Research map / Reinforcement learning / Visual and representation RL
N2N learning: Network to Network Compression via Policy Gradient Reinforcement Learning
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 |
|---|---|---|---|
| Use layer-removal policy to compress the teacher | VGG-13 · MNIST | accuracy: 99.54% -> 99.55% | No clear effect |
| Add layer shrinkage after layer removal (ablation) | VGG-19 student · CIFAR-10 | accuracy: 92.05% -> 91.64% | Hurt |
| Add layer shrinkage after layer removal (ablation) | ResNet-18 student · CIFAR-10 | accuracy: 91.97% -> 91.81% | Hurt |
| Add layer shrinkage after layer removal (ablation) | ResNet-34 student · CIFAR-10 | compression: 5.5x -> 10.2x | Helped |
| Add layer shrinkage after layer removal (ablation) | ResNet-18 student · SVHN | compression: 4.97x -> 19.8x | Helped |
| Use N2N architecture search instead of pruning | VGG-13 · MNIST | accuracy: 99.12% -> 99.55% | Helped |
| Use N2N architecture search instead of pruning | VGG-19 · CIFAR-10 | accuracy: 91.06% -> 92.05% | Helped |
| Use N2N-generated architecture instead of hand-designed SqueezeNet1.1 | ResNet-18 teacher · SVHN | accuracy: 89.34% -> 95.38% | Helped |
| Use N2N-generated architecture instead of FitNet-4 | ResNet-18 teacher · CIFAR-10 | accuracy: 91.33% -> 91.81% | Helped |
| Initialize compression policy from a pretrained ResNet-18 policy | ResNet-34 · first 10 iterations | reward: 0.50 -> 0.81 | Helped |
| Initialize compression policy from a pretrained ResNet-34 policy | ResNet-18 · first 10 iterations | reward: 0.53 -> 0.76 | Helped |
| Initialize compression policy from a pretrained VGG-11 policy | VGG-19 · first 10 iterations | reward: -0.07 -> 0.52 | Helped |
Closest papers by meaning
- Model compression via distillation and quantization (ICLR 2018 · similarity 0.89)
- DECORE: Deep Compression With Reinforcement Learning (CVPR 2022 · similarity 0.88)
- Few Sample Knowledge Distillation for Efficient Network Compression (CVPR 2020 · similarity 0.88)
- Topology-Aware Network Pruning using Multi-stage Graph Embedding and Reinforcement Learning (ICML 2022 · similarity 0.88)
- Model Preserving Compression for Neural Networks (NeurIPS 2022 · similarity 0.88)
- Auto-Compressing Networks (NeurIPS 2025 · similarity 0.87)
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.