Research map / Architectures and efficiency / Pruning and efficient tuning
Differentiable Transportation Pruning
Yunqiang Li, Jan C. van Gemert, Torsten Hoefler, Bert Moons, Evangelos Eleftheriou, Bram-Ernst Verhoef
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 |
|---|---|---|---|
| use a single Sinkhorn iteration per step instead of many (ablation) | Differentiable Transportation Pruning · CIFAR-10 ResNet-56 | accuracy and training cost | No clear effect |
| train soft-mask learning instead of no pruning mask | ResNet-50 training on ImageNet · training overhead on A100 | training time overhead: ~0.5% overhead | No clear effect |
| learn importance scores end-to-end via optimal transport instead of L1+one-shot | ResNet-56 pruning · CIFAR-10 at 95% pruning ratio | accuracy (%): 82.79 -> 86.18 | Improved |
| use optimal-transport soft masks instead of GReg-1 | ResNet-56 pruning · CIFAR-10 at 92.5% pruning ratio | accuracy (%): 88.39 -> 88.03 | Hurt |
| use soft-mask transportation pruning instead of GReg-2 on ImageNet | ResNet-50 pruning · ImageNet at 3.06x speedup | top-1 accuracy (%): 73.90 -> 74.26 | Improved |
| use transportation pruning instead of ChipNet for global sparsity | MobileNetV2 · CIFAR-100 at 80% sparsity | accuracy (%): 52.96 -> 68.90 | Improved |
| prune ResNet-34 with the proposed method | ResNet-34 · ImageNet at 1.32x speedup | top-1 accuracy (%): 73.31 -> 74.28 | 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.
- ChipNet — ChipNet: Budget-Aware Pruning with Heaviside Continuous Approximations (ICLR 2021) · 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
- Accelerate CNNs from Three Dimensions: A Comprehensive Pruning Framework (ICML 2021 · similarity 0.87)
- How Sparse Can We Prune A Deep Network: A Fundamental Limit Perspective (NeurIPS 2024 · similarity 0.87)
- Connectivity Matters: Neural Network Pruning Through the Lens of Effective Sparsity (JMLR 2023 · similarity 0.87)
- Device-Wise Federated Network Pruning (CVPR 2024 · similarity 0.87)
- GDeR: Safeguarding Efficiency, Balancing, and Robustness via Prototypical Graph Pruning (NeurIPS 2024 · similarity 0.86)
- Sparsest Models Elude Pruning: An Exposé of Pruning’s Current Capabilities (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.
