Research map / Architectures and efficiency / Pruning and efficient tuning
Discrimination-aware Network Pruning for Deep Model Compression
Jing Liu, Bohan Zhuang, Zhuangwei Zhuang, Yong Fang Guo, Junzhou Huang, Jinhui Zhu, Mingkui Tan
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 |
|---|---|---|---|
| add discrimination-aware loss (DCP vs random channel selection) (ablation) | DCP · CIFAR-10 ResNet-56 | Top-1 error (%): 6.66 -> 6.28 | Improved |
| add discrimination-aware loss (DCP vs WM+) (ablation) | DCP · CIFAR-10 VGGNet | Top-1 error (%): 6.12 -> 5.71 | Improved |
| add adaptive stopping conditions (Adapt-DCP vs DCP) | DCP · CIFAR-10 ResNet-56 | #Param. reduction (%): 49.67 -> 68.48 | Improved |
| use stopping condition 2 instead of 1 (ablation) | Adapt-DCP · CIFAR-10 VGGNet (eps 1e-4) | FLOPs reduction (%): 30.80 -> 69.81 | Improved |
| use stopping condition 2 instead of 1 (ablation) | Adapt-DCP · CIFAR-10 ResNet-56 | Top-1 error (%): 6.36 -> 6.23 | Improved |
| tighten tolerance epsilon from 5e-4 to 1e-4 (ablation) | Adapt-DCP · CIFAR-10 VGGNet | Top-1 error (%): 6.36 -> 5.45 | Improved |
| add reconstruction loss to cross-entropy lambda weight 1 vs 0 (ablation) | DCP · CIFAR-10 ResNet-56 | Top-1 error (%): 5.94 -> 5.80 | Improved |
| single round fine-tuning instead of conference version (ablation) | DCP · CIFAR-10 ResNet-56 | Pruning time (hours): 7.58 -> 2.83 | Improved |
| single round fine-tuning (ablation) | DCP · ILSVRC-12 ResNet-18 | Top-1 error (%): 35.88 -> 35.89 | No clear effect |
| add feature reusing (ablation) | DCP · CIFAR-10 ResNet-56 | Top-1 error (%): 6.26 -> 6.28 | No clear effect |
| increase B from 1 to 4 (ablation) | DCP · CIFAR-10 ResNet-56 | Top-1 error (%): 6.22 -> 6.32 | Mixed |
| add DKP kernel pruning to DCP | DCP · CIFAR-10 ResNet-56 | Top-1 error (%): 6.28 -> 6.23 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ThiNet — ThiNet: A Filter Level Pruning Method for Deep Neural Network Compression (ICCV 2017) · better on 3 of 3 shared results
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
- Discrimination-aware Channel Pruning for Deep Neural Networks (NeurIPS 2018 · similarity 0.95)
- Network Pruning via Performance Maximization (CVPR 2021 · similarity 0.90)
- Rethinking the Smaller-Norm-Less-Informative Assumption in Channel Pruning of Convolution Layers (ICLR 2018 · similarity 0.90)
- Learning Pruning-Friendly Networks via Frank-Wolfe: One-Shot, Any-Sparsity, And No Retraining (ICLR 2022 · similarity 0.90)
- REPrune: Channel Pruning via Kernel Representative Selection (AAAI 2024 · similarity 0.89)
- How Sparse Can We Prune A Deep Network: A Fundamental Limit Perspective (NeurIPS 2024 · similarity 0.89)
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.
