Abstract:
We introduce two generative classifiers that classify based on the pairwise similarities between samples or on the Euclidean features describing the samples: the regulari...Show MoreMetadata
Abstract:
We introduce two generative classifiers that classify based on the pairwise similarities between samples or on the Euclidean features describing the samples: the regularized local similarity discriminant analysis classifier for similarities and the local Bayesian discriminant analysis classifier for Euclidean features. Both new classifiers provide low-variance probability estimates of class labels from low-bias probabilistic models in their respective domains. We combine these two novel classifiers in a naive Bayes framework to form a classifier that fuses similarity and feature information to produce accurate probability estimates for the class labels. Experimental results on several benchmark datasets demonstrate that the two classifiers improve upon the state-of-the-art in their respective domains, and that the fused classifier adaptively uses the best information for classification.
Published in: 2009 12th International Conference on Information Fusion
Date of Conference: 06-09 July 2009
Date Added to IEEE Xplore: 18 August 2009
Print ISBN:978-0-9824-4380-4
Conference Location: Seattle, WA, USA