Research map / Graph learning / Graph self-supervised learning
Simple and Asymmetric Graph Contrastive Learning without Augmentations
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use GraphACL instead of the strongest reported baseline | node classification with GCN representations · heterophilic datasets: Squirrel, Chameleon, Crocodile, Wisconsin, Cornell, Texas, Roman, arXiv-year | accuracy: 54.05 ± 0.13 vs 52.94 ± 0.88; 69.12 ± 0.24 vs 68.74 ± 0.49; 66.17 ± 0.24 vs 62.98 ± 0.51; 69.22 ± 0.40 vs 65.28 ± 0.52; | Helped |
| Use GraphACL instead of the strongest reported baseline | node classification with GCN representations · homophilic datasets: Cora, Citeseer, Pubmed, Computer, Photo, Arxiv | accuracy: 84.20 ± 0.31 vs 84.00 ± 0.40; 73.63 ± 0.22 vs 73.26 ± 0.50; 82.02 ± 0.15 vs 81.82 ± 0.50; 89.80 ± 0.25 vs 89.69 ± 0.37; | Helped |
| Use GraphACL rather than L-GCL | node classification with GCN representations · Actor | accuracy: 30.03 ± 0.13 vs 32.55 ± 1.18 | Hurt |
| Replace the GraphACL objective with the combination loss L_COM (ablation) | GraphACL · Cora, Pubmed, Cornell, arXiv-year | accuracy: 84.20 ± 0.31 -> 83.08 ± 0.75; 82.02 ± 0.15 -> 81.31 ± 0.92; 59.33 ± 1.48 -> 58.10 ± 2.89; 47.21 ± 0.39 -> 45.79 ± 0.62 | Hurt |
| Add the asymmetric encoder architecture (ablation) | GraphACL without asymmetric encoder · Cora and Cornell | accuracy: 82.21 ± 0.60 -> 84.20 ± 0.31; 57.13 ± 0.13 -> 59.33 ± 1.48 | Helped |
| Add the uniformity loss with negative samples (ablation) | GraphACL without uniformity loss · Cora | accuracy: 81.39 ± 0.20 -> 84.20 ± 0.31 | Helped |
| Add the uniformity loss with negative samples (ablation) | GraphACL without uniformity loss · arXiv-year | accuracy: 46.54 ± 0.28 -> 47.21 ± 0.39 | Hurt |
| Add both the asymmetric encoder architecture and uniformity loss (ablation) | GraphACL without both components · Cornell and arXiv-year | accuracy: 42.03 ± 1.21 -> 59.33 ± 1.48; 42.71 ± 0.20 -> 47.21 ± 0.39 | Helped |
Closest papers by meaning
- S3GCL: Spectral, Swift, Spatial Graph Contrastive Learning (ICML 2024 · similarity 0.92)
- Beyond Homophily: Graph Contrastive Learning with Macro-Micro Message Passing (AAAI 2025 · similarity 0.91)
- Hierarchical Topology Isomorphism Expertise Embedded Graph Contrastive Learning (AAAI 2024 · similarity 0.91)
- Does GCL Need a Large Number of Negative Samples? Enhancing Graph Contrastive Learning with Effective and Efficient Negative Sampling (AAAI 2025 · similarity 0.91)
- ReGCL: Rethinking Message Passing in Graph Contrastive Learning (AAAI 2024 · similarity 0.91)
- Edge Contrastive Learning: An Augmentation-Free Graph Contrastive Learning Model (AAAI 2025 · similarity 0.91)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.