Research map / Architectures and efficiency / Pruning and efficient tuning
Operation-Aware Soft Channel Pruning using Differentiable Masks
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 |
|---|---|---|---|
| jointly learn differentiable channel masks and network parameters using BN and ReLU activation statistics | channel pruning · CIFAR-10 | accuracy: 92.59 -> 93.77 | Helped |
| jointly learn differentiable channel masks and network parameters using BN and ReLU activation statistics | channel pruning · CIFAR-100 | accuracy drop: 2.72 -> 0.62 | Helped |
| add ReLU-aware pruning to BN-based channel pruning (ablation) | SCP · CIFAR-100, ResNet-164 | accuracy drop: 1.33 -> 0.62 | Helped |
| add ReLU-aware pruning to BN-based channel pruning (ablation) | SCP · CIFAR-100, DenseNet-40 | accuracy drop: 1.79 -> 0.77 | Helped |
| add ReLU-aware pruning to BN-based channel pruning (ablation) | SCP · CIFAR-100, VGGNet-19 | accuracy drop: 1.33 -> 0.41 | Helped |
| add ReLU-aware pruning to BN-based channel pruning (ablation) | SCP · CIFAR-100, VGGNet-16 | accuracy drop: 5.21 -> 3.55 | Helped |
| increase sparsity-loss weight lambda (ablation) | SCP on DenseNet-40 · CIFAR-100 | FLOPs reduction: 28.99 -> 67.82 | Helped |
| increase sparsity-loss weight lambda (ablation) | SCP on DenseNet-40 · CIFAR-100 | accuracy: 74.91 -> 73.17 | Hurt |
| increase confidence-interval parameter s (ablation) | SCP on DenseNet-40 · CIFAR-100 | FLOPs reduction: 45.73 -> 60.81 | Helped |
| increase confidence-interval parameter s (ablation) | SCP on DenseNet-40 · CIFAR-100 | accuracy: 74.59 -> 73.38 | Hurt |
| use target-ratio training instead of exhaustive lambda search (ablation) | SCP on ResNet-50 · ILSVRC-12 validation | top-1 accuracy drop: 1.33 -> 1.43 | No clear effect |
Closest papers by meaning
- Discrimination-aware Channel Pruning for Deep Neural Networks (NeurIPS 2018 · similarity 0.91)
- Rethinking the Smaller-Norm-Less-Informative Assumption in Channel Pruning of Convolution Layers (ICLR 2018 · similarity 0.90)
- Revisiting Random Channel Pruning for Neural Network Compression (CVPR 2022 · similarity 0.89)
- DMCP: Differentiable Markov Channel Pruning for Neural Networks (CVPR 2020 · similarity 0.89)
- CHIP: CHannel Independence-based Pruning for Compact Neural Networks (NeurIPS 2021 · similarity 0.88)
- REPrune: Channel Pruning via Kernel Representative Selection (AAAI 2024 · similarity 0.86)
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.