Research map / Architectures and efficiency / Pruning and efficient tuning
MEST: Accurate and Fast Memory-Economic Sparse Training Framework on the Edge
Geng Yuan, Xiaolong Ma, Wei Niu, Zhengang Li, Zhenglun Kong, Ning Liu, Yifan Gong, Zheng Zhan, Chaoyang He, Qing Jin, Siyue Wang, Minghai Qin and 4 more
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 elastic mutation (decaying mutation ratio) (ablation) | MEST · CIFAR-100 98% sparsity | accuracy (%): 62.58 -> 64.59 | Improved |
| add elastic mutation (ablation) | MEST · CIFAR-10 90% sparsity | accuracy (%): 92.12 -> 92.56 | Improved |
| add soft memory bound (EM&S) on top of EM (ablation) | MEST+EM · CIFAR-100 98% sparsity | accuracy (%): 64.59 -> 67.16 | Improved |
| add soft memory bound (EM&S) on top of EM (ablation) | MEST+EM · CIFAR-10 98% sparsity | accuracy (%): 89.22 -> 90.51 | Improved |
| add soft memory bound (ablation) | MEST+EM · ImageNet ResNet-50 90% sparsity | Top-1 accuracy (%): 73.63 -> 75.00 | Improved |
| add soft memory bound (ablation) | MEST+EM · ImageNet ResNet-50 80% sparsity | Top-1 accuracy (%): 75.75 -> 75.73 | No clear effect |
| add data-efficient two-phase training (DE) (ablation) | MEST+EM&S · ImageNet ResNet-50 80% sparsity | Top-1 accuracy (%): 75.73 -> 75.70 | No clear effect |
| add data-efficient training (DE) (ablation) | MEST 1.7x+EM&S · ImageNet ResNet-50 90% sparsity | Top-1 accuracy (%): 76.13 -> 76.08 | No clear effect |
| add DE for extra speedup (ablation) | MEST · CIFAR-100 ResNet-32 90% sparsity | training speedup: 10% to 15% | Improved |
| use MEST+EM&S instead of RigL at similar FLOPs | RigL · ImageNet 90% sparsity ResNet-50 | Top-1 accuracy (%): 72.0 -> 72.37 | Improved |
| use channel (structured) sparsity instead of unstructured/fine-grained (ablation) | MEST · CIFAR-100 | accuracy | Hurt |
| use block or pattern sparsity instead of unstructured (ablation) | MEST · CIFAR-100 ResNet-32 | training acceleration: up to 2.3x | Improved |
Closest papers by meaning
- Sparsity in Deep Learning: Pruning and growth for efficient inference and training in neural networks (JMLR 2021 · similarity 0.87)
- SparseProp: Efficient Sparse Backpropagation for Faster Training of Neural Networks at the Edge (ICML 2023 · similarity 0.87)
- Fast and Controllable Post-training Sparsity: Learning Optimal Sparsity Allocation with Global Constraint in Minutes (AAAI 2024 · similarity 0.86)
- Navigating Extremes: Dynamic Sparsity in Large Output Spaces (NeurIPS 2024 · similarity 0.86)
- Sparse maximal update parameterization: A holistic approach to sparse training dynamics (NeurIPS 2024 · similarity 0.86)
- Sparse Cocktail: Every Sparse Pattern Every Sparse Ratio All At Once (ICML 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.
