Research map / Architectures and efficiency / Quantization
Symmetry Regularization and Saturating Nonlinearity for Robust Quantization
Sein Park, Yeongsang Jang, Eunhyeok Park
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 |
|---|---|---|---|
| add SymReg to the baseline (ablation) | MobileNet-V2 PTQ ACIQ · CIFAR-100 2-bit weight | top-1 accuracy (%): 6.92 -> 25.01 | Improved |
| add SatNL to the baseline (ablation) | MobileNet-V2 PTQ ACIQ · CIFAR-100 2-bit weight | top-1 accuracy (%): 6.92 -> 14.45 | Improved |
| add SatNL to SymReg (ablation) | MobileNet-V2 PTQ ACIQ · CIFAR-100 2-bit weight | top-1 accuracy (%): 25.01 -> 33.68 | Improved |
| add SatNL to ASAM (ablation) | MobileNet-V2 PTQ ACIQ · CIFAR-100 2-bit weight | top-1 accuracy (%): 11.80 -> 25.77 | Improved |
| add ASAM to SymReg+SatNL (All) (ablation) | MobileNet-V2 PTQ ACIQ · CIFAR-100 3-bit weight | top-1 accuracy (%): 69.66 -> 72.17 | Improved |
| add ASAM to SymReg+SatNL (All) (ablation) | MobileNet-V2 PTQ ACIQ · CIFAR-100 2-bit weight | top-1 accuracy (%): 33.68 -> 39.27 | Improved |
| add SymReg and SatNL to full-precision training (ablation) | MobileNet-V2 · CIFAR-100 FP | top-1 accuracy (%): 74.70 -> 74.41 | Hurt |
| add ASAM to SymReg+SatNL (All) (ablation) | MobileNet-V2 · CIFAR-100 FP | top-1 accuracy (%): 74.41 -> 75.33 | Improved |
| add ASAM to KURE (ablation) | MobileNet-V2 PTQ ACIQ · CIFAR-100 3-bit weight | top-1 accuracy (%): 71.26 -> 72.48 | Improved |
| add All to KURE (ablation) | MobileNet-V2 PTQ ACIQ · CIFAR-100 3-bit weight | top-1 accuracy (%): 72.48 -> 73.34 | Improved |
| train with SymReg+SatNL+ASAM (ablation) | ResNet-18 PTQ QDrop · ImageNet W4/A4 | top-1 accuracy (%): 68.91 -> 69.93 | Improved |
| train with SymReg+SatNL+ASAM | MobileNet-V2 PTQ ACIQ · ImageNet W4/A4 | top-1 accuracy (%): 18.15 -> 58.30 | Improved |
Closest papers by meaning
- Robust Quantization: One Model to Rule Them All (NeurIPS 2020 · similarity 0.89)
- Gradient $\ell_1$ Regularization for Quantization Robustness (ICLR 2020 · similarity 0.88)
- Robust Training of Neural Networks at Arbitrary Precision and Sparsity (ICLR 2026 · similarity 0.88)
- SYQ: Learning Symmetric Quantization for Efficient Deep Neural Networks (CVPR 2018 · similarity 0.88)
- QuantDemoire: Quantization with Outlier Aware for Image Demoiréing (NeurIPS 2026 · similarity 0.84)
- Sharpness-Aware Data Generation for Zero-shot Quantization (ICML 2024 · 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.
