Research map / Architectures and efficiency / Quantization
Zero-Shot Adversarial Quantization
Check 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 |
|---|---|---|---|
| Use ZAQ instead of the strongest reported data-free baseline (GDFQ) | CIFAR10 MobileNetV2 W6A6 · CIFAR10 | top-1 accuracy: 91.27 -> 92.15 | Helped |
| Use ZAQ instead of GDFQ | CIFAR100 ResNet18 W4A4 · CIFAR100 | top-1 accuracy: 71.53 -> 72.67 | Helped |
| Use ZAQ instead of ZeroQ | ImageNet ResNet50 W2A2 · ImageNet | top-1 accuracy: 63.12 -> 65.52 | Helped |
| Use ZAQ instead of ZeroQ | Cityscapes DeepLabV3 (MobileNetV2) W4A4 · Cityscapes | mIoU: 52.73 -> 55.12 | Helped |
| Use ZAQ instead of ZeroQ | CamVid DeepLabV3 (MobileNetV2) W4A4 · CamVid | mIoU: 43.83 -> 47.62 | Helped |
| Use ZAQ instead of FT using original training data | VOC2012 SSD (MobileNetV2) W4A4 · VOC2012 | mAP: 64.28 -> 64.44 | Helped |
| Add intermediate inter-channel discrepancy modeling (ablation) | ZAQ · ImageNet and Cityscapes | performance: 1\sim2\% | Helped |
| Add activation regularization (ablation) | ZAQ · ImageNet and Cityscapes | performance: about 0.5\% | Helped |
| Replace CRM with Gram matrix (ablation) | ZAQ on CIFAR100 ResNet18 W4A4 · CIFAR100 | top-1 accuracy: 72.67 -> 45.32 | Hurt |
| Replace CRM with AT feature alignment (ablation) | ZAQ on CIFAR100 ResNet18 W4A4 · CIFAR100 | top-1 accuracy: 72.67 -> 61.80 | Hurt |
| Use ZAQ instead of GDFQ for data-free fine-tuning | CIFAR100 · GPU time | GPU time: 7.6 h -> 3.2 h | Helped |
| Use ZAQ instead of ZeroQ for data-free fine-tuning | Cityscapes · GPU time | GPU time: 12.7 h -> 5.4 h | Helped |
Closest papers by meaning
- SynQ: Accurate Zero-shot Quantization by Synthesis-aware Fine-tuning (ICLR 2025 · similarity 0.91)
- Sharpness-Aware Data Generation for Zero-shot Quantization (ICML 2024 · similarity 0.90)
- On the Zero-shot Adversarial Robustness of Vision-Language Models: A Truly Zero-shot and Training-free Approach (CVPR 2025 · similarity 0.88)
- Zero-Shot Quantization via Weight-Space Arithmetic (NeurIPS 2026 · similarity 0.87)
- Enhancing Diversity for Data-free Quantization (CVPR 2025 · similarity 0.87)
- One-Step Forward and Backtrack: Overcoming Zig-Zagging in Loss-Aware Quantization Training (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.