Research map / Architectures and efficiency / Pruning and efficient tuning
Pruning Filter in Filter
Fanxu Meng, Hao Cheng, Ke Li, Huixiang Luo, Xiaowei Guo, Guangming Lu, Xing Sun
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 |
|---|---|---|---|
| Stripe-wise pruning (SWP) instead of group-wise pruning, same Skeleton (ablation) | ResNet56 · CIFAR-10 | accuracy at equal params/FLOPs | Improved |
| Stripe-wise instead of group-wise pruning (ablation) | ResNet56 · CIFAR-10 | pruning ratio before a layer is fully removed: 76.64% -> 87.36% | Improved |
| Train only the Filter Skeleton with fixed filter weights (ablation) | VGG16 · CIFAR-10 | test accuracy: 79.83 | No clear effect |
| Train only the Filter Skeleton with fixed filter weights (ablation) | ResNet56 · CIFAR-10 | test accuracy: 83.82 | No clear effect |
| Train only the Filter Skeleton with fixed filter weights (ablation) | MobileNetV2 · CIFAR-10 | test accuracy: 83.52 | No clear effect |
| Increase alpha from 0.8e-5 to 1.4e-5 at delta 0.05 (ablation) | SWP ResNet56 · CIFAR-10 | accuracy: 92.73 -> 92.12 | Hurt |
| Increase delta from 0.01 to 0.09 at alpha 1e-5 (ablation) | SWP ResNet56 · CIFAR-10 | params (M): 0.45 -> 0.12 | Improved |
| Increase delta from 0.01 to 0.09 at alpha 1e-5 (ablation) | SWP ResNet56 · CIFAR-10 | accuracy: 93.25 -> 91.83 | Mixed |
| Increase delta from 0.01 to 0.09 at alpha 1e-5 (ablation) | SWP ResNet56 · CIFAR-10 | FLOPs (M): 111.68 -> 29.72 | Improved |
| Apply SWP after Network Slimming | VGG16 · CIFAR-10 | FLOPs: 1.44 -> 1.09 | Improved |
| Apply SWP after DCP | VGG19 · CIFAR-10 | FLOPs: 10.36 -> 3.40 | Improved |
| Apply SWP after GBN | ResNet56 · CIFAR-10 | FLOPs: 0.30 -> 0.24 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- HRank — HRank: Filter Pruning Using High-Rank Feature Map (CVPR 2020) · better on 2 of 2 shared results
- DMCP — DMCP: Differentiable Markov Channel Pruning for Neural Networks (CVPR 2020) · better on 1 of 1 shared result
- ThiNet — ThiNet: A Filter Level Pruning Method for Deep Neural Network Compression (ICCV 2017) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Learning Filter Pruning Criteria for Deep Convolutional Neural Networks Acceleration (CVPR 2020 · similarity 0.83)
- Rethinking the Pruning Criteria for Convolutional Neural Network (NeurIPS 2021 · similarity 0.81)
- Entropy Induced Pruning Framework for Convolutional Neural Networks (AAAI 2024 · similarity 0.81)
- Pruning Filters for Efficient ConvNets (ICLR 2017 · similarity 0.81)
- Gate Decorator: Global Filter Pruning Method for Accelerating Deep Convolutional Neural Networks (NeurIPS 2019 · similarity 0.81)
- Dynamic Structure Pruning for Compressing CNNs (AAAI 2023 · similarity 0.81)
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.
