Research map / Architectures and efficiency / Quantization
Robust Quantization: One Model to Rule Them All
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 KURE regularization | ResNet-50 · ImageNet, quantized weight/activation configurations in Table 1 | accuracy: 71.8 -> 75.6 (4 / FP); 62.9 -> 73.6 (3 / FP); 10.3 -> 64.2 (2 / FP); 74.8 -> 76.2 (6 / 6); 72.9 -> 75.8 (5 / 5); 70 -> 7 | Helped |
| Add KURE regularization | ResNet-18 · ImageNet, quantized weight/activation configurations in Table 1 | accuracy: 62.6 -> 68.3 (4 / FP); 52.4 -> 62.6 (3 / FP); 0.5 -> 40.2 (2 / FP); 68.6 -> 70 (6 / 6); 65.4 -> 69.7 (5 / 5); 59.8 -> 66 | Helped |
| Add KURE regularization | MobileNet-V2 · ImageNet, quantized weight/activation configurations in Table 1 | accuracy: 60.4 -> 67.6 (4 / FP); 31.8 -> 56.6 (3 / FP); 69.7 -> 70 (6 / 6); 64.6 -> 66.9 (5 / 5); 48.1 -> 59 (4 / 4); 3.7 -> 24.4 | Helped |
| Add KURE regularization | MobileNet-V2 · ImageNet, full precision | accuracy: 71.8 -> 71.3 | Hurt |
| Use KURE regularization instead of L1 regularization | ResNet-18 · ImageNet, 4 / 4 | accuracy: 55.32 -> 66.9 | Helped |
| Use KURE regularization instead of L1 regularization | ResNet-18 · ImageNet, 8 / 8 | accuracy: 69.92 -> 70.2 | Helped |
| Use KURE regularization instead of L1 regularization | ResNet-18 · ImageNet, 6 / 6 | accuracy: 66.39 -> 70 | Helped |
| Add KURE regularization | ResNet-50 · ImageNet, weights quantized at nearest power-of-2 step size | accuracy: 63.6 -> 74.2 (4 / FP); 53.2 -> 71.6 (3 / FP) | Helped |
| Add KURE regularization | ResNet-18 · ImageNet, weights quantized at nearest power-of-2 step size | accuracy: 61.4 -> 66.2 (4 / FP); 37.5 -> 55.8 (3 / FP) | Helped |
Closest papers by meaning
- Gradient $\ell_1$ Regularization for Quantization Robustness (ICLR 2020 · similarity 0.91)
- Robust Training of Neural Networks at Arbitrary Precision and Sparsity (ICLR 2026 · similarity 0.89)
- Adaptive Quantization for Deep Neural Network (AAAI 2018 · similarity 0.88)
- One-Shot Model for Mixed-Precision Quantization (CVPR 2023 · similarity 0.88)
- Zero-Shot Quantization via Weight-Space Arithmetic (NeurIPS 2026 · similarity 0.87)
- Robustness-Guided Image Synthesis for Data-Free Quantization (AAAI 2024 · 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.