Research map / Architectures and efficiency / Pruning and efficient tuning
Exploring Structural Sparsity of Deep Networks Via Inverse Scale Spaces
Yanwei Fu, Chen Liu, Donghao Li, Zuyuan Zhong, Xinwei Sun, Jinshan Zeng, Yuan Yao
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 |
|---|---|---|---|
| add momentum and weight decay to DessiLBI (ablation) | DessiLBI · ImageNet-2012 ResNet-18 | top-1 accuracy (%): 65.26 -> 70.55 | Improved |
| add momentum to DessiLBI (ablation) | DessiLBI · ImageNet-2012 ResNet-18 | top-1 accuracy (%): 65.26 -> 68.55 | Improved |
| DessiLBI Mom-Wd instead of SGD Mom-Wd | DessiLBI · ImageNet-2012 ResNet-18 | top-1 accuracy (%): 69.76 -> 70.55 | Improved |
| DessiLBI Mom-Wd instead of SGD Mom-Wd | DessiLBI · ImageNet-2012 AlexNet | top-1 accuracy (%): 56.55 -> 57.09 | Improved |
| DessiLBI Mom-Wd instead of SGD Nesterov | DessiLBI · ImageNet-2012 ResNet-18 | top-1 accuracy (%): 70.19 -> 70.55 | Improved |
| fine-tuning instead of retraining pruned structure (ablation) | DessiLBI · CIFAR10 weight pruning VGG16 | sparse accuracy (%): 93.60 -> 94.18 | Improved |
| fine-tuning instead of retraining pruned structure (ablation) | DessiLBI · CIFAR10 filter pruning ResNet56 | sparse accuracy (%): 93.15 -> 93.94 | Improved |
| fine-tune only 10 epochs instead of full fine-tuning (ablation) | DessiLBI · CIFAR10 filter pruning ResNet56 | sparse accuracy (%): 93.94 -> 89.97 | Hurt |
| pruning layers (layer selection) instead of filter pruning (ablation) | DessiLBI · CIFAR10 ResNet56 | sparse accuracy (%): MFLOP 78 (62%) -> 56 (45%); accuracy 93.94 -> 93.47 | Mixed |
| DessiLBI filter pruning vs HRank | DessiLBI · CIFAR10 VGG16 | sparse accuracy (%): 93.43 -> 93.59 (FT 200) | Mixed |
| DessiLBI weight pruning vs Iterative-Pruning-B at sparsity 0.10 | DessiLBI · CIFAR10 ResNet50 | sparse accuracy (%): 73.75 -> 92.64 | Improved |
| early stopped DessiLBI for winning tickets vs one-shot and iterative pruning | DessiLBI · CIFAR10 winning tickets | accuracy | 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.
- HRank — HRank: Filter Pruning Using High-Rank Feature Map (CVPR 2020) · better on 2 of 2 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
- DessiLBI: Exploring Structural Sparsity of Deep Networks via Differential Inclusion Paths (ICML 2020 · similarity 0.95)
- Random Sparse Lifts: Construction, Analysis and Convergence of finite sparse networks (ICLR 2024 · similarity 0.88)
- Sparse maximal update parameterization: A holistic approach to sparse training dynamics (NeurIPS 2024 · similarity 0.88)
- Position: A Theory of Deep Learning Must Include Compositional Sparsity (ICML 2025 · similarity 0.87)
- Scaling Laws for Sparsely-Connected Foundation Models (ICLR 2024 · similarity 0.87)
- Investigating the Overlooked Hessian Structure: From CNNs to LLMs (ICML 2025 · 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.
