Research map / Label-efficient and robust learning / Recognition and long-tail
Making Binary Classification From Multiple Unlabeled Datasets Almost Free of Supervision
Yuhao Wu, Xiaobo Xia, Jun Yu, Bo Han, Gang Niu, Masashi Sugiyama, Tongliang Liu
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 ReMPE (regrouping) instead of standard MPE with loss-based confident examples | MOS · F-MNIST (binary, m=10 unlabeled sets) | class-prior estimation error x100: 4.76 -> 2.81 | Improved |
| use ReMPE instead of standard MPE with latent representations | MOS · CIFAR-10 (binary, m=10 unlabeled sets) | class-prior estimation error x100: 3.17 -> 2.15 | Improved |
| use ReMPE instead of standard MPE with loss-based confident examples | MOS · K-MNIST (binary, m=10 unlabeled sets) | class-prior estimation error x100: 3.99 -> 4.08 | No clear effect |
| use BBE instead of ReMPE with loss-based confident examples | MOS · CIFAR-10 (binary, m=10 unlabeled sets) | class-prior estimation error x100: 3.44 -> 2.41 | Improved |
| collect confident examples by latent representations rather than loss distribution (ReMPE) | MOS · CIFAR-10 (binary, m=10 unlabeled sets) | class-prior estimation error x100: 3.44 -> 2.15 | Improved |
| collect confident examples by latent representations rather than loss distribution (BBE) | MOS · K-MNIST (binary, m=10 unlabeled sets) | class-prior estimation error x100: 4.49 -> 4.75 | Hurt |
| use estimated class priors (MOS, latent+BBE) instead of true class priors for MCM | MCM · MNIST (binary, MCM classifier) | accuracy (%): 97.09 -> 97.05 | No clear effect |
| use estimated class priors (MOS, loss+Standard MPE) instead of true priors for MCM | MCM · K-MNIST (binary, MCM classifier) | accuracy (%): 90.35 -> 89.38 | Hurt |
| use estimated class priors (MOS, prob+ReMPE) instead of true priors for MCM | MCM · CIFAR-10 (binary, MCM classifier) | accuracy (%): 86.29 -> 86.26 | No clear effect |
Closest papers by meaning
- Consistent Multi-Class Classification from Multiple Unlabeled Datasets (ICLR 2024 · similarity 0.89)
- Binary Classification from Multiple Unlabeled Datasets via Surrogate Set Classification (ICML 2021 · similarity 0.88)
- Class Prior Estimation with Biased Positives and Unlabeled Examples (AAAI 2020 · similarity 0.87)
- On the Minimal Supervision for Training Any Binary Classifier from Only Unlabeled Data (ICLR 2019 · similarity 0.87)
- AUC Optimization from Multiple Unlabeled Datasets (AAAI 2024 · similarity 0.85)
- Unsupervised Learning for Class Distribution Mismatch (ICML 2025 · 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.
