Research map / Architectures and efficiency / Pruning and efficient tuning
Balanced Column-Wise Block Pruning for Maximizing GPU Parallelism
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 column-wise block pruning with balanced pruning ratios across tiles instead of cross-wise pruning | VGG-16 · ImageNet | inference time (ms): 3.0 -> 2.5 | Helped |
| Use column-wise block pruning with balanced pruning ratios across tiles instead of filter pruning | ResNet-50 · ImageNet | inference time (ms): 7.8 -> 7.5 | Helped |
| Increase BCBP pruning ratio from 41.3% to 84.7% | ResNet-18 · ImageNet | inference time (ms): 4.6 -> 1.5 | Helped |
| Use tile width 16 instead of 32 | VBP · ImageNet, VGG-16 | WIT: 10.4 -> 13.9 | Hurt |
| Increase pruning ratio from 60% to 80% | VBP · ImageNet, VGG-16 | WIT: 10.4 -> 6.8 | Helped |
| Add pruning-ratio recalibration across workload-balanced blocks (ablation) | BCBP · ImageNet, VGG-16 | information loss | Helped |
| Reduce WBB width from 18 to 9 (ablation) | BCBP · ImageNet, VGG-16 | elapsed time: 3.79%, 5.29%, and 4.07% | Helped |
| Use BCBP rather than filter pruning at similar pruning ratios | ResNet-18 · ImageNet | inference time (ms): 4.3 -> 4.6 | Hurt |
| Use BCBP rather than filter pruning at similar pruning ratios | ResNet-50 · ImageNet | inference time (ms): 7.2 -> 7.5 | Hurt |
Closest papers by meaning
- DEPrune: Depth-wise Separable Convolution Pruning for Maximizing GPU Parallelism (NeurIPS 2024 · similarity 0.89)
- Pruning Filters for Efficient ConvNets (ICLR 2017 · similarity 0.86)
- DARB: A Density-Adaptive Regular-Block Pruning for Deep Neural Networks (AAAI 2020 · similarity 0.86)
- GDeR: Safeguarding Efficiency, Balancing, and Robustness via Prototypical Graph Pruning (NeurIPS 2024 · similarity 0.85)
- How Sparse Can We Prune A Deep Network: A Fundamental Limit Perspective (NeurIPS 2024 · similarity 0.85)
- Numerical Pruning for Efficient Autoregressive Models (AAAI 2025 · similarity 0.84)
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.