Research map / Learning theory and algorithms / Submodular and game algorithms
Rethinking Data-Free Quantization as a Zero-Sum Game
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 AdaSG instead of the strongest listed baseline | ResNet-20 quantization · CIFAR-10 3w3a | accuracy: 77.07 -> 84.14 | Helped |
| Use AdaSG instead of the strongest listed baseline | ResNet-20 quantization · CIFAR-10 4w4a | accuracy: 92.13 -> 92.10 | Hurt |
| Use AdaSG instead of the strongest listed baseline | ResNet-20 quantization · CIFAR-100 3w3a | accuracy: 48.25 -> 52.76 | Helped |
| Use AdaSG instead of the strongest listed baseline | ResNet-18 quantization · ImageNet 3w3a | accuracy: 23.37 -> 37.04 | Helped |
| Use AdaSG instead of the strongest listed baseline | MobileNetV2 quantization · ImageNet 4w4a | accuracy: 66.47 -> 65.15 | Hurt |
| Use AdaSG instead of the strongest listed baseline | ResNet-50 quantization · ImageNet 5w5a | accuracy: 76.00 -> 76.03 | Helped |
| Add agreement-sample loss L_as to AdaSG (ablation) | ResNet-18 quantization · ImageNet 3w3a | accuracy: 14.86 -> 37.04 | Helped |
| Add agreement-sample loss L_as to AdaSG (ablation) | ResNet-18 quantization · ImageNet 5w5a | accuracy: 69.15 -> 70.29 | Helped |
| Add disagreement-sample loss L_ds to AdaSG (ablation) | ResNet-18 quantization · ImageNet 3w3a | accuracy: 15.01 -> 37.04 | Helped |
| Add disagreement-sample loss L_ds to AdaSG (ablation) | ResNet-18 quantization · ImageNet 5w5a | accuracy: 70.06 -> 70.29 | Helped |
| Add batch-normalization-statistics loss L_BNS to AdaSG (ablation) | ResNet-18 quantization · ImageNet 3w3a | accuracy: 20.98 -> 37.04 | Helped |
| Add batch-normalization-statistics loss L_BNS to AdaSG (ablation) | ResNet-18 quantization · ImageNet 5w5a | accuracy: 67.35 -> 70.29 | Helped |
Closest papers by meaning
- Predicting Quantization Price for Selecting PTQ Configurations Before Deployment (NeurIPS 2026 · similarity 0.84)
- Understanding the Unfairness in Network Quantization (ICML 2025 · similarity 0.83)
- QuantDemoire: Quantization with Outlier Aware for Image Demoiréing (NeurIPS 2026 · similarity 0.83)
- Exploring Query Efficient Data Generation Towards Data-Free Model Stealing in Hard Label Setting (AAAI 2025 · similarity 0.83)
- Adaptive Dataset Quantization (AAAI 2025 · similarity 0.83)
- Enhancing Diversity for Data-free Quantization (CVPR 2025 · similarity 0.83)
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.