Research map / Label-efficient and robust learning / Knowledge distillation
MixSKD: Self-Knowledge Distillation from Mixup for Image Recognition
Chuanguang Yang, Zhulin An, Helong Zhou, Linhang Cai, Xiang Zhi, Jiwen Wu, Yongjun Xu, Qian Zhang
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 Mixup task loss with auxiliary branches (ablation) | MixSKD (ResNet-50) · ImageNet | top-1: 77.08 -> 77.68 | Improved |
| add feature l2 Self-KD (ablation) | MixSKD (ResNet-50) · ImageNet | top-1: 77.68 -> 78.14 | Improved |
| add adversarial discriminator loss to feature Self-KD (ablation) | MixSKD (ResNet-50) · ImageNet | top-1: 78.14 -> 78.47 | Improved |
| add logit Self-KD over auxiliary branches (ablation) | MixSKD (ResNet-50) · ImageNet | top-1: 77.68 -> 78.06 | Improved |
| add self-teacher training and backbone logit Self-KD (ablation) | MixSKD (ResNet-50) · ImageNet | top-1: 77.68 -> 78.23 | Improved |
| combine all logit Self-KD losses (ablation) | MixSKD (ResNet-50) · ImageNet | top-1: 77.68 -> 78.54 | Improved |
| combine feature and logit Self-KD (ablation) | MixSKD (ResNet-50) · ImageNet | top-1: 77.68 -> 78.76 | Improved |
| build self-teacher from all stages instead of stage 1 only (ablation) | MixSKD (ResNet-50) · ImageNet | top-1: 78.01 -> 78.76 | Improved |
| build self-teacher from all stages instead of stage 4 only (ablation) | MixSKD (ResNet-50) · ImageNet | top-1: 78.42 -> 78.76 | Improved |
| set temperature T=3 instead of T=1 (ablation) | MixSKD (ResNet-50) · ImageNet | top-1: 77.94 -> 78.76 | Improved |
| set temperature T=3 instead of T=5 (ablation) | MixSKD (ResNet-50) · ImageNet | top-1: 78.37 -> 78.76 | Improved |
| add MixSKD to WRN-16-2 under FGSM attack eps=0.0001 | WRN-16-2 + Mixup · CIFAR-100 | top-1 under FGSM: 63.1 -> 67.7 | Improved |
Closest papers by meaning
- Refine Myself by Teaching Myself: Feature Refinement via Self-Knowledge Distillation (CVPR 2021 · similarity 0.89)
- SDPGO: Efficient Self-Distillation Training Meets Proximal Gradient Optimization (NeurIPS 2025 · similarity 0.88)
- Black-Box Few-Shot Knowledge Distillation (ECCV 2022 · similarity 0.87)
- CrossKD: Cross-Head Knowledge Distillation for Object Detection (CVPR 2024 · similarity 0.86)
- Multi-label Self Knowledge Distillation (AAAI 2025 · similarity 0.86)
- Wasserstein Distance Rivals Kullback-Leibler Divergence for Knowledge Distillation (NeurIPS 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.
