Learning Graph Augmentations to Learn Graph Representations

Kaveh Hassani, Amir Hosein Khas Ahmadi

Abstract

Devising augmentations for graph contrastive learning is challenging due to their irregular structure, drastic distribution shifts, and nonequivalent feature spaces across datasets. We introduce LG2AR, Learning Graph Augmentations to Learn Graph Representations, which is an end-to-end automatic graph augmentation framework that helps encoders learn generalizable representations on both node and graph levels. LG2AR consists of a probabilistic policy that learns a distribution over augmentations and a set of probabilistic augmentation heads that learn distributions over augmentation parameters. We show that LG2AR achieves state-of-the-art results on 18 out of 20 graph-level and node-level benchmarks compared to previous unsupervised models under both linear and semi-supervised evaluation protocols. The source code will be released here: https://github.com/kavehhassani/lg2ar

Benchmark Results

Tested on 6 datasets 13 baseline values reported too low

Tables show presented results in global context (best 3 papers + GCN + MLP).

Classic

Amazon-Photo

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	GraphTARIF code ↗ Hyb global	0.9703 ±0.0019	Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification code ↗
2	GAT GNN baseline actual #2 vs paper #269 267 places worse	0.9260 ±0.0040	actual #2 vs paper #269 267 places worse paper 0.9260 ±0.0040 prev best: Graph Contrastive Learning with Adaptive Augmentation (WWW 2020) best today: Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification (2026) Improvement within reported uncertainty; not highlighted. paper 0.9260 ±0.0040 Δ at pub 0.0004 Δ today 0.0409
3	GAT+T code ↗ Hyb global	0.9660 ±0.0007	Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification code ↗
7	SAGE code ↗ GNN global	0.9643 ±0.0027	Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification code ↗
8	GCN GNN baseline actual #8 vs paper #294 286 places worse	0.9632 ±0.0008 best reported	actual #8 vs paper #294 286 places worse paper 0.9240 ±0.0020 best at pub: Graph Contrastive Learning with Adaptive Augmentation (WWW 2020) best today: Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification (2026) paper 0.9240 ±0.0020 Δ at pub 0.0002 Δ today 0.0392
95	LG2AR + GRU Hyb ★ proposed	0.9410 ±0.0040	This paper
172	BGRL GNN baseline actual #172 vs paper #190 18 places worse	0.9320 ±0.0030	actual #172 vs paper #190 18 places worse paper 0.9320 ±0.0030 prev best: Augmentation-Free Self-Supervised Learning on Graphs (AAAI 2021) Improvement within reported uncertainty; not highlighted. paper 0.9320 ±0.0030 Δ at pub 0.0013
191	GCA GNN baseline actual #191 vs paper #324 133 places worse	0.9320 ±0.0060 best reported	actual #191 vs paper #324 133 places worse paper 0.9220 ±0.0020 best at pub: Calibrating and Improving Graph Contrastive Learning (TMLR 2021) best today: Decoupling Weighing and Selecting for Integrating Multiple Graph Pre-training Tasks (ICLR 2024) paper 0.9220 ±0.0020 Δ at pub 0.0093 Δ today 0.0100
203	GRACE GNN baseline actual #203 vs paper #237 34 places worse	0.9280 ±0.0050	actual #203 vs paper #237 34 places worse paper 0.9280 ±0.0050 prev best: Augmentation-Free Self-Supervised Learning on Graphs (AAAI 2021) Improvement within reported uncertainty; not highlighted. paper 0.9280 ±0.0050 Δ at pub 0.0034
206	MVGRL GNN baseline actual #206 vs paper #362 156 places worse	0.9310 ±0.0040 best reported	actual #206 vs paper #362 156 places worse paper 0.9170 ±0.0010 best at pub: Graph Contrastive Learning with Adaptive Augmentation (WWW 2020) best today: Decoupling Weighing and Selecting for Integrating Multiple Graph Pre-training Tasks (ICLR 2024) paper 0.9170 ±0.0010 Δ at pub 0.0004 Δ today 0.0140
213	DGI GNN baseline actual #213 vs paper #366 153 places worse	0.9309 ±0.0008 best reported	actual #213 vs paper #366 153 places worse paper 0.9160 ±0.0020 best at pub: Graph Contrastive Learning with Adaptive Augmentation (WWW 2020) best today: Noise-Resilient Unsupervised Graph Representation Learning via Multi-Hop Feature Quality Estimation (2024) paper 0.9160 ±0.0020 Δ at pub 0.0001 Δ today 0.0149
270	LG2AR + DeepSet Hyb ★ proposed	0.9260 ±0.0050	This paper
300	LG2AR + Random Hyb ★ proposed	0.9240 ±0.0060	This paper
317	VGAE GNN baseline actual #317 vs paper #324 7 places worse	0.9224 best reported	actual #317 vs paper #324 7 places worse paper 0.9220 ±0.0010 best at pub: Barlow Graph Auto-Encoder for Unsupervised Network Embedding (2021) Difference within reported uncertainty; not counted as underperformance. paper 0.9220 ±0.0010 Δ at pub 0.0004
328	GIN GNN baseline actual #328 vs paper #402 74 places worse	0.9211 ±0.0032 best reported	actual #328 vs paper #402 74 places worse paper 0.9090 ±0.0050 best today: Union Subgraph Neural Networks (AAAI 2023) paper 0.9090 ±0.0050 Δ today 0.0121
352	GAE GNN baseline actual #352 vs paper #366 14 places worse	0.9185 ±0.0006 best reported	actual #352 vs paper #366 14 places worse paper 0.9160 ±0.0010 best at pub: Barlow Graph Auto-Encoder for Unsupervised Network Embedding (2021) best today: Graph Masked Autoencoders with Transformers (2022) paper 0.9160 ±0.0010 Δ at pub 0.0008 Δ today 0.0025
390	MLP code ↗ Trad global	0.9128 ±0.0063	Clarify Confused Nodes via Separated Learning code ↗
408	GMI GNN baseline actual #408 vs paper #416 8 places worse	0.9070 ±0.0020	actual #408 vs paper #416 8 places worse paper 0.9070 ±0.0020 prev best: Graph Contrastive Learning with Adaptive Augmentation (WWW 2020) Improvement within reported uncertainty; not highlighted. paper 0.9070 ±0.0020 Δ at pub 0.0002
443	DeepWalk + feat. Walk baseline	0.9010 ±0.0010	This paper
444	DeepWalk Walk baseline actual #444 vs paper #465 21 places worse	0.9005 ±0.0008 best reported	actual #444 vs paper #465 21 places worse paper 0.8940 ±0.0010 best at pub: Barlow Graph Auto-Encoder for Unsupervised Network Embedding (2021) best today: ENGAGE: Explanation Guided Data Augmentation for Graph Representation Learning (2023) paper 0.8940 ±0.0010 Δ at pub 0.0065 Δ today 0.0065
457	Node2Vec Walk baseline actual #457 vs paper #457	0.8970 ±0.0010	actual #457 vs paper #457 paper 0.8970 ±0.0010 prev best: Graph Contrastive Learning with Adaptive Augmentation (WWW 2020) Improvement within reported uncertainty; not highlighted. paper 0.8970 ±0.0010 Δ at pub 0.0003
531	Raw features Trad baseline actual #531 vs paper #540 9 places worse	0.7999 ±0.0090 best reported	actual #531 vs paper #540 9 places worse paper 0.7850 best at pub: Graph Contrastive Learning with Adaptive Augmentation (WWW 2020) best today: Model-Driven Graph Contrastive Learning (2025) paper 0.7850 Δ at pub 0.0003 Δ today 0.0149

Coauthor-Physics

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	GrokFormer code ↗ GT global	0.9831 ±0.0018	GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers code ↗
2	PolyFormer code ↗ GT global	0.9808 ±0.0027	GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers code ↗
3	SGFormer code ↗ GT global	0.9796 ±0.0081	GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers code ↗
6	GAT GNN baseline actual #6 vs paper #116 110 places worse	0.9782 ±0.0028 best reported	actual #6 vs paper #116 110 places worse paper 0.9550 ±0.0020 best at pub: SelfGNN: Self-supervised Graph Neural Networks without explicit negative sampling (WWW 2021) best today: GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers (ICML 2024) paper 0.9550 ±0.0020 Δ at pub 0.0070 Δ today 0.0232
9	GCN GNN baseline actual #9 vs paper #84 75 places worse	0.9774 ±0.0035 best reported	actual #9 vs paper #84 75 places worse paper 0.9570 ±0.0020 best at pub: SelfGNN: Self-supervised Graph Neural Networks without explicit negative sampling (WWW 2021) best today: GrokFormer: Graph Fourier Kolmogorov-Arnold Transformers (ICML 2024) paper 0.9570 ±0.0020 Δ at pub 0.0080 Δ today 0.0204
17	SAGE code ↗ GNN global	0.9739 ±0.0018	Clarify Confused Nodes via Separated Learning code ↗
58	MLP code ↗ Trad global	0.9597 ±0.0007	Graph Transductive Sharpening: Leveraging Unlabeled Predictions in Node Classification code ↗
63	GCA GNN baseline actual #63 vs paper #84 21 places worse	0.9590 ±0.0020 best reported	actual #63 vs paper #84 21 places worse paper 0.9570 best at pub: Barlow Graph Auto-Encoder for Unsupervised Network Embedding (2021) best today: Unifying Graph Contrastive Learning with Flexible Contextual Scopes (2022) paper 0.9570 Δ at pub 0.0003 Δ today 0.0020
69	BGRL GNN baseline actual #69 vs paper #84 15 places worse	0.9570 ±0.0010	actual #69 vs paper #84 15 places worse paper 0.9570 ±0.0010 prev best: Barlow Graph Auto-Encoder for Unsupervised Network Embedding (2021) Improvement within reported uncertainty; not highlighted. paper 0.9570 ±0.0010 Δ at pub 0.0014
88	LG2AR + GRU Hyb ★ proposed	0.9570 ±0.0020	This paper
120	LG2AR + DeepSet Hyb ★ proposed	0.9550 ±0.0010	This paper
125	MVGRL GNN baseline actual #125 vs paper #136 11 places worse	0.9544 ±0.0027 best reported	actual #125 vs paper #136 11 places worse paper 0.9530 best at pub: Graph Contrastive Learning with Adaptive Augmentation (WWW 2020) best today: HAT-GAE: Self-Supervised Graph Auto-encoders with Hierarchical Adaptive Masking and Trainable Corruption (2023) Difference within reported uncertainty; not counted as underperformance. paper 0.9530 Δ at pub 0.0003 Δ today 0.0014
128	LG2AR + Random Hyb ★ proposed	0.9540 ±0.0010	This paper
137	GIN GNN baseline actual #137 vs paper #144 7 places worse	0.9530 ±0.0010 best reported	actual #137 vs paper #144 7 places worse paper 0.9520 ±0.0010 best today: Unsupervised Graph Neural Architecture Search with Disentangled Self-supervision (NeurIPS 2024) Difference within reported uncertainty; not counted as underperformance. paper 0.9520 ±0.0010 Δ today 0.0010
146	GAE GNN baseline actual #146 vs paper #161 15 places worse	0.9520 ±0.0010 best reported	actual #146 vs paper #161 15 places worse paper 0.9492 ±0.0010 best at pub: Graph Contrastive Learning with Adaptive Augmentation (WWW 2020) best today: Revisiting Graph Autoencoders as Implicit Contrastive Learners (KDD 2024) paper 0.9492 ±0.0010 Δ today 0.0028
163	DeepWalk Walk baseline actual #163 vs paper #262 99 places worse	0.9490 best reported	actual #163 vs paper #262 99 places worse paper 0.9180 ±0.0020 best at pub: Barlow Graph Auto-Encoder for Unsupervised Network Embedding (2021) paper 0.9180 ±0.0020 Δ at pub 0.0310
162	DeepWalk + feat. Walk baseline	0.9490 ±0.0010	This paper
167	DGI GNN baseline actual #167 vs paper #174 7 places worse	0.9489 ±0.0009 best reported	actual #167 vs paper #174 7 places worse paper 0.9450 ±0.0050 best at pub: Graph Contrastive Learning with Adaptive Augmentation (WWW 2020) best today: Balancing Graph Embedding Smoothness in Self-Supervised Learning via Information-Theoretic Decomposition (WWW 2025) Difference within reported uncertainty; not counted as underperformance. paper 0.9450 ±0.0050 Δ at pub 0.0001 Δ today 0.0039
174	VGAE GNN baseline actual #174 vs paper #174	0.9452 best reported	actual #174 vs paper #174 paper 0.9450 best at pub: Graph Contrastive Learning with Adaptive Augmentation (WWW 2020) best today: Beyond Smoothing: Unsupervised Graph Representation Learning with Edge Heterophily Discriminating (AAAI 2022) paper 0.9450 Δ at pub 0.0002 Δ today 0.0002
182	Raw features Trad baseline actual #182 vs paper #206 24 places worse	0.9360	actual #182 vs paper #206 24 places worse paper 0.9360 prev best: Graph Contrastive Learning with Adaptive Augmentation (WWW 2020) best today: Model-Driven Graph Contrastive Learning (2025) paper 0.9360 Δ at pub 0.0002 Δ today 0.0072
268	Node2Vec Walk baseline actual #268 vs paper #268	0.9120	actual #268 vs paper #268 paper 0.9120 prev best: Graph Contrastive Learning with Adaptive Augmentation (WWW 2020) Improvement within reported uncertainty; not highlighted. paper 0.9120 Δ at pub 0.0001

WikiCS

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	GCN GNN baseline actual #1 vs paper #201 200 places worse	0.9302 ±0.0011 best reported	actual #1 vs paper #201 200 places worse paper 0.7720 ±0.0010 best today: Augmentation-Free Graph Contrastive Learning with Performance Guarantee (2022) paper 0.7720 ±0.0010 Δ today 0.1582
2	RSE + RGCN (70b) ext data global	0.8852 ±0.0019	Unleashing the Potential of Text-attributed Graphs: Automatic Relation Decomposition via Large …
3	LEMP+T ext data Hyb global	0.8825	Language Models as Messengers: Enhancing Message Passing in Heterophilic Graph Learning
9	SAGE GNN global	0.8771	Language Models as Messengers: Enhancing Message Passing in Heterophilic Graph Learning
31	GAT GNN baseline actual #31 vs paper #186 155 places worse	0.8533 best reported	actual #31 vs paper #186 155 places worse paper 0.7770 ±0.0010 best today: Language Models as Messengers: Enhancing Message Passing in Heterophilic Graph Learning (2025) paper 0.7770 ±0.0010 Δ today 0.0763
54	GIN GNN baseline actual #54 vs paper #218 164 places worse	0.8303 ±0.0021 best reported	actual #54 vs paper #218 164 places worse paper 0.7590 ±0.0070 best today: Unleashing the Potential of Text-attributed Graphs: Automatic Relation Decomposition via Large Language Models (2024) paper 0.7590 ±0.0070 Δ today 0.0713
94	GRACE GNN baseline actual #94 vs paper #96 2 places worse	0.8014 ±0.0048 best reported	actual #94 vs paper #96 2 places worse paper 0.8010 ±0.0050 best at pub: Large-Scale Representation Learning on Graphs via Bootstrapping (ICLR 2021) best today: Adaptive Negative Scheduling for Graph Contrastive Learning (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.8010 ±0.0050 Δ at pub 0.0004 Δ today 0.0004
100	BGRL GNN baseline actual #100 vs paper #100	0.8000 ±0.0010	actual #100 vs paper #100 paper 0.8000 ±0.0010 prev best: Large-Scale Representation Learning on Graphs via Bootstrapping (ICLR 2021) Improvement within reported uncertainty; not highlighted. paper 0.8000 ±0.0010 Δ at pub 0.0002
151	DGI GNN baseline actual #151 vs paper #226 75 places worse	0.7540 ±0.0010	actual #151 vs paper #226 75 places worse paper 0.7540 ±0.0010 prev best: Graph Barlow Twins: A self-supervised representation learning framework for graphs (2021) Improvement within reported uncertainty; not highlighted. paper 0.7540 ±0.0010 Δ at pub 0.0005
162	GCA GNN baseline actual #162 vs paper #162	0.7840 ±0.0010	actual #162 vs paper #162 paper 0.7840 ±0.0010 prev best: Graph Barlow Twins: A self-supervised representation learning framework for graphs (2021) Improvement within reported uncertainty; not highlighted. paper 0.7840 ±0.0010 Δ at pub 0.0005
181	GAE GNN baseline actual #181 vs paper #265 84 places worse	0.7020	actual #181 vs paper #265 84 places worse paper 0.7020 prev best: Barlow Graph Auto-Encoder for Unsupervised Network Embedding (2021) best today: Negative-Free Self-Supervised Gaussian Embedding of Graphs (2024) paper 0.7020 Δ at pub 0.0005 Δ today 0.0767
180	VGAE GNN baseline actual #180 vs paper #222 42 places worse	0.7787 ±0.0053 best reported	actual #180 vs paper #222 42 places worse paper 0.7560 ±0.0020 best at pub: Barlow Graph Auto-Encoder for Unsupervised Network Embedding (2021) best today: Negative-Free Self-Supervised Gaussian Embedding of Graphs (2024) paper 0.7560 ±0.0020 Δ at pub 0.0003 Δ today 0.0227
185	LG2AR + GRU Hyb ★ proposed	0.7780 ±0.0050	This paper
190	MVGRL GNN baseline actual #190 vs paper #192 2 places worse	0.7757 ±0.0046 best reported	actual #190 vs paper #192 2 places worse paper 0.7750 best at pub: Graph Barlow Twins: A self-supervised representation learning framework for graphs (2021) best today: Negative-Free Self-Supervised Gaussian Embedding of Graphs (2024) Difference within reported uncertainty; not counted as underperformance. paper 0.7750 Δ at pub 0.0002 Δ today 0.0007
198	DeepWalk Walk baseline actual #198 vs paper #238 40 places worse	0.7721 best reported	actual #198 vs paper #238 40 places worse paper 0.7440 ±0.0010 best at pub: Barlow Graph Auto-Encoder for Unsupervised Network Embedding (2021) paper 0.7440 ±0.0010 Δ at pub 0.0281
200	DeepWalk + feat. Walk baseline actual #200 vs paper #201 1 place worse	0.7721 ±0.0300 best reported	actual #200 vs paper #201 1 place worse paper 0.7720 best at pub: Graph Barlow Twins: A self-supervised representation learning framework for graphs (2021) Difference within reported uncertainty; not counted as underperformance. paper 0.7720 Δ at pub 0.0001
217	LG2AR + DeepSet Hyb ★ proposed	0.7620 ±0.0090	This paper
218	LG2AR + Random Hyb ★ proposed	0.7620 ±0.0070	This paper
234	GMI GNN baseline actual #234 vs paper #234	0.7490 ±0.0010	actual #234 vs paper #234 paper 0.7490 ±0.0010 prev best: Barlow Graph Auto-Encoder for Unsupervised Network Embedding (2021) Improvement within reported uncertainty; not highlighted. paper 0.7490 ±0.0010 Δ at pub 0.0005
237	MLP Trad global	0.7469 ±0.0085	View Space: Learning Representation across Arbitrary Graphs
256	Raw features Trad baseline	0.7198	This paper
258	Node2Vec Walk baseline actual #258 vs paper #258	0.7180 ±0.0010	actual #258 vs paper #258 paper 0.7180 ±0.0010 prev best: Augmentation-Free Self-Supervised Learning on Graphs (AAAI 2021) Improvement within reported uncertainty; not highlighted. paper 0.7180 ±0.0010 Δ at pub 0.0001

Coauthor-CS

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	DoG code ↗ GNN global	0.9690 ±0.0030	Diffusion on Graph: Augmentation of Graph Structure for Node Classification code ↗
2	GWN-fa code ↗ Hyb global	0.9667 ±0.0026	Graph Wave Networks code ↗
3	GWN-sym code ↗ GNN global	0.9666 ±0.0026	Graph Wave Networks code ↗
6	GraphSAGE code ↗ GNN global	0.9593 ±0.0025	Graph Wave Networks code ↗
12	GCN GNN baseline actual #12 vs paper #69 57 places worse	0.9514 ±0.0025 best reported	actual #12 vs paper #69 57 places worse paper 0.9300 ±0.0030 best at pub: Unifying Graph Convolutional Neural Networks and Label Propagation (2020) best today: Graph Wave Networks (WWW 2025) paper 0.9300 ±0.0030 Δ at pub 0.0060 Δ today 0.0214
21	GAT GNN baseline actual #21 vs paper #107 86 places worse	0.9437 ±0.0028 best reported	actual #21 vs paper #107 86 places worse paper 0.9230 ±0.0020 best at pub: Unifying Graph Convolutional Neural Networks and Label Propagation (2020) best today: Graph Wave Networks (WWW 2025) paper 0.9230 ±0.0020 Δ at pub 0.0150 Δ today 0.0207
39	LG2AR + GRU Hyb ★ proposed	0.9360 ±0.0030	This paper
43	DGI GNN baseline actual #43 vs paper #113 70 places worse	0.9350 ±0.0030 best reported	actual #43 vs paper #113 70 places worse paper 0.9220 ±0.0060 best at pub: Towards Graph Self-Supervised Learning with Contrastive Adjusted Zooming (2021) best today: PyG-SSL: A Graph Self-Supervised Learning Toolkit (2024) paper 0.9220 ±0.0060 Δ today 0.0130
51	GCA GNN baseline actual #51 vs paper #63 12 places worse	0.9332 ±0.0012 best reported	actual #51 vs paper #63 12 places worse paper 0.9310 best today: Beyond Smoothing: Unsupervised Graph Representation Learning with Edge Heterophily Discriminating (AAAI 2022) paper 0.9310 Δ today 0.0022
53	BGRL GNN baseline actual #53 vs paper #54 1 place worse	0.9331 ±0.0013 best reported	actual #53 vs paper #54 1 place worse paper 0.9330 ±0.0010 best today: Adaptive Negative Scheduling for Graph Contrastive Learning (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.9330 ±0.0010 Δ today 0.0001
69	GRACE GNN baseline actual #69 vs paper #148 79 places worse	0.9301 ±0.0003 best reported	actual #69 vs paper #148 79 places worse paper 0.9110 ±0.0020 best at pub: Towards Graph Self-Supervised Learning with Contrastive Adjusted Zooming (2021) best today: A Fused Gromov-Wasserstein Approach to Subgraph Contrastive Learning (TMLR 2025) paper 0.9110 ±0.0020 Δ at pub 0.0170 Δ today 0.0191
88	MLP Trad global	0.9280 ±0.0020	Revisiting the Necessity of Graph Learning and Common Graph Benchmarks
105	LG2AR + DeepSet Hyb ★ proposed	0.9240 ±0.0030	This paper
108	LG2AR + Random Hyb ★ proposed	0.9230 ±0.0020	This paper
117	VGAE GNN baseline actual #117 vs paper #118 1 place worse	0.9211 ±0.0009 best reported	actual #117 vs paper #118 1 place worse paper 0.9210 ±0.0010 best at pub: Graph Contrastive Learning with Adaptive Augmentation (WWW 2020) best today: Rethinking Graph Contrastive Learning through Relative Similarity Preservation (IJCAI 2025) Difference within reported uncertainty; not counted as underperformance. paper 0.9210 ±0.0010 Δ at pub 0.0001 Δ today 0.0001
116	MVGRL GNN baseline actual #116 vs paper #118 2 places worse	0.9211 ±0.0012 best reported	actual #116 vs paper #118 2 places worse paper 0.9210 ±0.0010 best at pub: Graph Contrastive Learning with Adaptive Augmentation (WWW 2020) best today: Adaptive Negative Scheduling for Graph Contrastive Learning (2026) Difference within reported uncertainty; not counted as underperformance. paper 0.9210 ±0.0010 Δ at pub 0.0001 Δ today 0.0001
138	GIN GNN baseline	0.9140 ±0.0020	This paper
158	Raw features Trad baseline actual #158 vs paper #158	0.9040	actual #158 vs paper #158 paper 0.9040 prev best: Graph Contrastive Learning with Adaptive Augmentation (WWW 2020) paper 0.9040 Δ at pub 0.0003
163	GAE GNN baseline actual #163 vs paper #163	0.9005 ±0.0035 best reported	actual #163 vs paper #163 paper 0.9000 ±0.0070 best at pub: Graph Contrastive Learning with Adaptive Augmentation (WWW 2020) best today: HAT-GAE: Self-Supervised Graph Auto-encoders with Hierarchical Adaptive Masking and Trainable Corruption (2023) Difference within reported uncertainty; not counted as underperformance. paper 0.9000 ±0.0070 Δ at pub 0.0001 Δ today 0.0005
176	DeepWalk + feat. Walk baseline	0.8770	This paper
184	Node2Vec Walk baseline actual #184 vs paper #184	0.8510	actual #184 vs paper #184 paper 0.8510 prev best: Graph Contrastive Learning with Adaptive Augmentation (WWW 2020) Improvement within reported uncertainty; not highlighted. paper 0.8510 Δ at pub 0.0002
188	DeepWalk Walk baseline actual #188 vs paper #188	0.8461 ±0.0022 best reported	actual #188 vs paper #188 paper 0.8460 ±0.0020 best at pub: Graph Contrastive Learning with Adaptive Augmentation (WWW 2020) best today: Negative-Free Self-Supervised Gaussian Embedding of Graphs (2024) Difference within reported uncertainty; not counted as underperformance. paper 0.8460 ±0.0020 Δ at pub 0.0001 Δ today 0.0001

TU Dortmund

MUTAG

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	MSH-GNN GNN global	0.9910 ±0.0030	Multi-Scale Harmonic Encoding for Feature-Wise Graph Message Passing
2	ID-GNN+BEC Hyb global	0.9830 ±0.0360	Depth-Adaptive Graph Neural Networks via Learnable Bakry-Émery Curvature
3	SupCosine Hyb global	0.9830 ±0.0250	Supervised Contrastive Learning with Structure Inference for Graph Classification
39	GIN GNN baseline actual #39 vs paper #237 198 places worse	0.9330 ±0.0290 best reported	actual #39 vs paper #237 198 places worse paper 0.8940 ±0.0560 best at pub: ProtGNN: Towards Self-Explaining Graph Neural Networks (AAAI 2021) Difference within reported uncertainty; not counted as underperformance. paper 0.8940 ±0.0560 Δ at pub 0.0390
63	GCN GNN baseline actual #63 vs paper #432 369 places worse	0.9220 ±0.0440 best reported	actual #63 vs paper #432 369 places worse paper 0.8560 ±0.0580 best at pub: Breaking the Limits of Message Passing Graph Neural Networks (ICML 2021) best today: Depth-Adaptive Graph Neural Networks via Learnable Bakry-Émery Curvature (KDD 2025) Difference within reported uncertainty; not counted as underperformance. paper 0.8560 ±0.0580 Δ at pub 0.0350 Δ today 0.0660
159	WL Trad baseline actual #159 vs paper #597 438 places worse	0.9040 ±0.0570 best reported	actual #159 vs paper #597 438 places worse paper 0.8070 ±0.0300 best at pub: Understanding Isomorphism Bias in Graph Data Sets (2019) best today: How Powerful are $K$-hop Message Passing Graph Neural Networks (NeurIPS 2022) paper 0.8070 ±0.0300 Δ at pub 0.0550 Δ today 0.0970
158	GraphSAGE code ↗ GNN global	0.9040 ±0.0780	Asynchronous Neural Networks for Learning in Graphs code ↗
183	GAT GNN baseline actual #183 vs paper #237 54 places worse	0.9010 ±0.0580 best reported	actual #183 vs paper #237 54 places worse paper 0.8940 ±0.0610 best at pub: Breaking the Limits of Message Passing Graph Neural Networks (ICML 2021) best today: The Expressive Power of Graph Neural Networks: A Survey (2023) Difference within reported uncertainty; not counted as underperformance. paper 0.8940 ±0.0610 Δ at pub 0.0070 Δ today 0.0070
195	LG2AR + GRU Hyb ★ proposed	0.9000 ±0.0060	This paper
215	ADGCL GNN baseline	0.8970 ±0.0100	This paper
257	InfoGraph GNN baseline actual #257 vs paper #272 15 places worse	0.8930 ±0.0514 best reported	actual #257 vs paper #272 15 places worse paper 0.8900 ±0.0110 best at pub: Graph Contrastive Learning Automated (ICML 2021) best today: Improving Expressivity of GNNs with Subgraph-specific Factor Embedded Normalization (KDD 2023) Difference within reported uncertainty; not counted as underperformance. paper 0.8900 ±0.0110 Δ at pub 0.0001 Δ today 0.0030
283	LG2AR + DeepSet Hyb ★ proposed	0.8890 ±0.0060	This paper
305	LG2AR + Random Hyb ★ proposed	0.8860 ±0.0050	This paper
332	MLG SVM baseline actual #332 vs paper #334 2 places worse	0.8794 ±0.0161 best reported	actual #332 vs paper #334 2 places worse paper 0.8790 ±0.0160 best at pub: The Multiscale Laplacian Graph Kernel (NeurIPS 2016) Difference within reported uncertainty; not counted as underperformance. paper 0.8790 ±0.0160 Δ at pub 0.0004
351	JOAO GNN baseline	0.8770 ±0.0080	This paper
354	GraphCL GNN baseline actual #354 vs paper #396 42 places worse	0.8766 ±0.0103 best reported	actual #354 vs paper #396 42 places worse paper 0.8680 ±0.0140 best at pub: Graph Contrastive Learning Automated (ICML 2021) best today: Graph Masked Autoencoders with Transformers (2022) Difference within reported uncertainty; not counted as underperformance. paper 0.8680 ±0.0140 Δ today 0.0086
365	DGK SVM baseline actual #365 vs paper #366 1 place worse	0.8744 ±0.0272 best reported	actual #365 vs paper #366 1 place worse paper 0.8740 ±0.0270 best at pub: Provably Powerful Graph Networks (NeurIPS 2019) best today: Model-Driven Graph Contrastive Learning (2025) Difference within reported uncertainty; not counted as underperformance. paper 0.8740 ±0.0270 Δ at pub 0.0004 Δ today 0.0004
403	MLP code ↗ Trad global	0.8660 ±0.0495	Breaking the Limits of Message Passing Graph Neural Networks code ↗
407	graph2vec Walk baseline actual #407 vs paper #515 108 places worse	0.8644 ±0.0125 best reported	actual #407 vs paper #515 108 places worse paper 0.8320 ±0.0960 best at pub: CogDL: A Comprehensive Library for Graph Deep Learning (WWW 2021) best today: Graph Masked Autoencoders with Transformers (2022) Difference within reported uncertainty; not counted as underperformance. paper 0.8320 ±0.0960 Δ at pub 0.0048 Δ today 0.0324
426	SP Trad baseline actual #426 vs paper #443 17 places worse	0.8580 ±0.0020 best reported	actual #426 vs paper #443 17 places worse paper 0.8520 ±0.0240 best at pub: Propagation Kernels (2014) Difference within reported uncertainty; not counted as underperformance. paper 0.8520 ±0.0240 Δ at pub 0.0060
472	Sub2Vec Walk baseline actual #472 vs paper #786 314 places worse	0.8447 ±0.0037 best reported	actual #472 vs paper #786 314 places worse paper 0.6110 ±0.1580 best at pub: Learning Graph Representation by Aggregating Subgraphs via Mutual Information Maximization (2021) best today: Graph Masked Autoencoders with Transformers (2022) Difference within reported uncertainty; not counted as underperformance. paper 0.6110 ±0.1580 Δ at pub 0.0050 Δ today 0.2337
500	random walk Walk baseline	0.8370 ±0.0150	This paper
506	GK SVM baseline actual #506 vs paper #575 69 places worse	0.8350 ±0.0060 best reported	actual #506 vs paper #575 69 places worse paper 0.8170 ±0.0210 best at pub: Contrastive Multi-View Representation Learning on Graphs (ICML 2020) best today: Supervised Contrastive Learning with Structure Inference for Graph Classification (2022) Difference within reported uncertainty; not counted as underperformance. paper 0.8170 ±0.0210 Δ today 0.0180
746	Node2Vec Walk baseline actual #746 vs paper #746	0.7263 ±0.1020 best reported	actual #746 vs paper #746 paper 0.7260 ±0.1020 best at pub: Graph Contrastive Learning Automated (ICML 2021) best today: Model-Driven Graph Contrastive Learning (2025) Difference within reported uncertainty; not counted as underperformance. paper 0.7260 ±0.1020 Δ at pub 0.0003 Δ today 0.0003

COLLAB

Global leaderboard →

#	Model	Accuracy ↑	Paper
1	WL-MLP GT global	0.9792	1-WL Expressiveness Is (Almost) All You Need
2	MSH-GNN GNN global	0.9640 ±0.0070	Multi-Scale Harmonic Encoding for Feature-Wise Graph Message Passing
3	CoCN code ↗ GNN global	0.8722 ±0.0013	Scalable Graph Compressed Convolutions code ↗
23	GCN code ↗ GNN global	0.8260 ±0.0220	Scalable Expressiveness through Preprocessed Graph Perturbations code ↗
146	InfoGraph GNN baseline actual #146 vs paper #294 148 places worse	0.7940 best reported	actual #146 vs paper #294 148 places worse paper 0.7070 ±0.0110 best at pub: CogDL: A Comprehensive Library for Graph Deep Learning (WWW 2021) paper 0.7070 ±0.0110 Δ at pub 0.0870
151	MLP Trad global	0.7920 ±0.0240	Simplifying Graph Convolutional Networks with Redundancy-Free Neighbors
183	LG2AR + GRU Hyb ★ proposed	0.7780 ±0.0020	This paper
184	LG2AR + DeepSet Hyb ★ proposed	0.7780 ±0.0020	This paper
190	LG2AR + Random Hyb ★ proposed	0.7760 ±0.0020	This paper

SOTA performance (not significantly different from the best) Proposed method (this paper) Baseline / comparison (this paper) Global context (other papers) Best at pub — reported result worse than the best known when published (best reported in green) Best at pub — reported value improved over the best known when published Improved at pub — later paper surpassed this result Best today — OK at publication; today's leaderboard best is elsewhere