English

PAC-Bayes with Minimax for Confidence-Rated Transduction

Machine Learning 2015-01-19 v1 Machine Learning

Abstract

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.

Keywords

Cite

@article{arxiv.1501.03838,
  title  = {PAC-Bayes with Minimax for Confidence-Rated Transduction},
  author = {Akshay Balsubramani and Yoav Freund},
  journal= {arXiv preprint arXiv:1501.03838},
  year   = {2015}
}