Research map / Label-efficient and robust learning / Noisy labels
MentorNet: Learning Data-Driven Curriculum for Very Deep Neural Networks on Corrupted Labels
Read the paperCheck 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 |
|---|---|---|---|
| Use the data-driven MentorNet curriculum instead of the strongest listed baseline | ResNet-101 · CIFAR-10 and CIFAR-100 with noise fractions 0.2, 0.4, and 0.8 | validation accuracy: CIFAR-100: 0.70 -> 0.73; 0.55 -> 0.68; 0.16 -> 0.35. CIFAR-10: 0.89 -> 0.92; 0.85 -> 0.89; 0.35 -> 0.49 | Helped |
| Use the data-driven MentorNet curriculum instead of the strongest listed baseline | Inception · CIFAR-10 and CIFAR-100 with noise fractions 0.2, 0.4, and 0.8 | validation accuracy: CIFAR-100: 0.44 -> 0.46; 0.39 -> 0.41; 0.17 -> 0.20. CIFAR-10: 0.80 -> 0.79; 0.74 -> 0.76; 0.44 -> 0.46 | Mixed |
| Add MentorNet to the fully regularized model | Inception-ResNet v2 · ImageNet ILSVRC12, noise 0.4 | P@1: 0.628 -> 0.651 | Helped |
| Add MentorNet to the fully regularized model | Inception-ResNet v2 · ImageNet ILSVRC12, noise 0.4 | P@5: 0.845 -> 0.859 | Helped |
| Add MentorNet to the unregularized model | Inception-ResNet v2 · ImageNet ILSVRC12, noise 0.4 | P@1: 0.575 -> 0.590 | Helped |
| Add MentorNet to the unregularized model | Inception-ResNet v2 · ImageNet ILSVRC12, noise 0.4 | P@5: 0.809 -> 0.814 | Helped |
| Use MentorNet with additional verification labels | Inception-ResNet v2 · WebVision Entire; ILSVRC12 validation | top-1 accuracy: 0.625 -> 0.642 | Helped |
| Use MentorNet with additional verification labels | Inception-ResNet v2 · WebVision Entire; WebVision validation | top-1 accuracy: 0.708 -> 0.726 | Helped |
| Use MentorNet rather than FullModel | Inception-ResNet v2 · WebVision Mini; ILSVRC12 validation | top-1 accuracy: 0.585 -> 0.638 | Helped |
Closest papers by meaning
- On The Power of Curriculum Learning in Training Deep Networks (ICML 2019 · similarity 0.88)
- Fidelity-Weighted Learning (ICLR 2018 · similarity 0.87)
- Learning Student Networks in the Wild (CVPR 2021 · similarity 0.87)
- Robust Curriculum Learning: from clean label detection to noisy label self-correction (ICLR 2021 · similarity 0.87)
- Learning with Structural Labels for Learning with Noisy Labels (CVPR 2024 · similarity 0.85)
- Curriculum-Enhanced Residual Soft An-Isotropic Normalization for Over-Smoothness in Deep GNNs (AAAI 2024 · similarity 0.84)
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.