PAC-Bayes with Minimax for Confidence-Rated Transduction
This work addresses the challenge of providing reliable confidence estimates in transductive learning for applications like safety-critical systems, though it appears incremental as it builds on existing PAC-Bayes and minimax frameworks.
The paper tackles the problem of confidence-rated transductive prediction using an ensemble of binary classifiers when unlabeled test data are known in advance, deriving minimax optimal rules and applying PAC-Bayes analysis to obtain data-dependent performance guarantees without distributional assumptions.
We consider using an ensemble of binary classifiers for transductive prediction, when unlabeled test data are known in advance. We derive minimax optimal rules for confidence-rated prediction in this setting. By using PAC-Bayes analysis on these rules, we obtain data-dependent performance guarantees without distributional assumptions on the data. Our analysis techniques are readily extended to a setting in which the predictor is allowed to abstain.