conftrace_

2015 COLT COLT 2015

Optimally Combining Classifiers Using Unlabeled Data

Abstract

We develop a worst-case analysis of aggregation of classifier ensembles for binary classification. The task of predicting to minimize error is formulated as a game played over a given set of unlabeled data (a transductive setting), where prior label information is encoded as constraints on the game. The minimax solution of this game identifies cases where a weighted combination of the classifiers can perform significantly better than any single classifier.

🐣 Hot Topic Early Bird - binary classification

Authors

Akshay Balsubramani, Yoav Freund

Topics

Machine Learning > Core Methods > Classification Machine Learning > Learning Types > Semi-Supervised Learning Machine Learning > Optimization & Theory > Theory Machine Learning > Core Methods > Ensemble Methods

Keywords

semi-supervised learning binary classification ensemble learning transductive learning classifier aggregation classifier ensemble weighted combination transductive setting

More from COLT 2015

All COLT 2015 papers →

Label optimal regret bounds for online local learning

Efficient Learning of Linear Separators under Bounded Noise

Efficient Representations for Lifelong Learning and Autoencoding

Minimax Fixed-Design Linear Regression

Escaping the Local Minima via Simulated Annealing: Optimization of Approximately Convex Functions

Bandit Convex Optimization: \sqrtT Regret in One Dimension