Research map / Graph learning / Graph self-supervised learning
All in a Row: Compressed Convolution Networks for 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 structure features alongside node features (both vs. node features only) (ablation) | CoCN · Chameleon, Cornell, MUTAG, IMDB-B | accuracy: MUTAG 85.12 -> 87.08 (+1.96); IMDB-B 75.09 -> 77.26; Chameleon 77.98 -> 79.17; Cornell 79.19 -> 81.08 | Mixed |
| use only structure features (no node features) (ablation) | CoCN · MUTAG | accuracy: 88.35 vs 87.08 with both features | No clear effect |
| increase number of permutations (ablation) | CoCN · MUTAG, Chameleon, Cornell, Wisconsin | accuracy | Helped |
| stack multiple compressed convolution layers with unit step (ablation) | CoCN · network depth analysis | accuracy | Hurt |
| enlarge receptive field via kernel size and pooling until it exceeds number of input nodes (ablation) | CoCN · Chameleon | accuracy | Mixed |
| use kernel size k=5 instead of 3, 7, or 9 (ablation) | CoCN · Chameleon | accuracy | Helped |
| Cluster-GCN mini-batch training instead of full-batch | CoCN · amazon-ratings, questions, genius | accuracy / ROC AUC: CoCN performs comparably to full-batch classification results | No clear effect |
| proposed CoCN vs. convolutional GNN and graph pooling baselines on graph classification | CoCN · PROTEINS, NCI1, IMDB-B, IMDB-M, COLLAB | accuracy: COLLAB 86.15 vs 82.13 (DIFFPOOL) | Helped |
| proposed CoCN vs. baselines on MUTAG graph classification | CoCN · MUTAG | accuracy: 87.08 vs 88.95 (PATCHY-SAN) | Hurt |
| proposed CoCN vs. baselines on node classification | CoCN · Texas, Wisconsin, Actor | accuracy: Texas 85.21 vs 84.86 GGCN is actually higher for CoCN; Wisconsin 86.88 vs 87.45 ACMII-GCN; Actor 36.35 vs 37.54 GGCN | Hurt |
| proposed CoCN on filtered Chameleon and Squirrel (duplicate-free) | CoCN · Filtered Chameleon / Squirrel | accuracy: Filtered Chameleon 41.95 (rank 1) vs FSGNN 40.61; Squirrel 39.69 vs GCN 39.47 | Helped |
Closest papers by meaning
- Simplifying Graph Convolutional Networks (ICML 2019 · similarity 0.92)
- Beyond Graph Convolutional Network: An Interpretable Regularizer-Centered Optimization Framework (AAAI 2023 · similarity 0.92)
- Deeper Insights Into Graph Convolutional Networks for Semi-Supervised Learning (AAAI 2018 · similarity 0.91)
- Learning Convolutional Neural Networks for Graphs (ICML 2016 · similarity 0.91)
- Simple Spectral Graph Convolution (ICLR 2021 · similarity 0.91)
- Generalizing CNNs to graphs with learnable neighborhood quantization (NeurIPS 2024 · similarity 0.89)
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.