Research map / Architectures and efficiency / Pruning and efficient tuning
Towards Fully Sparse Training: Information Restoration with Spatial Similarity
Weixiang Xu, Xiangyu He, Ke Cheng, Peisong Wang, Jian Cheng
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 2:4 sparsity on W, Wᵀ, and Xᵀ with Information Restoration | FST with ResNet-18 · ImageNet | Top-1 accuracy: 71.2 -> 71.0 | Hurt |
| Add 2:4 sparsity on W and Wᵀ | FST with ResNet-18 · ImageNet | Top-1 accuracy: 71.2 -> 71.3 | Improved |
| Add 2:4 sparsity on W, Wᵀ, and Xᵀ with Information Restoration | FST with ResNet-50 · ImageNet | Top-1 accuracy: 77.3 -> 77.0 | Hurt |
| Add 2:4 sparsity on W, Wᵀ, and Xᵀ with Information Restoration | FST · PASCAL VOC, VGG16-SSD | mAP: 75.8 -> 76.2 | Improved |
| Add fully sparse training | R50 Faster-RCNN · PASCAL VOC | mAP: 79.5 -> 79.4 | Hurt |
| Add fully sparse training | VGG16-SSD · COCO2017 | mAP: 25.5 -> 25.5 | No clear effect |
| Add fully sparse training | R50 Faster-RCNN · COCO2017 | mAP: 37.4 -> 37.2 | Hurt |
| Remove sparsity on W (ablation) | FST with ResNet-18 · ImageNet | Top-1 accuracy: 71.2 -> 71.2 | No clear effect |
| Add sparsity on Wᵀ (ablation) | FST with ResNet-18 · ImageNet | Top-1 accuracy: 71.2 -> 71.3 | Improved |
| Add fixed-position sparsity on Xᵀ without Information Restoration (ablation) | FST with ResNet-18 · ImageNet | Top-1 accuracy: 71.3 -> 69.4 | Hurt |
| Add Information Restoration (ablation) | FST with ResNet-18 and sparse Xᵀ · ImageNet | Top-1 accuracy: 69.4 -> 71.0 | Improved |
Closest papers by meaning
- Sparser, Better, Deeper, Stronger: Improving Static Sparse Training with Exact Orthogonal Initialization (ICML 2024 · similarity 0.87)
- Navigating Extremes: Dynamic Sparsity in Large Output Spaces (NeurIPS 2024 · similarity 0.87)
- Two Sparse Matrices are Better than One: Sparsifying Neural Networks with Double Sparse Factorization (ICLR 2025 · similarity 0.87)
- Dynamic Sparse Training with Structured Sparsity (ICLR 2024 · similarity 0.86)
- S-STE: Continuous Pruning Function for Efficient 2:4 Sparse Pre-training (NeurIPS 2024 · similarity 0.86)
- Sparse Cocktail: Every Sparse Pattern Every Sparse Ratio All At Once (ICML 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.
