Information theoretic combination of pattern classifiers

PATTERN RECOGNITION

卷:43

Article

Meynet, Julien^1,2  Thiran, Jean-Philippe² 

[1] Yahoo France R&D, F-38130 Echirolles, France

[2] Ecole Polytech Fed Lausanne, Signal Proc Labs, LTS5, CH-1015 Lausanne, Switzerland

关键词: Machine learning;    Pattern recognition;    Classifier combination;    Information theory;    Mutual information;    Diversity;   

DOI  :  10.1016/j.patcog.2010.04.013

来源: Elsevier

【 摘 要 】

Combining several classifiers has proved to be an effective machine learning technique. Two concepts clearly influence the performances of an ensemble of classifiers: the diversity between classifiers and the individual accuracies of the classifiers. In this paper we propose an information theoretic framework to establish a link between these quantities. As they appear to be contradictory, we propose an information theoretic score (ITS) that expresses a trade-off between individual accuracy and diversity. This technique can be directly used, for example, for selecting an optimal ensemble in a pool of classifiers. We perform experiments in the context of overproduction and selection of classifiers, showing that the selection based on the ITS outperforms state-of-the-art diversity-based selection techniques. (C) 2010 Elsevier Ltd. All rights reserved.

【 授权许可】

Free