Research map / Architectures and efficiency / Quantization
Gradient $\ell_1$ Regularization for Quantization Robustness
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 gradient ℓ1 regularization | VGG-like · CIFAR-10 (8,4) | test accuracy (%): 79.10 -> 89.74 | Helped |
| add gradient ℓ1 regularization | VGG-like · CIFAR-10 (6,4) | test accuracy (%): 78.84 -> 89.78 | Helped |
| add gradient ℓ1 regularization | VGG-like · CIFAR-10 (4,4) | test accuracy (%): 11.47 -> 85.99 | Helped |
| add gradient ℓ1 regularization | ResNet-18 · CIFAR-10 (8,4) | test accuracy (%): 85.51 -> 88.70 | Helped |
| add gradient ℓ1 regularization | ResNet-18 · CIFAR-10 (6,4) | test accuracy (%): 85.35 -> 88.45 | Helped |
| add gradient ℓ1 regularization | ResNet-18 · CIFAR-10 (4,4) | test accuracy (%): 83.98 -> 87.62 | Helped |
| add gradient ℓ1 regularization | ResNet-18 · ImageNet (8,8) | test accuracy: 69.20 -> 69.92 | Helped |
| add gradient ℓ1 regularization | ResNet-18 · ImageNet (6,6) | test accuracy: 63.80 -> 66.39 | Helped |
| add gradient ℓ1 regularization | ResNet-18 · ImageNet (4,4) | test accuracy: 0.30 -> 0.22 | Hurt |
| increase regularization strength to λ=0.05 (ablation) | L1-regularized ResNet-18 · ImageNet (4,4) | test accuracy: 0.22 -> 55.32 | Helped |
| replace gradient ℓ1 regularization with gradient ℓ2 regularization | quantization-ready models · CIFAR-10 and ImageNet post-training quantization | test accuracy | Hurt |
| use Defensive Quantization regularization instead of gradient ℓ1 regularization | quantization-ready models · CIFAR-10 ResNet-18 and ImageNet ResNet-18 | test accuracy | Hurt |
Closest papers by meaning
- Robust Quantization: One Model to Rule Them All (NeurIPS 2020 · similarity 0.91)
- PARQ: Piecewise-Affine Regularized Quantization (ICML 2025 · similarity 0.89)
- Zero-Shot Quantization via Weight-Space Arithmetic (NeurIPS 2026 · similarity 0.89)
- One-Step Forward and Backtrack: Overcoming Zig-Zagging in Loss-Aware Quantization Training (AAAI 2024 · similarity 0.89)
- MagR: Weight Magnitude Reduction for Enhancing Post-Training Quantization (NeurIPS 2024 · similarity 0.88)
- Quantization without Tears (CVPR 2025 · similarity 0.88)
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.