Research map / Architectures and efficiency / Pruning and efficient tuning
Layer Freezing & Data Sieving: Missing Pieces of a Generic Framework for Sparse Training
Geng Yuan, Yanyu Li, Sheng Li, Zhenglun Kong, Sergey Tulyakov, Xulong Tang, Yanzhi Wang, Jian Ren
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 layer freezing with 20% FLOPs reduction (ablation) | MEST sparse training · CIFAR-100, ResNet-32, 90% sparsity | accuracy: 71.30 -> 71.29 | No clear effect |
| Add layer freezing with 20% FLOPs reduction (ablation) | MEST sparse training · CIFAR-100, ResNet-32, 60% sparsity | accuracy: 73.97 -> 73.76 | Hurt |
| Add layer freezing with 35% FLOPs reduction (ablation) | MEST sparse training · CIFAR-100, ResNet-32, 90% sparsity | accuracy: 71.30 -> 70.26 | Hurt |
| Add data sieving with 20% FLOPs reduction (ablation) | MEST sparse training · CIFAR-100, ResNet-32, 90% sparsity | accuracy: 71.3 -> 71.33 | No clear effect |
| Add data sieving with 35% FLOPs reduction (ablation) | MEST sparse training · CIFAR-100, ResNet-32, 90% sparsity | accuracy: 71.3 -> 70.59 | Hurt |
| Combine freezing and sieving instead of freezing only at 35% FLOPs reduction (ablation) | SpFDE · CIFAR-100, ResNet-32, 90% sparsity | accuracy: 70.26 -> 71.25 | Improved |
| Combine freezing and sieving instead of data sieving only at 35% FLOPs reduction (ablation) | SpFDE · CIFAR-100, ResNet-32, 90% sparsity | accuracy: 70.59 -> 71.25 | Improved |
| Use single-shot freezing instead of periodic freezing (ablation) | SpFDE · CIFAR-100, ResNet-32, 90% sparsity | accuracy: 70.86 -> 71.30 | Improved |
| Use single-shot freezing instead of single-shot freezing & resume (ablation) | SpFDE · CIFAR-100, ResNet-32, 60% sparsity | accuracy: 73.49 -> 73.61 | Improved |
| Raise data sieving update ratio from 10% to 30% (remove 15%) (ablation) | SpFDE · CIFAR-100, ResNet-32, 90% sparsity | accuracy: 70.68 -> 71.35 | Improved |
| Raise data sieving update ratio from 30% to 50% (remove 15%) (ablation) | SpFDE · CIFAR-100, ResNet-32, 90% sparsity | accuracy: 71.35 -> 70.92 | Hurt |
| Replace MEST with SpFDE 15%+15% at 80% sparsity | SpFDE · ImageNet, ResNet-50 | top-1 accuracy: 75.70 -> 75.75 | Improved |
Closest papers by meaning
- Sparsity in Deep Learning: Pruning and growth for efficient inference and training in neural networks (JMLR 2021 · similarity 0.86)
- MEST: Accurate and Fast Memory-Economic Sparse Training Framework on the Edge (NeurIPS 2021 · similarity 0.86)
- Sparse Cocktail: Every Sparse Pattern Every Sparse Ratio All At Once (ICML 2024 · similarity 0.86)
- On The Concurrence of Layer-wise Preconditioning Methods and Provable Feature Learning (ICML 2025 · similarity 0.85)
- A Single-Step, Sharpness-Aware Minimization is All You Need to Achieve Efficient and Accurate Sparse Training (NeurIPS 2024 · similarity 0.85)
- Random Sparse Lifts: Construction, Analysis and Convergence of finite sparse networks (ICLR 2024 · similarity 0.85)
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.
