Research map / Architectures and efficiency / Quantization
ZeroQ: A Novel Zero Shot Quantization Framework
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 mixed-precision weights with 6-bit activations instead of OCS W6A6 | ResNet50 · ImageNet | Top-1 accuracy: 74.80 -> 77.43 | Helped |
| Use mixed-precision weights with 8-bit activations instead of zero-shot OMSE | ResNet50 · ImageNet | Top-1 accuracy: 70.06 -> 75.80 | Helped |
| Use ZeroQ W8A8 instead of DFQ W8A8 | MobileNetV2 · ImageNet | Top-1 accuracy: 71.20 -> 72.91 | Helped |
| Use mixed-precision weights with 6-bit activations instead of Integer-Only W6A6 | MobileNetV2 · ImageNet | Top-1 accuracy: 70.90 -> 72.85 | Helped |
| Use mixed-precision weights and 8-bit activations instead of fine-tuned DFC W4A4 | ResNet18 · ImageNet | Top-1 accuracy: 68.06 -> 69.05 | Helped |
| Replace Gaussian input data with batch-normalization-matched Distilled Data for sensitivity and activation-range estimation (ablation) | ZeroQ · ImageNet, ResNet50 | Top-1 accuracy: 75.44 -> 75.80 | Helped |
| Replace Gaussian input data with batch-normalization-matched Distilled Data for sensitivity and activation-range estimation (ablation) | ZeroQ · ImageNet, MobileNetV2 | Top-1 accuracy: 66.73 -> 68.83 | Helped |
| Replace Gaussian input data with batch-normalization-matched Distilled Data (ablation) | ZeroQ · ImageNet, ResNet18 | Top-1 accuracy: 67.87 -> 68.45 | Helped |
| Replace Gaussian input data with batch-normalization-matched Distilled Data (ablation) | ZeroQ · ImageNet, ShuffleNet | Top-1 accuracy: 56.23 -> 57.50 | Helped |
| Replace Gaussian input data with batch-normalization-matched Distilled Data (ablation) | ZeroQ · ImageNet, SqueezeNext | Top-1 accuracy: 48.41 -> 59.23 | Helped |
| Replace Gaussian input data with training data for sensitivity and activation-range estimation (ablation) | ZeroQ · ImageNet, ResNet50 | Top-1 accuracy: 75.44 -> 75.95 | Helped |
| Replace Gaussian input data with training data for sensitivity and activation-range estimation (ablation) | ZeroQ · ImageNet, MobileNetV2 | Top-1 accuracy: 66.73 -> 69.06 | Helped |
Closest papers by meaning
- It's All in the Teacher: Zero-Shot Quantization Brought Closer to the Teacher (CVPR 2022 · similarity 0.93)
- SynQ: Accurate Zero-shot Quantization by Synthesis-aware Fine-tuning (ICLR 2025 · similarity 0.92)
- Sharpness-Aware Data Generation for Zero-shot Quantization (ICML 2024 · similarity 0.91)
- Zero-Shot Quantization via Weight-Space Arithmetic (NeurIPS 2026 · similarity 0.90)
- Learning from Loss Landscape: Generalizable Mixed-Precision Quantization via Adaptive Sharpness-Aware Gradient Aligning (ICML 2025 · similarity 0.87)
- Efficient Multi-bit Quantization Network Training via Weight Bias Correction and Bit-wise Coreset Sampling (NeurIPS 2025 · 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.