Research map / Label-efficient and robust learning / Knowledge distillation
MixMix: All You Need for Data-Free Compression Are Feature and Data Mixing
Yuhang Li, Feng Zhu, Ruihao Gong, Mingzhu Shen, Xin Dong, Fengwei Yu, Shaoqing Lu, Shi Gu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 Data Mixing to Feature Mixing with m'=1 (ablation) | MixMix · ResNet-50 structured pruning | top-1 accuracy: 64.46 -> 68.16 | Improved |
| add Data Mixing to Feature Mixing with m'=2 (ablation) | MixMix · ResNet-50 structured pruning | top-1 accuracy: 68.87 -> 69.29 | Improved |
| add Data Mixing to Feature Mixing with m'=3 (ablation) | MixMix · ResNet-50 structured pruning | top-1 accuracy: 69.49 -> 69.80 | Improved |
| add Data Mixing to Feature Mixing with m'=4 (ablation) | MixMix · ResNet-50 structured pruning | top-1 accuracy: 69.54 -> 69.47 | No clear effect |
| mix more models in Feature Mixing (m' 1 -> 3) (ablation) | MixMix · ResNet-50 structured pruning | top-1 accuracy: 64.46 -> 69.49 | Improved |
| mix more models in Feature Mixing (m' 3 -> 4) (ablation) | MixMix · ResNet-50 structured pruning | top-1 accuracy: 69.49 -> 69.54 | No clear effect |
| use MixMix data instead of ZeroQ data | BRECQ PTQ · MobileNetV2 W4A4 | top-1 accuracy: 49.83 -> 64.01 | Improved |
| use MixMix data instead of DeepInversion data | structured pruning · MobileNetV2 | top-1 accuracy: 15.32 -> 42.47 | Improved |
| use MixMix data instead of DeepInversion data | structured pruning · MNasNet | top-1 accuracy: 22.62 -> 43.41 | Improved |
| use MixMix data instead of KW data | LSQ QAT · MobileNetV2 W4A4 | top-1 accuracy: 66.07 -> 67.74 | Improved |
| add MixMix objective to DAFL | DAFL · CIFAR10 ResNet-34 -> ResNet-18 | accuracy: 92.22 -> 94.79 | Improved |
| use MixMix data instead of DeepInversion data | data-free KD · CIFAR10 VGG11 -> VGG11 | accuracy: 90.78 -> 93.97 | Improved |
Closest papers by meaning
- AutoMix: Unveiling the Power of Mixup for Stronger Classifiers (ECCV 2022 · similarity 0.86)
- DiffuseMix: Label-Preserving Data Augmentation with Diffusion Models (CVPR 2024 · similarity 0.86)
- Adversarial AutoMixup (ICLR 2024 · similarity 0.84)
- SnapMix: Semantically Proportional Mixing for Augmenting Fine-grained Data (AAAI 2021 · similarity 0.84)
- DP-Mix: Mixup-based Data Augmentation for Differentially Private Learning (NeurIPS 2023 · similarity 0.84)
- Data-Free Quantization via Pseudo-label Filtering (CVPR 2024 · 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.
