Research map / Architectures and efficiency / Pruning and efficient tuning
Accelerate CNNs from Three Dimensions: A Comprehensive Pruning Framework
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 |
|---|---|---|---|
| Prune depth, width, and resolution together using the learned MAP | ResNet-50 · ImageNet | accuracy: 75.45% -> 75.90% | Helped |
| Prune depth, width, and resolution together rather than using a single-dimension policy | ResNet-101 · TinyImageNet | accuracy: 64.82% -> 65.27% | Helped |
| Prune depth, width, and resolution together rather than using width pruning alone | DenseNet-121 · ImageNet | accuracy: 73.58% -> 73.68% | No clear effect |
| Replace an unrestricted polynomial MAP with the proposed low-rank MAP | MAP regression on ResNet-56 · CIFAR-10 | evaluation error: 1.28% -> 0.25% | Helped |
| Use a degree-5 normal polynomial instead of the low-rank MAP | MAP regression on ResNet-56 · CIFAR-10 | evaluation error: 0.25% -> 2.31% | Hurt |
| Add ℓ2 regularization to a normal degree-5 polynomial (ablation) | MAP regression on ResNet-56 · CIFAR-10 | evaluation error: 2.31% -> 2.28% | No clear effect |
| Lower the polynomial degree to K=1 or K=2 (ablation) | MAP-based pruning · CIFAR-10 and TinyImageNet | pruning policy quality | Hurt |
| Use polynomial degree K≥3 rather than K≤2 (ablation) | MAP-based pruning · CIFAR-10 and TinyImageNet | predicted pruning policy | Helped |
| Use rank 1 for the MAP rather than a larger rank (ablation) | MAP-based pruning · CIFAR-10 | predicted optimal dimensions | No clear effect |
| Reduce the polynomial degree from K=5 to K=3 in the specialized MAP (ablation) | MAP regression on ResNet-56 · CIFAR-10 | evaluation error: 0.25% -> 0.33% | Hurt |
Closest papers by meaning
- Structured Pruning of Neural Networks With Budget-Aware Regularization (CVPR 2019 · similarity 0.92)
- Pruning from Scratch (AAAI 2020 · similarity 0.91)
- Dynamic Structure Pruning for Compressing CNNs (AAAI 2023 · similarity 0.91)
- Pruning Filters for Efficient ConvNets (ICLR 2017 · similarity 0.91)
- UPDP: A Unified Progressive Depth Pruner for CNN and Vision Transformer (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.