Research map / Graph learning / Graph self-supervised learning
Self-Supervised Node Representation Learning via Node-to-Neighbourhood Alignment
Wei Dong, Dawei Yan, Peng Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| add topology-aware positive sampling (TAPS) instead of a random neighbour as the single positive (ablation) | N2N (JL) · Cora | micro-F1: 83.46 -> 85.46 | Improved |
| add TAPS instead of a random neighbour as the single positive (ablation) | N2N (JL) · Pubmed | micro-F1: 86.20 -> 91.08 | Improved |
| add TAPS instead of a random neighbour as the single positive (ablation) | N2N (URL) · Amazon Photo | micro-F1: 86.07 -> 89.40 | Improved |
| sample more TAPS positives (1 to 5) (ablation) | N2N (JL) · Cora | micro-F1: 85.46 -> 87.10 | Improved |
| use full-neighbourhood N2N instead of one TAPS positive (ablation) | N2N (JL) · Coauthor CS | micro-F1: 92.07 -> 93.06 | Improved |
| add whitening and decorrelation (GSD) constraint to MLP-MSE (ablation) | NF-N2N (W) · Cora | micro-F1: 36.30 -> 89.00 | Improved |
| add whitening and decorrelation (GSD) constraint to MLP-MSE (ablation) | NF-N2N (W) · Com-Amazon (pure-structured) | micro-F1: 44.25 -> 87.28 | Improved |
| replace InfoNCE negative-pair contrastive baseline with NF-N2N (W) | NF-N2N · Cora Full | micro-F1: 51.29 -> 61.76 | Improved |
| add auto- or cross-correlation regularisation to whitening (WA, WC) over W alone (ablation) | NF-N2N · Cora | micro-F1: 89.00 -> 88.80 | No clear effect |
| use NF-N2N (W) on feature-less large graphs vs MLP on adjacency | NF-N2N (W) · Com-Youtube (pure-structured) | micro-F1: 38.25 -> 46.87 | Improved |
| use negative-free NF-N2N instead of contrastive baseline for memory | NF-N2N (W) · Coauthor Physics | GPU memory (MB): 3,292 -> 2,552 | Improved |
| use MLP encoder with N2N instead of GNN encoders for training time | N2N-TAPS-1 (JL) · Coauthor Physics | time per epoch (ms): 11.31 -> 0.24 | Improved |
Closest papers by meaning
- Node Representation Learning in Graph via Node-to-Neighbourhood Mutual Information Maximization (CVPR 2022 · similarity 0.96)
- Contrastive Multi-View Representation Learning on Graphs (ICML 2020 · similarity 0.90)
- Sterling: Synergistic Representation Learning on Bipartite Graphs (AAAI 2024 · similarity 0.89)
- Node Identifiers: Compact, Discrete Representations for Efficient Graph Learning (ICLR 2025 · similarity 0.88)
- S3GCL: Spectral, Swift, Spatial Graph Contrastive Learning (ICML 2024 · similarity 0.88)
- Transitivity-Preserving Graph Representation Learning for Bridging Local Connectivity and Role-Based Similarity (AAAI 2024 · similarity 0.88)
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.
