Research map / Optimization / Implicit bias and generalization
Directional Pruning of Deep Neural Networks
Shih-Kang Chao, Zhanyu Wang, Yue Xing, Guang Cheng
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 |
|---|---|---|---|
| Use gRDA directional pruning instead of dense SGD | ResNet50 · ImageNet | testing accuracy: around 68% -> 73% | Improved |
| Use gRDA directional pruning instead of dense SGD | VGG16 · CIFAR-10, epoch 600 | testing accuracy: 90.8178(0.1779) -> 91.01(0.1464) | Improved |
| Use gRDA directional pruning instead of dense SGD | WRN28x10 · CIFAR-10, epoch 200 | testing accuracy: 94.173(0.1127) -> 94.459(0.12) | Improved |
| Use gRDA directional pruning instead of dense SGD | WRN28x10 · CIFAR-100, epoch 200 | testing accuracy: 76.529(0.169) -> 76.996(0.1713) | Improved |
| Add soft-thresholding to SGD updates | ResNet50 · ImageNet | time per iteration: 0.3964 (0.0183) -> 0.4582 (0.0166) | Hurt |
| Add soft-thresholding to SGD updates | ResNet50 · ImageNet | peak GPU memory: 14221 (376) -> 14159 (167) | No clear effect |
| Increase μ from 0.51 to 0.6 (ablation) | gRDA · CIFAR-10, VGG16, epoch 600 | sparsity: 61.4322(0.1143) -> 90.49(0.1593) | Improved |
| Increase μ from 0.51 to 0.7 (ablation) | gRDA · CIFAR-10, VGG16, epoch 600 | testing accuracy: 91.01(0.1464) -> 90.6433(0.224) | Hurt |
| Increase μ from 0.51 to 0.7 (ablation) | gRDA · CIFAR-10, VGG16, epoch 600 | sparsity: 61.4322(0.1143) -> 97.2167(0.0767) | Improved |
| Increase μ from 0.55 to 0.65 (ablation) | gRDA · CIFAR-100, WRN28x10, epoch 200 | sparsity: 90.2043(0.0508) -> 96.3042(0.0219) | Improved |
| Increase μ from 0.55 to 0.65 (ablation) | gRDA · CIFAR-100, WRN28x10, epoch 200 | testing accuracy: 76.996(0.1713) -> 76.853(0.1996) | No clear effect |
| Use the standard ImageNet step-down learning-rate schedule (ablation) | gRDA · ImageNet | testing accuracy: null | Hurt |
Closest papers by meaning
- The Unreasonable Effectiveness of Random Pruning: Return of the Most Naive Baseline for Sparse Training (ICLR 2022 · similarity 0.89)
- Pruning via Sparsity-indexed ODE: a Continuous Sparsity Viewpoint (ICML 2023 · similarity 0.89)
- Accelerate CNNs from Three Dimensions: A Comprehensive Pruning Framework (ICML 2021 · similarity 0.89)
- Learning Pruning-Friendly Networks via Frank-Wolfe: One-Shot, Any-Sparsity, And No Retraining (ICLR 2022 · similarity 0.89)
- How Sparse Can We Prune A Deep Network: A Fundamental Limit Perspective (NeurIPS 2024 · similarity 0.89)
- SAFE: Finding Sparse and Flat Minima to Improve Pruning (ICML 2025 · similarity 0.88)
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.
