data",
"analytics,”",
"In",
"Proc.",
"30th",
"The",
"Web",
"Conference",
"2021,",
"2021.",
"[20]",
"B.",
"Liu,",
"Y.",
"Guo,",
"and",
"X.",
"Chen,",
"“Pfa:",
"Privacy-preserving",
"federated",
"adaptation",
"for",
"effective",
"model",
"personalization,”",
"In",
"Proc.",
"30th",
"The",
"Web",
"Conference",
"2021,",
"2021.",
"[21]",
"J.",
"Wu,",
"Z.",
"Huang,",
"Y.",
"Ning,",
"H.",
"Wang,",
"E.",
"Chen,",
"J.",
"Yi,",
"and",
"B.",
"Zhou,",
"“Hierarchical",
"personalized",
"federated",
"learning",
"for",
"user",
"modeling,”",
"In",
"Proc.",
"30th",
"The",
"Web",
"Conference",
"2021,",
"2021.",
"[22]",
"Q.",
"Yang,",
"J.",
"Zhang,",
"W.",
"Hao,",
"G.",
"Spell,",
"and",
"L.",
"Carin,",
"“Flop:",
"Federated",
"learning",
"on",
"medical",
"datasets",
"using",
"partial",
"networks,”",
"In",
"Proc.",
"ACM",
"KDD’21,",
"2021.",
"[23]",
"Y.",
"Liu,",
"Y.",
"Kang,",
"C.",
"Xing,",
"T.",
"Chen,",
"and",
"Q.",
"Yang,",
"“A",
"secure",
"federated",
"transfer",
"learning",
"framework,”",
"IEEE",
"Intell.",
"Syst.,",
"vol.",
"35,",
"no.",
"4,",
"pp.",
"70–82,",
"2020.",
"[24]",
"H.",
"Yang,",
"H.",
"He,",
"W.",
"Zhang,",
"and",
"X.",
"Cao,",
"“Fedsteg:",
"A",
"federated",
"transfer",
"learning",
"framework",
"for",
"secure",
"image",
"steganalysis,”",
"IEEE",
"Trans.",
"Netw.",
"Sci.",
"Eng.,",
"vol.",
"8,",
"no.",
"2,",
"pp.",
"1084–1094,",
"2021.",
"[25]",
"D.",
"Dimitriadis,",
"K.",
"Kumatani,",
"R.",
"Gmyr,",
"Y.",
"Gaur,",
"and",
"S.",
"Eskimez,",
"“Federated",
"transfer",
"learning",
"with",
"dynamic",
"gradient",
"aggregation,”",
"arXiv",
"preprint",
"arXiv:2008.02452,",
"2020.",
"[26]",
"U.",
"Majeed,",
"S.",
"Hassan,",
"and",
"C.",
"Hong,",
"“Cross-silo",
"model-based",
"secure",
"federated",
"transfer",
"learning",
"for",
"flow-based",
"traffic",
"classifi-",
"cation,”",
"In",
"Proc.",
"ICOIN",
"2021,",
"2021.",
"[27]",
"H.",
"Seo,",
"J.",
"Park,",
"S.",
"Oh,",
"M.",
"Bennis,",
"and",
"S.-L.",
"Kim,",
"“Federated",
"knowledge",
"distillation,”",
"arXiv",
"preprint",
"arXiv:2011.02367,",
"2020.",
"[28]",
"C.",
"He,",
"M.",
"Annavaram,",
"and",
"S.",
"Avestimehr,",
"“Group",
"knowledge",
"transfer:",
"Federated",
"learning",
"of",
"large",
"cnns",
"at",
"the",
"edge,”",
"In",
"Proc.",
"NeurIPS",
"2020,",
"2020.",
"[29]",
"R.",
"Mishra,",
"H.",
"Gupta,",
"and",
"T.",
"Dutta,",
"“A",
"network",
"resource",
"aware",
"federated",
"learning",
"approach",
"using",
"knowledge",
"distillation,”",
"In",
"Proc.",
"INFOCOM",
"2021,",
"2021.",
"[30]",
"S.",
"Itahara,",
"T.",
"Nishio,",
"Y.",
"Koda,",
"M.",
"Morikura,",
"and",
"K.",
"Ya-",
"mamoto,",
"“Distillation-based",
"semi-supervised",
"federated",
"learning",
"for",
"communication-efficient",
"collaborative",
"training",
"with",
"non-iid",
"private",
"data,”",
"IEEE",
"Trans.",
"Mobile",
"Comput.,",
"pp.",
"1–1,",
"2021.",
"[31]",
"Y.",
"Chen,",
"X.",
"Sun,",
"and",
"Y.",
"Jin,",
"“Communication-efficient",
"federated",
"deep",
"learning",
"with",
"layer-wise",
"asynchronous",
"model",
"update",
"and",
"temporally",
"weighted",
"aggregation,”",
"IEEE",
"Trans.",
"Neur.",
"Net.",
"Learn.",
"Sys.,",
"vol.",
"31,",
"no.",
"10,",
"pp.",
"4229–4238,",
"2020.",
"[32]",
"F.",
"Sattler,",
"S.",
"Wiedemann,",
"K.-R.",
"Müller,",
"and",
"W.",
"Samek,",
"“Robust",
"and",
"communication-efficient",
"federated",
"learning",
"from",
"non-i.i.d.",
"data,”",
"IEEE",
"Trans.",
"Neur.",
"Net.",
"Learn.",
"Sys.,",
"vol.",
"31,",
"no.",
"9,",
"pp.",
"3400–3413,",
"2020.",
"[33]",
"L.",
"Nagalapatti",
"and",
"R.",
"Narayanam,",
"“Game",
"of",
"gradients:",
"Mitigat-",
"ing",
"irrelevant",
"clients",
"in",
"federated",
"learning,”",
"In",
"Proc.",
"AAAI",
"2021,",
"vol.",
"35,",
"no.",
"10,",
"pp.",
"9046–9054,",
"2021.",
"[34]",
"X.",
"Cao,",
"J.",
"Jia,",
"and",
"N.",
"Gong,",
"“Provably",
"secure",
"federated",
"learning",
"against",
"malicious",
"clients,”",
"In",
"Proc.",
"AAAI",
"2020,",
"vol.",
"35,",
"no.",
"8,",
"pp.",
"6885–6893,",
"2020.",
"[35]",
"J.",
"Hong,",
"Z.",
"Zhu,",
"S.",
"Yu,",
"Z.",
"Wang,",
"H.",
"Dodge,",
"and",
"J.",
"Zhou,",
"“Federated",
"adversarial",
"debiasing",
"for",
"fair",
"and",
"transferable",
"representations,”",
"In",
"Proc.",
"ACM",
"KDD’21,",
"vol.",
"1,",
"no.",
"1,",
"August",
"2021.",
"[36]",
"P.",
"Zhou,",
"L.",
"Wang,",
"L.",
"Guo,",
"S.",
"Gong,",
"and",
"B.",
"Zheng,",
"“A",
"privacy-",
"preserving",
"distributed",
"contextual",
"federated",
"online",
"learning",
"frame-",
"work",
"with",
"big",
"data",
"support",
"in",
"social",
"recommender",
"systems,”",
"IEEE",
"Trans.",
"Knowl.",
"Data",
"Eng.,",
"vol.",
"33,",
"no.",
"3,",
"pp.",
"824–838,",
"2021.",
"[37]",
"Z.",
"Pan,",
"L.",
"Hu,",
"W.",
"Tang,",
"J.",
"Li,",
"Y.",
"He,",
"and",
"Z.",
"Liu,",
"“Privacy-",
"preserving",
"multi-granular",
"federated",
"neural",
"architecture",
"search",
"a",
"general",
"framework,”",
"IEEE",
"Trans.",
"Knowl.",
"Data",
"Eng.,",
"pp.",
"1–1,",
"2021.",
"[38]",
"M.",
"Crawshaw,",
"“Multi-task",
"learning",
"with",
"deep",
"neural",
"networks:",
"A",
"survey,”",
"arXiv",
"preprint",
"arXiv:",
"2009.09796,",
"2020.",
"[39]",
"A.",
"Ruiz,",
"M.",
"Flynn,",
"and",
"A.",
"Bagnall,",
"“Benchmarking",
"multivari-",
"ate",
"time",
"series",
"classification",
"algorithms,”",
"arXiv",
"preprint",
"arXiv:",
"2007.13156,",
"2020.",
"[40]",
"J.",
"Lines",
"and",
"A.",
"Bagnall,",
"“Time",
"series",
"classification",
"with",
"ensembles",
"of",
"elastic",
"distance",
"measures,”",
"Data",
"Min.",
"Knowl.",
"Disc.,",
"vol.",
"29,",
"p.",
"565–592,",
"2015.",
"[41]",
"A.",
"Bagnall,",
"J.",
"Lines,",
"J.",
"Hills,",
"and",
"A.",
"Bostrom,",
"“Time",
"series",
"classi-",
"fication",
"with",
"cote:",
"the",
"collective",
"of",
"transformation-based",
"ensem-",
"bles,”",
"In",
"Proc.",
"ICDE",
"2016,",
"pp.",
"1548–1549,",
"2016.",
"[42]",
"J.",
"Lines,",
"S.",
"Taylor,",
"and",
"A.",
"Bagnall,",
"“Time",
"series",
"classification",
"with",
"hive-cote:",
"the",
"hierarchical",
"of",
"transformation-based",
"ensembles,”",
"ACM",
"Trans.",
"Knowl.",
"Discov.",
"D,",
"vol.",
"21,",
"no.",
"52,",
"pp.",
"1–35,",
"2018.",
"[43]",
"K.",
"Fauvel,",
"E.",
"Fromont,",
"V.",
"Masson,",
"P.",
"Faverdin,",
"and",
"A.",
"Termier,",
"“Lo-",
"cal",
"cascade",
"ensemble",
"for",
"multivariate",
"data",
"classification,”",
"arXiv",
"preprint",
"arXiv:2005.03645,",
"2020.",
"[44]",
"J.",
"Lines,",
"L.",
"Davis,",
"J.",
"Hills,",
"and",
"A.",
"Bagnall,",
"“A",
"shapelet",
"transform",
"for",
"time",
"series",
"classification,”",
"In",
"Proc.",
"ACM",
"KDD’12,",
"2012.",
"[45]",
"M.",
"Baydogan,",
"G.",
"Runger,",
"and",
"E.",
"Tuv,",
"“A",
"bag-of-features",
"frame-",
"work",
"to",
"classify",
"time",
"series,”",
"IEEE",
"Trans.",
"Pattern",
"Anal.,",
"vol.",
"35,",
"no.",
"11,",
"pp.",
"2796–2802,",
"2013.",
"[46]",
"A.",
"Dempster,",
"D.",
"Schmidt,",
"and",
"G.",
"Webb,",
"“Minirocket:",
"A",
"very",
"fast",
"(almost)",
"deterministic",
"transform",
"for",
"time",
"series",
"classification,”",
"In",
"Proc.",
"ACM",
"KDD’21,",
"2021.",
"[47]",
"M.",
"Baydogan",
"and",
"G.",
"Runger,",
"“Time",
"series",
"representation",
"and",
"similarity",
"based",
"on",
"local",
"auto",
"patterns,”",
"Data",
"Min.",
"Knowl.",
"Disc.,",
"vol.",
"30,",
"p.",
"476–509,",
"2016.",
"[48]",
"J.",
"Large,",
"A.",
"Bagnall,",
"S.",
"Malinowski,",
"and",
"R.",
"Tavenard,",
"“From",
"bop",
"to",
"boss",
"and",
"beyond:",
"time",
"series",
"classification",
"with",
"dictionary",
"based",
"classifier,”",
"arXiv",
"preprint",
"arXiv:1809.06751,",
"2018.",
"[49]",
"W.",
"Pei,",
"H.",
"Dibeklioglu,",
"D.",
"Tax,",
"and",
"L.",
"van",
"der",
"Maaten,",
"“Multivari-",
"ate",
"time-series",
"classification",
"using",
"the",
"hidden-unit",
"logistic",
"model,”",
"IEEE",
"Trans.",
"Neur.",
"Net.",
"Lear.,",
"vol.",
"29,",
"no.",
"4,",
"pp.",
"920–931,",
"2018."
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00011_1762755975/figures/2201.00011-Figure5-1.png
|
{
"x1": 48,
"x2": 384.1439208984375,
"y1": 158.53598022460938,
"y2": 163.48797607421875
}
|
200
|
1
| 2
|
Figure
|
{
"x1": 47.879999999999995,
"x2": 564.12,
"y1": 42.839999999999996,
"y2": 334.08
}
|
Fig. 1. The schematic diagram of EFDLS. Note that ‘FBST Framework’ and ‘DBWM Scheme’ denote the feature-based student-teacher framework deployed on each user and the distance-based weights matching scheme run on the server. ‘Conv x 9 128’ represents a 1-dimensional convolutional neural network, where its filter size and channel sizes are set to 9 and 128. ‘BN’ is a batch normalization module, and ‘ReLU’ is the rectified linear unit activation function.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00011_1762755975/figures/2201.00011-Figure1-1.png
|
{
"x1": 48,
"x2": 564,
"y1": 360.68701171875,
"y2": 392.6390075683594
}
|
200
|
3
| 7
|
Figure
|
{
"x1": 47.879999999999995,
"x2": 304.2,
"y1": 243,
"y2": 391.32
}
|
Fig. 3. MeanACC results with different values on 44 datasets when Nconn = 44 and Ntot = 44.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00011_1762755975/figures/2201.00011-Figure3-1.png
|
{
"x1": 48,
"x2": 300.00201416015625,
"y1": 417.2959899902344,
"y2": 432.2439880371094
}
|
200
|
2
| 7
|
Figure
|
{
"x1": 47.879999999999995,
"x2": 304.2,
"y1": 42.839999999999996,
"y2": 189
}
|
Fig. 2. MeanACC results obtained by EFDLS with different ratios of Nconn to Ntot on 44 datasets when Ntot = 44.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00011_1762755975/figures/2201.00011-Figure2-1.png
|
{
"x1": 48,
"x2": 299.9999694824219,
"y1": 215.28097534179688,
"y2": 230.22894287109375
}
|
200
|
4
| 8
|
Figure
|
{
"x1": 47.879999999999995,
"x2": 564.12,
"y1": 42.839999999999996,
"y2": 269.28
}
|
Fig. 4. Accuracy plot results showing the performance difference between two given algorithms. (a) Baseline vs. EFDLS; (b) FedAvg vs. EFDLS; (c) FedAvgM vs. EFDLS; (d) FedGrad vs. EFDLS; (e) FTL vs. EFDLS; (f) FTLS vs. EFDLS; (g) FKD vs. EFDLS.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00011_1762755975/figures/2201.00011-Figure4-1.png
|
{
"x1": 48,
"x2": 564,
"y1": 295.63201904296875,
"y2": 309.5840148925781
}
|
200
|
1
| 5
|
Table
|
{
"x1": 52.919999999999995,
"x2": 294.12,
"y1": 635.76,
"y2": 682.1999999999999
}
|
Table 1: Comparison of MORAL and DRLHP in Emergency.
|
[
"MORAL",
"5.76(±0.13)",
"40.08(±2.9)",
"25",
"3e6",
"(3e6)",
"DRLHP",
"5.62(±0.17)",
"12.32(±3.0)",
"1000",
"12e6",
"(-)",
"Extinguished",
"Fire",
"Nr.",
"of",
"Queries",
"Steps",
"(IRL)",
"People",
"Saved"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00012_1762755993/figures/2201.00012-Table1-1.png
|
{
"x1": 54.880001068115234,
"x2": 292.6678161621094,
"y1": 686.4740600585938,
"y2": 692.10498046875
}
|
200
|
6
| 5
|
Figure
|
{
"x1": 70.92,
"x2": 277.2,
"y1": 82.8,
"y2": 186.12
}
|
Figure 6: Average number of broken vases over three training runs.MORAL learns a safe policy, despite being provided with adversarial preferences.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00012_1762755993/figures/2201.00012-Figure6-1.png
|
{
"x1": 53.44900131225586,
"x2": 295.64874267578125,
"y1": 199.78810119628906,
"y2": 227.33697509765625
}
|
200
|
7
| 5
|
Figure
|
{
"x1": 318.96,
"x2": 557.28,
"y1": 82.8,
"y2": 227.16
}
|
Figure 7: Mean training curves of DRLHP and MORAL on preference ratios (3, 1, 1) (top), (1, 3, 1) (middle) and (1, 1, 3) (bottom).
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00012_1762755993/figures/2201.00012-Figure7-1.png
|
{
"x1": 317.677001953125,
"x2": 558.2007446289062,
"y1": 240.24610900878906,
"y2": 267.79498291015625
}
|
200
|
12
| 10
|
Figure
|
{
"x1": 55.8,
"x2": 292.32,
"y1": 84.96,
"y2": 203.4
}
|
Figure 12: Reward model architecture for DRLHP with convolutional (yellow) and linear (blue) layers. Actions are embedded through a linear layer and concatenatedwith the current state before being fed through subsequent layers.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00012_1762755993/figures/2201.00012-Figure12-1.png
|
{
"x1": 53.564998626708984,
"x2": 295.6488342285156,
"y1": 218.6970977783203,
"y2": 257.2039794921875
}
|
200
|
3
| 10
|
Table
|
{
"x1": 371.88,
"x2": 504,
"y1": 300.96,
"y2": 404.28
}
|
Table 3: AIRL hyperparameters in Emergency.
|
[
"Epochs",
"PPO",
"5",
"𝜖-clip",
"0.1",
"𝛾",
"0.999",
"Batch",
"Size",
"PPO",
"12",
"Environment",
"Steps",
"3e6",
"lr-PPO",
"5e-4",
"Batch",
"Size",
"Discriminator",
"512",
"Hyperparameter",
"Value",
"lr-Discriminator",
"5e-4"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00012_1762755993/figures/2201.00012-Table3-1.png
|
{
"x1": 345.43798828125,
"x2": 530.4227905273438,
"y1": 406.287109375,
"y2": 411.9179992675781
}
|
200
|
1
| 1
|
Figure
|
{
"x1": 326.88,
"x2": 548.28,
"y1": 83.88,
"y2": 187.2
}
|
Figure 1: Multi-Objective Reinforced Active Learning.
|
[
"Prior",
"PPO",
"Update",
"Query",
"Posterior",
"Mean",
"Step",
"2:",
"Active",
"MORL",
"Reward",
"Function",
"AIRL",
"Step",
"1:",
"IRL",
"Provides",
"Provides",
"Expert",
"k",
"Expert",
"1"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00012_1762755993/figures/2201.00012-Figure1-1.png
|
{
"x1": 329.5320129394531,
"x2": 546.6267700195312,
"y1": 201.0980987548828,
"y2": 206.72900390625
}
|
200
|
8
| 6
|
Figure
|
{
"x1": 347.76,
"x2": 523.0799999999999,
"y1": 81.36,
"y2": 257.4
}
|
Figure 8: (a) Average preference deviation as a function of the number of active and random queries. (b) Average preference deviation of active learning as a function of the proportion of noisy responses to queries.
|
[
"Active",
"Random",
"(0",
"noise)",
"(b)",
"tio",
"n",
"ev",
"ia",
"ce",
"D",
"er",
"en",
"Pr",
"ef",
"0.04",
"0.03",
"0.02",
"0.0",
"0.1",
"0.2",
"0.3",
"Preference",
"Noise",
"tio",
"n",
"(a)",
"Active",
"Random",
"ev",
"ia",
"ce",
"D",
"er",
"en",
"Pr",
"ef",
"0.08",
"0.06",
"0.04",
"0.02",
"5",
"10",
"15",
"20",
"25",
"30",
"35",
"40",
"45",
"50",
"#",
"Queries"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00012_1762755993/figures/2201.00012-Figure8-1.png
|
{
"x1": 317.9549865722656,
"x2": 559.8058471679688,
"y1": 277.86810302734375,
"y2": 316.3759765625
}
|
200
|
11
| 9
|
Figure
|
{
"x1": 319.68,
"x2": 557.28,
"y1": 320.76,
"y2": 435.24
}
|
Figure 11: Convolutional discriminator architecture for trainingAIRL in bigger environmentswith a parallel stream of convolutional (yellow) and linear (blue) layers.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00012_1762755993/figures/2201.00012-Figure11-1.png
|
{
"x1": 317.9549865722656,
"x2": 558.38916015625,
"y1": 449.9471130371094,
"y2": 477.4959716796875
}
|
200
|
10
| 9
|
Figure
|
{
"x1": 319.68,
"x2": 557.28,
"y1": 83.88,
"y2": 195.48
}
|
Figure 10: Linear discriminator architecture. A forward pass calculates activations of the networks 𝑔𝜃 and ℎ𝜃 respectively and combines them into the reward prediction 𝑓𝜃 .
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00012_1762755993/figures/2201.00012-Figure10-1.png
|
{
"x1": 317.9549865722656,
"x2": 558.4529418945312,
"y1": 209.88307189941406,
"y2": 239.39593505859375
}
|
200
|
9
| 9
|
Figure
|
{
"x1": 55.8,
"x2": 292.32,
"y1": 333,
"y2": 439.2
}
|
Figure 9: Actor-Critic architecture of the PPO agent consisting of convolutional (yellow) and linear (blue) layers. Regardless of the input dimension, we use 𝐶𝑜𝑢𝑡 output channels and kernel sizes of 2.
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00012_1762755993/figures/2201.00012-Figure9-1.png
|
{
"x1": 53.79798889160156,
"x2": 295.6492614746094,
"y1": 453.7931213378906,
"y2": 492.30096435546875
}
|
200
|
2
| 7
|
Table
|
{
"x1": 63.72,
"x2": 284.03999999999996,
"y1": 571.68,
"y2": 673.1999999999999
}
|
Table 2: Comparison of MORAL to previous work in terms of supported capabilities.
|
[
"Deep",
"×",
"×",
"✓",
"✓Learning",
"Multi-",
"∼",
"✓",
"×",
"✓Objective",
"Multiple",
"×",
"∼",
"×",
"✓Experts",
"Ethics-",
"Shaping",
"[47]",
"Policy-",
"Orchestration",
"[30]",
"DRLHP",
"[10]",
"MORAL"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00012_1762755993/figures/2201.00012-Table2-1.png
|
{
"x1": 53.50199890136719,
"x2": 294.0436096191406,
"y1": 675.5150756835938,
"y2": 692.10498046875
}
|
200
|
4
| 3
|
Figure
|
{
"x1": 340.91999999999996,
"x2": 535.3199999999999,
"y1": 526.68,
"y2": 663.12
}
|
Figure 4: The Delivery Environment consists of a primary goal (Deliver) and three different norms (Help a human, Clean a tile, Avoid the vase).
|
[] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00012_1762755993/figures/2201.00012-Figure4-1.png
|
{
"x1": 317.9549865722656,
"x2": 559.2984008789062,
"y1": 676.5110473632812,
"y2": 704.0599975585938
}
|
200
|
3
| 3
|
Figure
|
{
"x1": 350.64,
"x2": 524.88,
"y1": 86.03999999999999,
"y2": 199.79999999999998
}
|
Figure 3: Query efficiency of MORAL for finding a trade-off thatmatches the given preferences. Averaged over three random seeds.
|
[
"av",
"ed",
"le",
"S",
"Pe",
"op",
"6",
"5",
"4",
"3",
"2",
"1",
"0",
"ire",
"ed",
"F",
"gu",
"ish",
"Ex",
"tin",
"70",
"60",
"50",
"40",
"30",
"20",
"10",
"0",
"#",
"Queries",
"5",
"10",
"25"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00012_1762755993/figures/2201.00012-Figure3-1.png
|
{
"x1": 317.9549865722656,
"x2": 559.8060302734375,
"y1": 216.7600860595703,
"y2": 244.3089599609375
}
|
200
|
2
| 3
|
Figure
|
{
"x1": 72,
"x2": 273.24,
"y1": 89.64,
"y2": 205.56
}
|
Figure 2: Intermediate policies found during MORAL in the Emergency domain, compared to a manually computed CCS. MORAL approximates a Pareto-optimal solution that most closely matches the given preferences.
|
[
"0.30",
"0.25",
"0.20",
"0.15",
"0.10",
"MORAL",
"CCS",
"av",
"ed",
"le",
"S",
"Pe",
"op",
"6",
"5",
"4",
"3",
"2",
"1",
"0",
"0",
"10",
"20",
"30",
"40",
"50",
"60",
"Extinguished",
"Fire"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00012_1762755993/figures/2201.00012-Figure2-1.png
|
{
"x1": 53.79800033569336,
"x2": 295.5811462402344,
"y1": 227.41407775878906,
"y2": 265.92193603515625
}
|
200
|
6
| 11
|
Table
|
{
"x1": 107.64,
"x2": 240.12,
"y1": 567,
"y2": 669.24
}
|
Table 6: AIRL hyperparameters in Delivery.
|
[
"Epochs",
"PPO",
"5",
"𝜖-clip",
"0.1",
"𝛾",
"0.999",
"Batch",
"Size",
"PPO",
"4",
"Environment",
"Steps",
"6e6",
"lr-PPO",
"5e-4",
"Batch",
"Size",
"Discriminator",
"512",
"Hyperparameter",
"Value",
"lr-Discriminator",
"5e-5"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00012_1762755993/figures/2201.00012-Table6-1.png
|
{
"x1": 86.34300231933594,
"x2": 261.2048034667969,
"y1": 672.1480712890625,
"y2": 677.7789916992188
}
|
200
|
4
| 11
|
Table
|
{
"x1": 110.52,
"x2": 237.23999999999998,
"y1": 82.8,
"y2": 186.12
}
|
Table 4: Active learning hyperparameters in Emergency.
|
[
"Epochs",
"PPO",
"5",
"𝜖-clip",
"0.1",
"𝛾",
"0.999",
"Entropy",
"Regularization",
"0.25",
"Environment",
"Steps",
"6e6",
"#",
"Queries",
"25",
"Batch",
"Size",
"PPO",
"12",
"Hyperparameter",
"Value",
"lr-PPO",
"3e-4"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00012_1762755993/figures/2201.00012-Table4-1.png
|
{
"x1": 60.895999908447266,
"x2": 286.65081787109375,
"y1": 188.5410919189453,
"y2": 194.1719970703125
}
|
200
|
7
| 11
|
Table
|
{
"x1": 374.76,
"x2": 501.12,
"y1": 175.68,
"y2": 278.28
}
|
Table 7: Active learning hyperparameters in Delivery.
|
[
"Epochs",
"PPO",
"5",
"𝜖-clip",
"0.1",
"𝛾",
"0.999",
"Entropy",
"Regularization",
"0.25",
"Environment",
"Steps",
"8e6",
"#",
"Queries",
"25",
"Batch",
"Size",
"PPO",
"12",
"Hyperparameter",
"Value",
"lr-PPO",
"3e-4"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00012_1762755993/figures/2201.00012-Table7-1.png
|
{
"x1": 330.114990234375,
"x2": 545.7467651367188,
"y1": 281.068115234375,
"y2": 286.6990051269531
}
|
200
|
8
| 11
|
Table
|
{
"x1": 355.68,
"x2": 520.1999999999999,
"y1": 392.76,
"y2": 529.1999999999999
}
|
Table 8: Hyperparameter setup for DRLHP in Delivery.
|
[
"Epochs",
"PPO",
"5",
"𝜖-clip",
"0.1",
"𝛾",
"0.999",
"#",
"Queries",
"5000",
"Batch",
"Size",
"PPO",
"12",
"Batch",
"Size",
"Reward",
"Model",
"12",
"Entropy",
"Regularization",
"1",
"Environment",
"Steps",
"12e6",
"lr-Reward",
"Model",
"3e-5",
"Update",
"Reward",
"Model",
"Frequency",
"50",
"Hyperparameter",
"Value",
"lr-PPO",
"3e-4"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00012_1762755993/figures/2201.00012-Table8-1.png
|
{
"x1": 327.4020080566406,
"x2": 548.4588012695312,
"y1": 531.3291015625,
"y2": 536.9600219726562
}
|
200
|
5
| 11
|
Table
|
{
"x1": 91.8,
"x2": 256.32,
"y1": 264.96,
"y2": 400.32
}
|
Table 5: Hyperparameter setup for DRLHP in Emergency.
|
[
"Epochs",
"PPO",
"5",
"𝜖-clip",
"0.1",
"𝛾",
"0.999",
"#",
"Queries",
"1000",
"Batch",
"Size",
"PPO",
"12",
"Batch",
"Size",
"Reward",
"Model",
"32",
"Entropy",
"Regularization",
"1",
"Environment",
"Steps",
"12e6",
"lr-Reward",
"Model",
"3e-5",
"Update",
"Reward",
"Model",
"Frequency",
"50",
"Hyperparameter",
"Value",
"lr-PPO",
"3e-4"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00012_1762755993/figures/2201.00012-Table5-1.png
|
{
"x1": 58.183998107910156,
"x2": 289.3638000488281,
"y1": 402.6141052246094,
"y2": 408.2449951171875
}
|
200
|
5
| 4
|
Figure
|
{
"x1": 61.199999999999996,
"x2": 549.36,
"y1": 87.84,
"y2": 278.64
}
|
Figure 5: The convex coverage set found by MORAL for three reward dimensions. We plot two-dimensional projections of the attained explicit objectives, with colors indicating the third objective (top three panels). The colors in the bottom three panels show the deviation (8) to the respective preference vector𝑚 used during training. Gray circles around each policy indicate the relative amount of broken vases.
|
[
"0.040",
"0.035",
"0.030",
"0.025",
"0.020",
"0.015",
"0.010",
"0.005",
"0.040",
"0.035",
"0.030",
"0.025",
"0.020",
"0.015",
"0.010",
"0.005",
"0.040",
"0.035",
"0.030",
"0.025",
"0.020",
"0.015",
"0.010",
"0.005",
"6",
"5",
"4",
"3",
"6",
"5",
"4",
"3",
"6",
"5",
"4",
"3",
"n",
"Cl",
"ea",
"7",
"6",
"5",
"4",
"3",
"2",
"3",
"4",
"5",
"6",
"Deliver",
"He",
"lp",
"8",
"7",
"6",
"5",
"4",
"3",
"2",
"3",
"4",
"5",
"6",
"Deliver",
"n",
"Cl",
"ea",
"7",
"6",
"5",
"4",
"3",
"2",
"2",
"3",
"4",
"5",
"6",
"7",
"8",
"Help",
"n",
"Cl",
"ea",
"7",
"6",
"5",
"4",
"3",
"3",
"4",
"5",
"6",
"2",
"He",
"lp",
"8",
"7",
"6",
"5",
"4",
"3",
"2",
"3",
"4",
"5",
"6",
"n",
"Cl",
"ea",
"7",
"6",
"5",
"4",
"3",
"2",
"3",
"4",
"5",
"6",
"7",
"8",
"2"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00012_1762755993/figures/2201.00012-Figure5-1.png
|
{
"x1": 53.79800033569336,
"x2": 558.204345703125,
"y1": 297.4261169433594,
"y2": 335.9329833984375
}
|
200
|
2
| 6
|
Table
|
{
"x1": 65.88,
"x2": 531,
"y1": 157.32,
"y2": 732.6
}
|
Table 2: Evaluation of missing category identification in terms of Recall@K, F1-score@K and MAP. Bi-GTPPP-rand means that our model parameters Ep are initialized from glorot uniform distributions, and Bi-GTPPP-nonstatic means that parameters Ep are initialized via counting users’ visiting history on training data and fine-tune during training. Underlined results indicate the best baselines over each dataset and metric. “*” indicates that the improvement is statistically significant compared with the best baselines at p-value < 0.05 over independent samples t-tests.
|
[
"-",
"Bi-GTPPP-rand",
"outperforms",
"the",
"baseline",
"methods",
"over",
"all",
"evaluation",
"metrics",
"on",
"all",
"two",
"datasets.",
"And",
"Bi-GTPPP-nonstatic",
"further",
"improves",
"the",
"perfor-",
"mance",
"via",
"utilizing",
"priori",
"information",
"and",
"fine-",
"tuning",
"during",
"training.",
"On",
"NYC",
"dataset,",
"the",
"performance",
"improvement",
"of",
"Bi-GTPPP-nonstatic",
"on",
"Recall@1,",
"Recall@5,",
"Recall@10",
"comparing",
"tification.",
"POI",
"check-ins",
"are",
"more",
"dependent",
"on",
"spatio-temporal",
"information",
"(e.g.,",
"it",
"is",
"impossible",
"that",
"two",
"POI",
"check-ins",
"of",
"the",
"same",
"user",
"are",
"far",
"apart",
"in",
"space",
"distance,",
"but",
"the",
"time",
"interval",
"is",
"very",
"short.),",
"but",
"POI",
"category",
"check-ins",
"do",
"not",
"have",
"the",
"obvious",
"spatio-temporal",
"dependence",
"due",
"to",
"POIs",
"at",
"different",
"distances",
"may",
"belong",
"to",
"the",
"same",
"cat-",
"egory.",
"Therefore,",
"the",
"Bi-STDDP",
"can",
"not",
"address",
"the",
"missing",
"POI",
"category",
"identification",
"well.",
"-",
"PACE",
"predicts",
"user",
"preference",
"over",
"POIs,",
"user",
"context",
"and",
"POI",
"context",
"together",
"to",
"achieve",
"further",
"improvement",
"over",
"RNN-based",
"methods.",
"Another",
"great",
"improvement",
"is",
"achieved",
"by",
"STRNN.",
"It",
"in-",
"corporates",
"both",
"the",
"time-specific",
"transition",
"matri-",
"ces",
"and",
"distance-specific",
"transition",
"matrices",
"within",
"recurrent",
"architecture",
"in",
"each",
"layer.",
"Bi-STDDP",
"achieves",
"similar",
"performance",
"improvement",
"over",
"STRNN",
"via",
"incorporating",
"both",
"the",
"bi-directional",
"spatio-temporal",
"dependence",
"and",
"users’",
"dynamic",
"preferences,",
"and",
"they",
"are",
"the",
"best",
"methods",
"among",
"the",
"baselines",
"on",
"the",
"two",
"datasets.",
"However,",
"the",
"missing",
"POI",
"category",
"check-ins",
"identification",
"is",
"a",
"little",
"different",
"from",
"the",
"missing",
"POI",
"check-in",
"iden-",
"over",
"PRME-G",
"because",
"of",
"their",
"sequence",
"modeling",
"capability.",
"Bi-GTPPP-rand",
"0.4412",
"0.7088",
"0.8069",
"0.4412",
"0.2363",
"0.1467",
"0.5652",
"Bi-GTPPP-nonstatic",
"0.4454*",
"0.7211*",
"0.8146*",
"0.4454*",
"0.2404",
"0.1481",
"0.5721*",
"STRNN",
"0.4325",
"0.7071",
"0.8036",
"0.4325",
"0.2357",
"0.1461",
"0.5576",
"PACE",
"0.4251",
"0.6930",
"0.8033",
"0.4251",
"0.2310",
"0.1461",
"0.5568",
"Bi-STDDP",
"0.4283",
"0.7065",
"0.8022",
"0.4283",
"0.2355",
"0.1459",
"0.5571",
"RNN",
"0.4051",
"0.6773",
"0.7808",
"0.4051",
"0.2258",
"0.1420",
"0.5298",
"LSTM",
"0.4203",
"0.6956",
"0.7930",
"0.4203",
"0.2319",
"0.1442",
"0.5458",
"GRU",
"0.4189",
"0.6929",
"0.7912",
"0.4189",
"0.2310",
"0.1439",
"0.5445",
"PRME",
"0.3612",
"0.5265",
"0.6128",
"0.3612",
"0.1755",
"0.1114",
"0.4512",
"PRME-G",
"0.3638",
"0.5358",
"0.6204",
"0.3638",
"0.1786",
"0.1128",
"0.4558",
"TOP1",
"0.3806",
"0.5376",
"0.6385",
"0.3806",
"0.1792",
"0.1161",
"0.4694",
"TOP2",
"0.3967",
"0.6225",
"0.7013",
"0.3967",
"0.2075",
"0.1275",
"0.5109",
"Forward",
"0.3920",
"0.5950",
"0.6899",
"0.3920",
"0.1983",
"0.1254",
"0.4962",
"Backward",
"0.3924",
"0.5973",
"0.6919",
"0.3924",
"0.1991",
"0.1258",
"0.4965",
"TKY",
"Bi-GTPPP-rand",
"0.2425",
"0.5591",
"0.6702",
"0.2425",
"0.1864",
"0.1219",
"0.3898",
"Bi-GTPPP-nonstatic",
"0.2580*",
"0.5745*",
"0.6922*",
"0.2580*",
"0.1915*",
"0.1259*",
"0.4030*",
"STRNN",
"0.2331",
"0.5408",
"0.6689",
"0.2331",
"0.1803",
"0.1216",
"0.3787",
"PACE",
"0.2330",
"0.5401",
"0.6675",
"0.2330",
"0.1800",
"0.1214",
"0.3769",
"Bi-STDDP",
"0.2345",
"0.5409",
"0.6677",
"0.2345",
"0.1803",
"0.1214",
"0.3796",
"RNN",
"0.1873",
"0.4954",
"0.6164",
"0.1873",
"0.1651",
"0.1121",
"0.3292",
"LSTM",
"0.2095",
"0.5191",
"0.6504",
"0.2095",
"0.1730",
"0.1183",
"0.3555",
"GRU",
"0.2162",
"0.5234",
"0.6493",
"0.2162",
"0.1745",
"0.1181",
"0.3596",
"PRME",
"0.1039",
"0.3249",
"0.4442",
"0.1039",
"0.1083",
"0.0808",
"0.2163",
"PRME-G",
"0.1180",
"0.3532",
"0.4818",
"0.1180",
"0.1177",
"0.0876",
"0.2334",
"TOP1",
"0.0594",
"0.2888",
"0.4127",
"0.0594",
"0.0963",
"0.0750",
"0.1756",
"TOP2",
"0.1688",
"0.3842",
"0.5017",
"0.1688",
"0.1281",
"0.0912",
"0.3009",
"Forward",
"0.1576",
"0.3593",
"0.4877",
"0.1576",
"0.1198",
"0.0887",
"0.2636",
"Backward",
"0.1566",
"0.3505",
"0.4868",
"0.1566",
"0.1168",
"0.0885",
"0.2601",
"NYC",
"Recall@1",
"Recall@5",
"Recall@10",
"F1-score@1",
"F1-score@5",
"F1-score@10",
"MAP"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00014_1762756004/figures/2201.00014-Table2-1.png
|
{
"x1": 64.51200103759766,
"x2": 530.76318359375,
"y1": 121.70301818847656,
"y2": 154.8990478515625
}
|
200
|
5
| 9
|
Figure
|
{
"x1": 76.32,
"x2": 518.04,
"y1": 119.88,
"y2": 237.23999999999998
}
|
Figure 5: Performance of Bi-GTPPP with varying embedding dimension, LSTM output space and window width on NYC dataset evaluated by MAP.
|
[
"(c)",
"Impact",
"of",
"Window",
"Width",
"w",
"M",
"AP",
"0.400",
"0.395",
"0.390",
"0.385",
"0.380",
"0",
"2",
"4",
"6",
"8",
"10",
"12",
"14",
"16",
"18",
"20",
"Window",
"Width",
"(b)",
"Impact",
"of",
"LSTM",
"Output",
"Space",
"h",
"M",
"AP",
"0.400",
"0.395",
"0.390",
"0.385",
"0.380",
"0.375",
"0",
"200",
"512",
"800",
"1100",
"1400",
"LSTM",
"Output",
"Space",
"(a)",
"Impact",
"of",
"Embedding",
"Dimension",
"d",
"M",
"AP",
"0.400",
"0.395",
"0.390",
"0.385",
"0.380",
"0",
"128",
"200",
"300",
"400",
"500",
"Embedding",
"Dimension"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00014_1762756004/figures/2201.00014-Figure5-1.png
|
{
"x1": 64.51199340820312,
"x2": 530.7627563476562,
"y1": 254.06605529785156,
"y2": 268.33203125
}
|
200
|
3
| 7
|
Table
|
{
"x1": 320.76,
"x2": 531,
"y1": 483.84,
"y2": 574.1999999999999
}
|
Table 3: Impact of forward and backward sequences on NYC dataset evaluated by Recall@K and MAP.
|
[
"61858",
"to",
"10",
"×",
"250",
"=",
"2500,",
"which",
"is",
"a",
"25",
"times",
"reduction.",
"F-GTPPP",
"0.2394",
"0.5584",
"0.6787",
"0.3866",
"B-GTPPP",
"0.2403",
"0.5590",
"0.6785",
"0.3869",
"Bi-GTPPP",
"0.2580",
"0.5745",
"0.6922",
"0.4030",
"Recall@1",
"Recall@5",
"Recall@10",
"MAP"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00014_1762756004/figures/2201.00014-Table3-1.png
|
{
"x1": 309.593017578125,
"x2": 530.7633666992188,
"y1": 467.4139099121094,
"y2": 481.6808776855469
}
|
200
|
2
| 7
|
Figure
|
{
"x1": 69.48,
"x2": 530.64,
"y1": 113.03999999999999,
"y2": 408.24
}
|
Figure 2: The scatter plot of different methods in term of MAP over five runs.
|
[
"TKY",
"P",
"M",
"A",
"0.58",
"0.56",
"0.54",
"0.52",
"0.50",
"0.48",
"0.46",
"0.44",
"Forward",
"Backward",
"TOP1",
"TOP2",
"PRME",
"PRME-G",
"RNN",
"LSTM",
"GRU",
"STRNN",
"PACE",
"Bi-STDDP",
"Bi-GTPPP-rand",
"Bi-GTPPP-nonstatic",
"FPR",
"NYC",
"P",
"M",
"A",
"0.40",
"0.35",
"0.30",
"0.25",
"0.20",
"0.15",
"Forward",
"Backward",
"TOP1",
"TOP2",
"PRME",
"PRME-G",
"RNN",
"LSTM",
"GRU",
"STRNN",
"PACE",
"Bi-STDDP",
"Bi-GTPPP-rand",
"Bi-GTPPP-nonstatic",
"FPR"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00014_1762756004/figures/2201.00014-Figure2-1.png
|
{
"x1": 173.70399475097656,
"x2": 421.57421875,
"y1": 420.4030456542969,
"y2": 425.20501708984375
}
|
200
|
4
| 8
|
Figure
|
{
"x1": 76.32,
"x2": 518.04,
"y1": 452.52,
"y2": 570.6
}
|
Figure 4: Performance of fine-tuned bi-directional global transition patterns and personal preference on NYC dataset evaluated by Recall@K.
|
[
"(c)",
"Fine-tuned",
"Personal",
"Preference",
"Initialization",
"Fine-tuned",
"ll@",
"K",
"Re",
"ca",
"0.6",
"0.5",
"0.4",
"0.3",
"0.2",
"0.1",
"0.0",
"1",
"5",
"10",
"K",
"(b)",
"Fine-tuned",
"Backward",
"Transition",
"Patterns",
"Initialization",
"Fine-tuned",
"ll@",
"K",
"Re",
"ca",
"0.25",
"0.20",
"0.15",
"0.10",
"0.05",
"0.00",
"1",
"5",
"10",
"K",
"(a)",
"Fine-tuned",
"Forward",
"Transition",
"Patterns",
"Initialization",
"Fine-tuned",
"ll@",
"K",
"Re",
"ca",
"0.25",
"0.20",
"0.15",
"0.10",
"0.05",
"0.00",
"1",
"5",
"10",
"K"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00014_1762756004/figures/2201.00014-Figure4-1.png
|
{
"x1": 71.69900512695312,
"x2": 523.5798950195312,
"y1": 587.508056640625,
"y2": 592.31005859375
}
|
200
|
3
| 8
|
Figure
|
{
"x1": 69.84,
"x2": 528.12,
"y1": 118.8,
"y2": 401.4
}
|
Figure 3: The box plot of different methods in term of MAP over five runs.
|
[
"P",
"M",
"A",
"0.58",
"0.56",
"0.54",
"0.52",
"0.50",
"0.48",
"0.46",
"0.44",
"Forward",
"Backward",
"TOP1",
"TOP2",
"PRME",
"PRME-G",
"RNN",
"LSTM",
"GRU",
"STRNN",
"PACE",
"Bi-STDDP",
"Bi-GTPPP-rand",
"Bi-GTPPP-nonstatic",
"P",
"M",
"A",
"0.40",
"0.35",
"0.30",
"0.25",
"0.20",
"0.15",
"Forward",
"Backward",
"TOP1",
"TOP2",
"PRME",
"PRME-G",
"RNN",
"LSTM",
"GRU",
"STRNN",
"PACE",
"Bi-STDDP",
"Bi-GTPPP-rand",
"Bi-GTPPP-nonstatic"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00014_1762756004/figures/2201.00014-Figure3-1.png
|
{
"x1": 178.1269989013672,
"x2": 417.150390625,
"y1": 416.4150390625,
"y2": 421.2170104980469
}
|
200
|
1
| 4
|
Table
|
{
"x1": 320.76,
"x2": 531,
"y1": 330.47999999999996,
"y2": 462.59999999999997
}
|
Table 1: Statistics of the two datasets.
|
[
"•",
"TKY1",
"(Yang",
"et",
"al.,",
"2015)",
"is",
"a",
"dataset",
"similar",
"to",
"NYC",
"except",
"from",
"Tokyo.",
"•",
"NYC1",
"(Yang",
"et",
"al.,",
"2015)",
"is",
"a",
"dataset",
"from",
"Foursquare,",
"which",
"includes",
"long-term",
"(about",
"10",
"months)",
"check-in",
"data",
"in",
"New",
"York",
"city",
"collected",
"from",
"April",
"2012",
"to",
"February",
"2013.",
"NYC",
"1,083",
"38333",
"251",
"227,428",
"210.0",
"TKY",
"2,293",
"61858",
"247",
"573,703",
"250.2",
"Dataset",
"#user",
"#POI",
"#category",
"#check",
"in",
"#Avg.check-in"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00014_1762756004/figures/2201.00014-Table1-1.png
|
{
"x1": 359.97296142578125,
"x2": 480.3853454589844,
"y1": 324.7790222167969,
"y2": 329.58099365234375
}
|
200
|
1
| 4
|
Figure
|
{
"x1": 134.64,
"x2": 461.15999999999997,
"y1": 112.67999999999999,
"y2": 263.15999999999997
}
|
Figure 1: The proposed Bi-GTPPP model
|
[
"Sum",
"Sequence",
"Features",
"Input",
"Attention",
"Matching",
"Embedding",
"Output",
"Bi-GTPPP",
"Matching",
"Cell",
"Matching",
"Cell",
"Matching",
"Cell",
"Hidden",
"Layer",
"Hidden",
"Layer",
"c(l+w:l+1)c(l-w:l-1)",
"LSTM",
"LSTM",
"Embedding",
"Category",
"c(l+1)",
"Transition",
"Patterns",
"Backward",
"c(l-1)",
"Transition",
"Patterns",
"Forward",
"User",
"u",
"Personalized",
"Preferences",
"Softmax",
"Layer"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00014_1762756004/figures/2201.00014-Figure1-1.png
|
{
"x1": 231.0919952392578,
"x2": 364.1846923828125,
"y1": 277.7680358886719,
"y2": 282.57000732421875
}
|
200
|
1
| 0
|
Figure
|
{
"x1": 315.71999999999997,
"x2": 557.28,
"y1": 236.88,
"y2": 390.24
}
|
Figure 1: The same anomaly from multiple domains. The top part denotes the “Unusual End of Program” anomaly from three domains including BGL, Thunderbird, and Red Storm while the bottom part is the “Program Not Running” from four domains including BGL, Thunderbird, Spirit, and Liberty.
|
[
"Program",
"Not",
"Running",
"kernel:",
"GM:",
"LANai",
"is",
"not",
"running.",
"Allowing",
"port=0",
"open",
"for",
"debugging",
"Liberty:",
"kernel:",
"GM:",
"LANai",
"is",
"not",
"running.",
"Allowing",
"port=0",
"open",
"for",
"debugging",
"Spirit:",
"pbs",
"mom:",
"Bad",
"file",
"descriptor",
"(9)",
"in",
"tm",
"request,",
"job",
"[job]",
"not",
"running",
"Thunderbird:",
"rts",
"panic!",
"-",
"stopping",
"execution",
"BGL:",
"Unusual",
"End",
"of",
"Program",
"DMT",
"310",
"Command",
"Aborted:",
"SCSI",
"cmd:2A",
"LUN",
"2",
"DMT",
"310",
"T:299",
"a:",
"[...]",
"Red",
"Storm:",
"kernel:",
"mptscsih:",
"ioc0:",
"attempting",
"task",
"abort!",
"(sc=00000101bddee480)",
"Thunderbird:",
"rts:",
"kernel",
"terminated",
"for",
"reason",
"1004rts:",
"bad",
"message",
"header:",
"[...]",
"data",
"storage",
"interrupt",
"BGL:"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00016_1762756010/figures/2201.00016-Figure1-1.png
|
{
"x1": 315,
"x2": 557.9985961914062,
"y1": 414.1357421875,
"y2": 459.3880310058594
}
|
200
|
3
| 5
|
Table
|
{
"x1": 340.91999999999996,
"x2": 534.24,
"y1": 53.64,
"y2": 125.28
}
|
Table 3: Transfer result using fine-tuning and adapter based finetuning. Layers is the number of transformer encoder layers. Parameters is the number of trainable parameters in the model.
|
[
"4",
"1M",
"0.998",
"0.996",
"1",
"0.4M",
"0.997",
"0.990",
"Adapter",
"2",
"0.6M",
"0.997",
"0.996",
"4",
"28.5M",
"0.997",
"0.998",
"1",
"7.2M",
"0.998",
"0.997",
"Fine-tuning",
"2",
"14.3M",
"0.998",
"0.997",
"Method",
"Layers",
"Parameters",
"HDFS",
"Thunderbird"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00016_1762756010/figures/2201.00016-Table3-1.png
|
{
"x1": 314.99993896484375,
"x2": 558.0005493164062,
"y1": 136.22274780273438,
"y2": 161.54998779296875
}
|
200
|
6
| 5
|
Figure
|
{
"x1": 59.76,
"x2": 292.32,
"y1": 59.4,
"y2": 230.04
}
|
Figure 6: Loss on the dev set w.r.t training steps. The upper/bottom results are based on parameters pretrained on BGL/Thunderbird, thus BGL/Thunderbird are not shown. All results are using 1-layer Transformer encoder and the same learning rate.
|
[
"BGL",
"Lo",
"ss",
"60",
"40",
"20",
"6.3k",
"steps0",
"HDFS",
"Lo",
"ss",
"25",
"20",
"15",
"10",
"5",
"7.3k",
"steps0",
"Thunderbird",
"Lo",
"ss",
"60",
"40",
"20",
"0",
"20",
"25",
"Lo",
"ss",
"HDFS",
"from",
"scratch",
"fine-tuning",
"15",
"10",
"0",
"5"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00016_1762756010/figures/2201.00016-Figure6-1.png
|
{
"x1": 54,
"x2": 296.9984436035156,
"y1": 258.7427673339844,
"y2": 294.0330505371094
}
|
200
|
7
| 5
|
Figure
|
{
"x1": 352.08,
"x2": 509.03999999999996,
"y1": 396,
"y2": 513.72
}
|
Figure 7: Test performance on BGL (pretrained on Thunderbird) w.r.t.the number of training examples. 5k, 10k, 20k, 50k corresponding to the font 2.5%, 5%, 10%, 25% training data respectively. We show mean and standard deviation across 3 runs for all methods.
|
[
"from",
"scratch",
"fine-tuning",
"adapter",
"co",
"re",
"F1",
"S",
"0.7",
"0.6",
"0.5",
"0.4",
"0.3",
"0.2",
"0.1",
"0.0",
"5k",
"10k",
"20k",
"50k",
"Training",
"Samples"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00016_1762756010/figures/2201.00016-Figure7-1.png
|
{
"x1": 315,
"x2": 557.9986572265625,
"y1": 528.1177368164062,
"y2": 563.407958984375
}
|
200
|
2
| 1
|
Figure
|
{
"x1": 315.71999999999997,
"x2": 557.28,
"y1": 54.72,
"y2": 229.67999999999998
}
|
Figure 2: Logs and Templates. The top part is unstructured logs, we adopt Drain algorithm to extract log templates,then we match each log with its template, which is the middle part. The bottom part is structured inputs.
|
[
"Formatting",
"Structured",
"Inputs",
"Mapping",
"Drain",
"parsing",
"Log",
"templates",
"𝑇3:",
"(root)",
"CMD",
"(<*>",
"<*>)",
"𝐿1",
"𝐿2",
"𝐿3",
"𝐿4",
"𝐿5",
"𝑇2:",
"session",
"opened",
"for",
"user",
"<*>",
"by",
"<*>",
"𝑇1:",
"session",
"closed",
"for",
"user",
"<*>",
"Unstructured",
"Logs",
"𝐿5:",
"session",
"opened",
"for",
"user",
"<*>",
"by",
"<*>",
"𝐿4:",
"session",
"closed",
"for",
"user",
"<*>",
"𝐿3:(root)",
"CMD",
"(<*>",
"<*>)",
"𝐿2:",
"session",
"opened",
"for",
"user",
"<*>",
"by",
"<*>",
"𝐿1:",
"session",
"closed",
"for",
"user",
"<*>",
"𝐿1:",
"TIMES",
"8",
"crond(pam_unix)[2915]:",
"session",
"closed",
"for",
"user",
"root",
"𝐿2:",
"TIMES",
"dn228/dn228",
"crond(pam_unix)[2915]:",
"session",
"opened",
"for",
"user",
"root",
"by",
"(uid=0)",
"𝐿3:",
"TIMES",
"(root)",
"CMD",
"(run-parts",
"/etc/cron.hourly)",
"𝐿4:",
"TIMES",
"session",
"closed",
"for",
"user",
"root",
"𝐿5:",
"TIMES",
"session",
"opened",
"for",
"user",
"root",
"by",
"(uid=0)"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00016_1762756010/figures/2201.00016-Figure2-1.png
|
{
"x1": 315,
"x2": 557.9984130859375,
"y1": 252.90475463867188,
"y2": 288.1940002441406
}
|
200
|
4
| 2
|
Figure
|
{
"x1": 375.47999999999996,
"x2": 495,
"y1": 266.76,
"y2": 384.47999999999996
}
|
Figure 4: Encoder with light adapter. Where N is the number of transformer layers. The left part describes the traditional transformer encoder inserted by adapters, the right part is our light adapter, which is composed of the down- and up-projection layers.
|
[
"+Adapter",
"Up-projection",
"Down-projection",
"×N",
"Transformer",
"layer",
"Layer",
"Norm",
"+",
"Adapter",
"FFN",
"+",
"Attention",
"Multi-headed",
"Adapter",
"Layer",
"Norm"
] |
/home/yz979/palmer_scratch/chengye/SciMolmo-LaTeX-Source-Processing-Toolkit/pdfparser/pdf2latex/2022_new/temp/figures_2201.00016_1762756010/figures/2201.00016-Figure4-1.png
|
{
"x1": 315,
"x2": 558.0046997070312,
"y1": 396.71575927734375,
"y2": 432.00604248046875
}
|
End of preview. Expand
in Data Studio
No dataset card yet
- Downloads last month
- 486