Research map / Architectures and efficiency / Pruning and efficient tuning
DSA: More Efficient Budgeted Pruning via Differentiable Sparsity Allocation
Xuefei Ning, Tianchen Zhao, Wenshuo Li, Peng Lei, Yu Wang, Huazhong Yang
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 |
|---|---|---|---|
| gradient-based DSA instead of regularization (SSL) | ResNet-20 · CIFAR-10 75% FLOPs | accuracy: 91.08 -> 92.10 | Improved |
| gradient-based DSA instead of regularization (MorphNet) | ResNet-20 · CIFAR-10 33.3% FLOPs | accuracy: 88.72 -> 90.24 | Improved |
| DSA instead of SSL | ResNet-56 · CIFAR-10 33.3% FLOPs | accuracy: 91.32 -> 92.20 | Improved |
| DSA instead of LCCL (50% budget) | ResNet-20 · CIFAR-10 50% FLOPs | accuracy: 91.68 -> 91.38 | Hurt |
| DSA instead of FPGM (50% budget) | ResNet-56 · CIFAR-10 50% FLOPs | accuracy: 92.93 -> 92.91 | Hurt |
| DSA instead of FPGM | ResNet18 · ImageNet 60% FLOPs | top-1 accuracy drop: -2.47 -> -1.11 | Improved |
| DSA instead of FPGM | ResNet50 · ImageNet 60% FLOPs | top-1 accuracy drop: -1.12 -> -0.92 | Improved |
| DSA instead of AutoCompress | VGG-16 · CIFAR-10 | accuracy: 88.78 -> 90.16 | Improved |
| DSA instead of AutoCompress | ResNet-18 · CIFAR-10 | accuracy: 93.91 -> 94.19 | Improved |
| run DSA from scratch in one flow instead of pretrain/search/finetune | pruning pipeline · CIFAR-10 and ImageNet | total pruning cost: about 1.5x speed up | Improved |
| use task-loss gradients to allocate sparsity | DSA · ResNet-18 CIFAR-10 | sensitivity alignment | 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.
- MorphNet — MorphNet: Fast & Simple Resource-Constrained Structure Learning of Deep Networks (CVPR 2018) · better on 6 of 6 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
- SequentialAttention++ for Block Sparsification: Differentiable Pruning Meets Combinatorial Optimization (NeurIPS 2024 · similarity 0.87)
- A Unified Framework for Soft Threshold Pruning (ICLR 2023 · similarity 0.86)
- Exploring Learning Complexity for Efficient Downstream Dataset Pruning (ICLR 2025 · similarity 0.85)
- Sparsest Models Elude Pruning: An Exposé of Pruning’s Current Capabilities (ICML 2024 · similarity 0.85)
- Lightweight Dataset Pruning without Full Training via Example Difficulty and Prediction Uncertainty (ICML 2025 · similarity 0.85)
- MDP: Multidimensional Vision Model Pruning with Latency Constraint (CVPR 2025 · 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.
