A new arXiv preprint examines positive-unlabeled (PU) classification when the common SCAR assumption does not hold. The authors study logistic regression approaches, including a cluster-based method and Lasso-regularized variants, and add oversampling to improve performance. The work appears as a cross-listing in cs.AI and cs.LG.
A new arXiv paper addresses AUC maximization for imbalanced binary classification when reliable negative examples are unavailable. The authors tackle the realistic setting of biased positive-unlabeled data and incorporate confidence estimates into the learning procedure. This approach aims to improve ranking performance without requiring clean negative labels.
A new arXiv preprint addresses unlabeled-unlabeled (UU) learning, a setting where a binary classifier is trained from two unlabeled datasets that have differing class priors. The authors introduce importance weighting to handle distribution shift in this framework, which generalizes approaches such as positive-unlabeled learning. The work is listed as a cross-submission in the cs.AI category.