0311211 20240103190335.820090326235959.9 000259567200006 10 s. cav_un_epca*0311333978-3-540-87558-1Part II52-61BerlinSpringer2008KůrkováV.NerudaR.KoutníkJ. Probabilistic neural networks Feature extraction Recognition of numerals cav_un_auth*0101091 Grim Jiří UTIA-B Department of Pattern Recognition Ústav teorie informace a automatizace AV ČR, v. v. i. 1M0572 GA MŠk cav_un_auth*0001814 2C06019 GA MŠk CZ cav_un_auth*0216518 GA102/07/1594 GA ČR cav_un_auth*0228611 CEZ:AV0Z10750506 In order to design probabilistic neural networks in the framework of pattern recognition we estimate class-conditional probability distributions in the form of finite mixtures of product components. As the mixture components correspond to neurons we specify the properties of neurons in terms of component parameters. The probabilistic features defined by neuron outputs can be used to transform the classification problem without information loss and, simultaneously, the Shannon entropy of the feature space is minimized. We show that, instead of dimensionality reduction, the decision problem can be simplified by using binary approximation of the probabilistic features. In experiments the resulting binary features improve recognition accuracy but also they are nearly independent - in accordance with the minimum entropy property. Extrakce binárních příznaků pomocí pravděpodobnostních neuronových sítí cav_un_auth*0241921 ICANN 2008. International Conference on Artificial Neural Networks /18./ Prague 03.09.2008-06.09.2008 CZ 2009 IN http://hdl.handle.net/11104/0162890 2008 000259567200006 WOS cav_un_epca*0311333 Artificial Neural Networks - ICANN 2008 Part II 978-3-540-87558-1 52 61 Berlin Springer 2008 Lecture Notes in Computer Science 5164 Kůrková V. 340 Neruda R. 340 Koutník J. 340