Research map / Graph learning / GNN architectures
Auto Graph Encoder-Decoder for Neural Network Pruning
Sixing Yu, Arya Mazaheri, Ali Jannesari
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 |
|---|---|---|---|
| use learned graph embedding instead of handcrafted 11-feature AMC embedding (ablation) | RL pruning agent · CIFAR-10 ResNet-20, 50% FLOPs | error rate (%): 10.2% -> 5.38% | Improved |
| add stride-with-filters to stride-only handcrafted embedding (ablation) | AMC embedding · CIFAR-10 ResNet-20, 50% FLOPs | error rate (%): 31% -> 18% | Improved |
| use all 11 handcrafted features instead of stride and filters (ablation) | AMC embedding · CIFAR-10 ResNet-20, 50% FLOPs | error rate (%): 18% -> 10.2% | Improved |
| use learned graph embedding instead of no embedding (random search), 100 vs 300 episodes (ablation) | RL pruning agent · CIFAR-10 ResNet-20 | validation accuracy (%): 88.41% -> 94.6% | Improved |
| search with 100 episodes instead of 50 episodes (ablation) | AGMC · CIFAR-10 ResNet-20 | validation accuracy (%): 93.8% -> 94.6% | Improved |
| increase random search episodes from 200 to 300 (ablation) | RS · CIFAR-10 ResNet-20 | validation accuracy (%): 71% -> 88.41% | Improved |
| transfer encoder trained on ResNet-56 to ResNet-20 instead of training on ResNet-20 (ablation) | AGMC · CIFAR-10 ResNet-20, 50% FLOPs | validation accuracy (%): 94.6% -> 92.08% | Hurt |
| use AGMC instead of RS at 90% FLOPs reduction (ablation) | RL pruning agent · CIFAR-10 ResNet-56 | validation accuracy (%) | Improved |
| use AGMC instead of RS at 10% FLOPs reduction (ablation) | RL pruning agent · CIFAR-10 ResNet-56 | validation accuracy (%) | No clear effect |
| use AGMC instead of AMC | pruning policy · CIFAR-10 ResNet-56, 50% FLOPs | test accuracy (%): 90.2 -> 92.76 | Improved |
| use AGMC instead of AMC | pruning policy · CIFAR-10 ResNet-20, 50% FLOPs | test accuracy (%): 86.4 -> 91.42 | Improved |
| use AGMC instead of AMC | pruning policy · ImageNet MobileNet-v1, 40% FLOPs | test accuracy (%): 68.90 -> 69.40 | Improved |
Closest papers by meaning
- Topology-Aware Network Pruning using Multi-stage Graph Embedding and Reinforcement Learning (ICML 2022 · similarity 0.91)
- Serving Graph Compression for Graph Neural Networks (ICLR 2023 · similarity 0.87)
- Neural Epitome Search for Architecture-Agnostic Network Compression (ICLR 2020 · similarity 0.87)
- N2N learning: Network to Network Compression via Policy Gradient Reinforcement Learning (ICLR 2018 · similarity 0.87)
- Forget the Data and Fine-Tuning! Just Fold the Network to Compress (ICLR 2025 · similarity 0.86)
- Auto-Compressing Networks (NeurIPS 2025 · similarity 0.85)
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.
