Research map / Architectures and efficiency / Pruning and efficient tuning
Group Sparsity: The Hinge Between Filter Pruning and Decomposition for Network Compression
Check 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 |
|---|---|---|---|
| Use Hinge compression instead of KSE at the 50% FLOP operating point | ResNet-56 · CIFAR10 | Top-1 error: 6.77 -> 6.31 | Helped |
| Use Hinge compression instead of KSE at the 24% FLOP operating point | ResNet-56 · CIFAR10 | Top-1 error: 8.00 -> 7.35 | Helped |
| Use Hinge compression instead of SSS | ResNet-20 · CIFAR10 | Top-1 error: 9.15 -> 8.16 | Helped |
| Use Hinge compression instead of SSS | ResNet-164 · CIFAR10 | Top-1 error: 5.78 -> 5.4 | Helped |
| Use Hinge compression instead of SSS | ResNeXt-20 · CIFAR100 | Top-1 error: 30.60 -> 28.74 | Helped |
| Use Hinge compression instead of SSS | ResNeXt-164 · CIFAR100 | Top-1 error: 26.71 -> 22.56 | Helped |
| Add regularization factor annealing during compression (ablation) | Hinge on WRN · CIFAR100 | Top-1 error: 23.61 -> 21.79; 23.13 -> 22.06 | Helped |
| Replace ℓ1 with ℓ1/2 regularization (ablation) | Hinge on ResNet-56 · CIFAR10 | Top-1 error: 6.53 -> 6.31 | Helped |
| Replace ℓ1 with ℓ1-2 regularization (ablation) | Hinge on ResNet-56 · CIFAR10 | Top-1 error: 6.53 -> 6.56 | Hurt |
| Replace ℓ1 with logsum regularization (ablation) | Hinge on ResNet-56 · CIFAR10 | Top-1 error: 6.53 -> 6.53 | No clear effect |
| Set threshold T to 0.005 and α to 0.01 (ablation) | Hinge with ℓ1 regularization on ResNet-56 · CIFAR10 | Top-1 error: 6.53 -> 6.37 | Helped |
| Use Hinge instead of FPGM | ResNet50 · ImageNet2012 | Top-1 error: 25.17 -> 25.30 | Hurt |
Closest papers by meaning
- Group Fisher Pruning for Practical Network Compression (ICML 2021 · similarity 0.88)
- DPFPS: Dynamic and Progressive Filter Pruning for Compressing Convolutional Neural Networks from Scratch (AAAI 2021 · similarity 0.88)
- Convolutional Neural Network Pruning With Structural Redundancy Reduction (CVPR 2021 · similarity 0.88)
- Sparsest Models Elude Pruning: An Exposé of Pruning’s Current Capabilities (ICML 2024 · similarity 0.87)
- Towards Compact CNNs via Collaborative Compression (CVPR 2021 · similarity 0.87)
- Learning Efficient Image Super-Resolution Networks via Structure-Regularized Pruning (ICLR 2022 · 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.