Research map / Architectures and efficiency / Pruning and efficient tuning
Dynamic Structure Pruning for Compressing CNNs
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 |
|---|---|---|---|
| learn dynamic filter groups and prune group channels rather than use channel pruning | CNNs · ImageNet, ResNet50 | FLOPs pruned (%): 71.85 vs 55.50 | Helped |
| learn dynamic filter groups and prune group channels rather than use channel pruning | CNNs · ImageNet, ResNet50 | top-1 accuracy: 76.22 vs 76.15 | Helped |
| use 4 groups instead of 2 (ablation) | DSP · CIFAR-10, ResNet56 | test accuracy: 93.92 -> 94.19 | Helped |
| use 4 groups instead of 2 (ablation) | DSP · CIFAR-10, ResNet56 | FLOPs pruned (%): 58.56 -> 60.56 | Helped |
| use 4 groups instead of 3 (ablation) | DSP · CIFAR-10, ResNet56 | test accuracy: 94.08 -> 94.19 | Helped |
| use 4 groups instead of 3 (ablation) | DSP · CIFAR-10, ResNet56 | FLOPs pruned (%): 59.89 -> 60.56 | Helped |
| use 4 groups instead of 2 (ablation) | DSP · ImageNet, ResNet50 | top-1 accuracy: 76.16 -> 76.22 | Helped |
| use 4 groups instead of 2 (ablation) | DSP · ImageNet, ResNet50 | FLOPs pruned (%): 67.78 -> 71.85 | Helped |
| use 4 groups instead of 3 (ablation) | DSP · ImageNet, ResNet50 | top-1 accuracy: 76.20 -> 76.22 | Helped |
| use 4 groups instead of 3 (ablation) | DSP · ImageNet, ResNet50 | FLOPs pruned (%): 70.98 -> 71.85 | Helped |
| use 3 groups instead of 2 (ablation) | DSP · ImageNet inference on GPU | accuracy-latency tradeoff | Hurt |
| learn groups with differentiable group learning rather than average brute-force groupings | DSP · MNIST toy network, pruning rates 0.25, 0.5, and 0.75 | validation accuracy: 96.05 vs 92.85; 81.77 vs 66.33; 45.12 vs 41.63 | Helped |
Closest papers by meaning
- Convolutional Neural Network Pruning With Structural Redundancy Reduction (CVPR 2021 · similarity 0.91)
- Accelerate CNNs from Three Dimensions: A Comprehensive Pruning Framework (ICML 2021 · similarity 0.91)
- Towards Optimal Structured CNN Pruning via Generative Adversarial Learning (CVPR 2019 · similarity 0.91)
- Pruning from Scratch (AAAI 2020 · similarity 0.91)
- BilevelPruning: Unified Dynamic and Static Channel Pruning for Convolutional Neural Networks (CVPR 2024 · similarity 0.90)
- Structured Pruning of Neural Networks With Budget-Aware Regularization (CVPR 2019 · similarity 0.90)
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.