Research map / Architectures and efficiency / Pruning and efficient tuning
HRank: Filter Pruning Using High-Rank Feature Map
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 |
|---|---|---|---|
| Prune filters by low feature-map rank | HRank on VGG-16 · CIFAR-10 | FLOPs reduction: 53.5% vs. 34.3% | Helped |
| Prune filters by low feature-map rank | HRank on GoogLeNet · CIFAR-10 | Top-1 accuracy: 94.07% vs. 92.11% | Helped |
| Prune filters by low feature-map rank | HRank on ResNet-56 · CIFAR-10 | Top-1 accuracy: 93.17% vs. 93.01 | Helped |
| Prune filters by low feature-map rank | HRank on ResNet-110 · CIFAR-10 | FLOPs reduction: 58.2% vs. 48.5% | Helped |
| Prune filters by low feature-map rank | HRank on DenseNet-40 · CIFAR-10 | FLOPs reduction: 61.0% vs. 54.7% | Helped |
| Prune filters by low feature-map rank | HRank on ResNet-50 · ImageNet | Top-1 accuracy: 74.98% vs. 71.95% | Helped |
| Prune filters by low feature-map rank | HRank on ResNet-50 · ImageNet | Top-5 accuracy: 89.58% vs. 88.30% | Helped |
| Prune filters generating both low- and high-rank maps (Edge variant) (ablation) | HRank on ResNet-56 · CIFAR-10 | Top-1 accuracy | Hurt |
| Randomly prune filters instead of selecting by rank (ablation) | HRank on ResNet-56 · CIFAR-10 | Top-1 accuracy | Hurt |
| Prune high-rank filters instead of low-rank filters (ablation) | HRank on ResNet-56 · CIFAR-10 | Top-1 accuracy | Hurt |
| Freeze 15%-20% of reserved filters during fine-tuning (ablation) | HRank on ResNet-56 · CIFAR-10 | Top-1 accuracy: 93.17% -> 93.13% | No clear effect |
| Freeze 20%-25% of weights during fine-tuning (ablation) | HRank on ResNet-56 · CIFAR-10 | Top-1 accuracy: 93.17% -> 93.01% | Hurt |
Closest papers by meaning
- RePr: Improved Training of Convolutional Filters (CVPR 2019 · similarity 0.91)
- Provable Filter Pruning for Efficient Neural Networks (ICLR 2020 · similarity 0.90)
- Pruning Filters for Efficient ConvNets (ICLR 2017 · similarity 0.90)
- Towards Optimal Structured CNN Pruning via Generative Adversarial Learning (CVPR 2019 · similarity 0.90)
- Learning Filter Pruning Criteria for Deep Convolutional Neural Networks Acceleration (CVPR 2020 · similarity 0.90)
- Entropy Induced Pruning Framework for Convolutional Neural Networks (AAAI 2024 · 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.