Research map / Architectures and efficiency / Pruning and efficient tuning
A Gradient Flow Framework For Analyzing Network Pruning
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 |
|---|---|---|---|
| Use the proposed loss-preservation score biased toward removing small-magnitude parameters instead of plain loss preservation | Early structured pruning · CIFAR-10 and CIFAR-100 | accuracy: 91.43 -> 92.29 | Helped |
| Use the proposed loss-preservation score biased toward removing small-magnitude parameters instead of plain loss preservation | Early structured pruning · CIFAR-100 | accuracy: 67.88 -> 68.25 | Helped |
| Preserve large-magnitude parameters rather than using magnitude-agnostic pruning | Early structured pruning · CIFAR-10 | accuracy: 91.09 -> 91.41 | Helped |
| Preserve gradient norm using the absolute GraSP score rather than increase gradient norm (ablation) | GraSP · CIFAR-10, ResNet-56, 5 rounds | accuracy: 80.15 -> 91.25 | Helped |
| Preserve gradient norm using the absolute GraSP score rather than increase gradient norm (ablation) | GraSP · CIFAR-100, ResNet-56, 25 rounds | accuracy: 1.00 -> 68.02 | Helped |
| Use large temperature (T=200) rather than T=1 (ablation) | GraSP · CIFAR-10, ResNet-56, 5 rounds | accuracy: 80.15 -> 90.44 | Helped |
| Use large temperature (T=200) rather than T=1 (ablation) | GraSP · CIFAR-100, ResNet-56, 25 rounds | accuracy: 1.00 -> 57.55 | Helped |
| Use T=200 instead of T=1 for single-shot pruning (ablation) | GraSP · CIFAR-10, VGG-13, 1 round | accuracy: 91.32 -> 91.62 | Mixed |
| Use magnitude-based pruning rather than loss-preservation pruning | Early structured pruning · CIFAR-10 | accuracy: 91.09 -> 91.41 | Helped |
Closest papers by meaning
- Rethinking the Value of Network Pruning (ICLR 2019 · similarity 0.91)
- A Three-regime Model of Network Pruning (ICML 2023 · similarity 0.91)
- Pruning from Scratch (AAAI 2020 · similarity 0.90)
- Sparsest Models Elude Pruning: An Exposé of Pruning’s Current Capabilities (ICML 2024 · similarity 0.89)
- GDeR: Safeguarding Efficiency, Balancing, and Robustness via Prototypical Graph Pruning (NeurIPS 2024 · similarity 0.87)
- How Sparse Can We Prune A Deep Network: A Fundamental Limit Perspective (NeurIPS 2024 · similarity 0.87)
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.