Research map / Label-efficient and robust learning / Knowledge distillation
Training Deep Neural Networks in Generations: A More Tolerant Teacher Educates Better Students
Chenglin Yang, Lingxi Xie, Siyuan Qiao, Alan L. Yuille
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 |
|---|---|---|---|
| tolerant TSD patriarch instead of one-hot patriarch (ablation) | training in generations (ResNet-110) · CIFAR100 | best accuracy: 72.61% -> 73.72% | Improved |
| label smoothing patriarch instead of one-hot (ablation) | training in generations · CIFAR100 | best accuracy: 72.61% -> 73.46% | Improved |
| confidence penalty patriarch instead of one-hot (ablation) | training in generations · CIFAR100 | best accuracy: 72.61% -> 72.86% | No clear effect |
| increase u(eta) from 0.6 to 0.7 (ablation) | TSD patriarch · CIFAR100 | best accuracy: 73.72% -> 73.18% | Hurt |
| train in generations with tolerant teacher | DenseNet-100 · CIFAR100 | error: 22.20 -> 20.51 | Improved |
| ensemble models across generations | DenseNet-100 · CIFAR100 | error: 22.20 -> 17.25 | Improved |
| train in generations with tolerant teacher | DenseNet-190 · CIFAR100 | error: 17.22 -> 17.01 | No clear effect |
| tolerant patriarch alone (Gen #0) (ablation) | ResNet-18 · ILSVRC2012 | top-1 accuracy: 69.50 -> 67.48 | Hurt |
| train in 5 generations with tolerant teacher | ResNet-18 · ILSVRC2012 | top-1 accuracy: 69.50 -> 70.40 | Improved |
| train in generations with tolerant teacher | ResNet-18 features · Caltech256 transfer | accuracy: 75.94 -> 77.83 | Improved |
Closest papers by meaning
- Snapshot Distillation: Teacher-Student Optimization in One Generation (CVPR 2019 · similarity 0.88)
- On The Power of Curriculum Learning in Training Deep Networks (ICML 2019 · similarity 0.88)
- Excess Risk of Two-Layer ReLU Neural Networks in Teacher-Student Settings and its Superiority to Kernel Methods (ICLR 2023 · similarity 0.87)
- Adversarial Learning of Portable Student Networks (AAAI 2018 · similarity 0.87)
- Curriculum Learning by Transfer Learning: Theory and Experiments with Deep Networks (ICML 2018 · similarity 0.87)
- Hybrid Data-Free Knowledge Distillation (AAAI 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.
