PAC-Bayes with Minimax for Confidence-Rated Transduction

2015/01/15 by Akshay Balsubramani, Yoav Freund, Balsubramani, Akshay +1
Computer Science · #Algorithms and Data Compression #Blind Source Separation Techniques #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Machine Learning and Algorithms

paper · pdf · doi:10.48550/arxiv.1501.03838

openalex publication_date 2015/01/15 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

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.

Related

• PAC Classification based on PAC Estimates of Label Class Distributions
• PAC learning with stable and private predictions
• Protected probabilistic classification
• Distribution-free binary classification: prediction sets, confidence intervals and calibration
• PAC-Bayes Analysis Beyond the Usual Bounds