Research map / Label-efficient and robust learning / Knowledge distillation
Local Dense Logit Relations for Enhanced Knowledge Distillation
Liuchi Xu, Kang Liu, Jinshuai Liu, Lu Wang, Lisheng Xu, Jun Cheng
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 |
|---|---|---|---|
| use LDRLD instead of vanilla KD | ResNet32x4->ResNet8x4 · CIFAR-100 val | top-1 accuracy: 73.33 -> 77.20 | Improved |
| use LDRLD instead of vanilla KD | ResNet34->ResNet18 · ImageNet-1K val | top-1 accuracy: 70.66 -> 71.88 | Improved |
| use LDRLD instead of WTTM | ResNet34->ResNet18 · ImageNet-1K val | top-1 accuracy: 72.19 -> 71.88 | Hurt |
| add Adaptive Decay Weight (IRW and ERD) to LDRLD (ablation) | VGG13->VGG8 · CIFAR-100 | top-1 accuracy: 74.25 -> 74.46 | Improved |
| add Adaptive Decay Weight (IRW and ERD) to LDRLD (ablation) | ResNet50->MobileNetV2 · CIFAR-100 | top-1 accuracy: 69.23 -> 69.52 | Improved |
| add remaining non-target knowledge loss L_RNTK to L_Local (ablation) | ResNet32x4->ResNet8x4 · CIFAR-100 | top-1 accuracy: 76.55 -> 77.20 | Improved |
| add local relational loss L_Local to L_RNTK (ablation) | ResNet32x4->ResNet8x4 · CIFAR-100 | top-1 accuracy: 75.78 -> 77.20 | Improved |
| use recursion depth d=7 instead of smaller or larger depths (ablation) | LDRLD · CIFAR-100 | top-1 accuracy | Improved |
| use RNTK (d=7) instead of NCKD for non-target knowledge (ablation) | VGG13->VGG8 · CIFAR-100 | top-1 accuracy: 74.15 -> 74.28 | Improved |
| train with 0.1 noisy labels (ablation) | LDRLD · CIFAR-100 | top-1 accuracy: 77.20 -> 77.32 | No clear effect |
| combine LDRLD with DKD | ResNet32x4->ResNet8x4 · CIFAR-100 | top-1 accuracy: 77.20 -> 77.42 | Improved |
| combine LDRLD with IKD | ResNet32x4->ShuffleNetV2 · CIFAR-100 | top-1 accuracy: 77.33 -> 77.62 | Improved |
Closest papers by meaning
- Scaled Decoupled Distillation (CVPR 2024 · similarity 0.91)
- Enhancing Logits Distillation with Plug&Play Kendall's $\tau$ Ranking Loss (ICML 2025 · similarity 0.89)
- Maintaining Fairness in Logit-based Knowledge Distillation for Class-Incremental Learning (AAAI 2025 · similarity 0.89)
- PLD: A Choice-Theoretic List-Wise Knowledge Distillation (NeurIPS 2025 · similarity 0.88)
- Logit Standardization in Knowledge Distillation (CVPR 2024 · similarity 0.87)
- DeepKD: A Deeply Decoupled and Denoised Knowledge Distillation Trainer (NeurIPS 2025 · similarity 0.86)
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.
