Research map / Architectures and efficiency / Pruning and efficient tuning
Dynamic Network Surgery for Efficient DNNs
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 |
|---|---|---|---|
| Compress LeNet-5 with dynamic network surgery | LeNet-5 · MNIST | Top-1 error / params: 431K -> 4.0K | Helped |
| Compress LeNet-300-100 with DNS | LeNet-300-100 · MNIST | Top-1 error: 2.28% -> 1.99% | Helped |
| Compress AlexNet with DNS | AlexNet · ImageNet | Top-1 error: 43.42% -> 43.09% | Helped |
| Add connection splicing instead of pruning only on initial magnitude (ablation) | XOR network · XOR with noise | hidden neurons kept | Helped |
| Compress XOR network with DNS (40% weights zero) | XOR network · XOR with noise | prediction error: 0.31% -> 0.30% | Helped |
| DNS vs Han et al. pruning, per-layer remaining params | LeNet-5 · MNIST | remaining params: 8% -> 0.9% | Helped |
| DNS vs Han et al. pruning | LeNet-300-100 · MNIST | remaining params: 8% -> 1.8% | Helped |
| DNS vs Han et al. pruning | AlexNet · ImageNet | remaining params: 11% -> 5.7% | Helped |
| DNS vs Han et al. on conv2 of AlexNet | AlexNet conv2 · ImageNet | remaining params: 38% -> 40.6% | Hurt |
| DNS vs Han et al. on AlexNet top-1 error | AlexNet · ImageNet | Top-1 error: 42.77% -> 43.09% | No clear effect |
| DNS vs Han et al. training epochs | AlexNet · ImageNet | training epochs: at least 6.8 times decrease | Helped |
| DNS vs naive cut | AlexNet · ImageNet | Top-1 error: 57.18% -> 43.09% | Helped |
Closest papers by meaning
- DepthShrinker: A New Compression Paradigm Towards Boosting Real-Hardware Efficiency of Compact Neural Networks (ICML 2022 · similarity 0.86)
- "Lossless" Compression of Deep Neural Networks: A High-dimensional Neural Tangent Kernel Approach (NeurIPS 2022 · similarity 0.86)
- Dynamic Sparse Training: Find Efficient Sparse Network From Scratch With Trainable Masked Layers (ICLR 2020 · similarity 0.86)
- “Learning-Compression” Algorithms for Neural Net Pruning (CVPR 2018 · similarity 0.86)
- Topology-Aware Network Pruning using Multi-stage Graph Embedding and Reinforcement Learning (ICML 2022 · similarity 0.86)
- LEGACY: A Lightweight Dynamic Gradient Compression Strategy for Distributed Deep Learning (ICLR 2026 · 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.