Research map / Architectures and efficiency / Pruning and efficient tuning
Data-Driven Sparse Structure Selection for Deep Neural Networks
Zehao Huang, Naiyan Wang
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 |
|---|---|---|---|
| optimize scaling factors with APG instead of SGD with hard threshold (ablation) | SSS ResNet-32 · ImageNet val | top-1 error (%): 26.46 -> 25.82 | Improved |
| use APG instead of ISTA (ablation) | SSS block pruning · ImageNet val | convergence | Improved |
| prune ResNet-50 blocks to ResNet-41 | ResNet-50 · ImageNet val | top-1 error (%): 23.88 -> 24.56 | Hurt |
| prune ResNet-50 blocks to ResNet-32 | ResNet-50 · ImageNet val | top-1 error (%): 23.88 -> 25.82 | Hurt |
| prune ResNeXt-50 blocks and groups to ResNeXt-41 | ResNeXt-50 · ImageNet val | top-1 error (%): 22.43 -> 24.07 | Hurt |
| prune PeleeNet neurons to PeleeNet-A | PeleeNet · ImageNet val | top-1 error (%): 27.47 -> 27.85 | Hurt |
| prune conv1 to conv4 of VGG16 with conv5 penalty set to 0 | VGG16 · ImageNet val | top-1 error (%): 27.54 -> 31.47 | Hurt |
| use SSS ResNeXt-38 instead of DenseNet-121 | ResNeXt-38 · ImageNet val | top-5 error (%): 7.71 -> 7.50 | Improved |
| use SSS ResNet-32 instead of pruned ResNet-50 from channel pruning | ResNet-32 · ImageNet val | top-1 error (%): 27.70 -> 25.82 | Improved |
| use SSS ResNet-41 instead of ResNet-101-pruned | ResNet-41 · ImageNet val | top-1 error (%): 25.44 -> 24.56 | Improved |
| apply SSS block and group pruning to ResNeXt-164 | ResNeXt-164 · CIFAR-10 | FLOPs and accuracy | Improved |
| apply SSS neuron pruning to VGG | VGG on CIFAR · CIFAR-10 and CIFAR-100 | parameters and FLOPs | Mixed |
Closest papers by meaning
- Exploring Structural Sparsity of Deep Networks Via Inverse Scale Spaces (TPAMI 2022 · similarity 0.92)
- Random Sparse Lifts: Construction, Analysis and Convergence of finite sparse networks (ICLR 2024 · similarity 0.88)
- Sparsity in Deep Learning: Pruning and growth for efficient inference and training in neural networks (JMLR 2021 · similarity 0.88)
- DessiLBI: Exploring Structural Sparsity of Deep Networks via Differential Inclusion Paths (ICML 2020 · similarity 0.88)
- Growing Efficient Deep Networks by Structured Continuous Sparsification (ICLR 2021 · similarity 0.88)
- Parameter efficient training of deep convolutional neural networks by dynamic sparse reparameterization (ICML 2019 · 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.
