Research map / Architectures and efficiency / CNN design
RePr: Improved Training of Convolutional Filters
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 |
|---|---|---|---|
| Apply RePr with Ortho-ranked filter pruning | Vanilla CNN 3 layers · CIFAR-10 | test error: 27.9 -> 23.6 | Helped |
| Apply RePr with Ortho-ranked filter pruning | Vanilla CNN 3 layers · CIFAR-100 | test error: 52.8 -> 41.8 | Helped |
| Use RePr with Ortho rather than standard training | ResNet-20 · CIFAR-10 | test error: 8.4 -> 6.9 | Helped |
| Apply RePr for three rounds | ResNet-18 · ImageNet | top-1 test error: 30.41 -> 27.31 | Helped |
| Apply RePr for three rounds | VGG-16 · ImageNet | top-1 test error: 31.30 -> 25.50 | Helped |
| Train Inception-v1 features using RePr | VQA-LSTM-CNN · VQA v1 open-ended, All | accuracy: 60.3 -> 64.6 | Helped |
| Apply RePr to the ResNet-50 backbone | Faster R-CNN ResNet-50 RPN · COCO 2014 trainval35k + minival | mAP: 38.1 -> 41.1 | Helped |
| Add orthogonality loss to standard training (ablation) | C^3(32) · CIFAR-10 | accuracy: 72.1 -> 72.8 | Helped |
| Add orthogonality loss to RePr (ablation) | C^3(32) RePr · CIFAR-10 | accuracy: 76.4 -> 76.7 | Helped |
| Combine knowledge distillation with RePr (ablation) | C^3(32) · CIFAR-100 | accuracy: 58.2 -> 64.1 | Helped |
| Add standard dropout with probability 0.5 (ablation) | Three-layer ConvNet with RePr · CIFAR-10 | test accuracy | Hurt |
| Increase the RePr repeat count beyond a few iterations (ablation) | RePr · CIFAR-10 | test accuracy | Mixed |
Closest papers by meaning
- Provable Filter Pruning for Efficient Neural Networks (ICLR 2020 · similarity 0.93)
- CNN Filter DB: An Empirical Investigation of Trained Convolutional Filters (CVPR 2022 · similarity 0.91)
- Compressing Convolutional Neural Networks via Factorized Convolutional Filters (CVPR 2019 · similarity 0.91)
- Pruning Filters for Efficient ConvNets (ICLR 2017 · similarity 0.91)
- HRank: Filter Pruning Using High-Rank Feature Map (CVPR 2020 · similarity 0.91)
- REPrune: Channel Pruning via Kernel Representative Selection (AAAI 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.