Research map / Architectures and efficiency / RNNs and normalization
Channel Regeneration: Improving Channel Utilization for Compact DNNs
Ankit Sharma, Hassan Foroosh
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| Regenerate inhibited channels by resetting their BN gamma to the network median at regular intervals | compact CNNs · CIFAR-100 | accuracy: 69.10 -> 71.90 | Improved |
| Regenerate inhibited channels at regular intervals | MobileNetv2 · CIFAR-10 | accuracy: 90.74 -> 92.56 | Improved |
| Regenerate inhibited channels at regular intervals | compact architectures · Tiny-ImageNet | accuracy: 41.87 -> 46.40 | Improved |
| Regenerate inhibited channels at regular intervals | MobileNetv2 · ImageNet | accuracy: 71.70 -> 72.08 | Improved |
| Regenerate inhibited channels during segmentation training | MobileNetv2 · ADE20K | semantic segmentation performance: 69.69 -> 70.82 | Improved |
| Set regenerated BN gamma to the network median rather than the mean (ablation) | MobileNetv2 · CIFAR-100 | accuracy: 71.50 -> 71.90 | Improved |
| Set regenerated BN gamma to the network median rather than 0.2 (ablation) | MobileNetv2 · CIFAR-100 | accuracy: 71.25 -> 71.90 | Improved |
| Apply channel regeneration iteratively instead of only once (ablation) | MobileNetv2 · CIFAR-100 | accuracy | Improved |
| Vary the regeneration interval (ablation) | ResNet-34 · CIFAR-10 | accuracy: 95.28 -> 95.69 | Mixed |
| Raise BN gamma for inhibited channels to revive channel gradient updates | pruned models · CIFAR-100 | accuracy: 69.00 -> 69.42 | Improved |
| Raise BN gamma for inhibited channels in a pruned network | ResNet-18 pruned with FPGM · ImageNet | accuracy: 66.11 -> 66.38 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ShuffleNetv2 — ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design (ECCV 2018) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Rethinking the Smaller-Norm-Less-Informative Assumption in Channel Pruning of Convolution Layers (ICLR 2018 · similarity 0.86)
- Dynamic Neural Regeneration: Enhancing Deep Learning Generalization on Small Datasets (NeurIPS 2024 · similarity 0.86)
- Reborn Filters: Pruning Convolutional Neural Networks with Limited Data (AAAI 2020 · similarity 0.85)
- Learning Pruning-Friendly Networks via Frank-Wolfe: One-Shot, Any-Sparsity, And No Retraining (ICLR 2022 · similarity 0.85)
- Channel Gating Neural Networks (NeurIPS 2019 · similarity 0.85)
- On the Hardness of Training Deep Neural Networks Discretely (AAAI 2025 · similarity 0.83)
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.
