Title :
About perceptron realizations of Bayesian decisions
Author_Institution :
Inst. of Inf. Theory & Autom., Prague, Czech Republic
Abstract :
It is shown that one can imitate the Bayesian discrimination and classification of exponentially distributed random signals by the perceptrons with one hidden layer. The number of unknown weights just by 2 exceeds the number of parameters figuring in the exponential distribution. Learning is thus relatively easy
Keywords :
Bayes methods; Bayesian decisions; Bayesian discrimination; classification; exponentially distributed random signals; perceptron realizations; Acoustics; Bayesian methods; Computer aided analysis; Content addressable storage; Feature extraction; Information theory; Linearity; Random processes; Statistics;
Conference_Titel :
Neural Networks, 1996., IEEE International Conference on
Conference_Location :
Washington, DC
Print_ISBN :
0-7803-3210-5
DOI :
10.1109/ICNN.1996.548900
