Research map / Graph learning / Graph self-supervised learning
Disentangled Contrastive Learning on Graphs
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 |
|---|---|---|---|
| add multi-channel disentangled encoder with factor-wise contrastive loss (K>1) instead of K=1 (ablation) | DGCL · COLLAB | accuracy (%): 71.5 -> 81.2 | Helped |
| add multi-channel disentangled encoder (K>1) instead of K=1 (ablation) | DGCL · MUTAG | accuracy (%): 86.5 -> 92.1 | Helped |
| add multi-channel disentangled encoder (K>1) instead of K=1 (ablation) | DGCL · RDT-M5K | accuracy (%): 55.7 -> 56.1 | Helped |
| add prototype-based inference of latent factor distribution instead of uniform p(k|G) (ablation) | DGCL · MUTAG | accuracy (%): 89.3 -> 92.1 | Helped |
| add prototype-based inference of latent factor distribution instead of uniform p(k|G) (ablation) | DGCL · NCI1 | accuracy (%): 78.5 -> 81.9 | Helped |
| add prototype-based inference of latent factor distribution instead of uniform p(k|G) (ablation) | DGCL · COLLAB | accuracy (%): 77.5 -> 81.2 | Helped |
| add prototype-based inference of latent factor distribution instead of uniform p(k|G) (ablation) | DGCL · RDT-M5K | accuracy (%): 55.9 -> 56.1 | No clear effect |
| disentangled contrastive learning vs GraphCL under 1%/10%/20% label fine-tuning | DGCL · ogbg-molhiv | ROC-AUC (%): 67.6 -> 69.0 (1%), 70.6 -> 73.6 (10%) | Helped |
| increase number of channels K from 1 to 8 (ground-truth factor count) (ablation) | DGCL · Synthetic SBM graphs | Micro-F1 (%) | Helped |
| increase K beyond ground-truth 8 (ablation) | DGCL · Synthetic SBM graphs | Micro-F1 (%) | Hurt |
Closest papers by meaning
- Disentangled Graph Self-supervised Learning for Out-of-Distribution Generalization (ICML 2024 · similarity 0.92)
- Independence Promoted Graph Disentangled Networks (AAAI 2020 · similarity 0.92)
- A New Mechanism for Eliminating Implicit Conflict in Graph Contrastive Learning (AAAI 2024 · similarity 0.89)
- S3GCL: Spectral, Swift, Spatial Graph Contrastive Learning (ICML 2024 · similarity 0.88)
- Graph Contrastive Invariant Learning from the Causal Perspective (AAAI 2024 · similarity 0.88)
- Graph-based Unsupervised Disentangled Representation Learning via Multimodal Large Language Models (NeurIPS 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.