Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Hard Sample Matters a Lot in Zero-Shot Quantization
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 |
|---|---|---|---|
| use HAST instead of IntraQ | ResNet-20 quantized to W3A3 · CIFAR-10 | top-1 accuracy: 77.07 -> 88.34 | Helped |
| use HAST instead of IntraQ | ResNet-20 quantized to W3A3 · CIFAR-100 | top-1 accuracy: 48.25 -> 55.67 | Helped |
| use HAST instead of AIT | ResNet-18 quantized to W3A3 · ImageNet | top-1 accuracy: 36.70 -> 51.15 | Helped |
| use HAST instead of AIT | MobileNetV2 quantized to W5A5 and W4A4 · ImageNet | top-1 accuracy: 71.96 vs 71.72; 66.47 vs 65.60 | Hurt |
| add hard sample synthesis (ablation) | ZeroQ+IL, ResNet-18 W3A3 · ImageNet | top-1 accuracy: 44.68 -> 45.56 | Helped |
| add sample difficulty promotion (ablation) | ZeroQ+IL, ResNet-18 W3A3 · ImageNet | top-1 accuracy: 44.68 -> 48.47 | Helped |
| add feature alignment (ablation) | ZeroQ+IL, ResNet-18 W3A3 · ImageNet | top-1 accuracy: 44.68 -> 46.07 | Helped |
| add hard sample synthesis, sample difficulty promotion, and feature alignment (ablation) | ZeroQ+IL, ResNet-18 W3A3 · ImageNet | top-1 accuracy: 44.68 -> 51.15 | Helped |
| replace direct feature alignment with relaxed feature alignment (ablation) | HAST, ResNet-18 W3A3 · ImageNet | top-1 accuracy: 45.52 -> 51.15 | Helped |
| replace direct feature alignment with relaxed feature alignment (ablation) | HAST, ResNet-20 W3A3 · CIFAR-10 | top-1 accuracy: 83.92 -> 88.34 | Helped |
| reduce synthetic sample count from 5120 to 256 (ablation) | HAST, ResNet-18 W3A3 · ImageNet | top-1 accuracy: 51.15 -> 49.17 | Hurt |
| increase synthetic sample count from 256 to 5120 (ablation) | HAST, ResNet-18 W3A3 · ImageNet | top-1 accuracy: 49.17 -> 51.15 | Helped |
Closest papers by meaning
- SynQ: Accurate Zero-shot Quantization by Synthesis-aware Fine-tuning (ICLR 2025 · similarity 0.91)
- It's All in the Teacher: Zero-Shot Quantization Brought Closer to the Teacher (CVPR 2022 · similarity 0.90)
- Sharpness-Aware Data Generation for Zero-shot Quantization (ICML 2024 · similarity 0.89)
- Zero-Shot Quantization via Weight-Space Arithmetic (NeurIPS 2026 · similarity 0.85)
- One-Step Forward and Backtrack: Overcoming Zig-Zagging in Loss-Aware Quantization Training (AAAI 2024 · similarity 0.85)
- Data-Free Quantization via Pseudo-label Filtering (CVPR 2024 · similarity 0.85)
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.