Multiresolution-based committees of networks: a Bayesian point of view

It has been shown that a better classification with less generalization error can be obtained by combining outputs of multiple classifiers. This paper describes an extension to the combining method of ensemble classifier based on the multiresolution concept. We propose a new method using adaptive time-frequency representation (local discriminant basis) and show that our proposed method can be interpreted as a Bayesian committees of networks. In case of the "weighted majority" combining of classifiers, our proposed method applied to a face recognition problem gives not only highly increased accuracy but also simplified simplicity.