Research map / Architectures and efficiency / Pruning and efficient tuning
Harmonious Coexistence of Structured Weight Pruning and Ternarization for Deep Neural Networks
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 |
|---|---|---|---|
| add weight penalty clipping to naive group-Lasso plus ternarization (ablation) | ternarized pruned ResNet-18 · ImageNet val | top-1 accuracy: 65.12 -> 68.01 | Helped |
| add weight penalty clipping to naive group-Lasso plus ternarization (ablation) | ternarized pruned ResNet-20 · CIFAR-10 | accuracy: 90.01 -> 90.89 | Helped |
| add weight penalty clipping to naive group-Lasso plus ternarization (ablation) | ternarized pruned ResNet-56 · CIFAR-10 | accuracy: 92.01 -> 92.85 | Helped |
| add weight penalty clipping (ablation) | ternarized pruned ResNet-18 · ImageNet val | group sparsity (%): 18 -> 25 | Helped |
| add PE-wise pruning to ternarization (ablation) | ResNet-18 ternary · ImageNet val | compression rate: ~16x -> ~21.3x | Helped |
| add pruning to ternarization (ours vs ternarization only) (ablation) | ResNet-20 ternary · CIFAR-10 | accuracy: 91.62 -> 90.89 | Hurt |
| add pruning to ternarization (ours vs ternarization only) (ablation) | ResNet-56 ternary · CIFAR-10 | accuracy: 93.1 -> 92.85 | Hurt |
| use smaller PE group size (16 to 1) (ablation) | ours · CIFAR-10 ResNet-20 | accuracy: 90.89 -> 89.66 | Hurt |
| use smaller PE group size (16 to 1) (ablation) | ours · CIFAR-10 ResNet-20 | group sparsity (%): 25 -> 63.18 | Helped |
| PE-wise structured pruning on FPGA vs non-structured ternary (ablation) | ours · FPGA ResNet-20 conv layers | speedup: 1.56x to 1.73x | Helped |
Closest papers by meaning
- CLIP-Q: Deep Network Compression Learning by In-Parallel Pruning-Quantization (CVPR 2018 · similarity 0.90)
- Automatic Joint Structured Pruning and Quantization for Efficient Neural Network Training and Compression (CVPR 2025 · similarity 0.89)
- Structured Compression by Weight Encryption for Unstructured Pruning and Quantization (CVPR 2020 · similarity 0.89)
- AutoCompress: An Automatic DNN Structured Pruning Framework for Ultra-High Compression Rates (AAAI 2020 · similarity 0.89)
- Computation and Memory-Efficient Model Compression with Gradient Reweighting (NeurIPS 2025 · similarity 0.89)
- Towards Higher Ranks via Adversarial Weight Pruning (NeurIPS 2023 · similarity 0.89)
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.