Group Approach to Solving the Tasks of Recognition

Abstract

In this work semi-supervised learning was considered. To solve the problem of semi-supervised learning, CASVM and CANN algorithms were developed. The algorithms are based on a combination of collective cluster analysis and kernel methods. A probabilistic model of classification with use of cluster ensemble was proposed. Within the model, error probability of CANN was studied. Assumptions that make probability of error converge to zero were formulated. The proposed algorithms were experimentally tested on a hyper spectral image. It is shown that CASVM is more noise resistant than the standard SVM.