Research map / Label-efficient and robust learning / Knowledge distillation
Self-Distillation From the Last Mini-Batch for Consistency Regularization
Check 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 |
|---|---|---|---|
| Add last-mini-batch self-distillation | CNN classifiers · CIFAR-10 | top-1 error rate (%): 0.37% to 1.01% | Helped |
| Add last-mini-batch self-distillation | CNN classifiers · CIFAR-100 | top-1 error rate (%): 0.83% to 2.50% | Helped |
| Add last-mini-batch self-distillation | CNN classifiers · TinyImageNet | top-1 error rate (%): 0.81% to 3.17% | Helped |
| Add last-mini-batch self-distillation to CutOut | CutOut · CIFAR-10 | top-1 error rate (%): 0.45% to 1.14% | Helped |
| Add last-mini-batch self-distillation to CutOut | CutOut · CIFAR-100 | top-1 error rate (%): 0.84% to 2.74% | Helped |
| Add last-mini-batch self-distillation to CutMix | CutMix · CIFAR-10 | top-1 error rate (%): 0.42% | Helped |
| Add last-mini-batch self-distillation to CutMix | CutMix · CIFAR-100 | top-1 error rate (%): 1.02% | Helped |
| Add last-mini-batch self-distillation under symmetric label noise | DLB · CIFAR-100, noise rate p=0.1 | top-1 error rate (%): 2.38% | Helped |
| Add last-mini-batch self-distillation under symmetric label noise | DLB · CIFAR-100, noise rate p=0.6 | top-1 error rate (%): 4.44% | Helped |
| Add last-batch distillation loss (ablation) | DLB with ResNet-32 · CIFAR-100 | top-1 error rate (%): 28.01 -> 26.00 | Helped |
| Add last-batch distillation loss (ablation) | DLB with Vgg-16 · CIFAR-100 | top-1 error rate (%): 25.45 -> 23.88 | Helped |
| Add last-batch distillation loss (ablation) | DLB with Vgg-19 · CIFAR-100 | top-1 error rate (%): 25.73 -> 24.53 | Helped |
Closest papers by meaning
- Debiased Distillation for Consistency Regularization (AAAI 2025 · similarity 0.91)
- Dataset Distillation via Knowledge Distillation: Towards Efficient Self-Supervised Pre-training of Deep Networks (ICLR 2025 · similarity 0.90)
- DistiLLM: Towards Streamlined Distillation for Large Language Models (ICML 2024 · similarity 0.88)
- Gap Preserving Distillation by Building Bidirectional Mappings with A Dynamic Teacher (ICLR 2025 · similarity 0.88)
- DA-KD: Difficulty-Aware Knowledge Distillation for Efficient Large Language Models (ICML 2025 · similarity 0.88)
- Progressively Knowledge Distillation via Re-parameterizing Diffusion Reverse Process (AAAI 2024 · similarity 0.88)
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.