Research map / Label-efficient and robust learning / Recognition and long-tail
Exploiting Class Learnability in Noisy Data
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 bandit supervision to select classes instead of standard mini-batch training | Cifar100 classifier trained with added noisy classes · Cifar100 test, original classes | accuracy: 55.12 -> 67.14 | Helped |
| Add noisy classes to training compared with training on clean original data | ELU classifier · Cifar100 test, original classes | accuracy: 75.72 -> 55.12 | Hurt |
| Focus training on bandit-selected classes | Image tagging models on YFCC 100M and Recipe1m · Top-ranked classes | per-class F1-Score | Helped |
| Focus training on bandit-selected classes rather than seeking performance across all classes | Image tagging models on YFCC 100M and Recipe1m · Classes after the bandit model's F1 dropoff | per-class F1-Score | Hurt |
Closest papers by meaning
- Learning from Noisy Labels with No Change to the Training Process (ICML 2021 · similarity 0.87)
- Online Selective Classification with Limited Feedback (NeurIPS 2021 · similarity 0.87)
- Robustness of Accuracy Metric and its Inspirations in Learning with Noisy Labels (AAAI 2021 · similarity 0.87)
- Information-theoretic Limits of Online Classification with Noisy Labels (NeurIPS 2024 · similarity 0.87)
- Estimating Learnability in the Sublinear Data Regime (NeurIPS 2018 · similarity 0.87)
- Understanding Domain Generalization: A Noise Robustness Perspective (ICLR 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.