Research map / Architectures and efficiency / Pruning and efficient tuning
Balanced Sparsity for Efficient DNN Inference on GPU
Zhuliang Yao, Shijie Cao, Wencong Xiao, Chen Zhang, Lanshun Nie
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 |
|---|---|---|---|
| replace random sparsity with Balanced Sparsity | PTB LSTM inference · PTB | inference time (us): 370.9 -> 120.2 | Improved |
| replace dense computation with Balanced Sparsity | PTB LSTM inference · PTB | inference time (us): 294.1 -> 120.2 | Improved |
| replace block sparsity with Balanced Sparsity | PTB LSTM inference · PTB | inference time (us): 326.3 -> 120.2 | Improved |
| replace random sparsity with Balanced Sparsity | TIMIT CTC Bi-LSTM inference · TIMIT | inference time (us): 190.5 -> 83.9 | Improved |
| replace dense computation with Balanced Sparsity | TIMIT CTC Bi-LSTM inference · TIMIT | inference time (us): 117.9 -> 83.9 | Improved |
| replace block sparsity with Balanced Sparsity | TIMIT CTC Bi-LSTM inference · TIMIT | inference time (us): 212.8 -> 83.9 | Improved |
| replace random sparsity with Balanced Sparsity | VGG-16 inference · ImageNet ILSVRC-2012 | inference time (us): 33407.4 -> 5213.0 | Improved |
| use Balanced Sparsity pruning | VGG-16 · ImageNet ILSVRC-2012 | top-5 accuracy: 90.3% | No clear effect |
| use balance range 25 instead of other tested ranges (ablation) | Balanced Sparsity on PTB LSTM · PTB | perplexity: 78.3; 78.4; 78.4 | No clear effect |
| use block size 16*16 rather than 4*4 (ablation) | Block Sparsity on PTB LSTM · PTB, 80% sparsity | perplexity: 88.1 -> 99.5 | Hurt |
Closest papers by meaning
- Effective Interplay between Sparsity and Quantization: From Theory to Practice (ICLR 2025 · similarity 0.87)
- ICP: Immediate Compensation Pruning for Mid-to-high Sparsity (CVPR 2025 · similarity 0.86)
- Harnessing Manycore Processors with Distributed Memory for Accelerated Training of Sparse and Recurrent Models (AAAI 2024 · similarity 0.86)
- DEPrune: Depth-wise Separable Convolution Pruning for Maximizing GPU Parallelism (NeurIPS 2024 · similarity 0.86)
- DuoGPT: Training-free Dual Sparsity through Activation-aware Pruning in LLMs (NeurIPS 2025 · similarity 0.86)
- Polar Sparsity: High Throughput Batched LLM Inferencing with Scalable Contextual Sparsity (NeurIPS 2025 · 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.
