Research map / Label-efficient and robust learning / Recognition and long-tail
Unbiased Risk Estimators Can Mislead: A Case Study of Learning with Complementary Labels
Yu-Ting Chou, Gang Niu, Hsuan-Tien Lin, Masashi Sugiyama
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 SCL-EXP instead of URE | Linear model · MNIST | accuracy: 0.8503 -> 0.9019 | Improved |
| Use SCL-NL instead of GA | MLP · MNIST | accuracy: 0.9088 -> 0.9469 | Improved |
| Use SCL-EXP instead of URE | Linear model · Kuzushi-MNIST | accuracy: 0.5613 -> 0.6132 | Improved |
| Use SCL-NL instead of GA | MLP · Kuzushi-MNIST | accuracy: 0.6567 -> 0.7644 | Improved |
| Use SCL-EXP instead of NN | Linear model · Fashion-MNIST | accuracy: 0.7755 -> 0.8282 | Improved |
| Use SCL-NL instead of GA | MLP · Fashion-MNIST | accuracy: 0.8019 -> 0.8456 | Improved |
| Use SCL-FWD instead of GA | Resnet · CIFAR-10 | accuracy: 0.3738 -> 0.5058 | Improved |
| Use SCL-FWD instead of GA | DenseNet · CIFAR-10 | accuracy: 0.4108 -> 0.5457 | Improved |
| Replace URE with SCL-FWD for gradient estimation | Linear model · MNIST | gradient MSE: 1.9692E-03 -> 7.7511E-06 | Improved |
| Replace URE with SCL-FWD for gradient estimation | DenseNet · CIFAR-10 | gradient MSE: 5.0196E-02 -> 2.0237E-04 | Improved |
Closest papers by meaning
- Complementary-Label Learning for Arbitrary Losses and Models (ICML 2019 · similarity 0.90)
- Learning with Complementary Labels Revisited: The Selected-Completely-at-Random Setting Is More Practical (ICML 2024 · similarity 0.89)
- A Generalized Unbiased Risk Estimator for Learning with Augmented Classes (AAAI 2023 · similarity 0.88)
- Positive-Unlabeled Learning with Non-Negative Risk Estimator (NeurIPS 2017 · similarity 0.88)
- Loss factorization, weakly supervised learning and label noise robustness (ICML 2016 · similarity 0.87)
- Towards Unbiased Exploration in Partial Label Learning (JMLR 2024 · similarity 0.87)
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.
