Research map / Optimization / Nonconvex and smooth optimization
ProxQuant: Quantized Neural Networks via Proximal Operators
Yu Bai, Yu-Xiang Wang, Edo Liberty
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 |
|---|---|---|---|
| Use ProxQuant-Binary instead of BinaryConnect | ResNet binary quantization · CIFAR-10 | top-1 classification error (%): 9.54 (0.03) -> 9.35 (0.13); 8.61 (0.27) -> 8.53 (0.15); 8.23 (0.23) -> 7.95 (0.05); 7.97 (0.22) -> 7.70 (0.06) | Improved |
| Use ProxQuant-Ternary instead of TTQ | ResNet ternary quantization · CIFAR-10 | top-1 classification error (%): 8.87 -> 8.40 (0.13); 7.63 -> 7.65 (0.15); 7.02 -> 7.05 (0.08) | Mixed |
| Report the best of 4 ProxQuant-Ternary runs rather than the mean | ProxQuant-Ternary · CIFAR-10, ResNet ternary | top-1 classification error (%): 8.40 (0.13) -> 8.22; 7.65 (0.15) -> 7.53; 7.05 (0.08) -> 6.98 | Improved |
| Use ProxQuant-Binary instead of BinaryConnect | Binary LSTM · Penn Treebank | perplexity per word (PPW): 372.2 -> 288.5 | Improved |
| Use ALT-ProxQuant instead of ALT Straight-through | Multi-bit LSTM · Penn Treebank, 1-bit | perplexity per word (PPW): 104.7 -> 106.2 | Hurt |
| Use ALT-ProxQuant instead of ALT Straight-through | Multi-bit LSTM · Penn Treebank, 2-bit | perplexity per word (PPW): 90.2 -> 90.0 | Improved |
| Use ALT-ProxQuant instead of ALT Straight-through | Multi-bit LSTM · Penn Treebank, 3-bit | perplexity per word (PPW): 86.1 -> 87.2 | Hurt |
| Scale the alternating quantizer output by 0.3 (ablation) | ALT Straight-through LSTM · Penn Treebank, 1-, 2-, and 3-bit | perplexity per word (PPW): 116.7, 94.3, 87.3 -> 104.7, 90.2, 86.1 | Improved |
Closest papers by meaning
- Network Quantization With Element-Wise Gradient Scaling (CVPR 2021 · similarity 0.88)
- A Statistical Framework for Low-bitwidth Training of Deep Neural Networks (NeurIPS 2020 · similarity 0.88)
- Proximal Mean-Field for Neural Network Quantization (ICCV 2019 · similarity 0.87)
- ProQuant: Progressive Quantization-aware Training for Edge MLLMs (NeurIPS 2026 · similarity 0.87)
- Demystifying and Generalizing BinaryConnect (NeurIPS 2021 · similarity 0.87)
- Robust Training of Neural Networks at Arbitrary Precision and Sparsity (ICLR 2026 · similarity 0.87)
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.
