Research map / Architectures and efficiency / CNN design
Approximated Oracle Filter Pruning for Destructive CNN Width Optimization
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 VGG-16 using AOFP to 124M FLOPs | VGG-16 · CIFAR-10 | Top-1 accuracy: 93.38 -> 93.84 | Helped |
| Prune VGG-16 using AOFP to 77M FLOPs | VGG-16 · CIFAR-10 | Top-1 accuracy: 93.38 -> 93.28 | Hurt |
| Prune ResNet-50 using AOFP to 2.58G FLOPs | ResNet-50 · ImageNet | Top-1 accuracy: 75.34 -> 75.63 | Helped |
| Prune ResNet-152 using AOFP to 2.85G FLOPs | ResNet-152 · ImageNet | Top-1 accuracy: 77.37 -> 76.40 | Hurt |
| Replace AOFP's layer-specific widths with uniform slimming at a comparable FLOP budget | VGG-16 · CIFAR-10 | Top-1 accuracy: 93.84 -> 92.88 | Hurt |
| Replace AOFP's layer-specific widths with uniform slimming at a comparable FLOP budget | VGG-16 · CIFAR-10 | GPU throughput: 13903 -> 10224 | Hurt |
| Use the AOFP-discovered VGG width structure instead of the tidy baseline structure | VGG-16 · CIFAR-10 | Top-1 accuracy: 93.38 -> 94.03 | Helped |
| Train the re-designed VGG structure from scratch rather than retaining AOFP pruning initialization (ablation) | VGG re-design · CIFAR-10 | Top-1 accuracy: 94.03 -> 93.69 | Hurt |
| Use the AOFP-discovered ResNet-50 width structure instead of the tidy baseline structure | ResNet-50 · ImageNet | Top-1 accuracy: 75.34 -> 76.47 | Helped |
| Train the re-designed ResNet-50 structure from scratch rather than retaining AOFP pruning initialization (ablation) | ResNet-50 re-design · ImageNet | Top-1 accuracy: 76.47 -> 76.30 | Hurt |
| Recompute importance scores after each pruning step instead of keeping the initial scores fixed (ablation) | Oracle filter pruning on AlexNet · ImageNet | Top-1 accuracy | Helped |
Closest papers by meaning
- Pruning Filters for Efficient ConvNets (ICLR 2017 · similarity 0.91)
- Provable Filter Pruning for Efficient Neural Networks (ICLR 2020 · similarity 0.91)
- Towards Optimal Structured CNN Pruning via Generative Adversarial Learning (CVPR 2019 · similarity 0.90)
- Auto-Balanced Filter Pruning for Efficient Convolutional Neural Networks (AAAI 2018 · similarity 0.90)
- Interspace Pruning: Using Adaptive Filter Representations To Improve Training of Sparse CNNs (CVPR 2022 · similarity 0.90)
- Entropy Induced Pruning Framework for Convolutional Neural Networks (AAAI 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.