diff --git "a/parse/train/Bkeeca4Kvr/Bkeeca4Kvr_middle.json" "b/parse/train/Bkeeca4Kvr/Bkeeca4Kvr_middle.json"
new file mode 100644--- /dev/null
+++ "b/parse/train/Bkeeca4Kvr/Bkeeca4Kvr_middle.json"
@@ -0,0 +1,51311 @@
+{
+ "pdf_info": [
+ {
+ "preproc_blocks": [
+ {
+ "type": "title",
+ "bbox": [
+ 107,
+ 79,
+ 502,
+ 116
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 79,
+ 505,
+ 99
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 79,
+ 505,
+ 99
+ ],
+ "score": 1.0,
+ "content": "FEW-SHOT LEARNING ON GRAPHS VIA SUPER-",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 106,
+ 101,
+ 473,
+ 118
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 101,
+ 473,
+ 118
+ ],
+ "score": 1.0,
+ "content": "CLASSES BASED ON GRAPH SPECTRAL MEASURES",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ }
+ ],
+ "index": 0.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 113,
+ 135,
+ 320,
+ 147
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 111,
+ 135,
+ 321,
+ 148
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 111,
+ 135,
+ 321,
+ 148
+ ],
+ "score": 1.0,
+ "content": "Jatin Chauhan, Deepak Nathani, Manohar Kaul",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ }
+ ],
+ "index": 2
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 113,
+ 147,
+ 423,
+ 180
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 112,
+ 146,
+ 248,
+ 158
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 112,
+ 146,
+ 248,
+ 158
+ ],
+ "score": 1.0,
+ "content": "Department of Computer Science",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 111,
+ 156,
+ 282,
+ 169
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 111,
+ 156,
+ 282,
+ 169
+ ],
+ "score": 1.0,
+ "content": "Indian Institute of Technology Hyderabad",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 112,
+ 168,
+ 424,
+ 181
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 112,
+ 168,
+ 424,
+ 181
+ ],
+ "score": 1.0,
+ "content": "{chauhanjatin100,deepakn1019,manohar.kaul}@gmail.com",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ }
+ ],
+ "index": 4
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 278,
+ 208,
+ 333,
+ 220
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 276,
+ 208,
+ 335,
+ 221
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 276,
+ 208,
+ 335,
+ 221
+ ],
+ "score": 1.0,
+ "content": "ABSTRACT",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ }
+ ],
+ "index": 6
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 143,
+ 236,
+ 468,
+ 423
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 142,
+ 237,
+ 469,
+ 249
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 142,
+ 237,
+ 469,
+ 249
+ ],
+ "score": 1.0,
+ "content": "We propose to study the problem of few-shot graph classification in graph neu-",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 141,
+ 248,
+ 469,
+ 260
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 248,
+ 469,
+ 260
+ ],
+ "score": 1.0,
+ "content": "ral networks (GNNs) to recognize unseen classes, given limited labeled graph",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 141,
+ 259,
+ 470,
+ 270
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 259,
+ 470,
+ 270
+ ],
+ "score": 1.0,
+ "content": "examples. Despite several interesting GNN variants being proposed recently for",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 141,
+ 270,
+ 469,
+ 281
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 270,
+ 469,
+ 281
+ ],
+ "score": 1.0,
+ "content": "node and graph classification tasks, when faced with scarce labeled examples in",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 141,
+ 280,
+ 469,
+ 292
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 280,
+ 469,
+ 292
+ ],
+ "score": 1.0,
+ "content": "the few-shot setting, these GNNs exhibit significant loss in classification perfor-",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 141,
+ 291,
+ 470,
+ 303
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 291,
+ 470,
+ 303
+ ],
+ "score": 1.0,
+ "content": "mance. Here, we present an approach where a probability measure is assigned",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 141,
+ 302,
+ 470,
+ 314
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 302,
+ 470,
+ 314
+ ],
+ "score": 1.0,
+ "content": "to each graph based on the spectrum of the graph’s normalized Laplacian. This",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 141,
+ 312,
+ 469,
+ 326
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 312,
+ 469,
+ 326
+ ],
+ "score": 1.0,
+ "content": "enables us to accordingly cluster the graph base-labels associated with each graph",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 141,
+ 324,
+ 469,
+ 337
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 324,
+ 257,
+ 337
+ ],
+ "score": 1.0,
+ "content": "into super-classes, where the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 257,
+ 324,
+ 270,
+ 334
+ ],
+ "score": 0.84,
+ "content": "L ^ { p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 270,
+ 324,
+ 469,
+ 337
+ ],
+ "score": 1.0,
+ "content": "Wasserstein distance serves as our underlying dis-",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 141,
+ 335,
+ 470,
+ 348
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 335,
+ 470,
+ 348
+ ],
+ "score": 1.0,
+ "content": "tance metric. Subsequently, a super-graph constructed based on the super-classes",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 141,
+ 346,
+ 469,
+ 358
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 346,
+ 469,
+ 358
+ ],
+ "score": 1.0,
+ "content": "is then fed to our proposed GNN framework which exploits the latent inter-class",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 141,
+ 356,
+ 470,
+ 370
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 356,
+ 470,
+ 370
+ ],
+ "score": 1.0,
+ "content": "relationships made explicit by the super-graph to achieve better class label sep-",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 141,
+ 368,
+ 469,
+ 379
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 368,
+ 469,
+ 379
+ ],
+ "score": 1.0,
+ "content": "aration among the graphs. We conduct exhaustive empirical evaluations of our",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 141,
+ 380,
+ 470,
+ 391
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 380,
+ 470,
+ 391
+ ],
+ "score": 1.0,
+ "content": "proposed method and show that it outperforms both the adaptation of state-of-",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 142,
+ 390,
+ 469,
+ 401
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 142,
+ 390,
+ 469,
+ 401
+ ],
+ "score": 1.0,
+ "content": "the-art graph classification methods to few-shot scenario and our naive baseline",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 142,
+ 401,
+ 469,
+ 412
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 142,
+ 401,
+ 469,
+ 412
+ ],
+ "score": 1.0,
+ "content": "GNNs. Additionally, we also extend and study the behavior of our method to",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 141,
+ 412,
+ 329,
+ 424
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 412,
+ 329,
+ 424
+ ],
+ "score": 1.0,
+ "content": "semi-supervised and active learning scenarios.",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ }
+ ],
+ "index": 15
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 451,
+ 205,
+ 463
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 450,
+ 208,
+ 466
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 450,
+ 208,
+ 466
+ ],
+ "score": 1.0,
+ "content": "1 INTRODUCTION",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ }
+ ],
+ "index": 24
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 478,
+ 505,
+ 599
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 477,
+ 506,
+ 492
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 477,
+ 506,
+ 492
+ ],
+ "score": 1.0,
+ "content": "The need to analyze graph structured data coupled with the ubiquitous nature of graphs (Borgwardt",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 105,
+ 488,
+ 505,
+ 503
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 488,
+ 505,
+ 503
+ ],
+ "score": 1.0,
+ "content": "et al., 2005; Duvenaud et al., 2015; Backstrom & Leskovec, 2010; Chau et al., 2011), has given",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 105,
+ 500,
+ 505,
+ 513
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 500,
+ 505,
+ 513
+ ],
+ "score": 1.0,
+ "content": "greater impetus to research interest in developing graph neural networks (GNNs) (Defferrard et al.,",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 105,
+ 511,
+ 505,
+ 524
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 511,
+ 505,
+ 524
+ ],
+ "score": 1.0,
+ "content": "2016; Kipf & Welling, 2016; Hamilton et al., 2017; Velikovi et al., 2018) for learning tasks on such",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 105,
+ 521,
+ 506,
+ 537
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 521,
+ 506,
+ 537
+ ],
+ "score": 1.0,
+ "content": "graphs. The overarching theme in GNNs is for each node’s feature vector to be generated by passing,",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 105,
+ 533,
+ 506,
+ 547
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 533,
+ 419,
+ 547
+ ],
+ "score": 1.0,
+ "content": "transforming, and recursively aggregating feature information from a given",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 420,
+ 534,
+ 426,
+ 543
+ ],
+ "score": 0.79,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 427,
+ 533,
+ 506,
+ 547
+ ],
+ "score": 1.0,
+ "content": "-hop neighborhood",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 105,
+ 543,
+ 506,
+ 558
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 543,
+ 506,
+ 558
+ ],
+ "score": 1.0,
+ "content": "surrounding the node. However, GNNs still fall short in the ”few-shot” learning setting, where the",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ },
+ {
+ "bbox": [
+ 105,
+ 554,
+ 505,
+ 569
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 554,
+ 505,
+ 569
+ ],
+ "score": 1.0,
+ "content": "classifier must generalize well after seeing abundant base-class samples (while training) and very",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 105,
+ 565,
+ 505,
+ 579
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 565,
+ 505,
+ 579
+ ],
+ "score": 1.0,
+ "content": "few (or even zero) samples from a novel class (while testing). Given the scarcity and difficulty",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ },
+ {
+ "bbox": [
+ 105,
+ 577,
+ 505,
+ 590
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 577,
+ 505,
+ 590
+ ],
+ "score": 1.0,
+ "content": "involved with generation of labeled graph samples, it becomes all the more important to solve the",
+ "type": "text"
+ }
+ ],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 105,
+ 588,
+ 326,
+ 601
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 588,
+ 326,
+ 601
+ ],
+ "score": 1.0,
+ "content": "problem of graph classification in the few-shot setting.",
+ "type": "text"
+ }
+ ],
+ "index": 35
+ }
+ ],
+ "index": 30
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 605,
+ 504,
+ 671
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 604,
+ 506,
+ 617
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 604,
+ 506,
+ 617
+ ],
+ "score": 1.0,
+ "content": "Limitations and challenges: Recent work by Xu et. al. (Xu et al., 2019) indicated that most",
+ "type": "text"
+ }
+ ],
+ "index": 36
+ },
+ {
+ "bbox": [
+ 105,
+ 615,
+ 506,
+ 630
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 615,
+ 506,
+ 630
+ ],
+ "score": 1.0,
+ "content": "recently proposed GNNs were designed based on empirical intuition and heuristic approaches. They",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 105,
+ 626,
+ 506,
+ 640
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 626,
+ 506,
+ 640
+ ],
+ "score": 1.0,
+ "content": "studied the representational power of these GNNs and identified that most neighborhood aggregation",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ },
+ {
+ "bbox": [
+ 106,
+ 638,
+ 505,
+ 651
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 638,
+ 505,
+ 651
+ ],
+ "score": 1.0,
+ "content": "and graph-pooling schemes had diminished discriminative power. They rectified this problem with",
+ "type": "text"
+ }
+ ],
+ "index": 39
+ },
+ {
+ "bbox": [
+ 106,
+ 649,
+ 506,
+ 662
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 649,
+ 506,
+ 662
+ ],
+ "score": 1.0,
+ "content": "the introduction of a novel injective neighborhood aggregation scheme, making it as strong as the",
+ "type": "text"
+ }
+ ],
+ "index": 40
+ },
+ {
+ "bbox": [
+ 106,
+ 660,
+ 424,
+ 672
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 660,
+ 424,
+ 672
+ ],
+ "score": 1.0,
+ "content": "Weisfeiler-Lehman (WL) graph isomorphism test (Weisfeiler & Leman, 1968).",
+ "type": "text"
+ }
+ ],
+ "index": 41
+ }
+ ],
+ "index": 38.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 677,
+ 504,
+ 731
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 676,
+ 505,
+ 690
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 676,
+ 505,
+ 690
+ ],
+ "score": 1.0,
+ "content": "Nevertheless, the problem posed by extremely scarce novel-class samples in the few-shot setting",
+ "type": "text"
+ }
+ ],
+ "index": 42
+ },
+ {
+ "bbox": [
+ 106,
+ 688,
+ 504,
+ 699
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 688,
+ 504,
+ 699
+ ],
+ "score": 1.0,
+ "content": "remains to persist as a formidable challenge, as it requires more rounds of aggregation to affect",
+ "type": "text"
+ }
+ ],
+ "index": 43
+ },
+ {
+ "bbox": [
+ 105,
+ 698,
+ 506,
+ 713
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 698,
+ 506,
+ 713
+ ],
+ "score": 1.0,
+ "content": "larger neighborhoods and hence necessitate greater depth in the GNN. However, when it comes to",
+ "type": "text"
+ }
+ ],
+ "index": 44
+ },
+ {
+ "bbox": [
+ 105,
+ 708,
+ 505,
+ 723
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 708,
+ 505,
+ 723
+ ],
+ "score": 1.0,
+ "content": "GNNs, experimental studies have shown that an increase in the number of layers results in dramatic",
+ "type": "text"
+ }
+ ],
+ "index": 45
+ },
+ {
+ "bbox": [
+ 105,
+ 720,
+ 360,
+ 733
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 720,
+ 360,
+ 733
+ ],
+ "score": 1.0,
+ "content": "performance drops in GNNs (Wu et al., 2019; Li et al., 2018b).",
+ "type": "text"
+ }
+ ],
+ "index": 46
+ }
+ ],
+ "index": 44
+ }
+ ],
+ "page_idx": 0,
+ "page_size": [
+ 612,
+ 792
+ ],
+ "discarded_blocks": [
+ {
+ "type": "discarded",
+ "bbox": [
+ 107,
+ 27,
+ 294,
+ 37
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "score": 1.0,
+ "content": "Published as a conference paper at ICLR 2020",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 303,
+ 751,
+ 308,
+ 760
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 302,
+ 751,
+ 308,
+ 761
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 302,
+ 751,
+ 308,
+ 761
+ ],
+ "score": 1.0,
+ "content": "1",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ }
+ ],
+ "para_blocks": [
+ {
+ "type": "title",
+ "bbox": [
+ 107,
+ 79,
+ 502,
+ 116
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 79,
+ 505,
+ 99
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 79,
+ 505,
+ 99
+ ],
+ "score": 1.0,
+ "content": "FEW-SHOT LEARNING ON GRAPHS VIA SUPER-",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 106,
+ 101,
+ 473,
+ 118
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 101,
+ 473,
+ 118
+ ],
+ "score": 1.0,
+ "content": "CLASSES BASED ON GRAPH SPECTRAL MEASURES",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ }
+ ],
+ "index": 0.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 113,
+ 135,
+ 320,
+ 147
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 111,
+ 135,
+ 321,
+ 148
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 111,
+ 135,
+ 321,
+ 148
+ ],
+ "score": 1.0,
+ "content": "Jatin Chauhan, Deepak Nathani, Manohar Kaul",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ }
+ ],
+ "index": 2,
+ "bbox_fs": [
+ 111,
+ 135,
+ 321,
+ 148
+ ]
+ },
+ {
+ "type": "list",
+ "bbox": [
+ 113,
+ 147,
+ 423,
+ 180
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 112,
+ 146,
+ 248,
+ 158
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 112,
+ 146,
+ 248,
+ 158
+ ],
+ "score": 1.0,
+ "content": "Department of Computer Science",
+ "type": "text"
+ }
+ ],
+ "index": 3,
+ "is_list_start_line": true
+ },
+ {
+ "bbox": [
+ 111,
+ 156,
+ 282,
+ 169
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 111,
+ 156,
+ 282,
+ 169
+ ],
+ "score": 1.0,
+ "content": "Indian Institute of Technology Hyderabad",
+ "type": "text"
+ }
+ ],
+ "index": 4,
+ "is_list_start_line": true
+ },
+ {
+ "bbox": [
+ 112,
+ 168,
+ 424,
+ 181
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 112,
+ 168,
+ 424,
+ 181
+ ],
+ "score": 1.0,
+ "content": "{chauhanjatin100,deepakn1019,manohar.kaul}@gmail.com",
+ "type": "text"
+ }
+ ],
+ "index": 5,
+ "is_list_start_line": true
+ }
+ ],
+ "index": 4,
+ "bbox_fs": [
+ 111,
+ 146,
+ 424,
+ 181
+ ]
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 278,
+ 208,
+ 333,
+ 220
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 276,
+ 208,
+ 335,
+ 221
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 276,
+ 208,
+ 335,
+ 221
+ ],
+ "score": 1.0,
+ "content": "ABSTRACT",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ }
+ ],
+ "index": 6
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 143,
+ 236,
+ 468,
+ 423
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 142,
+ 237,
+ 469,
+ 249
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 142,
+ 237,
+ 469,
+ 249
+ ],
+ "score": 1.0,
+ "content": "We propose to study the problem of few-shot graph classification in graph neu-",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 141,
+ 248,
+ 469,
+ 260
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 248,
+ 469,
+ 260
+ ],
+ "score": 1.0,
+ "content": "ral networks (GNNs) to recognize unseen classes, given limited labeled graph",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 141,
+ 259,
+ 470,
+ 270
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 259,
+ 470,
+ 270
+ ],
+ "score": 1.0,
+ "content": "examples. Despite several interesting GNN variants being proposed recently for",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 141,
+ 270,
+ 469,
+ 281
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 270,
+ 469,
+ 281
+ ],
+ "score": 1.0,
+ "content": "node and graph classification tasks, when faced with scarce labeled examples in",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 141,
+ 280,
+ 469,
+ 292
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 280,
+ 469,
+ 292
+ ],
+ "score": 1.0,
+ "content": "the few-shot setting, these GNNs exhibit significant loss in classification perfor-",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 141,
+ 291,
+ 470,
+ 303
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 291,
+ 470,
+ 303
+ ],
+ "score": 1.0,
+ "content": "mance. Here, we present an approach where a probability measure is assigned",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 141,
+ 302,
+ 470,
+ 314
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 302,
+ 470,
+ 314
+ ],
+ "score": 1.0,
+ "content": "to each graph based on the spectrum of the graph’s normalized Laplacian. This",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 141,
+ 312,
+ 469,
+ 326
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 312,
+ 469,
+ 326
+ ],
+ "score": 1.0,
+ "content": "enables us to accordingly cluster the graph base-labels associated with each graph",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 141,
+ 324,
+ 469,
+ 337
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 324,
+ 257,
+ 337
+ ],
+ "score": 1.0,
+ "content": "into super-classes, where the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 257,
+ 324,
+ 270,
+ 334
+ ],
+ "score": 0.84,
+ "content": "L ^ { p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 270,
+ 324,
+ 469,
+ 337
+ ],
+ "score": 1.0,
+ "content": "Wasserstein distance serves as our underlying dis-",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 141,
+ 335,
+ 470,
+ 348
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 335,
+ 470,
+ 348
+ ],
+ "score": 1.0,
+ "content": "tance metric. Subsequently, a super-graph constructed based on the super-classes",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 141,
+ 346,
+ 469,
+ 358
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 346,
+ 469,
+ 358
+ ],
+ "score": 1.0,
+ "content": "is then fed to our proposed GNN framework which exploits the latent inter-class",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 141,
+ 356,
+ 470,
+ 370
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 356,
+ 470,
+ 370
+ ],
+ "score": 1.0,
+ "content": "relationships made explicit by the super-graph to achieve better class label sep-",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 141,
+ 368,
+ 469,
+ 379
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 368,
+ 469,
+ 379
+ ],
+ "score": 1.0,
+ "content": "aration among the graphs. We conduct exhaustive empirical evaluations of our",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 141,
+ 380,
+ 470,
+ 391
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 380,
+ 470,
+ 391
+ ],
+ "score": 1.0,
+ "content": "proposed method and show that it outperforms both the adaptation of state-of-",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 142,
+ 390,
+ 469,
+ 401
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 142,
+ 390,
+ 469,
+ 401
+ ],
+ "score": 1.0,
+ "content": "the-art graph classification methods to few-shot scenario and our naive baseline",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 142,
+ 401,
+ 469,
+ 412
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 142,
+ 401,
+ 469,
+ 412
+ ],
+ "score": 1.0,
+ "content": "GNNs. Additionally, we also extend and study the behavior of our method to",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 141,
+ 412,
+ 329,
+ 424
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 141,
+ 412,
+ 329,
+ 424
+ ],
+ "score": 1.0,
+ "content": "semi-supervised and active learning scenarios.",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ }
+ ],
+ "index": 15,
+ "bbox_fs": [
+ 141,
+ 237,
+ 470,
+ 424
+ ]
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 451,
+ 205,
+ 463
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 450,
+ 208,
+ 466
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 450,
+ 208,
+ 466
+ ],
+ "score": 1.0,
+ "content": "1 INTRODUCTION",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ }
+ ],
+ "index": 24
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 478,
+ 505,
+ 599
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 477,
+ 506,
+ 492
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 477,
+ 506,
+ 492
+ ],
+ "score": 1.0,
+ "content": "The need to analyze graph structured data coupled with the ubiquitous nature of graphs (Borgwardt",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 105,
+ 488,
+ 505,
+ 503
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 488,
+ 505,
+ 503
+ ],
+ "score": 1.0,
+ "content": "et al., 2005; Duvenaud et al., 2015; Backstrom & Leskovec, 2010; Chau et al., 2011), has given",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 105,
+ 500,
+ 505,
+ 513
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 500,
+ 505,
+ 513
+ ],
+ "score": 1.0,
+ "content": "greater impetus to research interest in developing graph neural networks (GNNs) (Defferrard et al.,",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 105,
+ 511,
+ 505,
+ 524
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 511,
+ 505,
+ 524
+ ],
+ "score": 1.0,
+ "content": "2016; Kipf & Welling, 2016; Hamilton et al., 2017; Velikovi et al., 2018) for learning tasks on such",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 105,
+ 521,
+ 506,
+ 537
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 521,
+ 506,
+ 537
+ ],
+ "score": 1.0,
+ "content": "graphs. The overarching theme in GNNs is for each node’s feature vector to be generated by passing,",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 105,
+ 533,
+ 506,
+ 547
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 533,
+ 419,
+ 547
+ ],
+ "score": 1.0,
+ "content": "transforming, and recursively aggregating feature information from a given",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 420,
+ 534,
+ 426,
+ 543
+ ],
+ "score": 0.79,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 427,
+ 533,
+ 506,
+ 547
+ ],
+ "score": 1.0,
+ "content": "-hop neighborhood",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 105,
+ 543,
+ 506,
+ 558
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 543,
+ 506,
+ 558
+ ],
+ "score": 1.0,
+ "content": "surrounding the node. However, GNNs still fall short in the ”few-shot” learning setting, where the",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ },
+ {
+ "bbox": [
+ 105,
+ 554,
+ 505,
+ 569
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 554,
+ 505,
+ 569
+ ],
+ "score": 1.0,
+ "content": "classifier must generalize well after seeing abundant base-class samples (while training) and very",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 105,
+ 565,
+ 505,
+ 579
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 565,
+ 505,
+ 579
+ ],
+ "score": 1.0,
+ "content": "few (or even zero) samples from a novel class (while testing). Given the scarcity and difficulty",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ },
+ {
+ "bbox": [
+ 105,
+ 577,
+ 505,
+ 590
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 577,
+ 505,
+ 590
+ ],
+ "score": 1.0,
+ "content": "involved with generation of labeled graph samples, it becomes all the more important to solve the",
+ "type": "text"
+ }
+ ],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 105,
+ 588,
+ 326,
+ 601
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 588,
+ 326,
+ 601
+ ],
+ "score": 1.0,
+ "content": "problem of graph classification in the few-shot setting.",
+ "type": "text"
+ }
+ ],
+ "index": 35
+ }
+ ],
+ "index": 30,
+ "bbox_fs": [
+ 105,
+ 477,
+ 506,
+ 601
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 605,
+ 504,
+ 671
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 604,
+ 506,
+ 617
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 604,
+ 506,
+ 617
+ ],
+ "score": 1.0,
+ "content": "Limitations and challenges: Recent work by Xu et. al. (Xu et al., 2019) indicated that most",
+ "type": "text"
+ }
+ ],
+ "index": 36
+ },
+ {
+ "bbox": [
+ 105,
+ 615,
+ 506,
+ 630
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 615,
+ 506,
+ 630
+ ],
+ "score": 1.0,
+ "content": "recently proposed GNNs were designed based on empirical intuition and heuristic approaches. They",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 105,
+ 626,
+ 506,
+ 640
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 626,
+ 506,
+ 640
+ ],
+ "score": 1.0,
+ "content": "studied the representational power of these GNNs and identified that most neighborhood aggregation",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ },
+ {
+ "bbox": [
+ 106,
+ 638,
+ 505,
+ 651
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 638,
+ 505,
+ 651
+ ],
+ "score": 1.0,
+ "content": "and graph-pooling schemes had diminished discriminative power. They rectified this problem with",
+ "type": "text"
+ }
+ ],
+ "index": 39
+ },
+ {
+ "bbox": [
+ 106,
+ 649,
+ 506,
+ 662
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 649,
+ 506,
+ 662
+ ],
+ "score": 1.0,
+ "content": "the introduction of a novel injective neighborhood aggregation scheme, making it as strong as the",
+ "type": "text"
+ }
+ ],
+ "index": 40
+ },
+ {
+ "bbox": [
+ 106,
+ 660,
+ 424,
+ 672
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 660,
+ 424,
+ 672
+ ],
+ "score": 1.0,
+ "content": "Weisfeiler-Lehman (WL) graph isomorphism test (Weisfeiler & Leman, 1968).",
+ "type": "text"
+ }
+ ],
+ "index": 41
+ }
+ ],
+ "index": 38.5,
+ "bbox_fs": [
+ 105,
+ 604,
+ 506,
+ 672
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 677,
+ 504,
+ 731
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 676,
+ 505,
+ 690
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 676,
+ 505,
+ 690
+ ],
+ "score": 1.0,
+ "content": "Nevertheless, the problem posed by extremely scarce novel-class samples in the few-shot setting",
+ "type": "text"
+ }
+ ],
+ "index": 42
+ },
+ {
+ "bbox": [
+ 106,
+ 688,
+ 504,
+ 699
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 688,
+ 504,
+ 699
+ ],
+ "score": 1.0,
+ "content": "remains to persist as a formidable challenge, as it requires more rounds of aggregation to affect",
+ "type": "text"
+ }
+ ],
+ "index": 43
+ },
+ {
+ "bbox": [
+ 105,
+ 698,
+ 506,
+ 713
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 698,
+ 506,
+ 713
+ ],
+ "score": 1.0,
+ "content": "larger neighborhoods and hence necessitate greater depth in the GNN. However, when it comes to",
+ "type": "text"
+ }
+ ],
+ "index": 44
+ },
+ {
+ "bbox": [
+ 105,
+ 708,
+ 505,
+ 723
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 708,
+ 505,
+ 723
+ ],
+ "score": 1.0,
+ "content": "GNNs, experimental studies have shown that an increase in the number of layers results in dramatic",
+ "type": "text"
+ }
+ ],
+ "index": 45
+ },
+ {
+ "bbox": [
+ 105,
+ 720,
+ 360,
+ 733
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 720,
+ 360,
+ 733
+ ],
+ "score": 1.0,
+ "content": "performance drops in GNNs (Wu et al., 2019; Li et al., 2018b).",
+ "type": "text"
+ }
+ ],
+ "index": 46
+ }
+ ],
+ "index": 44,
+ "bbox_fs": [
+ 105,
+ 676,
+ 506,
+ 733
+ ]
+ }
+ ]
+ },
+ {
+ "preproc_blocks": [
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 82,
+ 505,
+ 182
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 82,
+ 505,
+ 95
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 82,
+ 505,
+ 95
+ ],
+ "score": 1.0,
+ "content": "Our work: Motivated by the aforementioned observations and challenges, our method does the",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 106,
+ 94,
+ 505,
+ 106
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 94,
+ 505,
+ 106
+ ],
+ "score": 1.0,
+ "content": "following. We begin with a once-off preprocessing step. We assign a probability measure to each",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 105,
+ 104,
+ 506,
+ 118
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 104,
+ 506,
+ 118
+ ],
+ "score": 1.0,
+ "content": "graph, which we refer to as a graph spectral measure (similar to (Gu et al., 2015)), based on the",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ },
+ {
+ "bbox": [
+ 105,
+ 115,
+ 506,
+ 128
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 115,
+ 506,
+ 128
+ ],
+ "score": 1.0,
+ "content": "spectrum of the graph’s normalized Laplacian matrix representation. Given this metric space of",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 105,
+ 126,
+ 506,
+ 139
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 126,
+ 349,
+ 139
+ ],
+ "score": 1.0,
+ "content": "graph spectral measures and the underlying distance as the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 349,
+ 127,
+ 362,
+ 137
+ ],
+ "score": 0.84,
+ "content": "L ^ { p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 362,
+ 126,
+ 506,
+ 139
+ ],
+ "score": 1.0,
+ "content": "Wasserstein distance, we compute",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 106,
+ 137,
+ 505,
+ 150
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 137,
+ 505,
+ 150
+ ],
+ "score": 1.0,
+ "content": "Wasserstein barycenters (Agueh & Carlier, 2011) for each set of graphs specific to a base class and",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 105,
+ 148,
+ 505,
+ 162
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 148,
+ 505,
+ 162
+ ],
+ "score": 1.0,
+ "content": "term these barycenters as prototype graphs. With this set of prototype graphs for each base class",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 105,
+ 159,
+ 505,
+ 172
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 159,
+ 505,
+ 172
+ ],
+ "score": 1.0,
+ "content": "label, we cluster the spectral measures associated with each prototype graph in Wasserstein space to",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 105,
+ 172,
+ 210,
+ 182
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 172,
+ 210,
+ 182
+ ],
+ "score": 1.0,
+ "content": "create a super-class label.",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ }
+ ],
+ "index": 4
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 187,
+ 505,
+ 242
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 186,
+ 505,
+ 200
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 186,
+ 505,
+ 200
+ ],
+ "score": 1.0,
+ "content": "Utilizing this super-class information, we then build a graph of graphs called a super-graph.",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 105,
+ 198,
+ 505,
+ 210
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 198,
+ 505,
+ 210
+ ],
+ "score": 1.0,
+ "content": "The intuition behind this is to exploit the non-explicit and latent inter-class relationships between",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 104,
+ 208,
+ 506,
+ 222
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 208,
+ 506,
+ 222
+ ],
+ "score": 1.0,
+ "content": "graphs via their spectral measures and use a GNN on this to also introduce a relational inductive",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 105,
+ 219,
+ 505,
+ 233
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 219,
+ 505,
+ 233
+ ],
+ "score": 1.0,
+ "content": "bias (Battaglia et al., 2018), which in turn affords us an improved sample complexity and hence",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 105,
+ 230,
+ 402,
+ 243
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 230,
+ 402,
+ 243
+ ],
+ "score": 1.0,
+ "content": "better combinatorial generalization given such few samples to begin with.",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ }
+ ],
+ "index": 11
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 248,
+ 505,
+ 357
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 247,
+ 506,
+ 261
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 247,
+ 506,
+ 261
+ ],
+ "score": 1.0,
+ "content": "Given, the super-classes and the super-graph, we train our proposed GNN model for few-shot learn-",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 104,
+ 258,
+ 507,
+ 272
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 258,
+ 507,
+ 272
+ ],
+ "score": 1.0,
+ "content": "ing on graphs. Our GNN consists of a graph isomorphism network (GIN) Xu et al. (2019) as a",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 105,
+ 270,
+ 505,
+ 282
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 270,
+ 178,
+ 282
+ ],
+ "score": 1.0,
+ "content": "feature extractor",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 178,
+ 270,
+ 201,
+ 282
+ ],
+ "score": 0.92,
+ "content": "F _ { \\theta } ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 201,
+ 270,
+ 505,
+ 282
+ ],
+ "score": 1.0,
+ "content": "to generate graph embeddings; on which subsequently acts our classifier",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 107,
+ 279,
+ 507,
+ 294
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 281,
+ 126,
+ 293
+ ],
+ "score": 0.89,
+ "content": "C ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 126,
+ 279,
+ 270,
+ 294
+ ],
+ "score": 1.0,
+ "content": "comprising of two components: (i)",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 270,
+ 281,
+ 291,
+ 291
+ ],
+ "score": 0.88,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 292,
+ 279,
+ 507,
+ 294
+ ],
+ "score": 1.0,
+ "content": ": a MLP layer to learn and predict the super class as-",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 104,
+ 289,
+ 507,
+ 306
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 289,
+ 216,
+ 306
+ ],
+ "score": 1.0,
+ "content": "sociated to a graph, and (ii)",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 217,
+ 291,
+ 244,
+ 302
+ ],
+ "score": 0.9,
+ "content": "\\Dot { C } ^ { G A T }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 244,
+ 289,
+ 507,
+ 306
+ ],
+ "score": 1.0,
+ "content": ": a graph attention network (GAT) to predict the actual class label",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 104,
+ 302,
+ 506,
+ 316
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 302,
+ 464,
+ 316
+ ],
+ "score": 1.0,
+ "content": "of a graph. The overall loss function is a sum of the cross-entropy losses associated with",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 465,
+ 303,
+ 487,
+ 313
+ ],
+ "score": 0.88,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 487,
+ 302,
+ 506,
+ 316
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 107,
+ 310,
+ 506,
+ 329
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 313,
+ 134,
+ 324
+ ],
+ "score": 0.88,
+ "content": "C ^ { G A \\breve { T } }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 134,
+ 310,
+ 506,
+ 329
+ ],
+ "score": 1.0,
+ "content": ". We follow initialization based strategy (Chen et al., 2019), with a training and fine-tuning",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 105,
+ 324,
+ 504,
+ 336
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 324,
+ 439,
+ 336
+ ],
+ "score": 1.0,
+ "content": "phase, so that in the fine-tuning phase, the pre-trained parameters associated with",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 440,
+ 325,
+ 462,
+ 336
+ ],
+ "score": 0.92,
+ "content": "F _ { \\theta } ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 462,
+ 324,
+ 482,
+ 336
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 482,
+ 325,
+ 504,
+ 335
+ ],
+ "score": 0.88,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 106,
+ 336,
+ 505,
+ 348
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 336,
+ 505,
+ 348
+ ],
+ "score": 1.0,
+ "content": "are frozen, and the few novel labeled graph samples are used to update the weights and attention",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 104,
+ 344,
+ 182,
+ 359
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 344,
+ 182,
+ 359
+ ],
+ "score": 1.0,
+ "content": "learned by CGAT .",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ }
+ ],
+ "index": 18.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 363,
+ 505,
+ 441
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 362,
+ 505,
+ 377
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 362,
+ 505,
+ 377
+ ],
+ "score": 1.0,
+ "content": "Our contributions: To the best of our knowledge, we are the first to introduce few shot learning",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 105,
+ 375,
+ 505,
+ 387
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 375,
+ 505,
+ 387
+ ],
+ "score": 1.0,
+ "content": "on graphs for graph classification. Next, we propose an architecture that makes use of the graph’s",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 105,
+ 386,
+ 506,
+ 398
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 386,
+ 506,
+ 398
+ ],
+ "score": 1.0,
+ "content": "spectral measures to generate a set of super-classes and a super-graph to better model the latent",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 105,
+ 397,
+ 505,
+ 408
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 397,
+ 505,
+ 408
+ ],
+ "score": 1.0,
+ "content": "relations between classes, followed by our GNN trained using an initialization method. Finally, we",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 105,
+ 407,
+ 505,
+ 420
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 407,
+ 505,
+ 420
+ ],
+ "score": 1.0,
+ "content": "conduct extensive experiments to gain insight into our method. For example, in the 20-shot setting",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 105,
+ 417,
+ 504,
+ 431
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 417,
+ 450,
+ 431
+ ],
+ "score": 1.0,
+ "content": "on the TRIANGLES dataset, our method shows a substantial improvement of nearly",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 450,
+ 418,
+ 465,
+ 429
+ ],
+ "score": 0.86,
+ "content": "7 \\%",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 465,
+ 417,
+ 484,
+ 431
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 484,
+ 418,
+ 504,
+ 429
+ ],
+ "score": 0.87,
+ "content": "2 0 \\%",
+ "type": "inline_equation"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 105,
+ 430,
+ 333,
+ 442
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 430,
+ 333,
+ 442
+ ],
+ "score": 1.0,
+ "content": "over DL-based and unsupervised baselines, respectively.",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ }
+ ],
+ "index": 27
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 471,
+ 210,
+ 484
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 104,
+ 470,
+ 213,
+ 487
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 470,
+ 213,
+ 487
+ ],
+ "score": 1.0,
+ "content": "2 RELATED WORK",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ }
+ ],
+ "index": 31
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 506,
+ 505,
+ 627
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 506,
+ 505,
+ 519
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 506,
+ 505,
+ 519
+ ],
+ "score": 1.0,
+ "content": "Few-shot learning in the computer vision community was first introduced by (Fei-Fei et al., 2006)",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 105,
+ 518,
+ 505,
+ 530
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 518,
+ 505,
+ 530
+ ],
+ "score": 1.0,
+ "content": "with the intuition that learning the underlying properties of the base classes given abundant samples",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ },
+ {
+ "bbox": [
+ 105,
+ 527,
+ 505,
+ 542
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 527,
+ 505,
+ 542
+ ],
+ "score": 1.0,
+ "content": "can help generalize better to unseen classes with few-labeled samples available. Various learning",
+ "type": "text"
+ }
+ ],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 105,
+ 540,
+ 505,
+ 552
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 540,
+ 505,
+ 552
+ ],
+ "score": 1.0,
+ "content": "algorithms have been proposed in the image domain, among which a broad category of initializa-",
+ "type": "text"
+ }
+ ],
+ "index": 35
+ },
+ {
+ "bbox": [
+ 105,
+ 549,
+ 505,
+ 563
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 549,
+ 505,
+ 563
+ ],
+ "score": 1.0,
+ "content": "tion based methods aim to learn transferable knowledge from training classes, so that the model",
+ "type": "text"
+ }
+ ],
+ "index": 36
+ },
+ {
+ "bbox": [
+ 105,
+ 562,
+ 505,
+ 573
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 562,
+ 505,
+ 573
+ ],
+ "score": 1.0,
+ "content": "can be adapted to unseen classes with limited labeled examples (Finn et al., 2017); (Rusu et al.,",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 105,
+ 572,
+ 505,
+ 585
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 572,
+ 505,
+ 585
+ ],
+ "score": 1.0,
+ "content": "2018); (Nichol et al., 2018). Recently proposed and widely accepted Initialization based methods",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ },
+ {
+ "bbox": [
+ 105,
+ 583,
+ 506,
+ 595
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 583,
+ 506,
+ 595
+ ],
+ "score": 1.0,
+ "content": "can broadly be classified into: (i) methods that learn good model parameters with limited labeled",
+ "type": "text"
+ }
+ ],
+ "index": 39
+ },
+ {
+ "bbox": [
+ 105,
+ 594,
+ 504,
+ 606
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 594,
+ 504,
+ 606
+ ],
+ "score": 1.0,
+ "content": "examples and a small number of gradient update steps (Finn et al., 2017) and (ii) methods that learn",
+ "type": "text"
+ }
+ ],
+ "index": 40
+ },
+ {
+ "bbox": [
+ 105,
+ 605,
+ 506,
+ 617
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 605,
+ 506,
+ 617
+ ],
+ "score": 1.0,
+ "content": "an optimizer (Ravi & Larochelle, 2017). We refer the interested reader to Chen et. al. (Chen et al.,",
+ "type": "text"
+ }
+ ],
+ "index": 41
+ },
+ {
+ "bbox": [
+ 105,
+ 615,
+ 368,
+ 628
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 615,
+ 368,
+ 628
+ ],
+ "score": 1.0,
+ "content": "2019) for more examples of few-shot learning methods in vision.",
+ "type": "text"
+ }
+ ],
+ "index": 42
+ }
+ ],
+ "index": 37
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 632,
+ 505,
+ 731
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 633,
+ 505,
+ 645
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 633,
+ 505,
+ 645
+ ],
+ "score": 1.0,
+ "content": "Graph neural networks (GNNs) were first introduced in (Gori et al., 2005); (Scarselli et al., 2009)",
+ "type": "text"
+ }
+ ],
+ "index": 43
+ },
+ {
+ "bbox": [
+ 105,
+ 644,
+ 506,
+ 656
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 644,
+ 506,
+ 656
+ ],
+ "score": 1.0,
+ "content": "as recurrent message passing algorithms. Subsequent work (Bruna et al., 2014); (Henaff et al.,",
+ "type": "text"
+ }
+ ],
+ "index": 44
+ },
+ {
+ "bbox": [
+ 105,
+ 655,
+ 506,
+ 668
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 655,
+ 506,
+ 668
+ ],
+ "score": 1.0,
+ "content": "2015) proposed to learn smooth spectral multipliers of the graph Laplacian, but incurred higher",
+ "type": "text"
+ }
+ ],
+ "index": 45
+ },
+ {
+ "bbox": [
+ 105,
+ 666,
+ 506,
+ 678
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 666,
+ 506,
+ 678
+ ],
+ "score": 1.0,
+ "content": "computational cost. This computational bottleneck was later resolved (Defferrard et al., 2016); (Kipf",
+ "type": "text"
+ }
+ ],
+ "index": 46
+ },
+ {
+ "bbox": [
+ 105,
+ 677,
+ 506,
+ 690
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 677,
+ 506,
+ 690
+ ],
+ "score": 1.0,
+ "content": "& Welling, 2016) by learning polynomials of the graph Laplacian. GNNs are a natural extension to",
+ "type": "text"
+ }
+ ],
+ "index": 47
+ },
+ {
+ "bbox": [
+ 106,
+ 688,
+ 505,
+ 700
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 688,
+ 505,
+ 700
+ ],
+ "score": 1.0,
+ "content": "Convolutional neural networks (CNNs) on non-Euclidean data. Recent work (Velikovi et al., 2018)",
+ "type": "text"
+ }
+ ],
+ "index": 48
+ },
+ {
+ "bbox": [
+ 105,
+ 698,
+ 506,
+ 712
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 698,
+ 506,
+ 712
+ ],
+ "score": 1.0,
+ "content": "introduced the concept of self-attention in GNNs, which allows each node to provide attention to",
+ "type": "text"
+ }
+ ],
+ "index": 49
+ },
+ {
+ "bbox": [
+ 105,
+ 709,
+ 506,
+ 723
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 709,
+ 506,
+ 723
+ ],
+ "score": 1.0,
+ "content": "the enclosing neighborhood resulting in improved learning. We refer the reader to (Bronstein et al.,",
+ "type": "text"
+ }
+ ],
+ "index": 50
+ },
+ {
+ "bbox": [
+ 106,
+ 720,
+ 272,
+ 732
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 720,
+ 272,
+ 732
+ ],
+ "score": 1.0,
+ "content": "2016) for detailed information on GNNs.",
+ "type": "text"
+ }
+ ],
+ "index": 51
+ }
+ ],
+ "index": 47
+ }
+ ],
+ "page_idx": 1,
+ "page_size": [
+ 612,
+ 792
+ ],
+ "discarded_blocks": [
+ {
+ "type": "discarded",
+ "bbox": [
+ 108,
+ 27,
+ 293,
+ 37
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "score": 1.0,
+ "content": "Published as a conference paper at ICLR 2020",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 302,
+ 751,
+ 309,
+ 760
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 301,
+ 750,
+ 310,
+ 763
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 301,
+ 750,
+ 310,
+ 763
+ ],
+ "score": 1.0,
+ "content": "2",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ }
+ ],
+ "para_blocks": [
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 82,
+ 505,
+ 182
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 82,
+ 505,
+ 95
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 82,
+ 505,
+ 95
+ ],
+ "score": 1.0,
+ "content": "Our work: Motivated by the aforementioned observations and challenges, our method does the",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 106,
+ 94,
+ 505,
+ 106
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 94,
+ 505,
+ 106
+ ],
+ "score": 1.0,
+ "content": "following. We begin with a once-off preprocessing step. We assign a probability measure to each",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 105,
+ 104,
+ 506,
+ 118
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 104,
+ 506,
+ 118
+ ],
+ "score": 1.0,
+ "content": "graph, which we refer to as a graph spectral measure (similar to (Gu et al., 2015)), based on the",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ },
+ {
+ "bbox": [
+ 105,
+ 115,
+ 506,
+ 128
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 115,
+ 506,
+ 128
+ ],
+ "score": 1.0,
+ "content": "spectrum of the graph’s normalized Laplacian matrix representation. Given this metric space of",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 105,
+ 126,
+ 506,
+ 139
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 126,
+ 349,
+ 139
+ ],
+ "score": 1.0,
+ "content": "graph spectral measures and the underlying distance as the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 349,
+ 127,
+ 362,
+ 137
+ ],
+ "score": 0.84,
+ "content": "L ^ { p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 362,
+ 126,
+ 506,
+ 139
+ ],
+ "score": 1.0,
+ "content": "Wasserstein distance, we compute",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 106,
+ 137,
+ 505,
+ 150
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 137,
+ 505,
+ 150
+ ],
+ "score": 1.0,
+ "content": "Wasserstein barycenters (Agueh & Carlier, 2011) for each set of graphs specific to a base class and",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 105,
+ 148,
+ 505,
+ 162
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 148,
+ 505,
+ 162
+ ],
+ "score": 1.0,
+ "content": "term these barycenters as prototype graphs. With this set of prototype graphs for each base class",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 105,
+ 159,
+ 505,
+ 172
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 159,
+ 505,
+ 172
+ ],
+ "score": 1.0,
+ "content": "label, we cluster the spectral measures associated with each prototype graph in Wasserstein space to",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 105,
+ 172,
+ 210,
+ 182
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 172,
+ 210,
+ 182
+ ],
+ "score": 1.0,
+ "content": "create a super-class label.",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ }
+ ],
+ "index": 4,
+ "bbox_fs": [
+ 105,
+ 82,
+ 506,
+ 182
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 187,
+ 505,
+ 242
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 186,
+ 505,
+ 200
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 186,
+ 505,
+ 200
+ ],
+ "score": 1.0,
+ "content": "Utilizing this super-class information, we then build a graph of graphs called a super-graph.",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 105,
+ 198,
+ 505,
+ 210
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 198,
+ 505,
+ 210
+ ],
+ "score": 1.0,
+ "content": "The intuition behind this is to exploit the non-explicit and latent inter-class relationships between",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 104,
+ 208,
+ 506,
+ 222
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 208,
+ 506,
+ 222
+ ],
+ "score": 1.0,
+ "content": "graphs via their spectral measures and use a GNN on this to also introduce a relational inductive",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 105,
+ 219,
+ 505,
+ 233
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 219,
+ 505,
+ 233
+ ],
+ "score": 1.0,
+ "content": "bias (Battaglia et al., 2018), which in turn affords us an improved sample complexity and hence",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 105,
+ 230,
+ 402,
+ 243
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 230,
+ 402,
+ 243
+ ],
+ "score": 1.0,
+ "content": "better combinatorial generalization given such few samples to begin with.",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ }
+ ],
+ "index": 11,
+ "bbox_fs": [
+ 104,
+ 186,
+ 506,
+ 243
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 248,
+ 505,
+ 357
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 247,
+ 506,
+ 261
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 247,
+ 506,
+ 261
+ ],
+ "score": 1.0,
+ "content": "Given, the super-classes and the super-graph, we train our proposed GNN model for few-shot learn-",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 104,
+ 258,
+ 507,
+ 272
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 258,
+ 507,
+ 272
+ ],
+ "score": 1.0,
+ "content": "ing on graphs. Our GNN consists of a graph isomorphism network (GIN) Xu et al. (2019) as a",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 105,
+ 270,
+ 505,
+ 282
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 270,
+ 178,
+ 282
+ ],
+ "score": 1.0,
+ "content": "feature extractor",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 178,
+ 270,
+ 201,
+ 282
+ ],
+ "score": 0.92,
+ "content": "F _ { \\theta } ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 201,
+ 270,
+ 505,
+ 282
+ ],
+ "score": 1.0,
+ "content": "to generate graph embeddings; on which subsequently acts our classifier",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 107,
+ 279,
+ 507,
+ 294
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 281,
+ 126,
+ 293
+ ],
+ "score": 0.89,
+ "content": "C ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 126,
+ 279,
+ 270,
+ 294
+ ],
+ "score": 1.0,
+ "content": "comprising of two components: (i)",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 270,
+ 281,
+ 291,
+ 291
+ ],
+ "score": 0.88,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 292,
+ 279,
+ 507,
+ 294
+ ],
+ "score": 1.0,
+ "content": ": a MLP layer to learn and predict the super class as-",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 104,
+ 289,
+ 507,
+ 306
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 289,
+ 216,
+ 306
+ ],
+ "score": 1.0,
+ "content": "sociated to a graph, and (ii)",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 217,
+ 291,
+ 244,
+ 302
+ ],
+ "score": 0.9,
+ "content": "\\Dot { C } ^ { G A T }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 244,
+ 289,
+ 507,
+ 306
+ ],
+ "score": 1.0,
+ "content": ": a graph attention network (GAT) to predict the actual class label",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 104,
+ 302,
+ 506,
+ 316
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 302,
+ 464,
+ 316
+ ],
+ "score": 1.0,
+ "content": "of a graph. The overall loss function is a sum of the cross-entropy losses associated with",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 465,
+ 303,
+ 487,
+ 313
+ ],
+ "score": 0.88,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 487,
+ 302,
+ 506,
+ 316
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 107,
+ 310,
+ 506,
+ 329
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 313,
+ 134,
+ 324
+ ],
+ "score": 0.88,
+ "content": "C ^ { G A \\breve { T } }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 134,
+ 310,
+ 506,
+ 329
+ ],
+ "score": 1.0,
+ "content": ". We follow initialization based strategy (Chen et al., 2019), with a training and fine-tuning",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 105,
+ 324,
+ 504,
+ 336
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 324,
+ 439,
+ 336
+ ],
+ "score": 1.0,
+ "content": "phase, so that in the fine-tuning phase, the pre-trained parameters associated with",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 440,
+ 325,
+ 462,
+ 336
+ ],
+ "score": 0.92,
+ "content": "F _ { \\theta } ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 462,
+ 324,
+ 482,
+ 336
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 482,
+ 325,
+ 504,
+ 335
+ ],
+ "score": 0.88,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 106,
+ 336,
+ 505,
+ 348
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 336,
+ 505,
+ 348
+ ],
+ "score": 1.0,
+ "content": "are frozen, and the few novel labeled graph samples are used to update the weights and attention",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 104,
+ 344,
+ 182,
+ 359
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 344,
+ 182,
+ 359
+ ],
+ "score": 1.0,
+ "content": "learned by CGAT .",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ }
+ ],
+ "index": 18.5,
+ "bbox_fs": [
+ 104,
+ 247,
+ 507,
+ 359
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 363,
+ 505,
+ 441
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 362,
+ 505,
+ 377
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 362,
+ 505,
+ 377
+ ],
+ "score": 1.0,
+ "content": "Our contributions: To the best of our knowledge, we are the first to introduce few shot learning",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 105,
+ 375,
+ 505,
+ 387
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 375,
+ 505,
+ 387
+ ],
+ "score": 1.0,
+ "content": "on graphs for graph classification. Next, we propose an architecture that makes use of the graph’s",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 105,
+ 386,
+ 506,
+ 398
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 386,
+ 506,
+ 398
+ ],
+ "score": 1.0,
+ "content": "spectral measures to generate a set of super-classes and a super-graph to better model the latent",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 105,
+ 397,
+ 505,
+ 408
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 397,
+ 505,
+ 408
+ ],
+ "score": 1.0,
+ "content": "relations between classes, followed by our GNN trained using an initialization method. Finally, we",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 105,
+ 407,
+ 505,
+ 420
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 407,
+ 505,
+ 420
+ ],
+ "score": 1.0,
+ "content": "conduct extensive experiments to gain insight into our method. For example, in the 20-shot setting",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 105,
+ 417,
+ 504,
+ 431
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 417,
+ 450,
+ 431
+ ],
+ "score": 1.0,
+ "content": "on the TRIANGLES dataset, our method shows a substantial improvement of nearly",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 450,
+ 418,
+ 465,
+ 429
+ ],
+ "score": 0.86,
+ "content": "7 \\%",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 465,
+ 417,
+ 484,
+ 431
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 484,
+ 418,
+ 504,
+ 429
+ ],
+ "score": 0.87,
+ "content": "2 0 \\%",
+ "type": "inline_equation"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 105,
+ 430,
+ 333,
+ 442
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 430,
+ 333,
+ 442
+ ],
+ "score": 1.0,
+ "content": "over DL-based and unsupervised baselines, respectively.",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ }
+ ],
+ "index": 27,
+ "bbox_fs": [
+ 105,
+ 362,
+ 506,
+ 442
+ ]
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 471,
+ 210,
+ 484
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 104,
+ 470,
+ 213,
+ 487
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 470,
+ 213,
+ 487
+ ],
+ "score": 1.0,
+ "content": "2 RELATED WORK",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ }
+ ],
+ "index": 31
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 506,
+ 505,
+ 627
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 506,
+ 505,
+ 519
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 506,
+ 505,
+ 519
+ ],
+ "score": 1.0,
+ "content": "Few-shot learning in the computer vision community was first introduced by (Fei-Fei et al., 2006)",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 105,
+ 518,
+ 505,
+ 530
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 518,
+ 505,
+ 530
+ ],
+ "score": 1.0,
+ "content": "with the intuition that learning the underlying properties of the base classes given abundant samples",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ },
+ {
+ "bbox": [
+ 105,
+ 527,
+ 505,
+ 542
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 527,
+ 505,
+ 542
+ ],
+ "score": 1.0,
+ "content": "can help generalize better to unseen classes with few-labeled samples available. Various learning",
+ "type": "text"
+ }
+ ],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 105,
+ 540,
+ 505,
+ 552
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 540,
+ 505,
+ 552
+ ],
+ "score": 1.0,
+ "content": "algorithms have been proposed in the image domain, among which a broad category of initializa-",
+ "type": "text"
+ }
+ ],
+ "index": 35
+ },
+ {
+ "bbox": [
+ 105,
+ 549,
+ 505,
+ 563
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 549,
+ 505,
+ 563
+ ],
+ "score": 1.0,
+ "content": "tion based methods aim to learn transferable knowledge from training classes, so that the model",
+ "type": "text"
+ }
+ ],
+ "index": 36
+ },
+ {
+ "bbox": [
+ 105,
+ 562,
+ 505,
+ 573
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 562,
+ 505,
+ 573
+ ],
+ "score": 1.0,
+ "content": "can be adapted to unseen classes with limited labeled examples (Finn et al., 2017); (Rusu et al.,",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 105,
+ 572,
+ 505,
+ 585
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 572,
+ 505,
+ 585
+ ],
+ "score": 1.0,
+ "content": "2018); (Nichol et al., 2018). Recently proposed and widely accepted Initialization based methods",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ },
+ {
+ "bbox": [
+ 105,
+ 583,
+ 506,
+ 595
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 583,
+ 506,
+ 595
+ ],
+ "score": 1.0,
+ "content": "can broadly be classified into: (i) methods that learn good model parameters with limited labeled",
+ "type": "text"
+ }
+ ],
+ "index": 39
+ },
+ {
+ "bbox": [
+ 105,
+ 594,
+ 504,
+ 606
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 594,
+ 504,
+ 606
+ ],
+ "score": 1.0,
+ "content": "examples and a small number of gradient update steps (Finn et al., 2017) and (ii) methods that learn",
+ "type": "text"
+ }
+ ],
+ "index": 40
+ },
+ {
+ "bbox": [
+ 105,
+ 605,
+ 506,
+ 617
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 605,
+ 506,
+ 617
+ ],
+ "score": 1.0,
+ "content": "an optimizer (Ravi & Larochelle, 2017). We refer the interested reader to Chen et. al. (Chen et al.,",
+ "type": "text"
+ }
+ ],
+ "index": 41
+ },
+ {
+ "bbox": [
+ 105,
+ 615,
+ 368,
+ 628
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 615,
+ 368,
+ 628
+ ],
+ "score": 1.0,
+ "content": "2019) for more examples of few-shot learning methods in vision.",
+ "type": "text"
+ }
+ ],
+ "index": 42
+ }
+ ],
+ "index": 37,
+ "bbox_fs": [
+ 105,
+ 506,
+ 506,
+ 628
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 632,
+ 505,
+ 731
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 633,
+ 505,
+ 645
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 633,
+ 505,
+ 645
+ ],
+ "score": 1.0,
+ "content": "Graph neural networks (GNNs) were first introduced in (Gori et al., 2005); (Scarselli et al., 2009)",
+ "type": "text"
+ }
+ ],
+ "index": 43
+ },
+ {
+ "bbox": [
+ 105,
+ 644,
+ 506,
+ 656
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 644,
+ 506,
+ 656
+ ],
+ "score": 1.0,
+ "content": "as recurrent message passing algorithms. Subsequent work (Bruna et al., 2014); (Henaff et al.,",
+ "type": "text"
+ }
+ ],
+ "index": 44
+ },
+ {
+ "bbox": [
+ 105,
+ 655,
+ 506,
+ 668
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 655,
+ 506,
+ 668
+ ],
+ "score": 1.0,
+ "content": "2015) proposed to learn smooth spectral multipliers of the graph Laplacian, but incurred higher",
+ "type": "text"
+ }
+ ],
+ "index": 45
+ },
+ {
+ "bbox": [
+ 105,
+ 666,
+ 506,
+ 678
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 666,
+ 506,
+ 678
+ ],
+ "score": 1.0,
+ "content": "computational cost. This computational bottleneck was later resolved (Defferrard et al., 2016); (Kipf",
+ "type": "text"
+ }
+ ],
+ "index": 46
+ },
+ {
+ "bbox": [
+ 105,
+ 677,
+ 506,
+ 690
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 677,
+ 506,
+ 690
+ ],
+ "score": 1.0,
+ "content": "& Welling, 2016) by learning polynomials of the graph Laplacian. GNNs are a natural extension to",
+ "type": "text"
+ }
+ ],
+ "index": 47
+ },
+ {
+ "bbox": [
+ 106,
+ 688,
+ 505,
+ 700
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 688,
+ 505,
+ 700
+ ],
+ "score": 1.0,
+ "content": "Convolutional neural networks (CNNs) on non-Euclidean data. Recent work (Velikovi et al., 2018)",
+ "type": "text"
+ }
+ ],
+ "index": 48
+ },
+ {
+ "bbox": [
+ 105,
+ 698,
+ 506,
+ 712
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 698,
+ 506,
+ 712
+ ],
+ "score": 1.0,
+ "content": "introduced the concept of self-attention in GNNs, which allows each node to provide attention to",
+ "type": "text"
+ }
+ ],
+ "index": 49
+ },
+ {
+ "bbox": [
+ 105,
+ 709,
+ 506,
+ 723
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 709,
+ 506,
+ 723
+ ],
+ "score": 1.0,
+ "content": "the enclosing neighborhood resulting in improved learning. We refer the reader to (Bronstein et al.,",
+ "type": "text"
+ }
+ ],
+ "index": 50
+ },
+ {
+ "bbox": [
+ 106,
+ 720,
+ 272,
+ 732
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 720,
+ 272,
+ 732
+ ],
+ "score": 1.0,
+ "content": "2016) for detailed information on GNNs.",
+ "type": "text"
+ }
+ ],
+ "index": 51
+ }
+ ],
+ "index": 47,
+ "bbox_fs": [
+ 105,
+ 633,
+ 506,
+ 732
+ ]
+ }
+ ]
+ },
+ {
+ "preproc_blocks": [
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 82,
+ 505,
+ 137
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 82,
+ 504,
+ 94
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 82,
+ 504,
+ 94
+ ],
+ "score": 1.0,
+ "content": "Despite all the success of GNNs, few-shot classification remains an under-addressed problem. Some",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 105,
+ 93,
+ 505,
+ 106
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 93,
+ 505,
+ 106
+ ],
+ "score": 1.0,
+ "content": "recent attempts have focused on solving the few-shot learning on graph data where GNNs are either",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 105,
+ 104,
+ 505,
+ 118
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 104,
+ 505,
+ 118
+ ],
+ "score": 1.0,
+ "content": "trained via co-training and self-training (Li et al., 2018a), or extended by stacking transposed graph",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ },
+ {
+ "bbox": [
+ 105,
+ 115,
+ 505,
+ 128
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 115,
+ 505,
+ 128
+ ],
+ "score": 1.0,
+ "content": "convolutional layers imputing a structural regularizer (Zhang et al., 2019) - however, both these",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 105,
+ 127,
+ 301,
+ 138
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 127,
+ 301,
+ 138
+ ],
+ "score": 1.0,
+ "content": "works focus only on the node classification task.",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ }
+ ],
+ "index": 2
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 143,
+ 504,
+ 165
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 142,
+ 504,
+ 156
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 142,
+ 504,
+ 156
+ ],
+ "score": 1.0,
+ "content": "To the best of our knowledge, there does not exist any work pertaining few-shot learning on graphs",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 106,
+ 154,
+ 447,
+ 166
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 154,
+ 447,
+ 166
+ ],
+ "score": 1.0,
+ "content": "focusing on the graph classification task, thus providing the motivation for this work.",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ }
+ ],
+ "index": 5.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 171,
+ 505,
+ 489
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 171,
+ 505,
+ 184
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 171,
+ 505,
+ 184
+ ],
+ "score": 1.0,
+ "content": "Comparison to few-shot learning on images: Few shot learning (FSL) has gained wide-spread",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 105,
+ 181,
+ 505,
+ 195
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 181,
+ 505,
+ 195
+ ],
+ "score": 1.0,
+ "content": "traction in the image domain in recent years. However the success of FSL in images is not eas-",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 106,
+ 193,
+ 505,
+ 206
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 193,
+ 505,
+ 206
+ ],
+ "score": 1.0,
+ "content": "ily translated to the graph domain for the following reasons: (a) Images are typically represented",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 106,
+ 205,
+ 505,
+ 216
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 205,
+ 505,
+ 216
+ ],
+ "score": 1.0,
+ "content": "in Euclidean space and thus can easily be manipulated and handled using well-known metrics like",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 105,
+ 215,
+ 506,
+ 227
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 215,
+ 178,
+ 227
+ ],
+ "score": 1.0,
+ "content": "cosine similarity,",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 178,
+ 215,
+ 191,
+ 227
+ ],
+ "score": 0.92,
+ "content": "L _ { p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 192,
+ 215,
+ 506,
+ 227
+ ],
+ "score": 1.0,
+ "content": "norms etc. However, graphs come from non-Euclidean domains and exhibit",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 106,
+ 227,
+ 505,
+ 238
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 227,
+ 505,
+ 238
+ ],
+ "score": 1.0,
+ "content": "much more complex relationships and interdependency between objects. Furthermore, the notion",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 106,
+ 237,
+ 505,
+ 249
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 237,
+ 505,
+ 249
+ ],
+ "score": 1.0,
+ "content": "of a distance between graphs is also not straightforward and requires construction of graph kernels",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 105,
+ 248,
+ 505,
+ 261
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 248,
+ 505,
+ 261
+ ],
+ "score": 1.0,
+ "content": "or the use of standard metrics on graph embeddings (Kriege et al., 2019). Additionally, such graph",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 105,
+ 258,
+ 506,
+ 272
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 258,
+ 506,
+ 272
+ ],
+ "score": 1.0,
+ "content": "kernels dont capture higher order relations very well. (b) In the FSL setting on images, the number",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 105,
+ 270,
+ 505,
+ 283
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 270,
+ 505,
+ 283
+ ],
+ "score": 1.0,
+ "content": "of training samples from various classes is also abundantly more than what is available for graph",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 106,
+ 281,
+ 505,
+ 294
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 281,
+ 505,
+ 294
+ ],
+ "score": 1.0,
+ "content": "datasets. The image domain allows training generative models to learn the task distribution and can",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 105,
+ 291,
+ 505,
+ 305
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 291,
+ 505,
+ 305
+ ],
+ "score": 1.0,
+ "content": "further be used to generate samples for data augmentation, which act as very good priors. In contrast,",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 105,
+ 302,
+ 505,
+ 316
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 302,
+ 505,
+ 316
+ ],
+ "score": 1.0,
+ "content": "graph generative models are still in their infancy and work in very restricted settings. Furthermore,",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 104,
+ 313,
+ 506,
+ 327
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 313,
+ 506,
+ 327
+ ],
+ "score": 1.0,
+ "content": "methods like cropping and rotation to improve the models can’t be used for graphs given the per-",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 105,
+ 324,
+ 505,
+ 338
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 324,
+ 505,
+ 338
+ ],
+ "score": 1.0,
+ "content": "mutation invariant nature of graphs. Additionally, removal of any component from the graph can",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 106,
+ 336,
+ 506,
+ 349
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 336,
+ 506,
+ 349
+ ],
+ "score": 1.0,
+ "content": "adversely affect its structural properties, such as in biological datasets. (c) The image domain has",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 106,
+ 347,
+ 506,
+ 359
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 347,
+ 506,
+ 359
+ ],
+ "score": 1.0,
+ "content": "very well-known regularization methods (e.g. Tikhonov, Lasso) that help generalize much better to",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ },
+ {
+ "bbox": [
+ 106,
+ 357,
+ 505,
+ 370
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 357,
+ 505,
+ 370
+ ],
+ "score": 1.0,
+ "content": "novel datasets. Although, they dont bring any extra supervised information and hence cannot fully",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 105,
+ 368,
+ 505,
+ 381
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 368,
+ 505,
+ 381
+ ],
+ "score": 1.0,
+ "content": "address the problem of FSL in the image domain. To the best of our knowledge, this is still an open",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 105,
+ 379,
+ 505,
+ 392
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 379,
+ 505,
+ 392
+ ],
+ "score": 1.0,
+ "content": "research problem in the image domain. On the other hand, in the graph domain, our work would",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 106,
+ 390,
+ 505,
+ 403
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 390,
+ 505,
+ 403
+ ],
+ "score": 1.0,
+ "content": "be a first step towards graph classification in an FSL setting, which would then hopefully pave the",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 105,
+ 401,
+ 506,
+ 415
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 401,
+ 506,
+ 415
+ ],
+ "score": 1.0,
+ "content": "path for better FSL graph regularizers. (d) Transfer learning has led to substantial improvements",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 106,
+ 412,
+ 505,
+ 425
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 412,
+ 505,
+ 425
+ ],
+ "score": 1.0,
+ "content": "on various image related tasks due to the high degree of transferability of feature extractors. Thus,",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 106,
+ 423,
+ 506,
+ 436
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 423,
+ 506,
+ 436
+ ],
+ "score": 1.0,
+ "content": "downstream tasks like few-shot learning can be performed well with high quality feature extractor",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 105,
+ 434,
+ 506,
+ 448
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 434,
+ 506,
+ 448
+ ],
+ "score": 1.0,
+ "content": "models, such as Resnet variants trained on Imagenet. Transfer learning, or for that matter even good",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ },
+ {
+ "bbox": [
+ 105,
+ 445,
+ 506,
+ 458
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 445,
+ 506,
+ 458
+ ],
+ "score": 1.0,
+ "content": "feature extractors, remains a daunting challenge in the graph domain. For graphs, there neither exists",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 105,
+ 456,
+ 505,
+ 469
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 456,
+ 505,
+ 469
+ ],
+ "score": 1.0,
+ "content": "a dataset which can serve as a pivot for high quality feature learning, nor does there exist a Graph",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ },
+ {
+ "bbox": [
+ 105,
+ 466,
+ 505,
+ 480
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 466,
+ 505,
+ 480
+ ],
+ "score": 1.0,
+ "content": "NN which can capture the higher order relations between various categories of graphs, thus making",
+ "type": "text"
+ }
+ ],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 106,
+ 478,
+ 245,
+ 491
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 478,
+ 245,
+ 491
+ ],
+ "score": 1.0,
+ "content": "this a highly challenging problem.",
+ "type": "text"
+ }
+ ],
+ "index": 35
+ }
+ ],
+ "index": 21
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 507,
+ 207,
+ 519
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 506,
+ 209,
+ 522
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 506,
+ 209,
+ 522
+ ],
+ "score": 1.0,
+ "content": "3 PRELIMINARIES",
+ "type": "text"
+ }
+ ],
+ "index": 36
+ }
+ ],
+ "index": 36
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 533,
+ 505,
+ 577
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 533,
+ 505,
+ 545
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 533,
+ 505,
+ 545
+ ],
+ "score": 1.0,
+ "content": "In this section, we introduce our notation and provide the necessary background for our few-shot",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 105,
+ 544,
+ 505,
+ 556
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 544,
+ 505,
+ 556
+ ],
+ "score": 1.0,
+ "content": "learning setup on graphs. We begin by describing the various data sample types, followed by our",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ },
+ {
+ "bbox": [
+ 105,
+ 555,
+ 505,
+ 568
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 555,
+ 505,
+ 568
+ ],
+ "score": 1.0,
+ "content": "learning procedure, in order to formally define few-shot learning on graphs. Finally, we define the",
+ "type": "text"
+ }
+ ],
+ "index": 39
+ },
+ {
+ "bbox": [
+ 105,
+ 566,
+ 306,
+ 579
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 566,
+ 306,
+ 579
+ ],
+ "score": 1.0,
+ "content": "graph spectral distance between a pair of graphs.",
+ "type": "text"
+ }
+ ],
+ "index": 40
+ }
+ ],
+ "index": 38.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 582,
+ 505,
+ 658
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 582,
+ 506,
+ 595
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 582,
+ 198,
+ 595
+ ],
+ "score": 1.0,
+ "content": "Data sample sets: Let",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 198,
+ 583,
+ 206,
+ 594
+ ],
+ "score": 0.81,
+ "content": "\\mathcal { G }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 206,
+ 582,
+ 403,
+ 595
+ ],
+ "score": 1.0,
+ "content": "denote a set of undirected unweighted graphs and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 403,
+ 583,
+ 412,
+ 594
+ ],
+ "score": 0.81,
+ "content": "\\mathcal { V }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 412,
+ 582,
+ 506,
+ 595
+ ],
+ "score": 1.0,
+ "content": "be the set of associated",
+ "type": "text"
+ }
+ ],
+ "index": 41
+ },
+ {
+ "bbox": [
+ 106,
+ 594,
+ 505,
+ 606
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 594,
+ 505,
+ 606
+ ],
+ "score": 1.0,
+ "content": "class labels. We consider two disjoint populations of labeled graphs consisting of i.i.d. graph sam-",
+ "type": "text"
+ }
+ ],
+ "index": 42
+ },
+ {
+ "bbox": [
+ 105,
+ 601,
+ 506,
+ 624
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 601,
+ 269,
+ 624
+ ],
+ "score": 1.0,
+ "content": "ples, the set of base class labeled graphs",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 270,
+ 605,
+ 371,
+ 619
+ ],
+ "score": 0.92,
+ "content": "\\mathbf { \\bar { { G } } } _ { B } = \\{ ( g _ { i } ^ { ( B ) } , y _ { i } ^ { ( B ) } ) \\bar \\} _ { i = 1 } ^ { n } \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 371,
+ 607,
+ 506,
+ 619
+ ],
+ "score": 1.0,
+ "content": "and the set of novel class labeled",
+ "type": "text"
+ }
+ ],
+ "index": 43
+ },
+ {
+ "bbox": [
+ 104,
+ 615,
+ 507,
+ 642
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 617,
+ 136,
+ 638
+ ],
+ "score": 1.0,
+ "content": "graphs",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 136,
+ 619,
+ 239,
+ 634
+ ],
+ "score": 0.93,
+ "content": "G _ { N } = \\{ ( g _ { i } ^ { ( N ) } , y _ { i } ^ { ( N ) } ) \\} _ { i = 1 } ^ { m }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 195,
+ 615,
+ 284,
+ 642
+ ],
+ "score": 1.0,
+ "content": "y(N )i )}mi=1, where g(Bi",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 270,
+ 619,
+ 387,
+ 634
+ ],
+ "score": 0.78,
+ "content": "g _ { i } ^ { ( B ) } , g _ { i } ^ { ( N ) } \\in \\mathcal { G } , y _ { i } ^ { ( B ) } \\in \\mathcal { V } ^ { ( B ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 336,
+ 615,
+ 421,
+ 637
+ ],
+ "score": 1.0,
+ "content": "y(B)i ∈ Y (B), and y(Ni",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 408,
+ 618,
+ 461,
+ 633
+ ],
+ "score": 0.91,
+ "content": "y _ { i } ^ { ( N ) } \\in \\mathcal { y } ^ { ( N ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 462,
+ 618,
+ 507,
+ 635
+ ],
+ "score": 1.0,
+ "content": ". Here, the",
+ "type": "text"
+ }
+ ],
+ "index": 44
+ },
+ {
+ "bbox": [
+ 104,
+ 631,
+ 506,
+ 648
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 631,
+ 299,
+ 648
+ ],
+ "score": 1.0,
+ "content": "set of base and novel class labels are denoted by",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 299,
+ 634,
+ 379,
+ 646
+ ],
+ "score": 0.91,
+ "content": "{ \\mathcal { V } } ^ { ( B ) } = \\{ 1 , \\ldots , K \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 380,
+ 631,
+ 397,
+ 648
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 398,
+ 633,
+ 501,
+ 646
+ ],
+ "score": 0.91,
+ "content": "\\mathcal { V } ^ { ( N ) } = \\{ K + 1 , \\ldots , K ^ { \\prime } \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 502,
+ 631,
+ 506,
+ 648
+ ],
+ "score": 1.0,
+ "content": ",",
+ "type": "text"
+ }
+ ],
+ "index": 45
+ },
+ {
+ "bbox": [
+ 104,
+ 644,
+ 505,
+ 659
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 644,
+ 185,
+ 659
+ ],
+ "score": 1.0,
+ "content": "respectively, where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 186,
+ 646,
+ 221,
+ 657
+ ],
+ "score": 0.91,
+ "content": "K ^ { \\prime } > K",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 221,
+ 644,
+ 247,
+ 659
+ ],
+ "score": 1.0,
+ "content": ". Both",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 248,
+ 645,
+ 269,
+ 657
+ ],
+ "score": 0.91,
+ "content": "\\mathcal { V } ^ { ( B ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 269,
+ 644,
+ 287,
+ 659
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 288,
+ 646,
+ 309,
+ 658
+ ],
+ "score": 0.91,
+ "content": "\\mathcal { V } ^ { ( N ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 309,
+ 644,
+ 400,
+ 659
+ ],
+ "score": 1.0,
+ "content": "are disjoint subsets of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 401,
+ 647,
+ 409,
+ 657
+ ],
+ "score": 0.75,
+ "content": "\\mathcal { V }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 409,
+ 644,
+ 426,
+ 659
+ ],
+ "score": 1.0,
+ "content": ", so,",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 427,
+ 646,
+ 499,
+ 658
+ ],
+ "score": 0.91,
+ "content": "\\mathcal { V } ^ { ( B ) } \\cap \\mathcal { V } ^ { ( N ) } = \\emptyset",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 500,
+ 644,
+ 505,
+ 659
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 46
+ }
+ ],
+ "index": 43.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 663,
+ 505,
+ 700
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 662,
+ 505,
+ 676
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 662,
+ 147,
+ 676
+ ],
+ "score": 1.0,
+ "content": "Note that",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 147,
+ 665,
+ 182,
+ 674
+ ],
+ "score": 0.87,
+ "content": "m \\ll n",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 182,
+ 662,
+ 505,
+ 676
+ ],
+ "score": 1.0,
+ "content": ", i.e., there are far fewer novel class labeled graphs compared to the base class",
+ "type": "text"
+ }
+ ],
+ "index": 47
+ },
+ {
+ "bbox": [
+ 105,
+ 673,
+ 505,
+ 687
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 673,
+ 204,
+ 687
+ ],
+ "score": 1.0,
+ "content": "labeled ones. Besides",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 205,
+ 675,
+ 221,
+ 685
+ ],
+ "score": 0.87,
+ "content": "G _ { B }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 221,
+ 673,
+ 243,
+ 687
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 243,
+ 675,
+ 259,
+ 686
+ ],
+ "score": 0.88,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 259,
+ 673,
+ 356,
+ 687
+ ],
+ "score": 1.0,
+ "content": ", we consider a set of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 357,
+ 676,
+ 362,
+ 685
+ ],
+ "score": 0.77,
+ "content": "t",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 362,
+ 673,
+ 471,
+ 687
+ ],
+ "score": 1.0,
+ "content": "unlabeled unseen graphs",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 471,
+ 675,
+ 505,
+ 686
+ ],
+ "score": 0.88,
+ "content": "\\begin{array} { l l } { G _ { U } } & { : = } \\end{array}",
+ "type": "inline_equation"
+ }
+ ],
+ "index": 48
+ },
+ {
+ "bbox": [
+ 106,
+ 685,
+ 344,
+ 702
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 685,
+ 292,
+ 700
+ ],
+ "score": 0.89,
+ "content": "\\{ g _ { 1 } ^ { ( U ) } , \\ldots , g _ { t } ^ { ( U ) } \\mid g _ { i } ^ { ( U ) } \\in \\pi _ { 1 } ( G _ { N } ) , i = 1 \\ldots t \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 293,
+ 685,
+ 344,
+ 702
+ ],
+ "score": 1.0,
+ "content": ", for testing1.",
+ "type": "text"
+ }
+ ],
+ "index": 49
+ }
+ ],
+ "index": 48
+ }
+ ],
+ "page_idx": 2,
+ "page_size": [
+ 612,
+ 792
+ ],
+ "discarded_blocks": [
+ {
+ "type": "discarded",
+ "bbox": [
+ 105,
+ 711,
+ 504,
+ 732
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 118,
+ 708,
+ 506,
+ 724
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 118,
+ 708,
+ 197,
+ 724
+ ],
+ "score": 1.0,
+ "content": "1We use the notation",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 198,
+ 712,
+ 219,
+ 722
+ ],
+ "score": 0.91,
+ "content": "\\pi _ { 1 } ( p )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 220,
+ 708,
+ 237,
+ 724
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 237,
+ 712,
+ 259,
+ 722
+ ],
+ "score": 0.9,
+ "content": "\\pi _ { 2 } ( p )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 259,
+ 708,
+ 467,
+ 724
+ ],
+ "score": 1.0,
+ "content": "to denote the left and right projection of an ordered pair",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 467,
+ 714,
+ 473,
+ 722
+ ],
+ "score": 0.79,
+ "content": "p",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 473,
+ 708,
+ 506,
+ 724
+ ],
+ "score": 1.0,
+ "content": ", respec-",
+ "type": "text"
+ }
+ ]
+ },
+ {
+ "bbox": [
+ 104,
+ 718,
+ 132,
+ 734
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 718,
+ 132,
+ 734
+ ],
+ "score": 1.0,
+ "content": "tively.",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 108,
+ 27,
+ 293,
+ 37
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "score": 1.0,
+ "content": "Published as a conference paper at ICLR 2020",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 302,
+ 751,
+ 309,
+ 760
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 301,
+ 750,
+ 310,
+ 762
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 301,
+ 750,
+ 310,
+ 762
+ ],
+ "score": 1.0,
+ "content": "3",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ }
+ ],
+ "para_blocks": [
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 82,
+ 505,
+ 137
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 82,
+ 504,
+ 94
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 82,
+ 504,
+ 94
+ ],
+ "score": 1.0,
+ "content": "Despite all the success of GNNs, few-shot classification remains an under-addressed problem. Some",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 105,
+ 93,
+ 505,
+ 106
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 93,
+ 505,
+ 106
+ ],
+ "score": 1.0,
+ "content": "recent attempts have focused on solving the few-shot learning on graph data where GNNs are either",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 105,
+ 104,
+ 505,
+ 118
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 104,
+ 505,
+ 118
+ ],
+ "score": 1.0,
+ "content": "trained via co-training and self-training (Li et al., 2018a), or extended by stacking transposed graph",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ },
+ {
+ "bbox": [
+ 105,
+ 115,
+ 505,
+ 128
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 115,
+ 505,
+ 128
+ ],
+ "score": 1.0,
+ "content": "convolutional layers imputing a structural regularizer (Zhang et al., 2019) - however, both these",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 105,
+ 127,
+ 301,
+ 138
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 127,
+ 301,
+ 138
+ ],
+ "score": 1.0,
+ "content": "works focus only on the node classification task.",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ }
+ ],
+ "index": 2,
+ "bbox_fs": [
+ 105,
+ 82,
+ 505,
+ 138
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 143,
+ 504,
+ 165
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 142,
+ 504,
+ 156
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 142,
+ 504,
+ 156
+ ],
+ "score": 1.0,
+ "content": "To the best of our knowledge, there does not exist any work pertaining few-shot learning on graphs",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 106,
+ 154,
+ 447,
+ 166
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 154,
+ 447,
+ 166
+ ],
+ "score": 1.0,
+ "content": "focusing on the graph classification task, thus providing the motivation for this work.",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ }
+ ],
+ "index": 5.5,
+ "bbox_fs": [
+ 106,
+ 142,
+ 504,
+ 166
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 171,
+ 505,
+ 489
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 171,
+ 505,
+ 184
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 171,
+ 505,
+ 184
+ ],
+ "score": 1.0,
+ "content": "Comparison to few-shot learning on images: Few shot learning (FSL) has gained wide-spread",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 105,
+ 181,
+ 505,
+ 195
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 181,
+ 505,
+ 195
+ ],
+ "score": 1.0,
+ "content": "traction in the image domain in recent years. However the success of FSL in images is not eas-",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 106,
+ 193,
+ 505,
+ 206
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 193,
+ 505,
+ 206
+ ],
+ "score": 1.0,
+ "content": "ily translated to the graph domain for the following reasons: (a) Images are typically represented",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 106,
+ 205,
+ 505,
+ 216
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 205,
+ 505,
+ 216
+ ],
+ "score": 1.0,
+ "content": "in Euclidean space and thus can easily be manipulated and handled using well-known metrics like",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 105,
+ 215,
+ 506,
+ 227
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 215,
+ 178,
+ 227
+ ],
+ "score": 1.0,
+ "content": "cosine similarity,",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 178,
+ 215,
+ 191,
+ 227
+ ],
+ "score": 0.92,
+ "content": "L _ { p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 192,
+ 215,
+ 506,
+ 227
+ ],
+ "score": 1.0,
+ "content": "norms etc. However, graphs come from non-Euclidean domains and exhibit",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 106,
+ 227,
+ 505,
+ 238
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 227,
+ 505,
+ 238
+ ],
+ "score": 1.0,
+ "content": "much more complex relationships and interdependency between objects. Furthermore, the notion",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 106,
+ 237,
+ 505,
+ 249
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 237,
+ 505,
+ 249
+ ],
+ "score": 1.0,
+ "content": "of a distance between graphs is also not straightforward and requires construction of graph kernels",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 105,
+ 248,
+ 505,
+ 261
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 248,
+ 505,
+ 261
+ ],
+ "score": 1.0,
+ "content": "or the use of standard metrics on graph embeddings (Kriege et al., 2019). Additionally, such graph",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 105,
+ 258,
+ 506,
+ 272
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 258,
+ 506,
+ 272
+ ],
+ "score": 1.0,
+ "content": "kernels dont capture higher order relations very well. (b) In the FSL setting on images, the number",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 105,
+ 270,
+ 505,
+ 283
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 270,
+ 505,
+ 283
+ ],
+ "score": 1.0,
+ "content": "of training samples from various classes is also abundantly more than what is available for graph",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 106,
+ 281,
+ 505,
+ 294
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 281,
+ 505,
+ 294
+ ],
+ "score": 1.0,
+ "content": "datasets. The image domain allows training generative models to learn the task distribution and can",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 105,
+ 291,
+ 505,
+ 305
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 291,
+ 505,
+ 305
+ ],
+ "score": 1.0,
+ "content": "further be used to generate samples for data augmentation, which act as very good priors. In contrast,",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 105,
+ 302,
+ 505,
+ 316
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 302,
+ 505,
+ 316
+ ],
+ "score": 1.0,
+ "content": "graph generative models are still in their infancy and work in very restricted settings. Furthermore,",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 104,
+ 313,
+ 506,
+ 327
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 313,
+ 506,
+ 327
+ ],
+ "score": 1.0,
+ "content": "methods like cropping and rotation to improve the models can’t be used for graphs given the per-",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 105,
+ 324,
+ 505,
+ 338
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 324,
+ 505,
+ 338
+ ],
+ "score": 1.0,
+ "content": "mutation invariant nature of graphs. Additionally, removal of any component from the graph can",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 106,
+ 336,
+ 506,
+ 349
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 336,
+ 506,
+ 349
+ ],
+ "score": 1.0,
+ "content": "adversely affect its structural properties, such as in biological datasets. (c) The image domain has",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 106,
+ 347,
+ 506,
+ 359
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 347,
+ 506,
+ 359
+ ],
+ "score": 1.0,
+ "content": "very well-known regularization methods (e.g. Tikhonov, Lasso) that help generalize much better to",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ },
+ {
+ "bbox": [
+ 106,
+ 357,
+ 505,
+ 370
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 357,
+ 505,
+ 370
+ ],
+ "score": 1.0,
+ "content": "novel datasets. Although, they dont bring any extra supervised information and hence cannot fully",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 105,
+ 368,
+ 505,
+ 381
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 368,
+ 505,
+ 381
+ ],
+ "score": 1.0,
+ "content": "address the problem of FSL in the image domain. To the best of our knowledge, this is still an open",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 105,
+ 379,
+ 505,
+ 392
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 379,
+ 505,
+ 392
+ ],
+ "score": 1.0,
+ "content": "research problem in the image domain. On the other hand, in the graph domain, our work would",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 106,
+ 390,
+ 505,
+ 403
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 390,
+ 505,
+ 403
+ ],
+ "score": 1.0,
+ "content": "be a first step towards graph classification in an FSL setting, which would then hopefully pave the",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 105,
+ 401,
+ 506,
+ 415
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 401,
+ 506,
+ 415
+ ],
+ "score": 1.0,
+ "content": "path for better FSL graph regularizers. (d) Transfer learning has led to substantial improvements",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 106,
+ 412,
+ 505,
+ 425
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 412,
+ 505,
+ 425
+ ],
+ "score": 1.0,
+ "content": "on various image related tasks due to the high degree of transferability of feature extractors. Thus,",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 106,
+ 423,
+ 506,
+ 436
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 423,
+ 506,
+ 436
+ ],
+ "score": 1.0,
+ "content": "downstream tasks like few-shot learning can be performed well with high quality feature extractor",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 105,
+ 434,
+ 506,
+ 448
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 434,
+ 506,
+ 448
+ ],
+ "score": 1.0,
+ "content": "models, such as Resnet variants trained on Imagenet. Transfer learning, or for that matter even good",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ },
+ {
+ "bbox": [
+ 105,
+ 445,
+ 506,
+ 458
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 445,
+ 506,
+ 458
+ ],
+ "score": 1.0,
+ "content": "feature extractors, remains a daunting challenge in the graph domain. For graphs, there neither exists",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 105,
+ 456,
+ 505,
+ 469
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 456,
+ 505,
+ 469
+ ],
+ "score": 1.0,
+ "content": "a dataset which can serve as a pivot for high quality feature learning, nor does there exist a Graph",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ },
+ {
+ "bbox": [
+ 105,
+ 466,
+ 505,
+ 480
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 466,
+ 505,
+ 480
+ ],
+ "score": 1.0,
+ "content": "NN which can capture the higher order relations between various categories of graphs, thus making",
+ "type": "text"
+ }
+ ],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 106,
+ 478,
+ 245,
+ 491
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 478,
+ 245,
+ 491
+ ],
+ "score": 1.0,
+ "content": "this a highly challenging problem.",
+ "type": "text"
+ }
+ ],
+ "index": 35
+ }
+ ],
+ "index": 21,
+ "bbox_fs": [
+ 104,
+ 171,
+ 506,
+ 491
+ ]
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 507,
+ 207,
+ 519
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 506,
+ 209,
+ 522
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 506,
+ 209,
+ 522
+ ],
+ "score": 1.0,
+ "content": "3 PRELIMINARIES",
+ "type": "text"
+ }
+ ],
+ "index": 36
+ }
+ ],
+ "index": 36
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 533,
+ 505,
+ 577
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 533,
+ 505,
+ 545
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 533,
+ 505,
+ 545
+ ],
+ "score": 1.0,
+ "content": "In this section, we introduce our notation and provide the necessary background for our few-shot",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 105,
+ 544,
+ 505,
+ 556
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 544,
+ 505,
+ 556
+ ],
+ "score": 1.0,
+ "content": "learning setup on graphs. We begin by describing the various data sample types, followed by our",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ },
+ {
+ "bbox": [
+ 105,
+ 555,
+ 505,
+ 568
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 555,
+ 505,
+ 568
+ ],
+ "score": 1.0,
+ "content": "learning procedure, in order to formally define few-shot learning on graphs. Finally, we define the",
+ "type": "text"
+ }
+ ],
+ "index": 39
+ },
+ {
+ "bbox": [
+ 105,
+ 566,
+ 306,
+ 579
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 566,
+ 306,
+ 579
+ ],
+ "score": 1.0,
+ "content": "graph spectral distance between a pair of graphs.",
+ "type": "text"
+ }
+ ],
+ "index": 40
+ }
+ ],
+ "index": 38.5,
+ "bbox_fs": [
+ 105,
+ 533,
+ 505,
+ 579
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 582,
+ 505,
+ 658
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 582,
+ 506,
+ 595
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 582,
+ 198,
+ 595
+ ],
+ "score": 1.0,
+ "content": "Data sample sets: Let",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 198,
+ 583,
+ 206,
+ 594
+ ],
+ "score": 0.81,
+ "content": "\\mathcal { G }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 206,
+ 582,
+ 403,
+ 595
+ ],
+ "score": 1.0,
+ "content": "denote a set of undirected unweighted graphs and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 403,
+ 583,
+ 412,
+ 594
+ ],
+ "score": 0.81,
+ "content": "\\mathcal { V }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 412,
+ 582,
+ 506,
+ 595
+ ],
+ "score": 1.0,
+ "content": "be the set of associated",
+ "type": "text"
+ }
+ ],
+ "index": 41
+ },
+ {
+ "bbox": [
+ 106,
+ 594,
+ 505,
+ 606
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 594,
+ 505,
+ 606
+ ],
+ "score": 1.0,
+ "content": "class labels. We consider two disjoint populations of labeled graphs consisting of i.i.d. graph sam-",
+ "type": "text"
+ }
+ ],
+ "index": 42
+ },
+ {
+ "bbox": [
+ 105,
+ 601,
+ 506,
+ 624
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 601,
+ 269,
+ 624
+ ],
+ "score": 1.0,
+ "content": "ples, the set of base class labeled graphs",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 270,
+ 605,
+ 371,
+ 619
+ ],
+ "score": 0.92,
+ "content": "\\mathbf { \\bar { { G } } } _ { B } = \\{ ( g _ { i } ^ { ( B ) } , y _ { i } ^ { ( B ) } ) \\bar \\} _ { i = 1 } ^ { n } \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 371,
+ 607,
+ 506,
+ 619
+ ],
+ "score": 1.0,
+ "content": "and the set of novel class labeled",
+ "type": "text"
+ }
+ ],
+ "index": 43
+ },
+ {
+ "bbox": [
+ 104,
+ 615,
+ 507,
+ 642
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 617,
+ 136,
+ 638
+ ],
+ "score": 1.0,
+ "content": "graphs",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 136,
+ 619,
+ 239,
+ 634
+ ],
+ "score": 0.93,
+ "content": "G _ { N } = \\{ ( g _ { i } ^ { ( N ) } , y _ { i } ^ { ( N ) } ) \\} _ { i = 1 } ^ { m }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 195,
+ 615,
+ 284,
+ 642
+ ],
+ "score": 1.0,
+ "content": "y(N )i )}mi=1, where g(Bi",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 270,
+ 619,
+ 387,
+ 634
+ ],
+ "score": 0.78,
+ "content": "g _ { i } ^ { ( B ) } , g _ { i } ^ { ( N ) } \\in \\mathcal { G } , y _ { i } ^ { ( B ) } \\in \\mathcal { V } ^ { ( B ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 336,
+ 615,
+ 421,
+ 637
+ ],
+ "score": 1.0,
+ "content": "y(B)i ∈ Y (B), and y(Ni",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 408,
+ 618,
+ 461,
+ 633
+ ],
+ "score": 0.91,
+ "content": "y _ { i } ^ { ( N ) } \\in \\mathcal { y } ^ { ( N ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 462,
+ 618,
+ 507,
+ 635
+ ],
+ "score": 1.0,
+ "content": ". Here, the",
+ "type": "text"
+ }
+ ],
+ "index": 44
+ },
+ {
+ "bbox": [
+ 104,
+ 631,
+ 506,
+ 648
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 631,
+ 299,
+ 648
+ ],
+ "score": 1.0,
+ "content": "set of base and novel class labels are denoted by",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 299,
+ 634,
+ 379,
+ 646
+ ],
+ "score": 0.91,
+ "content": "{ \\mathcal { V } } ^ { ( B ) } = \\{ 1 , \\ldots , K \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 380,
+ 631,
+ 397,
+ 648
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 398,
+ 633,
+ 501,
+ 646
+ ],
+ "score": 0.91,
+ "content": "\\mathcal { V } ^ { ( N ) } = \\{ K + 1 , \\ldots , K ^ { \\prime } \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 502,
+ 631,
+ 506,
+ 648
+ ],
+ "score": 1.0,
+ "content": ",",
+ "type": "text"
+ }
+ ],
+ "index": 45
+ },
+ {
+ "bbox": [
+ 104,
+ 644,
+ 505,
+ 659
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 644,
+ 185,
+ 659
+ ],
+ "score": 1.0,
+ "content": "respectively, where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 186,
+ 646,
+ 221,
+ 657
+ ],
+ "score": 0.91,
+ "content": "K ^ { \\prime } > K",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 221,
+ 644,
+ 247,
+ 659
+ ],
+ "score": 1.0,
+ "content": ". Both",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 248,
+ 645,
+ 269,
+ 657
+ ],
+ "score": 0.91,
+ "content": "\\mathcal { V } ^ { ( B ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 269,
+ 644,
+ 287,
+ 659
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 288,
+ 646,
+ 309,
+ 658
+ ],
+ "score": 0.91,
+ "content": "\\mathcal { V } ^ { ( N ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 309,
+ 644,
+ 400,
+ 659
+ ],
+ "score": 1.0,
+ "content": "are disjoint subsets of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 401,
+ 647,
+ 409,
+ 657
+ ],
+ "score": 0.75,
+ "content": "\\mathcal { V }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 409,
+ 644,
+ 426,
+ 659
+ ],
+ "score": 1.0,
+ "content": ", so,",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 427,
+ 646,
+ 499,
+ 658
+ ],
+ "score": 0.91,
+ "content": "\\mathcal { V } ^ { ( B ) } \\cap \\mathcal { V } ^ { ( N ) } = \\emptyset",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 500,
+ 644,
+ 505,
+ 659
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 46
+ }
+ ],
+ "index": 43.5,
+ "bbox_fs": [
+ 104,
+ 582,
+ 507,
+ 659
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 663,
+ 505,
+ 700
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 662,
+ 505,
+ 676
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 662,
+ 147,
+ 676
+ ],
+ "score": 1.0,
+ "content": "Note that",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 147,
+ 665,
+ 182,
+ 674
+ ],
+ "score": 0.87,
+ "content": "m \\ll n",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 182,
+ 662,
+ 505,
+ 676
+ ],
+ "score": 1.0,
+ "content": ", i.e., there are far fewer novel class labeled graphs compared to the base class",
+ "type": "text"
+ }
+ ],
+ "index": 47
+ },
+ {
+ "bbox": [
+ 105,
+ 673,
+ 505,
+ 687
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 673,
+ 204,
+ 687
+ ],
+ "score": 1.0,
+ "content": "labeled ones. Besides",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 205,
+ 675,
+ 221,
+ 685
+ ],
+ "score": 0.87,
+ "content": "G _ { B }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 221,
+ 673,
+ 243,
+ 687
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 243,
+ 675,
+ 259,
+ 686
+ ],
+ "score": 0.88,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 259,
+ 673,
+ 356,
+ 687
+ ],
+ "score": 1.0,
+ "content": ", we consider a set of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 357,
+ 676,
+ 362,
+ 685
+ ],
+ "score": 0.77,
+ "content": "t",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 362,
+ 673,
+ 471,
+ 687
+ ],
+ "score": 1.0,
+ "content": "unlabeled unseen graphs",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 471,
+ 675,
+ 505,
+ 686
+ ],
+ "score": 0.88,
+ "content": "\\begin{array} { l l } { G _ { U } } & { : = } \\end{array}",
+ "type": "inline_equation"
+ }
+ ],
+ "index": 48
+ },
+ {
+ "bbox": [
+ 106,
+ 685,
+ 344,
+ 702
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 685,
+ 292,
+ 700
+ ],
+ "score": 0.89,
+ "content": "\\{ g _ { 1 } ^ { ( U ) } , \\ldots , g _ { t } ^ { ( U ) } \\mid g _ { i } ^ { ( U ) } \\in \\pi _ { 1 } ( G _ { N } ) , i = 1 \\ldots t \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 293,
+ 685,
+ 344,
+ 702
+ ],
+ "score": 1.0,
+ "content": ", for testing1.",
+ "type": "text"
+ }
+ ],
+ "index": 49
+ }
+ ],
+ "index": 48,
+ "bbox_fs": [
+ 105,
+ 662,
+ 505,
+ 702
+ ]
+ }
+ ]
+ },
+ {
+ "preproc_blocks": [
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 82,
+ 504,
+ 105
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 104,
+ 80,
+ 506,
+ 97
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 80,
+ 506,
+ 97
+ ],
+ "score": 1.0,
+ "content": "Learning procedure: Inspired by the initialization based methods, we similarly follow a two-stage",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 106,
+ 93,
+ 290,
+ 108
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 93,
+ 290,
+ 108
+ ],
+ "score": 1.0,
+ "content": "approach of training followed by fine-tuning.",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ }
+ ],
+ "index": 0.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 110,
+ 505,
+ 177
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 110,
+ 505,
+ 123
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 110,
+ 309,
+ 123
+ ],
+ "score": 1.0,
+ "content": "During training, we train a graph feature extractor",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 309,
+ 110,
+ 344,
+ 122
+ ],
+ "score": 0.94,
+ "content": "F _ { \\theta } ( G _ { B } )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 344,
+ 110,
+ 447,
+ 123
+ ],
+ "score": 1.0,
+ "content": "with network parameters",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 447,
+ 111,
+ 453,
+ 120
+ ],
+ "score": 0.78,
+ "content": "\\theta",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 454,
+ 110,
+ 505,
+ 123
+ ],
+ "score": 1.0,
+ "content": "followed by",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ },
+ {
+ "bbox": [
+ 105,
+ 121,
+ 505,
+ 134
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 121,
+ 152,
+ 134
+ ],
+ "score": 1.0,
+ "content": "a classifier",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 153,
+ 122,
+ 184,
+ 133
+ ],
+ "score": 0.93,
+ "content": "C ( G _ { B } )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 184,
+ 121,
+ 250,
+ 134
+ ],
+ "score": 1.0,
+ "content": "on graphs from",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 251,
+ 122,
+ 266,
+ 132
+ ],
+ "score": 0.9,
+ "content": "G _ { B }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 267,
+ 121,
+ 505,
+ 134
+ ],
+ "score": 1.0,
+ "content": ", where the loss function is the standard cross-entropy loss",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 106,
+ 132,
+ 506,
+ 145
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 133,
+ 118,
+ 144
+ ],
+ "score": 0.87,
+ "content": "\\mathcal { L } _ { c }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 119,
+ 132,
+ 506,
+ 145
+ ],
+ "score": 1.0,
+ "content": ". In order to better recognize and generalize well on samples from novel classes, in the fine-",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 105,
+ 143,
+ 505,
+ 155
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 143,
+ 294,
+ 155
+ ],
+ "score": 1.0,
+ "content": "tuning phase, the pre-trained feature extractor",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 295,
+ 143,
+ 317,
+ 155
+ ],
+ "score": 0.92,
+ "content": "F _ { \\theta } ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 318,
+ 143,
+ 505,
+ 155
+ ],
+ "score": 1.0,
+ "content": "along with its trained parameters is fixed and",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 105,
+ 153,
+ 506,
+ 168
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 153,
+ 160,
+ 168
+ ],
+ "score": 1.0,
+ "content": "the classifier",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 160,
+ 154,
+ 192,
+ 166
+ ],
+ "score": 0.92,
+ "content": "C ( G _ { N } )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 192,
+ 153,
+ 425,
+ 168
+ ],
+ "score": 1.0,
+ "content": "is trained on the novel class labeled graph samples from",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 425,
+ 154,
+ 441,
+ 165
+ ],
+ "score": 0.89,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 442,
+ 153,
+ 506,
+ 168
+ ],
+ "score": 1.0,
+ "content": ", with the same",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 104,
+ 162,
+ 142,
+ 179
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 162,
+ 124,
+ 179
+ ],
+ "score": 1.0,
+ "content": "loss",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 124,
+ 166,
+ 137,
+ 176
+ ],
+ "score": 0.88,
+ "content": "\\mathcal { L } _ { c }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 137,
+ 162,
+ 142,
+ 179
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ }
+ ],
+ "index": 4.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 182,
+ 504,
+ 204
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 182,
+ 505,
+ 194
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 182,
+ 505,
+ 194
+ ],
+ "score": 1.0,
+ "content": "Now, given the classification of data samples and the two-stage learning method, our problem of",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 106,
+ 192,
+ 345,
+ 205
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 192,
+ 345,
+ 205
+ ],
+ "score": 1.0,
+ "content": "few-shot classification on graphs can be defined as follows.",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ }
+ ],
+ "index": 8.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 209,
+ 505,
+ 266
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 209,
+ 504,
+ 223
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 209,
+ 220,
+ 223
+ ],
+ "score": 1.0,
+ "content": "Problem definition: Given",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 220,
+ 212,
+ 227,
+ 220
+ ],
+ "score": 0.65,
+ "content": "n",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 228,
+ 209,
+ 355,
+ 223
+ ],
+ "score": 1.0,
+ "content": "base-class labeled graphs from",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 356,
+ 210,
+ 371,
+ 221
+ ],
+ "score": 0.88,
+ "content": "G _ { B }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 372,
+ 209,
+ 494,
+ 223
+ ],
+ "score": 1.0,
+ "content": "during the training phase and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 494,
+ 212,
+ 504,
+ 220
+ ],
+ "score": 0.66,
+ "content": "m",
+ "type": "inline_equation"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 105,
+ 221,
+ 505,
+ 234
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 221,
+ 240,
+ 234
+ ],
+ "score": 1.0,
+ "content": "novel-class labeled graphs from",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 240,
+ 221,
+ 257,
+ 232
+ ],
+ "score": 0.89,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 257,
+ 221,
+ 409,
+ 234
+ ],
+ "score": 1.0,
+ "content": "during the fine-tuning phase, where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 410,
+ 222,
+ 445,
+ 232
+ ],
+ "score": 0.88,
+ "content": "m \\ll n",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 446,
+ 221,
+ 505,
+ 234
+ ],
+ "score": 1.0,
+ "content": ", the objective",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 106,
+ 232,
+ 505,
+ 244
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 232,
+ 291,
+ 244
+ ],
+ "score": 1.0,
+ "content": "of few-shot graph classification is to classify",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 291,
+ 233,
+ 297,
+ 242
+ ],
+ "score": 0.74,
+ "content": "t",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 297,
+ 232,
+ 429,
+ 244
+ ],
+ "score": 1.0,
+ "content": "unseen test graph samples from",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 430,
+ 232,
+ 445,
+ 243
+ ],
+ "score": 0.87,
+ "content": "G _ { U }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 446,
+ 232,
+ 505,
+ 244
+ ],
+ "score": 1.0,
+ "content": ". Moreover, if",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 107,
+ 242,
+ 506,
+ 256
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 243,
+ 142,
+ 254
+ ],
+ "score": 0.91,
+ "content": "m = q T",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 143,
+ 242,
+ 173,
+ 256
+ ],
+ "score": 1.0,
+ "content": ", where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 174,
+ 243,
+ 230,
+ 254
+ ],
+ "score": 0.92,
+ "content": "T = K ^ { \\prime } - K",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 230,
+ 242,
+ 405,
+ 256
+ ],
+ "score": 1.0,
+ "content": ", i.e., each novel class label appears exactly",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 405,
+ 245,
+ 412,
+ 255
+ ],
+ "score": 0.8,
+ "content": "q",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 412,
+ 242,
+ 447,
+ 256
+ ],
+ "score": 1.0,
+ "content": "times in",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 448,
+ 243,
+ 464,
+ 254
+ ],
+ "score": 0.9,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 464,
+ 242,
+ 506,
+ 256
+ ],
+ "score": 1.0,
+ "content": ", then this",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 105,
+ 253,
+ 311,
+ 268
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 253,
+ 214,
+ 268
+ ],
+ "score": 1.0,
+ "content": "setting is referred to as the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 214,
+ 255,
+ 221,
+ 265
+ ],
+ "score": 0.8,
+ "content": "q",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 221,
+ 253,
+ 244,
+ 268
+ ],
+ "score": 1.0,
+ "content": "-shot,",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 244,
+ 254,
+ 253,
+ 264
+ ],
+ "score": 0.81,
+ "content": "T",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 253,
+ 253,
+ 311,
+ 268
+ ],
+ "score": 1.0,
+ "content": "-way learning.",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ }
+ ],
+ "index": 12
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 270,
+ 505,
+ 365
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 270,
+ 505,
+ 284
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 270,
+ 338,
+ 284
+ ],
+ "score": 1.0,
+ "content": "Graph spectral distance: Let us consider the graphs in",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 338,
+ 271,
+ 346,
+ 282
+ ],
+ "score": 0.83,
+ "content": "\\mathcal { G }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 346,
+ 270,
+ 505,
+ 284
+ ],
+ "score": 1.0,
+ "content": ". The normalized Laplacian of a graph",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 106,
+ 280,
+ 506,
+ 297
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 282,
+ 135,
+ 295
+ ],
+ "score": 0.9,
+ "content": "g \\in { \\mathcal { G } }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 135,
+ 280,
+ 191,
+ 297
+ ],
+ "score": 1.0,
+ "content": "is defined as",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 191,
+ 281,
+ 295,
+ 295
+ ],
+ "score": 0.92,
+ "content": "\\Delta _ { g } = I - D ^ { - 1 / 2 } A D ^ { 1 / 2 }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 296,
+ 280,
+ 328,
+ 297
+ ],
+ "score": 1.0,
+ "content": ", where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 329,
+ 283,
+ 337,
+ 293
+ ],
+ "score": 0.81,
+ "content": "A",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 338,
+ 280,
+ 357,
+ 297
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 357,
+ 283,
+ 367,
+ 293
+ ],
+ "score": 0.83,
+ "content": "D",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 367,
+ 280,
+ 506,
+ 297
+ ],
+ "score": 1.0,
+ "content": "are the adjacency and the degree",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 101,
+ 289,
+ 505,
+ 316
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 101,
+ 289,
+ 182,
+ 316
+ ],
+ "score": 1.0,
+ "content": "matrices of graph",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 182,
+ 299,
+ 189,
+ 308
+ ],
+ "score": 0.74,
+ "content": "g",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 190,
+ 289,
+ 360,
+ 316
+ ],
+ "score": 1.0,
+ "content": ", respectively. The set of eigenvalues of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 360,
+ 297,
+ 374,
+ 309
+ ],
+ "score": 0.91,
+ "content": "\\Delta _ { g }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 375,
+ 289,
+ 416,
+ 316
+ ],
+ "score": 1.0,
+ "content": "given by",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 416,
+ 294,
+ 449,
+ 309
+ ],
+ "score": 0.92,
+ "content": "\\{ \\lambda _ { i } \\} _ { i = 1 } ^ { | V | }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 450,
+ 289,
+ 505,
+ 316
+ ],
+ "score": 1.0,
+ "content": "is called the",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 104,
+ 307,
+ 506,
+ 321
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 307,
+ 157,
+ 321
+ ],
+ "score": 1.0,
+ "content": "spectrum of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 172,
+ 307,
+ 248,
+ 321
+ ],
+ "score": 1.0,
+ "content": "and is denoted by",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 268,
+ 307,
+ 416,
+ 321
+ ],
+ "score": 1.0,
+ "content": ". It is well known that the spectrum",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 417,
+ 309,
+ 437,
+ 320
+ ],
+ "score": 0.88,
+ "content": "\\sigma ( g )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 437,
+ 307,
+ 506,
+ 321
+ ],
+ "score": 1.0,
+ "content": "of a normalized",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 105,
+ 319,
+ 505,
+ 331
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 319,
+ 274,
+ 331
+ ],
+ "score": 1.0,
+ "content": "Laplacian matrix is contained in interval",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 274,
+ 319,
+ 295,
+ 331
+ ],
+ "score": 0.47,
+ "content": "[ 0 , 2 ]",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 295,
+ 319,
+ 401,
+ 331
+ ],
+ "score": 1.0,
+ "content": ". We assign a Dirac mass",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 401,
+ 319,
+ 415,
+ 331
+ ],
+ "score": 0.88,
+ "content": "\\delta _ { \\lambda _ { i } }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 415,
+ 319,
+ 505,
+ 331
+ ],
+ "score": 1.0,
+ "content": "concentrated on each",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 106,
+ 328,
+ 506,
+ 344
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 330,
+ 152,
+ 342
+ ],
+ "score": 0.92,
+ "content": "\\lambda _ { i } \\ \\in \\ \\sigma ( g )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 153,
+ 328,
+ 330,
+ 344
+ ],
+ "score": 1.0,
+ "content": ", thus associating a probability measure to",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 330,
+ 330,
+ 350,
+ 342
+ ],
+ "score": 0.92,
+ "content": "\\sigma ( g )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 350,
+ 328,
+ 409,
+ 344
+ ],
+ "score": 1.0,
+ "content": "supported on",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 410,
+ 330,
+ 430,
+ 342
+ ],
+ "score": 0.82,
+ "content": "[ 0 , 2 ]",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 430,
+ 328,
+ 506,
+ 344
+ ],
+ "score": 1.0,
+ "content": ", called the graph",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 105,
+ 340,
+ 505,
+ 354
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 340,
+ 178,
+ 354
+ ],
+ "score": 1.0,
+ "content": "spectral measure",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 179,
+ 343,
+ 201,
+ 354
+ ],
+ "score": 0.89,
+ "content": "\\mu _ { \\sigma ( g ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 201,
+ 340,
+ 278,
+ 354
+ ],
+ "score": 1.0,
+ "content": ". Furthermore, let",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 278,
+ 341,
+ 314,
+ 353
+ ],
+ "score": 0.93,
+ "content": "P ( [ 0 , 2 ] )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 315,
+ 340,
+ 505,
+ 354
+ ],
+ "score": 1.0,
+ "content": "be the set of probability measures on interval",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 107,
+ 351,
+ 133,
+ 366
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 353,
+ 127,
+ 365
+ ],
+ "score": 0.4,
+ "content": "[ 0 , 2 ]",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 128,
+ 351,
+ 133,
+ 366
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ }
+ ],
+ "index": 18.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 105,
+ 369,
+ 504,
+ 393
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 369,
+ 505,
+ 383
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 369,
+ 185,
+ 383
+ ],
+ "score": 1.0,
+ "content": "We now define the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 185,
+ 371,
+ 191,
+ 381
+ ],
+ "score": 0.81,
+ "content": "p",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 192,
+ 369,
+ 505,
+ 383
+ ],
+ "score": 1.0,
+ "content": "-th Wasserstein distance between probability measures, which we later use to",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ },
+ {
+ "bbox": [
+ 106,
+ 381,
+ 321,
+ 394
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 381,
+ 321,
+ 394
+ ],
+ "score": 1.0,
+ "content": "define the spectral distance between a pair of graphs.",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ }
+ ],
+ "index": 23.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 402,
+ 505,
+ 436
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 402,
+ 505,
+ 415
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 402,
+ 176,
+ 415
+ ],
+ "score": 1.0,
+ "content": "Definition 1 Let",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 177,
+ 403,
+ 222,
+ 415
+ ],
+ "score": 0.91,
+ "content": "p \\in [ 1 , \\infty )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 222,
+ 402,
+ 254,
+ 415
+ ],
+ "score": 1.0,
+ "content": "and let",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 254,
+ 403,
+ 371,
+ 415
+ ],
+ "score": 0.91,
+ "content": "c : [ 0 , 2 ] \\times [ 0 , 2 ] [ 0 , + \\infty ]",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 371,
+ 402,
+ 505,
+ 415
+ ],
+ "score": 1.0,
+ "content": "be the cost function between the",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 104,
+ 414,
+ 506,
+ 427
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 414,
+ 193,
+ 427
+ ],
+ "score": 1.0,
+ "content": "probability measures",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 193,
+ 414,
+ 257,
+ 426
+ ],
+ "score": 0.92,
+ "content": "\\mu , \\nu \\in P ( [ 0 , 2 ] )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 258,
+ 414,
+ 299,
+ 427
+ ],
+ "score": 1.0,
+ "content": ". Then the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 300,
+ 416,
+ 306,
+ 426
+ ],
+ "score": 0.72,
+ "content": "p",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 306,
+ 414,
+ 479,
+ 427
+ ],
+ "score": 1.0,
+ "content": "-th Wasserstein distance between measures",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 479,
+ 416,
+ 486,
+ 426
+ ],
+ "score": 0.71,
+ "content": "\\mu",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 487,
+ 414,
+ 506,
+ 427
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 106,
+ 425,
+ 160,
+ 439
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 427,
+ 114,
+ 435
+ ],
+ "score": 0.64,
+ "content": "\\nu",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 114,
+ 425,
+ 160,
+ 439
+ ],
+ "score": 1.0,
+ "content": "is given by",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ }
+ ],
+ "index": 26
+ },
+ {
+ "type": "interline_equation",
+ "bbox": [
+ 188,
+ 435,
+ 421,
+ 472
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 188,
+ 435,
+ 421,
+ 472
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 188,
+ 435,
+ 421,
+ 472
+ ],
+ "score": 0.93,
+ "content": "W _ { p } ( \\mu , \\nu ) = \\left( \\operatorname* { i n f } _ { \\gamma } \\int _ { [ 0 , 2 ] \\times [ 0 , 2 ] } c ( x , y ) ^ { p } d \\gamma \\mid \\gamma \\in \\Pi ( \\mu , \\nu ) \\right) ^ { \\frac { 1 } { p } }",
+ "type": "interline_equation",
+ "image_path": "34613c1c0832df651fd54184395645c27fd1a9bf9f646499c259923761e9bd54.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 28.5,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 188,
+ 435,
+ 421,
+ 453.5
+ ],
+ "spans": [],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 188,
+ 453.5,
+ 421,
+ 472.0
+ ],
+ "spans": [],
+ "index": 29
+ }
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 474,
+ 505,
+ 497
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 473,
+ 505,
+ 487
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 473,
+ 133,
+ 487
+ ],
+ "score": 1.0,
+ "content": "where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 133,
+ 474,
+ 165,
+ 487
+ ],
+ "score": 0.93,
+ "content": "\\Pi ( \\mu , \\nu )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 165,
+ 473,
+ 431,
+ 487
+ ],
+ "score": 1.0,
+ "content": "is the set of transport plans, i.e., the collection of all measures on",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 431,
+ 474,
+ 484,
+ 487
+ ],
+ "score": 0.91,
+ "content": "[ 0 , 2 ] \\times [ 0 , 2 ]",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 484,
+ 473,
+ 505,
+ 487
+ ],
+ "score": 1.0,
+ "content": "with",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 105,
+ 486,
+ 186,
+ 498
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 486,
+ 149,
+ 498
+ ],
+ "score": 1.0,
+ "content": "marginals",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 149,
+ 488,
+ 156,
+ 497
+ ],
+ "score": 0.79,
+ "content": "\\mu",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 157,
+ 486,
+ 175,
+ 498
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 175,
+ 488,
+ 181,
+ 495
+ ],
+ "score": 0.73,
+ "content": "\\nu",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 182,
+ 486,
+ 186,
+ 498
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ }
+ ],
+ "index": 30.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 105,
+ 507,
+ 505,
+ 531
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 507,
+ 505,
+ 520
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 507,
+ 246,
+ 520
+ ],
+ "score": 1.0,
+ "content": "Given the general definition of the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 246,
+ 509,
+ 253,
+ 519
+ ],
+ "score": 0.83,
+ "content": "p",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 254,
+ 507,
+ 505,
+ 520
+ ],
+ "score": 1.0,
+ "content": "-th Wasserstein distance between probability measures and the",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 105,
+ 518,
+ 487,
+ 531
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 518,
+ 476,
+ 531
+ ],
+ "score": 1.0,
+ "content": "graph spectral measure, we can now define the spectral distance between a pair of graphs in",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 476,
+ 519,
+ 484,
+ 529
+ ],
+ "score": 0.83,
+ "content": "\\mathcal { G }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 484,
+ 518,
+ 487,
+ 531
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ }
+ ],
+ "index": 32.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 103,
+ 538,
+ 469,
+ 552
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 538,
+ 469,
+ 553
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 538,
+ 234,
+ 553
+ ],
+ "score": 1.0,
+ "content": "Definition 2 Given two graphs",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 234,
+ 540,
+ 271,
+ 551
+ ],
+ "score": 0.92,
+ "content": "g , g ^ { \\prime } \\in \\mathcal { G }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 272,
+ 538,
+ 469,
+ 553
+ ],
+ "score": 1.0,
+ "content": ", the spectral distance between them is defined as",
+ "type": "text"
+ }
+ ],
+ "index": 34
+ }
+ ],
+ "index": 34
+ },
+ {
+ "type": "interline_equation",
+ "bbox": [
+ 238,
+ 556,
+ 372,
+ 572
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 238,
+ 556,
+ 372,
+ 572
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 238,
+ 556,
+ 372,
+ 572
+ ],
+ "score": 0.92,
+ "content": "W ^ { p } ( g , g ^ { \\prime } ) : = W _ { p } \\left( \\mu _ { \\sigma ( g ) } , \\mu _ { \\sigma ( g ^ { \\prime } ) } \\right)",
+ "type": "interline_equation",
+ "image_path": "d89709e1a2a9eee3ee2245c73557bd920b69e751fe1e8e0d895a71cecea20d9f.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 35,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 238,
+ 556,
+ 372,
+ 572
+ ],
+ "spans": [],
+ "index": 35
+ }
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 582,
+ 504,
+ 617
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 582,
+ 506,
+ 596
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 582,
+ 147,
+ 596
+ ],
+ "score": 1.0,
+ "content": "In words,",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 148,
+ 583,
+ 189,
+ 595
+ ],
+ "score": 0.92,
+ "content": "W ^ { p } ( g , g ^ { \\prime } )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 189,
+ 582,
+ 506,
+ 596
+ ],
+ "score": 1.0,
+ "content": "is the optimal cost of moving mass from the graph spectral measure of graph",
+ "type": "text"
+ }
+ ],
+ "index": 36
+ },
+ {
+ "bbox": [
+ 107,
+ 594,
+ 506,
+ 606
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 597,
+ 113,
+ 606
+ ],
+ "score": 0.75,
+ "content": "g",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 113,
+ 594,
+ 181,
+ 606
+ ],
+ "score": 1.0,
+ "content": "to that of graph",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 181,
+ 595,
+ 190,
+ 606
+ ],
+ "score": 0.86,
+ "content": "g ^ { \\prime }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 190,
+ 594,
+ 432,
+ 606
+ ],
+ "score": 1.0,
+ "content": ", where the cost of moving unit mass is proportional to the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 432,
+ 596,
+ 438,
+ 606
+ ],
+ "score": 0.81,
+ "content": "p",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 438,
+ 594,
+ 506,
+ 606
+ ],
+ "score": 1.0,
+ "content": "-th power of the",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 105,
+ 604,
+ 300,
+ 618
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 604,
+ 270,
+ 618
+ ],
+ "score": 1.0,
+ "content": "difference of real-eigenvalues in interval",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 271,
+ 605,
+ 295,
+ 617
+ ],
+ "score": 0.92,
+ "content": "[ 0 , 2 ] ^ { 2 }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 295,
+ 604,
+ 300,
+ 618
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ }
+ ],
+ "index": 37
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 107,
+ 632,
+ 200,
+ 645
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 630,
+ 201,
+ 648
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 630,
+ 201,
+ 648
+ ],
+ "score": 1.0,
+ "content": "4 OUR METHOD",
+ "type": "text"
+ }
+ ],
+ "index": 39
+ }
+ ],
+ "index": 39
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 658,
+ 505,
+ 703
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 658,
+ 506,
+ 671
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 658,
+ 506,
+ 671
+ ],
+ "score": 1.0,
+ "content": "We present our proposed approach here. First, given abundant base-class labels, we cluster them into",
+ "type": "text"
+ }
+ ],
+ "index": 40
+ },
+ {
+ "bbox": [
+ 105,
+ 669,
+ 506,
+ 682
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 669,
+ 506,
+ 682
+ ],
+ "score": 1.0,
+ "content": "super-classes by computing prototype graphs from each class, followed by clustering the prototype",
+ "type": "text"
+ }
+ ],
+ "index": 41
+ },
+ {
+ "bbox": [
+ 105,
+ 681,
+ 506,
+ 693
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 681,
+ 506,
+ 693
+ ],
+ "score": 1.0,
+ "content": "graphs based on their spectral properties. This clustering of prototype graphs induces a natural",
+ "type": "text"
+ }
+ ],
+ "index": 42
+ },
+ {
+ "bbox": [
+ 105,
+ 691,
+ 505,
+ 704
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 691,
+ 505,
+ 704
+ ],
+ "score": 1.0,
+ "content": "clustering on their corresponding class labels, resulting in super-classes (as outlined in Section 4.1).",
+ "type": "text"
+ }
+ ],
+ "index": 43
+ }
+ ],
+ "index": 41.5
+ }
+ ],
+ "page_idx": 3,
+ "page_size": [
+ 612,
+ 792
+ ],
+ "discarded_blocks": [
+ {
+ "type": "discarded",
+ "bbox": [
+ 106,
+ 711,
+ 505,
+ 732
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 119,
+ 710,
+ 505,
+ 723
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 119,
+ 710,
+ 281,
+ 723
+ ],
+ "score": 1.0,
+ "content": "2In practice, extremely fast computation of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 282,
+ 711,
+ 320,
+ 722
+ ],
+ "score": 0.92,
+ "content": "W ^ { p } ( g , g ^ { \\prime } )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 320,
+ 710,
+ 505,
+ 723
+ ],
+ "score": 1.0,
+ "content": "is achieved using a regularized optimal transport",
+ "type": "text"
+ }
+ ]
+ },
+ {
+ "bbox": [
+ 106,
+ 720,
+ 371,
+ 732
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 720,
+ 371,
+ 732
+ ],
+ "score": 1.0,
+ "content": "(OT) (Genevay et al., 2016), which makes use of the Sinkhorn algorithm.",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 107,
+ 27,
+ 293,
+ 37
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "score": 1.0,
+ "content": "Published as a conference paper at ICLR 2020",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 302,
+ 751,
+ 309,
+ 759
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 301,
+ 750,
+ 310,
+ 762
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 301,
+ 750,
+ 310,
+ 762
+ ],
+ "score": 1.0,
+ "content": "4",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ }
+ ],
+ "para_blocks": [
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 82,
+ 504,
+ 105
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 104,
+ 80,
+ 506,
+ 97
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 80,
+ 506,
+ 97
+ ],
+ "score": 1.0,
+ "content": "Learning procedure: Inspired by the initialization based methods, we similarly follow a two-stage",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 106,
+ 93,
+ 290,
+ 108
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 93,
+ 290,
+ 108
+ ],
+ "score": 1.0,
+ "content": "approach of training followed by fine-tuning.",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ }
+ ],
+ "index": 0.5,
+ "bbox_fs": [
+ 104,
+ 80,
+ 506,
+ 108
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 110,
+ 505,
+ 177
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 110,
+ 505,
+ 123
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 110,
+ 309,
+ 123
+ ],
+ "score": 1.0,
+ "content": "During training, we train a graph feature extractor",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 309,
+ 110,
+ 344,
+ 122
+ ],
+ "score": 0.94,
+ "content": "F _ { \\theta } ( G _ { B } )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 344,
+ 110,
+ 447,
+ 123
+ ],
+ "score": 1.0,
+ "content": "with network parameters",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 447,
+ 111,
+ 453,
+ 120
+ ],
+ "score": 0.78,
+ "content": "\\theta",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 454,
+ 110,
+ 505,
+ 123
+ ],
+ "score": 1.0,
+ "content": "followed by",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ },
+ {
+ "bbox": [
+ 105,
+ 121,
+ 505,
+ 134
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 121,
+ 152,
+ 134
+ ],
+ "score": 1.0,
+ "content": "a classifier",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 153,
+ 122,
+ 184,
+ 133
+ ],
+ "score": 0.93,
+ "content": "C ( G _ { B } )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 184,
+ 121,
+ 250,
+ 134
+ ],
+ "score": 1.0,
+ "content": "on graphs from",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 251,
+ 122,
+ 266,
+ 132
+ ],
+ "score": 0.9,
+ "content": "G _ { B }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 267,
+ 121,
+ 505,
+ 134
+ ],
+ "score": 1.0,
+ "content": ", where the loss function is the standard cross-entropy loss",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 106,
+ 132,
+ 506,
+ 145
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 133,
+ 118,
+ 144
+ ],
+ "score": 0.87,
+ "content": "\\mathcal { L } _ { c }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 119,
+ 132,
+ 506,
+ 145
+ ],
+ "score": 1.0,
+ "content": ". In order to better recognize and generalize well on samples from novel classes, in the fine-",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 105,
+ 143,
+ 505,
+ 155
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 143,
+ 294,
+ 155
+ ],
+ "score": 1.0,
+ "content": "tuning phase, the pre-trained feature extractor",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 295,
+ 143,
+ 317,
+ 155
+ ],
+ "score": 0.92,
+ "content": "F _ { \\theta } ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 318,
+ 143,
+ 505,
+ 155
+ ],
+ "score": 1.0,
+ "content": "along with its trained parameters is fixed and",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 105,
+ 153,
+ 506,
+ 168
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 153,
+ 160,
+ 168
+ ],
+ "score": 1.0,
+ "content": "the classifier",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 160,
+ 154,
+ 192,
+ 166
+ ],
+ "score": 0.92,
+ "content": "C ( G _ { N } )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 192,
+ 153,
+ 425,
+ 168
+ ],
+ "score": 1.0,
+ "content": "is trained on the novel class labeled graph samples from",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 425,
+ 154,
+ 441,
+ 165
+ ],
+ "score": 0.89,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 442,
+ 153,
+ 506,
+ 168
+ ],
+ "score": 1.0,
+ "content": ", with the same",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 104,
+ 162,
+ 142,
+ 179
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 162,
+ 124,
+ 179
+ ],
+ "score": 1.0,
+ "content": "loss",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 124,
+ 166,
+ 137,
+ 176
+ ],
+ "score": 0.88,
+ "content": "\\mathcal { L } _ { c }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 137,
+ 162,
+ 142,
+ 179
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ }
+ ],
+ "index": 4.5,
+ "bbox_fs": [
+ 104,
+ 110,
+ 506,
+ 179
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 182,
+ 504,
+ 204
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 182,
+ 505,
+ 194
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 182,
+ 505,
+ 194
+ ],
+ "score": 1.0,
+ "content": "Now, given the classification of data samples and the two-stage learning method, our problem of",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 106,
+ 192,
+ 345,
+ 205
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 192,
+ 345,
+ 205
+ ],
+ "score": 1.0,
+ "content": "few-shot classification on graphs can be defined as follows.",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ }
+ ],
+ "index": 8.5,
+ "bbox_fs": [
+ 106,
+ 182,
+ 505,
+ 205
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 209,
+ 505,
+ 266
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 209,
+ 504,
+ 223
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 209,
+ 220,
+ 223
+ ],
+ "score": 1.0,
+ "content": "Problem definition: Given",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 220,
+ 212,
+ 227,
+ 220
+ ],
+ "score": 0.65,
+ "content": "n",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 228,
+ 209,
+ 355,
+ 223
+ ],
+ "score": 1.0,
+ "content": "base-class labeled graphs from",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 356,
+ 210,
+ 371,
+ 221
+ ],
+ "score": 0.88,
+ "content": "G _ { B }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 372,
+ 209,
+ 494,
+ 223
+ ],
+ "score": 1.0,
+ "content": "during the training phase and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 494,
+ 212,
+ 504,
+ 220
+ ],
+ "score": 0.66,
+ "content": "m",
+ "type": "inline_equation"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 105,
+ 221,
+ 505,
+ 234
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 221,
+ 240,
+ 234
+ ],
+ "score": 1.0,
+ "content": "novel-class labeled graphs from",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 240,
+ 221,
+ 257,
+ 232
+ ],
+ "score": 0.89,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 257,
+ 221,
+ 409,
+ 234
+ ],
+ "score": 1.0,
+ "content": "during the fine-tuning phase, where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 410,
+ 222,
+ 445,
+ 232
+ ],
+ "score": 0.88,
+ "content": "m \\ll n",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 446,
+ 221,
+ 505,
+ 234
+ ],
+ "score": 1.0,
+ "content": ", the objective",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 106,
+ 232,
+ 505,
+ 244
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 232,
+ 291,
+ 244
+ ],
+ "score": 1.0,
+ "content": "of few-shot graph classification is to classify",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 291,
+ 233,
+ 297,
+ 242
+ ],
+ "score": 0.74,
+ "content": "t",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 297,
+ 232,
+ 429,
+ 244
+ ],
+ "score": 1.0,
+ "content": "unseen test graph samples from",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 430,
+ 232,
+ 445,
+ 243
+ ],
+ "score": 0.87,
+ "content": "G _ { U }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 446,
+ 232,
+ 505,
+ 244
+ ],
+ "score": 1.0,
+ "content": ". Moreover, if",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 107,
+ 242,
+ 506,
+ 256
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 243,
+ 142,
+ 254
+ ],
+ "score": 0.91,
+ "content": "m = q T",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 143,
+ 242,
+ 173,
+ 256
+ ],
+ "score": 1.0,
+ "content": ", where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 174,
+ 243,
+ 230,
+ 254
+ ],
+ "score": 0.92,
+ "content": "T = K ^ { \\prime } - K",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 230,
+ 242,
+ 405,
+ 256
+ ],
+ "score": 1.0,
+ "content": ", i.e., each novel class label appears exactly",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 405,
+ 245,
+ 412,
+ 255
+ ],
+ "score": 0.8,
+ "content": "q",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 412,
+ 242,
+ 447,
+ 256
+ ],
+ "score": 1.0,
+ "content": "times in",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 448,
+ 243,
+ 464,
+ 254
+ ],
+ "score": 0.9,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 464,
+ 242,
+ 506,
+ 256
+ ],
+ "score": 1.0,
+ "content": ", then this",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 105,
+ 253,
+ 311,
+ 268
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 253,
+ 214,
+ 268
+ ],
+ "score": 1.0,
+ "content": "setting is referred to as the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 214,
+ 255,
+ 221,
+ 265
+ ],
+ "score": 0.8,
+ "content": "q",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 221,
+ 253,
+ 244,
+ 268
+ ],
+ "score": 1.0,
+ "content": "-shot,",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 244,
+ 254,
+ 253,
+ 264
+ ],
+ "score": 0.81,
+ "content": "T",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 253,
+ 253,
+ 311,
+ 268
+ ],
+ "score": 1.0,
+ "content": "-way learning.",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ }
+ ],
+ "index": 12,
+ "bbox_fs": [
+ 105,
+ 209,
+ 506,
+ 268
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 270,
+ 505,
+ 365
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 270,
+ 505,
+ 284
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 270,
+ 338,
+ 284
+ ],
+ "score": 1.0,
+ "content": "Graph spectral distance: Let us consider the graphs in",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 338,
+ 271,
+ 346,
+ 282
+ ],
+ "score": 0.83,
+ "content": "\\mathcal { G }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 346,
+ 270,
+ 505,
+ 284
+ ],
+ "score": 1.0,
+ "content": ". The normalized Laplacian of a graph",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 106,
+ 280,
+ 506,
+ 297
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 282,
+ 135,
+ 295
+ ],
+ "score": 0.9,
+ "content": "g \\in { \\mathcal { G } }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 135,
+ 280,
+ 191,
+ 297
+ ],
+ "score": 1.0,
+ "content": "is defined as",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 191,
+ 281,
+ 295,
+ 295
+ ],
+ "score": 0.92,
+ "content": "\\Delta _ { g } = I - D ^ { - 1 / 2 } A D ^ { 1 / 2 }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 296,
+ 280,
+ 328,
+ 297
+ ],
+ "score": 1.0,
+ "content": ", where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 329,
+ 283,
+ 337,
+ 293
+ ],
+ "score": 0.81,
+ "content": "A",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 338,
+ 280,
+ 357,
+ 297
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 357,
+ 283,
+ 367,
+ 293
+ ],
+ "score": 0.83,
+ "content": "D",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 367,
+ 280,
+ 506,
+ 297
+ ],
+ "score": 1.0,
+ "content": "are the adjacency and the degree",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 101,
+ 289,
+ 505,
+ 316
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 101,
+ 289,
+ 182,
+ 316
+ ],
+ "score": 1.0,
+ "content": "matrices of graph",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 182,
+ 299,
+ 189,
+ 308
+ ],
+ "score": 0.74,
+ "content": "g",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 190,
+ 289,
+ 360,
+ 316
+ ],
+ "score": 1.0,
+ "content": ", respectively. The set of eigenvalues of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 360,
+ 297,
+ 374,
+ 309
+ ],
+ "score": 0.91,
+ "content": "\\Delta _ { g }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 375,
+ 289,
+ 416,
+ 316
+ ],
+ "score": 1.0,
+ "content": "given by",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 416,
+ 294,
+ 449,
+ 309
+ ],
+ "score": 0.92,
+ "content": "\\{ \\lambda _ { i } \\} _ { i = 1 } ^ { | V | }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 450,
+ 289,
+ 505,
+ 316
+ ],
+ "score": 1.0,
+ "content": "is called the",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 104,
+ 307,
+ 506,
+ 321
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 307,
+ 157,
+ 321
+ ],
+ "score": 1.0,
+ "content": "spectrum of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 172,
+ 307,
+ 248,
+ 321
+ ],
+ "score": 1.0,
+ "content": "and is denoted by",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 268,
+ 307,
+ 416,
+ 321
+ ],
+ "score": 1.0,
+ "content": ". It is well known that the spectrum",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 417,
+ 309,
+ 437,
+ 320
+ ],
+ "score": 0.88,
+ "content": "\\sigma ( g )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 437,
+ 307,
+ 506,
+ 321
+ ],
+ "score": 1.0,
+ "content": "of a normalized",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 105,
+ 319,
+ 505,
+ 331
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 319,
+ 274,
+ 331
+ ],
+ "score": 1.0,
+ "content": "Laplacian matrix is contained in interval",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 274,
+ 319,
+ 295,
+ 331
+ ],
+ "score": 0.47,
+ "content": "[ 0 , 2 ]",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 295,
+ 319,
+ 401,
+ 331
+ ],
+ "score": 1.0,
+ "content": ". We assign a Dirac mass",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 401,
+ 319,
+ 415,
+ 331
+ ],
+ "score": 0.88,
+ "content": "\\delta _ { \\lambda _ { i } }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 415,
+ 319,
+ 505,
+ 331
+ ],
+ "score": 1.0,
+ "content": "concentrated on each",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 106,
+ 328,
+ 506,
+ 344
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 330,
+ 152,
+ 342
+ ],
+ "score": 0.92,
+ "content": "\\lambda _ { i } \\ \\in \\ \\sigma ( g )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 153,
+ 328,
+ 330,
+ 344
+ ],
+ "score": 1.0,
+ "content": ", thus associating a probability measure to",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 330,
+ 330,
+ 350,
+ 342
+ ],
+ "score": 0.92,
+ "content": "\\sigma ( g )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 350,
+ 328,
+ 409,
+ 344
+ ],
+ "score": 1.0,
+ "content": "supported on",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 410,
+ 330,
+ 430,
+ 342
+ ],
+ "score": 0.82,
+ "content": "[ 0 , 2 ]",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 430,
+ 328,
+ 506,
+ 344
+ ],
+ "score": 1.0,
+ "content": ", called the graph",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 105,
+ 340,
+ 505,
+ 354
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 340,
+ 178,
+ 354
+ ],
+ "score": 1.0,
+ "content": "spectral measure",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 179,
+ 343,
+ 201,
+ 354
+ ],
+ "score": 0.89,
+ "content": "\\mu _ { \\sigma ( g ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 201,
+ 340,
+ 278,
+ 354
+ ],
+ "score": 1.0,
+ "content": ". Furthermore, let",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 278,
+ 341,
+ 314,
+ 353
+ ],
+ "score": 0.93,
+ "content": "P ( [ 0 , 2 ] )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 315,
+ 340,
+ 505,
+ 354
+ ],
+ "score": 1.0,
+ "content": "be the set of probability measures on interval",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 107,
+ 351,
+ 133,
+ 366
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 353,
+ 127,
+ 365
+ ],
+ "score": 0.4,
+ "content": "[ 0 , 2 ]",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 128,
+ 351,
+ 133,
+ 366
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ }
+ ],
+ "index": 18.5,
+ "bbox_fs": [
+ 101,
+ 270,
+ 506,
+ 366
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 105,
+ 369,
+ 504,
+ 393
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 369,
+ 505,
+ 383
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 369,
+ 185,
+ 383
+ ],
+ "score": 1.0,
+ "content": "We now define the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 185,
+ 371,
+ 191,
+ 381
+ ],
+ "score": 0.81,
+ "content": "p",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 192,
+ 369,
+ 505,
+ 383
+ ],
+ "score": 1.0,
+ "content": "-th Wasserstein distance between probability measures, which we later use to",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ },
+ {
+ "bbox": [
+ 106,
+ 381,
+ 321,
+ 394
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 381,
+ 321,
+ 394
+ ],
+ "score": 1.0,
+ "content": "define the spectral distance between a pair of graphs.",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ }
+ ],
+ "index": 23.5,
+ "bbox_fs": [
+ 105,
+ 369,
+ 505,
+ 394
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 402,
+ 505,
+ 436
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 402,
+ 505,
+ 415
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 402,
+ 176,
+ 415
+ ],
+ "score": 1.0,
+ "content": "Definition 1 Let",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 177,
+ 403,
+ 222,
+ 415
+ ],
+ "score": 0.91,
+ "content": "p \\in [ 1 , \\infty )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 222,
+ 402,
+ 254,
+ 415
+ ],
+ "score": 1.0,
+ "content": "and let",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 254,
+ 403,
+ 371,
+ 415
+ ],
+ "score": 0.91,
+ "content": "c : [ 0 , 2 ] \\times [ 0 , 2 ] [ 0 , + \\infty ]",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 371,
+ 402,
+ 505,
+ 415
+ ],
+ "score": 1.0,
+ "content": "be the cost function between the",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 104,
+ 414,
+ 506,
+ 427
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 414,
+ 193,
+ 427
+ ],
+ "score": 1.0,
+ "content": "probability measures",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 193,
+ 414,
+ 257,
+ 426
+ ],
+ "score": 0.92,
+ "content": "\\mu , \\nu \\in P ( [ 0 , 2 ] )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 258,
+ 414,
+ 299,
+ 427
+ ],
+ "score": 1.0,
+ "content": ". Then the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 300,
+ 416,
+ 306,
+ 426
+ ],
+ "score": 0.72,
+ "content": "p",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 306,
+ 414,
+ 479,
+ 427
+ ],
+ "score": 1.0,
+ "content": "-th Wasserstein distance between measures",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 479,
+ 416,
+ 486,
+ 426
+ ],
+ "score": 0.71,
+ "content": "\\mu",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 487,
+ 414,
+ 506,
+ 427
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 106,
+ 425,
+ 160,
+ 439
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 427,
+ 114,
+ 435
+ ],
+ "score": 0.64,
+ "content": "\\nu",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 114,
+ 425,
+ 160,
+ 439
+ ],
+ "score": 1.0,
+ "content": "is given by",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ }
+ ],
+ "index": 26,
+ "bbox_fs": [
+ 104,
+ 402,
+ 506,
+ 439
+ ]
+ },
+ {
+ "type": "interline_equation",
+ "bbox": [
+ 188,
+ 435,
+ 421,
+ 472
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 188,
+ 435,
+ 421,
+ 472
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 188,
+ 435,
+ 421,
+ 472
+ ],
+ "score": 0.93,
+ "content": "W _ { p } ( \\mu , \\nu ) = \\left( \\operatorname* { i n f } _ { \\gamma } \\int _ { [ 0 , 2 ] \\times [ 0 , 2 ] } c ( x , y ) ^ { p } d \\gamma \\mid \\gamma \\in \\Pi ( \\mu , \\nu ) \\right) ^ { \\frac { 1 } { p } }",
+ "type": "interline_equation",
+ "image_path": "34613c1c0832df651fd54184395645c27fd1a9bf9f646499c259923761e9bd54.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 28.5,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 188,
+ 435,
+ 421,
+ 453.5
+ ],
+ "spans": [],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 188,
+ 453.5,
+ 421,
+ 472.0
+ ],
+ "spans": [],
+ "index": 29
+ }
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 474,
+ 505,
+ 497
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 473,
+ 505,
+ 487
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 473,
+ 133,
+ 487
+ ],
+ "score": 1.0,
+ "content": "where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 133,
+ 474,
+ 165,
+ 487
+ ],
+ "score": 0.93,
+ "content": "\\Pi ( \\mu , \\nu )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 165,
+ 473,
+ 431,
+ 487
+ ],
+ "score": 1.0,
+ "content": "is the set of transport plans, i.e., the collection of all measures on",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 431,
+ 474,
+ 484,
+ 487
+ ],
+ "score": 0.91,
+ "content": "[ 0 , 2 ] \\times [ 0 , 2 ]",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 484,
+ 473,
+ 505,
+ 487
+ ],
+ "score": 1.0,
+ "content": "with",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 105,
+ 486,
+ 186,
+ 498
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 486,
+ 149,
+ 498
+ ],
+ "score": 1.0,
+ "content": "marginals",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 149,
+ 488,
+ 156,
+ 497
+ ],
+ "score": 0.79,
+ "content": "\\mu",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 157,
+ 486,
+ 175,
+ 498
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 175,
+ 488,
+ 181,
+ 495
+ ],
+ "score": 0.73,
+ "content": "\\nu",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 182,
+ 486,
+ 186,
+ 498
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ }
+ ],
+ "index": 30.5,
+ "bbox_fs": [
+ 105,
+ 473,
+ 505,
+ 498
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 105,
+ 507,
+ 505,
+ 531
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 507,
+ 505,
+ 520
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 507,
+ 246,
+ 520
+ ],
+ "score": 1.0,
+ "content": "Given the general definition of the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 246,
+ 509,
+ 253,
+ 519
+ ],
+ "score": 0.83,
+ "content": "p",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 254,
+ 507,
+ 505,
+ 520
+ ],
+ "score": 1.0,
+ "content": "-th Wasserstein distance between probability measures and the",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 105,
+ 518,
+ 487,
+ 531
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 518,
+ 476,
+ 531
+ ],
+ "score": 1.0,
+ "content": "graph spectral measure, we can now define the spectral distance between a pair of graphs in",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 476,
+ 519,
+ 484,
+ 529
+ ],
+ "score": 0.83,
+ "content": "\\mathcal { G }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 484,
+ 518,
+ 487,
+ 531
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ }
+ ],
+ "index": 32.5,
+ "bbox_fs": [
+ 105,
+ 507,
+ 505,
+ 531
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 103,
+ 538,
+ 469,
+ 552
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 538,
+ 469,
+ 553
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 538,
+ 234,
+ 553
+ ],
+ "score": 1.0,
+ "content": "Definition 2 Given two graphs",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 234,
+ 540,
+ 271,
+ 551
+ ],
+ "score": 0.92,
+ "content": "g , g ^ { \\prime } \\in \\mathcal { G }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 272,
+ 538,
+ 469,
+ 553
+ ],
+ "score": 1.0,
+ "content": ", the spectral distance between them is defined as",
+ "type": "text"
+ }
+ ],
+ "index": 34
+ }
+ ],
+ "index": 34,
+ "bbox_fs": [
+ 105,
+ 538,
+ 469,
+ 553
+ ]
+ },
+ {
+ "type": "interline_equation",
+ "bbox": [
+ 238,
+ 556,
+ 372,
+ 572
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 238,
+ 556,
+ 372,
+ 572
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 238,
+ 556,
+ 372,
+ 572
+ ],
+ "score": 0.92,
+ "content": "W ^ { p } ( g , g ^ { \\prime } ) : = W _ { p } \\left( \\mu _ { \\sigma ( g ) } , \\mu _ { \\sigma ( g ^ { \\prime } ) } \\right)",
+ "type": "interline_equation",
+ "image_path": "d89709e1a2a9eee3ee2245c73557bd920b69e751fe1e8e0d895a71cecea20d9f.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 35,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 238,
+ 556,
+ 372,
+ 572
+ ],
+ "spans": [],
+ "index": 35
+ }
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 582,
+ 504,
+ 617
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 582,
+ 506,
+ 596
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 582,
+ 147,
+ 596
+ ],
+ "score": 1.0,
+ "content": "In words,",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 148,
+ 583,
+ 189,
+ 595
+ ],
+ "score": 0.92,
+ "content": "W ^ { p } ( g , g ^ { \\prime } )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 189,
+ 582,
+ 506,
+ 596
+ ],
+ "score": 1.0,
+ "content": "is the optimal cost of moving mass from the graph spectral measure of graph",
+ "type": "text"
+ }
+ ],
+ "index": 36
+ },
+ {
+ "bbox": [
+ 107,
+ 594,
+ 506,
+ 606
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 597,
+ 113,
+ 606
+ ],
+ "score": 0.75,
+ "content": "g",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 113,
+ 594,
+ 181,
+ 606
+ ],
+ "score": 1.0,
+ "content": "to that of graph",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 181,
+ 595,
+ 190,
+ 606
+ ],
+ "score": 0.86,
+ "content": "g ^ { \\prime }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 190,
+ 594,
+ 432,
+ 606
+ ],
+ "score": 1.0,
+ "content": ", where the cost of moving unit mass is proportional to the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 432,
+ 596,
+ 438,
+ 606
+ ],
+ "score": 0.81,
+ "content": "p",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 438,
+ 594,
+ 506,
+ 606
+ ],
+ "score": 1.0,
+ "content": "-th power of the",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 105,
+ 604,
+ 300,
+ 618
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 604,
+ 270,
+ 618
+ ],
+ "score": 1.0,
+ "content": "difference of real-eigenvalues in interval",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 271,
+ 605,
+ 295,
+ 617
+ ],
+ "score": 0.92,
+ "content": "[ 0 , 2 ] ^ { 2 }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 295,
+ 604,
+ 300,
+ 618
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ }
+ ],
+ "index": 37,
+ "bbox_fs": [
+ 105,
+ 582,
+ 506,
+ 618
+ ]
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 107,
+ 632,
+ 200,
+ 645
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 630,
+ 201,
+ 648
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 630,
+ 201,
+ 648
+ ],
+ "score": 1.0,
+ "content": "4 OUR METHOD",
+ "type": "text"
+ }
+ ],
+ "index": 39
+ }
+ ],
+ "index": 39
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 658,
+ 505,
+ 703
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 658,
+ 506,
+ 671
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 658,
+ 506,
+ 671
+ ],
+ "score": 1.0,
+ "content": "We present our proposed approach here. First, given abundant base-class labels, we cluster them into",
+ "type": "text"
+ }
+ ],
+ "index": 40
+ },
+ {
+ "bbox": [
+ 105,
+ 669,
+ 506,
+ 682
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 669,
+ 506,
+ 682
+ ],
+ "score": 1.0,
+ "content": "super-classes by computing prototype graphs from each class, followed by clustering the prototype",
+ "type": "text"
+ }
+ ],
+ "index": 41
+ },
+ {
+ "bbox": [
+ 105,
+ 681,
+ 506,
+ 693
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 681,
+ 506,
+ 693
+ ],
+ "score": 1.0,
+ "content": "graphs based on their spectral properties. This clustering of prototype graphs induces a natural",
+ "type": "text"
+ }
+ ],
+ "index": 42
+ },
+ {
+ "bbox": [
+ 105,
+ 691,
+ 505,
+ 704
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 691,
+ 505,
+ 704
+ ],
+ "score": 1.0,
+ "content": "clustering on their corresponding class labels, resulting in super-classes (as outlined in Section 4.1).",
+ "type": "text"
+ }
+ ],
+ "index": 43
+ }
+ ],
+ "index": 41.5,
+ "bbox_fs": [
+ 105,
+ 658,
+ 506,
+ 704
+ ]
+ }
+ ]
+ },
+ {
+ "preproc_blocks": [
+ {
+ "type": "image",
+ "bbox": [
+ 109,
+ 83,
+ 503,
+ 194
+ ],
+ "blocks": [
+ {
+ "type": "image_body",
+ "bbox": [
+ 109,
+ 83,
+ 503,
+ 194
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 109,
+ 83,
+ 503,
+ 194
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 109,
+ 83,
+ 503,
+ 194
+ ],
+ "score": 0.968,
+ "type": "image",
+ "image_path": "393d3119a11d19636ddb73ae543b4196b5496b3b1defea32cbff310a102d2d91.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 1,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 109,
+ 83,
+ 503,
+ 120.0
+ ],
+ "spans": [],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 109,
+ 120.0,
+ 503,
+ 157.0
+ ],
+ "spans": [],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 109,
+ 157.0,
+ 503,
+ 194.0
+ ],
+ "spans": [],
+ "index": 2
+ }
+ ]
+ },
+ {
+ "type": "image_caption",
+ "bbox": [
+ 163,
+ 203,
+ 447,
+ 216
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 163,
+ 203,
+ 447,
+ 217
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 163,
+ 203,
+ 447,
+ 217
+ ],
+ "score": 1.0,
+ "content": "Figure 1: The training (left) and fine-tuning (right) stages of our GNN.",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ }
+ ],
+ "index": 3
+ }
+ ],
+ "index": 2.0
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 236,
+ 505,
+ 281
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 236,
+ 505,
+ 249
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 236,
+ 505,
+ 249
+ ],
+ "score": 1.0,
+ "content": "These super-classes are then used in the creation of a super-graph used further down by our GNN.",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 105,
+ 247,
+ 505,
+ 261
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 247,
+ 505,
+ 261
+ ],
+ "score": 1.0,
+ "content": "Note that the creation of super-classes, followed by building a super-graph are a once-off process.",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 104,
+ 257,
+ 506,
+ 272
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 257,
+ 506,
+ 272
+ ],
+ "score": 1.0,
+ "content": "The prototype graphs as well as the super-classes for the base classes can be stored in memory for",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 106,
+ 269,
+ 155,
+ 282
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 269,
+ 155,
+ 282
+ ],
+ "score": 1.0,
+ "content": "further use.",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ }
+ ],
+ "index": 5.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 286,
+ 505,
+ 342
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 287,
+ 505,
+ 299
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 287,
+ 505,
+ 299
+ ],
+ "score": 1.0,
+ "content": "Next, we explain our graph neural network’s architecture which comprises of a feature extractor",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 107,
+ 297,
+ 506,
+ 310
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 298,
+ 129,
+ 309
+ ],
+ "score": 0.91,
+ "content": "F _ { \\theta } ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 129,
+ 297,
+ 193,
+ 310
+ ],
+ "score": 1.0,
+ "content": "and a classifier",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 193,
+ 298,
+ 213,
+ 309
+ ],
+ "score": 0.91,
+ "content": "C ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 213,
+ 297,
+ 374,
+ 310
+ ],
+ "score": 1.0,
+ "content": ", described in Section 4.2. The classifier",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 375,
+ 298,
+ 394,
+ 309
+ ],
+ "score": 0.92,
+ "content": "C ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 394,
+ 297,
+ 506,
+ 310
+ ],
+ "score": 1.0,
+ "content": "is further subdivided into a",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 105,
+ 307,
+ 506,
+ 321
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 307,
+ 145,
+ 321
+ ],
+ "score": 1.0,
+ "content": "classifier",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 145,
+ 309,
+ 167,
+ 318
+ ],
+ "score": 0.87,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 167,
+ 307,
+ 506,
+ 321
+ ],
+ "score": 1.0,
+ "content": "that predicts the superclass of a graph feature vector and a graph attention network",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 104,
+ 316,
+ 506,
+ 335
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 316,
+ 136,
+ 335
+ ],
+ "score": 1.0,
+ "content": "(GAT)",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 136,
+ 319,
+ 164,
+ 330
+ ],
+ "score": 0.88,
+ "content": "C ^ { G A T }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 164,
+ 316,
+ 506,
+ 335
+ ],
+ "score": 1.0,
+ "content": "to predict the graph’s class label. Figure 1 illustrates the training and fine-tuning",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 105,
+ 330,
+ 189,
+ 342
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 330,
+ 189,
+ 342
+ ],
+ "score": 1.0,
+ "content": "phases of our GNN.",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ }
+ ],
+ "index": 10
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 356,
+ 256,
+ 367
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 356,
+ 257,
+ 368
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 356,
+ 257,
+ 368
+ ],
+ "score": 1.0,
+ "content": "4.1 COMPUTING SUPER CLASSES",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ }
+ ],
+ "index": 13
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 376,
+ 505,
+ 414
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 376,
+ 504,
+ 389
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 376,
+ 504,
+ 389
+ ],
+ "score": 1.0,
+ "content": "In order to exploit inter-class relationships between base-class labels, we cluster them in the follow-",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 104,
+ 387,
+ 503,
+ 401
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 388,
+ 260,
+ 401
+ ],
+ "score": 1.0,
+ "content": "ing manner. First, we partition the set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 261,
+ 389,
+ 276,
+ 400
+ ],
+ "score": 0.89,
+ "content": "G _ { B }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 276,
+ 388,
+ 369,
+ 401
+ ],
+ "score": 1.0,
+ "content": "into class-specific sets",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 370,
+ 387,
+ 388,
+ 399
+ ],
+ "score": 0.9,
+ "content": "G ^ { ( i ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 388,
+ 388,
+ 406,
+ 401
+ ],
+ "score": 1.0,
+ "content": ", for",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 406,
+ 389,
+ 454,
+ 399
+ ],
+ "score": 0.88,
+ "content": "i = 1 \\dots K",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 454,
+ 388,
+ 485,
+ 401
+ ],
+ "score": 1.0,
+ "content": ", where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 486,
+ 388,
+ 503,
+ 399
+ ],
+ "score": 0.91,
+ "content": "G ^ { ( i ) }",
+ "type": "inline_equation"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 101,
+ 394,
+ 375,
+ 419
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 101,
+ 394,
+ 268,
+ 419
+ ],
+ "score": 1.0,
+ "content": "is the set of graphs with base-class label",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 269,
+ 402,
+ 273,
+ 411
+ ],
+ "score": 0.69,
+ "content": "i",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 273,
+ 394,
+ 303,
+ 419
+ ],
+ "score": 1.0,
+ "content": ". Thus,",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 303,
+ 400,
+ 372,
+ 415
+ ],
+ "score": 0.93,
+ "content": "\\begin{array} { r } { G _ { B } = \\bigsqcup _ { i = 1 } ^ { K } G ^ { ( i ) } } \\end{array}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 373,
+ 394,
+ 375,
+ 419
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ }
+ ],
+ "index": 15
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 418,
+ 504,
+ 442
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 418,
+ 505,
+ 431
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 418,
+ 505,
+ 431
+ ],
+ "score": 1.0,
+ "content": "Then, we compute class prototype graphs for each class-specific set. The class prototype graph for",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 105,
+ 429,
+ 252,
+ 443
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 429,
+ 128,
+ 443
+ ],
+ "score": 1.0,
+ "content": "class",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 128,
+ 430,
+ 133,
+ 439
+ ],
+ "score": 0.75,
+ "content": "i",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 133,
+ 429,
+ 195,
+ 443
+ ],
+ "score": 1.0,
+ "content": "represented by",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 195,
+ 431,
+ 205,
+ 441
+ ],
+ "score": 0.84,
+ "content": "p _ { i }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 205,
+ 429,
+ 252,
+ 443
+ ],
+ "score": 1.0,
+ "content": "is given by",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ }
+ ],
+ "index": 17.5
+ },
+ {
+ "type": "interline_equation",
+ "bbox": [
+ 225,
+ 447,
+ 386,
+ 485
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 225,
+ 447,
+ 386,
+ 485
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 225,
+ 447,
+ 386,
+ 485
+ ],
+ "score": 0.95,
+ "content": "p _ { i } = \\operatorname * { a r g m i n } _ { g _ { i } \\in \\pi _ { 1 } ( G ^ { ( i ) } ) } \\frac { 1 } { | G ^ { ( i ) } | } \\sum _ { j = 1 } ^ { | G ^ { ( i ) } | } W ^ { p } ( g _ { i } , g _ { j } )",
+ "type": "interline_equation",
+ "image_path": "184a8c2a4d8c67644a73819913d2c42712d2ae5fc43d95dd4c9e69267eebce09.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 19.5,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 225,
+ 447,
+ 386,
+ 466.0
+ ],
+ "spans": [],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 225,
+ 466.0,
+ 386,
+ 485.0
+ ],
+ "spans": [],
+ "index": 20
+ }
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 490,
+ 505,
+ 523
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 489,
+ 505,
+ 503
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 489,
+ 256,
+ 503
+ ],
+ "score": 1.0,
+ "content": "Essentially, the class prototype graph",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 256,
+ 492,
+ 266,
+ 501
+ ],
+ "score": 0.85,
+ "content": "p _ { i }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 266,
+ 489,
+ 295,
+ 503
+ ],
+ "score": 1.0,
+ "content": "for the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 296,
+ 491,
+ 300,
+ 500
+ ],
+ "score": 0.77,
+ "content": "i",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 301,
+ 489,
+ 505,
+ 503
+ ],
+ "score": 1.0,
+ "content": "-th class is the graph with the least average spectral",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 105,
+ 501,
+ 505,
+ 514
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 501,
+ 357,
+ 514
+ ],
+ "score": 1.0,
+ "content": "distance to the rest of the graphs in the same class. Given these",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 358,
+ 502,
+ 368,
+ 511
+ ],
+ "score": 0.84,
+ "content": "K",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 369,
+ 501,
+ 505,
+ 514
+ ],
+ "score": 1.0,
+ "content": "prototypes, we cluster them using",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 105,
+ 512,
+ 280,
+ 523
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 512,
+ 234,
+ 523
+ ],
+ "score": 1.0,
+ "content": "Lloyd’s method (also known as",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 234,
+ 513,
+ 240,
+ 522
+ ],
+ "score": 0.8,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 241,
+ 512,
+ 280,
+ 523
+ ],
+ "score": 1.0,
+ "content": "-means)3.",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ }
+ ],
+ "index": 22
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 528,
+ 505,
+ 574
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 528,
+ 506,
+ 542
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 528,
+ 266,
+ 542
+ ],
+ "score": 1.0,
+ "content": "Clustering prototype graphs: Given",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 267,
+ 529,
+ 277,
+ 539
+ ],
+ "score": 0.82,
+ "content": "K",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 277,
+ 528,
+ 368,
+ 542
+ ],
+ "score": 1.0,
+ "content": "unlabeled prototypes",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 368,
+ 529,
+ 464,
+ 541
+ ],
+ "score": 0.9,
+ "content": "p _ { 1 } , \\dotsc , p _ { K } \\in \\pi _ { 1 } ( G _ { B } )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 464,
+ 528,
+ 506,
+ 542
+ ],
+ "score": 1.0,
+ "content": "and their",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 104,
+ 539,
+ 506,
+ 555
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 539,
+ 225,
+ 555
+ ],
+ "score": 1.0,
+ "content": "associated spectral measures",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 225,
+ 540,
+ 351,
+ 553
+ ],
+ "score": 0.91,
+ "content": " { \\mu } _ { \\sigma ( p _ { 1 } ) } , \\ldots , { \\mu } _ { \\sigma ( p _ { K } ) } \\in P ( [ 0 , 2 ] )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 351,
+ 539,
+ 506,
+ 555
+ ],
+ "score": 1.0,
+ "content": ". We rename the spectral measures as",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 107,
+ 551,
+ 504,
+ 563
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 553,
+ 151,
+ 562
+ ],
+ "score": 0.9,
+ "content": "s _ { 1 } , \\ldots , s _ { K }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 152,
+ 551,
+ 497,
+ 563
+ ],
+ "score": 1.0,
+ "content": "to ease notation. Thus, our goal is to associate these spectral measures to at most",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 497,
+ 551,
+ 504,
+ 561
+ ],
+ "score": 0.77,
+ "content": "k",
+ "type": "inline_equation"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 105,
+ 561,
+ 307,
+ 574
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 561,
+ 168,
+ 574
+ ],
+ "score": 1.0,
+ "content": "clusters, where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 169,
+ 562,
+ 194,
+ 573
+ ],
+ "score": 0.91,
+ "content": "k \\geq 1",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 194,
+ 561,
+ 307,
+ 574
+ ],
+ "score": 1.0,
+ "content": "is a user defined parameter.",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ }
+ ],
+ "index": 25.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 578,
+ 504,
+ 602
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 577,
+ 505,
+ 592
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 577,
+ 124,
+ 592
+ ],
+ "score": 1.0,
+ "content": "The",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 124,
+ 579,
+ 131,
+ 589
+ ],
+ "score": 0.81,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 132,
+ 577,
+ 225,
+ 592
+ ],
+ "score": 1.0,
+ "content": "-means problem finds a",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 226,
+ 579,
+ 232,
+ 589
+ ],
+ "score": 0.79,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 232,
+ 577,
+ 271,
+ 592
+ ],
+ "score": 1.0,
+ "content": "-partition",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 271,
+ 579,
+ 349,
+ 591
+ ],
+ "score": 0.95,
+ "content": "C = \\{ C _ { 1 } , \\ldots , C _ { k } \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 349,
+ 577,
+ 505,
+ 592
+ ],
+ "score": 1.0,
+ "content": "that minimizes the following objective",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 105,
+ 588,
+ 342,
+ 603
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 588,
+ 342,
+ 603
+ ],
+ "score": 1.0,
+ "content": "that represents the overall distortion error of the clustering",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ }
+ ],
+ "index": 28.5
+ },
+ {
+ "type": "interline_equation",
+ "bbox": [
+ 237,
+ 606,
+ 372,
+ 642
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 237,
+ 606,
+ 372,
+ 642
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 237,
+ 606,
+ 372,
+ 642
+ ],
+ "score": 0.94,
+ "content": "\\underset { C } { \\operatorname { a r g m i n } } \\sum _ { i = 1 } ^ { k } \\sum _ { s _ { i } \\in C _ { i } } W _ { p } ( s _ { i } , B ( C _ { i } ) )",
+ "type": "interline_equation",
+ "image_path": "4073ea7dd99c2452de4f406f1a96a9e85bbfa50c9d7fcdccc4fe50e72615b66f.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 30.5,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 237,
+ 606,
+ 372,
+ 624.0
+ ],
+ "spans": [],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 237,
+ 624.0,
+ 372,
+ 642.0
+ ],
+ "spans": [],
+ "index": 31
+ }
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 108,
+ 647,
+ 505,
+ 671
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 646,
+ 506,
+ 662
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 646,
+ 133,
+ 662
+ ],
+ "score": 1.0,
+ "content": "where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 134,
+ 650,
+ 143,
+ 659
+ ],
+ "score": 0.85,
+ "content": "s _ { i }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 143,
+ 646,
+ 269,
+ 662
+ ],
+ "score": 1.0,
+ "content": "is a prototype graph in cluster",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 269,
+ 648,
+ 281,
+ 659
+ ],
+ "score": 0.89,
+ "content": "C _ { i }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 281,
+ 646,
+ 300,
+ 662
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 300,
+ 648,
+ 327,
+ 660
+ ],
+ "score": 0.93,
+ "content": "B ( C _ { i } )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 327,
+ 646,
+ 506,
+ 662
+ ],
+ "score": 1.0,
+ "content": "is the Wasserstein barycenter of the cluster",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 107,
+ 659,
+ 248,
+ 672
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 659,
+ 118,
+ 670
+ ],
+ "score": 0.86,
+ "content": "C _ { i }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 118,
+ 659,
+ 248,
+ 672
+ ],
+ "score": 1.0,
+ "content": ". The barycenter is computed as",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ }
+ ],
+ "index": 32.5
+ },
+ {
+ "type": "interline_equation",
+ "bbox": [
+ 229,
+ 676,
+ 382,
+ 713
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 229,
+ 676,
+ 382,
+ 713
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 229,
+ 676,
+ 382,
+ 713
+ ],
+ "score": 0.95,
+ "content": "B ( C _ { i } ) = \\operatorname * { a r g m i n } _ { p \\in P ( [ 0 , 2 ] ) } \\sum _ { j = 1 } ^ { | C _ { i } | } W _ { p } ( p , s ( i , j ) )",
+ "type": "interline_equation",
+ "image_path": "e41f71ea816eed4ff86ff64b32490bcc3db78eef95e6867af8384888b318d74e.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 34.5,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 229,
+ 676,
+ 382,
+ 694.5
+ ],
+ "spans": [],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 229,
+ 694.5,
+ 382,
+ 713.0
+ ],
+ "spans": [],
+ "index": 35
+ }
+ ]
+ }
+ ],
+ "page_idx": 4,
+ "page_size": [
+ 612,
+ 792
+ ],
+ "discarded_blocks": [
+ {
+ "type": "discarded",
+ "bbox": [
+ 107,
+ 27,
+ 293,
+ 37
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "score": 1.0,
+ "content": "Published as a conference paper at ICLR 2020",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 118,
+ 721,
+ 428,
+ 732
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 118,
+ 719,
+ 428,
+ 734
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 118,
+ 719,
+ 274,
+ 734
+ ],
+ "score": 1.0,
+ "content": "3We used the seeding method suggested in",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 275,
+ 722,
+ 280,
+ 730
+ ],
+ "score": 0.81,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 280,
+ 719,
+ 311,
+ 734
+ ],
+ "score": 1.0,
+ "content": "-means+",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 311,
+ 725,
+ 316,
+ 730
+ ],
+ "score": 0.42,
+ "content": "^ +",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 317,
+ 719,
+ 428,
+ 734
+ ],
+ "score": 1.0,
+ "content": "(Arthur & Vassilvitskii, 2007)",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 302,
+ 751,
+ 309,
+ 760
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 301,
+ 750,
+ 310,
+ 762
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 301,
+ 750,
+ 310,
+ 762
+ ],
+ "score": 1.0,
+ "content": "5",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ }
+ ],
+ "para_blocks": [
+ {
+ "type": "image",
+ "bbox": [
+ 109,
+ 83,
+ 503,
+ 194
+ ],
+ "blocks": [
+ {
+ "type": "image_body",
+ "bbox": [
+ 109,
+ 83,
+ 503,
+ 194
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 109,
+ 83,
+ 503,
+ 194
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 109,
+ 83,
+ 503,
+ 194
+ ],
+ "score": 0.968,
+ "type": "image",
+ "image_path": "393d3119a11d19636ddb73ae543b4196b5496b3b1defea32cbff310a102d2d91.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 1,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 109,
+ 83,
+ 503,
+ 120.0
+ ],
+ "spans": [],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 109,
+ 120.0,
+ 503,
+ 157.0
+ ],
+ "spans": [],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 109,
+ 157.0,
+ 503,
+ 194.0
+ ],
+ "spans": [],
+ "index": 2
+ }
+ ]
+ },
+ {
+ "type": "image_caption",
+ "bbox": [
+ 163,
+ 203,
+ 447,
+ 216
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 163,
+ 203,
+ 447,
+ 217
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 163,
+ 203,
+ 447,
+ 217
+ ],
+ "score": 1.0,
+ "content": "Figure 1: The training (left) and fine-tuning (right) stages of our GNN.",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ }
+ ],
+ "index": 3
+ }
+ ],
+ "index": 2.0
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 236,
+ 505,
+ 281
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 236,
+ 505,
+ 249
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 236,
+ 505,
+ 249
+ ],
+ "score": 1.0,
+ "content": "These super-classes are then used in the creation of a super-graph used further down by our GNN.",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 105,
+ 247,
+ 505,
+ 261
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 247,
+ 505,
+ 261
+ ],
+ "score": 1.0,
+ "content": "Note that the creation of super-classes, followed by building a super-graph are a once-off process.",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 104,
+ 257,
+ 506,
+ 272
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 257,
+ 506,
+ 272
+ ],
+ "score": 1.0,
+ "content": "The prototype graphs as well as the super-classes for the base classes can be stored in memory for",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 106,
+ 269,
+ 155,
+ 282
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 269,
+ 155,
+ 282
+ ],
+ "score": 1.0,
+ "content": "further use.",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ }
+ ],
+ "index": 5.5,
+ "bbox_fs": [
+ 104,
+ 236,
+ 506,
+ 282
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 286,
+ 505,
+ 342
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 287,
+ 505,
+ 299
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 287,
+ 505,
+ 299
+ ],
+ "score": 1.0,
+ "content": "Next, we explain our graph neural network’s architecture which comprises of a feature extractor",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 107,
+ 297,
+ 506,
+ 310
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 298,
+ 129,
+ 309
+ ],
+ "score": 0.91,
+ "content": "F _ { \\theta } ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 129,
+ 297,
+ 193,
+ 310
+ ],
+ "score": 1.0,
+ "content": "and a classifier",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 193,
+ 298,
+ 213,
+ 309
+ ],
+ "score": 0.91,
+ "content": "C ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 213,
+ 297,
+ 374,
+ 310
+ ],
+ "score": 1.0,
+ "content": ", described in Section 4.2. The classifier",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 375,
+ 298,
+ 394,
+ 309
+ ],
+ "score": 0.92,
+ "content": "C ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 394,
+ 297,
+ 506,
+ 310
+ ],
+ "score": 1.0,
+ "content": "is further subdivided into a",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 105,
+ 307,
+ 506,
+ 321
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 307,
+ 145,
+ 321
+ ],
+ "score": 1.0,
+ "content": "classifier",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 145,
+ 309,
+ 167,
+ 318
+ ],
+ "score": 0.87,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 167,
+ 307,
+ 506,
+ 321
+ ],
+ "score": 1.0,
+ "content": "that predicts the superclass of a graph feature vector and a graph attention network",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 104,
+ 316,
+ 506,
+ 335
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 316,
+ 136,
+ 335
+ ],
+ "score": 1.0,
+ "content": "(GAT)",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 136,
+ 319,
+ 164,
+ 330
+ ],
+ "score": 0.88,
+ "content": "C ^ { G A T }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 164,
+ 316,
+ 506,
+ 335
+ ],
+ "score": 1.0,
+ "content": "to predict the graph’s class label. Figure 1 illustrates the training and fine-tuning",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 105,
+ 330,
+ 189,
+ 342
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 330,
+ 189,
+ 342
+ ],
+ "score": 1.0,
+ "content": "phases of our GNN.",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ }
+ ],
+ "index": 10,
+ "bbox_fs": [
+ 104,
+ 287,
+ 506,
+ 342
+ ]
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 356,
+ 256,
+ 367
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 356,
+ 257,
+ 368
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 356,
+ 257,
+ 368
+ ],
+ "score": 1.0,
+ "content": "4.1 COMPUTING SUPER CLASSES",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ }
+ ],
+ "index": 13
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 376,
+ 505,
+ 414
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 376,
+ 504,
+ 389
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 376,
+ 504,
+ 389
+ ],
+ "score": 1.0,
+ "content": "In order to exploit inter-class relationships between base-class labels, we cluster them in the follow-",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 104,
+ 387,
+ 503,
+ 401
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 388,
+ 260,
+ 401
+ ],
+ "score": 1.0,
+ "content": "ing manner. First, we partition the set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 261,
+ 389,
+ 276,
+ 400
+ ],
+ "score": 0.89,
+ "content": "G _ { B }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 276,
+ 388,
+ 369,
+ 401
+ ],
+ "score": 1.0,
+ "content": "into class-specific sets",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 370,
+ 387,
+ 388,
+ 399
+ ],
+ "score": 0.9,
+ "content": "G ^ { ( i ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 388,
+ 388,
+ 406,
+ 401
+ ],
+ "score": 1.0,
+ "content": ", for",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 406,
+ 389,
+ 454,
+ 399
+ ],
+ "score": 0.88,
+ "content": "i = 1 \\dots K",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 454,
+ 388,
+ 485,
+ 401
+ ],
+ "score": 1.0,
+ "content": ", where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 486,
+ 388,
+ 503,
+ 399
+ ],
+ "score": 0.91,
+ "content": "G ^ { ( i ) }",
+ "type": "inline_equation"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 101,
+ 394,
+ 375,
+ 419
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 101,
+ 394,
+ 268,
+ 419
+ ],
+ "score": 1.0,
+ "content": "is the set of graphs with base-class label",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 269,
+ 402,
+ 273,
+ 411
+ ],
+ "score": 0.69,
+ "content": "i",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 273,
+ 394,
+ 303,
+ 419
+ ],
+ "score": 1.0,
+ "content": ". Thus,",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 303,
+ 400,
+ 372,
+ 415
+ ],
+ "score": 0.93,
+ "content": "\\begin{array} { r } { G _ { B } = \\bigsqcup _ { i = 1 } ^ { K } G ^ { ( i ) } } \\end{array}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 373,
+ 394,
+ 375,
+ 419
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ }
+ ],
+ "index": 15,
+ "bbox_fs": [
+ 101,
+ 376,
+ 504,
+ 419
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 418,
+ 504,
+ 442
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 418,
+ 505,
+ 431
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 418,
+ 505,
+ 431
+ ],
+ "score": 1.0,
+ "content": "Then, we compute class prototype graphs for each class-specific set. The class prototype graph for",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 105,
+ 429,
+ 252,
+ 443
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 429,
+ 128,
+ 443
+ ],
+ "score": 1.0,
+ "content": "class",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 128,
+ 430,
+ 133,
+ 439
+ ],
+ "score": 0.75,
+ "content": "i",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 133,
+ 429,
+ 195,
+ 443
+ ],
+ "score": 1.0,
+ "content": "represented by",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 195,
+ 431,
+ 205,
+ 441
+ ],
+ "score": 0.84,
+ "content": "p _ { i }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 205,
+ 429,
+ 252,
+ 443
+ ],
+ "score": 1.0,
+ "content": "is given by",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ }
+ ],
+ "index": 17.5,
+ "bbox_fs": [
+ 105,
+ 418,
+ 505,
+ 443
+ ]
+ },
+ {
+ "type": "interline_equation",
+ "bbox": [
+ 225,
+ 447,
+ 386,
+ 485
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 225,
+ 447,
+ 386,
+ 485
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 225,
+ 447,
+ 386,
+ 485
+ ],
+ "score": 0.95,
+ "content": "p _ { i } = \\operatorname * { a r g m i n } _ { g _ { i } \\in \\pi _ { 1 } ( G ^ { ( i ) } ) } \\frac { 1 } { | G ^ { ( i ) } | } \\sum _ { j = 1 } ^ { | G ^ { ( i ) } | } W ^ { p } ( g _ { i } , g _ { j } )",
+ "type": "interline_equation",
+ "image_path": "184a8c2a4d8c67644a73819913d2c42712d2ae5fc43d95dd4c9e69267eebce09.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 19.5,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 225,
+ 447,
+ 386,
+ 466.0
+ ],
+ "spans": [],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 225,
+ 466.0,
+ 386,
+ 485.0
+ ],
+ "spans": [],
+ "index": 20
+ }
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 490,
+ 505,
+ 523
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 489,
+ 505,
+ 503
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 489,
+ 256,
+ 503
+ ],
+ "score": 1.0,
+ "content": "Essentially, the class prototype graph",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 256,
+ 492,
+ 266,
+ 501
+ ],
+ "score": 0.85,
+ "content": "p _ { i }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 266,
+ 489,
+ 295,
+ 503
+ ],
+ "score": 1.0,
+ "content": "for the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 296,
+ 491,
+ 300,
+ 500
+ ],
+ "score": 0.77,
+ "content": "i",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 301,
+ 489,
+ 505,
+ 503
+ ],
+ "score": 1.0,
+ "content": "-th class is the graph with the least average spectral",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 105,
+ 501,
+ 505,
+ 514
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 501,
+ 357,
+ 514
+ ],
+ "score": 1.0,
+ "content": "distance to the rest of the graphs in the same class. Given these",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 358,
+ 502,
+ 368,
+ 511
+ ],
+ "score": 0.84,
+ "content": "K",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 369,
+ 501,
+ 505,
+ 514
+ ],
+ "score": 1.0,
+ "content": "prototypes, we cluster them using",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 105,
+ 512,
+ 280,
+ 523
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 512,
+ 234,
+ 523
+ ],
+ "score": 1.0,
+ "content": "Lloyd’s method (also known as",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 234,
+ 513,
+ 240,
+ 522
+ ],
+ "score": 0.8,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 241,
+ 512,
+ 280,
+ 523
+ ],
+ "score": 1.0,
+ "content": "-means)3.",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ }
+ ],
+ "index": 22,
+ "bbox_fs": [
+ 105,
+ 489,
+ 505,
+ 523
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 528,
+ 505,
+ 574
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 528,
+ 506,
+ 542
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 528,
+ 266,
+ 542
+ ],
+ "score": 1.0,
+ "content": "Clustering prototype graphs: Given",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 267,
+ 529,
+ 277,
+ 539
+ ],
+ "score": 0.82,
+ "content": "K",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 277,
+ 528,
+ 368,
+ 542
+ ],
+ "score": 1.0,
+ "content": "unlabeled prototypes",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 368,
+ 529,
+ 464,
+ 541
+ ],
+ "score": 0.9,
+ "content": "p _ { 1 } , \\dotsc , p _ { K } \\in \\pi _ { 1 } ( G _ { B } )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 464,
+ 528,
+ 506,
+ 542
+ ],
+ "score": 1.0,
+ "content": "and their",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 104,
+ 539,
+ 506,
+ 555
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 539,
+ 225,
+ 555
+ ],
+ "score": 1.0,
+ "content": "associated spectral measures",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 225,
+ 540,
+ 351,
+ 553
+ ],
+ "score": 0.91,
+ "content": " { \\mu } _ { \\sigma ( p _ { 1 } ) } , \\ldots , { \\mu } _ { \\sigma ( p _ { K } ) } \\in P ( [ 0 , 2 ] )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 351,
+ 539,
+ 506,
+ 555
+ ],
+ "score": 1.0,
+ "content": ". We rename the spectral measures as",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 107,
+ 551,
+ 504,
+ 563
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 553,
+ 151,
+ 562
+ ],
+ "score": 0.9,
+ "content": "s _ { 1 } , \\ldots , s _ { K }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 152,
+ 551,
+ 497,
+ 563
+ ],
+ "score": 1.0,
+ "content": "to ease notation. Thus, our goal is to associate these spectral measures to at most",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 497,
+ 551,
+ 504,
+ 561
+ ],
+ "score": 0.77,
+ "content": "k",
+ "type": "inline_equation"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 105,
+ 561,
+ 307,
+ 574
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 561,
+ 168,
+ 574
+ ],
+ "score": 1.0,
+ "content": "clusters, where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 169,
+ 562,
+ 194,
+ 573
+ ],
+ "score": 0.91,
+ "content": "k \\geq 1",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 194,
+ 561,
+ 307,
+ 574
+ ],
+ "score": 1.0,
+ "content": "is a user defined parameter.",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ }
+ ],
+ "index": 25.5,
+ "bbox_fs": [
+ 104,
+ 528,
+ 506,
+ 574
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 578,
+ 504,
+ 602
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 577,
+ 505,
+ 592
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 577,
+ 124,
+ 592
+ ],
+ "score": 1.0,
+ "content": "The",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 124,
+ 579,
+ 131,
+ 589
+ ],
+ "score": 0.81,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 132,
+ 577,
+ 225,
+ 592
+ ],
+ "score": 1.0,
+ "content": "-means problem finds a",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 226,
+ 579,
+ 232,
+ 589
+ ],
+ "score": 0.79,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 232,
+ 577,
+ 271,
+ 592
+ ],
+ "score": 1.0,
+ "content": "-partition",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 271,
+ 579,
+ 349,
+ 591
+ ],
+ "score": 0.95,
+ "content": "C = \\{ C _ { 1 } , \\ldots , C _ { k } \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 349,
+ 577,
+ 505,
+ 592
+ ],
+ "score": 1.0,
+ "content": "that minimizes the following objective",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 105,
+ 588,
+ 342,
+ 603
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 588,
+ 342,
+ 603
+ ],
+ "score": 1.0,
+ "content": "that represents the overall distortion error of the clustering",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ }
+ ],
+ "index": 28.5,
+ "bbox_fs": [
+ 105,
+ 577,
+ 505,
+ 603
+ ]
+ },
+ {
+ "type": "interline_equation",
+ "bbox": [
+ 237,
+ 606,
+ 372,
+ 642
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 237,
+ 606,
+ 372,
+ 642
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 237,
+ 606,
+ 372,
+ 642
+ ],
+ "score": 0.94,
+ "content": "\\underset { C } { \\operatorname { a r g m i n } } \\sum _ { i = 1 } ^ { k } \\sum _ { s _ { i } \\in C _ { i } } W _ { p } ( s _ { i } , B ( C _ { i } ) )",
+ "type": "interline_equation",
+ "image_path": "4073ea7dd99c2452de4f406f1a96a9e85bbfa50c9d7fcdccc4fe50e72615b66f.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 30.5,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 237,
+ 606,
+ 372,
+ 624.0
+ ],
+ "spans": [],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 237,
+ 624.0,
+ 372,
+ 642.0
+ ],
+ "spans": [],
+ "index": 31
+ }
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 108,
+ 647,
+ 505,
+ 671
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 646,
+ 506,
+ 662
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 646,
+ 133,
+ 662
+ ],
+ "score": 1.0,
+ "content": "where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 134,
+ 650,
+ 143,
+ 659
+ ],
+ "score": 0.85,
+ "content": "s _ { i }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 143,
+ 646,
+ 269,
+ 662
+ ],
+ "score": 1.0,
+ "content": "is a prototype graph in cluster",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 269,
+ 648,
+ 281,
+ 659
+ ],
+ "score": 0.89,
+ "content": "C _ { i }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 281,
+ 646,
+ 300,
+ 662
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 300,
+ 648,
+ 327,
+ 660
+ ],
+ "score": 0.93,
+ "content": "B ( C _ { i } )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 327,
+ 646,
+ 506,
+ 662
+ ],
+ "score": 1.0,
+ "content": "is the Wasserstein barycenter of the cluster",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 107,
+ 659,
+ 248,
+ 672
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 659,
+ 118,
+ 670
+ ],
+ "score": 0.86,
+ "content": "C _ { i }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 118,
+ 659,
+ 248,
+ 672
+ ],
+ "score": 1.0,
+ "content": ". The barycenter is computed as",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ }
+ ],
+ "index": 32.5,
+ "bbox_fs": [
+ 105,
+ 646,
+ 506,
+ 672
+ ]
+ },
+ {
+ "type": "interline_equation",
+ "bbox": [
+ 229,
+ 676,
+ 382,
+ 713
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 229,
+ 676,
+ 382,
+ 713
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 229,
+ 676,
+ 382,
+ 713
+ ],
+ "score": 0.95,
+ "content": "B ( C _ { i } ) = \\operatorname * { a r g m i n } _ { p \\in P ( [ 0 , 2 ] ) } \\sum _ { j = 1 } ^ { | C _ { i } | } W _ { p } ( p , s ( i , j ) )",
+ "type": "interline_equation",
+ "image_path": "e41f71ea816eed4ff86ff64b32490bcc3db78eef95e6867af8384888b318d74e.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 34.5,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 229,
+ 676,
+ 382,
+ 694.5
+ ],
+ "spans": [],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 229,
+ 694.5,
+ 382,
+ 713.0
+ ],
+ "spans": [],
+ "index": 35
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "preproc_blocks": [
+ {
+ "type": "image",
+ "bbox": [
+ 109,
+ 89,
+ 486,
+ 242
+ ],
+ "blocks": [
+ {
+ "type": "image_body",
+ "bbox": [
+ 109,
+ 89,
+ 486,
+ 242
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 109,
+ 89,
+ 486,
+ 242
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 109,
+ 89,
+ 486,
+ 242
+ ],
+ "score": 0.97,
+ "type": "image",
+ "image_path": "0908b6d8353222c6cc3822778fc0d6149a35077983a3938dc8d0e1af7f456d8a.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 1,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 109,
+ 89,
+ 486,
+ 140.0
+ ],
+ "spans": [],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 109,
+ 140.0,
+ 486,
+ 191.0
+ ],
+ "spans": [],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 109,
+ 191.0,
+ 486,
+ 242.0
+ ],
+ "spans": [],
+ "index": 2
+ }
+ ]
+ },
+ {
+ "type": "image_caption",
+ "bbox": [
+ 131,
+ 260,
+ 478,
+ 272
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 131,
+ 259,
+ 479,
+ 274
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 131,
+ 259,
+ 479,
+ 274
+ ],
+ "score": 1.0,
+ "content": "Figure 2: An illustration of our proposed Wasserstein super-class clustering algorithm.",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ }
+ ],
+ "index": 3
+ }
+ ],
+ "index": 2.0
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 294,
+ 381,
+ 306
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 294,
+ 383,
+ 308
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 294,
+ 133,
+ 308
+ ],
+ "score": 1.0,
+ "content": "where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 133,
+ 294,
+ 159,
+ 307
+ ],
+ "score": 0.93,
+ "content": "s ( i , j )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 160,
+ 294,
+ 208,
+ 308
+ ],
+ "score": 1.0,
+ "content": "denotes the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 208,
+ 295,
+ 214,
+ 306
+ ],
+ "score": 0.83,
+ "content": "j",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 214,
+ 294,
+ 320,
+ 308
+ ],
+ "score": 1.0,
+ "content": "-th spectral measure in the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 321,
+ 295,
+ 325,
+ 304
+ ],
+ "score": 0.76,
+ "content": "i",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 326,
+ 294,
+ 366,
+ 308
+ ],
+ "score": 1.0,
+ "content": "-th cluster",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 367,
+ 295,
+ 378,
+ 306
+ ],
+ "score": 0.88,
+ "content": "C _ { i }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 378,
+ 294,
+ 383,
+ 308
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ }
+ ],
+ "index": 4
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 312,
+ 505,
+ 347
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 311,
+ 506,
+ 326
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 311,
+ 391,
+ 326
+ ],
+ "score": 1.0,
+ "content": "Lloyd’s algorithm: Given an initial set of Wasserstein barycenters",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 392,
+ 311,
+ 491,
+ 325
+ ],
+ "score": 0.93,
+ "content": "B ^ { ( 1 ) } ( C _ { 1 } ) , \\ldots , B ^ { ( 1 ) } ( C _ { k } )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 492,
+ 311,
+ 506,
+ 326
+ ],
+ "score": 1.0,
+ "content": "of",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 105,
+ 323,
+ 505,
+ 337
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 323,
+ 211,
+ 337
+ ],
+ "score": 1.0,
+ "content": "spectral measures at step",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 212,
+ 324,
+ 239,
+ 334
+ ],
+ "score": 0.89,
+ "content": "t = 1",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 239,
+ 323,
+ 505,
+ 337
+ ],
+ "score": 1.0,
+ "content": ", one uses the standard Lloyd’s algorithm to find the solution by",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 105,
+ 334,
+ 424,
+ 348
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 334,
+ 424,
+ 348
+ ],
+ "score": 1.0,
+ "content": "alternating between the assignment (Equation 4) and update (Equation 5) steps",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ }
+ ],
+ "index": 6
+ },
+ {
+ "type": "interline_equation",
+ "bbox": [
+ 134,
+ 352,
+ 477,
+ 393
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 134,
+ 352,
+ 477,
+ 393
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 134,
+ 352,
+ 477,
+ 393
+ ],
+ "score": 0.91,
+ "content": "\\begin{array} { r l } & { \\quad C _ { i } ^ { ( t ) } = \\left\\{ s _ { p } : W _ { p } ( s _ { p } , B ^ { ( t ) } ( C _ { i } ) ) \\leq W _ { p } ( s _ { p } , B ^ { ( t ) } ( C _ { j } ) ) , \\forall j , 1 \\leq j \\leq k , 1 \\leq p \\leq K \\right\\} } \\\\ & { \\quad C _ { i } ^ { ( t + 1 ) } = B ( C _ { i } ^ { ( t ) } ) } \\end{array}",
+ "type": "interline_equation",
+ "image_path": "ad26653c6b604b388c4728b71459d97fa1b1aa7435fef4487de65c7c891e0de4.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 9,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 134,
+ 352,
+ 477,
+ 365.6666666666667
+ ],
+ "spans": [],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 134,
+ 365.6666666666667,
+ 477,
+ 379.33333333333337
+ ],
+ "spans": [],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 134,
+ 379.33333333333337,
+ 477,
+ 393.00000000000006
+ ],
+ "spans": [],
+ "index": 10
+ }
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 397,
+ 505,
+ 442
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 397,
+ 506,
+ 410
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 397,
+ 506,
+ 410
+ ],
+ "score": 1.0,
+ "content": "Lloyd’s algorithm is known to converge to a local minimum (except in pathological cases, where it",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 105,
+ 408,
+ 505,
+ 421
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 408,
+ 505,
+ 421
+ ],
+ "score": 1.0,
+ "content": "can oscillate between equivalent solutions). The final output is a grouping of the prototype graphs",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 105,
+ 418,
+ 506,
+ 432
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 418,
+ 125,
+ 432
+ ],
+ "score": 1.0,
+ "content": "into",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 125,
+ 420,
+ 132,
+ 429
+ ],
+ "score": 0.79,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 132,
+ 418,
+ 506,
+ 432
+ ],
+ "score": 1.0,
+ "content": "groups, which also induces a grouping of the corresponding base classes. We denote these",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 104,
+ 430,
+ 400,
+ 443
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 430,
+ 374,
+ 443
+ ],
+ "score": 1.0,
+ "content": "class groups as super-classes and denote the set of super-classes as",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 374,
+ 430,
+ 396,
+ 441
+ ],
+ "score": 0.89,
+ "content": "y ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 396,
+ 430,
+ 400,
+ 443
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ }
+ ],
+ "index": 12.5
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 456,
+ 266,
+ 468
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 456,
+ 268,
+ 469
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 456,
+ 268,
+ 469
+ ],
+ "score": 1.0,
+ "content": "4.2 OUR GRAPH NEURAL NETWORK",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ }
+ ],
+ "index": 15
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 477,
+ 505,
+ 511
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 477,
+ 505,
+ 489
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 477,
+ 505,
+ 489
+ ],
+ "score": 1.0,
+ "content": "Feature extractor: To apply standard neural network architectures for downstream tasks we must",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 105,
+ 488,
+ 505,
+ 501
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 488,
+ 505,
+ 501
+ ],
+ "score": 1.0,
+ "content": "embed the graphs in a finite dimensional vector space. We consider graph neural networks (GNNs)",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 105,
+ 499,
+ 331,
+ 512
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 499,
+ 331,
+ 512
+ ],
+ "score": 1.0,
+ "content": "that employ the following message-passing architecture",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ }
+ ],
+ "index": 17
+ },
+ {
+ "type": "interline_equation",
+ "bbox": [
+ 247,
+ 517,
+ 363,
+ 532
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 247,
+ 517,
+ 363,
+ 532
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 247,
+ 517,
+ 363,
+ 532
+ ],
+ "score": 0.92,
+ "content": "H ^ { ( j ) } = M ( A , H ^ { ( j - 1 ) } , \\theta ^ { ( j ) } )",
+ "type": "interline_equation",
+ "image_path": "a4db44c94ed2d34ff3fe4924ba2165185fceface82e7f5ec3ecdd6463052c7f7.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 19,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 247,
+ 517,
+ 363,
+ 532
+ ],
+ "spans": [],
+ "index": 19
+ }
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 539,
+ 505,
+ 586
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 537,
+ 506,
+ 554
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 537,
+ 133,
+ 554
+ ],
+ "score": 1.0,
+ "content": "where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 133,
+ 539,
+ 196,
+ 551
+ ],
+ "score": 0.93,
+ "content": "H ^ { ( j ) } \\in \\mathbb { R } ^ { | V | \\times d }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 196,
+ 537,
+ 426,
+ 554
+ ],
+ "score": 1.0,
+ "content": "are the node embeddings (i.e., messages) computed after",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 426,
+ 541,
+ 432,
+ 552
+ ],
+ "score": 0.83,
+ "content": "j",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 432,
+ 537,
+ 506,
+ 554
+ ],
+ "score": 1.0,
+ "content": "steps of the GNN",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 106,
+ 551,
+ 504,
+ 563
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 551,
+ 124,
+ 563
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 124,
+ 552,
+ 136,
+ 561
+ ],
+ "score": 0.83,
+ "content": "M",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 137,
+ 551,
+ 504,
+ 563
+ ],
+ "score": 1.0,
+ "content": "is the message propagation function which depends on the adjacency matrix of the graph",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 107,
+ 562,
+ 504,
+ 576
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 564,
+ 115,
+ 573
+ ],
+ "score": 0.73,
+ "content": "A",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 115,
+ 562,
+ 244,
+ 576
+ ],
+ "score": 1.0,
+ "content": ", the trainable parameters of the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 245,
+ 563,
+ 259,
+ 575
+ ],
+ "score": 0.9,
+ "content": "j ^ { t h }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 259,
+ 562,
+ 283,
+ 576
+ ],
+ "score": 1.0,
+ "content": "layer",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 283,
+ 562,
+ 299,
+ 574
+ ],
+ "score": 0.89,
+ "content": "\\theta ^ { ( j ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 299,
+ 562,
+ 394,
+ 576
+ ],
+ "score": 1.0,
+ "content": ", and node embeddings",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 394,
+ 562,
+ 424,
+ 574
+ ],
+ "score": 0.91,
+ "content": "H ^ { ( j - 1 ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 425,
+ 562,
+ 504,
+ 576
+ ],
+ "score": 1.0,
+ "content": "generated from the",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 106,
+ 575,
+ 164,
+ 587
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 575,
+ 164,
+ 587
+ ],
+ "score": 1.0,
+ "content": "previous step.",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ }
+ ],
+ "index": 21.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 590,
+ 505,
+ 658
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 591,
+ 505,
+ 604
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 591,
+ 418,
+ 604
+ ],
+ "score": 1.0,
+ "content": "A recently proposed GNN called the graph isomorphism network (GIN) by",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 418,
+ 591,
+ 432,
+ 601
+ ],
+ "score": 0.4,
+ "content": "\\mathrm { X u }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 433,
+ 591,
+ 505,
+ 604
+ ],
+ "score": 1.0,
+ "content": "et al. (2019) was",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 105,
+ 602,
+ 506,
+ 615
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 602,
+ 506,
+ 615
+ ],
+ "score": 1.0,
+ "content": "shown to be stronger than several popular GNN variants like GCN Kipf & Welling (2016) and",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 106,
+ 613,
+ 506,
+ 626
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 613,
+ 506,
+ 626
+ ],
+ "score": 1.0,
+ "content": "GraphSAGE Hamilton et al. (2017). What makes GIN so powerful and sets it apart from the other",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 105,
+ 623,
+ 505,
+ 637
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 623,
+ 505,
+ 637
+ ],
+ "score": 1.0,
+ "content": "GNN variants is its injective neighborhood aggregation scheme which allows it to be as powerful as",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 105,
+ 634,
+ 506,
+ 648
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 634,
+ 506,
+ 648
+ ],
+ "score": 1.0,
+ "content": "the Weisfeiler-Lehman (WL) graph isomorphism test. Motivated by this finding, we chose GIN as",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 106,
+ 646,
+ 430,
+ 659
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 646,
+ 430,
+ 659
+ ],
+ "score": 1.0,
+ "content": "our graph feature extractor. The message propagation scheme in GIN is given by",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ }
+ ],
+ "index": 26.5
+ },
+ {
+ "type": "interline_equation",
+ "bbox": [
+ 205,
+ 663,
+ 405,
+ 679
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 205,
+ 663,
+ 405,
+ 679
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 205,
+ 663,
+ 405,
+ 679
+ ],
+ "score": 0.91,
+ "content": "H ^ { ( j ) } = M L P ( ( 1 + \\epsilon ) ^ { j } \\odot H ^ { ( j - 1 ) } + A ^ { T } H ^ { ( j - 1 ) } )",
+ "type": "interline_equation",
+ "image_path": "ab87fbbec2534eec2d3ce3655b5c4dd0ba436535dfe260cc5acf999727c19ffe.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 30,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 205,
+ 663,
+ 405,
+ 679
+ ],
+ "spans": [],
+ "index": 30
+ }
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 685,
+ 504,
+ 732
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 685,
+ 505,
+ 698
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 685,
+ 131,
+ 698
+ ],
+ "score": 1.0,
+ "content": "Here,",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 131,
+ 688,
+ 136,
+ 695
+ ],
+ "score": 0.73,
+ "content": "\\epsilon",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 136,
+ 685,
+ 317,
+ 698
+ ],
+ "score": 1.0,
+ "content": "is a layer-wise learnable scalar parameter and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 318,
+ 686,
+ 344,
+ 695
+ ],
+ "score": 0.75,
+ "content": "M L P",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 344,
+ 685,
+ 505,
+ 698
+ ],
+ "score": 1.0,
+ "content": "represents a multi-layer perceptron with",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ },
+ {
+ "bbox": [
+ 105,
+ 696,
+ 506,
+ 709
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 696,
+ 455,
+ 709
+ ],
+ "score": 1.0,
+ "content": "layer-wise non-linearities for more expressive representations. The full GIN model run",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 456,
+ 697,
+ 464,
+ 707
+ ],
+ "score": 0.8,
+ "content": "R",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 465,
+ 696,
+ 506,
+ 709
+ ],
+ "score": 1.0,
+ "content": "iterations",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 105,
+ 707,
+ 506,
+ 721
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 707,
+ 404,
+ 721
+ ],
+ "score": 1.0,
+ "content": "of Equation 6 to generate final node embeddings which we represent by",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 405,
+ 707,
+ 427,
+ 718
+ ],
+ "score": 0.91,
+ "content": "H ^ { ( R ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 427,
+ 707,
+ 506,
+ 721
+ ],
+ "score": 1.0,
+ "content": ". As features from",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ },
+ {
+ "bbox": [
+ 105,
+ 719,
+ 503,
+ 733
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 719,
+ 483,
+ 733
+ ],
+ "score": 1.0,
+ "content": "earlier iterations can also be helpful in achieving higher discriminative power, embeddings",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 483,
+ 720,
+ 503,
+ 730
+ ],
+ "score": 0.89,
+ "content": "H ^ { ( j ) }",
+ "type": "inline_equation"
+ }
+ ],
+ "index": 34
+ }
+ ],
+ "index": 32.5
+ }
+ ],
+ "page_idx": 5,
+ "page_size": [
+ 612,
+ 792
+ ],
+ "discarded_blocks": [
+ {
+ "type": "discarded",
+ "bbox": [
+ 107,
+ 27,
+ 293,
+ 37
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 293,
+ 38
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 293,
+ 38
+ ],
+ "score": 1.0,
+ "content": "Published as a conference paper at ICLR 2020",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 302,
+ 751,
+ 309,
+ 760
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 302,
+ 750,
+ 310,
+ 761
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 302,
+ 750,
+ 310,
+ 761
+ ],
+ "score": 1.0,
+ "content": "6",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ }
+ ],
+ "para_blocks": [
+ {
+ "type": "image",
+ "bbox": [
+ 109,
+ 89,
+ 486,
+ 242
+ ],
+ "blocks": [
+ {
+ "type": "image_body",
+ "bbox": [
+ 109,
+ 89,
+ 486,
+ 242
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 109,
+ 89,
+ 486,
+ 242
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 109,
+ 89,
+ 486,
+ 242
+ ],
+ "score": 0.97,
+ "type": "image",
+ "image_path": "0908b6d8353222c6cc3822778fc0d6149a35077983a3938dc8d0e1af7f456d8a.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 1,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 109,
+ 89,
+ 486,
+ 140.0
+ ],
+ "spans": [],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 109,
+ 140.0,
+ 486,
+ 191.0
+ ],
+ "spans": [],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 109,
+ 191.0,
+ 486,
+ 242.0
+ ],
+ "spans": [],
+ "index": 2
+ }
+ ]
+ },
+ {
+ "type": "image_caption",
+ "bbox": [
+ 131,
+ 260,
+ 478,
+ 272
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 131,
+ 259,
+ 479,
+ 274
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 131,
+ 259,
+ 479,
+ 274
+ ],
+ "score": 1.0,
+ "content": "Figure 2: An illustration of our proposed Wasserstein super-class clustering algorithm.",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ }
+ ],
+ "index": 3
+ }
+ ],
+ "index": 2.0
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 294,
+ 381,
+ 306
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 294,
+ 383,
+ 308
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 294,
+ 133,
+ 308
+ ],
+ "score": 1.0,
+ "content": "where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 133,
+ 294,
+ 159,
+ 307
+ ],
+ "score": 0.93,
+ "content": "s ( i , j )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 160,
+ 294,
+ 208,
+ 308
+ ],
+ "score": 1.0,
+ "content": "denotes the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 208,
+ 295,
+ 214,
+ 306
+ ],
+ "score": 0.83,
+ "content": "j",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 214,
+ 294,
+ 320,
+ 308
+ ],
+ "score": 1.0,
+ "content": "-th spectral measure in the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 321,
+ 295,
+ 325,
+ 304
+ ],
+ "score": 0.76,
+ "content": "i",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 326,
+ 294,
+ 366,
+ 308
+ ],
+ "score": 1.0,
+ "content": "-th cluster",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 367,
+ 295,
+ 378,
+ 306
+ ],
+ "score": 0.88,
+ "content": "C _ { i }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 378,
+ 294,
+ 383,
+ 308
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ }
+ ],
+ "index": 4,
+ "bbox_fs": [
+ 106,
+ 294,
+ 383,
+ 308
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 312,
+ 505,
+ 347
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 311,
+ 506,
+ 326
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 311,
+ 391,
+ 326
+ ],
+ "score": 1.0,
+ "content": "Lloyd’s algorithm: Given an initial set of Wasserstein barycenters",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 392,
+ 311,
+ 491,
+ 325
+ ],
+ "score": 0.93,
+ "content": "B ^ { ( 1 ) } ( C _ { 1 } ) , \\ldots , B ^ { ( 1 ) } ( C _ { k } )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 492,
+ 311,
+ 506,
+ 326
+ ],
+ "score": 1.0,
+ "content": "of",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 105,
+ 323,
+ 505,
+ 337
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 323,
+ 211,
+ 337
+ ],
+ "score": 1.0,
+ "content": "spectral measures at step",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 212,
+ 324,
+ 239,
+ 334
+ ],
+ "score": 0.89,
+ "content": "t = 1",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 239,
+ 323,
+ 505,
+ 337
+ ],
+ "score": 1.0,
+ "content": ", one uses the standard Lloyd’s algorithm to find the solution by",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 105,
+ 334,
+ 424,
+ 348
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 334,
+ 424,
+ 348
+ ],
+ "score": 1.0,
+ "content": "alternating between the assignment (Equation 4) and update (Equation 5) steps",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ }
+ ],
+ "index": 6,
+ "bbox_fs": [
+ 105,
+ 311,
+ 506,
+ 348
+ ]
+ },
+ {
+ "type": "interline_equation",
+ "bbox": [
+ 134,
+ 352,
+ 477,
+ 393
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 134,
+ 352,
+ 477,
+ 393
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 134,
+ 352,
+ 477,
+ 393
+ ],
+ "score": 0.91,
+ "content": "\\begin{array} { r l } & { \\quad C _ { i } ^ { ( t ) } = \\left\\{ s _ { p } : W _ { p } ( s _ { p } , B ^ { ( t ) } ( C _ { i } ) ) \\leq W _ { p } ( s _ { p } , B ^ { ( t ) } ( C _ { j } ) ) , \\forall j , 1 \\leq j \\leq k , 1 \\leq p \\leq K \\right\\} } \\\\ & { \\quad C _ { i } ^ { ( t + 1 ) } = B ( C _ { i } ^ { ( t ) } ) } \\end{array}",
+ "type": "interline_equation",
+ "image_path": "ad26653c6b604b388c4728b71459d97fa1b1aa7435fef4487de65c7c891e0de4.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 9,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 134,
+ 352,
+ 477,
+ 365.6666666666667
+ ],
+ "spans": [],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 134,
+ 365.6666666666667,
+ 477,
+ 379.33333333333337
+ ],
+ "spans": [],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 134,
+ 379.33333333333337,
+ 477,
+ 393.00000000000006
+ ],
+ "spans": [],
+ "index": 10
+ }
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 397,
+ 505,
+ 442
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 397,
+ 506,
+ 410
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 397,
+ 506,
+ 410
+ ],
+ "score": 1.0,
+ "content": "Lloyd’s algorithm is known to converge to a local minimum (except in pathological cases, where it",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 105,
+ 408,
+ 505,
+ 421
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 408,
+ 505,
+ 421
+ ],
+ "score": 1.0,
+ "content": "can oscillate between equivalent solutions). The final output is a grouping of the prototype graphs",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 105,
+ 418,
+ 506,
+ 432
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 418,
+ 125,
+ 432
+ ],
+ "score": 1.0,
+ "content": "into",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 125,
+ 420,
+ 132,
+ 429
+ ],
+ "score": 0.79,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 132,
+ 418,
+ 506,
+ 432
+ ],
+ "score": 1.0,
+ "content": "groups, which also induces a grouping of the corresponding base classes. We denote these",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 104,
+ 430,
+ 400,
+ 443
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 430,
+ 374,
+ 443
+ ],
+ "score": 1.0,
+ "content": "class groups as super-classes and denote the set of super-classes as",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 374,
+ 430,
+ 396,
+ 441
+ ],
+ "score": 0.89,
+ "content": "y ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 396,
+ 430,
+ 400,
+ 443
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ }
+ ],
+ "index": 12.5,
+ "bbox_fs": [
+ 104,
+ 397,
+ 506,
+ 443
+ ]
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 456,
+ 266,
+ 468
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 456,
+ 268,
+ 469
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 456,
+ 268,
+ 469
+ ],
+ "score": 1.0,
+ "content": "4.2 OUR GRAPH NEURAL NETWORK",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ }
+ ],
+ "index": 15
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 477,
+ 505,
+ 511
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 477,
+ 505,
+ 489
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 477,
+ 505,
+ 489
+ ],
+ "score": 1.0,
+ "content": "Feature extractor: To apply standard neural network architectures for downstream tasks we must",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 105,
+ 488,
+ 505,
+ 501
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 488,
+ 505,
+ 501
+ ],
+ "score": 1.0,
+ "content": "embed the graphs in a finite dimensional vector space. We consider graph neural networks (GNNs)",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 105,
+ 499,
+ 331,
+ 512
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 499,
+ 331,
+ 512
+ ],
+ "score": 1.0,
+ "content": "that employ the following message-passing architecture",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ }
+ ],
+ "index": 17,
+ "bbox_fs": [
+ 105,
+ 477,
+ 505,
+ 512
+ ]
+ },
+ {
+ "type": "interline_equation",
+ "bbox": [
+ 247,
+ 517,
+ 363,
+ 532
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 247,
+ 517,
+ 363,
+ 532
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 247,
+ 517,
+ 363,
+ 532
+ ],
+ "score": 0.92,
+ "content": "H ^ { ( j ) } = M ( A , H ^ { ( j - 1 ) } , \\theta ^ { ( j ) } )",
+ "type": "interline_equation",
+ "image_path": "a4db44c94ed2d34ff3fe4924ba2165185fceface82e7f5ec3ecdd6463052c7f7.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 19,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 247,
+ 517,
+ 363,
+ 532
+ ],
+ "spans": [],
+ "index": 19
+ }
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 539,
+ 505,
+ 586
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 537,
+ 506,
+ 554
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 537,
+ 133,
+ 554
+ ],
+ "score": 1.0,
+ "content": "where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 133,
+ 539,
+ 196,
+ 551
+ ],
+ "score": 0.93,
+ "content": "H ^ { ( j ) } \\in \\mathbb { R } ^ { | V | \\times d }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 196,
+ 537,
+ 426,
+ 554
+ ],
+ "score": 1.0,
+ "content": "are the node embeddings (i.e., messages) computed after",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 426,
+ 541,
+ 432,
+ 552
+ ],
+ "score": 0.83,
+ "content": "j",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 432,
+ 537,
+ 506,
+ 554
+ ],
+ "score": 1.0,
+ "content": "steps of the GNN",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 106,
+ 551,
+ 504,
+ 563
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 551,
+ 124,
+ 563
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 124,
+ 552,
+ 136,
+ 561
+ ],
+ "score": 0.83,
+ "content": "M",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 137,
+ 551,
+ 504,
+ 563
+ ],
+ "score": 1.0,
+ "content": "is the message propagation function which depends on the adjacency matrix of the graph",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 107,
+ 562,
+ 504,
+ 576
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 564,
+ 115,
+ 573
+ ],
+ "score": 0.73,
+ "content": "A",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 115,
+ 562,
+ 244,
+ 576
+ ],
+ "score": 1.0,
+ "content": ", the trainable parameters of the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 245,
+ 563,
+ 259,
+ 575
+ ],
+ "score": 0.9,
+ "content": "j ^ { t h }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 259,
+ 562,
+ 283,
+ 576
+ ],
+ "score": 1.0,
+ "content": "layer",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 283,
+ 562,
+ 299,
+ 574
+ ],
+ "score": 0.89,
+ "content": "\\theta ^ { ( j ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 299,
+ 562,
+ 394,
+ 576
+ ],
+ "score": 1.0,
+ "content": ", and node embeddings",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 394,
+ 562,
+ 424,
+ 574
+ ],
+ "score": 0.91,
+ "content": "H ^ { ( j - 1 ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 425,
+ 562,
+ 504,
+ 576
+ ],
+ "score": 1.0,
+ "content": "generated from the",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 106,
+ 575,
+ 164,
+ 587
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 575,
+ 164,
+ 587
+ ],
+ "score": 1.0,
+ "content": "previous step.",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ }
+ ],
+ "index": 21.5,
+ "bbox_fs": [
+ 105,
+ 537,
+ 506,
+ 587
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 590,
+ 505,
+ 658
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 591,
+ 505,
+ 604
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 591,
+ 418,
+ 604
+ ],
+ "score": 1.0,
+ "content": "A recently proposed GNN called the graph isomorphism network (GIN) by",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 418,
+ 591,
+ 432,
+ 601
+ ],
+ "score": 0.4,
+ "content": "\\mathrm { X u }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 433,
+ 591,
+ 505,
+ 604
+ ],
+ "score": 1.0,
+ "content": "et al. (2019) was",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 105,
+ 602,
+ 506,
+ 615
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 602,
+ 506,
+ 615
+ ],
+ "score": 1.0,
+ "content": "shown to be stronger than several popular GNN variants like GCN Kipf & Welling (2016) and",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 106,
+ 613,
+ 506,
+ 626
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 613,
+ 506,
+ 626
+ ],
+ "score": 1.0,
+ "content": "GraphSAGE Hamilton et al. (2017). What makes GIN so powerful and sets it apart from the other",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 105,
+ 623,
+ 505,
+ 637
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 623,
+ 505,
+ 637
+ ],
+ "score": 1.0,
+ "content": "GNN variants is its injective neighborhood aggregation scheme which allows it to be as powerful as",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 105,
+ 634,
+ 506,
+ 648
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 634,
+ 506,
+ 648
+ ],
+ "score": 1.0,
+ "content": "the Weisfeiler-Lehman (WL) graph isomorphism test. Motivated by this finding, we chose GIN as",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 106,
+ 646,
+ 430,
+ 659
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 646,
+ 430,
+ 659
+ ],
+ "score": 1.0,
+ "content": "our graph feature extractor. The message propagation scheme in GIN is given by",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ }
+ ],
+ "index": 26.5,
+ "bbox_fs": [
+ 105,
+ 591,
+ 506,
+ 659
+ ]
+ },
+ {
+ "type": "interline_equation",
+ "bbox": [
+ 205,
+ 663,
+ 405,
+ 679
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 205,
+ 663,
+ 405,
+ 679
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 205,
+ 663,
+ 405,
+ 679
+ ],
+ "score": 0.91,
+ "content": "H ^ { ( j ) } = M L P ( ( 1 + \\epsilon ) ^ { j } \\odot H ^ { ( j - 1 ) } + A ^ { T } H ^ { ( j - 1 ) } )",
+ "type": "interline_equation",
+ "image_path": "ab87fbbec2534eec2d3ce3655b5c4dd0ba436535dfe260cc5acf999727c19ffe.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 30,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 205,
+ 663,
+ 405,
+ 679
+ ],
+ "spans": [],
+ "index": 30
+ }
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 685,
+ 504,
+ 732
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 685,
+ 505,
+ 698
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 685,
+ 131,
+ 698
+ ],
+ "score": 1.0,
+ "content": "Here,",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 131,
+ 688,
+ 136,
+ 695
+ ],
+ "score": 0.73,
+ "content": "\\epsilon",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 136,
+ 685,
+ 317,
+ 698
+ ],
+ "score": 1.0,
+ "content": "is a layer-wise learnable scalar parameter and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 318,
+ 686,
+ 344,
+ 695
+ ],
+ "score": 0.75,
+ "content": "M L P",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 344,
+ 685,
+ 505,
+ 698
+ ],
+ "score": 1.0,
+ "content": "represents a multi-layer perceptron with",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ },
+ {
+ "bbox": [
+ 105,
+ 696,
+ 506,
+ 709
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 696,
+ 455,
+ 709
+ ],
+ "score": 1.0,
+ "content": "layer-wise non-linearities for more expressive representations. The full GIN model run",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 456,
+ 697,
+ 464,
+ 707
+ ],
+ "score": 0.8,
+ "content": "R",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 465,
+ 696,
+ 506,
+ 709
+ ],
+ "score": 1.0,
+ "content": "iterations",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 105,
+ 707,
+ 506,
+ 721
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 707,
+ 404,
+ 721
+ ],
+ "score": 1.0,
+ "content": "of Equation 6 to generate final node embeddings which we represent by",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 405,
+ 707,
+ 427,
+ 718
+ ],
+ "score": 0.91,
+ "content": "H ^ { ( R ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 427,
+ 707,
+ 506,
+ 721
+ ],
+ "score": 1.0,
+ "content": ". As features from",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ },
+ {
+ "bbox": [
+ 105,
+ 719,
+ 503,
+ 733
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 719,
+ 483,
+ 733
+ ],
+ "score": 1.0,
+ "content": "earlier iterations can also be helpful in achieving higher discriminative power, embeddings",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 483,
+ 720,
+ 503,
+ 730
+ ],
+ "score": 0.89,
+ "content": "H ^ { ( j ) }",
+ "type": "inline_equation"
+ }
+ ],
+ "index": 34
+ }
+ ],
+ "index": 32.5,
+ "bbox_fs": [
+ 105,
+ 685,
+ 506,
+ 733
+ ]
+ }
+ ]
+ },
+ {
+ "preproc_blocks": [
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 82,
+ 270,
+ 93
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 82,
+ 270,
+ 94
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 82,
+ 141,
+ 94
+ ],
+ "score": 1.0,
+ "content": "from all",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 141,
+ 83,
+ 150,
+ 92
+ ],
+ "score": 0.82,
+ "content": "R",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 150,
+ 82,
+ 270,
+ 94
+ ],
+ "score": 1.0,
+ "content": "iterations are concatenated as",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ }
+ ],
+ "index": 0
+ },
+ {
+ "type": "interline_equation",
+ "bbox": [
+ 267,
+ 100,
+ 340,
+ 132
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 267,
+ 100,
+ 340,
+ 132
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 267,
+ 100,
+ 340,
+ 132
+ ],
+ "score": 0.94,
+ "content": "H _ { g } = \\left\\| _ { j = 1 } ^ { R } H ^ { ( j ) } , \\right\\|",
+ "type": "interline_equation",
+ "image_path": "efd9868b636e27c4f3f13c23a86a3abe1b3f5758487b5acfc443c0f0c77fd42d.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 1.5,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 267,
+ 100,
+ 340,
+ 116.0
+ ],
+ "spans": [],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 267,
+ 116.0,
+ 340,
+ 132.0
+ ],
+ "spans": [],
+ "index": 2
+ }
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 141,
+ 505,
+ 178
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 102,
+ 137,
+ 507,
+ 158
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 102,
+ 137,
+ 131,
+ 158
+ ],
+ "score": 1.0,
+ "content": "Here,",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 132,
+ 141,
+ 216,
+ 156
+ ],
+ "score": 0.93,
+ "content": "\\begin{array} { r } { H ^ { ( j ) } = \\sum _ { v \\in V } H _ { v } ^ { ( j ) } } \\end{array}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 216,
+ 137,
+ 248,
+ 158
+ ],
+ "score": 1.0,
+ "content": ", where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 248,
+ 141,
+ 269,
+ 156
+ ],
+ "score": 0.93,
+ "content": "H _ { i } ^ { ( j ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 269,
+ 137,
+ 330,
+ 158
+ ],
+ "score": 1.0,
+ "content": "represents the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 330,
+ 145,
+ 335,
+ 154
+ ],
+ "score": 0.73,
+ "content": "i",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 335,
+ 137,
+ 450,
+ 158
+ ],
+ "score": 1.0,
+ "content": "-th node’s embedding in the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 450,
+ 145,
+ 456,
+ 155
+ ],
+ "score": 0.83,
+ "content": "j",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 457,
+ 137,
+ 507,
+ 158
+ ],
+ "score": 1.0,
+ "content": "-th iteration",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 105,
+ 154,
+ 505,
+ 169
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 154,
+ 123,
+ 169
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 124,
+ 156,
+ 130,
+ 168
+ ],
+ "score": 0.83,
+ "content": "\\parallel",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 131,
+ 154,
+ 268,
+ 169
+ ],
+ "score": 1.0,
+ "content": "denotes a concatenation operator.",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 269,
+ 156,
+ 283,
+ 168
+ ],
+ "score": 0.9,
+ "content": "H _ { g }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 283,
+ 154,
+ 470,
+ 169
+ ],
+ "score": 1.0,
+ "content": "now contains the graph embedding of a graph",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 471,
+ 158,
+ 478,
+ 167
+ ],
+ "score": 0.79,
+ "content": "g",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 478,
+ 154,
+ 505,
+ 169
+ ],
+ "score": 1.0,
+ "content": "and is",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 105,
+ 167,
+ 213,
+ 179
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 167,
+ 213,
+ 179
+ ],
+ "score": 1.0,
+ "content": "passed on to the classifier.",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ }
+ ],
+ "index": 4
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 183,
+ 505,
+ 228
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 183,
+ 505,
+ 196
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 183,
+ 505,
+ 196
+ ],
+ "score": 1.0,
+ "content": "Classifier: Here, our objective is to improve the class separation produced by the graph embeddings",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 105,
+ 194,
+ 506,
+ 207
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 194,
+ 202,
+ 207
+ ],
+ "score": 1.0,
+ "content": "of the feature extractor",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 202,
+ 195,
+ 225,
+ 207
+ ],
+ "score": 0.92,
+ "content": "F _ { \\theta } ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 225,
+ 194,
+ 506,
+ 207
+ ],
+ "score": 1.0,
+ "content": "and we do this by building a “graph of graph embeddings”, called a",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 104,
+ 205,
+ 505,
+ 219
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 205,
+ 158,
+ 219
+ ],
+ "score": 1.0,
+ "content": "super-graph",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 158,
+ 206,
+ 177,
+ 217
+ ],
+ "score": 0.9,
+ "content": "g ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 177,
+ 205,
+ 471,
+ 219
+ ],
+ "score": 1.0,
+ "content": ", where each node is a graph feature vector. We then employ our classifier",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 472,
+ 205,
+ 491,
+ 218
+ ],
+ "score": 0.91,
+ "content": "C ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 492,
+ 205,
+ 505,
+ 219
+ ],
+ "score": 1.0,
+ "content": "on",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 105,
+ 216,
+ 483,
+ 230
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 216,
+ 483,
+ 230
+ ],
+ "score": 1.0,
+ "content": "this super-graph to achieve better separation among the graph classes in the embedding space.",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ }
+ ],
+ "index": 7.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 233,
+ 505,
+ 332
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 231,
+ 506,
+ 247
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 231,
+ 291,
+ 247
+ ],
+ "score": 1.0,
+ "content": "During training, we first build the super-graph",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 291,
+ 234,
+ 311,
+ 245
+ ],
+ "score": 0.91,
+ "content": "g ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 311,
+ 231,
+ 506,
+ 247
+ ],
+ "score": 1.0,
+ "content": "on a batch of base-labeled graphs as a collection",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 106,
+ 244,
+ 506,
+ 257
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 244,
+ 118,
+ 257
+ ],
+ "score": 1.0,
+ "content": "of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 118,
+ 245,
+ 124,
+ 254
+ ],
+ "score": 0.79,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 125,
+ 244,
+ 272,
+ 257
+ ],
+ "score": 1.0,
+ "content": "-NN graphs, where each constituent",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 272,
+ 245,
+ 279,
+ 254
+ ],
+ "score": 0.79,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 279,
+ 244,
+ 506,
+ 257
+ ],
+ "score": 1.0,
+ "content": "-NN graph is built on the graphs belonging to the same",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 104,
+ 253,
+ 507,
+ 268
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 253,
+ 156,
+ 268
+ ],
+ "score": 1.0,
+ "content": "super-class.",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 156,
+ 256,
+ 176,
+ 267
+ ],
+ "score": 0.9,
+ "content": "g ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 176,
+ 253,
+ 429,
+ 268
+ ],
+ "score": 1.0,
+ "content": "is then passed through a multi-layered graph attention network",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 429,
+ 254,
+ 456,
+ 265
+ ],
+ "score": 0.9,
+ "content": "C ^ { \\forall G A \\mathbf { \\breve { T } } }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 457,
+ 253,
+ 507,
+ 268
+ ],
+ "score": 1.0,
+ "content": "to learn the",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 105,
+ 266,
+ 505,
+ 279
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 266,
+ 344,
+ 279
+ ],
+ "score": 1.0,
+ "content": "associated class probabilities. The features extracted from",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 344,
+ 266,
+ 366,
+ 277
+ ],
+ "score": 0.89,
+ "content": "F _ { \\theta } ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 367,
+ 266,
+ 505,
+ 279
+ ],
+ "score": 1.0,
+ "content": "are passed into the MLP network",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 107,
+ 273,
+ 507,
+ 291
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 277,
+ 128,
+ 287
+ ],
+ "score": 0.87,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 129,
+ 273,
+ 306,
+ 291
+ ],
+ "score": 1.0,
+ "content": "to learn the associated super-class labels.",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 306,
+ 277,
+ 327,
+ 287
+ ],
+ "score": 0.89,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 328,
+ 273,
+ 348,
+ 291
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 349,
+ 277,
+ 376,
+ 287
+ ],
+ "score": 0.87,
+ "content": "C ^ { \\widecheck { G } A T }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 376,
+ 273,
+ 507,
+ 291
+ ],
+ "score": 1.0,
+ "content": "combine to form our classifier",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 107,
+ 284,
+ 507,
+ 303
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 288,
+ 126,
+ 300
+ ],
+ "score": 0.86,
+ "content": "C ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 126,
+ 284,
+ 298,
+ 303
+ ],
+ "score": 1.0,
+ "content": ". The cross-entropy losses associated with",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 299,
+ 289,
+ 321,
+ 299
+ ],
+ "score": 0.89,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 321,
+ 284,
+ 340,
+ 303
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 340,
+ 288,
+ 367,
+ 298
+ ],
+ "score": 0.89,
+ "content": "C ^ { G A T }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 368,
+ 284,
+ 507,
+ 303
+ ],
+ "score": 1.0,
+ "content": "are added to give the overall loss",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 104,
+ 296,
+ 507,
+ 313
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 296,
+ 120,
+ 313
+ ],
+ "score": 1.0,
+ "content": "for",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 121,
+ 299,
+ 140,
+ 311
+ ],
+ "score": 0.9,
+ "content": "C ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 140,
+ 296,
+ 305,
+ 313
+ ],
+ "score": 1.0,
+ "content": ". The intuition behind the construction of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 305,
+ 300,
+ 324,
+ 311
+ ],
+ "score": 0.91,
+ "content": "g ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 325,
+ 296,
+ 356,
+ 313
+ ],
+ "score": 1.0,
+ "content": "to train",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 357,
+ 298,
+ 384,
+ 309
+ ],
+ "score": 0.9,
+ "content": "C ^ { G A T }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 384,
+ 296,
+ 507,
+ 313
+ ],
+ "score": 1.0,
+ "content": "on was to further improve the",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 106,
+ 310,
+ 505,
+ 323
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 310,
+ 505,
+ 323
+ ],
+ "score": 1.0,
+ "content": "existing cluster separation based on graph spectral measures by introducing a relational inductive",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 104,
+ 318,
+ 359,
+ 335
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 318,
+ 359,
+ 335
+ ],
+ "score": 1.0,
+ "content": "bias (Battaglia et al., 2018) that is inherent to the GNN CGAT .",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ }
+ ],
+ "index": 14
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 338,
+ 504,
+ 393
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 337,
+ 506,
+ 350
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 337,
+ 506,
+ 350
+ ],
+ "score": 1.0,
+ "content": "Recall that we adopt an initialization method (described in 3). In our fine-tuning stage, novel class",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 105,
+ 348,
+ 506,
+ 362
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 348,
+ 188,
+ 362
+ ],
+ "score": 1.0,
+ "content": "labeled graphs from",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 189,
+ 349,
+ 205,
+ 360
+ ],
+ "score": 0.89,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 205,
+ 348,
+ 506,
+ 362
+ ],
+ "score": 1.0,
+ "content": "are input to the network. The pre-trained parameters learned by the feature",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 105,
+ 358,
+ 506,
+ 373
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 358,
+ 145,
+ 373
+ ],
+ "score": 1.0,
+ "content": "extractor",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 145,
+ 360,
+ 167,
+ 372
+ ],
+ "score": 0.92,
+ "content": "F _ { \\theta } ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 168,
+ 358,
+ 223,
+ 373
+ ],
+ "score": 1.0,
+ "content": "are fixed and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 223,
+ 360,
+ 245,
+ 370
+ ],
+ "score": 0.89,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 246,
+ 358,
+ 506,
+ 373
+ ],
+ "score": 1.0,
+ "content": "is used to infer the novel graph’s super-class label, followed by",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 103,
+ 366,
+ 507,
+ 387
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 103,
+ 366,
+ 507,
+ 387
+ ],
+ "score": 1.0,
+ "content": "creation of super-graph on the novel graph samples and finally updating the parameters in CGAT",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 105,
+ 381,
+ 496,
+ 395
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 381,
+ 476,
+ 395
+ ],
+ "score": 1.0,
+ "content": "through the loss. Finally the evaluation is performed on the samples from the unseen test set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 476,
+ 382,
+ 492,
+ 393
+ ],
+ "score": 0.87,
+ "content": "G _ { U }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 492,
+ 381,
+ 496,
+ 395
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ }
+ ],
+ "index": 21
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 404,
+ 505,
+ 492
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 403,
+ 506,
+ 416
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 403,
+ 506,
+ 416
+ ],
+ "score": 1.0,
+ "content": "Discussion: We make the assumption that the novel test classes belong to the same set of",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 105,
+ 415,
+ 506,
+ 427
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 415,
+ 506,
+ 427
+ ],
+ "score": 1.0,
+ "content": "super-classes from the training graphs. The reason being that the novel class labeled samples are so",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 105,
+ 425,
+ 505,
+ 438
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 425,
+ 505,
+ 438
+ ],
+ "score": 1.0,
+ "content": "much fewer than the base class labeled samples, that the resulting super-graph ends up being ex-",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 106,
+ 438,
+ 505,
+ 448
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 438,
+ 505,
+ 448
+ ],
+ "score": 1.0,
+ "content": "tremely sparse and deviates a lot from the shape of the super-graph from the base classes; therefore",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 104,
+ 444,
+ 507,
+ 462
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 444,
+ 182,
+ 462
+ ],
+ "score": 1.0,
+ "content": "it severely hinders",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 183,
+ 447,
+ 215,
+ 458
+ ],
+ "score": 0.91,
+ "content": "C ^ { G A T } \\mathrm { s }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 215,
+ 444,
+ 507,
+ 462
+ ],
+ "score": 1.0,
+ "content": "ability to effectively aggregate information from the embeddings of the",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 104,
+ 457,
+ 506,
+ 471
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 457,
+ 464,
+ 471
+ ],
+ "score": 1.0,
+ "content": "novel class labeled graphs. Instead, we pass the novel graph samples through our trained",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 465,
+ 459,
+ 487,
+ 469
+ ],
+ "score": 0.89,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 487,
+ 457,
+ 506,
+ 471
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 105,
+ 469,
+ 505,
+ 483
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 469,
+ 505,
+ 483
+ ],
+ "score": 1.0,
+ "content": "infer its super-class label and this works very effectively for us, as is evidenced by our empirical",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 105,
+ 481,
+ 138,
+ 492
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 481,
+ 138,
+ 492
+ ],
+ "score": 1.0,
+ "content": "results.",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ }
+ ],
+ "index": 27.5
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 511,
+ 256,
+ 524
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 510,
+ 258,
+ 526
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 510,
+ 258,
+ 526
+ ],
+ "score": 1.0,
+ "content": "5 EXPERIMENTAL RESULTS",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ }
+ ],
+ "index": 32
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 538,
+ 248,
+ 550
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 539,
+ 249,
+ 551
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 539,
+ 249,
+ 551
+ ],
+ "score": 1.0,
+ "content": "5.1 BASELINES AND DATASETS",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ }
+ ],
+ "index": 33
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 561,
+ 505,
+ 627
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 561,
+ 505,
+ 573
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 561,
+ 505,
+ 573
+ ],
+ "score": 1.0,
+ "content": "The standard graph classification datasets do not adequately satisfy the requirements for few-shot",
+ "type": "text"
+ }
+ ],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 105,
+ 572,
+ 506,
+ 585
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 572,
+ 506,
+ 585
+ ],
+ "score": 1.0,
+ "content": "learning due to the dearth of unique class labels. Hence, we pick four new classification datasets,",
+ "type": "text"
+ }
+ ],
+ "index": 35
+ },
+ {
+ "bbox": [
+ 105,
+ 582,
+ 506,
+ 596
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 582,
+ 506,
+ 596
+ ],
+ "score": 1.0,
+ "content": "namely, Letter-High, TRIANGLES, Reddit-12K, and ENZYMES. The details and statistics for these",
+ "type": "text"
+ }
+ ],
+ "index": 36
+ },
+ {
+ "bbox": [
+ 106,
+ 594,
+ 505,
+ 607
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 594,
+ 505,
+ 607
+ ],
+ "score": 1.0,
+ "content": "datasets are given in Appendix A.1. As there do not exist any standard state-of-the-art methods for",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 105,
+ 604,
+ 505,
+ 617
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 604,
+ 505,
+ 617
+ ],
+ "score": 1.0,
+ "content": "few-shot graph classification, we chose existing baselines for standard graph classification from both",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ },
+ {
+ "bbox": [
+ 105,
+ 616,
+ 262,
+ 628
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 616,
+ 262,
+ 628
+ ],
+ "score": 1.0,
+ "content": "supervised and unsupervised methods.",
+ "type": "text"
+ }
+ ],
+ "index": 39
+ }
+ ],
+ "index": 36.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 632,
+ 505,
+ 732
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 633,
+ 505,
+ 646
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 633,
+ 505,
+ 646
+ ],
+ "score": 1.0,
+ "content": "For supervised deep learning baselines, we chose - GIN (Xu et al. (2019)), CapsGNN (Xinyi & Chen",
+ "type": "text"
+ }
+ ],
+ "index": 40
+ },
+ {
+ "bbox": [
+ 105,
+ 642,
+ 506,
+ 657
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 642,
+ 506,
+ 657
+ ],
+ "score": 1.0,
+ "content": "(2019)), and Diffpool (Lee et al. (2019)). We ran these methods with similar settings as ours, i.e.,",
+ "type": "text"
+ }
+ ],
+ "index": 41
+ },
+ {
+ "bbox": [
+ 106,
+ 655,
+ 506,
+ 668
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 655,
+ 506,
+ 668
+ ],
+ "score": 1.0,
+ "content": "by partitioning the main model into feature extraction and classifier sub-models to compare them in",
+ "type": "text"
+ }
+ ],
+ "index": 42
+ },
+ {
+ "bbox": [
+ 104,
+ 665,
+ 505,
+ 679
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 665,
+ 505,
+ 679
+ ],
+ "score": 1.0,
+ "content": "a fair and informative manner. From the unsupervised category, we consider 4 powerful SOTA",
+ "type": "text"
+ }
+ ],
+ "index": 43
+ },
+ {
+ "bbox": [
+ 105,
+ 676,
+ 506,
+ 690
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 676,
+ 506,
+ 690
+ ],
+ "score": 1.0,
+ "content": "methods - AWE (Ivanov & Burnaev (2018)), Graph2Vec (Narayanan et al. (2017)), Weisfeiler-",
+ "type": "text"
+ }
+ ],
+ "index": 44
+ },
+ {
+ "bbox": [
+ 105,
+ 687,
+ 505,
+ 700
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 687,
+ 505,
+ 700
+ ],
+ "score": 1.0,
+ "content": "Lehman subtree Kernel (Shervashidze et al. (2011)), and Graphlet count kernel (Shervashidze et al.",
+ "type": "text"
+ }
+ ],
+ "index": 45
+ },
+ {
+ "bbox": [
+ 106,
+ 699,
+ 505,
+ 712
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 699,
+ 505,
+ 712
+ ],
+ "score": 1.0,
+ "content": "(2009)). Since we want to analyze the few-shot classification abilities of these models, we essentially",
+ "type": "text"
+ }
+ ],
+ "index": 46
+ },
+ {
+ "bbox": [
+ 105,
+ 709,
+ 505,
+ 722
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 709,
+ 425,
+ 722
+ ],
+ "score": 1.0,
+ "content": "want to find out how well these algorithms can achieve class separation. We use",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 425,
+ 710,
+ 432,
+ 720
+ ],
+ "score": 0.8,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 432,
+ 709,
+ 505,
+ 722
+ ],
+ "score": 1.0,
+ "content": "-NN search on the",
+ "type": "text"
+ }
+ ],
+ "index": 47
+ },
+ {
+ "bbox": [
+ 106,
+ 721,
+ 266,
+ 733
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 721,
+ 266,
+ 733
+ ],
+ "score": 1.0,
+ "content": "output embeddings of these algorithms.",
+ "type": "text"
+ }
+ ],
+ "index": 48
+ }
+ ],
+ "index": 44
+ }
+ ],
+ "page_idx": 6,
+ "page_size": [
+ 612,
+ 792
+ ],
+ "discarded_blocks": [
+ {
+ "type": "discarded",
+ "bbox": [
+ 107,
+ 26,
+ 293,
+ 37
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 25,
+ 294,
+ 38
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 25,
+ 294,
+ 38
+ ],
+ "score": 1.0,
+ "content": "Published as a conference paper at ICLR 2020",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 303,
+ 751,
+ 309,
+ 759
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 302,
+ 750,
+ 309,
+ 762
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 302,
+ 750,
+ 309,
+ 762
+ ],
+ "score": 1.0,
+ "content": "7",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ }
+ ],
+ "para_blocks": [
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 82,
+ 270,
+ 93
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 82,
+ 270,
+ 94
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 82,
+ 141,
+ 94
+ ],
+ "score": 1.0,
+ "content": "from all",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 141,
+ 83,
+ 150,
+ 92
+ ],
+ "score": 0.82,
+ "content": "R",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 150,
+ 82,
+ 270,
+ 94
+ ],
+ "score": 1.0,
+ "content": "iterations are concatenated as",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ }
+ ],
+ "index": 0,
+ "bbox_fs": [
+ 106,
+ 82,
+ 270,
+ 94
+ ]
+ },
+ {
+ "type": "interline_equation",
+ "bbox": [
+ 267,
+ 100,
+ 340,
+ 132
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 267,
+ 100,
+ 340,
+ 132
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 267,
+ 100,
+ 340,
+ 132
+ ],
+ "score": 0.94,
+ "content": "H _ { g } = \\left\\| _ { j = 1 } ^ { R } H ^ { ( j ) } , \\right\\|",
+ "type": "interline_equation",
+ "image_path": "efd9868b636e27c4f3f13c23a86a3abe1b3f5758487b5acfc443c0f0c77fd42d.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 1.5,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 267,
+ 100,
+ 340,
+ 116.0
+ ],
+ "spans": [],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 267,
+ 116.0,
+ 340,
+ 132.0
+ ],
+ "spans": [],
+ "index": 2
+ }
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 141,
+ 505,
+ 178
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 102,
+ 137,
+ 507,
+ 158
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 102,
+ 137,
+ 131,
+ 158
+ ],
+ "score": 1.0,
+ "content": "Here,",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 132,
+ 141,
+ 216,
+ 156
+ ],
+ "score": 0.93,
+ "content": "\\begin{array} { r } { H ^ { ( j ) } = \\sum _ { v \\in V } H _ { v } ^ { ( j ) } } \\end{array}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 216,
+ 137,
+ 248,
+ 158
+ ],
+ "score": 1.0,
+ "content": ", where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 248,
+ 141,
+ 269,
+ 156
+ ],
+ "score": 0.93,
+ "content": "H _ { i } ^ { ( j ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 269,
+ 137,
+ 330,
+ 158
+ ],
+ "score": 1.0,
+ "content": "represents the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 330,
+ 145,
+ 335,
+ 154
+ ],
+ "score": 0.73,
+ "content": "i",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 335,
+ 137,
+ 450,
+ 158
+ ],
+ "score": 1.0,
+ "content": "-th node’s embedding in the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 450,
+ 145,
+ 456,
+ 155
+ ],
+ "score": 0.83,
+ "content": "j",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 457,
+ 137,
+ 507,
+ 158
+ ],
+ "score": 1.0,
+ "content": "-th iteration",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 105,
+ 154,
+ 505,
+ 169
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 154,
+ 123,
+ 169
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 124,
+ 156,
+ 130,
+ 168
+ ],
+ "score": 0.83,
+ "content": "\\parallel",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 131,
+ 154,
+ 268,
+ 169
+ ],
+ "score": 1.0,
+ "content": "denotes a concatenation operator.",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 269,
+ 156,
+ 283,
+ 168
+ ],
+ "score": 0.9,
+ "content": "H _ { g }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 283,
+ 154,
+ 470,
+ 169
+ ],
+ "score": 1.0,
+ "content": "now contains the graph embedding of a graph",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 471,
+ 158,
+ 478,
+ 167
+ ],
+ "score": 0.79,
+ "content": "g",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 478,
+ 154,
+ 505,
+ 169
+ ],
+ "score": 1.0,
+ "content": "and is",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 105,
+ 167,
+ 213,
+ 179
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 167,
+ 213,
+ 179
+ ],
+ "score": 1.0,
+ "content": "passed on to the classifier.",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ }
+ ],
+ "index": 4,
+ "bbox_fs": [
+ 102,
+ 137,
+ 507,
+ 179
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 183,
+ 505,
+ 228
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 183,
+ 505,
+ 196
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 183,
+ 505,
+ 196
+ ],
+ "score": 1.0,
+ "content": "Classifier: Here, our objective is to improve the class separation produced by the graph embeddings",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 105,
+ 194,
+ 506,
+ 207
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 194,
+ 202,
+ 207
+ ],
+ "score": 1.0,
+ "content": "of the feature extractor",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 202,
+ 195,
+ 225,
+ 207
+ ],
+ "score": 0.92,
+ "content": "F _ { \\theta } ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 225,
+ 194,
+ 506,
+ 207
+ ],
+ "score": 1.0,
+ "content": "and we do this by building a “graph of graph embeddings”, called a",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 104,
+ 205,
+ 505,
+ 219
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 205,
+ 158,
+ 219
+ ],
+ "score": 1.0,
+ "content": "super-graph",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 158,
+ 206,
+ 177,
+ 217
+ ],
+ "score": 0.9,
+ "content": "g ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 177,
+ 205,
+ 471,
+ 219
+ ],
+ "score": 1.0,
+ "content": ", where each node is a graph feature vector. We then employ our classifier",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 472,
+ 205,
+ 491,
+ 218
+ ],
+ "score": 0.91,
+ "content": "C ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 492,
+ 205,
+ 505,
+ 219
+ ],
+ "score": 1.0,
+ "content": "on",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 105,
+ 216,
+ 483,
+ 230
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 216,
+ 483,
+ 230
+ ],
+ "score": 1.0,
+ "content": "this super-graph to achieve better separation among the graph classes in the embedding space.",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ }
+ ],
+ "index": 7.5,
+ "bbox_fs": [
+ 104,
+ 183,
+ 506,
+ 230
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 233,
+ 505,
+ 332
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 231,
+ 506,
+ 247
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 231,
+ 291,
+ 247
+ ],
+ "score": 1.0,
+ "content": "During training, we first build the super-graph",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 291,
+ 234,
+ 311,
+ 245
+ ],
+ "score": 0.91,
+ "content": "g ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 311,
+ 231,
+ 506,
+ 247
+ ],
+ "score": 1.0,
+ "content": "on a batch of base-labeled graphs as a collection",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 106,
+ 244,
+ 506,
+ 257
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 244,
+ 118,
+ 257
+ ],
+ "score": 1.0,
+ "content": "of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 118,
+ 245,
+ 124,
+ 254
+ ],
+ "score": 0.79,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 125,
+ 244,
+ 272,
+ 257
+ ],
+ "score": 1.0,
+ "content": "-NN graphs, where each constituent",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 272,
+ 245,
+ 279,
+ 254
+ ],
+ "score": 0.79,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 279,
+ 244,
+ 506,
+ 257
+ ],
+ "score": 1.0,
+ "content": "-NN graph is built on the graphs belonging to the same",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 104,
+ 253,
+ 507,
+ 268
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 253,
+ 156,
+ 268
+ ],
+ "score": 1.0,
+ "content": "super-class.",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 156,
+ 256,
+ 176,
+ 267
+ ],
+ "score": 0.9,
+ "content": "g ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 176,
+ 253,
+ 429,
+ 268
+ ],
+ "score": 1.0,
+ "content": "is then passed through a multi-layered graph attention network",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 429,
+ 254,
+ 456,
+ 265
+ ],
+ "score": 0.9,
+ "content": "C ^ { \\forall G A \\mathbf { \\breve { T } } }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 457,
+ 253,
+ 507,
+ 268
+ ],
+ "score": 1.0,
+ "content": "to learn the",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 105,
+ 266,
+ 505,
+ 279
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 266,
+ 344,
+ 279
+ ],
+ "score": 1.0,
+ "content": "associated class probabilities. The features extracted from",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 344,
+ 266,
+ 366,
+ 277
+ ],
+ "score": 0.89,
+ "content": "F _ { \\theta } ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 367,
+ 266,
+ 505,
+ 279
+ ],
+ "score": 1.0,
+ "content": "are passed into the MLP network",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 107,
+ 273,
+ 507,
+ 291
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 277,
+ 128,
+ 287
+ ],
+ "score": 0.87,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 129,
+ 273,
+ 306,
+ 291
+ ],
+ "score": 1.0,
+ "content": "to learn the associated super-class labels.",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 306,
+ 277,
+ 327,
+ 287
+ ],
+ "score": 0.89,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 328,
+ 273,
+ 348,
+ 291
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 349,
+ 277,
+ 376,
+ 287
+ ],
+ "score": 0.87,
+ "content": "C ^ { \\widecheck { G } A T }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 376,
+ 273,
+ 507,
+ 291
+ ],
+ "score": 1.0,
+ "content": "combine to form our classifier",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 107,
+ 284,
+ 507,
+ 303
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 288,
+ 126,
+ 300
+ ],
+ "score": 0.86,
+ "content": "C ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 126,
+ 284,
+ 298,
+ 303
+ ],
+ "score": 1.0,
+ "content": ". The cross-entropy losses associated with",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 299,
+ 289,
+ 321,
+ 299
+ ],
+ "score": 0.89,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 321,
+ 284,
+ 340,
+ 303
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 340,
+ 288,
+ 367,
+ 298
+ ],
+ "score": 0.89,
+ "content": "C ^ { G A T }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 368,
+ 284,
+ 507,
+ 303
+ ],
+ "score": 1.0,
+ "content": "are added to give the overall loss",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 104,
+ 296,
+ 507,
+ 313
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 296,
+ 120,
+ 313
+ ],
+ "score": 1.0,
+ "content": "for",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 121,
+ 299,
+ 140,
+ 311
+ ],
+ "score": 0.9,
+ "content": "C ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 140,
+ 296,
+ 305,
+ 313
+ ],
+ "score": 1.0,
+ "content": ". The intuition behind the construction of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 305,
+ 300,
+ 324,
+ 311
+ ],
+ "score": 0.91,
+ "content": "g ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 325,
+ 296,
+ 356,
+ 313
+ ],
+ "score": 1.0,
+ "content": "to train",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 357,
+ 298,
+ 384,
+ 309
+ ],
+ "score": 0.9,
+ "content": "C ^ { G A T }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 384,
+ 296,
+ 507,
+ 313
+ ],
+ "score": 1.0,
+ "content": "on was to further improve the",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 106,
+ 310,
+ 505,
+ 323
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 310,
+ 505,
+ 323
+ ],
+ "score": 1.0,
+ "content": "existing cluster separation based on graph spectral measures by introducing a relational inductive",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 104,
+ 318,
+ 359,
+ 335
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 318,
+ 359,
+ 335
+ ],
+ "score": 1.0,
+ "content": "bias (Battaglia et al., 2018) that is inherent to the GNN CGAT .",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ }
+ ],
+ "index": 14,
+ "bbox_fs": [
+ 104,
+ 231,
+ 507,
+ 335
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 338,
+ 504,
+ 393
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 337,
+ 506,
+ 350
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 337,
+ 506,
+ 350
+ ],
+ "score": 1.0,
+ "content": "Recall that we adopt an initialization method (described in 3). In our fine-tuning stage, novel class",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 105,
+ 348,
+ 506,
+ 362
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 348,
+ 188,
+ 362
+ ],
+ "score": 1.0,
+ "content": "labeled graphs from",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 189,
+ 349,
+ 205,
+ 360
+ ],
+ "score": 0.89,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 205,
+ 348,
+ 506,
+ 362
+ ],
+ "score": 1.0,
+ "content": "are input to the network. The pre-trained parameters learned by the feature",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 105,
+ 358,
+ 506,
+ 373
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 358,
+ 145,
+ 373
+ ],
+ "score": 1.0,
+ "content": "extractor",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 145,
+ 360,
+ 167,
+ 372
+ ],
+ "score": 0.92,
+ "content": "F _ { \\theta } ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 168,
+ 358,
+ 223,
+ 373
+ ],
+ "score": 1.0,
+ "content": "are fixed and",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 223,
+ 360,
+ 245,
+ 370
+ ],
+ "score": 0.89,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 246,
+ 358,
+ 506,
+ 373
+ ],
+ "score": 1.0,
+ "content": "is used to infer the novel graph’s super-class label, followed by",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 103,
+ 366,
+ 507,
+ 387
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 103,
+ 366,
+ 507,
+ 387
+ ],
+ "score": 1.0,
+ "content": "creation of super-graph on the novel graph samples and finally updating the parameters in CGAT",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 105,
+ 381,
+ 496,
+ 395
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 381,
+ 476,
+ 395
+ ],
+ "score": 1.0,
+ "content": "through the loss. Finally the evaluation is performed on the samples from the unseen test set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 476,
+ 382,
+ 492,
+ 393
+ ],
+ "score": 0.87,
+ "content": "G _ { U }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 492,
+ 381,
+ 496,
+ 395
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ }
+ ],
+ "index": 21,
+ "bbox_fs": [
+ 103,
+ 337,
+ 507,
+ 395
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 404,
+ 505,
+ 492
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 403,
+ 506,
+ 416
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 403,
+ 506,
+ 416
+ ],
+ "score": 1.0,
+ "content": "Discussion: We make the assumption that the novel test classes belong to the same set of",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 105,
+ 415,
+ 506,
+ 427
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 415,
+ 506,
+ 427
+ ],
+ "score": 1.0,
+ "content": "super-classes from the training graphs. The reason being that the novel class labeled samples are so",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 105,
+ 425,
+ 505,
+ 438
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 425,
+ 505,
+ 438
+ ],
+ "score": 1.0,
+ "content": "much fewer than the base class labeled samples, that the resulting super-graph ends up being ex-",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 106,
+ 438,
+ 505,
+ 448
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 438,
+ 505,
+ 448
+ ],
+ "score": 1.0,
+ "content": "tremely sparse and deviates a lot from the shape of the super-graph from the base classes; therefore",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 104,
+ 444,
+ 507,
+ 462
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 444,
+ 182,
+ 462
+ ],
+ "score": 1.0,
+ "content": "it severely hinders",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 183,
+ 447,
+ 215,
+ 458
+ ],
+ "score": 0.91,
+ "content": "C ^ { G A T } \\mathrm { s }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 215,
+ 444,
+ 507,
+ 462
+ ],
+ "score": 1.0,
+ "content": "ability to effectively aggregate information from the embeddings of the",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 104,
+ 457,
+ 506,
+ 471
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 457,
+ 464,
+ 471
+ ],
+ "score": 1.0,
+ "content": "novel class labeled graphs. Instead, we pass the novel graph samples through our trained",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 465,
+ 459,
+ 487,
+ 469
+ ],
+ "score": 0.89,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 487,
+ 457,
+ 506,
+ 471
+ ],
+ "score": 1.0,
+ "content": "and",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 105,
+ 469,
+ 505,
+ 483
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 469,
+ 505,
+ 483
+ ],
+ "score": 1.0,
+ "content": "infer its super-class label and this works very effectively for us, as is evidenced by our empirical",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 105,
+ 481,
+ 138,
+ 492
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 481,
+ 138,
+ 492
+ ],
+ "score": 1.0,
+ "content": "results.",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ }
+ ],
+ "index": 27.5,
+ "bbox_fs": [
+ 104,
+ 403,
+ 507,
+ 492
+ ]
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 511,
+ 256,
+ 524
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 510,
+ 258,
+ 526
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 510,
+ 258,
+ 526
+ ],
+ "score": 1.0,
+ "content": "5 EXPERIMENTAL RESULTS",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ }
+ ],
+ "index": 32
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 538,
+ 248,
+ 550
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 539,
+ 249,
+ 551
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 539,
+ 249,
+ 551
+ ],
+ "score": 1.0,
+ "content": "5.1 BASELINES AND DATASETS",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ }
+ ],
+ "index": 33
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 561,
+ 505,
+ 627
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 561,
+ 505,
+ 573
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 561,
+ 505,
+ 573
+ ],
+ "score": 1.0,
+ "content": "The standard graph classification datasets do not adequately satisfy the requirements for few-shot",
+ "type": "text"
+ }
+ ],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 105,
+ 572,
+ 506,
+ 585
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 572,
+ 506,
+ 585
+ ],
+ "score": 1.0,
+ "content": "learning due to the dearth of unique class labels. Hence, we pick four new classification datasets,",
+ "type": "text"
+ }
+ ],
+ "index": 35
+ },
+ {
+ "bbox": [
+ 105,
+ 582,
+ 506,
+ 596
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 582,
+ 506,
+ 596
+ ],
+ "score": 1.0,
+ "content": "namely, Letter-High, TRIANGLES, Reddit-12K, and ENZYMES. The details and statistics for these",
+ "type": "text"
+ }
+ ],
+ "index": 36
+ },
+ {
+ "bbox": [
+ 106,
+ 594,
+ 505,
+ 607
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 594,
+ 505,
+ 607
+ ],
+ "score": 1.0,
+ "content": "datasets are given in Appendix A.1. As there do not exist any standard state-of-the-art methods for",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 105,
+ 604,
+ 505,
+ 617
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 604,
+ 505,
+ 617
+ ],
+ "score": 1.0,
+ "content": "few-shot graph classification, we chose existing baselines for standard graph classification from both",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ },
+ {
+ "bbox": [
+ 105,
+ 616,
+ 262,
+ 628
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 616,
+ 262,
+ 628
+ ],
+ "score": 1.0,
+ "content": "supervised and unsupervised methods.",
+ "type": "text"
+ }
+ ],
+ "index": 39
+ }
+ ],
+ "index": 36.5,
+ "bbox_fs": [
+ 105,
+ 561,
+ 506,
+ 628
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 632,
+ 505,
+ 732
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 633,
+ 505,
+ 646
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 633,
+ 505,
+ 646
+ ],
+ "score": 1.0,
+ "content": "For supervised deep learning baselines, we chose - GIN (Xu et al. (2019)), CapsGNN (Xinyi & Chen",
+ "type": "text"
+ }
+ ],
+ "index": 40
+ },
+ {
+ "bbox": [
+ 105,
+ 642,
+ 506,
+ 657
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 642,
+ 506,
+ 657
+ ],
+ "score": 1.0,
+ "content": "(2019)), and Diffpool (Lee et al. (2019)). We ran these methods with similar settings as ours, i.e.,",
+ "type": "text"
+ }
+ ],
+ "index": 41
+ },
+ {
+ "bbox": [
+ 106,
+ 655,
+ 506,
+ 668
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 655,
+ 506,
+ 668
+ ],
+ "score": 1.0,
+ "content": "by partitioning the main model into feature extraction and classifier sub-models to compare them in",
+ "type": "text"
+ }
+ ],
+ "index": 42
+ },
+ {
+ "bbox": [
+ 104,
+ 665,
+ 505,
+ 679
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 665,
+ 505,
+ 679
+ ],
+ "score": 1.0,
+ "content": "a fair and informative manner. From the unsupervised category, we consider 4 powerful SOTA",
+ "type": "text"
+ }
+ ],
+ "index": 43
+ },
+ {
+ "bbox": [
+ 105,
+ 676,
+ 506,
+ 690
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 676,
+ 506,
+ 690
+ ],
+ "score": 1.0,
+ "content": "methods - AWE (Ivanov & Burnaev (2018)), Graph2Vec (Narayanan et al. (2017)), Weisfeiler-",
+ "type": "text"
+ }
+ ],
+ "index": 44
+ },
+ {
+ "bbox": [
+ 105,
+ 687,
+ 505,
+ 700
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 687,
+ 505,
+ 700
+ ],
+ "score": 1.0,
+ "content": "Lehman subtree Kernel (Shervashidze et al. (2011)), and Graphlet count kernel (Shervashidze et al.",
+ "type": "text"
+ }
+ ],
+ "index": 45
+ },
+ {
+ "bbox": [
+ 106,
+ 699,
+ 505,
+ 712
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 699,
+ 505,
+ 712
+ ],
+ "score": 1.0,
+ "content": "(2009)). Since we want to analyze the few-shot classification abilities of these models, we essentially",
+ "type": "text"
+ }
+ ],
+ "index": 46
+ },
+ {
+ "bbox": [
+ 105,
+ 709,
+ 505,
+ 722
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 709,
+ 425,
+ 722
+ ],
+ "score": 1.0,
+ "content": "want to find out how well these algorithms can achieve class separation. We use",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 425,
+ 710,
+ 432,
+ 720
+ ],
+ "score": 0.8,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 432,
+ 709,
+ 505,
+ 722
+ ],
+ "score": 1.0,
+ "content": "-NN search on the",
+ "type": "text"
+ }
+ ],
+ "index": 47
+ },
+ {
+ "bbox": [
+ 106,
+ 721,
+ 266,
+ 733
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 721,
+ 266,
+ 733
+ ],
+ "score": 1.0,
+ "content": "output embeddings of these algorithms.",
+ "type": "text"
+ }
+ ],
+ "index": 48
+ }
+ ],
+ "index": 44,
+ "bbox_fs": [
+ 104,
+ 633,
+ 506,
+ 733
+ ]
+ }
+ ]
+ },
+ {
+ "preproc_blocks": [
+ {
+ "type": "table",
+ "bbox": [
+ 116,
+ 119,
+ 495,
+ 219
+ ],
+ "blocks": [
+ {
+ "type": "table_caption",
+ "bbox": [
+ 106,
+ 80,
+ 504,
+ 103
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 80,
+ 505,
+ 93
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 80,
+ 505,
+ 93
+ ],
+ "score": 1.0,
+ "content": "Table 1: Results for various few-shot scenarios on Letter-High and TRIANGLES datasets. The best",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 105,
+ 91,
+ 407,
+ 103
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 91,
+ 407,
+ 103
+ ],
+ "score": 1.0,
+ "content": "results are highlighted in bold while the second best results are underlined.",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ }
+ ],
+ "index": 0.5
+ },
+ {
+ "type": "table_body",
+ "bbox": [
+ 116,
+ 119,
+ 495,
+ 219
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 116,
+ 119,
+ 495,
+ 219
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 119,
+ 495,
+ 219
+ ],
+ "score": 0.982,
+ "html": "| Method | Letter-High | TRIANGLES |
| 5-shot | 10-shot | 20-shot | 5-shot | 10-shot | 20-shot |
| WL | 65.27 ± 7.67 | 68.39± 4.69 | 72.69±3.02 | 51.25± 4.02 | 53.26 ± 2.95 | 57.74 ± 2.88 |
| Graphlet | 33.76 ± 6.94 | 37.59 ± 4.60 | 41.11 ± 3.71 | 40.17 ± 3.18 | 43.76 ± 3.09 | 45.90 ±2.65 |
| AWE | 40.60 ± 3.91 | 42.20 ±2.87 | 43.12 ± 1.00 | 39.36± 3.85 | 42.58 ± 3.11 | 44.98 ± 1.54 |
| Graph2Vec | 66.12 ± 5.21 | 68.17 ± 4.26 | 70.28 ± 2.81 | 48.38 ± 3.85 | 50.16 ± 4.15 | 54.90 ± 3.01 |
| Diffpool | 58.69 ± 6.39 | 61.59 ± 5.21 | 64.67 ± 3.21 | 64.17 ± 5.87 | 67.12 ± 4.29 | 73.27 ± 3.29 |
| CapsGNN | 56.60 ± 7.86 | 60.67 ± 5.24 | 63.97 ± 3.69 | 65.40 ± 6.13 | 68.37 ± 3.67 | 73.06 ± 3.64 |
| GIN | 65.83 ± 7.17 | 69.16 ± 5.14 | 73.28 ± 2.17 | 63.80 ± 5.61 | 67.30 ± 4.35 | 72.55 ± 1.97 |
| GIN-k-NN | 63.52 ± 7.27 | 65.66± 8.69 | 67.45± 8.76 | 58.34 ± 3.91 | 61.55± 3.19 | 63.45± 2.76 |
| OurMethod-GCN | 68.69 ± 6.50 | 72.80 ± 4.12 | 75.17 ± 3.11 | 69.37 ± 4.92 | 73.11 ± 3.94 | 77.86 ± 2.84 |
| OurMethod-GAT | 69.91 ± 5.90 | 73.28± 3.46 | 77.38 ± 1.58 | 71.40 ± 4.34 | 75.60± 3.67 | 80.04 ± 2.20 |
",
+ "type": "table",
+ "image_path": "8342046b53e913f031efd7336ea4bc6015e333001246da39f501904f0b0e11e3.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 3,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 116,
+ 119,
+ 495,
+ 152.33333333333334
+ ],
+ "spans": [],
+ "index": 2
+ },
+ {
+ "bbox": [
+ 116,
+ 152.33333333333334,
+ 495,
+ 185.66666666666669
+ ],
+ "spans": [],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 116,
+ 185.66666666666669,
+ 495,
+ 219.00000000000003
+ ],
+ "spans": [],
+ "index": 4
+ }
+ ]
+ }
+ ],
+ "index": 1.75
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 108,
+ 240,
+ 504,
+ 273
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 240,
+ 504,
+ 252
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 240,
+ 504,
+ 252
+ ],
+ "score": 1.0,
+ "content": "Further configuration and implementation details for the baselines can be found in Appendix A.2.",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 106,
+ 250,
+ 505,
+ 263
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 250,
+ 505,
+ 263
+ ],
+ "score": 1.0,
+ "content": "We also emphasize the benefit of using a GNN as a classifier by showing the adaptation of our model",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 105,
+ 261,
+ 495,
+ 275
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 261,
+ 495,
+ 275
+ ],
+ "score": 1.0,
+ "content": "to semi-supervised fine-tuning (in Appendix A.5) and active learning (in Appendix A.6) settings.",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ }
+ ],
+ "index": 6
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 288,
+ 217,
+ 299
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 286,
+ 219,
+ 301
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 286,
+ 219,
+ 301
+ ],
+ "score": 1.0,
+ "content": "5.2 FEW-SHOT RESULTS",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ }
+ ],
+ "index": 8
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 109,
+ 309,
+ 504,
+ 342
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 107,
+ 309,
+ 505,
+ 321
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 309,
+ 505,
+ 321
+ ],
+ "score": 1.0,
+ "content": "We consider two variants of our model as naive baselines. In the first variant, we replace our GAT",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 107,
+ 320,
+ 505,
+ 332
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 320,
+ 505,
+ 332
+ ],
+ "score": 1.0,
+ "content": "classifier with GCN Kipf & Welling (2016). We call this model OurMethod-GCN. This variant is",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 106,
+ 331,
+ 287,
+ 343
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 331,
+ 287,
+ 343
+ ],
+ "score": 1.0,
+ "content": "used to justify the choice of GAT over GCN.",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ }
+ ],
+ "index": 10
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 347,
+ 505,
+ 392
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 347,
+ 505,
+ 360
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 347,
+ 365,
+ 360
+ ],
+ "score": 1.0,
+ "content": "In the second variant, we replace the entire classifier with the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 365,
+ 348,
+ 372,
+ 358
+ ],
+ "score": 0.78,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 372,
+ 347,
+ 505,
+ 360
+ ],
+ "score": 1.0,
+ "content": "-NN algorithm over the features",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 106,
+ 360,
+ 505,
+ 370
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 360,
+ 434,
+ 370
+ ],
+ "score": 1.0,
+ "content": "extracted from various layers of the feature extractor. We call this variant GIN-",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 435,
+ 360,
+ 441,
+ 369
+ ],
+ "score": 0.57,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 442,
+ 360,
+ 505,
+ 370
+ ],
+ "score": 1.0,
+ "content": "-NN and this is",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 106,
+ 370,
+ 506,
+ 382
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 370,
+ 506,
+ 382
+ ],
+ "score": 1.0,
+ "content": "introduced to emphasize the significance of building a super-graph and using a GAT on it as a",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 106,
+ 381,
+ 300,
+ 392
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 381,
+ 300,
+ 392
+ ],
+ "score": 1.0,
+ "content": "classifier to exploit the relational inductive bias.",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ }
+ ],
+ "index": 13.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 397,
+ 505,
+ 464
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 396,
+ 505,
+ 410
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 396,
+ 266,
+ 410
+ ],
+ "score": 1.0,
+ "content": "The results for all the datasets in various",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 266,
+ 399,
+ 272,
+ 409
+ ],
+ "score": 0.78,
+ "content": "q",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 272,
+ 396,
+ 361,
+ 410
+ ],
+ "score": 1.0,
+ "content": "-shot scenarios, where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 361,
+ 397,
+ 423,
+ 410
+ ],
+ "score": 0.93,
+ "content": "q \\in \\{ 5 , 1 0 , 2 0 \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 423,
+ 396,
+ 505,
+ 410
+ ],
+ "score": 1.0,
+ "content": "are given in Table 1.",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 105,
+ 408,
+ 506,
+ 421
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 408,
+ 506,
+ 421
+ ],
+ "score": 1.0,
+ "content": "We run each model 50 times and report averaged results. In every run, we select a different novel",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 106,
+ 420,
+ 505,
+ 431
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 420,
+ 167,
+ 431
+ ],
+ "score": 1.0,
+ "content": "labeled subset",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 167,
+ 420,
+ 184,
+ 430
+ ],
+ "score": 0.89,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 184,
+ 420,
+ 505,
+ 431
+ ],
+ "score": 1.0,
+ "content": "for fine-tuning the classifiers of the models. The evaluation for all models is",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 105,
+ 430,
+ 505,
+ 443
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 430,
+ 405,
+ 443
+ ],
+ "score": 1.0,
+ "content": "done by randomly selecting a subset of 500 samples from the testing set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 405,
+ 431,
+ 421,
+ 442
+ ],
+ "score": 0.9,
+ "content": "G _ { U }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 421,
+ 430,
+ 505,
+ 443
+ ],
+ "score": 1.0,
+ "content": "for Letter-High and",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 105,
+ 441,
+ 505,
+ 453
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 441,
+ 505,
+ 453
+ ],
+ "score": 1.0,
+ "content": "TRIANGLES, whereas over 150 for ENZYMES and 300 for Reddit dataset and averaging over 10",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 106,
+ 451,
+ 204,
+ 464
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 451,
+ 204,
+ 464
+ ],
+ "score": 1.0,
+ "content": "such random selections.",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ }
+ ],
+ "index": 18.5
+ },
+ {
+ "type": "table",
+ "bbox": [
+ 117,
+ 523,
+ 494,
+ 624
+ ],
+ "blocks": [
+ {
+ "type": "table_caption",
+ "bbox": [
+ 106,
+ 484,
+ 504,
+ 507
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 485,
+ 505,
+ 496
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 485,
+ 344,
+ 496
+ ],
+ "score": 1.0,
+ "content": "Table 2: Results for various few-shot scenarios on Reddit-",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 345,
+ 485,
+ 363,
+ 495
+ ],
+ "score": 0.52,
+ "content": "I 2 K",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 363,
+ 485,
+ 505,
+ 496
+ ],
+ "score": 1.0,
+ "content": "and ENZYMES datasets. The best",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 105,
+ 496,
+ 407,
+ 508
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 496,
+ 407,
+ 508
+ ],
+ "score": 1.0,
+ "content": "results are highlighted in bold while the second best results are underlined.",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ }
+ ],
+ "index": 22.5
+ },
+ {
+ "type": "table_body",
+ "bbox": [
+ 117,
+ 523,
+ 494,
+ 624
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 117,
+ 523,
+ 494,
+ 624
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 117,
+ 523,
+ 494,
+ 624
+ ],
+ "score": 0.98,
+ "html": "| Method | Reddit-12K | ENZYMES |
| 5-shot | 10-shot | 20-shot | 5-shot | 10-shot | 20-shot |
| WL | 40.26± 5.17 | 42.57±3.69 | 44.41±3.43 | 55.78± 4.72 | 58.47±3.84 | 60.1±3.18 |
| Graphlet | 33.76 ± 6.94 | 37.59 ± 4.60 | 41.11 ± 3.71 | 53.17 ± 5.92 | 55.30 ±3.78 | 56.90 ±3.79 |
| AWE | 30.24± 2.34 | 33.44 ±2.04 | 36.13 ± 1.89 | 43.75 ±1.85 | 45.58 ± 2.11 | 49.98 ± 1.54 |
| Graph2Vec | 27.85 ± 4.21 | 29.97 ± 3.17 | 32.75 ± 2.02 | 55.88 ± 4.86 | 58.22 ± 4.30 | 62.28 ± 4.14 |
| Diffpool | 35.24 ± 5.69 | 37.43 ± 3.94 | 39.11 ± 3.52 | 45.64 ± 4.56 | 49.64 ± 4.23 | 54.27 ± 3.94 |
| CapsGNN | 36.58 ±4.28 | 39.16 ± 3.73 | 41.27 ± 3.12 | 52.67 ± 5.51 | 55.31 ± 4.23 | 59.34 ± 4.02 |
| GIN | 40.36 ± 4.69 | 43.70 ± 3.98 | 46.28 ± 3.49 | 55.73 ± 5.80 | 58.83 ±5.32 | 61.12 ± 4.64 |
| GIN-k-NN | 41.31± 2.84 | 43.58±2.80 | 45.12 ± 2.19 | 57.24 ± 7.06 | 59.34± 5.24 | 60.49±3.48 |
| OurMethod-GCN | 40.77 ± 4.32 | 44.28 ± 3.86 | 48.67 ± 4.22 | 54.34 ± 5.64 | 58.16 ± 4.39 | 60.86 ± 3.74 |
| OurMethod-GAT | 41.59 ± 4.12 | 45.67± 3.68 | 50.34± 2.71 | 55.42 ± 5.74 | 60.64 ± 3.84 | 62.81 ± 3.56 |
",
+ "type": "table",
+ "image_path": "b955792ef88c002b65b8845aeab3812c807e8919a39ab39ae60f47bbce68009d.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 25,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 117,
+ 523,
+ 494,
+ 556.6666666666666
+ ],
+ "spans": [],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 117,
+ 556.6666666666666,
+ 494,
+ 590.3333333333333
+ ],
+ "spans": [],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 117,
+ 590.3333333333333,
+ 494,
+ 623.9999999999999
+ ],
+ "spans": [],
+ "index": 26
+ }
+ ]
+ }
+ ],
+ "index": 23.75
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 643,
+ 505,
+ 732
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 643,
+ 505,
+ 657
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 643,
+ 505,
+ 657
+ ],
+ "score": 1.0,
+ "content": "The results clearly show that our proposed method and its GCN variant (i.e., OurMethod-GCN)",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 106,
+ 654,
+ 505,
+ 666
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 654,
+ 231,
+ 666
+ ],
+ "score": 1.0,
+ "content": "outperform the baselines. GIN-",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 232,
+ 655,
+ 238,
+ 665
+ ],
+ "score": 0.76,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 239,
+ 654,
+ 505,
+ 666
+ ],
+ "score": 1.0,
+ "content": "-NN shows significant degradation in results on nearly all the three",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 105,
+ 666,
+ 505,
+ 678
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 666,
+ 505,
+ 678
+ ],
+ "score": 1.0,
+ "content": "paradigms for all datasets with exceptions on 5-shot and 10-shot on ENZYMES dataset, thus strongly",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 106,
+ 677,
+ 505,
+ 689
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 677,
+ 505,
+ 689
+ ],
+ "score": 1.0,
+ "content": "indicating that the improvements of our method can primarily be attributed to our GNN classifier",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 105,
+ 686,
+ 506,
+ 701
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 686,
+ 506,
+ 701
+ ],
+ "score": 1.0,
+ "content": "fed with the super-graph constructed from our proposed method. The improvements in results are",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ },
+ {
+ "bbox": [
+ 105,
+ 699,
+ 505,
+ 711
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 699,
+ 505,
+ 711
+ ],
+ "score": 1.0,
+ "content": "higher on TRIANGLES and Reddit datasets in contrast to Letter-High and ENZYMES, which can be",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 106,
+ 710,
+ 505,
+ 722
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 710,
+ 505,
+ 722
+ ],
+ "score": 1.0,
+ "content": "attributed to the smaller size of the graphs in Letter-High making it difficult to distinguish based on",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ },
+ {
+ "bbox": [
+ 105,
+ 721,
+ 505,
+ 733
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 721,
+ 505,
+ 733
+ ],
+ "score": 1.0,
+ "content": "graph spectra alone, whereas the complex and highly inter-related structure of enzymes makes it dif-",
+ "type": "text"
+ }
+ ],
+ "index": 34
+ }
+ ],
+ "index": 30.5
+ }
+ ],
+ "page_idx": 7,
+ "page_size": [
+ 612,
+ 792
+ ],
+ "discarded_blocks": [
+ {
+ "type": "discarded",
+ "bbox": [
+ 108,
+ 27,
+ 293,
+ 37
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "score": 1.0,
+ "content": "Published as a conference paper at ICLR 2020",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 302,
+ 751,
+ 308,
+ 760
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 300,
+ 750,
+ 309,
+ 761
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 300,
+ 750,
+ 309,
+ 761
+ ],
+ "score": 1.0,
+ "content": "8",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ }
+ ],
+ "para_blocks": [
+ {
+ "type": "table",
+ "bbox": [
+ 116,
+ 119,
+ 495,
+ 219
+ ],
+ "blocks": [
+ {
+ "type": "table_caption",
+ "bbox": [
+ 106,
+ 80,
+ 504,
+ 103
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 80,
+ 505,
+ 93
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 80,
+ 505,
+ 93
+ ],
+ "score": 1.0,
+ "content": "Table 1: Results for various few-shot scenarios on Letter-High and TRIANGLES datasets. The best",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 105,
+ 91,
+ 407,
+ 103
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 91,
+ 407,
+ 103
+ ],
+ "score": 1.0,
+ "content": "results are highlighted in bold while the second best results are underlined.",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ }
+ ],
+ "index": 0.5
+ },
+ {
+ "type": "table_body",
+ "bbox": [
+ 116,
+ 119,
+ 495,
+ 219
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 116,
+ 119,
+ 495,
+ 219
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 119,
+ 495,
+ 219
+ ],
+ "score": 0.982,
+ "html": "| Method | Letter-High | TRIANGLES |
| 5-shot | 10-shot | 20-shot | 5-shot | 10-shot | 20-shot |
| WL | 65.27 ± 7.67 | 68.39± 4.69 | 72.69±3.02 | 51.25± 4.02 | 53.26 ± 2.95 | 57.74 ± 2.88 |
| Graphlet | 33.76 ± 6.94 | 37.59 ± 4.60 | 41.11 ± 3.71 | 40.17 ± 3.18 | 43.76 ± 3.09 | 45.90 ±2.65 |
| AWE | 40.60 ± 3.91 | 42.20 ±2.87 | 43.12 ± 1.00 | 39.36± 3.85 | 42.58 ± 3.11 | 44.98 ± 1.54 |
| Graph2Vec | 66.12 ± 5.21 | 68.17 ± 4.26 | 70.28 ± 2.81 | 48.38 ± 3.85 | 50.16 ± 4.15 | 54.90 ± 3.01 |
| Diffpool | 58.69 ± 6.39 | 61.59 ± 5.21 | 64.67 ± 3.21 | 64.17 ± 5.87 | 67.12 ± 4.29 | 73.27 ± 3.29 |
| CapsGNN | 56.60 ± 7.86 | 60.67 ± 5.24 | 63.97 ± 3.69 | 65.40 ± 6.13 | 68.37 ± 3.67 | 73.06 ± 3.64 |
| GIN | 65.83 ± 7.17 | 69.16 ± 5.14 | 73.28 ± 2.17 | 63.80 ± 5.61 | 67.30 ± 4.35 | 72.55 ± 1.97 |
| GIN-k-NN | 63.52 ± 7.27 | 65.66± 8.69 | 67.45± 8.76 | 58.34 ± 3.91 | 61.55± 3.19 | 63.45± 2.76 |
| OurMethod-GCN | 68.69 ± 6.50 | 72.80 ± 4.12 | 75.17 ± 3.11 | 69.37 ± 4.92 | 73.11 ± 3.94 | 77.86 ± 2.84 |
| OurMethod-GAT | 69.91 ± 5.90 | 73.28± 3.46 | 77.38 ± 1.58 | 71.40 ± 4.34 | 75.60± 3.67 | 80.04 ± 2.20 |
",
+ "type": "table",
+ "image_path": "8342046b53e913f031efd7336ea4bc6015e333001246da39f501904f0b0e11e3.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 3,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 116,
+ 119,
+ 495,
+ 152.33333333333334
+ ],
+ "spans": [],
+ "index": 2
+ },
+ {
+ "bbox": [
+ 116,
+ 152.33333333333334,
+ 495,
+ 185.66666666666669
+ ],
+ "spans": [],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 116,
+ 185.66666666666669,
+ 495,
+ 219.00000000000003
+ ],
+ "spans": [],
+ "index": 4
+ }
+ ]
+ }
+ ],
+ "index": 1.75
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 108,
+ 240,
+ 504,
+ 273
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 240,
+ 504,
+ 252
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 240,
+ 504,
+ 252
+ ],
+ "score": 1.0,
+ "content": "Further configuration and implementation details for the baselines can be found in Appendix A.2.",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 106,
+ 250,
+ 505,
+ 263
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 250,
+ 505,
+ 263
+ ],
+ "score": 1.0,
+ "content": "We also emphasize the benefit of using a GNN as a classifier by showing the adaptation of our model",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 105,
+ 261,
+ 495,
+ 275
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 261,
+ 495,
+ 275
+ ],
+ "score": 1.0,
+ "content": "to semi-supervised fine-tuning (in Appendix A.5) and active learning (in Appendix A.6) settings.",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ }
+ ],
+ "index": 6,
+ "bbox_fs": [
+ 105,
+ 240,
+ 505,
+ 275
+ ]
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 288,
+ 217,
+ 299
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 286,
+ 219,
+ 301
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 286,
+ 219,
+ 301
+ ],
+ "score": 1.0,
+ "content": "5.2 FEW-SHOT RESULTS",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ }
+ ],
+ "index": 8
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 109,
+ 309,
+ 504,
+ 342
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 107,
+ 309,
+ 505,
+ 321
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 309,
+ 505,
+ 321
+ ],
+ "score": 1.0,
+ "content": "We consider two variants of our model as naive baselines. In the first variant, we replace our GAT",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 107,
+ 320,
+ 505,
+ 332
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 320,
+ 505,
+ 332
+ ],
+ "score": 1.0,
+ "content": "classifier with GCN Kipf & Welling (2016). We call this model OurMethod-GCN. This variant is",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 106,
+ 331,
+ 287,
+ 343
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 331,
+ 287,
+ 343
+ ],
+ "score": 1.0,
+ "content": "used to justify the choice of GAT over GCN.",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ }
+ ],
+ "index": 10,
+ "bbox_fs": [
+ 106,
+ 309,
+ 505,
+ 343
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 347,
+ 505,
+ 392
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 347,
+ 505,
+ 360
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 347,
+ 365,
+ 360
+ ],
+ "score": 1.0,
+ "content": "In the second variant, we replace the entire classifier with the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 365,
+ 348,
+ 372,
+ 358
+ ],
+ "score": 0.78,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 372,
+ 347,
+ 505,
+ 360
+ ],
+ "score": 1.0,
+ "content": "-NN algorithm over the features",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 106,
+ 360,
+ 505,
+ 370
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 360,
+ 434,
+ 370
+ ],
+ "score": 1.0,
+ "content": "extracted from various layers of the feature extractor. We call this variant GIN-",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 435,
+ 360,
+ 441,
+ 369
+ ],
+ "score": 0.57,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 442,
+ 360,
+ 505,
+ 370
+ ],
+ "score": 1.0,
+ "content": "-NN and this is",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 106,
+ 370,
+ 506,
+ 382
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 370,
+ 506,
+ 382
+ ],
+ "score": 1.0,
+ "content": "introduced to emphasize the significance of building a super-graph and using a GAT on it as a",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 106,
+ 381,
+ 300,
+ 392
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 381,
+ 300,
+ 392
+ ],
+ "score": 1.0,
+ "content": "classifier to exploit the relational inductive bias.",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ }
+ ],
+ "index": 13.5,
+ "bbox_fs": [
+ 105,
+ 347,
+ 506,
+ 392
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 397,
+ 505,
+ 464
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 396,
+ 505,
+ 410
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 396,
+ 266,
+ 410
+ ],
+ "score": 1.0,
+ "content": "The results for all the datasets in various",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 266,
+ 399,
+ 272,
+ 409
+ ],
+ "score": 0.78,
+ "content": "q",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 272,
+ 396,
+ 361,
+ 410
+ ],
+ "score": 1.0,
+ "content": "-shot scenarios, where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 361,
+ 397,
+ 423,
+ 410
+ ],
+ "score": 0.93,
+ "content": "q \\in \\{ 5 , 1 0 , 2 0 \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 423,
+ 396,
+ 505,
+ 410
+ ],
+ "score": 1.0,
+ "content": "are given in Table 1.",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 105,
+ 408,
+ 506,
+ 421
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 408,
+ 506,
+ 421
+ ],
+ "score": 1.0,
+ "content": "We run each model 50 times and report averaged results. In every run, we select a different novel",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 106,
+ 420,
+ 505,
+ 431
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 420,
+ 167,
+ 431
+ ],
+ "score": 1.0,
+ "content": "labeled subset",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 167,
+ 420,
+ 184,
+ 430
+ ],
+ "score": 0.89,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 184,
+ 420,
+ 505,
+ 431
+ ],
+ "score": 1.0,
+ "content": "for fine-tuning the classifiers of the models. The evaluation for all models is",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 105,
+ 430,
+ 505,
+ 443
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 430,
+ 405,
+ 443
+ ],
+ "score": 1.0,
+ "content": "done by randomly selecting a subset of 500 samples from the testing set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 405,
+ 431,
+ 421,
+ 442
+ ],
+ "score": 0.9,
+ "content": "G _ { U }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 421,
+ 430,
+ 505,
+ 443
+ ],
+ "score": 1.0,
+ "content": "for Letter-High and",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 105,
+ 441,
+ 505,
+ 453
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 441,
+ 505,
+ 453
+ ],
+ "score": 1.0,
+ "content": "TRIANGLES, whereas over 150 for ENZYMES and 300 for Reddit dataset and averaging over 10",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 106,
+ 451,
+ 204,
+ 464
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 451,
+ 204,
+ 464
+ ],
+ "score": 1.0,
+ "content": "such random selections.",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ }
+ ],
+ "index": 18.5,
+ "bbox_fs": [
+ 105,
+ 396,
+ 506,
+ 464
+ ]
+ },
+ {
+ "type": "table",
+ "bbox": [
+ 117,
+ 523,
+ 494,
+ 624
+ ],
+ "blocks": [
+ {
+ "type": "table_caption",
+ "bbox": [
+ 106,
+ 484,
+ 504,
+ 507
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 485,
+ 505,
+ 496
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 485,
+ 344,
+ 496
+ ],
+ "score": 1.0,
+ "content": "Table 2: Results for various few-shot scenarios on Reddit-",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 345,
+ 485,
+ 363,
+ 495
+ ],
+ "score": 0.52,
+ "content": "I 2 K",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 363,
+ 485,
+ 505,
+ 496
+ ],
+ "score": 1.0,
+ "content": "and ENZYMES datasets. The best",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 105,
+ 496,
+ 407,
+ 508
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 496,
+ 407,
+ 508
+ ],
+ "score": 1.0,
+ "content": "results are highlighted in bold while the second best results are underlined.",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ }
+ ],
+ "index": 22.5
+ },
+ {
+ "type": "table_body",
+ "bbox": [
+ 117,
+ 523,
+ 494,
+ 624
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 117,
+ 523,
+ 494,
+ 624
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 117,
+ 523,
+ 494,
+ 624
+ ],
+ "score": 0.98,
+ "html": "| Method | Reddit-12K | ENZYMES |
| 5-shot | 10-shot | 20-shot | 5-shot | 10-shot | 20-shot |
| WL | 40.26± 5.17 | 42.57±3.69 | 44.41±3.43 | 55.78± 4.72 | 58.47±3.84 | 60.1±3.18 |
| Graphlet | 33.76 ± 6.94 | 37.59 ± 4.60 | 41.11 ± 3.71 | 53.17 ± 5.92 | 55.30 ±3.78 | 56.90 ±3.79 |
| AWE | 30.24± 2.34 | 33.44 ±2.04 | 36.13 ± 1.89 | 43.75 ±1.85 | 45.58 ± 2.11 | 49.98 ± 1.54 |
| Graph2Vec | 27.85 ± 4.21 | 29.97 ± 3.17 | 32.75 ± 2.02 | 55.88 ± 4.86 | 58.22 ± 4.30 | 62.28 ± 4.14 |
| Diffpool | 35.24 ± 5.69 | 37.43 ± 3.94 | 39.11 ± 3.52 | 45.64 ± 4.56 | 49.64 ± 4.23 | 54.27 ± 3.94 |
| CapsGNN | 36.58 ±4.28 | 39.16 ± 3.73 | 41.27 ± 3.12 | 52.67 ± 5.51 | 55.31 ± 4.23 | 59.34 ± 4.02 |
| GIN | 40.36 ± 4.69 | 43.70 ± 3.98 | 46.28 ± 3.49 | 55.73 ± 5.80 | 58.83 ±5.32 | 61.12 ± 4.64 |
| GIN-k-NN | 41.31± 2.84 | 43.58±2.80 | 45.12 ± 2.19 | 57.24 ± 7.06 | 59.34± 5.24 | 60.49±3.48 |
| OurMethod-GCN | 40.77 ± 4.32 | 44.28 ± 3.86 | 48.67 ± 4.22 | 54.34 ± 5.64 | 58.16 ± 4.39 | 60.86 ± 3.74 |
| OurMethod-GAT | 41.59 ± 4.12 | 45.67± 3.68 | 50.34± 2.71 | 55.42 ± 5.74 | 60.64 ± 3.84 | 62.81 ± 3.56 |
",
+ "type": "table",
+ "image_path": "b955792ef88c002b65b8845aeab3812c807e8919a39ab39ae60f47bbce68009d.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 25,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 117,
+ 523,
+ 494,
+ 556.6666666666666
+ ],
+ "spans": [],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 117,
+ 556.6666666666666,
+ 494,
+ 590.3333333333333
+ ],
+ "spans": [],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 117,
+ 590.3333333333333,
+ 494,
+ 623.9999999999999
+ ],
+ "spans": [],
+ "index": 26
+ }
+ ]
+ }
+ ],
+ "index": 23.75
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 643,
+ 505,
+ 732
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 643,
+ 505,
+ 657
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 643,
+ 505,
+ 657
+ ],
+ "score": 1.0,
+ "content": "The results clearly show that our proposed method and its GCN variant (i.e., OurMethod-GCN)",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 106,
+ 654,
+ 505,
+ 666
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 654,
+ 231,
+ 666
+ ],
+ "score": 1.0,
+ "content": "outperform the baselines. GIN-",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 232,
+ 655,
+ 238,
+ 665
+ ],
+ "score": 0.76,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 239,
+ 654,
+ 505,
+ 666
+ ],
+ "score": 1.0,
+ "content": "-NN shows significant degradation in results on nearly all the three",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 105,
+ 666,
+ 505,
+ 678
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 666,
+ 505,
+ 678
+ ],
+ "score": 1.0,
+ "content": "paradigms for all datasets with exceptions on 5-shot and 10-shot on ENZYMES dataset, thus strongly",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 106,
+ 677,
+ 505,
+ 689
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 677,
+ 505,
+ 689
+ ],
+ "score": 1.0,
+ "content": "indicating that the improvements of our method can primarily be attributed to our GNN classifier",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 105,
+ 686,
+ 506,
+ 701
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 686,
+ 506,
+ 701
+ ],
+ "score": 1.0,
+ "content": "fed with the super-graph constructed from our proposed method. The improvements in results are",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ },
+ {
+ "bbox": [
+ 105,
+ 699,
+ 505,
+ 711
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 699,
+ 505,
+ 711
+ ],
+ "score": 1.0,
+ "content": "higher on TRIANGLES and Reddit datasets in contrast to Letter-High and ENZYMES, which can be",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 106,
+ 710,
+ 505,
+ 722
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 710,
+ 505,
+ 722
+ ],
+ "score": 1.0,
+ "content": "attributed to the smaller size of the graphs in Letter-High making it difficult to distinguish based on",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ },
+ {
+ "bbox": [
+ 105,
+ 721,
+ 505,
+ 733
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 721,
+ 505,
+ 733
+ ],
+ "score": 1.0,
+ "content": "graph spectra alone, whereas the complex and highly inter-related structure of enzymes makes it dif-",
+ "type": "text"
+ }
+ ],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 106,
+ 247,
+ 505,
+ 259
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 247,
+ 505,
+ 259
+ ],
+ "score": 1.0,
+ "content": "ficult for the DL based feature extractors as well as the graph kernel methods to segregate the classes",
+ "type": "text",
+ "cross_page": true
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 105,
+ 258,
+ 506,
+ 270
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 258,
+ 506,
+ 270
+ ],
+ "score": 1.0,
+ "content": "in the feature and graph space respectively. GIN and WL show much better results as compared to",
+ "type": "text",
+ "cross_page": true
+ }
+ ],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 105,
+ 268,
+ 505,
+ 282
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 268,
+ 212,
+ 282
+ ],
+ "score": 1.0,
+ "content": "other baselines for all the",
+ "type": "text",
+ "cross_page": true
+ },
+ {
+ "bbox": [
+ 213,
+ 271,
+ 219,
+ 281
+ ],
+ "score": 0.8,
+ "content": "q",
+ "type": "inline_equation",
+ "cross_page": true
+ },
+ {
+ "bbox": [
+ 219,
+ 268,
+ 505,
+ 282
+ ],
+ "score": 1.0,
+ "content": "-shot scenarios, whereas AWE and Graphlet Kernel show significantly",
+ "type": "text",
+ "cross_page": true
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 105,
+ 280,
+ 505,
+ 292
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 280,
+ 505,
+ 292
+ ],
+ "score": 1.0,
+ "content": "low results, unable to capture the properties of the graphs well. The DL baselines apart from GIN",
+ "type": "text",
+ "cross_page": true
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 106,
+ 291,
+ 505,
+ 303
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 291,
+ 505,
+ 303
+ ],
+ "score": 1.0,
+ "content": "on the other hand show improvements on the TRIANGLES dataset performing close to GIN, where",
+ "type": "text",
+ "cross_page": true
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 105,
+ 301,
+ 505,
+ 315
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 301,
+ 505,
+ 315
+ ],
+ "score": 1.0,
+ "content": "the unsupervised methods fails to capture the local node properties, however still perform poorly on",
+ "type": "text",
+ "cross_page": true
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 105,
+ 312,
+ 506,
+ 326
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 312,
+ 506,
+ 326
+ ],
+ "score": 1.0,
+ "content": "other datasets. For the 20-shot scenario on TRIANGLES, our GAT variant shows an improvement",
+ "type": "text",
+ "cross_page": true
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 105,
+ 323,
+ 505,
+ 337
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 323,
+ 147,
+ 337
+ ],
+ "score": 1.0,
+ "content": "of around",
+ "type": "text",
+ "cross_page": true
+ },
+ {
+ "bbox": [
+ 147,
+ 324,
+ 162,
+ 334
+ ],
+ "score": 0.88,
+ "content": "7 \\%",
+ "type": "inline_equation",
+ "cross_page": true
+ },
+ {
+ "bbox": [
+ 162,
+ 323,
+ 296,
+ 337
+ ],
+ "score": 1.0,
+ "content": "over DL baselines and more than",
+ "type": "text",
+ "cross_page": true
+ },
+ {
+ "bbox": [
+ 297,
+ 324,
+ 316,
+ 334
+ ],
+ "score": 0.89,
+ "content": "2 0 \\%",
+ "type": "inline_equation",
+ "cross_page": true
+ },
+ {
+ "bbox": [
+ 316,
+ 323,
+ 505,
+ 337
+ ],
+ "score": 1.0,
+ "content": "when compared to unsupervised methods. The",
+ "type": "text",
+ "cross_page": true
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 105,
+ 334,
+ 505,
+ 347
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 334,
+ 254,
+ 347
+ ],
+ "score": 1.0,
+ "content": "substantial improvements of around",
+ "type": "text",
+ "cross_page": true
+ },
+ {
+ "bbox": [
+ 254,
+ 334,
+ 270,
+ 345
+ ],
+ "score": 0.86,
+ "content": "4 \\%",
+ "type": "inline_equation",
+ "cross_page": true
+ },
+ {
+ "bbox": [
+ 270,
+ 334,
+ 505,
+ 347
+ ],
+ "score": 1.0,
+ "content": "on Reddit dataset shows the superiority of our model for",
+ "type": "text",
+ "cross_page": true
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 105,
+ 345,
+ 506,
+ 358
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 345,
+ 315,
+ 358
+ ],
+ "score": 1.0,
+ "content": "both the variants - GAT and GCN. Furthermore, the",
+ "type": "text",
+ "cross_page": true
+ },
+ {
+ "bbox": [
+ 316,
+ 347,
+ 321,
+ 356
+ ],
+ "score": 0.66,
+ "content": "t",
+ "type": "inline_equation",
+ "cross_page": true
+ },
+ {
+ "bbox": [
+ 321,
+ 345,
+ 506,
+ 358
+ ],
+ "score": 1.0,
+ "content": "-SNE plots in Figure 3 show a substantial and",
+ "type": "text",
+ "cross_page": true
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 106,
+ 357,
+ 505,
+ 369
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 357,
+ 505,
+ 369
+ ],
+ "score": 1.0,
+ "content": "interesting separation of class labels which strongly indicate that a good feature extractor in con-",
+ "type": "text",
+ "cross_page": true
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 105,
+ 367,
+ 505,
+ 380
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 367,
+ 505,
+ 380
+ ],
+ "score": 1.0,
+ "content": "junction with a GNN perform well as a combination. The t-SNE plots for ENZYMES, Reddit, and",
+ "type": "text",
+ "cross_page": true
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 105,
+ 378,
+ 333,
+ 392
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 378,
+ 333,
+ 392
+ ],
+ "score": 1.0,
+ "content": "Letter-High are shown in figures 4, 5 and 6 respectively.",
+ "type": "text",
+ "cross_page": true
+ }
+ ],
+ "index": 18
+ }
+ ],
+ "index": 30.5,
+ "bbox_fs": [
+ 105,
+ 643,
+ 506,
+ 733
+ ]
+ }
+ ]
+ },
+ {
+ "preproc_blocks": [
+ {
+ "type": "image",
+ "bbox": [
+ 123,
+ 61,
+ 489,
+ 182
+ ],
+ "blocks": [
+ {
+ "type": "image_body",
+ "bbox": [
+ 123,
+ 61,
+ 489,
+ 182
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 123,
+ 61,
+ 489,
+ 182
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 123,
+ 61,
+ 489,
+ 182
+ ],
+ "score": 0.899,
+ "type": "image",
+ "image_path": "c52ff756e9f560ae3e4f1bc969a402fd9e4832578474fe3bc86a8e21779491da.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 1,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 123,
+ 61,
+ 489,
+ 101.33333333333334
+ ],
+ "spans": [],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 123,
+ 101.33333333333334,
+ 489,
+ 141.66666666666669
+ ],
+ "spans": [],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 123,
+ 141.66666666666669,
+ 489,
+ 182.00000000000003
+ ],
+ "spans": [],
+ "index": 2
+ }
+ ]
+ },
+ {
+ "type": "image_caption",
+ "bbox": [
+ 106,
+ 191,
+ 506,
+ 225
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 191,
+ 505,
+ 204
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 191,
+ 505,
+ 204
+ ],
+ "score": 1.0,
+ "content": "Figure 3: Visualization: t-SNE plots of the computed embeddings of test graphs on 20-shot scenario",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 106,
+ 202,
+ 505,
+ 214
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 202,
+ 505,
+ 214
+ ],
+ "score": 1.0,
+ "content": "from OurMethod-GAT (left), GIN (middle) and WL Kernel (right) on TRIANGLES dataset. The",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 106,
+ 213,
+ 495,
+ 226
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 213,
+ 495,
+ 226
+ ],
+ "score": 1.0,
+ "content": "embeddings for both our model and GIN are taken from the final layers of the respective models.",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ }
+ ],
+ "index": 4
+ }
+ ],
+ "index": 2.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 247,
+ 505,
+ 389
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 247,
+ 505,
+ 259
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 247,
+ 505,
+ 259
+ ],
+ "score": 1.0,
+ "content": "ficult for the DL based feature extractors as well as the graph kernel methods to segregate the classes",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 105,
+ 258,
+ 506,
+ 270
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 258,
+ 506,
+ 270
+ ],
+ "score": 1.0,
+ "content": "in the feature and graph space respectively. GIN and WL show much better results as compared to",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 105,
+ 268,
+ 505,
+ 282
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 268,
+ 212,
+ 282
+ ],
+ "score": 1.0,
+ "content": "other baselines for all the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 213,
+ 271,
+ 219,
+ 281
+ ],
+ "score": 0.8,
+ "content": "q",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 219,
+ 268,
+ 505,
+ 282
+ ],
+ "score": 1.0,
+ "content": "-shot scenarios, whereas AWE and Graphlet Kernel show significantly",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 105,
+ 280,
+ 505,
+ 292
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 280,
+ 505,
+ 292
+ ],
+ "score": 1.0,
+ "content": "low results, unable to capture the properties of the graphs well. The DL baselines apart from GIN",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 106,
+ 291,
+ 505,
+ 303
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 291,
+ 505,
+ 303
+ ],
+ "score": 1.0,
+ "content": "on the other hand show improvements on the TRIANGLES dataset performing close to GIN, where",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 105,
+ 301,
+ 505,
+ 315
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 301,
+ 505,
+ 315
+ ],
+ "score": 1.0,
+ "content": "the unsupervised methods fails to capture the local node properties, however still perform poorly on",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 105,
+ 312,
+ 506,
+ 326
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 312,
+ 506,
+ 326
+ ],
+ "score": 1.0,
+ "content": "other datasets. For the 20-shot scenario on TRIANGLES, our GAT variant shows an improvement",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 105,
+ 323,
+ 505,
+ 337
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 323,
+ 147,
+ 337
+ ],
+ "score": 1.0,
+ "content": "of around",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 147,
+ 324,
+ 162,
+ 334
+ ],
+ "score": 0.88,
+ "content": "7 \\%",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 162,
+ 323,
+ 296,
+ 337
+ ],
+ "score": 1.0,
+ "content": "over DL baselines and more than",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 297,
+ 324,
+ 316,
+ 334
+ ],
+ "score": 0.89,
+ "content": "2 0 \\%",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 316,
+ 323,
+ 505,
+ 337
+ ],
+ "score": 1.0,
+ "content": "when compared to unsupervised methods. The",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 105,
+ 334,
+ 505,
+ 347
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 334,
+ 254,
+ 347
+ ],
+ "score": 1.0,
+ "content": "substantial improvements of around",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 254,
+ 334,
+ 270,
+ 345
+ ],
+ "score": 0.86,
+ "content": "4 \\%",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 270,
+ 334,
+ 505,
+ 347
+ ],
+ "score": 1.0,
+ "content": "on Reddit dataset shows the superiority of our model for",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 105,
+ 345,
+ 506,
+ 358
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 345,
+ 315,
+ 358
+ ],
+ "score": 1.0,
+ "content": "both the variants - GAT and GCN. Furthermore, the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 316,
+ 347,
+ 321,
+ 356
+ ],
+ "score": 0.66,
+ "content": "t",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 321,
+ 345,
+ 506,
+ 358
+ ],
+ "score": 1.0,
+ "content": "-SNE plots in Figure 3 show a substantial and",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 106,
+ 357,
+ 505,
+ 369
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 357,
+ 505,
+ 369
+ ],
+ "score": 1.0,
+ "content": "interesting separation of class labels which strongly indicate that a good feature extractor in con-",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 105,
+ 367,
+ 505,
+ 380
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 367,
+ 505,
+ 380
+ ],
+ "score": 1.0,
+ "content": "junction with a GNN perform well as a combination. The t-SNE plots for ENZYMES, Reddit, and",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 105,
+ 378,
+ 333,
+ 392
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 378,
+ 333,
+ 392
+ ],
+ "score": 1.0,
+ "content": "Letter-High are shown in figures 4, 5 and 6 respectively.",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ }
+ ],
+ "index": 12
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 107,
+ 404,
+ 352,
+ 415
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 403,
+ 353,
+ 416
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 403,
+ 353,
+ 416
+ ],
+ "score": 1.0,
+ "content": "5.3 ABLATION STUDY ON NUMBER OF SUPER-CLASSES",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ }
+ ],
+ "index": 19
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 424,
+ 505,
+ 535
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 424,
+ 505,
+ 437
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 424,
+ 479,
+ 437
+ ],
+ "score": 1.0,
+ "content": "Here, we study the behavior of our proposed network model without the super-class classifier",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 479,
+ 425,
+ 501,
+ 435
+ ],
+ "score": 0.88,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 501,
+ 424,
+ 505,
+ 437
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 105,
+ 435,
+ 505,
+ 448
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 435,
+ 505,
+ 448
+ ],
+ "score": 1.0,
+ "content": "In Table 3 (10 and 20-shot setting), we observe a marked increase with the addition of our classifier",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 105,
+ 446,
+ 505,
+ 460
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 446,
+ 505,
+ 460
+ ],
+ "score": 1.0,
+ "content": "which uses the super-class information and the super-graph based on spectral measures to guide",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 106,
+ 455,
+ 506,
+ 473
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 457,
+ 134,
+ 468
+ ],
+ "score": 0.88,
+ "content": "C ^ { G A T }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 134,
+ 455,
+ 506,
+ 473
+ ],
+ "score": 1.0,
+ "content": "towards improving the class separation of the graphs during both the training and fine-tuning",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ },
+ {
+ "bbox": [
+ 104,
+ 468,
+ 506,
+ 482
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 468,
+ 506,
+ 482
+ ],
+ "score": 1.0,
+ "content": "stages. Using super-classes help in reducing the sample complexity of the large Hypothesis space",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 105,
+ 480,
+ 506,
+ 492
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 480,
+ 506,
+ 492
+ ],
+ "score": 1.0,
+ "content": "and makes tuning of the model parameters easier during fine-tuning stage with less samples and",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 105,
+ 491,
+ 505,
+ 502
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 491,
+ 505,
+ 502
+ ],
+ "score": 1.0,
+ "content": "few iterations. Negligible differences are observed on ENZYMES dataset since both the number of",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 104,
+ 500,
+ 505,
+ 516
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 500,
+ 505,
+ 516
+ ],
+ "score": 1.0,
+ "content": "training classes as well as test classes are low, thus, the model performs equally well on removing",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 105,
+ 513,
+ 505,
+ 525
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 513,
+ 505,
+ 525
+ ],
+ "score": 1.0,
+ "content": "super-classes. This is because of the latent inter-class representations can still be captured between",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 105,
+ 524,
+ 313,
+ 536
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 524,
+ 313,
+ 536
+ ],
+ "score": 1.0,
+ "content": "few classes especially during the fine-tuning phase.",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ }
+ ],
+ "index": 24.5
+ },
+ {
+ "type": "table",
+ "bbox": [
+ 137,
+ 588,
+ 474,
+ 659
+ ],
+ "blocks": [
+ {
+ "type": "table_caption",
+ "bbox": [
+ 107,
+ 546,
+ 504,
+ 570
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 545,
+ 505,
+ 559
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 545,
+ 358,
+ 559
+ ],
+ "score": 1.0,
+ "content": "Table 3: Ablation Study: “No-SC” represents our classifier",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 359,
+ 547,
+ 378,
+ 559
+ ],
+ "score": 0.88,
+ "content": "C ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 379,
+ 545,
+ 415,
+ 559
+ ],
+ "score": 1.0,
+ "content": "without",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 416,
+ 547,
+ 437,
+ 557
+ ],
+ "score": 0.84,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 438,
+ 545,
+ 505,
+ 559
+ ],
+ "score": 1.0,
+ "content": "and “With-SC”",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 105,
+ 556,
+ 313,
+ 571
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 556,
+ 280,
+ 571
+ ],
+ "score": 1.0,
+ "content": "represents C(.) with both Csup and CGAT",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 279,
+ 558,
+ 313,
+ 570
+ ],
+ "score": 1.0,
+ "content": "present.",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ }
+ ],
+ "index": 30.5
+ },
+ {
+ "type": "table_body",
+ "bbox": [
+ 137,
+ 588,
+ 474,
+ 659
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 137,
+ 588,
+ 474,
+ 659
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 137,
+ 588,
+ 474,
+ 659
+ ],
+ "score": 0.974,
+ "html": "| Dataset | 10-shot | 20-shot |
| No-SC | With-SC | No-SC | With-SC |
| Letter-High | 71.13 ± 3.64 | 73.61 ± 3.19 | 75.23 ± 2.48 | 77.42 ± 1.47 |
| TRIANGLES | 74.03 ± 3.89 | 76.49 ± 3.26 | 76.89 ± 2.63 | 80.14 ± 1.88 |
| Reddit-12K | 43.76 ± 4.34 | 45.35 ± 4.06 | 48.19 ± 4.01 | 50.36 ± 3.04 |
| ENZYMES | 59.97 ± 3.98 | 59.58 ± 4.32 | 62.7 ± 3.63 | 62.39 ± 3.48 |
",
+ "type": "table",
+ "image_path": "e052f38d2b04a5429a0c303b7baded7ce27984a6f914d9210609d5da78404829.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 33,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 137,
+ 588,
+ 474,
+ 611.6666666666666
+ ],
+ "spans": [],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 137,
+ 611.6666666666666,
+ 474,
+ 635.3333333333333
+ ],
+ "spans": [],
+ "index": 33
+ },
+ {
+ "bbox": [
+ 137,
+ 635.3333333333333,
+ 474,
+ 658.9999999999999
+ ],
+ "spans": [],
+ "index": 34
+ }
+ ]
+ }
+ ],
+ "index": 31.75
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 107,
+ 678,
+ 345,
+ 689
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 677,
+ 346,
+ 690
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 677,
+ 346,
+ 690
+ ],
+ "score": 1.0,
+ "content": "5.4 SENSITIVITY ANALYSIS OF VARIOUS ATTRIBUTES",
+ "type": "text"
+ }
+ ],
+ "index": 35
+ }
+ ],
+ "index": 35
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 699,
+ 505,
+ 732
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 698,
+ 506,
+ 712
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 698,
+ 506,
+ 712
+ ],
+ "score": 1.0,
+ "content": "Our proposed method contains two crucial attributes. We analyze our model by varying: (i) the",
+ "type": "text"
+ }
+ ],
+ "index": 36
+ },
+ {
+ "bbox": [
+ 106,
+ 709,
+ 505,
+ 721
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 709,
+ 250,
+ 721
+ ],
+ "score": 1.0,
+ "content": "number of super-classes and (ii) the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 250,
+ 710,
+ 257,
+ 720
+ ],
+ "score": 0.83,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 257,
+ 709,
+ 505,
+ 721
+ ],
+ "score": 1.0,
+ "content": "-value in super-graph construction. The effect of varying these",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 106,
+ 721,
+ 505,
+ 732
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 721,
+ 505,
+ 732
+ ],
+ "score": 1.0,
+ "content": "attributes on model accuracy are shown in Tables 4 and 5, respectively. For TRIANGLES and Letter-",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ }
+ ],
+ "index": 37
+ }
+ ],
+ "page_idx": 8,
+ "page_size": [
+ 612,
+ 792
+ ],
+ "discarded_blocks": [
+ {
+ "type": "discarded",
+ "bbox": [
+ 107,
+ 27,
+ 293,
+ 37
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "score": 1.0,
+ "content": "Published as a conference paper at ICLR 2020",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 302,
+ 751,
+ 308,
+ 759
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 302,
+ 751,
+ 309,
+ 762
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 302,
+ 751,
+ 309,
+ 762
+ ],
+ "score": 1.0,
+ "content": "9",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ }
+ ],
+ "para_blocks": [
+ {
+ "type": "image",
+ "bbox": [
+ 123,
+ 61,
+ 489,
+ 182
+ ],
+ "blocks": [
+ {
+ "type": "image_body",
+ "bbox": [
+ 123,
+ 61,
+ 489,
+ 182
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 123,
+ 61,
+ 489,
+ 182
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 123,
+ 61,
+ 489,
+ 182
+ ],
+ "score": 0.899,
+ "type": "image",
+ "image_path": "c52ff756e9f560ae3e4f1bc969a402fd9e4832578474fe3bc86a8e21779491da.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 1,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 123,
+ 61,
+ 489,
+ 101.33333333333334
+ ],
+ "spans": [],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 123,
+ 101.33333333333334,
+ 489,
+ 141.66666666666669
+ ],
+ "spans": [],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 123,
+ 141.66666666666669,
+ 489,
+ 182.00000000000003
+ ],
+ "spans": [],
+ "index": 2
+ }
+ ]
+ },
+ {
+ "type": "image_caption",
+ "bbox": [
+ 106,
+ 191,
+ 506,
+ 225
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 191,
+ 505,
+ 204
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 191,
+ 505,
+ 204
+ ],
+ "score": 1.0,
+ "content": "Figure 3: Visualization: t-SNE plots of the computed embeddings of test graphs on 20-shot scenario",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 106,
+ 202,
+ 505,
+ 214
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 202,
+ 505,
+ 214
+ ],
+ "score": 1.0,
+ "content": "from OurMethod-GAT (left), GIN (middle) and WL Kernel (right) on TRIANGLES dataset. The",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 106,
+ 213,
+ 495,
+ 226
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 213,
+ 495,
+ 226
+ ],
+ "score": 1.0,
+ "content": "embeddings for both our model and GIN are taken from the final layers of the respective models.",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ }
+ ],
+ "index": 4
+ }
+ ],
+ "index": 2.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 247,
+ 505,
+ 389
+ ],
+ "lines": [],
+ "index": 12,
+ "bbox_fs": [
+ 105,
+ 247,
+ 506,
+ 392
+ ],
+ "lines_deleted": true
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 107,
+ 404,
+ 352,
+ 415
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 403,
+ 353,
+ 416
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 403,
+ 353,
+ 416
+ ],
+ "score": 1.0,
+ "content": "5.3 ABLATION STUDY ON NUMBER OF SUPER-CLASSES",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ }
+ ],
+ "index": 19
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 424,
+ 505,
+ 535
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 424,
+ 505,
+ 437
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 424,
+ 479,
+ 437
+ ],
+ "score": 1.0,
+ "content": "Here, we study the behavior of our proposed network model without the super-class classifier",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 479,
+ 425,
+ 501,
+ 435
+ ],
+ "score": 0.88,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 501,
+ 424,
+ 505,
+ 437
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 105,
+ 435,
+ 505,
+ 448
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 435,
+ 505,
+ 448
+ ],
+ "score": 1.0,
+ "content": "In Table 3 (10 and 20-shot setting), we observe a marked increase with the addition of our classifier",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 105,
+ 446,
+ 505,
+ 460
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 446,
+ 505,
+ 460
+ ],
+ "score": 1.0,
+ "content": "which uses the super-class information and the super-graph based on spectral measures to guide",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 106,
+ 455,
+ 506,
+ 473
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 457,
+ 134,
+ 468
+ ],
+ "score": 0.88,
+ "content": "C ^ { G A T }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 134,
+ 455,
+ 506,
+ 473
+ ],
+ "score": 1.0,
+ "content": "towards improving the class separation of the graphs during both the training and fine-tuning",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ },
+ {
+ "bbox": [
+ 104,
+ 468,
+ 506,
+ 482
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 468,
+ 506,
+ 482
+ ],
+ "score": 1.0,
+ "content": "stages. Using super-classes help in reducing the sample complexity of the large Hypothesis space",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 105,
+ 480,
+ 506,
+ 492
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 480,
+ 506,
+ 492
+ ],
+ "score": 1.0,
+ "content": "and makes tuning of the model parameters easier during fine-tuning stage with less samples and",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 105,
+ 491,
+ 505,
+ 502
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 491,
+ 505,
+ 502
+ ],
+ "score": 1.0,
+ "content": "few iterations. Negligible differences are observed on ENZYMES dataset since both the number of",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 104,
+ 500,
+ 505,
+ 516
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 500,
+ 505,
+ 516
+ ],
+ "score": 1.0,
+ "content": "training classes as well as test classes are low, thus, the model performs equally well on removing",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 105,
+ 513,
+ 505,
+ 525
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 513,
+ 505,
+ 525
+ ],
+ "score": 1.0,
+ "content": "super-classes. This is because of the latent inter-class representations can still be captured between",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 105,
+ 524,
+ 313,
+ 536
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 524,
+ 313,
+ 536
+ ],
+ "score": 1.0,
+ "content": "few classes especially during the fine-tuning phase.",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ }
+ ],
+ "index": 24.5,
+ "bbox_fs": [
+ 104,
+ 424,
+ 506,
+ 536
+ ]
+ },
+ {
+ "type": "table",
+ "bbox": [
+ 137,
+ 588,
+ 474,
+ 659
+ ],
+ "blocks": [
+ {
+ "type": "table_caption",
+ "bbox": [
+ 107,
+ 546,
+ 504,
+ 570
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 545,
+ 505,
+ 559
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 545,
+ 358,
+ 559
+ ],
+ "score": 1.0,
+ "content": "Table 3: Ablation Study: “No-SC” represents our classifier",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 359,
+ 547,
+ 378,
+ 559
+ ],
+ "score": 0.88,
+ "content": "C ( . )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 379,
+ 545,
+ 415,
+ 559
+ ],
+ "score": 1.0,
+ "content": "without",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 416,
+ 547,
+ 437,
+ 557
+ ],
+ "score": 0.84,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 438,
+ 545,
+ 505,
+ 559
+ ],
+ "score": 1.0,
+ "content": "and “With-SC”",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 105,
+ 556,
+ 313,
+ 571
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 556,
+ 280,
+ 571
+ ],
+ "score": 1.0,
+ "content": "represents C(.) with both Csup and CGAT",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 279,
+ 558,
+ 313,
+ 570
+ ],
+ "score": 1.0,
+ "content": "present.",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ }
+ ],
+ "index": 30.5
+ },
+ {
+ "type": "table_body",
+ "bbox": [
+ 137,
+ 588,
+ 474,
+ 659
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 137,
+ 588,
+ 474,
+ 659
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 137,
+ 588,
+ 474,
+ 659
+ ],
+ "score": 0.974,
+ "html": "| Dataset | 10-shot | 20-shot |
| No-SC | With-SC | No-SC | With-SC |
| Letter-High | 71.13 ± 3.64 | 73.61 ± 3.19 | 75.23 ± 2.48 | 77.42 ± 1.47 |
| TRIANGLES | 74.03 ± 3.89 | 76.49 ± 3.26 | 76.89 ± 2.63 | 80.14 ± 1.88 |
| Reddit-12K | 43.76 ± 4.34 | 45.35 ± 4.06 | 48.19 ± 4.01 | 50.36 ± 3.04 |
| ENZYMES | 59.97 ± 3.98 | 59.58 ± 4.32 | 62.7 ± 3.63 | 62.39 ± 3.48 |
",
+ "type": "table",
+ "image_path": "e052f38d2b04a5429a0c303b7baded7ce27984a6f914d9210609d5da78404829.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 33,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 137,
+ 588,
+ 474,
+ 611.6666666666666
+ ],
+ "spans": [],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 137,
+ 611.6666666666666,
+ 474,
+ 635.3333333333333
+ ],
+ "spans": [],
+ "index": 33
+ },
+ {
+ "bbox": [
+ 137,
+ 635.3333333333333,
+ 474,
+ 658.9999999999999
+ ],
+ "spans": [],
+ "index": 34
+ }
+ ]
+ }
+ ],
+ "index": 31.75
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 107,
+ 678,
+ 345,
+ 689
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 677,
+ 346,
+ 690
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 677,
+ 346,
+ 690
+ ],
+ "score": 1.0,
+ "content": "5.4 SENSITIVITY ANALYSIS OF VARIOUS ATTRIBUTES",
+ "type": "text"
+ }
+ ],
+ "index": 35
+ }
+ ],
+ "index": 35
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 699,
+ 505,
+ 732
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 698,
+ 506,
+ 712
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 698,
+ 506,
+ 712
+ ],
+ "score": 1.0,
+ "content": "Our proposed method contains two crucial attributes. We analyze our model by varying: (i) the",
+ "type": "text"
+ }
+ ],
+ "index": 36
+ },
+ {
+ "bbox": [
+ 106,
+ 709,
+ 505,
+ 721
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 709,
+ 250,
+ 721
+ ],
+ "score": 1.0,
+ "content": "number of super-classes and (ii) the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 250,
+ 710,
+ 257,
+ 720
+ ],
+ "score": 0.83,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 257,
+ 709,
+ 505,
+ 721
+ ],
+ "score": 1.0,
+ "content": "-value in super-graph construction. The effect of varying these",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 106,
+ 721,
+ 505,
+ 732
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 721,
+ 505,
+ 732
+ ],
+ "score": 1.0,
+ "content": "attributes on model accuracy are shown in Tables 4 and 5, respectively. For TRIANGLES and Letter-",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ }
+ ],
+ "index": 37,
+ "bbox_fs": [
+ 105,
+ 698,
+ 506,
+ 732
+ ]
+ }
+ ]
+ },
+ {
+ "preproc_blocks": [
+ {
+ "type": "table",
+ "bbox": [
+ 112,
+ 145,
+ 501,
+ 203
+ ],
+ "blocks": [
+ {
+ "type": "table_caption",
+ "bbox": [
+ 106,
+ 80,
+ 505,
+ 113
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 79,
+ 505,
+ 92
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 79,
+ 505,
+ 92
+ ],
+ "score": 1.0,
+ "content": "Table 4: Model analysis over number of super-classes in 20-shot scenario. There is no evaluation for",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 105,
+ 91,
+ 506,
+ 103
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 91,
+ 506,
+ 103
+ ],
+ "score": 1.0,
+ "content": "5 super-classes on ENZYMES since the number of training classes is 4. Default value of parameter",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 106,
+ 102,
+ 165,
+ 113
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 103,
+ 113,
+ 112
+ ],
+ "score": 0.7,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 114,
+ 102,
+ 165,
+ 113
+ ],
+ "score": 1.0,
+ "content": "is fixed at 2.",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ }
+ ],
+ "index": 1
+ },
+ {
+ "type": "table_caption",
+ "bbox": [
+ 117,
+ 134,
+ 151,
+ 145
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 117,
+ 133,
+ 153,
+ 146
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 117,
+ 133,
+ 153,
+ 146
+ ],
+ "score": 1.0,
+ "content": "Dataset",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ }
+ ],
+ "index": 3
+ },
+ {
+ "type": "table_caption",
+ "bbox": [
+ 321,
+ 134,
+ 353,
+ 144
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 320,
+ 133,
+ 355,
+ 146
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 320,
+ 133,
+ 355,
+ 146
+ ],
+ "score": 1.0,
+ "content": "20-shot",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ }
+ ],
+ "index": 4
+ },
+ {
+ "type": "table_body",
+ "bbox": [
+ 112,
+ 145,
+ 501,
+ 203
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 112,
+ 145,
+ 501,
+ 203
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 112,
+ 145,
+ 501,
+ 203
+ ],
+ "score": 0.932,
+ "html": " | 1 | 2 | 3 | 4 | 5 |
| Letter-High | 74.43± 2.61 | 76.61 ± 1.67 | 77.51 ± 1.49 | 76.31 ± 1.98 | 75.05 ± 2.29 |
| TRIANGLES | 76.43 ± 2.87 | 79.55 ± 1.91 | 80.51 ± 1.72 | 78.91 ± 2.09 | 78.25 ± 2.40 |
| Reddit-12K | 48.32 ± 4.09 | 50.67 ± 2.94 | 50.10 ± 3.02 | 49.52 ± 4.02 | 48.33 ± 4.08 |
| ENZYMES | 62.34 ± 4.11 | 62.13 ± 4.01 | 60.16 ± 3.81 | 59.34 ± 3.98 | |
",
+ "type": "table",
+ "image_path": "9297c3837aad9fbb9b836f8f7340cdcabaf13a6f26dc6e5a9682212b53b232ae.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 6,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 112,
+ 145,
+ 501,
+ 164.33333333333334
+ ],
+ "spans": [],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 112,
+ 164.33333333333334,
+ 501,
+ 183.66666666666669
+ ],
+ "spans": [],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 112,
+ 183.66666666666669,
+ 501,
+ 203.00000000000003
+ ],
+ "spans": [],
+ "index": 7
+ }
+ ]
+ },
+ {
+ "type": "table_caption",
+ "bbox": [
+ 106,
+ 223,
+ 504,
+ 246
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 223,
+ 505,
+ 236
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 223,
+ 312,
+ 236
+ ],
+ "score": 1.0,
+ "content": "Table 5: Model analysis over number of neighbors",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 312,
+ 224,
+ 325,
+ 235
+ ],
+ "score": 0.61,
+ "content": "( k )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 325,
+ 223,
+ 505,
+ 236
+ ],
+ "score": 1.0,
+ "content": "in super-graph for 20-shot scenario. Default",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 106,
+ 234,
+ 308,
+ 246
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 234,
+ 308,
+ 246
+ ],
+ "score": 1.0,
+ "content": "value for the number of super-classes is fixed at 3.",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ }
+ ],
+ "index": 8.5
+ }
+ ],
+ "index": 4
+ },
+ {
+ "type": "table",
+ "bbox": [
+ 112,
+ 277,
+ 500,
+ 335
+ ],
+ "blocks": [
+ {
+ "type": "table_caption",
+ "bbox": [
+ 117,
+ 267,
+ 151,
+ 277
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 116,
+ 265,
+ 153,
+ 279
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 265,
+ 153,
+ 279
+ ],
+ "score": 1.0,
+ "content": "Dataset",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ }
+ ],
+ "index": 10
+ },
+ {
+ "type": "table_caption",
+ "bbox": [
+ 321,
+ 267,
+ 354,
+ 277
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 320,
+ 266,
+ 355,
+ 279
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 320,
+ 266,
+ 355,
+ 279
+ ],
+ "score": 1.0,
+ "content": "20-shot",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ }
+ ],
+ "index": 11
+ },
+ {
+ "type": "table_body",
+ "bbox": [
+ 112,
+ 277,
+ 500,
+ 335
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 112,
+ 277,
+ 500,
+ 335
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 112,
+ 277,
+ 500,
+ 335
+ ],
+ "score": 0.961,
+ "html": " | 2 | | 4 | 6 | 8 | Heuristic |
| Letter-High | 77.33 ± 1.71 | | 76.61 ± 1.67 | 75.63 ± 2.49 | 74.66 ± 2.61 | 74.35 ± 2.48 |
| TRIANGLES | 80.77 ± 1.57 | | 79.85 ± 1.59 | 79.45 ± 1.97 | 78.93 ± 2.04 | 79.42 ± 3.16 |
| Reddit-12K | 50.48 ± 3.02 | | 46.37 ± 3.03 | 44.12 ± 2.98 | 43.88 ± 3.24 | 44.82 ± 2.83 |
| ENZYMES | 62.34 ± 4.11 | | 61.42 ± 4.42 | 60.23 ± 5.10 | 59.67 ± 4.77 | 61.07 ± 4.68 |
",
+ "type": "table",
+ "image_path": "239ed734fe3bd68fda8888be536259aa30b7bbc47460f944ed1eafdd58b397e4.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 13,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 112,
+ 277,
+ 500,
+ 296.3333333333333
+ ],
+ "spans": [],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 112,
+ 296.3333333333333,
+ 500,
+ 315.66666666666663
+ ],
+ "spans": [],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 112,
+ 315.66666666666663,
+ 500,
+ 334.99999999999994
+ ],
+ "spans": [],
+ "index": 14
+ }
+ ]
+ }
+ ],
+ "index": 11
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 366,
+ 505,
+ 465
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 365,
+ 505,
+ 380
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 365,
+ 505,
+ 380
+ ],
+ "score": 1.0,
+ "content": "High datasets, as we increase the number of super-classes, we observe the accuracy improving",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 106,
+ 378,
+ 505,
+ 389
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 378,
+ 505,
+ 389
+ ],
+ "score": 1.0,
+ "content": "steadily up to 3 super-classes and then dropping from there onwards. For super-classes less than",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 105,
+ 389,
+ 505,
+ 400
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 389,
+ 198,
+ 400
+ ],
+ "score": 1.0,
+ "content": "3, we observe that the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 199,
+ 389,
+ 205,
+ 398
+ ],
+ "score": 0.77,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 206,
+ 389,
+ 505,
+ 400
+ ],
+ "score": 1.0,
+ "content": "-NN graph does not respect the class boundaries that are already imposed",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 105,
+ 399,
+ 506,
+ 412
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 399,
+ 506,
+ 412
+ ],
+ "score": 1.0,
+ "content": "by the graph spectral measures, thus connecting more arbitrary classes. For Reddit we observe",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 105,
+ 410,
+ 506,
+ 423
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 410,
+ 506,
+ 423
+ ],
+ "score": 1.0,
+ "content": "the performance is slightly better on using 2 super-classes and for ENZYMES similar performances",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 105,
+ 420,
+ 505,
+ 434
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 420,
+ 505,
+ 434
+ ],
+ "score": 1.0,
+ "content": "are observed for both 1 and 2 super-classes as described in the ablation study. On the other hand,",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 105,
+ 432,
+ 505,
+ 445
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 432,
+ 505,
+ 445
+ ],
+ "score": 1.0,
+ "content": "increasing the number of super-classes past 3, makes each super-class cluster very sparse with few",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 105,
+ 442,
+ 506,
+ 456
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 442,
+ 506,
+ 456
+ ],
+ "score": 1.0,
+ "content": "graph classes within, leading to an underflow of information between the graph classes. The same",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 106,
+ 454,
+ 242,
+ 466
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 454,
+ 242,
+ 466
+ ],
+ "score": 1.0,
+ "content": "effect is observed for all datasets.",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ }
+ ],
+ "index": 19
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 470,
+ 505,
+ 570
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 470,
+ 505,
+ 483
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 470,
+ 124,
+ 483
+ ],
+ "score": 1.0,
+ "content": "The",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 125,
+ 471,
+ 132,
+ 481
+ ],
+ "score": 0.82,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 132,
+ 470,
+ 505,
+ 483
+ ],
+ "score": 1.0,
+ "content": "-value or the number of neighbors of each node belonging to the same connected component",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 105,
+ 481,
+ 505,
+ 495
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 481,
+ 505,
+ 495
+ ],
+ "score": 1.0,
+ "content": "in the super-graph (i.e., belonging to the same super-class) is another salient parameter upon which",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 106,
+ 493,
+ 506,
+ 506
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 493,
+ 506,
+ 506
+ ],
+ "score": 1.0,
+ "content": "hinges the information flow (via message passing) between the graphs of the same super-class. We",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 106,
+ 504,
+ 505,
+ 516
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 504,
+ 208,
+ 516
+ ],
+ "score": 1.0,
+ "content": "analyze our model with",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 208,
+ 504,
+ 216,
+ 514
+ ],
+ "score": 0.79,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 216,
+ 504,
+ 290,
+ 516
+ ],
+ "score": 1.0,
+ "content": "values in the set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 290,
+ 504,
+ 334,
+ 516
+ ],
+ "score": 0.93,
+ "content": "\\{ 2 , 4 , 6 , 8 \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 334,
+ 504,
+ 505,
+ 516
+ ],
+ "score": 1.0,
+ "content": "and a commonly used heuristic method,",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 106,
+ 515,
+ 505,
+ 529
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 516,
+ 250,
+ 529
+ ],
+ "score": 1.0,
+ "content": "whereby each graph is connected to",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 250,
+ 515,
+ 268,
+ 527
+ ],
+ "score": 0.94,
+ "content": "\\sqrt { b _ { s } }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 268,
+ 516,
+ 505,
+ 529
+ ],
+ "score": 1.0,
+ "content": "nearest neighboring graphs based on the Euclidean similar-",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 106,
+ 527,
+ 505,
+ 540
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 527,
+ 272,
+ 540
+ ],
+ "score": 1.0,
+ "content": "ity of their feature representations, where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 272,
+ 527,
+ 282,
+ 538
+ ],
+ "score": 0.88,
+ "content": "b _ { s }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 282,
+ 527,
+ 505,
+ 540
+ ],
+ "score": 1.0,
+ "content": "is the number of samples in the mini-batch correspond-",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 105,
+ 538,
+ 505,
+ 551
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 538,
+ 189,
+ 551
+ ],
+ "score": 1.0,
+ "content": "ing to super-classes",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 190,
+ 540,
+ 196,
+ 548
+ ],
+ "score": 0.5,
+ "content": "s",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 196,
+ 538,
+ 505,
+ 551
+ ],
+ "score": 1.0,
+ "content": ". We achieve best results with 2-NN graphs per super-class and increasing",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 106,
+ 548,
+ 505,
+ 562
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 549,
+ 113,
+ 559
+ ],
+ "score": 0.78,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 114,
+ 548,
+ 505,
+ 562
+ ],
+ "score": 1.0,
+ "content": "beyond it leads to denser graphs with unnecessary connections between classes belonging to the",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ },
+ {
+ "bbox": [
+ 106,
+ 560,
+ 178,
+ 572
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 560,
+ 178,
+ 572
+ ],
+ "score": 1.0,
+ "content": "same super-class.",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ }
+ ],
+ "index": 28
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 107,
+ 600,
+ 195,
+ 613
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 599,
+ 197,
+ 616
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 599,
+ 197,
+ 616
+ ],
+ "score": 1.0,
+ "content": "6 CONCLUSION",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ }
+ ],
+ "index": 33
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 632,
+ 505,
+ 732
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 633,
+ 506,
+ 646
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 633,
+ 506,
+ 646
+ ],
+ "score": 1.0,
+ "content": "In this paper, we investigated the problem of few-shot learning on graphs for the graph classification",
+ "type": "text"
+ }
+ ],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 106,
+ 644,
+ 505,
+ 656
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 644,
+ 505,
+ 656
+ ],
+ "score": 1.0,
+ "content": "task. We explicitly created a super-graph on the base-labeled graphs and then grouped / clustered",
+ "type": "text"
+ }
+ ],
+ "index": 35
+ },
+ {
+ "bbox": [
+ 106,
+ 655,
+ 505,
+ 667
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 655,
+ 505,
+ 667
+ ],
+ "score": 1.0,
+ "content": "their associated class labels into super-classes, based on the graph spectral measures attributed to",
+ "type": "text"
+ }
+ ],
+ "index": 36
+ },
+ {
+ "bbox": [
+ 106,
+ 665,
+ 505,
+ 678
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 665,
+ 187,
+ 678
+ ],
+ "score": 1.0,
+ "content": "each graph and the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 187,
+ 666,
+ 199,
+ 676
+ ],
+ "score": 0.85,
+ "content": "L ^ { p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 199,
+ 665,
+ 505,
+ 678
+ ],
+ "score": 1.0,
+ "content": "-Wasserstein distances between them. We found that training our GNN on",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 106,
+ 677,
+ 505,
+ 689
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 677,
+ 505,
+ 689
+ ],
+ "score": 1.0,
+ "content": "the super-graph along with the auxiliary super-classes resulted in a marked improvement over state-",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ },
+ {
+ "bbox": [
+ 106,
+ 687,
+ 505,
+ 701
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 687,
+ 505,
+ 701
+ ],
+ "score": 1.0,
+ "content": "of-the-art GNNs. A promising future work is to propose new GNN models that break away from",
+ "type": "text"
+ }
+ ],
+ "index": 39
+ },
+ {
+ "bbox": [
+ 105,
+ 699,
+ 505,
+ 712
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 699,
+ 505,
+ 712
+ ],
+ "score": 1.0,
+ "content": "current neighborhood aggregation schemes to specifically overcome the obstacle posed by few-shot",
+ "type": "text"
+ }
+ ],
+ "index": 40
+ },
+ {
+ "bbox": [
+ 105,
+ 709,
+ 506,
+ 723
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 709,
+ 506,
+ 723
+ ],
+ "score": 1.0,
+ "content": "learning on graphs. Our source-code and dataset splits have been made public in an attempt to attract",
+ "type": "text"
+ }
+ ],
+ "index": 41
+ },
+ {
+ "bbox": [
+ 105,
+ 721,
+ 349,
+ 734
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 721,
+ 349,
+ 734
+ ],
+ "score": 1.0,
+ "content": "more attention to the context of few-shot learning on graphs.",
+ "type": "text"
+ }
+ ],
+ "index": 42
+ }
+ ],
+ "index": 38
+ }
+ ],
+ "page_idx": 9,
+ "page_size": [
+ 612,
+ 792
+ ],
+ "discarded_blocks": [
+ {
+ "type": "discarded",
+ "bbox": [
+ 108,
+ 27,
+ 293,
+ 37
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "score": 1.0,
+ "content": "Published as a conference paper at ICLR 2020",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 301,
+ 751,
+ 311,
+ 760
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 299,
+ 750,
+ 313,
+ 765
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 299,
+ 750,
+ 313,
+ 765
+ ],
+ "score": 1.0,
+ "content": "",
+ "type": "text",
+ "height": 15,
+ "width": 14
+ }
+ ]
+ }
+ ]
+ }
+ ],
+ "para_blocks": [
+ {
+ "type": "table",
+ "bbox": [
+ 112,
+ 145,
+ 501,
+ 203
+ ],
+ "blocks": [
+ {
+ "type": "table_caption",
+ "bbox": [
+ 106,
+ 80,
+ 505,
+ 113
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 79,
+ 505,
+ 92
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 79,
+ 505,
+ 92
+ ],
+ "score": 1.0,
+ "content": "Table 4: Model analysis over number of super-classes in 20-shot scenario. There is no evaluation for",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 105,
+ 91,
+ 506,
+ 103
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 91,
+ 506,
+ 103
+ ],
+ "score": 1.0,
+ "content": "5 super-classes on ENZYMES since the number of training classes is 4. Default value of parameter",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 106,
+ 102,
+ 165,
+ 113
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 103,
+ 113,
+ 112
+ ],
+ "score": 0.7,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 114,
+ 102,
+ 165,
+ 113
+ ],
+ "score": 1.0,
+ "content": "is fixed at 2.",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ }
+ ],
+ "index": 1
+ },
+ {
+ "type": "table_caption",
+ "bbox": [
+ 117,
+ 134,
+ 151,
+ 145
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 117,
+ 133,
+ 153,
+ 146
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 117,
+ 133,
+ 153,
+ 146
+ ],
+ "score": 1.0,
+ "content": "Dataset",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ }
+ ],
+ "index": 3
+ },
+ {
+ "type": "table_caption",
+ "bbox": [
+ 321,
+ 134,
+ 353,
+ 144
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 320,
+ 133,
+ 355,
+ 146
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 320,
+ 133,
+ 355,
+ 146
+ ],
+ "score": 1.0,
+ "content": "20-shot",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ }
+ ],
+ "index": 4
+ },
+ {
+ "type": "table_body",
+ "bbox": [
+ 112,
+ 145,
+ 501,
+ 203
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 112,
+ 145,
+ 501,
+ 203
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 112,
+ 145,
+ 501,
+ 203
+ ],
+ "score": 0.932,
+ "html": " | 1 | 2 | 3 | 4 | 5 |
| Letter-High | 74.43± 2.61 | 76.61 ± 1.67 | 77.51 ± 1.49 | 76.31 ± 1.98 | 75.05 ± 2.29 |
| TRIANGLES | 76.43 ± 2.87 | 79.55 ± 1.91 | 80.51 ± 1.72 | 78.91 ± 2.09 | 78.25 ± 2.40 |
| Reddit-12K | 48.32 ± 4.09 | 50.67 ± 2.94 | 50.10 ± 3.02 | 49.52 ± 4.02 | 48.33 ± 4.08 |
| ENZYMES | 62.34 ± 4.11 | 62.13 ± 4.01 | 60.16 ± 3.81 | 59.34 ± 3.98 | |
",
+ "type": "table",
+ "image_path": "9297c3837aad9fbb9b836f8f7340cdcabaf13a6f26dc6e5a9682212b53b232ae.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 6,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 112,
+ 145,
+ 501,
+ 164.33333333333334
+ ],
+ "spans": [],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 112,
+ 164.33333333333334,
+ 501,
+ 183.66666666666669
+ ],
+ "spans": [],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 112,
+ 183.66666666666669,
+ 501,
+ 203.00000000000003
+ ],
+ "spans": [],
+ "index": 7
+ }
+ ]
+ },
+ {
+ "type": "table_caption",
+ "bbox": [
+ 106,
+ 223,
+ 504,
+ 246
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 223,
+ 505,
+ 236
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 223,
+ 312,
+ 236
+ ],
+ "score": 1.0,
+ "content": "Table 5: Model analysis over number of neighbors",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 312,
+ 224,
+ 325,
+ 235
+ ],
+ "score": 0.61,
+ "content": "( k )",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 325,
+ 223,
+ 505,
+ 236
+ ],
+ "score": 1.0,
+ "content": "in super-graph for 20-shot scenario. Default",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 106,
+ 234,
+ 308,
+ 246
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 234,
+ 308,
+ 246
+ ],
+ "score": 1.0,
+ "content": "value for the number of super-classes is fixed at 3.",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ }
+ ],
+ "index": 8.5
+ }
+ ],
+ "index": 4
+ },
+ {
+ "type": "table",
+ "bbox": [
+ 112,
+ 277,
+ 500,
+ 335
+ ],
+ "blocks": [
+ {
+ "type": "table_caption",
+ "bbox": [
+ 117,
+ 267,
+ 151,
+ 277
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 116,
+ 265,
+ 153,
+ 279
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 265,
+ 153,
+ 279
+ ],
+ "score": 1.0,
+ "content": "Dataset",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ }
+ ],
+ "index": 10
+ },
+ {
+ "type": "table_caption",
+ "bbox": [
+ 321,
+ 267,
+ 354,
+ 277
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 320,
+ 266,
+ 355,
+ 279
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 320,
+ 266,
+ 355,
+ 279
+ ],
+ "score": 1.0,
+ "content": "20-shot",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ }
+ ],
+ "index": 11
+ },
+ {
+ "type": "table_body",
+ "bbox": [
+ 112,
+ 277,
+ 500,
+ 335
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 112,
+ 277,
+ 500,
+ 335
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 112,
+ 277,
+ 500,
+ 335
+ ],
+ "score": 0.961,
+ "html": " | 2 | | 4 | 6 | 8 | Heuristic |
| Letter-High | 77.33 ± 1.71 | | 76.61 ± 1.67 | 75.63 ± 2.49 | 74.66 ± 2.61 | 74.35 ± 2.48 |
| TRIANGLES | 80.77 ± 1.57 | | 79.85 ± 1.59 | 79.45 ± 1.97 | 78.93 ± 2.04 | 79.42 ± 3.16 |
| Reddit-12K | 50.48 ± 3.02 | | 46.37 ± 3.03 | 44.12 ± 2.98 | 43.88 ± 3.24 | 44.82 ± 2.83 |
| ENZYMES | 62.34 ± 4.11 | | 61.42 ± 4.42 | 60.23 ± 5.10 | 59.67 ± 4.77 | 61.07 ± 4.68 |
",
+ "type": "table",
+ "image_path": "239ed734fe3bd68fda8888be536259aa30b7bbc47460f944ed1eafdd58b397e4.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 13,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 112,
+ 277,
+ 500,
+ 296.3333333333333
+ ],
+ "spans": [],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 112,
+ 296.3333333333333,
+ 500,
+ 315.66666666666663
+ ],
+ "spans": [],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 112,
+ 315.66666666666663,
+ 500,
+ 334.99999999999994
+ ],
+ "spans": [],
+ "index": 14
+ }
+ ]
+ }
+ ],
+ "index": 11
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 366,
+ 505,
+ 465
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 365,
+ 505,
+ 380
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 365,
+ 505,
+ 380
+ ],
+ "score": 1.0,
+ "content": "High datasets, as we increase the number of super-classes, we observe the accuracy improving",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 106,
+ 378,
+ 505,
+ 389
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 378,
+ 505,
+ 389
+ ],
+ "score": 1.0,
+ "content": "steadily up to 3 super-classes and then dropping from there onwards. For super-classes less than",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 105,
+ 389,
+ 505,
+ 400
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 389,
+ 198,
+ 400
+ ],
+ "score": 1.0,
+ "content": "3, we observe that the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 199,
+ 389,
+ 205,
+ 398
+ ],
+ "score": 0.77,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 206,
+ 389,
+ 505,
+ 400
+ ],
+ "score": 1.0,
+ "content": "-NN graph does not respect the class boundaries that are already imposed",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 105,
+ 399,
+ 506,
+ 412
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 399,
+ 506,
+ 412
+ ],
+ "score": 1.0,
+ "content": "by the graph spectral measures, thus connecting more arbitrary classes. For Reddit we observe",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 105,
+ 410,
+ 506,
+ 423
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 410,
+ 506,
+ 423
+ ],
+ "score": 1.0,
+ "content": "the performance is slightly better on using 2 super-classes and for ENZYMES similar performances",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 105,
+ 420,
+ 505,
+ 434
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 420,
+ 505,
+ 434
+ ],
+ "score": 1.0,
+ "content": "are observed for both 1 and 2 super-classes as described in the ablation study. On the other hand,",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 105,
+ 432,
+ 505,
+ 445
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 432,
+ 505,
+ 445
+ ],
+ "score": 1.0,
+ "content": "increasing the number of super-classes past 3, makes each super-class cluster very sparse with few",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 105,
+ 442,
+ 506,
+ 456
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 442,
+ 506,
+ 456
+ ],
+ "score": 1.0,
+ "content": "graph classes within, leading to an underflow of information between the graph classes. The same",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 106,
+ 454,
+ 242,
+ 466
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 454,
+ 242,
+ 466
+ ],
+ "score": 1.0,
+ "content": "effect is observed for all datasets.",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ }
+ ],
+ "index": 19,
+ "bbox_fs": [
+ 105,
+ 365,
+ 506,
+ 466
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 470,
+ 505,
+ 570
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 470,
+ 505,
+ 483
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 470,
+ 124,
+ 483
+ ],
+ "score": 1.0,
+ "content": "The",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 125,
+ 471,
+ 132,
+ 481
+ ],
+ "score": 0.82,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 132,
+ 470,
+ 505,
+ 483
+ ],
+ "score": 1.0,
+ "content": "-value or the number of neighbors of each node belonging to the same connected component",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 105,
+ 481,
+ 505,
+ 495
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 481,
+ 505,
+ 495
+ ],
+ "score": 1.0,
+ "content": "in the super-graph (i.e., belonging to the same super-class) is another salient parameter upon which",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 106,
+ 493,
+ 506,
+ 506
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 493,
+ 506,
+ 506
+ ],
+ "score": 1.0,
+ "content": "hinges the information flow (via message passing) between the graphs of the same super-class. We",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 106,
+ 504,
+ 505,
+ 516
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 504,
+ 208,
+ 516
+ ],
+ "score": 1.0,
+ "content": "analyze our model with",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 208,
+ 504,
+ 216,
+ 514
+ ],
+ "score": 0.79,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 216,
+ 504,
+ 290,
+ 516
+ ],
+ "score": 1.0,
+ "content": "values in the set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 290,
+ 504,
+ 334,
+ 516
+ ],
+ "score": 0.93,
+ "content": "\\{ 2 , 4 , 6 , 8 \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 334,
+ 504,
+ 505,
+ 516
+ ],
+ "score": 1.0,
+ "content": "and a commonly used heuristic method,",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 106,
+ 515,
+ 505,
+ 529
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 516,
+ 250,
+ 529
+ ],
+ "score": 1.0,
+ "content": "whereby each graph is connected to",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 250,
+ 515,
+ 268,
+ 527
+ ],
+ "score": 0.94,
+ "content": "\\sqrt { b _ { s } }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 268,
+ 516,
+ 505,
+ 529
+ ],
+ "score": 1.0,
+ "content": "nearest neighboring graphs based on the Euclidean similar-",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 106,
+ 527,
+ 505,
+ 540
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 527,
+ 272,
+ 540
+ ],
+ "score": 1.0,
+ "content": "ity of their feature representations, where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 272,
+ 527,
+ 282,
+ 538
+ ],
+ "score": 0.88,
+ "content": "b _ { s }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 282,
+ 527,
+ 505,
+ 540
+ ],
+ "score": 1.0,
+ "content": "is the number of samples in the mini-batch correspond-",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 105,
+ 538,
+ 505,
+ 551
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 538,
+ 189,
+ 551
+ ],
+ "score": 1.0,
+ "content": "ing to super-classes",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 190,
+ 540,
+ 196,
+ 548
+ ],
+ "score": 0.5,
+ "content": "s",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 196,
+ 538,
+ 505,
+ 551
+ ],
+ "score": 1.0,
+ "content": ". We achieve best results with 2-NN graphs per super-class and increasing",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 106,
+ 548,
+ 505,
+ 562
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 549,
+ 113,
+ 559
+ ],
+ "score": 0.78,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 114,
+ 548,
+ 505,
+ 562
+ ],
+ "score": 1.0,
+ "content": "beyond it leads to denser graphs with unnecessary connections between classes belonging to the",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ },
+ {
+ "bbox": [
+ 106,
+ 560,
+ 178,
+ 572
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 560,
+ 178,
+ 572
+ ],
+ "score": 1.0,
+ "content": "same super-class.",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ }
+ ],
+ "index": 28,
+ "bbox_fs": [
+ 105,
+ 470,
+ 506,
+ 572
+ ]
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 107,
+ 600,
+ 195,
+ 613
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 599,
+ 197,
+ 616
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 599,
+ 197,
+ 616
+ ],
+ "score": 1.0,
+ "content": "6 CONCLUSION",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ }
+ ],
+ "index": 33
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 632,
+ 505,
+ 732
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 633,
+ 506,
+ 646
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 633,
+ 506,
+ 646
+ ],
+ "score": 1.0,
+ "content": "In this paper, we investigated the problem of few-shot learning on graphs for the graph classification",
+ "type": "text"
+ }
+ ],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 106,
+ 644,
+ 505,
+ 656
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 644,
+ 505,
+ 656
+ ],
+ "score": 1.0,
+ "content": "task. We explicitly created a super-graph on the base-labeled graphs and then grouped / clustered",
+ "type": "text"
+ }
+ ],
+ "index": 35
+ },
+ {
+ "bbox": [
+ 106,
+ 655,
+ 505,
+ 667
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 655,
+ 505,
+ 667
+ ],
+ "score": 1.0,
+ "content": "their associated class labels into super-classes, based on the graph spectral measures attributed to",
+ "type": "text"
+ }
+ ],
+ "index": 36
+ },
+ {
+ "bbox": [
+ 106,
+ 665,
+ 505,
+ 678
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 665,
+ 187,
+ 678
+ ],
+ "score": 1.0,
+ "content": "each graph and the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 187,
+ 666,
+ 199,
+ 676
+ ],
+ "score": 0.85,
+ "content": "L ^ { p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 199,
+ 665,
+ 505,
+ 678
+ ],
+ "score": 1.0,
+ "content": "-Wasserstein distances between them. We found that training our GNN on",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 106,
+ 677,
+ 505,
+ 689
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 677,
+ 505,
+ 689
+ ],
+ "score": 1.0,
+ "content": "the super-graph along with the auxiliary super-classes resulted in a marked improvement over state-",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ },
+ {
+ "bbox": [
+ 106,
+ 687,
+ 505,
+ 701
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 687,
+ 505,
+ 701
+ ],
+ "score": 1.0,
+ "content": "of-the-art GNNs. A promising future work is to propose new GNN models that break away from",
+ "type": "text"
+ }
+ ],
+ "index": 39
+ },
+ {
+ "bbox": [
+ 105,
+ 699,
+ 505,
+ 712
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 699,
+ 505,
+ 712
+ ],
+ "score": 1.0,
+ "content": "current neighborhood aggregation schemes to specifically overcome the obstacle posed by few-shot",
+ "type": "text"
+ }
+ ],
+ "index": 40
+ },
+ {
+ "bbox": [
+ 105,
+ 709,
+ 506,
+ 723
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 709,
+ 506,
+ 723
+ ],
+ "score": 1.0,
+ "content": "learning on graphs. Our source-code and dataset splits have been made public in an attempt to attract",
+ "type": "text"
+ }
+ ],
+ "index": 41
+ },
+ {
+ "bbox": [
+ 105,
+ 721,
+ 349,
+ 734
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 721,
+ 349,
+ 734
+ ],
+ "score": 1.0,
+ "content": "more attention to the context of few-shot learning on graphs.",
+ "type": "text"
+ }
+ ],
+ "index": 42
+ }
+ ],
+ "index": 38,
+ "bbox_fs": [
+ 105,
+ 633,
+ 506,
+ 734
+ ]
+ }
+ ]
+ },
+ {
+ "preproc_blocks": [
+ {
+ "type": "title",
+ "bbox": [
+ 107,
+ 81,
+ 176,
+ 94
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 82,
+ 176,
+ 94
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 82,
+ 176,
+ 94
+ ],
+ "score": 1.0,
+ "content": "REFERENCES",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ }
+ ],
+ "index": 0
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 100,
+ 504,
+ 122
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 98,
+ 505,
+ 113
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 98,
+ 505,
+ 113
+ ],
+ "score": 1.0,
+ "content": "Martial Agueh and Guillaume Carlier. Barycenters in the wasserstein space. SIAM J. Math. Analysis,",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 116,
+ 110,
+ 206,
+ 123
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 110,
+ 206,
+ 123
+ ],
+ "score": 1.0,
+ "content": "43(2):904–924, 2011.",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ }
+ ],
+ "index": 1.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 128,
+ 505,
+ 162
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 129,
+ 504,
+ 141
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 129,
+ 298,
+ 141
+ ],
+ "score": 1.0,
+ "content": "David Arthur and Sergei Vassilvitskii. K-means",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 298,
+ 130,
+ 309,
+ 139
+ ],
+ "score": 0.58,
+ "content": "^ { + + }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 310,
+ 129,
+ 504,
+ 141
+ ],
+ "score": 1.0,
+ "content": ": The advantages of careful seeding. In Proceed-",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 115,
+ 138,
+ 505,
+ 154
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 138,
+ 505,
+ 154
+ ],
+ "score": 1.0,
+ "content": "ings of the Eighteenth Annual ACM-SIAM Symposium on Discrete Algorithms, SODA ’07, pp.",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 117,
+ 151,
+ 192,
+ 162
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 117,
+ 151,
+ 192,
+ 162
+ ],
+ "score": 1.0,
+ "content": "1027–1035, 2007.",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ }
+ ],
+ "index": 4
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 169,
+ 504,
+ 202
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 169,
+ 504,
+ 181
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 169,
+ 504,
+ 181
+ ],
+ "score": 1.0,
+ "content": "Lars Backstrom and Jure Leskovec. Supervised random walks: Predicting and recommending links",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 115,
+ 180,
+ 503,
+ 192
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 180,
+ 503,
+ 192
+ ],
+ "score": 1.0,
+ "content": "in social networks. CoRR, abs/1011.4071, 2010. URL http://arxiv.org/abs/1011.",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 115,
+ 191,
+ 146,
+ 203
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 191,
+ 146,
+ 203
+ ],
+ "score": 1.0,
+ "content": "4071.",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ }
+ ],
+ "index": 7
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 209,
+ 505,
+ 287
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 209,
+ 505,
+ 222
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 209,
+ 505,
+ 222
+ ],
+ "score": 1.0,
+ "content": "Peter Battaglia, Jessica Blake Chandler Hamrick, Victor Bapst, Alvaro Sanchez, Vinicius Zam-",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 115,
+ 219,
+ 505,
+ 233
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 219,
+ 505,
+ 233
+ ],
+ "score": 1.0,
+ "content": "baldi, Mateusz Malinowski, Andrea Tacchetti, David Raposo, Adam Santoro, Ryan Faulkner,",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 115,
+ 231,
+ 505,
+ 244
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 231,
+ 505,
+ 244
+ ],
+ "score": 1.0,
+ "content": "Caglar Gulcehre, Francis Song, Andy Ballard, Justin Gilmer, George E. Dahl, Ashish Vaswani,",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 116,
+ 242,
+ 505,
+ 255
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 242,
+ 505,
+ 255
+ ],
+ "score": 1.0,
+ "content": "Kelsey Allen, Charles Nash, Victoria Jayne Langston, Chris Dyer, Nicolas Heess, Daan Wier-",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 115,
+ 253,
+ 505,
+ 265
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 253,
+ 505,
+ 265
+ ],
+ "score": 1.0,
+ "content": "stra, Pushmeet Kohli, Matt Botvinick, Oriol Vinyals, Yujia Li, and Razvan Pascanu. Re-",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 115,
+ 263,
+ 504,
+ 277
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 263,
+ 504,
+ 277
+ ],
+ "score": 1.0,
+ "content": "lational inductive biases, deep learning, and graph networks. arXiv, 2018. URL https:",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 117,
+ 276,
+ 301,
+ 287
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 117,
+ 276,
+ 301,
+ 287
+ ],
+ "score": 1.0,
+ "content": "//arxiv.org/pdf/1806.01261.pdf.",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ }
+ ],
+ "index": 12
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 293,
+ 505,
+ 338
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 293,
+ 506,
+ 306
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 293,
+ 506,
+ 306
+ ],
+ "score": 1.0,
+ "content": "Karsten M. Borgwardt, Cheng Soon Ong, Stefan Schonauer, S. V. N. Vishwanathan, Alex J. Smola, ¨",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 115,
+ 304,
+ 505,
+ 317
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 304,
+ 505,
+ 317
+ ],
+ "score": 1.0,
+ "content": "and Hans-Peter Kriegel. Protein function prediction via graph kernels. Bioinformatics, 21(1):",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 115,
+ 315,
+ 505,
+ 329
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 315,
+ 505,
+ 329
+ ],
+ "score": 1.0,
+ "content": "47–56, January 2005. ISSN 1367-4803. doi: 10.1093/bioinformatics/bti1007. URL http:",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 117,
+ 327,
+ 379,
+ 338
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 117,
+ 327,
+ 379,
+ 338
+ ],
+ "score": 1.0,
+ "content": "//dx.doi.org/10.1093/bioinformatics/bti1007.",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ }
+ ],
+ "index": 17.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 344,
+ 505,
+ 378
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 344,
+ 505,
+ 356
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 344,
+ 505,
+ 356
+ ],
+ "score": 1.0,
+ "content": "Michael M. Bronstein, Joan Bruna, Yann LeCun, Arthur Szlam, and Pierre Vandergheynst. Ge-",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 115,
+ 356,
+ 505,
+ 367
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 356,
+ 505,
+ 367
+ ],
+ "score": 1.0,
+ "content": "ometric deep learning: going beyond euclidean data. CoRR, abs/1611.08097, 2016. URL",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 117,
+ 367,
+ 307,
+ 379
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 117,
+ 367,
+ 307,
+ 379
+ ],
+ "score": 1.0,
+ "content": "http://arxiv.org/abs/1611.08097.",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ }
+ ],
+ "index": 21
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 384,
+ 505,
+ 418
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 384,
+ 504,
+ 397
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 384,
+ 504,
+ 397
+ ],
+ "score": 1.0,
+ "content": "Joan Bruna, Wojciech Zaremba, Arthur Szlam, and Yann Lecun. Spectral networks and lo-",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ },
+ {
+ "bbox": [
+ 116,
+ 396,
+ 505,
+ 408
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 396,
+ 505,
+ 408
+ ],
+ "score": 1.0,
+ "content": "cally connected networks on graphs. In International Conference on Learning Representations",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 116,
+ 407,
+ 271,
+ 419
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 407,
+ 271,
+ 419
+ ],
+ "score": 1.0,
+ "content": "(ICLR2014), CBLS, April 2014, 2014.",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ }
+ ],
+ "index": 24
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 424,
+ 505,
+ 459
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 424,
+ 505,
+ 438
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 424,
+ 505,
+ 438
+ ],
+ "score": 1.0,
+ "content": "Duen Horng Chau, Carey Nachenberg, Jeffrey Wilhelm, Adam Wright, and Christos Faloutsos.",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 115,
+ 435,
+ 505,
+ 449
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 435,
+ 505,
+ 449
+ ],
+ "score": 1.0,
+ "content": "Polonium: Tera-scale graph mining and inference for malware detection. In SIAM INTERNA-",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 116,
+ 446,
+ 412,
+ 459
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 446,
+ 412,
+ 459
+ ],
+ "score": 1.0,
+ "content": "TIONAL CONFERENCE ON DATA MINING (SDM), pp. 131–142, 2011.",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ }
+ ],
+ "index": 27
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 464,
+ 505,
+ 498
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 464,
+ 506,
+ 478
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 464,
+ 506,
+ 478
+ ],
+ "score": 1.0,
+ "content": "Wei-Yu Chen, Yen-Cheng Liu, Zsolt Kira, Yu-Chiang Frank Wang, and Jia-Bin Huang. A closer",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 116,
+ 476,
+ 505,
+ 488
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 476,
+ 505,
+ 488
+ ],
+ "score": 1.0,
+ "content": "look at few-shot classification. In International Conference on Learning Representations, 2019.",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 116,
+ 486,
+ 394,
+ 500
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 486,
+ 325,
+ 500
+ ],
+ "score": 1.0,
+ "content": "URL https://openreview.net/forum?id",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 325,
+ 488,
+ 331,
+ 496
+ ],
+ "score": 0.53,
+ "content": "=",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 332,
+ 486,
+ 394,
+ 500
+ ],
+ "score": 1.0,
+ "content": "HkxLXnAcFQ.",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ }
+ ],
+ "index": 30
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 505,
+ 505,
+ 549
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 505,
+ 505,
+ 518
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 505,
+ 505,
+ 518
+ ],
+ "score": 1.0,
+ "content": "Michael Defferrard, Xavier Bresson, and Pierre Vandergheynst. Convolutional neural networks on ¨",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 115,
+ 515,
+ 506,
+ 529
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 515,
+ 506,
+ 529
+ ],
+ "score": 1.0,
+ "content": "graphs with fast localized spectral filtering. In D. D. Lee, M. Sugiyama, U. V. Luxburg, I. Guyon,",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ },
+ {
+ "bbox": [
+ 115,
+ 527,
+ 505,
+ 540
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 527,
+ 505,
+ 540
+ ],
+ "score": 1.0,
+ "content": "and R. Garnett (eds.), Advances in Neural Information Processing Systems 29, pp. 3844–3852.",
+ "type": "text"
+ }
+ ],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 117,
+ 538,
+ 239,
+ 550
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 117,
+ 538,
+ 239,
+ 550
+ ],
+ "score": 1.0,
+ "content": "Curran Associates, Inc., 2016.",
+ "type": "text"
+ }
+ ],
+ "index": 35
+ }
+ ],
+ "index": 33.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 556,
+ 505,
+ 611
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 556,
+ 506,
+ 569
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 556,
+ 506,
+ 569
+ ],
+ "score": 1.0,
+ "content": "David K Duvenaud, Dougal Maclaurin, Jorge Iparraguirre, Rafael Bombarell, Timothy Hirzel, Alan",
+ "type": "text"
+ }
+ ],
+ "index": 36
+ },
+ {
+ "bbox": [
+ 117,
+ 567,
+ 505,
+ 579
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 117,
+ 567,
+ 505,
+ 579
+ ],
+ "score": 1.0,
+ "content": "Aspuru-Guzik, and Ryan P Adams. Convolutional networks on graphs for learning molecular",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 115,
+ 578,
+ 506,
+ 591
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 578,
+ 506,
+ 591
+ ],
+ "score": 1.0,
+ "content": "fingerprints. In C. Cortes, N. D. Lawrence, D. D. Lee, M. Sugiyama, and R. Garnett (eds.),",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ },
+ {
+ "bbox": [
+ 114,
+ 587,
+ 506,
+ 603
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 114,
+ 587,
+ 506,
+ 603
+ ],
+ "score": 1.0,
+ "content": "Advances in Neural Information Processing Systems 28, pp. 2224–2232. Curran Associates, Inc.,",
+ "type": "text"
+ }
+ ],
+ "index": 39
+ },
+ {
+ "bbox": [
+ 115,
+ 599,
+ 142,
+ 612
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 599,
+ 142,
+ 612
+ ],
+ "score": 1.0,
+ "content": "2015.",
+ "type": "text"
+ }
+ ],
+ "index": 40
+ }
+ ],
+ "index": 38
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 108,
+ 618,
+ 503,
+ 652
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 617,
+ 504,
+ 631
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 617,
+ 504,
+ 631
+ ],
+ "score": 1.0,
+ "content": "Li Fei-Fei, Rob Fergus, and Pietro Perona. One-shot learning of object categories. IEEE Trans.",
+ "type": "text"
+ }
+ ],
+ "index": 41
+ },
+ {
+ "bbox": [
+ 116,
+ 629,
+ 505,
+ 641
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 629,
+ 505,
+ 641
+ ],
+ "score": 1.0,
+ "content": "Pattern Anal. Mach. Intell., 28(4):594–611, April 2006. ISSN 0162-8828. doi: 10.1109/TPAMI.",
+ "type": "text"
+ }
+ ],
+ "index": 42
+ },
+ {
+ "bbox": [
+ 116,
+ 639,
+ 404,
+ 653
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 639,
+ 404,
+ 653
+ ],
+ "score": 1.0,
+ "content": "2006.79. URL https://doi.org/10.1109/TPAMI.2006.79.",
+ "type": "text"
+ }
+ ],
+ "index": 43
+ }
+ ],
+ "index": 42
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 105,
+ 658,
+ 504,
+ 691
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 657,
+ 505,
+ 671
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 657,
+ 505,
+ 671
+ ],
+ "score": 1.0,
+ "content": "Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model-agnostic meta-learning for fast adaptation",
+ "type": "text"
+ }
+ ],
+ "index": 44
+ },
+ {
+ "bbox": [
+ 116,
+ 669,
+ 504,
+ 681
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 669,
+ 504,
+ 681
+ ],
+ "score": 1.0,
+ "content": "of deep networks. CoRR, abs/1703.03400, 2017. URL http://arxiv.org/abs/1703.",
+ "type": "text"
+ }
+ ],
+ "index": 45
+ },
+ {
+ "bbox": [
+ 116,
+ 680,
+ 152,
+ 692
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 680,
+ 152,
+ 692
+ ],
+ "score": 1.0,
+ "content": "03400.",
+ "type": "text"
+ }
+ ],
+ "index": 46
+ }
+ ],
+ "index": 45
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 698,
+ 504,
+ 731
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 696,
+ 505,
+ 713
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 696,
+ 505,
+ 713
+ ],
+ "score": 1.0,
+ "content": "Aude Genevay, Marco Cuturi, Gabriel Peyre, and Francis Bach. Stochastic optimization for large- ´",
+ "type": "text"
+ }
+ ],
+ "index": 47
+ },
+ {
+ "bbox": [
+ 115,
+ 710,
+ 505,
+ 722
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 710,
+ 505,
+ 722
+ ],
+ "score": 1.0,
+ "content": "scale optimal transport. In Advances in Neural Information Processing Systems 29, pp. 3440–",
+ "type": "text"
+ }
+ ],
+ "index": 48
+ },
+ {
+ "bbox": [
+ 115,
+ 719,
+ 167,
+ 732
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 719,
+ 167,
+ 732
+ ],
+ "score": 1.0,
+ "content": "3448. 2016.",
+ "type": "text"
+ }
+ ],
+ "index": 49
+ }
+ ],
+ "index": 48
+ }
+ ],
+ "page_idx": 10,
+ "page_size": [
+ 612,
+ 792
+ ],
+ "discarded_blocks": [
+ {
+ "type": "discarded",
+ "bbox": [
+ 107,
+ 27,
+ 293,
+ 37
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "score": 1.0,
+ "content": "Published as a conference paper at ICLR 2020",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 300,
+ 751,
+ 310,
+ 760
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 299,
+ 750,
+ 312,
+ 765
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 299,
+ 750,
+ 312,
+ 765
+ ],
+ "score": 1.0,
+ "content": "",
+ "type": "text",
+ "height": 15,
+ "width": 13
+ }
+ ]
+ }
+ ]
+ }
+ ],
+ "para_blocks": [
+ {
+ "type": "title",
+ "bbox": [
+ 107,
+ 81,
+ 176,
+ 94
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 82,
+ 176,
+ 94
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 82,
+ 176,
+ 94
+ ],
+ "score": 1.0,
+ "content": "REFERENCES",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ }
+ ],
+ "index": 0
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 100,
+ 504,
+ 122
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 98,
+ 505,
+ 113
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 98,
+ 505,
+ 113
+ ],
+ "score": 1.0,
+ "content": "Martial Agueh and Guillaume Carlier. Barycenters in the wasserstein space. SIAM J. Math. Analysis,",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 116,
+ 110,
+ 206,
+ 123
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 110,
+ 206,
+ 123
+ ],
+ "score": 1.0,
+ "content": "43(2):904–924, 2011.",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ }
+ ],
+ "index": 1.5,
+ "bbox_fs": [
+ 105,
+ 98,
+ 505,
+ 123
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 128,
+ 505,
+ 162
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 129,
+ 504,
+ 141
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 129,
+ 298,
+ 141
+ ],
+ "score": 1.0,
+ "content": "David Arthur and Sergei Vassilvitskii. K-means",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 298,
+ 130,
+ 309,
+ 139
+ ],
+ "score": 0.58,
+ "content": "^ { + + }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 310,
+ 129,
+ 504,
+ 141
+ ],
+ "score": 1.0,
+ "content": ": The advantages of careful seeding. In Proceed-",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 115,
+ 138,
+ 505,
+ 154
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 138,
+ 505,
+ 154
+ ],
+ "score": 1.0,
+ "content": "ings of the Eighteenth Annual ACM-SIAM Symposium on Discrete Algorithms, SODA ’07, pp.",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 117,
+ 151,
+ 192,
+ 162
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 117,
+ 151,
+ 192,
+ 162
+ ],
+ "score": 1.0,
+ "content": "1027–1035, 2007.",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ }
+ ],
+ "index": 4,
+ "bbox_fs": [
+ 106,
+ 129,
+ 505,
+ 162
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 169,
+ 504,
+ 202
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 169,
+ 504,
+ 181
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 169,
+ 504,
+ 181
+ ],
+ "score": 1.0,
+ "content": "Lars Backstrom and Jure Leskovec. Supervised random walks: Predicting and recommending links",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 115,
+ 180,
+ 503,
+ 192
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 180,
+ 503,
+ 192
+ ],
+ "score": 1.0,
+ "content": "in social networks. CoRR, abs/1011.4071, 2010. URL http://arxiv.org/abs/1011.",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 115,
+ 191,
+ 146,
+ 203
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 191,
+ 146,
+ 203
+ ],
+ "score": 1.0,
+ "content": "4071.",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ }
+ ],
+ "index": 7,
+ "bbox_fs": [
+ 105,
+ 169,
+ 504,
+ 203
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 209,
+ 505,
+ 287
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 209,
+ 505,
+ 222
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 209,
+ 505,
+ 222
+ ],
+ "score": 1.0,
+ "content": "Peter Battaglia, Jessica Blake Chandler Hamrick, Victor Bapst, Alvaro Sanchez, Vinicius Zam-",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 115,
+ 219,
+ 505,
+ 233
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 219,
+ 505,
+ 233
+ ],
+ "score": 1.0,
+ "content": "baldi, Mateusz Malinowski, Andrea Tacchetti, David Raposo, Adam Santoro, Ryan Faulkner,",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 115,
+ 231,
+ 505,
+ 244
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 231,
+ 505,
+ 244
+ ],
+ "score": 1.0,
+ "content": "Caglar Gulcehre, Francis Song, Andy Ballard, Justin Gilmer, George E. Dahl, Ashish Vaswani,",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 116,
+ 242,
+ 505,
+ 255
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 242,
+ 505,
+ 255
+ ],
+ "score": 1.0,
+ "content": "Kelsey Allen, Charles Nash, Victoria Jayne Langston, Chris Dyer, Nicolas Heess, Daan Wier-",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 115,
+ 253,
+ 505,
+ 265
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 253,
+ 505,
+ 265
+ ],
+ "score": 1.0,
+ "content": "stra, Pushmeet Kohli, Matt Botvinick, Oriol Vinyals, Yujia Li, and Razvan Pascanu. Re-",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 115,
+ 263,
+ 504,
+ 277
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 263,
+ 504,
+ 277
+ ],
+ "score": 1.0,
+ "content": "lational inductive biases, deep learning, and graph networks. arXiv, 2018. URL https:",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 117,
+ 276,
+ 301,
+ 287
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 117,
+ 276,
+ 301,
+ 287
+ ],
+ "score": 1.0,
+ "content": "//arxiv.org/pdf/1806.01261.pdf.",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ }
+ ],
+ "index": 12,
+ "bbox_fs": [
+ 105,
+ 209,
+ 505,
+ 287
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 293,
+ 505,
+ 338
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 293,
+ 506,
+ 306
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 293,
+ 506,
+ 306
+ ],
+ "score": 1.0,
+ "content": "Karsten M. Borgwardt, Cheng Soon Ong, Stefan Schonauer, S. V. N. Vishwanathan, Alex J. Smola, ¨",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 115,
+ 304,
+ 505,
+ 317
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 304,
+ 505,
+ 317
+ ],
+ "score": 1.0,
+ "content": "and Hans-Peter Kriegel. Protein function prediction via graph kernels. Bioinformatics, 21(1):",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 115,
+ 315,
+ 505,
+ 329
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 315,
+ 505,
+ 329
+ ],
+ "score": 1.0,
+ "content": "47–56, January 2005. ISSN 1367-4803. doi: 10.1093/bioinformatics/bti1007. URL http:",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 117,
+ 327,
+ 379,
+ 338
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 117,
+ 327,
+ 379,
+ 338
+ ],
+ "score": 1.0,
+ "content": "//dx.doi.org/10.1093/bioinformatics/bti1007.",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ }
+ ],
+ "index": 17.5,
+ "bbox_fs": [
+ 106,
+ 293,
+ 506,
+ 338
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 344,
+ 505,
+ 378
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 344,
+ 505,
+ 356
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 344,
+ 505,
+ 356
+ ],
+ "score": 1.0,
+ "content": "Michael M. Bronstein, Joan Bruna, Yann LeCun, Arthur Szlam, and Pierre Vandergheynst. Ge-",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 115,
+ 356,
+ 505,
+ 367
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 356,
+ 505,
+ 367
+ ],
+ "score": 1.0,
+ "content": "ometric deep learning: going beyond euclidean data. CoRR, abs/1611.08097, 2016. URL",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 117,
+ 367,
+ 307,
+ 379
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 117,
+ 367,
+ 307,
+ 379
+ ],
+ "score": 1.0,
+ "content": "http://arxiv.org/abs/1611.08097.",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ }
+ ],
+ "index": 21,
+ "bbox_fs": [
+ 106,
+ 344,
+ 505,
+ 379
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 384,
+ 505,
+ 418
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 384,
+ 504,
+ 397
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 384,
+ 504,
+ 397
+ ],
+ "score": 1.0,
+ "content": "Joan Bruna, Wojciech Zaremba, Arthur Szlam, and Yann Lecun. Spectral networks and lo-",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ },
+ {
+ "bbox": [
+ 116,
+ 396,
+ 505,
+ 408
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 396,
+ 505,
+ 408
+ ],
+ "score": 1.0,
+ "content": "cally connected networks on graphs. In International Conference on Learning Representations",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 116,
+ 407,
+ 271,
+ 419
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 407,
+ 271,
+ 419
+ ],
+ "score": 1.0,
+ "content": "(ICLR2014), CBLS, April 2014, 2014.",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ }
+ ],
+ "index": 24,
+ "bbox_fs": [
+ 105,
+ 384,
+ 505,
+ 419
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 424,
+ 505,
+ 459
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 424,
+ 505,
+ 438
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 424,
+ 505,
+ 438
+ ],
+ "score": 1.0,
+ "content": "Duen Horng Chau, Carey Nachenberg, Jeffrey Wilhelm, Adam Wright, and Christos Faloutsos.",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 115,
+ 435,
+ 505,
+ 449
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 435,
+ 505,
+ 449
+ ],
+ "score": 1.0,
+ "content": "Polonium: Tera-scale graph mining and inference for malware detection. In SIAM INTERNA-",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 116,
+ 446,
+ 412,
+ 459
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 446,
+ 412,
+ 459
+ ],
+ "score": 1.0,
+ "content": "TIONAL CONFERENCE ON DATA MINING (SDM), pp. 131–142, 2011.",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ }
+ ],
+ "index": 27,
+ "bbox_fs": [
+ 105,
+ 424,
+ 505,
+ 459
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 464,
+ 505,
+ 498
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 464,
+ 506,
+ 478
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 464,
+ 506,
+ 478
+ ],
+ "score": 1.0,
+ "content": "Wei-Yu Chen, Yen-Cheng Liu, Zsolt Kira, Yu-Chiang Frank Wang, and Jia-Bin Huang. A closer",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 116,
+ 476,
+ 505,
+ 488
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 476,
+ 505,
+ 488
+ ],
+ "score": 1.0,
+ "content": "look at few-shot classification. In International Conference on Learning Representations, 2019.",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 116,
+ 486,
+ 394,
+ 500
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 486,
+ 325,
+ 500
+ ],
+ "score": 1.0,
+ "content": "URL https://openreview.net/forum?id",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 325,
+ 488,
+ 331,
+ 496
+ ],
+ "score": 0.53,
+ "content": "=",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 332,
+ 486,
+ 394,
+ 500
+ ],
+ "score": 1.0,
+ "content": "HkxLXnAcFQ.",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ }
+ ],
+ "index": 30,
+ "bbox_fs": [
+ 106,
+ 464,
+ 506,
+ 500
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 505,
+ 505,
+ 549
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 505,
+ 505,
+ 518
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 505,
+ 505,
+ 518
+ ],
+ "score": 1.0,
+ "content": "Michael Defferrard, Xavier Bresson, and Pierre Vandergheynst. Convolutional neural networks on ¨",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 115,
+ 515,
+ 506,
+ 529
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 515,
+ 506,
+ 529
+ ],
+ "score": 1.0,
+ "content": "graphs with fast localized spectral filtering. In D. D. Lee, M. Sugiyama, U. V. Luxburg, I. Guyon,",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ },
+ {
+ "bbox": [
+ 115,
+ 527,
+ 505,
+ 540
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 527,
+ 505,
+ 540
+ ],
+ "score": 1.0,
+ "content": "and R. Garnett (eds.), Advances in Neural Information Processing Systems 29, pp. 3844–3852.",
+ "type": "text"
+ }
+ ],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 117,
+ 538,
+ 239,
+ 550
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 117,
+ 538,
+ 239,
+ 550
+ ],
+ "score": 1.0,
+ "content": "Curran Associates, Inc., 2016.",
+ "type": "text"
+ }
+ ],
+ "index": 35
+ }
+ ],
+ "index": 33.5,
+ "bbox_fs": [
+ 105,
+ 505,
+ 506,
+ 550
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 556,
+ 505,
+ 611
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 556,
+ 506,
+ 569
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 556,
+ 506,
+ 569
+ ],
+ "score": 1.0,
+ "content": "David K Duvenaud, Dougal Maclaurin, Jorge Iparraguirre, Rafael Bombarell, Timothy Hirzel, Alan",
+ "type": "text"
+ }
+ ],
+ "index": 36
+ },
+ {
+ "bbox": [
+ 117,
+ 567,
+ 505,
+ 579
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 117,
+ 567,
+ 505,
+ 579
+ ],
+ "score": 1.0,
+ "content": "Aspuru-Guzik, and Ryan P Adams. Convolutional networks on graphs for learning molecular",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 115,
+ 578,
+ 506,
+ 591
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 578,
+ 506,
+ 591
+ ],
+ "score": 1.0,
+ "content": "fingerprints. In C. Cortes, N. D. Lawrence, D. D. Lee, M. Sugiyama, and R. Garnett (eds.),",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ },
+ {
+ "bbox": [
+ 114,
+ 587,
+ 506,
+ 603
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 114,
+ 587,
+ 506,
+ 603
+ ],
+ "score": 1.0,
+ "content": "Advances in Neural Information Processing Systems 28, pp. 2224–2232. Curran Associates, Inc.,",
+ "type": "text"
+ }
+ ],
+ "index": 39
+ },
+ {
+ "bbox": [
+ 115,
+ 599,
+ 142,
+ 612
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 599,
+ 142,
+ 612
+ ],
+ "score": 1.0,
+ "content": "2015.",
+ "type": "text"
+ }
+ ],
+ "index": 40
+ }
+ ],
+ "index": 38,
+ "bbox_fs": [
+ 106,
+ 556,
+ 506,
+ 612
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 108,
+ 618,
+ 503,
+ 652
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 617,
+ 504,
+ 631
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 617,
+ 504,
+ 631
+ ],
+ "score": 1.0,
+ "content": "Li Fei-Fei, Rob Fergus, and Pietro Perona. One-shot learning of object categories. IEEE Trans.",
+ "type": "text"
+ }
+ ],
+ "index": 41
+ },
+ {
+ "bbox": [
+ 116,
+ 629,
+ 505,
+ 641
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 629,
+ 505,
+ 641
+ ],
+ "score": 1.0,
+ "content": "Pattern Anal. Mach. Intell., 28(4):594–611, April 2006. ISSN 0162-8828. doi: 10.1109/TPAMI.",
+ "type": "text"
+ }
+ ],
+ "index": 42
+ },
+ {
+ "bbox": [
+ 116,
+ 639,
+ 404,
+ 653
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 639,
+ 404,
+ 653
+ ],
+ "score": 1.0,
+ "content": "2006.79. URL https://doi.org/10.1109/TPAMI.2006.79.",
+ "type": "text"
+ }
+ ],
+ "index": 43
+ }
+ ],
+ "index": 42,
+ "bbox_fs": [
+ 105,
+ 617,
+ 505,
+ 653
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 105,
+ 658,
+ 504,
+ 691
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 657,
+ 505,
+ 671
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 657,
+ 505,
+ 671
+ ],
+ "score": 1.0,
+ "content": "Chelsea Finn, Pieter Abbeel, and Sergey Levine. Model-agnostic meta-learning for fast adaptation",
+ "type": "text"
+ }
+ ],
+ "index": 44
+ },
+ {
+ "bbox": [
+ 116,
+ 669,
+ 504,
+ 681
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 669,
+ 504,
+ 681
+ ],
+ "score": 1.0,
+ "content": "of deep networks. CoRR, abs/1703.03400, 2017. URL http://arxiv.org/abs/1703.",
+ "type": "text"
+ }
+ ],
+ "index": 45
+ },
+ {
+ "bbox": [
+ 116,
+ 680,
+ 152,
+ 692
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 680,
+ 152,
+ 692
+ ],
+ "score": 1.0,
+ "content": "03400.",
+ "type": "text"
+ }
+ ],
+ "index": 46
+ }
+ ],
+ "index": 45,
+ "bbox_fs": [
+ 105,
+ 657,
+ 505,
+ 692
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 698,
+ 504,
+ 731
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 696,
+ 505,
+ 713
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 696,
+ 505,
+ 713
+ ],
+ "score": 1.0,
+ "content": "Aude Genevay, Marco Cuturi, Gabriel Peyre, and Francis Bach. Stochastic optimization for large- ´",
+ "type": "text"
+ }
+ ],
+ "index": 47
+ },
+ {
+ "bbox": [
+ 115,
+ 710,
+ 505,
+ 722
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 710,
+ 505,
+ 722
+ ],
+ "score": 1.0,
+ "content": "scale optimal transport. In Advances in Neural Information Processing Systems 29, pp. 3440–",
+ "type": "text"
+ }
+ ],
+ "index": 48
+ },
+ {
+ "bbox": [
+ 115,
+ 719,
+ 167,
+ 732
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 719,
+ 167,
+ 732
+ ],
+ "score": 1.0,
+ "content": "3448. 2016.",
+ "type": "text"
+ }
+ ],
+ "index": 49
+ }
+ ],
+ "index": 48,
+ "bbox_fs": [
+ 105,
+ 696,
+ 505,
+ 732
+ ]
+ }
+ ]
+ },
+ {
+ "preproc_blocks": [
+ {
+ "type": "text",
+ "bbox": [
+ 108,
+ 82,
+ 504,
+ 116
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 82,
+ 505,
+ 95
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 82,
+ 505,
+ 95
+ ],
+ "score": 1.0,
+ "content": "Marco Gori, Gabriele Monfardini, and Franco Scarselli. A new model for learning in graph domains.",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 115,
+ 93,
+ 505,
+ 106
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 93,
+ 505,
+ 106
+ ],
+ "score": 1.0,
+ "content": "In Proceedings. 2005 IEEE International Joint Conference on Neural Networks, 2005., volume 2,",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 115,
+ 105,
+ 223,
+ 116
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 105,
+ 223,
+ 116
+ ],
+ "score": 1.0,
+ "content": "pp. 729–734. IEEE, 2005.",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ }
+ ],
+ "index": 1
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 104,
+ 123,
+ 504,
+ 147
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 104,
+ 122,
+ 505,
+ 138
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 122,
+ 505,
+ 138
+ ],
+ "score": 1.0,
+ "content": "Jiao Gu, Bobo Hua, and Shiping Liu. Spectral distances on graphs. Discrete Applied Mathematics,",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 117,
+ 135,
+ 212,
+ 147
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 117,
+ 135,
+ 212,
+ 147
+ ],
+ "score": 1.0,
+ "content": "190-191:56 – 74, 2015.",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ }
+ ],
+ "index": 3.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 155,
+ 504,
+ 178
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 153,
+ 506,
+ 169
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 153,
+ 506,
+ 169
+ ],
+ "score": 1.0,
+ "content": "William L. Hamilton, Rex Ying, and Jure Leskovec. Inductive representation learning on large",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 115,
+ 165,
+ 486,
+ 179
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 165,
+ 486,
+ 179
+ ],
+ "score": 1.0,
+ "content": "graphs. CoRR, abs/1706.02216, 2017. URL http://arxiv.org/abs/1706.02216.",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ }
+ ],
+ "index": 5.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 105,
+ 186,
+ 504,
+ 209
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 185,
+ 505,
+ 198
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 185,
+ 505,
+ 198
+ ],
+ "score": 1.0,
+ "content": "Mikael Henaff, Joan Bruna, and Yann LeCun. Deep convolutional networks on graph-structured",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 116,
+ 196,
+ 476,
+ 209
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 196,
+ 476,
+ 209
+ ],
+ "score": 1.0,
+ "content": "data. CoRR, abs/1506.05163, 2015. URL http://arxiv.org/abs/1506.05163.",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ }
+ ],
+ "index": 7.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 216,
+ 505,
+ 272
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 217,
+ 505,
+ 229
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 217,
+ 505,
+ 229
+ ],
+ "score": 1.0,
+ "content": "Sergey Ivanov and Evgeny Burnaev. Anonymous walk embeddings. In Jennifer Dy and An-",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 115,
+ 226,
+ 506,
+ 241
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 226,
+ 506,
+ 241
+ ],
+ "score": 1.0,
+ "content": "dreas Krause (eds.), Proceedings of the 35th International Conference on Machine Learning,",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 116,
+ 239,
+ 505,
+ 251
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 239,
+ 505,
+ 251
+ ],
+ "score": 1.0,
+ "content": "volume 80 of Proceedings of Machine Learning Research, pp. 2186–2195, Stockholmsmssan,",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 115,
+ 248,
+ 505,
+ 263
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 248,
+ 505,
+ 263
+ ],
+ "score": 1.0,
+ "content": "Stockholm Sweden, 10–15 Jul 2018. PMLR. URL http://proceedings.mlr.press/",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 116,
+ 261,
+ 229,
+ 272
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 261,
+ 229,
+ 272
+ ],
+ "score": 1.0,
+ "content": "v80/ivanov18a.html.",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ }
+ ],
+ "index": 11
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 280,
+ 503,
+ 315
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 280,
+ 505,
+ 293
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 280,
+ 505,
+ 293
+ ],
+ "score": 1.0,
+ "content": "Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization, 2014. URL",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 116,
+ 291,
+ 505,
+ 304
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 291,
+ 505,
+ 304
+ ],
+ "score": 1.0,
+ "content": "http://arxiv.org/abs/1412.6980. cite arxiv:1412.6980Comment: Published as a con-",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 116,
+ 302,
+ 505,
+ 315
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 302,
+ 505,
+ 315
+ ],
+ "score": 1.0,
+ "content": "ference paper at the 3rd International Conference for Learning Representations, San Diego, 2015.",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ }
+ ],
+ "index": 15
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 322,
+ 502,
+ 345
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 322,
+ 504,
+ 335
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 322,
+ 504,
+ 335
+ ],
+ "score": 1.0,
+ "content": "Thomas N. Kipf and Max Welling. Semi-supervised classification with graph convolutional net-",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 116,
+ 333,
+ 483,
+ 346
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 333,
+ 483,
+ 346
+ ],
+ "score": 1.0,
+ "content": "works. CoRR, abs/1609.02907, 2016. URL http://arxiv.org/abs/1609.02907.",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ }
+ ],
+ "index": 17.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 352,
+ 503,
+ 376
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 104,
+ 351,
+ 505,
+ 367
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 351,
+ 505,
+ 367
+ ],
+ "score": 1.0,
+ "content": "Nils M. Kriege, Fredrik D. Johansson, and Christopher Morris. A survey on graph kernels. ArXiv,",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 116,
+ 364,
+ 210,
+ 376
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 364,
+ 210,
+ 376
+ ],
+ "score": 1.0,
+ "content": "abs/1903.11835, 2019.",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ }
+ ],
+ "index": 19.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 384,
+ 505,
+ 440
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 383,
+ 505,
+ 396
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 383,
+ 505,
+ 396
+ ],
+ "score": 1.0,
+ "content": "Junhyun Lee, Inyeop Lee, and Jaewoo Kang. Self-attention graph pooling. In Kamalika Chaud-",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 115,
+ 393,
+ 505,
+ 408
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 393,
+ 505,
+ 408
+ ],
+ "score": 1.0,
+ "content": "huri and Ruslan Salakhutdinov (eds.), Proceedings of the 36th International Conference on",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 115,
+ 405,
+ 505,
+ 419
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 405,
+ 505,
+ 419
+ ],
+ "score": 1.0,
+ "content": "Machine Learning, volume 97 of Proceedings of Machine Learning Research, pp. 3734–3743,",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ },
+ {
+ "bbox": [
+ 115,
+ 416,
+ 504,
+ 429
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 416,
+ 504,
+ 429
+ ],
+ "score": 1.0,
+ "content": "Long Beach, California, USA, 09–15 Jun 2019. PMLR. URL http://proceedings.mlr.",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 115,
+ 429,
+ 247,
+ 440
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 429,
+ 247,
+ 440
+ ],
+ "score": 1.0,
+ "content": "press/v97/lee19c.html.",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ }
+ ],
+ "index": 23
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 447,
+ 504,
+ 481
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 447,
+ 506,
+ 460
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 447,
+ 506,
+ 460
+ ],
+ "score": 1.0,
+ "content": "Qimai Li, Zhichao Han, and Xiao-Ming Wu. Deeper insights into graph convolutional networks for",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 115,
+ 459,
+ 505,
+ 471
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 459,
+ 505,
+ 471
+ ],
+ "score": 1.0,
+ "content": "semi-supervised learning. CoRR, abs/1801.07606, 2018a. URL http://arxiv.org/abs/",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 116,
+ 469,
+ 182,
+ 481
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 469,
+ 182,
+ 481
+ ],
+ "score": 1.0,
+ "content": "1801.07606.",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ }
+ ],
+ "index": 27
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 105,
+ 489,
+ 504,
+ 512
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 107,
+ 489,
+ 505,
+ 502
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 489,
+ 505,
+ 502
+ ],
+ "score": 1.0,
+ "content": "Qimai Li, Zhichao Han, and Xiao-Ming Wu. Deeper insights into graph convolutional networks for",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 116,
+ 501,
+ 289,
+ 512
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 501,
+ 289,
+ 512
+ ],
+ "score": 1.0,
+ "content": "semi-supervised learning. In AAAI, 2018b.",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ }
+ ],
+ "index": 29.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 520,
+ 504,
+ 554
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 519,
+ 505,
+ 534
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 519,
+ 505,
+ 534
+ ],
+ "score": 1.0,
+ "content": "Annamalai Narayanan, Mahinthan Chandramohan, Rajasekar Venkatesan, Lihui Chen, Yang Liu,",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ },
+ {
+ "bbox": [
+ 116,
+ 532,
+ 505,
+ 544
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 532,
+ 505,
+ 544
+ ],
+ "score": 1.0,
+ "content": "and Shantanu Jaiswal. graph2vec: Learning distributed representations of graphs. CoRR,",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 116,
+ 542,
+ 424,
+ 554
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 542,
+ 424,
+ 554
+ ],
+ "score": 1.0,
+ "content": "abs/1707.05005, 2017. URL http://arxiv.org/abs/1707.05005.",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ }
+ ],
+ "index": 32
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 561,
+ 503,
+ 585
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 562,
+ 504,
+ 574
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 562,
+ 504,
+ 574
+ ],
+ "score": 1.0,
+ "content": "Alex Nichol, Joshua Achiam, and John Schulman. On first-order meta-learning algorithms. CoRR,",
+ "type": "text"
+ }
+ ],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 116,
+ 573,
+ 424,
+ 585
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 573,
+ 424,
+ 585
+ ],
+ "score": 1.0,
+ "content": "abs/1803.02999, 2018. URL http://arxiv.org/abs/1803.02999.",
+ "type": "text"
+ }
+ ],
+ "index": 35
+ }
+ ],
+ "index": 34.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 592,
+ 505,
+ 638
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 592,
+ 505,
+ 605
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 592,
+ 505,
+ 605
+ ],
+ "score": 1.0,
+ "content": "Sachin Ravi and Hugo Larochelle. Optimization as a model for few-shot learning. In 5th Interna-",
+ "type": "text"
+ }
+ ],
+ "index": 36
+ },
+ {
+ "bbox": [
+ 116,
+ 604,
+ 505,
+ 616
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 604,
+ 505,
+ 616
+ ],
+ "score": 1.0,
+ "content": "tional Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017,",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 116,
+ 614,
+ 505,
+ 628
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 614,
+ 505,
+ 628
+ ],
+ "score": 1.0,
+ "content": "Conference Track Proceedings. OpenReview.net, 2017. URL https://openreview.net/",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ },
+ {
+ "bbox": [
+ 116,
+ 626,
+ 229,
+ 637
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 626,
+ 164,
+ 637
+ ],
+ "score": 1.0,
+ "content": "forum?id",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 165,
+ 627,
+ 171,
+ 636
+ ],
+ "score": 0.5,
+ "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 172,
+ 626,
+ 229,
+ 637
+ ],
+ "score": 1.0,
+ "content": "rJY0-Kcll.",
+ "type": "text"
+ }
+ ],
+ "index": 39
+ }
+ ],
+ "index": 37.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 645,
+ 505,
+ 680
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 645,
+ 505,
+ 659
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 645,
+ 505,
+ 659
+ ],
+ "score": 1.0,
+ "content": "Andrei A. Rusu, Dushyant Rao, Jakub Sygnowski, Oriol Vinyals, Razvan Pascanu, Simon Osindero,",
+ "type": "text"
+ }
+ ],
+ "index": 40
+ },
+ {
+ "bbox": [
+ 115,
+ 656,
+ 505,
+ 669
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 656,
+ 505,
+ 669
+ ],
+ "score": 1.0,
+ "content": "and Raia Hadsell. Meta-learning with latent embedding optimization. CoRR, abs/1807.05960,",
+ "type": "text"
+ }
+ ],
+ "index": 41
+ },
+ {
+ "bbox": [
+ 116,
+ 667,
+ 356,
+ 680
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 667,
+ 356,
+ 680
+ ],
+ "score": 1.0,
+ "content": "2018. URL http://arxiv.org/abs/1807.05960.",
+ "type": "text"
+ }
+ ],
+ "index": 42
+ }
+ ],
+ "index": 41
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 687,
+ 504,
+ 732
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 687,
+ 505,
+ 700
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 687,
+ 505,
+ 700
+ ],
+ "score": 1.0,
+ "content": "Franco Scarselli, Marco Gori, Ah Chung Tsoi, Markus Hagenbuchner, and Gabriele Monfardini.",
+ "type": "text"
+ }
+ ],
+ "index": 43
+ },
+ {
+ "bbox": [
+ 116,
+ 699,
+ 505,
+ 711
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 699,
+ 505,
+ 711
+ ],
+ "score": 1.0,
+ "content": "The graph neural network model. Trans. Neur. Netw., 20(1):61–80, January 2009. ISSN 1045-",
+ "type": "text"
+ }
+ ],
+ "index": 44
+ },
+ {
+ "bbox": [
+ 115,
+ 709,
+ 504,
+ 722
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 709,
+ 504,
+ 722
+ ],
+ "score": 1.0,
+ "content": "9227. doi: 10.1109/TNN.2008.2005605. URL http://dx.doi.org/10.1109/TNN.",
+ "type": "text"
+ }
+ ],
+ "index": 45
+ },
+ {
+ "bbox": [
+ 116,
+ 721,
+ 193,
+ 732
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 721,
+ 193,
+ 732
+ ],
+ "score": 1.0,
+ "content": "2008.2005605.",
+ "type": "text"
+ }
+ ],
+ "index": 46
+ }
+ ],
+ "index": 44.5
+ }
+ ],
+ "page_idx": 11,
+ "page_size": [
+ 612,
+ 792
+ ],
+ "discarded_blocks": [
+ {
+ "type": "discarded",
+ "bbox": [
+ 107,
+ 27,
+ 294,
+ 37
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "score": 1.0,
+ "content": "Published as a conference paper at ICLR 2020",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 300,
+ 751,
+ 311,
+ 760
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 299,
+ 750,
+ 313,
+ 764
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 299,
+ 750,
+ 313,
+ 764
+ ],
+ "score": 1.0,
+ "content": "12",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ }
+ ],
+ "para_blocks": [
+ {
+ "type": "text",
+ "bbox": [
+ 108,
+ 82,
+ 504,
+ 116
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 82,
+ 505,
+ 95
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 82,
+ 505,
+ 95
+ ],
+ "score": 1.0,
+ "content": "Marco Gori, Gabriele Monfardini, and Franco Scarselli. A new model for learning in graph domains.",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 115,
+ 93,
+ 505,
+ 106
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 93,
+ 505,
+ 106
+ ],
+ "score": 1.0,
+ "content": "In Proceedings. 2005 IEEE International Joint Conference on Neural Networks, 2005., volume 2,",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 115,
+ 105,
+ 223,
+ 116
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 105,
+ 223,
+ 116
+ ],
+ "score": 1.0,
+ "content": "pp. 729–734. IEEE, 2005.",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ }
+ ],
+ "index": 1,
+ "bbox_fs": [
+ 106,
+ 82,
+ 505,
+ 116
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 104,
+ 123,
+ 504,
+ 147
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 104,
+ 122,
+ 505,
+ 138
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 122,
+ 505,
+ 138
+ ],
+ "score": 1.0,
+ "content": "Jiao Gu, Bobo Hua, and Shiping Liu. Spectral distances on graphs. Discrete Applied Mathematics,",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 117,
+ 135,
+ 212,
+ 147
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 117,
+ 135,
+ 212,
+ 147
+ ],
+ "score": 1.0,
+ "content": "190-191:56 – 74, 2015.",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ }
+ ],
+ "index": 3.5,
+ "bbox_fs": [
+ 104,
+ 122,
+ 505,
+ 147
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 155,
+ 504,
+ 178
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 153,
+ 506,
+ 169
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 153,
+ 506,
+ 169
+ ],
+ "score": 1.0,
+ "content": "William L. Hamilton, Rex Ying, and Jure Leskovec. Inductive representation learning on large",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 115,
+ 165,
+ 486,
+ 179
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 165,
+ 486,
+ 179
+ ],
+ "score": 1.0,
+ "content": "graphs. CoRR, abs/1706.02216, 2017. URL http://arxiv.org/abs/1706.02216.",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ }
+ ],
+ "index": 5.5,
+ "bbox_fs": [
+ 105,
+ 153,
+ 506,
+ 179
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 105,
+ 186,
+ 504,
+ 209
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 185,
+ 505,
+ 198
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 185,
+ 505,
+ 198
+ ],
+ "score": 1.0,
+ "content": "Mikael Henaff, Joan Bruna, and Yann LeCun. Deep convolutional networks on graph-structured",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 116,
+ 196,
+ 476,
+ 209
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 196,
+ 476,
+ 209
+ ],
+ "score": 1.0,
+ "content": "data. CoRR, abs/1506.05163, 2015. URL http://arxiv.org/abs/1506.05163.",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ }
+ ],
+ "index": 7.5,
+ "bbox_fs": [
+ 105,
+ 185,
+ 505,
+ 209
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 216,
+ 505,
+ 272
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 217,
+ 505,
+ 229
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 217,
+ 505,
+ 229
+ ],
+ "score": 1.0,
+ "content": "Sergey Ivanov and Evgeny Burnaev. Anonymous walk embeddings. In Jennifer Dy and An-",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 115,
+ 226,
+ 506,
+ 241
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 226,
+ 506,
+ 241
+ ],
+ "score": 1.0,
+ "content": "dreas Krause (eds.), Proceedings of the 35th International Conference on Machine Learning,",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 116,
+ 239,
+ 505,
+ 251
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 239,
+ 505,
+ 251
+ ],
+ "score": 1.0,
+ "content": "volume 80 of Proceedings of Machine Learning Research, pp. 2186–2195, Stockholmsmssan,",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 115,
+ 248,
+ 505,
+ 263
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 248,
+ 505,
+ 263
+ ],
+ "score": 1.0,
+ "content": "Stockholm Sweden, 10–15 Jul 2018. PMLR. URL http://proceedings.mlr.press/",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 116,
+ 261,
+ 229,
+ 272
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 261,
+ 229,
+ 272
+ ],
+ "score": 1.0,
+ "content": "v80/ivanov18a.html.",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ }
+ ],
+ "index": 11,
+ "bbox_fs": [
+ 106,
+ 217,
+ 506,
+ 272
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 280,
+ 503,
+ 315
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 280,
+ 505,
+ 293
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 280,
+ 505,
+ 293
+ ],
+ "score": 1.0,
+ "content": "Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization, 2014. URL",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 116,
+ 291,
+ 505,
+ 304
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 291,
+ 505,
+ 304
+ ],
+ "score": 1.0,
+ "content": "http://arxiv.org/abs/1412.6980. cite arxiv:1412.6980Comment: Published as a con-",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 116,
+ 302,
+ 505,
+ 315
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 302,
+ 505,
+ 315
+ ],
+ "score": 1.0,
+ "content": "ference paper at the 3rd International Conference for Learning Representations, San Diego, 2015.",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ }
+ ],
+ "index": 15,
+ "bbox_fs": [
+ 106,
+ 280,
+ 505,
+ 315
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 322,
+ 502,
+ 345
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 322,
+ 504,
+ 335
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 322,
+ 504,
+ 335
+ ],
+ "score": 1.0,
+ "content": "Thomas N. Kipf and Max Welling. Semi-supervised classification with graph convolutional net-",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 116,
+ 333,
+ 483,
+ 346
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 333,
+ 483,
+ 346
+ ],
+ "score": 1.0,
+ "content": "works. CoRR, abs/1609.02907, 2016. URL http://arxiv.org/abs/1609.02907.",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ }
+ ],
+ "index": 17.5,
+ "bbox_fs": [
+ 106,
+ 322,
+ 504,
+ 346
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 352,
+ 503,
+ 376
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 104,
+ 351,
+ 505,
+ 367
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 351,
+ 505,
+ 367
+ ],
+ "score": 1.0,
+ "content": "Nils M. Kriege, Fredrik D. Johansson, and Christopher Morris. A survey on graph kernels. ArXiv,",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 116,
+ 364,
+ 210,
+ 376
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 364,
+ 210,
+ 376
+ ],
+ "score": 1.0,
+ "content": "abs/1903.11835, 2019.",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ }
+ ],
+ "index": 19.5,
+ "bbox_fs": [
+ 104,
+ 351,
+ 505,
+ 376
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 384,
+ 505,
+ 440
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 383,
+ 505,
+ 396
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 383,
+ 505,
+ 396
+ ],
+ "score": 1.0,
+ "content": "Junhyun Lee, Inyeop Lee, and Jaewoo Kang. Self-attention graph pooling. In Kamalika Chaud-",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 115,
+ 393,
+ 505,
+ 408
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 393,
+ 505,
+ 408
+ ],
+ "score": 1.0,
+ "content": "huri and Ruslan Salakhutdinov (eds.), Proceedings of the 36th International Conference on",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 115,
+ 405,
+ 505,
+ 419
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 405,
+ 505,
+ 419
+ ],
+ "score": 1.0,
+ "content": "Machine Learning, volume 97 of Proceedings of Machine Learning Research, pp. 3734–3743,",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ },
+ {
+ "bbox": [
+ 115,
+ 416,
+ 504,
+ 429
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 416,
+ 504,
+ 429
+ ],
+ "score": 1.0,
+ "content": "Long Beach, California, USA, 09–15 Jun 2019. PMLR. URL http://proceedings.mlr.",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 115,
+ 429,
+ 247,
+ 440
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 429,
+ 247,
+ 440
+ ],
+ "score": 1.0,
+ "content": "press/v97/lee19c.html.",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ }
+ ],
+ "index": 23,
+ "bbox_fs": [
+ 105,
+ 383,
+ 505,
+ 440
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 447,
+ 504,
+ 481
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 447,
+ 506,
+ 460
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 447,
+ 506,
+ 460
+ ],
+ "score": 1.0,
+ "content": "Qimai Li, Zhichao Han, and Xiao-Ming Wu. Deeper insights into graph convolutional networks for",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 115,
+ 459,
+ 505,
+ 471
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 459,
+ 505,
+ 471
+ ],
+ "score": 1.0,
+ "content": "semi-supervised learning. CoRR, abs/1801.07606, 2018a. URL http://arxiv.org/abs/",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 116,
+ 469,
+ 182,
+ 481
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 469,
+ 182,
+ 481
+ ],
+ "score": 1.0,
+ "content": "1801.07606.",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ }
+ ],
+ "index": 27,
+ "bbox_fs": [
+ 105,
+ 447,
+ 506,
+ 481
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 105,
+ 489,
+ 504,
+ 512
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 107,
+ 489,
+ 505,
+ 502
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 107,
+ 489,
+ 505,
+ 502
+ ],
+ "score": 1.0,
+ "content": "Qimai Li, Zhichao Han, and Xiao-Ming Wu. Deeper insights into graph convolutional networks for",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 116,
+ 501,
+ 289,
+ 512
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 501,
+ 289,
+ 512
+ ],
+ "score": 1.0,
+ "content": "semi-supervised learning. In AAAI, 2018b.",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ }
+ ],
+ "index": 29.5,
+ "bbox_fs": [
+ 107,
+ 489,
+ 505,
+ 512
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 520,
+ 504,
+ 554
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 519,
+ 505,
+ 534
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 519,
+ 505,
+ 534
+ ],
+ "score": 1.0,
+ "content": "Annamalai Narayanan, Mahinthan Chandramohan, Rajasekar Venkatesan, Lihui Chen, Yang Liu,",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ },
+ {
+ "bbox": [
+ 116,
+ 532,
+ 505,
+ 544
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 532,
+ 505,
+ 544
+ ],
+ "score": 1.0,
+ "content": "and Shantanu Jaiswal. graph2vec: Learning distributed representations of graphs. CoRR,",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 116,
+ 542,
+ 424,
+ 554
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 542,
+ 424,
+ 554
+ ],
+ "score": 1.0,
+ "content": "abs/1707.05005, 2017. URL http://arxiv.org/abs/1707.05005.",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ }
+ ],
+ "index": 32,
+ "bbox_fs": [
+ 105,
+ 519,
+ 505,
+ 554
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 561,
+ 503,
+ 585
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 562,
+ 504,
+ 574
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 562,
+ 504,
+ 574
+ ],
+ "score": 1.0,
+ "content": "Alex Nichol, Joshua Achiam, and John Schulman. On first-order meta-learning algorithms. CoRR,",
+ "type": "text"
+ }
+ ],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 116,
+ 573,
+ 424,
+ 585
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 573,
+ 424,
+ 585
+ ],
+ "score": 1.0,
+ "content": "abs/1803.02999, 2018. URL http://arxiv.org/abs/1803.02999.",
+ "type": "text"
+ }
+ ],
+ "index": 35
+ }
+ ],
+ "index": 34.5,
+ "bbox_fs": [
+ 106,
+ 562,
+ 504,
+ 585
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 592,
+ 505,
+ 638
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 592,
+ 505,
+ 605
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 592,
+ 505,
+ 605
+ ],
+ "score": 1.0,
+ "content": "Sachin Ravi and Hugo Larochelle. Optimization as a model for few-shot learning. In 5th Interna-",
+ "type": "text"
+ }
+ ],
+ "index": 36
+ },
+ {
+ "bbox": [
+ 116,
+ 604,
+ 505,
+ 616
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 604,
+ 505,
+ 616
+ ],
+ "score": 1.0,
+ "content": "tional Conference on Learning Representations, ICLR 2017, Toulon, France, April 24-26, 2017,",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 116,
+ 614,
+ 505,
+ 628
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 614,
+ 505,
+ 628
+ ],
+ "score": 1.0,
+ "content": "Conference Track Proceedings. OpenReview.net, 2017. URL https://openreview.net/",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ },
+ {
+ "bbox": [
+ 116,
+ 626,
+ 229,
+ 637
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 626,
+ 164,
+ 637
+ ],
+ "score": 1.0,
+ "content": "forum?id",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 165,
+ 627,
+ 171,
+ 636
+ ],
+ "score": 0.5,
+ "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 172,
+ 626,
+ 229,
+ 637
+ ],
+ "score": 1.0,
+ "content": "rJY0-Kcll.",
+ "type": "text"
+ }
+ ],
+ "index": 39
+ }
+ ],
+ "index": 37.5,
+ "bbox_fs": [
+ 106,
+ 592,
+ 505,
+ 637
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 645,
+ 505,
+ 680
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 645,
+ 505,
+ 659
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 645,
+ 505,
+ 659
+ ],
+ "score": 1.0,
+ "content": "Andrei A. Rusu, Dushyant Rao, Jakub Sygnowski, Oriol Vinyals, Razvan Pascanu, Simon Osindero,",
+ "type": "text"
+ }
+ ],
+ "index": 40
+ },
+ {
+ "bbox": [
+ 115,
+ 656,
+ 505,
+ 669
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 656,
+ 505,
+ 669
+ ],
+ "score": 1.0,
+ "content": "and Raia Hadsell. Meta-learning with latent embedding optimization. CoRR, abs/1807.05960,",
+ "type": "text"
+ }
+ ],
+ "index": 41
+ },
+ {
+ "bbox": [
+ 116,
+ 667,
+ 356,
+ 680
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 667,
+ 356,
+ 680
+ ],
+ "score": 1.0,
+ "content": "2018. URL http://arxiv.org/abs/1807.05960.",
+ "type": "text"
+ }
+ ],
+ "index": 42
+ }
+ ],
+ "index": 41,
+ "bbox_fs": [
+ 106,
+ 645,
+ 505,
+ 680
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 687,
+ 504,
+ 732
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 687,
+ 505,
+ 700
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 687,
+ 505,
+ 700
+ ],
+ "score": 1.0,
+ "content": "Franco Scarselli, Marco Gori, Ah Chung Tsoi, Markus Hagenbuchner, and Gabriele Monfardini.",
+ "type": "text"
+ }
+ ],
+ "index": 43
+ },
+ {
+ "bbox": [
+ 116,
+ 699,
+ 505,
+ 711
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 699,
+ 505,
+ 711
+ ],
+ "score": 1.0,
+ "content": "The graph neural network model. Trans. Neur. Netw., 20(1):61–80, January 2009. ISSN 1045-",
+ "type": "text"
+ }
+ ],
+ "index": 44
+ },
+ {
+ "bbox": [
+ 115,
+ 709,
+ 504,
+ 722
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 709,
+ 504,
+ 722
+ ],
+ "score": 1.0,
+ "content": "9227. doi: 10.1109/TNN.2008.2005605. URL http://dx.doi.org/10.1109/TNN.",
+ "type": "text"
+ }
+ ],
+ "index": 45
+ },
+ {
+ "bbox": [
+ 116,
+ 721,
+ 193,
+ 732
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 721,
+ 193,
+ 732
+ ],
+ "score": 1.0,
+ "content": "2008.2005605.",
+ "type": "text"
+ }
+ ],
+ "index": 46
+ }
+ ],
+ "index": 44.5,
+ "bbox_fs": [
+ 105,
+ 687,
+ 505,
+ 732
+ ]
+ }
+ ]
+ },
+ {
+ "preproc_blocks": [
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 82,
+ 505,
+ 126
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 82,
+ 505,
+ 94
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 82,
+ 505,
+ 94
+ ],
+ "score": 1.0,
+ "content": "N. Shervashidze, SVN. Vishwanathan, TH. Petri, K. Mehlhorn, and KM. Borgwardt. Efficient",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 115,
+ 93,
+ 505,
+ 106
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 93,
+ 505,
+ 106
+ ],
+ "score": 1.0,
+ "content": "graphlet kernels for large graph comparison. In JMLR Workshop and Conference Proceed-",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 115,
+ 104,
+ 505,
+ 117
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 104,
+ 505,
+ 117
+ ],
+ "score": 1.0,
+ "content": "ings Volume 5: AISTATS 2009, pp. 488–495, Cambridge, MA, USA, April 2009. Max-Planck-",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ },
+ {
+ "bbox": [
+ 115,
+ 114,
+ 218,
+ 127
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 114,
+ 218,
+ 127
+ ],
+ "score": 1.0,
+ "content": "Gesellschaft, MIT Press.",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ }
+ ],
+ "index": 1.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 134,
+ 505,
+ 178
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 134,
+ 505,
+ 146
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 134,
+ 505,
+ 146
+ ],
+ "score": 1.0,
+ "content": "Nino Shervashidze, Pascal Schweitzer, Erik Jan van Leeuwen, Kurt Mehlhorn, and Karsten M.",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 115,
+ 145,
+ 505,
+ 157
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 145,
+ 505,
+ 157
+ ],
+ "score": 1.0,
+ "content": "Borgwardt. Weisfeiler-lehman graph kernels. J. Mach. Learn. Res., 12:2539–2561, Novem-",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 115,
+ 156,
+ 504,
+ 168
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 156,
+ 449,
+ 168
+ ],
+ "score": 1.0,
+ "content": "ber 2011. ISSN 1532-4435. URL http://dl.acm.org/citation.cfm?id",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 449,
+ 157,
+ 456,
+ 166
+ ],
+ "score": 0.35,
+ "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 456,
+ 156,
+ 504,
+ 168
+ ],
+ "score": 1.0,
+ "content": "1953048.",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 115,
+ 167,
+ 164,
+ 178
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 167,
+ 164,
+ 178
+ ],
+ "score": 1.0,
+ "content": "2078187.",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ }
+ ],
+ "index": 5.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 186,
+ 505,
+ 220
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 185,
+ 505,
+ 199
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 185,
+ 505,
+ 199
+ ],
+ "score": 1.0,
+ "content": "Petar Velikovi, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Li, and Yoshua Ben-",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 115,
+ 197,
+ 505,
+ 210
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 197,
+ 505,
+ 210
+ ],
+ "score": 1.0,
+ "content": "gio. Graph attention networks. In International Conference on Learning Representations, 2018.",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 115,
+ 207,
+ 389,
+ 222
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 207,
+ 325,
+ 222
+ ],
+ "score": 1.0,
+ "content": "URL https://openreview.net/forum?id",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 325,
+ 209,
+ 331,
+ 218
+ ],
+ "score": 0.53,
+ "content": "=",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 332,
+ 207,
+ 389,
+ 222
+ ],
+ "score": 1.0,
+ "content": "rJXMpikCZ.",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ }
+ ],
+ "index": 9
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 226,
+ 503,
+ 249
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 225,
+ 505,
+ 241
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 225,
+ 505,
+ 241
+ ],
+ "score": 1.0,
+ "content": "B. Yu. Weisfeiler and A. A. Leman. Reduction of a graph to a canonical form and an algebra arising",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 116,
+ 238,
+ 230,
+ 250
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 238,
+ 230,
+ 250
+ ],
+ "score": 1.0,
+ "content": "during this reduction. 1968.",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ }
+ ],
+ "index": 11.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 256,
+ 503,
+ 280
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 255,
+ 505,
+ 270
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 255,
+ 505,
+ 270
+ ],
+ "score": 1.0,
+ "content": "Zonghan Wu, Shirui Pan, Fengwen Chen, Guodong Long, Chengqi Zhang, and Philip S. Yu. A",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 115,
+ 268,
+ 437,
+ 280
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 268,
+ 437,
+ 280
+ ],
+ "score": 1.0,
+ "content": "comprehensive survey on graph neural networks. CoRR, abs/1901.00596, 2019.",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ }
+ ],
+ "index": 13.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 286,
+ 504,
+ 309
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 286,
+ 505,
+ 300
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 286,
+ 505,
+ 300
+ ],
+ "score": 1.0,
+ "content": "Zhang Xinyi and Lihui Chen. Capsule graph neural network. In International Conference on Learn-",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 116,
+ 297,
+ 505,
+ 311
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 297,
+ 434,
+ 311
+ ],
+ "score": 1.0,
+ "content": "ing Representations, 2019. URL https://openreview.net/forum?id",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 435,
+ 299,
+ 441,
+ 307
+ ],
+ "score": 0.38,
+ "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 442,
+ 297,
+ 505,
+ 311
+ ],
+ "score": 1.0,
+ "content": "Byl8BnRcYm.",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ }
+ ],
+ "index": 15.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 316,
+ 505,
+ 350
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 316,
+ 505,
+ 329
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 316,
+ 505,
+ 329
+ ],
+ "score": 1.0,
+ "content": "Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 115,
+ 327,
+ 505,
+ 340
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 327,
+ 505,
+ 340
+ ],
+ "score": 1.0,
+ "content": "networks? In International Conference on Learning Representations, 2019. URL https:",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 117,
+ 338,
+ 337,
+ 351
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 117,
+ 338,
+ 266,
+ 351
+ ],
+ "score": 1.0,
+ "content": "//openreview.net/forum?id",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 267,
+ 340,
+ 273,
+ 348
+ ],
+ "score": 0.58,
+ "content": "=",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 273,
+ 338,
+ 337,
+ 351
+ ],
+ "score": 1.0,
+ "content": "ryGs6iA5Km.",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ }
+ ],
+ "index": 18
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 357,
+ 504,
+ 390
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 357,
+ 506,
+ 370
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 357,
+ 506,
+ 370
+ ],
+ "score": 1.0,
+ "content": "Shengzhong Zhang, Ziang Zhou, Zengfeng Huang, and Zhongyu Wei. Few-shot classification on",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 115,
+ 368,
+ 505,
+ 381
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 368,
+ 505,
+ 381
+ ],
+ "score": 1.0,
+ "content": "graphs with structural regularized GCNs, 2019. URL https://openreview.net/forum?",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 116,
+ 380,
+ 199,
+ 391
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 380,
+ 199,
+ 391
+ ],
+ "score": 1.0,
+ "content": "id=r1znKiAcY7.",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ }
+ ],
+ "index": 21
+ }
+ ],
+ "page_idx": 12,
+ "page_size": [
+ 612,
+ 792
+ ],
+ "discarded_blocks": [
+ {
+ "type": "discarded",
+ "bbox": [
+ 107,
+ 27,
+ 293,
+ 37
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "score": 1.0,
+ "content": "Published as a conference paper at ICLR 2020",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 300,
+ 751,
+ 311,
+ 760
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 299,
+ 750,
+ 313,
+ 764
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 299,
+ 750,
+ 313,
+ 764
+ ],
+ "score": 1.0,
+ "content": "13",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ }
+ ],
+ "para_blocks": [
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 82,
+ 505,
+ 126
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 82,
+ 505,
+ 94
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 82,
+ 505,
+ 94
+ ],
+ "score": 1.0,
+ "content": "N. Shervashidze, SVN. Vishwanathan, TH. Petri, K. Mehlhorn, and KM. Borgwardt. Efficient",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 115,
+ 93,
+ 505,
+ 106
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 93,
+ 505,
+ 106
+ ],
+ "score": 1.0,
+ "content": "graphlet kernels for large graph comparison. In JMLR Workshop and Conference Proceed-",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 115,
+ 104,
+ 505,
+ 117
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 104,
+ 505,
+ 117
+ ],
+ "score": 1.0,
+ "content": "ings Volume 5: AISTATS 2009, pp. 488–495, Cambridge, MA, USA, April 2009. Max-Planck-",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ },
+ {
+ "bbox": [
+ 115,
+ 114,
+ 218,
+ 127
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 114,
+ 218,
+ 127
+ ],
+ "score": 1.0,
+ "content": "Gesellschaft, MIT Press.",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ }
+ ],
+ "index": 1.5,
+ "bbox_fs": [
+ 105,
+ 82,
+ 505,
+ 127
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 134,
+ 505,
+ 178
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 134,
+ 505,
+ 146
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 134,
+ 505,
+ 146
+ ],
+ "score": 1.0,
+ "content": "Nino Shervashidze, Pascal Schweitzer, Erik Jan van Leeuwen, Kurt Mehlhorn, and Karsten M.",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 115,
+ 145,
+ 505,
+ 157
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 145,
+ 505,
+ 157
+ ],
+ "score": 1.0,
+ "content": "Borgwardt. Weisfeiler-lehman graph kernels. J. Mach. Learn. Res., 12:2539–2561, Novem-",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 115,
+ 156,
+ 504,
+ 168
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 156,
+ 449,
+ 168
+ ],
+ "score": 1.0,
+ "content": "ber 2011. ISSN 1532-4435. URL http://dl.acm.org/citation.cfm?id",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 449,
+ 157,
+ 456,
+ 166
+ ],
+ "score": 0.35,
+ "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } } =",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 456,
+ 156,
+ 504,
+ 168
+ ],
+ "score": 1.0,
+ "content": "1953048.",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 115,
+ 167,
+ 164,
+ 178
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 167,
+ 164,
+ 178
+ ],
+ "score": 1.0,
+ "content": "2078187.",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ }
+ ],
+ "index": 5.5,
+ "bbox_fs": [
+ 105,
+ 134,
+ 505,
+ 178
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 186,
+ 505,
+ 220
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 185,
+ 505,
+ 199
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 185,
+ 505,
+ 199
+ ],
+ "score": 1.0,
+ "content": "Petar Velikovi, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Li, and Yoshua Ben-",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 115,
+ 197,
+ 505,
+ 210
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 197,
+ 505,
+ 210
+ ],
+ "score": 1.0,
+ "content": "gio. Graph attention networks. In International Conference on Learning Representations, 2018.",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 115,
+ 207,
+ 389,
+ 222
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 207,
+ 325,
+ 222
+ ],
+ "score": 1.0,
+ "content": "URL https://openreview.net/forum?id",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 325,
+ 209,
+ 331,
+ 218
+ ],
+ "score": 0.53,
+ "content": "=",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 332,
+ 207,
+ 389,
+ 222
+ ],
+ "score": 1.0,
+ "content": "rJXMpikCZ.",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ }
+ ],
+ "index": 9,
+ "bbox_fs": [
+ 105,
+ 185,
+ 505,
+ 222
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 226,
+ 503,
+ 249
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 225,
+ 505,
+ 241
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 225,
+ 505,
+ 241
+ ],
+ "score": 1.0,
+ "content": "B. Yu. Weisfeiler and A. A. Leman. Reduction of a graph to a canonical form and an algebra arising",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 116,
+ 238,
+ 230,
+ 250
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 238,
+ 230,
+ 250
+ ],
+ "score": 1.0,
+ "content": "during this reduction. 1968.",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ }
+ ],
+ "index": 11.5,
+ "bbox_fs": [
+ 105,
+ 225,
+ 505,
+ 250
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 256,
+ 503,
+ 280
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 255,
+ 505,
+ 270
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 255,
+ 505,
+ 270
+ ],
+ "score": 1.0,
+ "content": "Zonghan Wu, Shirui Pan, Fengwen Chen, Guodong Long, Chengqi Zhang, and Philip S. Yu. A",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 115,
+ 268,
+ 437,
+ 280
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 268,
+ 437,
+ 280
+ ],
+ "score": 1.0,
+ "content": "comprehensive survey on graph neural networks. CoRR, abs/1901.00596, 2019.",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ }
+ ],
+ "index": 13.5,
+ "bbox_fs": [
+ 106,
+ 255,
+ 505,
+ 280
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 286,
+ 504,
+ 309
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 286,
+ 505,
+ 300
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 286,
+ 505,
+ 300
+ ],
+ "score": 1.0,
+ "content": "Zhang Xinyi and Lihui Chen. Capsule graph neural network. In International Conference on Learn-",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 116,
+ 297,
+ 505,
+ 311
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 297,
+ 434,
+ 311
+ ],
+ "score": 1.0,
+ "content": "ing Representations, 2019. URL https://openreview.net/forum?id",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 435,
+ 299,
+ 441,
+ 307
+ ],
+ "score": 0.38,
+ "content": "\\underline { { \\underline { { \\mathbf { \\Pi } } } } }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 442,
+ 297,
+ 505,
+ 311
+ ],
+ "score": 1.0,
+ "content": "Byl8BnRcYm.",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ }
+ ],
+ "index": 15.5,
+ "bbox_fs": [
+ 105,
+ 286,
+ 505,
+ 311
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 316,
+ 505,
+ 350
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 316,
+ 505,
+ 329
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 316,
+ 505,
+ 329
+ ],
+ "score": 1.0,
+ "content": "Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. How powerful are graph neural",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 115,
+ 327,
+ 505,
+ 340
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 327,
+ 505,
+ 340
+ ],
+ "score": 1.0,
+ "content": "networks? In International Conference on Learning Representations, 2019. URL https:",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 117,
+ 338,
+ 337,
+ 351
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 117,
+ 338,
+ 266,
+ 351
+ ],
+ "score": 1.0,
+ "content": "//openreview.net/forum?id",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 267,
+ 340,
+ 273,
+ 348
+ ],
+ "score": 0.58,
+ "content": "=",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 273,
+ 338,
+ 337,
+ 351
+ ],
+ "score": 1.0,
+ "content": "ryGs6iA5Km.",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ }
+ ],
+ "index": 18,
+ "bbox_fs": [
+ 106,
+ 316,
+ 505,
+ 351
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 357,
+ 504,
+ 390
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 357,
+ 506,
+ 370
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 357,
+ 506,
+ 370
+ ],
+ "score": 1.0,
+ "content": "Shengzhong Zhang, Ziang Zhou, Zengfeng Huang, and Zhongyu Wei. Few-shot classification on",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 115,
+ 368,
+ 505,
+ 381
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 115,
+ 368,
+ 505,
+ 381
+ ],
+ "score": 1.0,
+ "content": "graphs with structural regularized GCNs, 2019. URL https://openreview.net/forum?",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 116,
+ 380,
+ 199,
+ 391
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 116,
+ 380,
+ 199,
+ 391
+ ],
+ "score": 1.0,
+ "content": "id=r1znKiAcY7.",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ }
+ ],
+ "index": 21,
+ "bbox_fs": [
+ 106,
+ 357,
+ 506,
+ 391
+ ]
+ }
+ ]
+ },
+ {
+ "preproc_blocks": [
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 81,
+ 182,
+ 93
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 79,
+ 185,
+ 97
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 79,
+ 185,
+ 97
+ ],
+ "score": 1.0,
+ "content": "A APPENDIX",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ }
+ ],
+ "index": 0
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 110,
+ 212,
+ 121
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 110,
+ 213,
+ 122
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 110,
+ 213,
+ 122
+ ],
+ "score": 1.0,
+ "content": "A.1 DATASET DETAILS",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ }
+ ],
+ "index": 1
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 133,
+ 505,
+ 188
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 133,
+ 505,
+ 146
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 133,
+ 505,
+ 146
+ ],
+ "score": 1.0,
+ "content": "We use 4 different datasets namely - Reddit-12K, ENZYMES, Letter-High and TRIANGLES to per-",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ },
+ {
+ "bbox": [
+ 106,
+ 144,
+ 505,
+ 156
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 144,
+ 505,
+ 156
+ ],
+ "score": 1.0,
+ "content": "form exhaustive empirical evaluation of our model on various real-world datasets varying from small",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 105,
+ 155,
+ 505,
+ 168
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 155,
+ 505,
+ 168
+ ],
+ "score": 1.0,
+ "content": "average graph size on Letter-High to large graphs like Reddit-12K. These datasets can be down-",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 105,
+ 164,
+ 505,
+ 179
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 164,
+ 505,
+ 179
+ ],
+ "score": 1.0,
+ "content": "loaded here 4. The dataset statistics are provided in Table 6, while the split statistics are provided in",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 105,
+ 175,
+ 140,
+ 189
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 175,
+ 140,
+ 189
+ ],
+ "score": 1.0,
+ "content": "Table 7",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ }
+ ],
+ "index": 4
+ },
+ {
+ "type": "table",
+ "bbox": [
+ 153,
+ 234,
+ 457,
+ 304
+ ],
+ "blocks": [
+ {
+ "type": "table_caption",
+ "bbox": [
+ 253,
+ 203,
+ 358,
+ 215
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 253,
+ 203,
+ 358,
+ 215
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 253,
+ 203,
+ 358,
+ 215
+ ],
+ "score": 1.0,
+ "content": "Table 6: Dataset Statistics",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ }
+ ],
+ "index": 7
+ },
+ {
+ "type": "table_body",
+ "bbox": [
+ 153,
+ 234,
+ 457,
+ 304
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 153,
+ 234,
+ 457,
+ 304
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 153,
+ 234,
+ 457,
+ 304
+ ],
+ "score": 0.958,
+ "html": "| DatasetName | # Classes | # Graphs | Avg # Nodes | Avg #Edges |
| Reddit-12K | | 11929 | 391.41 | 456.89 |
| ENZYMES | 11 6 | 600 | 32.63 | 62.14 |
| Letter-High | 15 | 2250 | 4.67 | 4.50 |
| TRIANGLES | 10 | 45000 | 20.85 | 35.50 |
",
+ "type": "table",
+ "image_path": "7f2a202ef08fea6e3b49b3ef6526d917c6df6c3f47300c2899ddc24614d5646d.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 9,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 153,
+ 234,
+ 457,
+ 257.3333333333333
+ ],
+ "spans": [],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 153,
+ 257.3333333333333,
+ 457,
+ 280.66666666666663
+ ],
+ "spans": [],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 153,
+ 280.66666666666663,
+ 457,
+ 303.99999999999994
+ ],
+ "spans": [],
+ "index": 10
+ }
+ ]
+ }
+ ],
+ "index": 8.0
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 328,
+ 504,
+ 372
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 327,
+ 506,
+ 341
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 327,
+ 506,
+ 341
+ ],
+ "score": 1.0,
+ "content": "Dataset Description: Reddit-12K datasets contains 11929 graphs where each graph corresponds to",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 105,
+ 339,
+ 506,
+ 352
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 339,
+ 506,
+ 352
+ ],
+ "score": 1.0,
+ "content": "a thread in which each node represents a user and each edge represents that one user has responded",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 104,
+ 349,
+ 506,
+ 363
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 349,
+ 506,
+ 363
+ ],
+ "score": 1.0,
+ "content": "to a comment from some other user. There are 11 different types of discussion forums corresponding",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 105,
+ 360,
+ 208,
+ 373
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 360,
+ 208,
+ 373
+ ],
+ "score": 1.0,
+ "content": "to each of the 11 classes.",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ }
+ ],
+ "index": 12.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 108,
+ 372,
+ 503,
+ 405
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 372,
+ 505,
+ 384
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 372,
+ 505,
+ 384
+ ],
+ "score": 1.0,
+ "content": "ENZYMES is a dataset of protein tertiary structures consisting of 600 enzymes from the BRENDA",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 105,
+ 382,
+ 505,
+ 396
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 382,
+ 505,
+ 396
+ ],
+ "score": 1.0,
+ "content": "enzyme database. The dataset contains 6 different graph categories corresponding to each different",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 105,
+ 393,
+ 196,
+ 408
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 393,
+ 196,
+ 408
+ ],
+ "score": 1.0,
+ "content": "top-level EC enzyme.",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ }
+ ],
+ "index": 16
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 405,
+ 504,
+ 427
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 405,
+ 505,
+ 417
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 405,
+ 505,
+ 417
+ ],
+ "score": 1.0,
+ "content": "TRIANGLES dataset contain 10 different classes where the classes are numbered from 1 to 10",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 106,
+ 416,
+ 425,
+ 429
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 416,
+ 425,
+ 429
+ ],
+ "score": 1.0,
+ "content": "corresponding to the number of triangles/3-cliques in each graph of the dataset.",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ }
+ ],
+ "index": 18.5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 428,
+ 502,
+ 460
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 426,
+ 504,
+ 440
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 426,
+ 504,
+ 440
+ ],
+ "score": 1.0,
+ "content": "Letter-High dataset contains graphs which represent distorted letter drawings from the english al-",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 106,
+ 438,
+ 503,
+ 450
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 438,
+ 145,
+ 450
+ ],
+ "score": 1.0,
+ "content": "phabets -",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 145,
+ 438,
+ 323,
+ 450
+ ],
+ "score": 0.92,
+ "content": "A , E , F , H , I , K , L , M , N , T , V , W , X , Y , Z",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 324,
+ 438,
+ 503,
+ 450
+ ],
+ "score": 1.0,
+ "content": ". Each graph is a prototype manual construc-",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 106,
+ 449,
+ 192,
+ 461
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 449,
+ 192,
+ 461
+ ],
+ "score": 1.0,
+ "content": "tion of the alphabets.",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ }
+ ],
+ "index": 21
+ },
+ {
+ "type": "table",
+ "bbox": [
+ 133,
+ 507,
+ 478,
+ 589
+ ],
+ "blocks": [
+ {
+ "type": "table_caption",
+ "bbox": [
+ 259,
+ 477,
+ 352,
+ 489
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 258,
+ 476,
+ 352,
+ 490
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 258,
+ 476,
+ 352,
+ 490
+ ],
+ "score": 1.0,
+ "content": "Table 7: Dataset Splits",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ }
+ ],
+ "index": 23
+ },
+ {
+ "type": "table_body",
+ "bbox": [
+ 133,
+ 507,
+ 478,
+ 589
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 133,
+ 507,
+ 478,
+ 589
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 133,
+ 507,
+ 478,
+ 589
+ ],
+ "score": 0.977,
+ "html": "| Dataset Name | #Train Classes | #Test Classes | # Training Graphs | # Validation Graphs | #Test Graphs |
| | | | | |
| Reddit-12K | 7 | 4 | 566 | 141 | 404 |
| ENZYMES | 4 11 | 2 4 | 320 1330 | 80 320 | 200 600 |
| Letter-High TRIANGLES | 7 | 3 | 1126 | 271 | 603 |
| | | | | |
",
+ "type": "table",
+ "image_path": "6d74ff990d783464ac74a44d21f0162800cf54953df543089fc8a0b3e9c03ed8.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 25,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 133,
+ 507,
+ 478,
+ 534.3333333333334
+ ],
+ "spans": [],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 133,
+ 534.3333333333334,
+ 478,
+ 561.6666666666667
+ ],
+ "spans": [],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 133,
+ 561.6666666666667,
+ 478,
+ 589.0000000000001
+ ],
+ "spans": [],
+ "index": 26
+ }
+ ]
+ }
+ ],
+ "index": 24.0
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 613,
+ 505,
+ 702
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 612,
+ 505,
+ 626
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 612,
+ 505,
+ 626
+ ],
+ "score": 1.0,
+ "content": "The validation graphs are used to assess model performance on training classes itself to check over-",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 105,
+ 624,
+ 506,
+ 637
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 624,
+ 506,
+ 637
+ ],
+ "score": 1.0,
+ "content": "fitting as well as for grid-search over hyperparameters. The actual train-testing class splits used for",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 105,
+ 635,
+ 506,
+ 648
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 635,
+ 506,
+ 648
+ ],
+ "score": 1.0,
+ "content": "this paper are provided with the code. Since the TRIANGLES dataset has a large number of sam-",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 106,
+ 646,
+ 506,
+ 659
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 646,
+ 506,
+ 659
+ ],
+ "score": 1.0,
+ "content": "ples, this makes it infeasible to run many baselines including DL and non-DL methods. Hence, we",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 106,
+ 657,
+ 505,
+ 669
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 657,
+ 505,
+ 669
+ ],
+ "score": 1.0,
+ "content": "sample 200 graphs from each class, making the total sample size 2000. Similarly we downsample",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ },
+ {
+ "bbox": [
+ 105,
+ 667,
+ 506,
+ 681
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 667,
+ 506,
+ 681
+ ],
+ "score": 1.0,
+ "content": "the number of graphs from 11929 to 1111 (nearly 101 graphs per class). Downsampling is per-",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 105,
+ 678,
+ 505,
+ 692
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 678,
+ 505,
+ 692
+ ],
+ "score": 1.0,
+ "content": "formed for Reddit-12K given extremely large graph sizes which makes the graph kernels as well as",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ },
+ {
+ "bbox": [
+ 106,
+ 691,
+ 290,
+ 703
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 691,
+ 290,
+ 703
+ ],
+ "score": 1.0,
+ "content": "some deep learning baselines extremely slow.",
+ "type": "text"
+ }
+ ],
+ "index": 34
+ }
+ ],
+ "index": 30.5
+ }
+ ],
+ "page_idx": 13,
+ "page_size": [
+ 612,
+ 792
+ ],
+ "discarded_blocks": [
+ {
+ "type": "discarded",
+ "bbox": [
+ 118,
+ 721,
+ 371,
+ 732
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 118,
+ 719,
+ 372,
+ 734
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 118,
+ 719,
+ 372,
+ 734
+ ],
+ "score": 1.0,
+ "content": "4https://ls11-www.cs.tu-dortmund.de/staff/morris/graphkerneldatasets",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 107,
+ 27,
+ 293,
+ 37
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "score": 1.0,
+ "content": "Published as a conference paper at ICLR 2020",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 300,
+ 752,
+ 311,
+ 760
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 299,
+ 750,
+ 313,
+ 764
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 299,
+ 750,
+ 313,
+ 764
+ ],
+ "score": 1.0,
+ "content": "",
+ "type": "text",
+ "height": 14,
+ "width": 14
+ }
+ ]
+ }
+ ]
+ }
+ ],
+ "para_blocks": [
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 81,
+ 182,
+ 93
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 79,
+ 185,
+ 97
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 79,
+ 185,
+ 97
+ ],
+ "score": 1.0,
+ "content": "A APPENDIX",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ }
+ ],
+ "index": 0
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 110,
+ 212,
+ 121
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 110,
+ 213,
+ 122
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 110,
+ 213,
+ 122
+ ],
+ "score": 1.0,
+ "content": "A.1 DATASET DETAILS",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ }
+ ],
+ "index": 1
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 133,
+ 505,
+ 188
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 133,
+ 505,
+ 146
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 133,
+ 505,
+ 146
+ ],
+ "score": 1.0,
+ "content": "We use 4 different datasets namely - Reddit-12K, ENZYMES, Letter-High and TRIANGLES to per-",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ },
+ {
+ "bbox": [
+ 106,
+ 144,
+ 505,
+ 156
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 144,
+ 505,
+ 156
+ ],
+ "score": 1.0,
+ "content": "form exhaustive empirical evaluation of our model on various real-world datasets varying from small",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 105,
+ 155,
+ 505,
+ 168
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 155,
+ 505,
+ 168
+ ],
+ "score": 1.0,
+ "content": "average graph size on Letter-High to large graphs like Reddit-12K. These datasets can be down-",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 105,
+ 164,
+ 505,
+ 179
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 164,
+ 505,
+ 179
+ ],
+ "score": 1.0,
+ "content": "loaded here 4. The dataset statistics are provided in Table 6, while the split statistics are provided in",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 105,
+ 175,
+ 140,
+ 189
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 175,
+ 140,
+ 189
+ ],
+ "score": 1.0,
+ "content": "Table 7",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ }
+ ],
+ "index": 4,
+ "bbox_fs": [
+ 105,
+ 133,
+ 505,
+ 189
+ ]
+ },
+ {
+ "type": "table",
+ "bbox": [
+ 153,
+ 234,
+ 457,
+ 304
+ ],
+ "blocks": [
+ {
+ "type": "table_caption",
+ "bbox": [
+ 253,
+ 203,
+ 358,
+ 215
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 253,
+ 203,
+ 358,
+ 215
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 253,
+ 203,
+ 358,
+ 215
+ ],
+ "score": 1.0,
+ "content": "Table 6: Dataset Statistics",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ }
+ ],
+ "index": 7
+ },
+ {
+ "type": "table_body",
+ "bbox": [
+ 153,
+ 234,
+ 457,
+ 304
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 153,
+ 234,
+ 457,
+ 304
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 153,
+ 234,
+ 457,
+ 304
+ ],
+ "score": 0.958,
+ "html": "| DatasetName | # Classes | # Graphs | Avg # Nodes | Avg #Edges |
| Reddit-12K | | 11929 | 391.41 | 456.89 |
| ENZYMES | 11 6 | 600 | 32.63 | 62.14 |
| Letter-High | 15 | 2250 | 4.67 | 4.50 |
| TRIANGLES | 10 | 45000 | 20.85 | 35.50 |
",
+ "type": "table",
+ "image_path": "7f2a202ef08fea6e3b49b3ef6526d917c6df6c3f47300c2899ddc24614d5646d.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 9,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 153,
+ 234,
+ 457,
+ 257.3333333333333
+ ],
+ "spans": [],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 153,
+ 257.3333333333333,
+ 457,
+ 280.66666666666663
+ ],
+ "spans": [],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 153,
+ 280.66666666666663,
+ 457,
+ 303.99999999999994
+ ],
+ "spans": [],
+ "index": 10
+ }
+ ]
+ }
+ ],
+ "index": 8.0
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 328,
+ 504,
+ 372
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 327,
+ 506,
+ 341
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 327,
+ 506,
+ 341
+ ],
+ "score": 1.0,
+ "content": "Dataset Description: Reddit-12K datasets contains 11929 graphs where each graph corresponds to",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 105,
+ 339,
+ 506,
+ 352
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 339,
+ 506,
+ 352
+ ],
+ "score": 1.0,
+ "content": "a thread in which each node represents a user and each edge represents that one user has responded",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 104,
+ 349,
+ 506,
+ 363
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 349,
+ 506,
+ 363
+ ],
+ "score": 1.0,
+ "content": "to a comment from some other user. There are 11 different types of discussion forums corresponding",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 105,
+ 360,
+ 208,
+ 373
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 360,
+ 208,
+ 373
+ ],
+ "score": 1.0,
+ "content": "to each of the 11 classes.",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ }
+ ],
+ "index": 12.5,
+ "bbox_fs": [
+ 104,
+ 327,
+ 506,
+ 373
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 108,
+ 372,
+ 503,
+ 405
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 372,
+ 505,
+ 384
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 372,
+ 505,
+ 384
+ ],
+ "score": 1.0,
+ "content": "ENZYMES is a dataset of protein tertiary structures consisting of 600 enzymes from the BRENDA",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 105,
+ 382,
+ 505,
+ 396
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 382,
+ 505,
+ 396
+ ],
+ "score": 1.0,
+ "content": "enzyme database. The dataset contains 6 different graph categories corresponding to each different",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 105,
+ 393,
+ 196,
+ 408
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 393,
+ 196,
+ 408
+ ],
+ "score": 1.0,
+ "content": "top-level EC enzyme.",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ }
+ ],
+ "index": 16,
+ "bbox_fs": [
+ 105,
+ 372,
+ 505,
+ 408
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 405,
+ 504,
+ 427
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 405,
+ 505,
+ 417
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 405,
+ 505,
+ 417
+ ],
+ "score": 1.0,
+ "content": "TRIANGLES dataset contain 10 different classes where the classes are numbered from 1 to 10",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 106,
+ 416,
+ 425,
+ 429
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 416,
+ 425,
+ 429
+ ],
+ "score": 1.0,
+ "content": "corresponding to the number of triangles/3-cliques in each graph of the dataset.",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ }
+ ],
+ "index": 18.5,
+ "bbox_fs": [
+ 106,
+ 405,
+ 505,
+ 429
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 428,
+ 502,
+ 460
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 426,
+ 504,
+ 440
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 426,
+ 504,
+ 440
+ ],
+ "score": 1.0,
+ "content": "Letter-High dataset contains graphs which represent distorted letter drawings from the english al-",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 106,
+ 438,
+ 503,
+ 450
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 438,
+ 145,
+ 450
+ ],
+ "score": 1.0,
+ "content": "phabets -",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 145,
+ 438,
+ 323,
+ 450
+ ],
+ "score": 0.92,
+ "content": "A , E , F , H , I , K , L , M , N , T , V , W , X , Y , Z",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 324,
+ 438,
+ 503,
+ 450
+ ],
+ "score": 1.0,
+ "content": ". Each graph is a prototype manual construc-",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 106,
+ 449,
+ 192,
+ 461
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 449,
+ 192,
+ 461
+ ],
+ "score": 1.0,
+ "content": "tion of the alphabets.",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ }
+ ],
+ "index": 21,
+ "bbox_fs": [
+ 105,
+ 426,
+ 504,
+ 461
+ ]
+ },
+ {
+ "type": "table",
+ "bbox": [
+ 133,
+ 507,
+ 478,
+ 589
+ ],
+ "blocks": [
+ {
+ "type": "table_caption",
+ "bbox": [
+ 259,
+ 477,
+ 352,
+ 489
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 258,
+ 476,
+ 352,
+ 490
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 258,
+ 476,
+ 352,
+ 490
+ ],
+ "score": 1.0,
+ "content": "Table 7: Dataset Splits",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ }
+ ],
+ "index": 23
+ },
+ {
+ "type": "table_body",
+ "bbox": [
+ 133,
+ 507,
+ 478,
+ 589
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 133,
+ 507,
+ 478,
+ 589
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 133,
+ 507,
+ 478,
+ 589
+ ],
+ "score": 0.977,
+ "html": "| Dataset Name | #Train Classes | #Test Classes | # Training Graphs | # Validation Graphs | #Test Graphs |
| | | | | |
| Reddit-12K | 7 | 4 | 566 | 141 | 404 |
| ENZYMES | 4 11 | 2 4 | 320 1330 | 80 320 | 200 600 |
| Letter-High TRIANGLES | 7 | 3 | 1126 | 271 | 603 |
| | | | | |
",
+ "type": "table",
+ "image_path": "6d74ff990d783464ac74a44d21f0162800cf54953df543089fc8a0b3e9c03ed8.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 25,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 133,
+ 507,
+ 478,
+ 534.3333333333334
+ ],
+ "spans": [],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 133,
+ 534.3333333333334,
+ 478,
+ 561.6666666666667
+ ],
+ "spans": [],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 133,
+ 561.6666666666667,
+ 478,
+ 589.0000000000001
+ ],
+ "spans": [],
+ "index": 26
+ }
+ ]
+ }
+ ],
+ "index": 24.0
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 613,
+ 505,
+ 702
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 612,
+ 505,
+ 626
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 612,
+ 505,
+ 626
+ ],
+ "score": 1.0,
+ "content": "The validation graphs are used to assess model performance on training classes itself to check over-",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 105,
+ 624,
+ 506,
+ 637
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 624,
+ 506,
+ 637
+ ],
+ "score": 1.0,
+ "content": "fitting as well as for grid-search over hyperparameters. The actual train-testing class splits used for",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 105,
+ 635,
+ 506,
+ 648
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 635,
+ 506,
+ 648
+ ],
+ "score": 1.0,
+ "content": "this paper are provided with the code. Since the TRIANGLES dataset has a large number of sam-",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 106,
+ 646,
+ 506,
+ 659
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 646,
+ 506,
+ 659
+ ],
+ "score": 1.0,
+ "content": "ples, this makes it infeasible to run many baselines including DL and non-DL methods. Hence, we",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 106,
+ 657,
+ 505,
+ 669
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 657,
+ 505,
+ 669
+ ],
+ "score": 1.0,
+ "content": "sample 200 graphs from each class, making the total sample size 2000. Similarly we downsample",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ },
+ {
+ "bbox": [
+ 105,
+ 667,
+ 506,
+ 681
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 667,
+ 506,
+ 681
+ ],
+ "score": 1.0,
+ "content": "the number of graphs from 11929 to 1111 (nearly 101 graphs per class). Downsampling is per-",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 105,
+ 678,
+ 505,
+ 692
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 678,
+ 505,
+ 692
+ ],
+ "score": 1.0,
+ "content": "formed for Reddit-12K given extremely large graph sizes which makes the graph kernels as well as",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ },
+ {
+ "bbox": [
+ 106,
+ 691,
+ 290,
+ 703
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 691,
+ 290,
+ 703
+ ],
+ "score": 1.0,
+ "content": "some deep learning baselines extremely slow.",
+ "type": "text"
+ }
+ ],
+ "index": 34
+ }
+ ],
+ "index": 30.5,
+ "bbox_fs": [
+ 105,
+ 612,
+ 506,
+ 703
+ ]
+ }
+ ]
+ },
+ {
+ "preproc_blocks": [
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 82,
+ 217,
+ 93
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 81,
+ 219,
+ 95
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 81,
+ 219,
+ 95
+ ],
+ "score": 1.0,
+ "content": "A.2 BASELINE DETAILS",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ }
+ ],
+ "index": 0
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 105,
+ 505,
+ 205
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 105,
+ 505,
+ 117
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 105,
+ 505,
+ 117
+ ],
+ "score": 1.0,
+ "content": "This section details the implementation of the baseline methods. Since, DL-based methods - GIN,",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 105,
+ 116,
+ 505,
+ 129
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 116,
+ 505,
+ 129
+ ],
+ "score": 1.0,
+ "content": "CapsGNN and DIFFPOOL have not been previously run on these datasets, we select the crucial",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ },
+ {
+ "bbox": [
+ 105,
+ 128,
+ 505,
+ 139
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 128,
+ 505,
+ 139
+ ],
+ "score": 1.0,
+ "content": "hyper-parameters - such as number of layers heuristically based on the results of standard graph",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 105,
+ 137,
+ 505,
+ 151
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 137,
+ 505,
+ 151
+ ],
+ "score": 1.0,
+ "content": "classification datasets on the best performing variants of these models. For these three methods we",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 105,
+ 150,
+ 505,
+ 162
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 150,
+ 505,
+ 162
+ ],
+ "score": 1.0,
+ "content": "take the novel layers proposed in the corresponding papers as their feature extractors, while down-",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 105,
+ 160,
+ 505,
+ 173
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 160,
+ 505,
+ 173
+ ],
+ "score": 1.0,
+ "content": "stream MLP layers are chosen as the classifier. The training and evaluation strategies are similar to",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 105,
+ 172,
+ 505,
+ 183
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 172,
+ 467,
+ 183
+ ],
+ "score": 1.0,
+ "content": "our model, i.e., the models are first trained in an end-to-end fashion on the training dataset",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 467,
+ 172,
+ 483,
+ 182
+ ],
+ "score": 0.89,
+ "content": "G _ { B }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 483,
+ 172,
+ 505,
+ 183
+ ],
+ "score": 1.0,
+ "content": "until",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 105,
+ 182,
+ 505,
+ 195
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 182,
+ 505,
+ 195
+ ],
+ "score": 1.0,
+ "content": "convergence with learning rate decay on loss plateau and then the classifier layers are fine-tuned",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 105,
+ 193,
+ 386,
+ 206
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 193,
+ 126,
+ 206
+ ],
+ "score": 1.0,
+ "content": "over",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 127,
+ 194,
+ 143,
+ 204
+ ],
+ "score": 0.89,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 143,
+ 193,
+ 386,
+ 206
+ ],
+ "score": 1.0,
+ "content": ", keeping the parameters of the feature extractor layers fixed.",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ }
+ ],
+ "index": 5
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 210,
+ 505,
+ 342
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 210,
+ 505,
+ 222
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 210,
+ 505,
+ 222
+ ],
+ "score": 1.0,
+ "content": "For the unsupervised models - WL subtree kernel, Graphlet Count kernel, AWE and Graph2Vec, the",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 105,
+ 221,
+ 506,
+ 234
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 221,
+ 203,
+ 234
+ ],
+ "score": 1.0,
+ "content": "evaluation is done using",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 204,
+ 222,
+ 210,
+ 231
+ ],
+ "score": 0.78,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 210,
+ 221,
+ 506,
+ 234
+ ],
+ "score": 1.0,
+ "content": "-NN search to assess the clustering quality of these models in our few-shot",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 105,
+ 231,
+ 506,
+ 245
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 231,
+ 506,
+ 245
+ ],
+ "score": 1.0,
+ "content": "scenario. We refrain from using high-level classifier models such as SVM or MLPs, since training",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 105,
+ 243,
+ 505,
+ 255
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 243,
+ 505,
+ 255
+ ],
+ "score": 1.0,
+ "content": "these classifiers on few-shot regime will not properly assess the abilities of these models to cluster",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 105,
+ 254,
+ 505,
+ 266
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 254,
+ 505,
+ 266
+ ],
+ "score": 1.0,
+ "content": "together graphs of similar class labels. We empirically found that using high level classifiers resulted",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 105,
+ 265,
+ 505,
+ 277
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 265,
+ 505,
+ 277
+ ],
+ "score": 1.0,
+ "content": "in higher deviations and lower mean accuracies. We choose the hyper-parameters for these models",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 105,
+ 276,
+ 505,
+ 288
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 276,
+ 505,
+ 288
+ ],
+ "score": 1.0,
+ "content": "using grid-search, since they are significantly faster and each one of these models have few highly",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 105,
+ 287,
+ 505,
+ 299
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 287,
+ 505,
+ 299
+ ],
+ "score": 1.0,
+ "content": "sensitive parameters which affect the model significantly. For these models, we perform a grid",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 105,
+ 298,
+ 505,
+ 310
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 298,
+ 199,
+ 310
+ ],
+ "score": 1.0,
+ "content": "search for selection of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 199,
+ 298,
+ 206,
+ 308
+ ],
+ "score": 0.82,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 207,
+ 298,
+ 234,
+ 310
+ ],
+ "score": 1.0,
+ "content": "in the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 234,
+ 298,
+ 240,
+ 308
+ ],
+ "score": 0.79,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 241,
+ 298,
+ 354,
+ 310
+ ],
+ "score": 1.0,
+ "content": "-NN algorithm from the set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 354,
+ 298,
+ 407,
+ 309
+ ],
+ "score": 0.93,
+ "content": "\\{ 1 , 2 , 3 , 4 , 5 \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 408,
+ 298,
+ 505,
+ 310
+ ],
+ "score": 1.0,
+ "content": "for the 5-shot scenario,",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 105,
+ 309,
+ 505,
+ 321
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 309,
+ 144,
+ 321
+ ],
+ "score": 1.0,
+ "content": "of which",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 144,
+ 309,
+ 170,
+ 319
+ ],
+ "score": 0.9,
+ "content": "k = 1",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 171,
+ 309,
+ 505,
+ 321
+ ],
+ "score": 1.0,
+ "content": "was found to perform the best. For higher shot scenario, the search was performed",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 105,
+ 320,
+ 505,
+ 332
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 320,
+ 155,
+ 332
+ ],
+ "score": 1.0,
+ "content": "over the set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 156,
+ 320,
+ 261,
+ 332
+ ],
+ "score": 0.91,
+ "content": "\\{ 1 , 2 , 3 , 4 , 5 , 6 , 7 , \\bar { 8 } , 9 , 1 0 \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 261,
+ 320,
+ 357,
+ 332
+ ],
+ "score": 1.0,
+ "content": ", where we again found",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 357,
+ 320,
+ 383,
+ 330
+ ],
+ "score": 0.9,
+ "content": "k = 1",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 384,
+ 320,
+ 505,
+ 332
+ ],
+ "score": 1.0,
+ "content": "to be the best. The validation",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 105,
+ 331,
+ 453,
+ 343
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 331,
+ 453,
+ 343
+ ],
+ "score": 1.0,
+ "content": "set is used to check overfitting and hyper-parameter selection on the baseline methods.",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ }
+ ],
+ "index": 15.5
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 362,
+ 228,
+ 373
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 361,
+ 230,
+ 375
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 361,
+ 230,
+ 375
+ ],
+ "score": 1.0,
+ "content": "A.3 OUR MODEL DETAILS",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ }
+ ],
+ "index": 22
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 385,
+ 505,
+ 506
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 385,
+ 505,
+ 398
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 385,
+ 505,
+ 398
+ ],
+ "score": 1.0,
+ "content": "This section provides the implementation details of our proposed model. Since, our feature extractor",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ },
+ {
+ "bbox": [
+ 106,
+ 396,
+ 505,
+ 409
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 396,
+ 505,
+ 409
+ ],
+ "score": 1.0,
+ "content": "model is GIN, we maintain similar parameter settings as recommended by their paper. As mentioned",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 105,
+ 407,
+ 505,
+ 420
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 407,
+ 505,
+ 420
+ ],
+ "score": 1.0,
+ "content": "in section 4.2, using embeddings from all iterations of the message passing network helps achieve",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 105,
+ 418,
+ 505,
+ 431
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 418,
+ 505,
+ 431
+ ],
+ "score": 1.0,
+ "content": "better discriminative power and improved gradient flow, therefore we employ the same strategy in",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 104,
+ 428,
+ 506,
+ 443
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 428,
+ 425,
+ 443
+ ],
+ "score": 1.0,
+ "content": "our feature extractor. The number of super-classes are selected from the set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 425,
+ 429,
+ 479,
+ 441
+ ],
+ "score": 0.91,
+ "content": "\\{ 1 , 2 , 3 , 4 , 5 \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 479,
+ 428,
+ 506,
+ 443
+ ],
+ "score": 1.0,
+ "content": "using",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 106,
+ 440,
+ 506,
+ 452
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 440,
+ 175,
+ 452
+ ],
+ "score": 1.0,
+ "content": "grid-search. The",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 175,
+ 441,
+ 182,
+ 450
+ ],
+ "score": 0.82,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 182,
+ 440,
+ 438,
+ 452
+ ],
+ "score": 1.0,
+ "content": "-value for construction of super-graph was selected from the set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 438,
+ 441,
+ 482,
+ 452
+ ],
+ "score": 0.9,
+ "content": "\\{ 2 , 4 , 6 , 8 \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 483,
+ 440,
+ 506,
+ 452
+ ],
+ "score": 1.0,
+ "content": ". The",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 105,
+ 450,
+ 506,
+ 464
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 450,
+ 506,
+ 464
+ ],
+ "score": 1.0,
+ "content": "feature extractor model uses batch-normalization between subsequent message passing layers. We",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 104,
+ 459,
+ 506,
+ 476
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 459,
+ 209,
+ 476
+ ],
+ "score": 1.0,
+ "content": "use dropout of 0.5 in the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 209,
+ 462,
+ 231,
+ 472
+ ],
+ "score": 0.89,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 231,
+ 459,
+ 281,
+ 476
+ ],
+ "score": 1.0,
+ "content": "layers. The",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 282,
+ 461,
+ 309,
+ 472
+ ],
+ "score": 0.9,
+ "content": "C ^ { G A T }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 309,
+ 459,
+ 506,
+ 476
+ ],
+ "score": 1.0,
+ "content": "layers undergo normalization of inputs between",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 106,
+ 473,
+ 505,
+ 485
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 473,
+ 505,
+ 485
+ ],
+ "score": 1.0,
+ "content": "subsequent layers along with a dropout of 0.5, however, the normalization mechanism in classifier",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ },
+ {
+ "bbox": [
+ 105,
+ 483,
+ 505,
+ 497
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 483,
+ 505,
+ 497
+ ],
+ "score": 1.0,
+ "content": "layers is different from batch-norm. We normalize each feature embedding to have Euclidean norm",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 106,
+ 495,
+ 209,
+ 507
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 495,
+ 209,
+ 507
+ ],
+ "score": 1.0,
+ "content": "with value 1. Essentially,",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ }
+ ],
+ "index": 28
+ },
+ {
+ "type": "interline_equation",
+ "bbox": [
+ 267,
+ 513,
+ 344,
+ 543
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 267,
+ 513,
+ 344,
+ 543
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 267,
+ 513,
+ 344,
+ 543
+ ],
+ "score": 0.94,
+ "content": "\\mathbf { x } _ { i n p u t } ^ { j + 1 } = \\frac { \\mathbf { x } _ { o u t } ^ { j } } { | | \\mathbf { x } _ { o u t } ^ { j } | | _ { 2 } }",
+ "type": "interline_equation",
+ "image_path": "3ae89ef6953aa96bd243b83d2892c1e11485b0e15e9c2d90ac6788465304f328.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 34.5,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 267,
+ 513,
+ 344,
+ 528.0
+ ],
+ "spans": [],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 267,
+ 528.0,
+ 344,
+ 543.0
+ ],
+ "spans": [],
+ "index": 35
+ }
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 554,
+ 505,
+ 701
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 103,
+ 554,
+ 507,
+ 573
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 103,
+ 554,
+ 133,
+ 573
+ ],
+ "score": 1.0,
+ "content": "where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 133,
+ 555,
+ 159,
+ 570
+ ],
+ "score": 0.93,
+ "content": "\\mathbf { x } _ { i n p u t } ^ { j + 1 }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 160,
+ 554,
+ 218,
+ 573
+ ],
+ "score": 1.0,
+ "content": "is the input of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 218,
+ 555,
+ 249,
+ 568
+ ],
+ "score": 0.92,
+ "content": "j + 1 ^ { t h }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 249,
+ 554,
+ 323,
+ 573
+ ],
+ "score": 1.0,
+ "content": "layer of classifier,",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 323,
+ 554,
+ 342,
+ 569
+ ],
+ "score": 0.92,
+ "content": "\\mathbf { x } _ { o u t } ^ { j }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 342,
+ 554,
+ 420,
+ 573
+ ],
+ "score": 1.0,
+ "content": "is the output of the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 420,
+ 555,
+ 434,
+ 568
+ ],
+ "score": 0.9,
+ "content": "j ^ { t h }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 434,
+ 554,
+ 507,
+ 573
+ ],
+ "score": 1.0,
+ "content": "layer. The inputs",
+ "type": "text"
+ }
+ ],
+ "index": 36
+ },
+ {
+ "bbox": [
+ 104,
+ 566,
+ 506,
+ 583
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 566,
+ 189,
+ 583
+ ],
+ "score": 1.0,
+ "content": "of the first layer of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 189,
+ 568,
+ 216,
+ 580
+ ],
+ "score": 0.9,
+ "content": "C ^ { G A T }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 217,
+ 566,
+ 506,
+ 583
+ ],
+ "score": 1.0,
+ "content": "also undergo the same transformation over the outputs of the feature",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 105,
+ 581,
+ 505,
+ 593
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 581,
+ 505,
+ 593
+ ],
+ "score": 1.0,
+ "content": "extractor model. We train our models with Adam (Kingma & Ba (2014)) with an initial learning rate",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ },
+ {
+ "bbox": [
+ 106,
+ 591,
+ 505,
+ 604
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 591,
+ 117,
+ 604
+ ],
+ "score": 1.0,
+ "content": "of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 118,
+ 591,
+ 139,
+ 602
+ ],
+ "score": 0.89,
+ "content": "1 0 ^ { - 3 }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 140,
+ 591,
+ 505,
+ 604
+ ],
+ "score": 1.0,
+ "content": "for 50 epochs. Each epoch has 10 iterations, where we randomly select a mini-batch from",
+ "type": "text"
+ }
+ ],
+ "index": 39
+ },
+ {
+ "bbox": [
+ 106,
+ 603,
+ 505,
+ 615
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 603,
+ 174,
+ 615
+ ],
+ "score": 1.0,
+ "content": "the training data",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 174,
+ 603,
+ 190,
+ 614
+ ],
+ "score": 0.88,
+ "content": "G _ { B }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 190,
+ 603,
+ 505,
+ 615
+ ],
+ "score": 1.0,
+ "content": ". The fine-tuning stage consists of 20 epochs with 10 iterations per epoch. We",
+ "type": "text"
+ }
+ ],
+ "index": 40
+ },
+ {
+ "bbox": [
+ 104,
+ 610,
+ 506,
+ 627
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 610,
+ 321,
+ 627
+ ],
+ "score": 1.0,
+ "content": "use a two-layer MLP over the final attention layer of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 321,
+ 612,
+ 348,
+ 624
+ ],
+ "score": 0.91,
+ "content": "C ^ { G A T ^ { \\bf 1 } }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 349,
+ 610,
+ 506,
+ 627
+ ],
+ "score": 1.0,
+ "content": "for classification. The attention layers",
+ "type": "text"
+ }
+ ],
+ "index": 41
+ },
+ {
+ "bbox": [
+ 106,
+ 624,
+ 505,
+ 637
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 624,
+ 505,
+ 637
+ ],
+ "score": 1.0,
+ "content": "use multi-head attention with 2 heads and leaky ReLU slope of 0.1 . The embeddings from both the",
+ "type": "text"
+ }
+ ],
+ "index": 42
+ },
+ {
+ "bbox": [
+ 106,
+ 636,
+ 505,
+ 647
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 636,
+ 320,
+ 647
+ ],
+ "score": 1.0,
+ "content": "attention heads are concatenated. For 20-shot, we set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 320,
+ 636,
+ 327,
+ 645
+ ],
+ "score": 0.79,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 327,
+ 636,
+ 505,
+ 647
+ ],
+ "score": 1.0,
+ "content": "to 2, number of super-classes to 3 and batch",
+ "type": "text"
+ }
+ ],
+ "index": 43
+ },
+ {
+ "bbox": [
+ 106,
+ 647,
+ 505,
+ 658
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 647,
+ 290,
+ 658
+ ],
+ "score": 1.0,
+ "content": "size to 128 on the Letter-High dataset, while",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 290,
+ 647,
+ 297,
+ 656
+ ],
+ "score": 0.81,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 298,
+ 647,
+ 505,
+ 658
+ ],
+ "score": 1.0,
+ "content": "is set to 2 and batch size 64 on Reddit, ENZYMES",
+ "type": "text"
+ }
+ ],
+ "index": 44
+ },
+ {
+ "bbox": [
+ 106,
+ 657,
+ 506,
+ 669
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 657,
+ 506,
+ 669
+ ],
+ "score": 1.0,
+ "content": "and TRIANGLES datasets. The number of super-classes for Reddit are set to 2, for ENZYMES it",
+ "type": "text"
+ }
+ ],
+ "index": 45
+ },
+ {
+ "bbox": [
+ 106,
+ 668,
+ 505,
+ 681
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 668,
+ 505,
+ 681
+ ],
+ "score": 1.0,
+ "content": "is set to 1 and for TRIANGLES are 3. For ENZYMES, there are negligible differences on using 1",
+ "type": "text"
+ }
+ ],
+ "index": 46
+ },
+ {
+ "bbox": [
+ 105,
+ 678,
+ 506,
+ 692
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 678,
+ 506,
+ 692
+ ],
+ "score": 1.0,
+ "content": "and 2 super-classes as shown in table 4. We used Python Optimal Transport (POT) library 5 for",
+ "type": "text"
+ }
+ ],
+ "index": 47
+ },
+ {
+ "bbox": [
+ 106,
+ 690,
+ 303,
+ 703
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 690,
+ 197,
+ 703
+ ],
+ "score": 1.0,
+ "content": "implementation of the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 197,
+ 691,
+ 203,
+ 702
+ ],
+ "score": 0.79,
+ "content": "p",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 204,
+ 690,
+ 303,
+ 703
+ ],
+ "score": 1.0,
+ "content": "-th Wasserstein distance.",
+ "type": "text"
+ }
+ ],
+ "index": 48
+ }
+ ],
+ "index": 42
+ }
+ ],
+ "page_idx": 14,
+ "page_size": [
+ 612,
+ 792
+ ],
+ "discarded_blocks": [
+ {
+ "type": "discarded",
+ "bbox": [
+ 119,
+ 721,
+ 279,
+ 731
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 119,
+ 720,
+ 280,
+ 733
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 119,
+ 720,
+ 280,
+ 733
+ ],
+ "score": 1.0,
+ "content": "5https://pot.readthedocs.io/en/stable/all.html",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 108,
+ 27,
+ 293,
+ 37
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "score": 1.0,
+ "content": "Published as a conference paper at ICLR 2020",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 300,
+ 751,
+ 311,
+ 760
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 299,
+ 750,
+ 312,
+ 764
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 299,
+ 750,
+ 312,
+ 764
+ ],
+ "score": 1.0,
+ "content": "15",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ }
+ ],
+ "para_blocks": [
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 82,
+ 217,
+ 93
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 81,
+ 219,
+ 95
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 81,
+ 219,
+ 95
+ ],
+ "score": 1.0,
+ "content": "A.2 BASELINE DETAILS",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ }
+ ],
+ "index": 0
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 105,
+ 505,
+ 205
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 105,
+ 505,
+ 117
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 105,
+ 505,
+ 117
+ ],
+ "score": 1.0,
+ "content": "This section details the implementation of the baseline methods. Since, DL-based methods - GIN,",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 105,
+ 116,
+ 505,
+ 129
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 116,
+ 505,
+ 129
+ ],
+ "score": 1.0,
+ "content": "CapsGNN and DIFFPOOL have not been previously run on these datasets, we select the crucial",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ },
+ {
+ "bbox": [
+ 105,
+ 128,
+ 505,
+ 139
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 128,
+ 505,
+ 139
+ ],
+ "score": 1.0,
+ "content": "hyper-parameters - such as number of layers heuristically based on the results of standard graph",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 105,
+ 137,
+ 505,
+ 151
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 137,
+ 505,
+ 151
+ ],
+ "score": 1.0,
+ "content": "classification datasets on the best performing variants of these models. For these three methods we",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 105,
+ 150,
+ 505,
+ 162
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 150,
+ 505,
+ 162
+ ],
+ "score": 1.0,
+ "content": "take the novel layers proposed in the corresponding papers as their feature extractors, while down-",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 105,
+ 160,
+ 505,
+ 173
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 160,
+ 505,
+ 173
+ ],
+ "score": 1.0,
+ "content": "stream MLP layers are chosen as the classifier. The training and evaluation strategies are similar to",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 105,
+ 172,
+ 505,
+ 183
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 172,
+ 467,
+ 183
+ ],
+ "score": 1.0,
+ "content": "our model, i.e., the models are first trained in an end-to-end fashion on the training dataset",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 467,
+ 172,
+ 483,
+ 182
+ ],
+ "score": 0.89,
+ "content": "G _ { B }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 483,
+ 172,
+ 505,
+ 183
+ ],
+ "score": 1.0,
+ "content": "until",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 105,
+ 182,
+ 505,
+ 195
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 182,
+ 505,
+ 195
+ ],
+ "score": 1.0,
+ "content": "convergence with learning rate decay on loss plateau and then the classifier layers are fine-tuned",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 105,
+ 193,
+ 386,
+ 206
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 193,
+ 126,
+ 206
+ ],
+ "score": 1.0,
+ "content": "over",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 127,
+ 194,
+ 143,
+ 204
+ ],
+ "score": 0.89,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 143,
+ 193,
+ 386,
+ 206
+ ],
+ "score": 1.0,
+ "content": ", keeping the parameters of the feature extractor layers fixed.",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ }
+ ],
+ "index": 5,
+ "bbox_fs": [
+ 105,
+ 105,
+ 505,
+ 206
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 210,
+ 505,
+ 342
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 210,
+ 505,
+ 222
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 210,
+ 505,
+ 222
+ ],
+ "score": 1.0,
+ "content": "For the unsupervised models - WL subtree kernel, Graphlet Count kernel, AWE and Graph2Vec, the",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 105,
+ 221,
+ 506,
+ 234
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 221,
+ 203,
+ 234
+ ],
+ "score": 1.0,
+ "content": "evaluation is done using",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 204,
+ 222,
+ 210,
+ 231
+ ],
+ "score": 0.78,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 210,
+ 221,
+ 506,
+ 234
+ ],
+ "score": 1.0,
+ "content": "-NN search to assess the clustering quality of these models in our few-shot",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 105,
+ 231,
+ 506,
+ 245
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 231,
+ 506,
+ 245
+ ],
+ "score": 1.0,
+ "content": "scenario. We refrain from using high-level classifier models such as SVM or MLPs, since training",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 105,
+ 243,
+ 505,
+ 255
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 243,
+ 505,
+ 255
+ ],
+ "score": 1.0,
+ "content": "these classifiers on few-shot regime will not properly assess the abilities of these models to cluster",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 105,
+ 254,
+ 505,
+ 266
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 254,
+ 505,
+ 266
+ ],
+ "score": 1.0,
+ "content": "together graphs of similar class labels. We empirically found that using high level classifiers resulted",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 105,
+ 265,
+ 505,
+ 277
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 265,
+ 505,
+ 277
+ ],
+ "score": 1.0,
+ "content": "in higher deviations and lower mean accuracies. We choose the hyper-parameters for these models",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 105,
+ 276,
+ 505,
+ 288
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 276,
+ 505,
+ 288
+ ],
+ "score": 1.0,
+ "content": "using grid-search, since they are significantly faster and each one of these models have few highly",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 105,
+ 287,
+ 505,
+ 299
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 287,
+ 505,
+ 299
+ ],
+ "score": 1.0,
+ "content": "sensitive parameters which affect the model significantly. For these models, we perform a grid",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 105,
+ 298,
+ 505,
+ 310
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 298,
+ 199,
+ 310
+ ],
+ "score": 1.0,
+ "content": "search for selection of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 199,
+ 298,
+ 206,
+ 308
+ ],
+ "score": 0.82,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 207,
+ 298,
+ 234,
+ 310
+ ],
+ "score": 1.0,
+ "content": "in the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 234,
+ 298,
+ 240,
+ 308
+ ],
+ "score": 0.79,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 241,
+ 298,
+ 354,
+ 310
+ ],
+ "score": 1.0,
+ "content": "-NN algorithm from the set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 354,
+ 298,
+ 407,
+ 309
+ ],
+ "score": 0.93,
+ "content": "\\{ 1 , 2 , 3 , 4 , 5 \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 408,
+ 298,
+ 505,
+ 310
+ ],
+ "score": 1.0,
+ "content": "for the 5-shot scenario,",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 105,
+ 309,
+ 505,
+ 321
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 309,
+ 144,
+ 321
+ ],
+ "score": 1.0,
+ "content": "of which",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 144,
+ 309,
+ 170,
+ 319
+ ],
+ "score": 0.9,
+ "content": "k = 1",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 171,
+ 309,
+ 505,
+ 321
+ ],
+ "score": 1.0,
+ "content": "was found to perform the best. For higher shot scenario, the search was performed",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 105,
+ 320,
+ 505,
+ 332
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 320,
+ 155,
+ 332
+ ],
+ "score": 1.0,
+ "content": "over the set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 156,
+ 320,
+ 261,
+ 332
+ ],
+ "score": 0.91,
+ "content": "\\{ 1 , 2 , 3 , 4 , 5 , 6 , 7 , \\bar { 8 } , 9 , 1 0 \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 261,
+ 320,
+ 357,
+ 332
+ ],
+ "score": 1.0,
+ "content": ", where we again found",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 357,
+ 320,
+ 383,
+ 330
+ ],
+ "score": 0.9,
+ "content": "k = 1",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 384,
+ 320,
+ 505,
+ 332
+ ],
+ "score": 1.0,
+ "content": "to be the best. The validation",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 105,
+ 331,
+ 453,
+ 343
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 331,
+ 453,
+ 343
+ ],
+ "score": 1.0,
+ "content": "set is used to check overfitting and hyper-parameter selection on the baseline methods.",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ }
+ ],
+ "index": 15.5,
+ "bbox_fs": [
+ 105,
+ 210,
+ 506,
+ 343
+ ]
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 362,
+ 228,
+ 373
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 361,
+ 230,
+ 375
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 361,
+ 230,
+ 375
+ ],
+ "score": 1.0,
+ "content": "A.3 OUR MODEL DETAILS",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ }
+ ],
+ "index": 22
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 385,
+ 505,
+ 506
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 385,
+ 505,
+ 398
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 385,
+ 505,
+ 398
+ ],
+ "score": 1.0,
+ "content": "This section provides the implementation details of our proposed model. Since, our feature extractor",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ },
+ {
+ "bbox": [
+ 106,
+ 396,
+ 505,
+ 409
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 396,
+ 505,
+ 409
+ ],
+ "score": 1.0,
+ "content": "model is GIN, we maintain similar parameter settings as recommended by their paper. As mentioned",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 105,
+ 407,
+ 505,
+ 420
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 407,
+ 505,
+ 420
+ ],
+ "score": 1.0,
+ "content": "in section 4.2, using embeddings from all iterations of the message passing network helps achieve",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 105,
+ 418,
+ 505,
+ 431
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 418,
+ 505,
+ 431
+ ],
+ "score": 1.0,
+ "content": "better discriminative power and improved gradient flow, therefore we employ the same strategy in",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 104,
+ 428,
+ 506,
+ 443
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 428,
+ 425,
+ 443
+ ],
+ "score": 1.0,
+ "content": "our feature extractor. The number of super-classes are selected from the set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 425,
+ 429,
+ 479,
+ 441
+ ],
+ "score": 0.91,
+ "content": "\\{ 1 , 2 , 3 , 4 , 5 \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 479,
+ 428,
+ 506,
+ 443
+ ],
+ "score": 1.0,
+ "content": "using",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 106,
+ 440,
+ 506,
+ 452
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 440,
+ 175,
+ 452
+ ],
+ "score": 1.0,
+ "content": "grid-search. The",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 175,
+ 441,
+ 182,
+ 450
+ ],
+ "score": 0.82,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 182,
+ 440,
+ 438,
+ 452
+ ],
+ "score": 1.0,
+ "content": "-value for construction of super-graph was selected from the set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 438,
+ 441,
+ 482,
+ 452
+ ],
+ "score": 0.9,
+ "content": "\\{ 2 , 4 , 6 , 8 \\}",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 483,
+ 440,
+ 506,
+ 452
+ ],
+ "score": 1.0,
+ "content": ". The",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 105,
+ 450,
+ 506,
+ 464
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 450,
+ 506,
+ 464
+ ],
+ "score": 1.0,
+ "content": "feature extractor model uses batch-normalization between subsequent message passing layers. We",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 104,
+ 459,
+ 506,
+ 476
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 459,
+ 209,
+ 476
+ ],
+ "score": 1.0,
+ "content": "use dropout of 0.5 in the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 209,
+ 462,
+ 231,
+ 472
+ ],
+ "score": 0.89,
+ "content": "C ^ { s u p }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 231,
+ 459,
+ 281,
+ 476
+ ],
+ "score": 1.0,
+ "content": "layers. The",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 282,
+ 461,
+ 309,
+ 472
+ ],
+ "score": 0.9,
+ "content": "C ^ { G A T }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 309,
+ 459,
+ 506,
+ 476
+ ],
+ "score": 1.0,
+ "content": "layers undergo normalization of inputs between",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ },
+ {
+ "bbox": [
+ 106,
+ 473,
+ 505,
+ 485
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 473,
+ 505,
+ 485
+ ],
+ "score": 1.0,
+ "content": "subsequent layers along with a dropout of 0.5, however, the normalization mechanism in classifier",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ },
+ {
+ "bbox": [
+ 105,
+ 483,
+ 505,
+ 497
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 483,
+ 505,
+ 497
+ ],
+ "score": 1.0,
+ "content": "layers is different from batch-norm. We normalize each feature embedding to have Euclidean norm",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 106,
+ 495,
+ 209,
+ 507
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 495,
+ 209,
+ 507
+ ],
+ "score": 1.0,
+ "content": "with value 1. Essentially,",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ }
+ ],
+ "index": 28,
+ "bbox_fs": [
+ 104,
+ 385,
+ 506,
+ 507
+ ]
+ },
+ {
+ "type": "interline_equation",
+ "bbox": [
+ 267,
+ 513,
+ 344,
+ 543
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 267,
+ 513,
+ 344,
+ 543
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 267,
+ 513,
+ 344,
+ 543
+ ],
+ "score": 0.94,
+ "content": "\\mathbf { x } _ { i n p u t } ^ { j + 1 } = \\frac { \\mathbf { x } _ { o u t } ^ { j } } { | | \\mathbf { x } _ { o u t } ^ { j } | | _ { 2 } }",
+ "type": "interline_equation",
+ "image_path": "3ae89ef6953aa96bd243b83d2892c1e11485b0e15e9c2d90ac6788465304f328.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 34.5,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 267,
+ 513,
+ 344,
+ 528.0
+ ],
+ "spans": [],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 267,
+ 528.0,
+ 344,
+ 543.0
+ ],
+ "spans": [],
+ "index": 35
+ }
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 554,
+ 505,
+ 701
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 103,
+ 554,
+ 507,
+ 573
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 103,
+ 554,
+ 133,
+ 573
+ ],
+ "score": 1.0,
+ "content": "where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 133,
+ 555,
+ 159,
+ 570
+ ],
+ "score": 0.93,
+ "content": "\\mathbf { x } _ { i n p u t } ^ { j + 1 }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 160,
+ 554,
+ 218,
+ 573
+ ],
+ "score": 1.0,
+ "content": "is the input of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 218,
+ 555,
+ 249,
+ 568
+ ],
+ "score": 0.92,
+ "content": "j + 1 ^ { t h }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 249,
+ 554,
+ 323,
+ 573
+ ],
+ "score": 1.0,
+ "content": "layer of classifier,",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 323,
+ 554,
+ 342,
+ 569
+ ],
+ "score": 0.92,
+ "content": "\\mathbf { x } _ { o u t } ^ { j }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 342,
+ 554,
+ 420,
+ 573
+ ],
+ "score": 1.0,
+ "content": "is the output of the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 420,
+ 555,
+ 434,
+ 568
+ ],
+ "score": 0.9,
+ "content": "j ^ { t h }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 434,
+ 554,
+ 507,
+ 573
+ ],
+ "score": 1.0,
+ "content": "layer. The inputs",
+ "type": "text"
+ }
+ ],
+ "index": 36
+ },
+ {
+ "bbox": [
+ 104,
+ 566,
+ 506,
+ 583
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 566,
+ 189,
+ 583
+ ],
+ "score": 1.0,
+ "content": "of the first layer of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 189,
+ 568,
+ 216,
+ 580
+ ],
+ "score": 0.9,
+ "content": "C ^ { G A T }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 217,
+ 566,
+ 506,
+ 583
+ ],
+ "score": 1.0,
+ "content": "also undergo the same transformation over the outputs of the feature",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 105,
+ 581,
+ 505,
+ 593
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 581,
+ 505,
+ 593
+ ],
+ "score": 1.0,
+ "content": "extractor model. We train our models with Adam (Kingma & Ba (2014)) with an initial learning rate",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ },
+ {
+ "bbox": [
+ 106,
+ 591,
+ 505,
+ 604
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 591,
+ 117,
+ 604
+ ],
+ "score": 1.0,
+ "content": "of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 118,
+ 591,
+ 139,
+ 602
+ ],
+ "score": 0.89,
+ "content": "1 0 ^ { - 3 }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 140,
+ 591,
+ 505,
+ 604
+ ],
+ "score": 1.0,
+ "content": "for 50 epochs. Each epoch has 10 iterations, where we randomly select a mini-batch from",
+ "type": "text"
+ }
+ ],
+ "index": 39
+ },
+ {
+ "bbox": [
+ 106,
+ 603,
+ 505,
+ 615
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 603,
+ 174,
+ 615
+ ],
+ "score": 1.0,
+ "content": "the training data",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 174,
+ 603,
+ 190,
+ 614
+ ],
+ "score": 0.88,
+ "content": "G _ { B }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 190,
+ 603,
+ 505,
+ 615
+ ],
+ "score": 1.0,
+ "content": ". The fine-tuning stage consists of 20 epochs with 10 iterations per epoch. We",
+ "type": "text"
+ }
+ ],
+ "index": 40
+ },
+ {
+ "bbox": [
+ 104,
+ 610,
+ 506,
+ 627
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 610,
+ 321,
+ 627
+ ],
+ "score": 1.0,
+ "content": "use a two-layer MLP over the final attention layer of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 321,
+ 612,
+ 348,
+ 624
+ ],
+ "score": 0.91,
+ "content": "C ^ { G A T ^ { \\bf 1 } }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 349,
+ 610,
+ 506,
+ 627
+ ],
+ "score": 1.0,
+ "content": "for classification. The attention layers",
+ "type": "text"
+ }
+ ],
+ "index": 41
+ },
+ {
+ "bbox": [
+ 106,
+ 624,
+ 505,
+ 637
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 624,
+ 505,
+ 637
+ ],
+ "score": 1.0,
+ "content": "use multi-head attention with 2 heads and leaky ReLU slope of 0.1 . The embeddings from both the",
+ "type": "text"
+ }
+ ],
+ "index": 42
+ },
+ {
+ "bbox": [
+ 106,
+ 636,
+ 505,
+ 647
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 636,
+ 320,
+ 647
+ ],
+ "score": 1.0,
+ "content": "attention heads are concatenated. For 20-shot, we set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 320,
+ 636,
+ 327,
+ 645
+ ],
+ "score": 0.79,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 327,
+ 636,
+ 505,
+ 647
+ ],
+ "score": 1.0,
+ "content": "to 2, number of super-classes to 3 and batch",
+ "type": "text"
+ }
+ ],
+ "index": 43
+ },
+ {
+ "bbox": [
+ 106,
+ 647,
+ 505,
+ 658
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 647,
+ 290,
+ 658
+ ],
+ "score": 1.0,
+ "content": "size to 128 on the Letter-High dataset, while",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 290,
+ 647,
+ 297,
+ 656
+ ],
+ "score": 0.81,
+ "content": "k",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 298,
+ 647,
+ 505,
+ 658
+ ],
+ "score": 1.0,
+ "content": "is set to 2 and batch size 64 on Reddit, ENZYMES",
+ "type": "text"
+ }
+ ],
+ "index": 44
+ },
+ {
+ "bbox": [
+ 106,
+ 657,
+ 506,
+ 669
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 657,
+ 506,
+ 669
+ ],
+ "score": 1.0,
+ "content": "and TRIANGLES datasets. The number of super-classes for Reddit are set to 2, for ENZYMES it",
+ "type": "text"
+ }
+ ],
+ "index": 45
+ },
+ {
+ "bbox": [
+ 106,
+ 668,
+ 505,
+ 681
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 668,
+ 505,
+ 681
+ ],
+ "score": 1.0,
+ "content": "is set to 1 and for TRIANGLES are 3. For ENZYMES, there are negligible differences on using 1",
+ "type": "text"
+ }
+ ],
+ "index": 46
+ },
+ {
+ "bbox": [
+ 105,
+ 678,
+ 506,
+ 692
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 678,
+ 506,
+ 692
+ ],
+ "score": 1.0,
+ "content": "and 2 super-classes as shown in table 4. We used Python Optimal Transport (POT) library 5 for",
+ "type": "text"
+ }
+ ],
+ "index": 47
+ },
+ {
+ "bbox": [
+ 106,
+ 690,
+ 303,
+ 703
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 690,
+ 197,
+ 703
+ ],
+ "score": 1.0,
+ "content": "implementation of the",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 197,
+ 691,
+ 203,
+ 702
+ ],
+ "score": 0.79,
+ "content": "p",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 204,
+ 690,
+ 303,
+ 703
+ ],
+ "score": 1.0,
+ "content": "-th Wasserstein distance.",
+ "type": "text"
+ }
+ ],
+ "index": 48
+ }
+ ],
+ "index": 42,
+ "bbox_fs": [
+ 103,
+ 554,
+ 507,
+ 703
+ ]
+ }
+ ]
+ },
+ {
+ "preproc_blocks": [
+ {
+ "type": "image",
+ "bbox": [
+ 122,
+ 49,
+ 493,
+ 197
+ ],
+ "blocks": [
+ {
+ "type": "image_body",
+ "bbox": [
+ 122,
+ 49,
+ 493,
+ 197
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 122,
+ 49,
+ 493,
+ 197
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 122,
+ 49,
+ 493,
+ 197
+ ],
+ "score": 0.882,
+ "type": "image",
+ "image_path": "29fdd17a77c3e8e015beb1724b352f5089f6e6dae91756dba71ded23dab355b7.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 1,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 122,
+ 49,
+ 493,
+ 98.33333333333334
+ ],
+ "spans": [],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 122,
+ 98.33333333333334,
+ 493,
+ 147.66666666666669
+ ],
+ "spans": [],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 122,
+ 147.66666666666669,
+ 493,
+ 197.00000000000003
+ ],
+ "spans": [],
+ "index": 2
+ }
+ ]
+ },
+ {
+ "type": "image_caption",
+ "bbox": [
+ 105,
+ 205,
+ 505,
+ 228
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 205,
+ 505,
+ 218
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 205,
+ 505,
+ 218
+ ],
+ "score": 1.0,
+ "content": "Figure 4: Visualization: t-SNE plots of the computed embeddings of test graphs on 20-shot scenario",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 105,
+ 216,
+ 465,
+ 229
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 216,
+ 465,
+ 229
+ ],
+ "score": 1.0,
+ "content": "from OurMethod-GAT (left), GIN (middle) and WL Kernel (right) on ENZYMES dataset.",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ }
+ ],
+ "index": 3.5
+ }
+ ],
+ "index": 2.25
+ },
+ {
+ "type": "image",
+ "bbox": [
+ 122,
+ 260,
+ 490,
+ 366
+ ],
+ "blocks": [
+ {
+ "type": "image_body",
+ "bbox": [
+ 122,
+ 260,
+ 490,
+ 366
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 122,
+ 260,
+ 490,
+ 366
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 122,
+ 260,
+ 490,
+ 366
+ ],
+ "score": 0.942,
+ "type": "image",
+ "image_path": "562f5e73ca715f0b27d1079b212c127e5869fabcb2d041d7a1dfb57df4883ff6.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 6,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 122,
+ 260,
+ 490,
+ 295.3333333333333
+ ],
+ "spans": [],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 122,
+ 295.3333333333333,
+ 490,
+ 330.66666666666663
+ ],
+ "spans": [],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 122,
+ 330.66666666666663,
+ 490,
+ 365.99999999999994
+ ],
+ "spans": [],
+ "index": 7
+ }
+ ]
+ },
+ {
+ "type": "image_caption",
+ "bbox": [
+ 106,
+ 375,
+ 506,
+ 397
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 375,
+ 505,
+ 387
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 375,
+ 505,
+ 387
+ ],
+ "score": 1.0,
+ "content": "Figure 5: Visualization: t-SNE plots of the computed embeddings of test graphs on 20-shot scenario",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 105,
+ 385,
+ 448,
+ 398
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 385,
+ 448,
+ 398
+ ],
+ "score": 1.0,
+ "content": "from OurMethod-GAT (left), GIN (middle) and WL Kernel (right) on Reddit dataset.",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ }
+ ],
+ "index": 8.5
+ }
+ ],
+ "index": 7.25
+ },
+ {
+ "type": "image",
+ "bbox": [
+ 122,
+ 428,
+ 489,
+ 535
+ ],
+ "blocks": [
+ {
+ "type": "image_body",
+ "bbox": [
+ 122,
+ 428,
+ 489,
+ 535
+ ],
+ "group_id": 2,
+ "lines": [
+ {
+ "bbox": [
+ 122,
+ 428,
+ 489,
+ 535
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 122,
+ 428,
+ 489,
+ 535
+ ],
+ "score": 0.962,
+ "type": "image",
+ "image_path": "5cc4f83d20953c2ca1efcadb37047fc6947a35899155b9dcb5b918db26496726.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 11,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 122,
+ 428,
+ 489,
+ 463.6666666666667
+ ],
+ "spans": [],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 122,
+ 463.6666666666667,
+ 489,
+ 499.33333333333337
+ ],
+ "spans": [],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 122,
+ 499.33333333333337,
+ 489,
+ 535.0
+ ],
+ "spans": [],
+ "index": 12
+ }
+ ]
+ },
+ {
+ "type": "image_caption",
+ "bbox": [
+ 106,
+ 544,
+ 505,
+ 567
+ ],
+ "group_id": 2,
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 544,
+ 505,
+ 557
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 544,
+ 505,
+ 557
+ ],
+ "score": 1.0,
+ "content": "Figure 6: Visualization: t-SNE plots of the computed embeddings of test graphs on 20-shot scenario",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 105,
+ 555,
+ 468,
+ 568
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 555,
+ 468,
+ 568
+ ],
+ "score": 1.0,
+ "content": "from OurMethod-GAT (left), GIN (middle) and WL Kernel (right) on Letter-High dataset.",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ }
+ ],
+ "index": 13.5
+ }
+ ],
+ "index": 12.25
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 586,
+ 225,
+ 597
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 585,
+ 226,
+ 598
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 585,
+ 226,
+ 598
+ ],
+ "score": 1.0,
+ "content": "A.4 SILHOUETTE SCORES",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ }
+ ],
+ "index": 15
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 606,
+ 505,
+ 732
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 606,
+ 505,
+ 618
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 606,
+ 505,
+ 618
+ ],
+ "score": 1.0,
+ "content": "To assess the clustering abilities of the models we analyze the silhouette scores of the test embed-",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 105,
+ 618,
+ 506,
+ 630
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 618,
+ 506,
+ 630
+ ],
+ "score": 1.0,
+ "content": "dings produced by the GAT variant of our method, GIN and WL Kernel. Silhouette coefficient",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 105,
+ 628,
+ 506,
+ 641
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 628,
+ 506,
+ 641
+ ],
+ "score": 1.0,
+ "content": "essentially measures the ratio of intra-class versus inter-class distance. The Silhouette Coefficient",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 106,
+ 640,
+ 504,
+ 651
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 640,
+ 504,
+ 651
+ ],
+ "score": 1.0,
+ "content": "is calculated using the mean intra-cluster distance (a) and the mean nearest-cluster distance (b) for",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 102,
+ 649,
+ 509,
+ 672
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 102,
+ 649,
+ 371,
+ 672
+ ],
+ "score": 1.0,
+ "content": "each sample. The Silhouette Coefficient for a sample is given by",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 371,
+ 650,
+ 407,
+ 667
+ ],
+ "score": 0.94,
+ "content": "\\frac { ( b - a ) } { m a x ( a , b ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 407,
+ 649,
+ 438,
+ 672
+ ],
+ "score": 1.0,
+ "content": ",where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 438,
+ 653,
+ 444,
+ 663
+ ],
+ "score": 0.73,
+ "content": "b",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 444,
+ 649,
+ 509,
+ 672
+ ],
+ "score": 1.0,
+ "content": "is the distance",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 105,
+ 666,
+ 505,
+ 678
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 666,
+ 505,
+ 678
+ ],
+ "score": 1.0,
+ "content": "between a sample and the nearest cluster that the sample is not a part of. The results for mean",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 106,
+ 677,
+ 505,
+ 689
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 677,
+ 505,
+ 689
+ ],
+ "score": 1.0,
+ "content": "silhouette coefficient over the test samples averaged over multiple runs are shown in Table 8. We",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 105,
+ 687,
+ 505,
+ 700
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 687,
+ 505,
+ 700
+ ],
+ "score": 1.0,
+ "content": "normalize the embeddings before calculating the silhouette coefficient. We can clearly see that our",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ },
+ {
+ "bbox": [
+ 105,
+ 699,
+ 505,
+ 711
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 699,
+ 505,
+ 711
+ ],
+ "score": 1.0,
+ "content": "model creates better clusters with low intra-cluster distance as well as high inter-cluster distance.",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 106,
+ 710,
+ 505,
+ 722
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 710,
+ 505,
+ 722
+ ],
+ "score": 1.0,
+ "content": "Note that the coefficient value for WL remains the same for all scenarios since it computes fixed",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 105,
+ 720,
+ 335,
+ 734
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 720,
+ 335,
+ 734
+ ],
+ "score": 1.0,
+ "content": "embeddings attributed to absence of any DL component.",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ }
+ ],
+ "index": 21
+ }
+ ],
+ "page_idx": 15,
+ "page_size": [
+ 612,
+ 792
+ ],
+ "discarded_blocks": [
+ {
+ "type": "discarded",
+ "bbox": [
+ 107,
+ 27,
+ 293,
+ 37
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "score": 1.0,
+ "content": "Published as a conference paper at ICLR 2020",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 300,
+ 751,
+ 311,
+ 760
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 299,
+ 750,
+ 313,
+ 764
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 299,
+ 750,
+ 313,
+ 764
+ ],
+ "score": 1.0,
+ "content": "16",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ }
+ ],
+ "para_blocks": [
+ {
+ "type": "image",
+ "bbox": [
+ 122,
+ 49,
+ 493,
+ 197
+ ],
+ "blocks": [
+ {
+ "type": "image_body",
+ "bbox": [
+ 122,
+ 49,
+ 493,
+ 197
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 122,
+ 49,
+ 493,
+ 197
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 122,
+ 49,
+ 493,
+ 197
+ ],
+ "score": 0.882,
+ "type": "image",
+ "image_path": "29fdd17a77c3e8e015beb1724b352f5089f6e6dae91756dba71ded23dab355b7.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 1,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 122,
+ 49,
+ 493,
+ 98.33333333333334
+ ],
+ "spans": [],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 122,
+ 98.33333333333334,
+ 493,
+ 147.66666666666669
+ ],
+ "spans": [],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 122,
+ 147.66666666666669,
+ 493,
+ 197.00000000000003
+ ],
+ "spans": [],
+ "index": 2
+ }
+ ]
+ },
+ {
+ "type": "image_caption",
+ "bbox": [
+ 105,
+ 205,
+ 505,
+ 228
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 205,
+ 505,
+ 218
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 205,
+ 505,
+ 218
+ ],
+ "score": 1.0,
+ "content": "Figure 4: Visualization: t-SNE plots of the computed embeddings of test graphs on 20-shot scenario",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 105,
+ 216,
+ 465,
+ 229
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 216,
+ 465,
+ 229
+ ],
+ "score": 1.0,
+ "content": "from OurMethod-GAT (left), GIN (middle) and WL Kernel (right) on ENZYMES dataset.",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ }
+ ],
+ "index": 3.5
+ }
+ ],
+ "index": 2.25
+ },
+ {
+ "type": "image",
+ "bbox": [
+ 122,
+ 260,
+ 490,
+ 366
+ ],
+ "blocks": [
+ {
+ "type": "image_body",
+ "bbox": [
+ 122,
+ 260,
+ 490,
+ 366
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 122,
+ 260,
+ 490,
+ 366
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 122,
+ 260,
+ 490,
+ 366
+ ],
+ "score": 0.942,
+ "type": "image",
+ "image_path": "562f5e73ca715f0b27d1079b212c127e5869fabcb2d041d7a1dfb57df4883ff6.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 6,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 122,
+ 260,
+ 490,
+ 295.3333333333333
+ ],
+ "spans": [],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 122,
+ 295.3333333333333,
+ 490,
+ 330.66666666666663
+ ],
+ "spans": [],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 122,
+ 330.66666666666663,
+ 490,
+ 365.99999999999994
+ ],
+ "spans": [],
+ "index": 7
+ }
+ ]
+ },
+ {
+ "type": "image_caption",
+ "bbox": [
+ 106,
+ 375,
+ 506,
+ 397
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 375,
+ 505,
+ 387
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 375,
+ 505,
+ 387
+ ],
+ "score": 1.0,
+ "content": "Figure 5: Visualization: t-SNE plots of the computed embeddings of test graphs on 20-shot scenario",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 105,
+ 385,
+ 448,
+ 398
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 385,
+ 448,
+ 398
+ ],
+ "score": 1.0,
+ "content": "from OurMethod-GAT (left), GIN (middle) and WL Kernel (right) on Reddit dataset.",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ }
+ ],
+ "index": 8.5
+ }
+ ],
+ "index": 7.25
+ },
+ {
+ "type": "image",
+ "bbox": [
+ 122,
+ 428,
+ 489,
+ 535
+ ],
+ "blocks": [
+ {
+ "type": "image_body",
+ "bbox": [
+ 122,
+ 428,
+ 489,
+ 535
+ ],
+ "group_id": 2,
+ "lines": [
+ {
+ "bbox": [
+ 122,
+ 428,
+ 489,
+ 535
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 122,
+ 428,
+ 489,
+ 535
+ ],
+ "score": 0.962,
+ "type": "image",
+ "image_path": "5cc4f83d20953c2ca1efcadb37047fc6947a35899155b9dcb5b918db26496726.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 11,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 122,
+ 428,
+ 489,
+ 463.6666666666667
+ ],
+ "spans": [],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 122,
+ 463.6666666666667,
+ 489,
+ 499.33333333333337
+ ],
+ "spans": [],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 122,
+ 499.33333333333337,
+ 489,
+ 535.0
+ ],
+ "spans": [],
+ "index": 12
+ }
+ ]
+ },
+ {
+ "type": "image_caption",
+ "bbox": [
+ 106,
+ 544,
+ 505,
+ 567
+ ],
+ "group_id": 2,
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 544,
+ 505,
+ 557
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 544,
+ 505,
+ 557
+ ],
+ "score": 1.0,
+ "content": "Figure 6: Visualization: t-SNE plots of the computed embeddings of test graphs on 20-shot scenario",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 105,
+ 555,
+ 468,
+ 568
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 555,
+ 468,
+ 568
+ ],
+ "score": 1.0,
+ "content": "from OurMethod-GAT (left), GIN (middle) and WL Kernel (right) on Letter-High dataset.",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ }
+ ],
+ "index": 13.5
+ }
+ ],
+ "index": 12.25
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 586,
+ 225,
+ 597
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 585,
+ 226,
+ 598
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 585,
+ 226,
+ 598
+ ],
+ "score": 1.0,
+ "content": "A.4 SILHOUETTE SCORES",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ }
+ ],
+ "index": 15
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 106,
+ 606,
+ 505,
+ 732
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 606,
+ 505,
+ 618
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 606,
+ 505,
+ 618
+ ],
+ "score": 1.0,
+ "content": "To assess the clustering abilities of the models we analyze the silhouette scores of the test embed-",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 105,
+ 618,
+ 506,
+ 630
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 618,
+ 506,
+ 630
+ ],
+ "score": 1.0,
+ "content": "dings produced by the GAT variant of our method, GIN and WL Kernel. Silhouette coefficient",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 105,
+ 628,
+ 506,
+ 641
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 628,
+ 506,
+ 641
+ ],
+ "score": 1.0,
+ "content": "essentially measures the ratio of intra-class versus inter-class distance. The Silhouette Coefficient",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 106,
+ 640,
+ 504,
+ 651
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 640,
+ 504,
+ 651
+ ],
+ "score": 1.0,
+ "content": "is calculated using the mean intra-cluster distance (a) and the mean nearest-cluster distance (b) for",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 102,
+ 649,
+ 509,
+ 672
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 102,
+ 649,
+ 371,
+ 672
+ ],
+ "score": 1.0,
+ "content": "each sample. The Silhouette Coefficient for a sample is given by",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 371,
+ 650,
+ 407,
+ 667
+ ],
+ "score": 0.94,
+ "content": "\\frac { ( b - a ) } { m a x ( a , b ) }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 407,
+ 649,
+ 438,
+ 672
+ ],
+ "score": 1.0,
+ "content": ",where",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 438,
+ 653,
+ 444,
+ 663
+ ],
+ "score": 0.73,
+ "content": "b",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 444,
+ 649,
+ 509,
+ 672
+ ],
+ "score": 1.0,
+ "content": "is the distance",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 105,
+ 666,
+ 505,
+ 678
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 666,
+ 505,
+ 678
+ ],
+ "score": 1.0,
+ "content": "between a sample and the nearest cluster that the sample is not a part of. The results for mean",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 106,
+ 677,
+ 505,
+ 689
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 677,
+ 505,
+ 689
+ ],
+ "score": 1.0,
+ "content": "silhouette coefficient over the test samples averaged over multiple runs are shown in Table 8. We",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 105,
+ 687,
+ 505,
+ 700
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 687,
+ 505,
+ 700
+ ],
+ "score": 1.0,
+ "content": "normalize the embeddings before calculating the silhouette coefficient. We can clearly see that our",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ },
+ {
+ "bbox": [
+ 105,
+ 699,
+ 505,
+ 711
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 699,
+ 505,
+ 711
+ ],
+ "score": 1.0,
+ "content": "model creates better clusters with low intra-cluster distance as well as high inter-cluster distance.",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 106,
+ 710,
+ 505,
+ 722
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 710,
+ 505,
+ 722
+ ],
+ "score": 1.0,
+ "content": "Note that the coefficient value for WL remains the same for all scenarios since it computes fixed",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 105,
+ 720,
+ 335,
+ 734
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 720,
+ 335,
+ 734
+ ],
+ "score": 1.0,
+ "content": "embeddings attributed to absence of any DL component.",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ }
+ ],
+ "index": 21,
+ "bbox_fs": [
+ 102,
+ 606,
+ 509,
+ 734
+ ]
+ }
+ ]
+ },
+ {
+ "preproc_blocks": [
+ {
+ "type": "table",
+ "bbox": [
+ 124,
+ 119,
+ 487,
+ 162
+ ],
+ "blocks": [
+ {
+ "type": "table_caption",
+ "bbox": [
+ 107,
+ 80,
+ 504,
+ 103
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 80,
+ 506,
+ 93
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 80,
+ 506,
+ 93
+ ],
+ "score": 1.0,
+ "content": "Table 8: Silhouette coefficients of the test classes for the three dominant models - GAT variant of",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 106,
+ 91,
+ 376,
+ 103
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 91,
+ 376,
+ 103
+ ],
+ "score": 1.0,
+ "content": "Our Method, GIN and WL. The best scores are highlighted in bold.",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ }
+ ],
+ "index": 0.5
+ },
+ {
+ "type": "table_body",
+ "bbox": [
+ 124,
+ 119,
+ 487,
+ 162
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 124,
+ 119,
+ 487,
+ 162
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 124,
+ 119,
+ 487,
+ 162
+ ],
+ "score": 0.966,
+ "html": "| Method | Reddit-12K | ENZYMES | Letter-High | TRIANGLES |
| 10-shot | 20-shot | 10-shot | 20-shot | 10-shot | 20-shot | 10-shot | 20-shot |
| GIN | -0.0566 | -0.0652 | 0.0168 | 0.0432 | 0.2157 | 0.2316 | 0.0373 | 0.1256 |
| WLKernel | -0.0626 | -0.0626 | 0.0366 | 0.0366 | 0.2490 | 0.2490 | 0.0186 | 0.0186 |
| OurMethod-GAT | -0.0553 | -0.0559 | 0.0296 | 0.1172 | 0.3494 | 0.3787 | 0.3824 | 0.4508 |
",
+ "type": "table",
+ "image_path": "47003b584db4ec8d671af55bb420905906ffbb2b85e175dedc50ef2660beaad4.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 3,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 124,
+ 119,
+ 487,
+ 133.33333333333334
+ ],
+ "spans": [],
+ "index": 2
+ },
+ {
+ "bbox": [
+ 124,
+ 133.33333333333334,
+ 487,
+ 147.66666666666669
+ ],
+ "spans": [],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 124,
+ 147.66666666666669,
+ 487,
+ 162.00000000000003
+ ],
+ "spans": [],
+ "index": 4
+ }
+ ]
+ }
+ ],
+ "index": 1.75
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 173,
+ 505,
+ 195
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 173,
+ 505,
+ 185
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 173,
+ 332,
+ 185
+ ],
+ "score": 1.0,
+ "content": "Table 9: Semi-supervised fine-tuning results for various",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 333,
+ 175,
+ 339,
+ 184
+ ],
+ "score": 0.79,
+ "content": "p",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 339,
+ 173,
+ 505,
+ 185
+ ],
+ "score": 1.0,
+ "content": "values on 10-shot and 20-shot scenarios,",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 106,
+ 183,
+ 469,
+ 196
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 183,
+ 469,
+ 196
+ ],
+ "score": 1.0,
+ "content": "where “No Semi-Sup” represents the fine-tuning stage without additional labeled samples.",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ }
+ ],
+ "index": 5.5
+ },
+ {
+ "type": "table",
+ "bbox": [
+ 126,
+ 212,
+ 485,
+ 289
+ ],
+ "blocks": [
+ {
+ "type": "table_body",
+ "bbox": [
+ 126,
+ 212,
+ 485,
+ 289
+ ],
+ "group_id": 2,
+ "lines": [
+ {
+ "bbox": [
+ 126,
+ 212,
+ 485,
+ 289
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 126,
+ 212,
+ 485,
+ 289
+ ],
+ "score": 0.947,
+ "html": "| Dataset | 10-shot | 20-shot |
| No Semi-Sup | 25 | 50 | No Semi-Sup | 25 | 50 |
| Letter-High | 73.21± 3.19 | 74.18 ± 2.58 | 74.65 ± 2.16 | 76.95 ± 1.79 | 77.79 ± 1.52 | 78.31 ± 1.11 |
| TRIANGLES | 75.83 ± 2.97 | 76.36± 2.59 | 77.8 ± 2.04 | 80.09 ±1.78 | 81.29 ± 1.98 | 81.87 ± 1.45 |
| Dataset | | 10-shot | | | 20-shot | |
| No Semi-Sup | 10 | 20 | No Semi-Sup | 10 | 20 |
| Reddit | 45.41± 3.79 | 45.88±3.32 | 46.01± 2.99 | 50.34± 2.77 | 50.76±2.52 | 51.17 ± 2.21 |
| ENZYMES | 60.13 ± 3.98 | 60.87 ± 3.24 | 61.25 ± 3.17 | 62.74 ± 3.64 | 63.10± 3.47 | 63.67 ± 3.18 |
",
+ "type": "table",
+ "image_path": "bbca2369fa72335fc5638686c7581496db6b789c5859f356022902292405a991.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 8,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 126,
+ 212,
+ 485,
+ 237.66666666666666
+ ],
+ "spans": [],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 126,
+ 237.66666666666666,
+ 485,
+ 263.3333333333333
+ ],
+ "spans": [],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 126,
+ 263.3333333333333,
+ 485,
+ 289.0
+ ],
+ "spans": [],
+ "index": 9
+ }
+ ]
+ }
+ ],
+ "index": 8
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 309,
+ 274,
+ 320
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 308,
+ 275,
+ 321
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 308,
+ 275,
+ 321
+ ],
+ "score": 1.0,
+ "content": "A.5 SEMI-SUPERVISED FINE-TUNING",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ }
+ ],
+ "index": 10
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 108,
+ 330,
+ 504,
+ 363
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 330,
+ 505,
+ 342
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 330,
+ 505,
+ 342
+ ],
+ "score": 1.0,
+ "content": "In many real-world learning scenarios, it is quite common to find abundant unlabelled data. Since",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 106,
+ 340,
+ 504,
+ 353
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 340,
+ 504,
+ 353
+ ],
+ "score": 1.0,
+ "content": "our model uses a GNN classifier, this makes it possible to use unlabelled data while learning through",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 105,
+ 351,
+ 505,
+ 365
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 351,
+ 505,
+ 365
+ ],
+ "score": 1.0,
+ "content": "message passing, where the fine tuning stage of our method is performed in semi-supervised settings.",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ }
+ ],
+ "index": 12
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 368,
+ 505,
+ 457
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 104,
+ 365,
+ 506,
+ 382
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 365,
+ 402,
+ 382
+ ],
+ "score": 1.0,
+ "content": "Essentially, while fine tuning the model, i.e., only training the classifier",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 403,
+ 367,
+ 430,
+ 379
+ ],
+ "score": 0.91,
+ "content": "C ^ { G A T }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 430,
+ 365,
+ 445,
+ 382
+ ],
+ "score": 1.0,
+ "content": "on",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 446,
+ 369,
+ 462,
+ 380
+ ],
+ "score": 0.9,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 462,
+ 365,
+ 506,
+ 382
+ ],
+ "score": 1.0,
+ "content": ", we addi-",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 105,
+ 379,
+ 505,
+ 392
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 379,
+ 155,
+ 392
+ ],
+ "score": 1.0,
+ "content": "tionally use",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 156,
+ 382,
+ 163,
+ 391
+ ],
+ "score": 0.75,
+ "content": "p",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 163,
+ 379,
+ 263,
+ 392
+ ],
+ "score": 1.0,
+ "content": "more graphs along with",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 263,
+ 380,
+ 279,
+ 390
+ ],
+ "score": 0.9,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 280,
+ 379,
+ 505,
+ 392
+ ],
+ "score": 1.0,
+ "content": ", whose labels are unknown. The learning objective for",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 105,
+ 390,
+ 506,
+ 403
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 390,
+ 506,
+ 403
+ ],
+ "score": 1.0,
+ "content": "fine tuning stage doesn’t change since the gradients are back-propagated from the labeled samples",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 105,
+ 401,
+ 506,
+ 414
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 401,
+ 506,
+ 414
+ ],
+ "score": 1.0,
+ "content": "only. In this setting, each node in the attention classifier can aggregate information from unlabelled",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 105,
+ 411,
+ 506,
+ 425
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 411,
+ 415,
+ 425
+ ],
+ "score": 1.0,
+ "content": "samples as well, thus allowing improved learning of the graphs features in",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 415,
+ 411,
+ 442,
+ 423
+ ],
+ "score": 0.93,
+ "content": "C ^ { G A T }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 443,
+ 411,
+ 506,
+ 425
+ ],
+ "score": 1.0,
+ "content": ". We show the",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 105,
+ 423,
+ 505,
+ 435
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 423,
+ 148,
+ 435
+ ],
+ "score": 1.0,
+ "content": "results for",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 149,
+ 425,
+ 155,
+ 435
+ ],
+ "score": 0.77,
+ "content": "p",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 156,
+ 423,
+ 441,
+ 435
+ ],
+ "score": 1.0,
+ "content": "values 25 and 50 on Letter-High and TRIANGLES datasets, whereas for",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 441,
+ 425,
+ 448,
+ 435
+ ],
+ "score": 0.81,
+ "content": "p",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 448,
+ 423,
+ 505,
+ 435
+ ],
+ "score": 1.0,
+ "content": "values 10 and",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 106,
+ 434,
+ 505,
+ 446
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 434,
+ 505,
+ 446
+ ],
+ "score": 1.0,
+ "content": "20 on Reddit and ENZYMES datasets. The results are shown in Table 9. We observe an increase in",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 106,
+ 445,
+ 447,
+ 458
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 445,
+ 447,
+ 458
+ ],
+ "score": 1.0,
+ "content": "the accuracy with increase in number of unlabeled samples during fine-tuning phase.",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ }
+ ],
+ "index": 17.5
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 471,
+ 289,
+ 482
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 470,
+ 290,
+ 483
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 470,
+ 290,
+ 483
+ ],
+ "score": 1.0,
+ "content": "A.6 ADAPTATION TO ACTIVE-LEARNING",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ }
+ ],
+ "index": 22
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 491,
+ 505,
+ 547
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 490,
+ 505,
+ 505
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 490,
+ 505,
+ 505
+ ],
+ "score": 1.0,
+ "content": "In this section, we show the adaptation of our model to highly practical active learning scenario.",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ },
+ {
+ "bbox": [
+ 106,
+ 503,
+ 505,
+ 515
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 503,
+ 505,
+ 515
+ ],
+ "score": 1.0,
+ "content": "In many real world applications, we might start with few samples per class, however as the number",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 105,
+ 513,
+ 505,
+ 527
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 513,
+ 505,
+ 527
+ ],
+ "score": 1.0,
+ "content": "of samples to classify from these classes increase over time, some of these samples can be used by",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 106,
+ 524,
+ 505,
+ 537
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 524,
+ 505,
+ 537
+ ],
+ "score": 1.0,
+ "content": "the model to adaptively learn and improve with very less human intervention, since the number of",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 105,
+ 536,
+ 402,
+ 547
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 536,
+ 402,
+ 547
+ ],
+ "score": 1.0,
+ "content": "number of samples to be queried for theirs label can always be controlled.",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ }
+ ],
+ "index": 25
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 552,
+ 505,
+ 586
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 552,
+ 505,
+ 564
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 552,
+ 505,
+ 564
+ ],
+ "score": 1.0,
+ "content": "To perform active-learning, we first select a random subset of size 100 for Letter-High and TRI-",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 106,
+ 564,
+ 505,
+ 574
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 564,
+ 505,
+ 574
+ ],
+ "score": 1.0,
+ "content": "ANGLES datasets as well as a random subset of size 40 for Reddit and ENZYMES datasets, which",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 105,
+ 572,
+ 506,
+ 588
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 572,
+ 153,
+ 588
+ ],
+ "score": 1.0,
+ "content": "we term as",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 153,
+ 574,
+ 190,
+ 586
+ ],
+ "score": 0.9,
+ "content": "G _ { r a n d o m }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 191,
+ 572,
+ 307,
+ 588
+ ],
+ "score": 1.0,
+ "content": ", then fine tune the model on",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 308,
+ 574,
+ 324,
+ 585
+ ],
+ "score": 0.89,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 325,
+ 572,
+ 463,
+ 588
+ ],
+ "score": 1.0,
+ "content": "and further evaluate the model on",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 464,
+ 574,
+ 501,
+ 586
+ ],
+ "score": 0.86,
+ "content": "G _ { r a n d o m }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 501,
+ 572,
+ 506,
+ 588
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ }
+ ],
+ "index": 29
+ },
+ {
+ "type": "table",
+ "bbox": [
+ 126,
+ 654,
+ 485,
+ 731
+ ],
+ "blocks": [
+ {
+ "type": "table_caption",
+ "bbox": [
+ 107,
+ 604,
+ 505,
+ 637
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 604,
+ 505,
+ 616
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 604,
+ 496,
+ 616
+ ],
+ "score": 1.0,
+ "content": "Table 10: Active Learning Results. The value below each shot represents the number samples",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 497,
+ 604,
+ 501,
+ 614
+ ],
+ "score": 0.47,
+ "content": "l",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 502,
+ 604,
+ 505,
+ 616
+ ],
+ "score": 1.0,
+ "content": ",",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ },
+ {
+ "bbox": [
+ 105,
+ 614,
+ 506,
+ 628
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 614,
+ 144,
+ 628
+ ],
+ "score": 1.0,
+ "content": "added to",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 144,
+ 615,
+ 160,
+ 626
+ ],
+ "score": 0.89,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 160,
+ 614,
+ 506,
+ 628
+ ],
+ "score": 1.0,
+ "content": "for second fine-tuning step, where “No AL” represents the model evaluation without",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 105,
+ 625,
+ 217,
+ 638
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 625,
+ 217,
+ 638
+ ],
+ "score": 1.0,
+ "content": "additional labeled samples.",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ }
+ ],
+ "index": 32
+ },
+ {
+ "type": "table_body",
+ "bbox": [
+ 126,
+ 654,
+ 485,
+ 731
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 126,
+ 654,
+ 485,
+ 731
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 126,
+ 654,
+ 485,
+ 731
+ ],
+ "score": 0.945,
+ "html": "| Dataset | 10-shot | 20-shot |
| No AL | 15 | 25 | No AL | 15 | 25 |
| Letter-High | 73.34± 3.37 | 75.03 ± 3.24 | 76.89 ± 2.16 | 77.06 ± 1.73 | 78.44 ± 1.52 | 79.28 ± 1.36 |
| TRIANGLES | 76.02 ± 2.54 | 78.44 ± 1.84 | 79.91 ± 1.28 | 80.27 ±1.84 | 81.74 ± 2.03 | 82.58 ± 1.57 |
| Dataset | 10-shot | 20-shot |
| No AL | 10 | 20 | No AL | 10 | 20 |
| Reddit | 45.41± 3.79 | 46.88±3.14 | 47.91± 2.99 | 50.43±2.66 | 51.76± 2.32 | 53.07± 2.21 |
| ENZYMES | 60.13 ± 3.98 | 61.57 ± 3.48 | 62.25 ± 3.06 | 62.74 ± 3.64 | 63.60±3.30 | 64.97 ± 3.11 |
",
+ "type": "table",
+ "image_path": "8a7cc9daae3583429066fddac91f1e1cd620856b408f3ed283bb8bd95e385ddf.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 35,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 126,
+ 654,
+ 485,
+ 679.6666666666666
+ ],
+ "spans": [],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 126,
+ 679.6666666666666,
+ 485,
+ 705.3333333333333
+ ],
+ "spans": [],
+ "index": 35
+ },
+ {
+ "bbox": [
+ 126,
+ 705.3333333333333,
+ 485,
+ 730.9999999999999
+ ],
+ "spans": [],
+ "index": 36
+ }
+ ]
+ }
+ ],
+ "index": 33.5
+ }
+ ],
+ "page_idx": 16,
+ "page_size": [
+ 612,
+ 792
+ ],
+ "discarded_blocks": [
+ {
+ "type": "discarded",
+ "bbox": [
+ 107,
+ 27,
+ 293,
+ 37
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "score": 1.0,
+ "content": "Published as a conference paper at ICLR 2020",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 300,
+ 751,
+ 311,
+ 760
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 299,
+ 750,
+ 313,
+ 765
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 299,
+ 750,
+ 313,
+ 765
+ ],
+ "score": 1.0,
+ "content": "",
+ "type": "text",
+ "height": 15,
+ "width": 14
+ }
+ ]
+ }
+ ]
+ }
+ ],
+ "para_blocks": [
+ {
+ "type": "table",
+ "bbox": [
+ 124,
+ 119,
+ 487,
+ 162
+ ],
+ "blocks": [
+ {
+ "type": "table_caption",
+ "bbox": [
+ 107,
+ 80,
+ 504,
+ 103
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 80,
+ 506,
+ 93
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 80,
+ 506,
+ 93
+ ],
+ "score": 1.0,
+ "content": "Table 8: Silhouette coefficients of the test classes for the three dominant models - GAT variant of",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 106,
+ 91,
+ 376,
+ 103
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 91,
+ 376,
+ 103
+ ],
+ "score": 1.0,
+ "content": "Our Method, GIN and WL. The best scores are highlighted in bold.",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ }
+ ],
+ "index": 0.5
+ },
+ {
+ "type": "table_body",
+ "bbox": [
+ 124,
+ 119,
+ 487,
+ 162
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 124,
+ 119,
+ 487,
+ 162
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 124,
+ 119,
+ 487,
+ 162
+ ],
+ "score": 0.966,
+ "html": "| Method | Reddit-12K | ENZYMES | Letter-High | TRIANGLES |
| 10-shot | 20-shot | 10-shot | 20-shot | 10-shot | 20-shot | 10-shot | 20-shot |
| GIN | -0.0566 | -0.0652 | 0.0168 | 0.0432 | 0.2157 | 0.2316 | 0.0373 | 0.1256 |
| WLKernel | -0.0626 | -0.0626 | 0.0366 | 0.0366 | 0.2490 | 0.2490 | 0.0186 | 0.0186 |
| OurMethod-GAT | -0.0553 | -0.0559 | 0.0296 | 0.1172 | 0.3494 | 0.3787 | 0.3824 | 0.4508 |
",
+ "type": "table",
+ "image_path": "47003b584db4ec8d671af55bb420905906ffbb2b85e175dedc50ef2660beaad4.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 3,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 124,
+ 119,
+ 487,
+ 133.33333333333334
+ ],
+ "spans": [],
+ "index": 2
+ },
+ {
+ "bbox": [
+ 124,
+ 133.33333333333334,
+ 487,
+ 147.66666666666669
+ ],
+ "spans": [],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 124,
+ 147.66666666666669,
+ 487,
+ 162.00000000000003
+ ],
+ "spans": [],
+ "index": 4
+ }
+ ]
+ }
+ ],
+ "index": 1.75
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 173,
+ 505,
+ 195
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 173,
+ 505,
+ 185
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 173,
+ 332,
+ 185
+ ],
+ "score": 1.0,
+ "content": "Table 9: Semi-supervised fine-tuning results for various",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 333,
+ 175,
+ 339,
+ 184
+ ],
+ "score": 0.79,
+ "content": "p",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 339,
+ 173,
+ 505,
+ 185
+ ],
+ "score": 1.0,
+ "content": "values on 10-shot and 20-shot scenarios,",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 106,
+ 183,
+ 469,
+ 196
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 183,
+ 469,
+ 196
+ ],
+ "score": 1.0,
+ "content": "where “No Semi-Sup” represents the fine-tuning stage without additional labeled samples.",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ }
+ ],
+ "index": 5.5,
+ "bbox_fs": [
+ 106,
+ 173,
+ 505,
+ 196
+ ]
+ },
+ {
+ "type": "table",
+ "bbox": [
+ 126,
+ 212,
+ 485,
+ 289
+ ],
+ "blocks": [
+ {
+ "type": "table_body",
+ "bbox": [
+ 126,
+ 212,
+ 485,
+ 289
+ ],
+ "group_id": 2,
+ "lines": [
+ {
+ "bbox": [
+ 126,
+ 212,
+ 485,
+ 289
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 126,
+ 212,
+ 485,
+ 289
+ ],
+ "score": 0.947,
+ "html": "| Dataset | 10-shot | 20-shot |
| No Semi-Sup | 25 | 50 | No Semi-Sup | 25 | 50 |
| Letter-High | 73.21± 3.19 | 74.18 ± 2.58 | 74.65 ± 2.16 | 76.95 ± 1.79 | 77.79 ± 1.52 | 78.31 ± 1.11 |
| TRIANGLES | 75.83 ± 2.97 | 76.36± 2.59 | 77.8 ± 2.04 | 80.09 ±1.78 | 81.29 ± 1.98 | 81.87 ± 1.45 |
| Dataset | | 10-shot | | | 20-shot | |
| No Semi-Sup | 10 | 20 | No Semi-Sup | 10 | 20 |
| Reddit | 45.41± 3.79 | 45.88±3.32 | 46.01± 2.99 | 50.34± 2.77 | 50.76±2.52 | 51.17 ± 2.21 |
| ENZYMES | 60.13 ± 3.98 | 60.87 ± 3.24 | 61.25 ± 3.17 | 62.74 ± 3.64 | 63.10± 3.47 | 63.67 ± 3.18 |
",
+ "type": "table",
+ "image_path": "bbca2369fa72335fc5638686c7581496db6b789c5859f356022902292405a991.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 8,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 126,
+ 212,
+ 485,
+ 237.66666666666666
+ ],
+ "spans": [],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 126,
+ 237.66666666666666,
+ 485,
+ 263.3333333333333
+ ],
+ "spans": [],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 126,
+ 263.3333333333333,
+ 485,
+ 289.0
+ ],
+ "spans": [],
+ "index": 9
+ }
+ ]
+ }
+ ],
+ "index": 8
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 309,
+ 274,
+ 320
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 308,
+ 275,
+ 321
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 308,
+ 275,
+ 321
+ ],
+ "score": 1.0,
+ "content": "A.5 SEMI-SUPERVISED FINE-TUNING",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ }
+ ],
+ "index": 10
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 108,
+ 330,
+ 504,
+ 363
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 330,
+ 505,
+ 342
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 330,
+ 505,
+ 342
+ ],
+ "score": 1.0,
+ "content": "In many real-world learning scenarios, it is quite common to find abundant unlabelled data. Since",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ },
+ {
+ "bbox": [
+ 106,
+ 340,
+ 504,
+ 353
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 340,
+ 504,
+ 353
+ ],
+ "score": 1.0,
+ "content": "our model uses a GNN classifier, this makes it possible to use unlabelled data while learning through",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ },
+ {
+ "bbox": [
+ 105,
+ 351,
+ 505,
+ 365
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 351,
+ 505,
+ 365
+ ],
+ "score": 1.0,
+ "content": "message passing, where the fine tuning stage of our method is performed in semi-supervised settings.",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ }
+ ],
+ "index": 12,
+ "bbox_fs": [
+ 105,
+ 330,
+ 505,
+ 365
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 368,
+ 505,
+ 457
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 104,
+ 365,
+ 506,
+ 382
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 365,
+ 402,
+ 382
+ ],
+ "score": 1.0,
+ "content": "Essentially, while fine tuning the model, i.e., only training the classifier",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 403,
+ 367,
+ 430,
+ 379
+ ],
+ "score": 0.91,
+ "content": "C ^ { G A T }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 430,
+ 365,
+ 445,
+ 382
+ ],
+ "score": 1.0,
+ "content": "on",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 446,
+ 369,
+ 462,
+ 380
+ ],
+ "score": 0.9,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 462,
+ 365,
+ 506,
+ 382
+ ],
+ "score": 1.0,
+ "content": ", we addi-",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 105,
+ 379,
+ 505,
+ 392
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 379,
+ 155,
+ 392
+ ],
+ "score": 1.0,
+ "content": "tionally use",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 156,
+ 382,
+ 163,
+ 391
+ ],
+ "score": 0.75,
+ "content": "p",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 163,
+ 379,
+ 263,
+ 392
+ ],
+ "score": 1.0,
+ "content": "more graphs along with",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 263,
+ 380,
+ 279,
+ 390
+ ],
+ "score": 0.9,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 280,
+ 379,
+ 505,
+ 392
+ ],
+ "score": 1.0,
+ "content": ", whose labels are unknown. The learning objective for",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 105,
+ 390,
+ 506,
+ 403
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 390,
+ 506,
+ 403
+ ],
+ "score": 1.0,
+ "content": "fine tuning stage doesn’t change since the gradients are back-propagated from the labeled samples",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 105,
+ 401,
+ 506,
+ 414
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 401,
+ 506,
+ 414
+ ],
+ "score": 1.0,
+ "content": "only. In this setting, each node in the attention classifier can aggregate information from unlabelled",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 105,
+ 411,
+ 506,
+ 425
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 411,
+ 415,
+ 425
+ ],
+ "score": 1.0,
+ "content": "samples as well, thus allowing improved learning of the graphs features in",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 415,
+ 411,
+ 442,
+ 423
+ ],
+ "score": 0.93,
+ "content": "C ^ { G A T }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 443,
+ 411,
+ 506,
+ 425
+ ],
+ "score": 1.0,
+ "content": ". We show the",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 105,
+ 423,
+ 505,
+ 435
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 423,
+ 148,
+ 435
+ ],
+ "score": 1.0,
+ "content": "results for",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 149,
+ 425,
+ 155,
+ 435
+ ],
+ "score": 0.77,
+ "content": "p",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 156,
+ 423,
+ 441,
+ 435
+ ],
+ "score": 1.0,
+ "content": "values 25 and 50 on Letter-High and TRIANGLES datasets, whereas for",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 441,
+ 425,
+ 448,
+ 435
+ ],
+ "score": 0.81,
+ "content": "p",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 448,
+ 423,
+ 505,
+ 435
+ ],
+ "score": 1.0,
+ "content": "values 10 and",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 106,
+ 434,
+ 505,
+ 446
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 434,
+ 505,
+ 446
+ ],
+ "score": 1.0,
+ "content": "20 on Reddit and ENZYMES datasets. The results are shown in Table 9. We observe an increase in",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 106,
+ 445,
+ 447,
+ 458
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 445,
+ 447,
+ 458
+ ],
+ "score": 1.0,
+ "content": "the accuracy with increase in number of unlabeled samples during fine-tuning phase.",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ }
+ ],
+ "index": 17.5,
+ "bbox_fs": [
+ 104,
+ 365,
+ 506,
+ 458
+ ]
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 108,
+ 471,
+ 289,
+ 482
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 470,
+ 290,
+ 483
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 470,
+ 290,
+ 483
+ ],
+ "score": 1.0,
+ "content": "A.6 ADAPTATION TO ACTIVE-LEARNING",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ }
+ ],
+ "index": 22
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 491,
+ 505,
+ 547
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 490,
+ 505,
+ 505
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 490,
+ 505,
+ 505
+ ],
+ "score": 1.0,
+ "content": "In this section, we show the adaptation of our model to highly practical active learning scenario.",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ },
+ {
+ "bbox": [
+ 106,
+ 503,
+ 505,
+ 515
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 503,
+ 505,
+ 515
+ ],
+ "score": 1.0,
+ "content": "In many real world applications, we might start with few samples per class, however as the number",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 105,
+ 513,
+ 505,
+ 527
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 513,
+ 505,
+ 527
+ ],
+ "score": 1.0,
+ "content": "of samples to classify from these classes increase over time, some of these samples can be used by",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 106,
+ 524,
+ 505,
+ 537
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 524,
+ 505,
+ 537
+ ],
+ "score": 1.0,
+ "content": "the model to adaptively learn and improve with very less human intervention, since the number of",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ },
+ {
+ "bbox": [
+ 105,
+ 536,
+ 402,
+ 547
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 536,
+ 402,
+ 547
+ ],
+ "score": 1.0,
+ "content": "number of samples to be queried for theirs label can always be controlled.",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ }
+ ],
+ "index": 25,
+ "bbox_fs": [
+ 105,
+ 490,
+ 505,
+ 547
+ ]
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 552,
+ 505,
+ 586
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 552,
+ 505,
+ 564
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 552,
+ 505,
+ 564
+ ],
+ "score": 1.0,
+ "content": "To perform active-learning, we first select a random subset of size 100 for Letter-High and TRI-",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 106,
+ 564,
+ 505,
+ 574
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 564,
+ 505,
+ 574
+ ],
+ "score": 1.0,
+ "content": "ANGLES datasets as well as a random subset of size 40 for Reddit and ENZYMES datasets, which",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 105,
+ 572,
+ 506,
+ 588
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 572,
+ 153,
+ 588
+ ],
+ "score": 1.0,
+ "content": "we term as",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 153,
+ 574,
+ 190,
+ 586
+ ],
+ "score": 0.9,
+ "content": "G _ { r a n d o m }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 191,
+ 572,
+ 307,
+ 588
+ ],
+ "score": 1.0,
+ "content": ", then fine tune the model on",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 308,
+ 574,
+ 324,
+ 585
+ ],
+ "score": 0.89,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 325,
+ 572,
+ 463,
+ 588
+ ],
+ "score": 1.0,
+ "content": "and further evaluate the model on",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 464,
+ 574,
+ 501,
+ 586
+ ],
+ "score": 0.86,
+ "content": "G _ { r a n d o m }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 501,
+ 572,
+ 506,
+ 588
+ ],
+ "score": 1.0,
+ "content": ".",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ }
+ ],
+ "index": 29,
+ "bbox_fs": [
+ 105,
+ 552,
+ 506,
+ 588
+ ]
+ },
+ {
+ "type": "table",
+ "bbox": [
+ 126,
+ 654,
+ 485,
+ 731
+ ],
+ "blocks": [
+ {
+ "type": "table_caption",
+ "bbox": [
+ 107,
+ 604,
+ 505,
+ 637
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 604,
+ 505,
+ 616
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 604,
+ 496,
+ 616
+ ],
+ "score": 1.0,
+ "content": "Table 10: Active Learning Results. The value below each shot represents the number samples",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 497,
+ 604,
+ 501,
+ 614
+ ],
+ "score": 0.47,
+ "content": "l",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 502,
+ 604,
+ 505,
+ 616
+ ],
+ "score": 1.0,
+ "content": ",",
+ "type": "text"
+ }
+ ],
+ "index": 31
+ },
+ {
+ "bbox": [
+ 105,
+ 614,
+ 506,
+ 628
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 614,
+ 144,
+ 628
+ ],
+ "score": 1.0,
+ "content": "added to",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 144,
+ 615,
+ 160,
+ 626
+ ],
+ "score": 0.89,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 160,
+ 614,
+ 506,
+ 628
+ ],
+ "score": 1.0,
+ "content": "for second fine-tuning step, where “No AL” represents the model evaluation without",
+ "type": "text"
+ }
+ ],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 105,
+ 625,
+ 217,
+ 638
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 625,
+ 217,
+ 638
+ ],
+ "score": 1.0,
+ "content": "additional labeled samples.",
+ "type": "text"
+ }
+ ],
+ "index": 33
+ }
+ ],
+ "index": 32
+ },
+ {
+ "type": "table_body",
+ "bbox": [
+ 126,
+ 654,
+ 485,
+ 731
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 126,
+ 654,
+ 485,
+ 731
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 126,
+ 654,
+ 485,
+ 731
+ ],
+ "score": 0.945,
+ "html": "| Dataset | 10-shot | 20-shot |
| No AL | 15 | 25 | No AL | 15 | 25 |
| Letter-High | 73.34± 3.37 | 75.03 ± 3.24 | 76.89 ± 2.16 | 77.06 ± 1.73 | 78.44 ± 1.52 | 79.28 ± 1.36 |
| TRIANGLES | 76.02 ± 2.54 | 78.44 ± 1.84 | 79.91 ± 1.28 | 80.27 ±1.84 | 81.74 ± 2.03 | 82.58 ± 1.57 |
| Dataset | 10-shot | 20-shot |
| No AL | 10 | 20 | No AL | 10 | 20 |
| Reddit | 45.41± 3.79 | 46.88±3.14 | 47.91± 2.99 | 50.43±2.66 | 51.76± 2.32 | 53.07± 2.21 |
| ENZYMES | 60.13 ± 3.98 | 61.57 ± 3.48 | 62.25 ± 3.06 | 62.74 ± 3.64 | 63.60±3.30 | 64.97 ± 3.11 |
",
+ "type": "table",
+ "image_path": "8a7cc9daae3583429066fddac91f1e1cd620856b408f3ed283bb8bd95e385ddf.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 35,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 126,
+ 654,
+ 485,
+ 679.6666666666666
+ ],
+ "spans": [],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 126,
+ 679.6666666666666,
+ 485,
+ 705.3333333333333
+ ],
+ "spans": [],
+ "index": 35
+ },
+ {
+ "bbox": [
+ 126,
+ 705.3333333333333,
+ 485,
+ 730.9999999999999
+ ],
+ "spans": [],
+ "index": 36
+ }
+ ]
+ }
+ ],
+ "index": 33.5
+ }
+ ]
+ },
+ {
+ "preproc_blocks": [
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 82,
+ 505,
+ 214
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 81,
+ 506,
+ 96
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 81,
+ 345,
+ 96
+ ],
+ "score": 1.0,
+ "content": "Thereafter, l relatively important samples are chosen from",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 345,
+ 83,
+ 383,
+ 94
+ ],
+ "score": 0.91,
+ "content": "G _ { r a n d o m }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 384,
+ 81,
+ 439,
+ 96
+ ],
+ "score": 1.0,
+ "content": "and added to",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 440,
+ 83,
+ 456,
+ 93
+ ],
+ "score": 0.89,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 457,
+ 81,
+ 506,
+ 96
+ ],
+ "score": 1.0,
+ "content": "for another",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 106,
+ 95,
+ 504,
+ 105
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 95,
+ 504,
+ 105
+ ],
+ "score": 1.0,
+ "content": "step of fine-tuning. There can be multiple strategies for defining relative importance of a sample.",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 105,
+ 105,
+ 505,
+ 117
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 105,
+ 505,
+ 117
+ ],
+ "score": 1.0,
+ "content": "For our purpose, we define a sample’s relative importance via its predicted class probability distri-",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ },
+ {
+ "bbox": [
+ 105,
+ 115,
+ 505,
+ 127
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 115,
+ 505,
+ 127
+ ],
+ "score": 1.0,
+ "content": "bution. We sort these samples in increasing order of the difference between their highest and second",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 105,
+ 126,
+ 505,
+ 138
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 126,
+ 331,
+ 138
+ ],
+ "score": 1.0,
+ "content": "highest predicted class probabilities and choose the first",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 331,
+ 127,
+ 336,
+ 136
+ ],
+ "score": 0.51,
+ "content": "l",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 336,
+ 126,
+ 505,
+ 138
+ ],
+ "score": 1.0,
+ "content": "samples from this sorted ranking. We call",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 105,
+ 137,
+ 505,
+ 150
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 137,
+ 417,
+ 150
+ ],
+ "score": 1.0,
+ "content": "this importance relative, since each sample is evaluated with respect to the set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 417,
+ 137,
+ 434,
+ 148
+ ],
+ "score": 0.9,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 434,
+ 137,
+ 505,
+ 150
+ ],
+ "score": 1.0,
+ "content": "and thus, there is",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 105,
+ 148,
+ 505,
+ 161
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 148,
+ 505,
+ 161
+ ],
+ "score": 1.0,
+ "content": "transductive flow of information among the samples, hence defining the relative embeddings in the",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 105,
+ 159,
+ 505,
+ 172
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 159,
+ 505,
+ 172
+ ],
+ "score": 1.0,
+ "content": "space. Intuitively speaking, we have chosen the samples lying closer to separation boundary with",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 105,
+ 169,
+ 506,
+ 183
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 169,
+ 147,
+ 183
+ ],
+ "score": 1.0,
+ "content": "respect to",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 148,
+ 171,
+ 164,
+ 181
+ ],
+ "score": 0.89,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 164,
+ 169,
+ 301,
+ 183
+ ],
+ "score": 1.0,
+ "content": ". The results for various values of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 301,
+ 171,
+ 306,
+ 180
+ ],
+ "score": 0.65,
+ "content": "l",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 306,
+ 169,
+ 506,
+ 183
+ ],
+ "score": 1.0,
+ "content": "are shown in Table 10. The evaluation is done as",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 105,
+ 181,
+ 505,
+ 194
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 181,
+ 253,
+ 194
+ ],
+ "score": 1.0,
+ "content": "mentioned earlier on the unseen set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 253,
+ 181,
+ 269,
+ 192
+ ],
+ "score": 0.87,
+ "content": "G _ { U }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 269,
+ 181,
+ 505,
+ 194
+ ],
+ "score": 1.0,
+ "content": ". We observe significant improvement for all the datasets.",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 105,
+ 191,
+ 505,
+ 205
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 191,
+ 505,
+ 205
+ ],
+ "score": 1.0,
+ "content": "This shows our model is capable of selecting important samples with respect to the few existing",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 106,
+ 204,
+ 216,
+ 216
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 204,
+ 216,
+ 216
+ ],
+ "score": 1.0,
+ "content": "samples and learn actively.",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ }
+ ],
+ "index": 5.5
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 107,
+ 236,
+ 344,
+ 248
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 236,
+ 346,
+ 249
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 236,
+ 346,
+ 249
+ ],
+ "score": 1.0,
+ "content": "A.7 PERFORMANCE OF MODEL WITH 1 SUPER-CLASS",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ }
+ ],
+ "index": 12
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 261,
+ 505,
+ 415
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 259,
+ 505,
+ 275
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 259,
+ 505,
+ 275
+ ],
+ "score": 1.0,
+ "content": "From table 4 in correspondence to tables 1 and 2, one can observe that the results obtained by using",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 106,
+ 272,
+ 505,
+ 285
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 272,
+ 505,
+ 285
+ ],
+ "score": 1.0,
+ "content": "only 1 super-class which is equivalent to removing the super-classes are still better in comparison",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 106,
+ 283,
+ 505,
+ 296
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 283,
+ 505,
+ 296
+ ],
+ "score": 1.0,
+ "content": "with many GNN and graph kernel baselines. By removing the super-classes and thus forming the",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 105,
+ 294,
+ 505,
+ 306
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 294,
+ 220,
+ 306
+ ],
+ "score": 1.0,
+ "content": "super-graph solely based on",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 221,
+ 295,
+ 228,
+ 304
+ ],
+ "score": 0.27,
+ "content": "\\mathbf { k }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 228,
+ 294,
+ 505,
+ 306
+ ],
+ "score": 1.0,
+ "content": "-nearest neighbor heuristic the GAT still learns latent inter class con-",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 105,
+ 304,
+ 505,
+ 318
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 304,
+ 505,
+ 318
+ ],
+ "score": 1.0,
+ "content": "nections via information flow better than the GNNs which use MLP as their classifier. The super",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 105,
+ 315,
+ 505,
+ 330
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 315,
+ 505,
+ 330
+ ],
+ "score": 1.0,
+ "content": "graph constructed in such scenario will have arbitrary connections between classes in the beginning,",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 105,
+ 327,
+ 505,
+ 339
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 327,
+ 505,
+ 339
+ ],
+ "score": 1.0,
+ "content": "however as the GIN feature extractor learns over time the segregation in the feature space increases",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 105,
+ 337,
+ 505,
+ 351
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 337,
+ 505,
+ 351
+ ],
+ "score": 1.0,
+ "content": "leading to better inter as well intra-class connections. Despite this, the performance with super-",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 105,
+ 348,
+ 505,
+ 361
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 348,
+ 505,
+ 361
+ ],
+ "score": 1.0,
+ "content": "classes is better as this inductive bias allows the model to initiate with a better alignment in the",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 106,
+ 360,
+ 505,
+ 372
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 360,
+ 505,
+ 372
+ ],
+ "score": 1.0,
+ "content": "feature space. The silhouette score comparison for OurMethod-GAT with 1 super-class and with the",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 106,
+ 371,
+ 505,
+ 383
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 371,
+ 505,
+ 383
+ ],
+ "score": 1.0,
+ "content": "best performing number of super-classes to GIN and WL clearly indicates the multifold benefits of",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ },
+ {
+ "bbox": [
+ 105,
+ 382,
+ 505,
+ 393
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 382,
+ 505,
+ 393
+ ],
+ "score": 1.0,
+ "content": "using GNNs as a classifier via super-graph construction. The t-SNE plots for OurMethod-GAT with",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 106,
+ 392,
+ 505,
+ 405
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 392,
+ 505,
+ 405
+ ],
+ "score": 1.0,
+ "content": "only 1 super-class, GIN and WL kernel on the datasets TRIANGLES, Reddit and Letter-High are",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 105,
+ 403,
+ 293,
+ 417
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 403,
+ 293,
+ 417
+ ],
+ "score": 1.0,
+ "content": "provided in the figures 7, 8 and 9 respectively.",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ }
+ ],
+ "index": 19.5
+ },
+ {
+ "type": "table",
+ "bbox": [
+ 129,
+ 494,
+ 483,
+ 538
+ ],
+ "blocks": [
+ {
+ "type": "table_caption",
+ "bbox": [
+ 107,
+ 433,
+ 505,
+ 478
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 433,
+ 505,
+ 445
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 433,
+ 505,
+ 445
+ ],
+ "score": 1.0,
+ "content": "Table 11: Silhouette coefficients of the test classes for three models - GAT variant of Our Method",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 105,
+ 443,
+ 506,
+ 457
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 443,
+ 506,
+ 457
+ ],
+ "score": 1.0,
+ "content": "for 1 super-class which is equivalent to not using any super-classes vs the best performing number of",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 105,
+ 456,
+ 505,
+ 468
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 456,
+ 505,
+ 468
+ ],
+ "score": 1.0,
+ "content": "super-classes as well as GIN and WL on 20-shot scenario. For GIN and WL both the sub-columns",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 105,
+ 467,
+ 396,
+ 479
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 467,
+ 396,
+ 479
+ ],
+ "score": 1.0,
+ "content": "contain the same values as they don’t have any concept of super-classes.",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ }
+ ],
+ "index": 28.5
+ },
+ {
+ "type": "table_body",
+ "bbox": [
+ 129,
+ 494,
+ 483,
+ 538
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 129,
+ 494,
+ 483,
+ 538
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 129,
+ 494,
+ 483,
+ 538
+ ],
+ "score": 0.729,
+ "html": "| Method | Reddit-12K | ENZYMES | Letter-High | TRIANGLES |
| 1-SC | 2-SC | 1-SC | 2-SC | 1-SC | 3-SC | 1-SC | 3-SC |
| GIN | -0.0652 | -0.0652 | 0.0432 | 0.0432 | 0.2316 | 0.2316 | 0.1256 | 0.1256 |
| WLKernel | -0.0626 | -0.0626 | 0.0366 | 0.0366 | 0.2490 | 0.2490 | 0.0186 | 0.0186 |
| OurMethod-GAT | -0.0593 | -0.0559 | 0.1172 | 0.0989 | 0.3519 | 0.3787 | 0.3975 | 0.4508 |
",
+ "type": "table",
+ "image_path": "c644275b549300264d9b69267a41dfc443e44a9e18a8ec592f85fda7c7f5a611.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 32,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 129,
+ 494,
+ 483,
+ 508.6666666666667
+ ],
+ "spans": [],
+ "index": 31
+ },
+ {
+ "bbox": [
+ 129,
+ 508.6666666666667,
+ 483,
+ 523.3333333333334
+ ],
+ "spans": [],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 129,
+ 523.3333333333334,
+ 483,
+ 538.0
+ ],
+ "spans": [],
+ "index": 33
+ }
+ ]
+ }
+ ],
+ "index": 30.25
+ },
+ {
+ "type": "image",
+ "bbox": [
+ 122,
+ 580,
+ 489,
+ 686
+ ],
+ "blocks": [
+ {
+ "type": "image_body",
+ "bbox": [
+ 122,
+ 580,
+ 489,
+ 686
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 122,
+ 580,
+ 489,
+ 686
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 122,
+ 580,
+ 489,
+ 686
+ ],
+ "score": 0.369,
+ "type": "image",
+ "image_path": "1bafd2a3349ea4e2fe9c394cca4ab791a2cd1f65cf071a516bbd1c14891f4fbb.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 35,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 122,
+ 580,
+ 489,
+ 615.3333333333334
+ ],
+ "spans": [],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 122,
+ 615.3333333333334,
+ 489,
+ 650.6666666666667
+ ],
+ "spans": [],
+ "index": 35
+ },
+ {
+ "bbox": [
+ 122,
+ 650.6666666666667,
+ 489,
+ 686.0000000000001
+ ],
+ "spans": [],
+ "index": 36
+ }
+ ]
+ },
+ {
+ "type": "image_caption",
+ "bbox": [
+ 106,
+ 695,
+ 505,
+ 729
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 694,
+ 505,
+ 709
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 694,
+ 505,
+ 709
+ ],
+ "score": 1.0,
+ "content": "Figure 7: Visualization: t-SNE plots of the computed embeddings of test graphs on 20-shot sce-",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 105,
+ 706,
+ 505,
+ 718
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 706,
+ 505,
+ 718
+ ],
+ "score": 1.0,
+ "content": "nario from OurMethod-GAT with only 1 super-class (left), GIN (middle) and WL Kernel (right) on",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ },
+ {
+ "bbox": [
+ 106,
+ 716,
+ 193,
+ 729
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 716,
+ 193,
+ 729
+ ],
+ "score": 1.0,
+ "content": "TRIANGLES dataset.",
+ "type": "text"
+ }
+ ],
+ "index": 39
+ }
+ ],
+ "index": 38
+ }
+ ],
+ "index": 36.5
+ }
+ ],
+ "page_idx": 17,
+ "page_size": [
+ 612,
+ 792
+ ],
+ "discarded_blocks": [
+ {
+ "type": "discarded",
+ "bbox": [
+ 108,
+ 27,
+ 293,
+ 37
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 26,
+ 294,
+ 38
+ ],
+ "score": 1.0,
+ "content": "Published as a conference paper at ICLR 2020",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 300,
+ 751,
+ 311,
+ 760
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 299,
+ 750,
+ 313,
+ 763
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 299,
+ 750,
+ 313,
+ 763
+ ],
+ "score": 1.0,
+ "content": "18",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ }
+ ],
+ "para_blocks": [
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 82,
+ 505,
+ 214
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 81,
+ 506,
+ 96
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 81,
+ 345,
+ 96
+ ],
+ "score": 1.0,
+ "content": "Thereafter, l relatively important samples are chosen from",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 345,
+ 83,
+ 383,
+ 94
+ ],
+ "score": 0.91,
+ "content": "G _ { r a n d o m }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 384,
+ 81,
+ 439,
+ 96
+ ],
+ "score": 1.0,
+ "content": "and added to",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 440,
+ 83,
+ 456,
+ 93
+ ],
+ "score": 0.89,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 457,
+ 81,
+ 506,
+ 96
+ ],
+ "score": 1.0,
+ "content": "for another",
+ "type": "text"
+ }
+ ],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 106,
+ 95,
+ 504,
+ 105
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 95,
+ 504,
+ 105
+ ],
+ "score": 1.0,
+ "content": "step of fine-tuning. There can be multiple strategies for defining relative importance of a sample.",
+ "type": "text"
+ }
+ ],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 105,
+ 105,
+ 505,
+ 117
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 105,
+ 505,
+ 117
+ ],
+ "score": 1.0,
+ "content": "For our purpose, we define a sample’s relative importance via its predicted class probability distri-",
+ "type": "text"
+ }
+ ],
+ "index": 2
+ },
+ {
+ "bbox": [
+ 105,
+ 115,
+ 505,
+ 127
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 115,
+ 505,
+ 127
+ ],
+ "score": 1.0,
+ "content": "bution. We sort these samples in increasing order of the difference between their highest and second",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 105,
+ 126,
+ 505,
+ 138
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 126,
+ 331,
+ 138
+ ],
+ "score": 1.0,
+ "content": "highest predicted class probabilities and choose the first",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 331,
+ 127,
+ 336,
+ 136
+ ],
+ "score": 0.51,
+ "content": "l",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 336,
+ 126,
+ 505,
+ 138
+ ],
+ "score": 1.0,
+ "content": "samples from this sorted ranking. We call",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 105,
+ 137,
+ 505,
+ 150
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 137,
+ 417,
+ 150
+ ],
+ "score": 1.0,
+ "content": "this importance relative, since each sample is evaluated with respect to the set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 417,
+ 137,
+ 434,
+ 148
+ ],
+ "score": 0.9,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 434,
+ 137,
+ 505,
+ 150
+ ],
+ "score": 1.0,
+ "content": "and thus, there is",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ },
+ {
+ "bbox": [
+ 105,
+ 148,
+ 505,
+ 161
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 148,
+ 505,
+ 161
+ ],
+ "score": 1.0,
+ "content": "transductive flow of information among the samples, hence defining the relative embeddings in the",
+ "type": "text"
+ }
+ ],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 105,
+ 159,
+ 505,
+ 172
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 159,
+ 505,
+ 172
+ ],
+ "score": 1.0,
+ "content": "space. Intuitively speaking, we have chosen the samples lying closer to separation boundary with",
+ "type": "text"
+ }
+ ],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 105,
+ 169,
+ 506,
+ 183
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 169,
+ 147,
+ 183
+ ],
+ "score": 1.0,
+ "content": "respect to",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 148,
+ 171,
+ 164,
+ 181
+ ],
+ "score": 0.89,
+ "content": "G _ { N }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 164,
+ 169,
+ 301,
+ 183
+ ],
+ "score": 1.0,
+ "content": ". The results for various values of",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 301,
+ 171,
+ 306,
+ 180
+ ],
+ "score": 0.65,
+ "content": "l",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 306,
+ 169,
+ 506,
+ 183
+ ],
+ "score": 1.0,
+ "content": "are shown in Table 10. The evaluation is done as",
+ "type": "text"
+ }
+ ],
+ "index": 8
+ },
+ {
+ "bbox": [
+ 105,
+ 181,
+ 505,
+ 194
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 181,
+ 253,
+ 194
+ ],
+ "score": 1.0,
+ "content": "mentioned earlier on the unseen set",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 253,
+ 181,
+ 269,
+ 192
+ ],
+ "score": 0.87,
+ "content": "G _ { U }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 269,
+ 181,
+ 505,
+ 194
+ ],
+ "score": 1.0,
+ "content": ". We observe significant improvement for all the datasets.",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 105,
+ 191,
+ 505,
+ 205
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 191,
+ 505,
+ 205
+ ],
+ "score": 1.0,
+ "content": "This shows our model is capable of selecting important samples with respect to the few existing",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 106,
+ 204,
+ 216,
+ 216
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 204,
+ 216,
+ 216
+ ],
+ "score": 1.0,
+ "content": "samples and learn actively.",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ }
+ ],
+ "index": 5.5,
+ "bbox_fs": [
+ 105,
+ 81,
+ 506,
+ 216
+ ]
+ },
+ {
+ "type": "title",
+ "bbox": [
+ 107,
+ 236,
+ 344,
+ 248
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 236,
+ 346,
+ 249
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 236,
+ 346,
+ 249
+ ],
+ "score": 1.0,
+ "content": "A.7 PERFORMANCE OF MODEL WITH 1 SUPER-CLASS",
+ "type": "text"
+ }
+ ],
+ "index": 12
+ }
+ ],
+ "index": 12
+ },
+ {
+ "type": "text",
+ "bbox": [
+ 107,
+ 261,
+ 505,
+ 415
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 259,
+ 505,
+ 275
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 259,
+ 505,
+ 275
+ ],
+ "score": 1.0,
+ "content": "From table 4 in correspondence to tables 1 and 2, one can observe that the results obtained by using",
+ "type": "text"
+ }
+ ],
+ "index": 13
+ },
+ {
+ "bbox": [
+ 106,
+ 272,
+ 505,
+ 285
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 272,
+ 505,
+ 285
+ ],
+ "score": 1.0,
+ "content": "only 1 super-class which is equivalent to removing the super-classes are still better in comparison",
+ "type": "text"
+ }
+ ],
+ "index": 14
+ },
+ {
+ "bbox": [
+ 106,
+ 283,
+ 505,
+ 296
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 283,
+ 505,
+ 296
+ ],
+ "score": 1.0,
+ "content": "with many GNN and graph kernel baselines. By removing the super-classes and thus forming the",
+ "type": "text"
+ }
+ ],
+ "index": 15
+ },
+ {
+ "bbox": [
+ 105,
+ 294,
+ 505,
+ 306
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 294,
+ 220,
+ 306
+ ],
+ "score": 1.0,
+ "content": "super-graph solely based on",
+ "type": "text"
+ },
+ {
+ "bbox": [
+ 221,
+ 295,
+ 228,
+ 304
+ ],
+ "score": 0.27,
+ "content": "\\mathbf { k }",
+ "type": "inline_equation"
+ },
+ {
+ "bbox": [
+ 228,
+ 294,
+ 505,
+ 306
+ ],
+ "score": 1.0,
+ "content": "-nearest neighbor heuristic the GAT still learns latent inter class con-",
+ "type": "text"
+ }
+ ],
+ "index": 16
+ },
+ {
+ "bbox": [
+ 105,
+ 304,
+ 505,
+ 318
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 304,
+ 505,
+ 318
+ ],
+ "score": 1.0,
+ "content": "nections via information flow better than the GNNs which use MLP as their classifier. The super",
+ "type": "text"
+ }
+ ],
+ "index": 17
+ },
+ {
+ "bbox": [
+ 105,
+ 315,
+ 505,
+ 330
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 315,
+ 505,
+ 330
+ ],
+ "score": 1.0,
+ "content": "graph constructed in such scenario will have arbitrary connections between classes in the beginning,",
+ "type": "text"
+ }
+ ],
+ "index": 18
+ },
+ {
+ "bbox": [
+ 105,
+ 327,
+ 505,
+ 339
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 327,
+ 505,
+ 339
+ ],
+ "score": 1.0,
+ "content": "however as the GIN feature extractor learns over time the segregation in the feature space increases",
+ "type": "text"
+ }
+ ],
+ "index": 19
+ },
+ {
+ "bbox": [
+ 105,
+ 337,
+ 505,
+ 351
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 337,
+ 505,
+ 351
+ ],
+ "score": 1.0,
+ "content": "leading to better inter as well intra-class connections. Despite this, the performance with super-",
+ "type": "text"
+ }
+ ],
+ "index": 20
+ },
+ {
+ "bbox": [
+ 105,
+ 348,
+ 505,
+ 361
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 348,
+ 505,
+ 361
+ ],
+ "score": 1.0,
+ "content": "classes is better as this inductive bias allows the model to initiate with a better alignment in the",
+ "type": "text"
+ }
+ ],
+ "index": 21
+ },
+ {
+ "bbox": [
+ 106,
+ 360,
+ 505,
+ 372
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 360,
+ 505,
+ 372
+ ],
+ "score": 1.0,
+ "content": "feature space. The silhouette score comparison for OurMethod-GAT with 1 super-class and with the",
+ "type": "text"
+ }
+ ],
+ "index": 22
+ },
+ {
+ "bbox": [
+ 106,
+ 371,
+ 505,
+ 383
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 371,
+ 505,
+ 383
+ ],
+ "score": 1.0,
+ "content": "best performing number of super-classes to GIN and WL clearly indicates the multifold benefits of",
+ "type": "text"
+ }
+ ],
+ "index": 23
+ },
+ {
+ "bbox": [
+ 105,
+ 382,
+ 505,
+ 393
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 382,
+ 505,
+ 393
+ ],
+ "score": 1.0,
+ "content": "using GNNs as a classifier via super-graph construction. The t-SNE plots for OurMethod-GAT with",
+ "type": "text"
+ }
+ ],
+ "index": 24
+ },
+ {
+ "bbox": [
+ 106,
+ 392,
+ 505,
+ 405
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 392,
+ 505,
+ 405
+ ],
+ "score": 1.0,
+ "content": "only 1 super-class, GIN and WL kernel on the datasets TRIANGLES, Reddit and Letter-High are",
+ "type": "text"
+ }
+ ],
+ "index": 25
+ },
+ {
+ "bbox": [
+ 105,
+ 403,
+ 293,
+ 417
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 403,
+ 293,
+ 417
+ ],
+ "score": 1.0,
+ "content": "provided in the figures 7, 8 and 9 respectively.",
+ "type": "text"
+ }
+ ],
+ "index": 26
+ }
+ ],
+ "index": 19.5,
+ "bbox_fs": [
+ 105,
+ 259,
+ 505,
+ 417
+ ]
+ },
+ {
+ "type": "table",
+ "bbox": [
+ 129,
+ 494,
+ 483,
+ 538
+ ],
+ "blocks": [
+ {
+ "type": "table_caption",
+ "bbox": [
+ 107,
+ 433,
+ 505,
+ 478
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 433,
+ 505,
+ 445
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 433,
+ 505,
+ 445
+ ],
+ "score": 1.0,
+ "content": "Table 11: Silhouette coefficients of the test classes for three models - GAT variant of Our Method",
+ "type": "text"
+ }
+ ],
+ "index": 27
+ },
+ {
+ "bbox": [
+ 105,
+ 443,
+ 506,
+ 457
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 443,
+ 506,
+ 457
+ ],
+ "score": 1.0,
+ "content": "for 1 super-class which is equivalent to not using any super-classes vs the best performing number of",
+ "type": "text"
+ }
+ ],
+ "index": 28
+ },
+ {
+ "bbox": [
+ 105,
+ 456,
+ 505,
+ 468
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 456,
+ 505,
+ 468
+ ],
+ "score": 1.0,
+ "content": "super-classes as well as GIN and WL on 20-shot scenario. For GIN and WL both the sub-columns",
+ "type": "text"
+ }
+ ],
+ "index": 29
+ },
+ {
+ "bbox": [
+ 105,
+ 467,
+ 396,
+ 479
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 467,
+ 396,
+ 479
+ ],
+ "score": 1.0,
+ "content": "contain the same values as they don’t have any concept of super-classes.",
+ "type": "text"
+ }
+ ],
+ "index": 30
+ }
+ ],
+ "index": 28.5
+ },
+ {
+ "type": "table_body",
+ "bbox": [
+ 129,
+ 494,
+ 483,
+ 538
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 129,
+ 494,
+ 483,
+ 538
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 129,
+ 494,
+ 483,
+ 538
+ ],
+ "score": 0.729,
+ "html": "| Method | Reddit-12K | ENZYMES | Letter-High | TRIANGLES |
| 1-SC | 2-SC | 1-SC | 2-SC | 1-SC | 3-SC | 1-SC | 3-SC |
| GIN | -0.0652 | -0.0652 | 0.0432 | 0.0432 | 0.2316 | 0.2316 | 0.1256 | 0.1256 |
| WLKernel | -0.0626 | -0.0626 | 0.0366 | 0.0366 | 0.2490 | 0.2490 | 0.0186 | 0.0186 |
| OurMethod-GAT | -0.0593 | -0.0559 | 0.1172 | 0.0989 | 0.3519 | 0.3787 | 0.3975 | 0.4508 |
",
+ "type": "table",
+ "image_path": "c644275b549300264d9b69267a41dfc443e44a9e18a8ec592f85fda7c7f5a611.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 32,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 129,
+ 494,
+ 483,
+ 508.6666666666667
+ ],
+ "spans": [],
+ "index": 31
+ },
+ {
+ "bbox": [
+ 129,
+ 508.6666666666667,
+ 483,
+ 523.3333333333334
+ ],
+ "spans": [],
+ "index": 32
+ },
+ {
+ "bbox": [
+ 129,
+ 523.3333333333334,
+ 483,
+ 538.0
+ ],
+ "spans": [],
+ "index": 33
+ }
+ ]
+ }
+ ],
+ "index": 30.25
+ },
+ {
+ "type": "image",
+ "bbox": [
+ 122,
+ 580,
+ 489,
+ 686
+ ],
+ "blocks": [
+ {
+ "type": "image_body",
+ "bbox": [
+ 122,
+ 580,
+ 489,
+ 686
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 122,
+ 580,
+ 489,
+ 686
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 122,
+ 580,
+ 489,
+ 686
+ ],
+ "score": 0.369,
+ "type": "image",
+ "image_path": "1bafd2a3349ea4e2fe9c394cca4ab791a2cd1f65cf071a516bbd1c14891f4fbb.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 35,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 122,
+ 580,
+ 489,
+ 615.3333333333334
+ ],
+ "spans": [],
+ "index": 34
+ },
+ {
+ "bbox": [
+ 122,
+ 615.3333333333334,
+ 489,
+ 650.6666666666667
+ ],
+ "spans": [],
+ "index": 35
+ },
+ {
+ "bbox": [
+ 122,
+ 650.6666666666667,
+ 489,
+ 686.0000000000001
+ ],
+ "spans": [],
+ "index": 36
+ }
+ ]
+ },
+ {
+ "type": "image_caption",
+ "bbox": [
+ 106,
+ 695,
+ 505,
+ 729
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 694,
+ 505,
+ 709
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 694,
+ 505,
+ 709
+ ],
+ "score": 1.0,
+ "content": "Figure 7: Visualization: t-SNE plots of the computed embeddings of test graphs on 20-shot sce-",
+ "type": "text"
+ }
+ ],
+ "index": 37
+ },
+ {
+ "bbox": [
+ 105,
+ 706,
+ 505,
+ 718
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 706,
+ 505,
+ 718
+ ],
+ "score": 1.0,
+ "content": "nario from OurMethod-GAT with only 1 super-class (left), GIN (middle) and WL Kernel (right) on",
+ "type": "text"
+ }
+ ],
+ "index": 38
+ },
+ {
+ "bbox": [
+ 106,
+ 716,
+ 193,
+ 729
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 716,
+ 193,
+ 729
+ ],
+ "score": 1.0,
+ "content": "TRIANGLES dataset.",
+ "type": "text"
+ }
+ ],
+ "index": 39
+ }
+ ],
+ "index": 38
+ }
+ ],
+ "index": 36.5
+ }
+ ]
+ },
+ {
+ "preproc_blocks": [
+ {
+ "type": "image",
+ "bbox": [
+ 123,
+ 53,
+ 491,
+ 197
+ ],
+ "blocks": [
+ {
+ "type": "image_body",
+ "bbox": [
+ 123,
+ 53,
+ 491,
+ 197
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 123,
+ 53,
+ 491,
+ 197
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 123,
+ 53,
+ 491,
+ 197
+ ],
+ "score": 0.9,
+ "type": "image",
+ "image_path": "2c26d5078ecb3af998cfa6a0df81aaa3205653b35f692f0d5124b1eb6194c0bd.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 1,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 123,
+ 53,
+ 491,
+ 101.0
+ ],
+ "spans": [],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 123,
+ 101.0,
+ 491,
+ 149.0
+ ],
+ "spans": [],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 123,
+ 149.0,
+ 491,
+ 197.0
+ ],
+ "spans": [],
+ "index": 2
+ }
+ ]
+ },
+ {
+ "type": "image_caption",
+ "bbox": [
+ 106,
+ 205,
+ 505,
+ 239
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 204,
+ 505,
+ 218
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 204,
+ 505,
+ 218
+ ],
+ "score": 1.0,
+ "content": "Figure 8: Visualization: t-SNE plots of the computed embeddings of test graphs on 20-shot scenario",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 104,
+ 215,
+ 505,
+ 229
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 215,
+ 505,
+ 229
+ ],
+ "score": 1.0,
+ "content": "from OurMethod-GAT with only 1 super-class (left), GIN (middle) and WL Kernel (right) on Reddit",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 105,
+ 226,
+ 140,
+ 240
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 226,
+ 140,
+ 240
+ ],
+ "score": 1.0,
+ "content": "dataset.",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ }
+ ],
+ "index": 4
+ }
+ ],
+ "index": 2.5
+ },
+ {
+ "type": "image",
+ "bbox": [
+ 122,
+ 270,
+ 489,
+ 379
+ ],
+ "blocks": [
+ {
+ "type": "image_body",
+ "bbox": [
+ 122,
+ 270,
+ 489,
+ 379
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 122,
+ 270,
+ 489,
+ 379
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 122,
+ 270,
+ 489,
+ 379
+ ],
+ "score": 0.953,
+ "type": "image",
+ "image_path": "6612059db44bda00c40722071d4b3b857b74cec9b39a0f864e9338819d122eed.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 7,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 122,
+ 270,
+ 489,
+ 306.3333333333333
+ ],
+ "spans": [],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 122,
+ 306.3333333333333,
+ 489,
+ 342.66666666666663
+ ],
+ "spans": [],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 122,
+ 342.66666666666663,
+ 489,
+ 378.99999999999994
+ ],
+ "spans": [],
+ "index": 8
+ }
+ ]
+ },
+ {
+ "type": "image_caption",
+ "bbox": [
+ 106,
+ 388,
+ 504,
+ 422
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 387,
+ 506,
+ 401
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 387,
+ 506,
+ 401
+ ],
+ "score": 1.0,
+ "content": "Figure 9: Visualization: t-SNE plots of the computed embeddings of test graphs on 20-shot scenario",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 105,
+ 398,
+ 505,
+ 412
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 398,
+ 505,
+ 412
+ ],
+ "score": 1.0,
+ "content": "from OurMethod-GAT with only 1 super-class (left), GIN (middle) and WL Kernel (right) on Letter-",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 105,
+ 408,
+ 163,
+ 424
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 408,
+ 163,
+ 424
+ ],
+ "score": 1.0,
+ "content": "High dataset.",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ }
+ ],
+ "index": 10
+ }
+ ],
+ "index": 8.5
+ }
+ ],
+ "page_idx": 18,
+ "page_size": [
+ 612,
+ 792
+ ],
+ "discarded_blocks": [
+ {
+ "type": "discarded",
+ "bbox": [
+ 107,
+ 27,
+ 293,
+ 37
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 106,
+ 25,
+ 294,
+ 38
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 106,
+ 25,
+ 294,
+ 38
+ ],
+ "score": 1.0,
+ "content": "Published as a conference paper at ICLR 2020",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ },
+ {
+ "type": "discarded",
+ "bbox": [
+ 300,
+ 751,
+ 311,
+ 760
+ ],
+ "lines": [
+ {
+ "bbox": [
+ 299,
+ 750,
+ 313,
+ 764
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 299,
+ 750,
+ 313,
+ 764
+ ],
+ "score": 1.0,
+ "content": "19",
+ "type": "text"
+ }
+ ]
+ }
+ ]
+ }
+ ],
+ "para_blocks": [
+ {
+ "type": "image",
+ "bbox": [
+ 123,
+ 53,
+ 491,
+ 197
+ ],
+ "blocks": [
+ {
+ "type": "image_body",
+ "bbox": [
+ 123,
+ 53,
+ 491,
+ 197
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 123,
+ 53,
+ 491,
+ 197
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 123,
+ 53,
+ 491,
+ 197
+ ],
+ "score": 0.9,
+ "type": "image",
+ "image_path": "2c26d5078ecb3af998cfa6a0df81aaa3205653b35f692f0d5124b1eb6194c0bd.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 1,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 123,
+ 53,
+ 491,
+ 101.0
+ ],
+ "spans": [],
+ "index": 0
+ },
+ {
+ "bbox": [
+ 123,
+ 101.0,
+ 491,
+ 149.0
+ ],
+ "spans": [],
+ "index": 1
+ },
+ {
+ "bbox": [
+ 123,
+ 149.0,
+ 491,
+ 197.0
+ ],
+ "spans": [],
+ "index": 2
+ }
+ ]
+ },
+ {
+ "type": "image_caption",
+ "bbox": [
+ 106,
+ 205,
+ 505,
+ 239
+ ],
+ "group_id": 0,
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 204,
+ 505,
+ 218
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 204,
+ 505,
+ 218
+ ],
+ "score": 1.0,
+ "content": "Figure 8: Visualization: t-SNE plots of the computed embeddings of test graphs on 20-shot scenario",
+ "type": "text"
+ }
+ ],
+ "index": 3
+ },
+ {
+ "bbox": [
+ 104,
+ 215,
+ 505,
+ 229
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 104,
+ 215,
+ 505,
+ 229
+ ],
+ "score": 1.0,
+ "content": "from OurMethod-GAT with only 1 super-class (left), GIN (middle) and WL Kernel (right) on Reddit",
+ "type": "text"
+ }
+ ],
+ "index": 4
+ },
+ {
+ "bbox": [
+ 105,
+ 226,
+ 140,
+ 240
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 226,
+ 140,
+ 240
+ ],
+ "score": 1.0,
+ "content": "dataset.",
+ "type": "text"
+ }
+ ],
+ "index": 5
+ }
+ ],
+ "index": 4
+ }
+ ],
+ "index": 2.5
+ },
+ {
+ "type": "image",
+ "bbox": [
+ 122,
+ 270,
+ 489,
+ 379
+ ],
+ "blocks": [
+ {
+ "type": "image_body",
+ "bbox": [
+ 122,
+ 270,
+ 489,
+ 379
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 122,
+ 270,
+ 489,
+ 379
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 122,
+ 270,
+ 489,
+ 379
+ ],
+ "score": 0.953,
+ "type": "image",
+ "image_path": "6612059db44bda00c40722071d4b3b857b74cec9b39a0f864e9338819d122eed.jpg"
+ }
+ ]
+ }
+ ],
+ "index": 7,
+ "virtual_lines": [
+ {
+ "bbox": [
+ 122,
+ 270,
+ 489,
+ 306.3333333333333
+ ],
+ "spans": [],
+ "index": 6
+ },
+ {
+ "bbox": [
+ 122,
+ 306.3333333333333,
+ 489,
+ 342.66666666666663
+ ],
+ "spans": [],
+ "index": 7
+ },
+ {
+ "bbox": [
+ 122,
+ 342.66666666666663,
+ 489,
+ 378.99999999999994
+ ],
+ "spans": [],
+ "index": 8
+ }
+ ]
+ },
+ {
+ "type": "image_caption",
+ "bbox": [
+ 106,
+ 388,
+ 504,
+ 422
+ ],
+ "group_id": 1,
+ "lines": [
+ {
+ "bbox": [
+ 105,
+ 387,
+ 506,
+ 401
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 387,
+ 506,
+ 401
+ ],
+ "score": 1.0,
+ "content": "Figure 9: Visualization: t-SNE plots of the computed embeddings of test graphs on 20-shot scenario",
+ "type": "text"
+ }
+ ],
+ "index": 9
+ },
+ {
+ "bbox": [
+ 105,
+ 398,
+ 505,
+ 412
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 398,
+ 505,
+ 412
+ ],
+ "score": 1.0,
+ "content": "from OurMethod-GAT with only 1 super-class (left), GIN (middle) and WL Kernel (right) on Letter-",
+ "type": "text"
+ }
+ ],
+ "index": 10
+ },
+ {
+ "bbox": [
+ 105,
+ 408,
+ 163,
+ 424
+ ],
+ "spans": [
+ {
+ "bbox": [
+ 105,
+ 408,
+ 163,
+ 424
+ ],
+ "score": 1.0,
+ "content": "High dataset.",
+ "type": "text"
+ }
+ ],
+ "index": 11
+ }
+ ],
+ "index": 10
+ }
+ ],
+ "index": 8.5
+ }
+ ]
+ }
+ ],
+ "_backend": "pipeline",
+ "_version_name": "2.2.2"
+}
\ No newline at end of file