Research map / Graph learning / GNN architectures
Recognizing Predictive Substructures With Subgraph Information Bottleneck
Junchi Yu, Tingyang Xu, Yu Rong, Yatao Bian, Junzhou Huang, Ran He
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 SIB subgraph aggregation to GCN | GCN · DD | accuracy: 0.725 -> 0.765 | Improved |
| add SIB to GCN | GCN · PROTEINS | accuracy: 0.719 -> 0.748 | Improved |
| add SIB to GIN | GIN · DD | accuracy: 0.730 -> 0.747 | Improved |
| add SIB to GIN | GIN · MUTAG | accuracy: 0.825 -> 0.839 | Improved |
| add SIB to GAT with DropEdge | GAT+DropEdge · DD | accuracy: 0.717 -> 0.776 | Improved |
| replace attention top-k threshold with SIB subgraph | GCN · QED interpretation | absolute property bias: 0.41 -> 0.38 | Improved |
| replace attention top-k with SIB | GCN · DRD2 interpretation | absolute property bias: 0.16 -> 0.06 | Improved |
| add connectivity loss L_con (ablation) | SIB · QED interpretation | absolute property bias: 0.46 -> 0.38 | Improved |
| add MI compression loss L_MI (ablation) | SIB · DRD2 interpretation | absolute property bias: 0.21 -> 0.06 | Improved |
| add SIB for denoising | GCN · MUTAG denoising | accuracy: 0.617 -> 0.684 | Improved |
| add SIB for 3D relevant structure extraction vs attention | ResDyGCN · S3DIS floor | mIoU: 0.464 -> 0.519 | Improved |
| vary connectivity loss weight alpha from 1 to 10 (ablation) | SIB · QED | subgraph size: 0.483 -> 0.466 | No clear effect |
Closest papers by meaning
- Graph Information Bottleneck for Subgraph Recognition (ICLR 2021 · similarity 0.95)
- Improving Subgraph Recognition With Variational Graph Information Bottleneck (CVPR 2022 · similarity 0.92)
- IBCS: Learning Information Bottleneck-Constrained Denoised Causal Subgraph for Graph Classification (TPAMI 2024 · similarity 0.90)
- Causal Subgraphs and Information Bottlenecks: Redefining OOD Robustness in Graph Neural Networks (ECCV 2024 · similarity 0.89)
- A Flexible, Equivariant Framework for Subgraph GNNs via Graph Products and Graph Coarsening (NeurIPS 2024 · similarity 0.88)
- Implicit Subgraph Neural Network (ICML 2025 · 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.
