Research map / Architectures and efficiency / Pruning and efficient tuning
Practical Network Acceleration With Tiny Sets
Guo-Hua Wang, Jianxin Wu
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 |
|---|---|---|---|
| drop blocks instead of pruning filters | ResNet-34 compression with 500 ImageNet-1k images · ImageNet-1k val, 16.3% latency reduction | Top-1 accuracy (%): 67.0 -> 72.5 | Improved |
| use the recoverability- and latency-based Practise block-selection criterion instead of CURL (ablation) | ResNet-34 block pruning · ImageNet-1k val, 500 images, 5 blocks dropped | Top-1 accuracy (%): 64.97 -> 69.35 | Improved |
| use the recoverability- and latency-based Practise block-selection criterion instead of L2 distance (ablation) | ResNet-34 block pruning · ImageNet-1k val, 500 images, 5 blocks dropped | Top-1 accuracy (%): 64.93 -> 69.35 | Improved |
| sample uniformly rather than drawing all examples from one class (ablation) | Practise · ImageNet-1k val, ResNet-34, 500 images | Top-1 accuracy (%): 70.9 -> 71.8 | Improved |
| increase finetuning iterations from 1000 to 2000, with 100 adaptor-optimization iterations and learning rate 0.02 (ablation) | Practise · ImageNet-1k val, ResNet-34 | Top-1 accuracy (%): 71.65 -> 71.83 | Improved |
| increase adaptor optimization from 100 to 1000 iterations, keeping finetuning at 2000 iterations and learning rate 0.02 (ablation) | Practise · ImageNet-1k val, ResNet-34 | Top-1 accuracy (%): 71.83 -> 71.82 | No clear effect |
Closest papers by meaning
- Practical Network Acceleration With Tiny Sets: Hypothesis, Theory, and Algorithm (TPAMI 2024 · similarity 0.97)
- The Unreasonable Effectiveness of Random Pruning: Return of the Most Naive Baseline for Sparse Training (ICLR 2022 · similarity 0.87)
- Dense for the Price of Sparse: Improved Performance of Sparsely Initialized Networks via a Subspace Offset (ICML 2021 · similarity 0.86)
- BlockSwap: Fisher-guided Block Substitution for Network Compression on a Budget (ICLR 2020 · similarity 0.86)
- Progressive Skeletonization: Trimming more fat from a network at initialization (ICLR 2021 · similarity 0.86)
- Sparsest Models Elude Pruning: An Exposé of Pruning’s Current Capabilities (ICML 2024 · 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.
