Researchers have developed a new framework called PUe to enhance Positive-Unlabeled (PU) learning by addressing selection bias in real-world datasets. This framework, building on prior work by Bekker et al., introduces a normalized inverse-probability-weighted PU risk formulation and integrates with modern cost-sensitive methods. Experiments on datasets like MNIST, CIFAR-10, and ADNI show PUe outperforms existing PU baselines when label distributions are uneven. AI
IMPACT This research could improve the accuracy of machine learning models trained on real-world data that often suffers from biased labeling.
RANK_REASON The cluster contains a research paper detailing a new machine learning framework.
- Alzheimer's Disease Neuroimaging Initiative
- Bekker et al.
- CIFAR-10
- Jeanne Guesdon
- MNIST database
- Positive Unlabeled (PU) learning
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
