Research map / Graph learning / GNN architectures
Understanding the Dynamics of DNNs Using Graph Modularity
Yao Lu, Wen Yang, Yunzhe Zhang, Zuohui Chen, Jinyin Chen, Qi Xuan, Zhen Wang, Xiaoniu Yang
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 |
|---|---|---|---|
| remove the plateau layers (last Conv-BN-ReLU layers) from VGG19 one by one (ablation) | VGG19 · CIFAR-10 | accuracy | No clear effect |
| remove stage-3 bottleneck blocks of ResNet152 where modularity plateaus (ablation) | ResNet152 · CIFAR-10 and ImageNet50 | accuracy | No clear effect |
| prune a VGG11 layer with small modularity increase (0.01) vs large (0.16) (ablation) | VGG11 · CIFAR-10 | accuracy drop: 0.01% vs 0.57% degradation | Hurt |
| prune a VGG11 layer with modularity increase 0.08 vs 0.26 (ablation) | VGG11 · CIFAR-100 | accuracy drop: 0.37% vs 2.11% | Hurt |
| modularity-guided layer pruning of ResNet56 vs Chen et al | ResNet56 · CIFAR-10 | FLOPs reduction: 34.80% -> 60.30% | Improved |
| modularity-guided layer pruning of ResNet56 vs Chen et al | ResNet56 · CIFAR-10 | top-1 accuracy (%): 93.29 -> 93.38 | Improved |
| modularity-guided layer pruning of ResNet56 vs DBP-0.5 | ResNet56 · CIFAR-10 | FLOPs reduction: 53.41% -> 60.30% | Improved |
| vary batch size N (200 to 5000) when building graphs (ablation) | modularity curve · CIFAR-10 (VGG16, ResNet18) | modularity curve | No clear effect |
| vary k from 3 to 11 in k-NN graph (ablation) | modularity curve · CIFAR-10 | modularity curve | No clear effect |
| increase k to 100-400 in k-NN graph (ablation) | modularity curve · CIFAR-10 | modularity value | Hurt |
| replace cosine similarity with Pearson correlation for the similarity matrix (ablation) | modularity curve · CIFAR-10 | modularity curve | No clear effect |
| use randomly initialized models instead of trained models (ablation) | modularity curve · CIFAR-10 (VGG16, ResNet18) | modularity trend | Hurt |
Closest papers by meaning
- Breaking Neural Network Scaling Laws with Modularity (ICLR 2025 · similarity 0.88)
- How Graph Neural Networks Learn: Lessons from Training Dynamics (ICML 2024 · similarity 0.87)
- Neural networks are more modular than single neurons suggest (NeurIPS 2026 · similarity 0.87)
- When Representations Align: Universality in Representation Learning Dynamics (ICML 2024 · similarity 0.87)
- Feature Learning Dynamics in Infinite-Depth Neural Networks (NeurIPS 2026 · similarity 0.86)
- Feature Distribution on Graph Topology Mediates the Effect of Graph Convolution: Homophily Perspective (ICML 2024 · similarity 0.86)
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.
