Research map / Graph learning / Graph self-supervised learning
HTG-GCL: Leveraging Hierarchical Topological Granularity from Cellular Complexes for Graph Contrastive Learning
Qirui Ji, Bin Qin, Yifan Jin, Yunze Zhao, Chuxiong Sun, Changwen Zheng, Jianwen Cao, Jiangmeng Li
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 |
|---|---|---|---|
| integrate multiple ring granularities (6, 9, 12) instead of a single ring size (ablation) | HTG-GCL · NCI1, MUTAG, IMDB-B, IMDB-M (unsupervised) | accuracy (%): 79.3 -> 80.7 (NCI1, ring 6 vs full) | Improved |
| add multi-granularity decoupled contrast with uncertainty weighting (MGDC) (ablation) | HTG-GCL with three granularities · NCI1 (unsupervised) | accuracy (%): 79.7 -> 80.7 | Improved |
| add MGDC (ablation) | HTG-GCL with three granularities · IMDB-M (unsupervised) | accuracy (%): 50.9 -> 51.5 | Improved |
| add MGDC to two-granularity variant (ring 9, 12) (ablation) | HTG-GCL two granularities · NCI1 (unsupervised) | accuracy (%): 80.1 -> 79.8 | Hurt |
| add MGDC to two-granularity variant (ring 6, 9) (ablation) | HTG-GCL two granularities · MUTAG (unsupervised) | accuracy (%): 89.9 -> 90.1 | Improved |
| use four granularities (6,9,12,15) instead of three (ablation) | HTG-GCL · NCI1 and NCI109 | accuracy | Hurt |
| set decoupled loss coefficient beta to 0.01 (ablation) | HTG-GCL · NCI1 and NCI109 | accuracy | Improved |
| use HTG-GCL instead of CellCLAT | unsupervised graph classification · TU datasets | accuracy (%): 79.4 -> 80.7 (NCI1); 50.6 -> 51.5 (IMDB-M) | Improved |
| use HTG-GCL in semi-supervised setting instead of CellCLAT | 10% labels with pseudo-labeling · TU datasets | accuracy (%): 69.5 -> 70.6 (NCI109) | Improved |
| increase batch size up to 128 (ablation) | HTG-GCL · NCI1 and NCI109 (20 epochs) | accuracy | Improved |
Closest papers by meaning
- TopoGCL: Topological Graph Contrastive Learning (AAAI 2024 · similarity 0.91)
- Hierarchical Topology Isomorphism Expertise Embedded Graph Contrastive Learning (AAAI 2024 · similarity 0.91)
- S3GCL: Spectral, Swift, Spatial Graph Contrastive Learning (ICML 2024 · similarity 0.88)
- Graph Contrastive Learning with Joint Spectral Augmentation of Attribute and Topology (AAAI 2025 · similarity 0.88)
- Beyond Homophily: Graph Contrastive Learning with Macro-Micro Message Passing (AAAI 2025 · similarity 0.88)
- ReGCL: Rethinking Message Passing in Graph Contrastive Learning (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.
