Combining multiple classifiers in probabilistic neural networks

0410332 - UTIA-B 20000048 RIV DE eng C - Conference Paper (international conference)
Grim, Jiří - Kittler, J. - Pudil, Pavel - Somol, Petr
Combining multiple classifiers in probabilistic neural networks.
Berlin: Springer, 2000. Lecture Notes in Computer Science., 1857. ISBN 3-540-67704-6. In: Multiple Classifier Systems. - (Kittler, J.; Roli, F.), s. 157-166
[First International Workshop MCS 2000. Cagliari (IT), 21.06.2000-23.06.2000]
Grant - others:GA AV(CZ) IAA2075703; MŠMT(CZ) VS96063; GA AV(CZ) KSK1075601
Program: IA; KS
Institutional research plan: AV0Z1075907
Subject RIV: BB - Applied Statistics, Operational Research

The paper summarizes main features of a new probabilistic approach to neural networks in the framework of statistical pattern recognition. Assuming approximation of class-conditional distributions by finite mixtures we identify formal neurons with the components of finite mixtures and therefore the EM algorithm can be used to optimize the parameters of neurons. In order to prevent the arising information loss we propose a parallel use of the output variables to design the Bayesian classifier.
Permanent Link: http://hdl.handle.net/11104/0130422