Title :
Using of the corrected reclassification method estimation for the recognition system tuning
Author :
Arbekov, I.M. ; Artamonov, S.M.
Author_Institution :
Sci. Centre of Neurocomput., Acad. of Sci., Moscow, Russia
Abstract :
A synthesis of a recognition system with variable structure for recognition of two pattern classes is considered. The system is a three layer neural network (NN). The NN consists of linear threshold elements. The first layer tuning uses the recognition exactness estimation for the reclassification method. This method is called "with examination on the training set".
Keywords :
multilayer perceptrons; pattern classification; threshold elements; corrected reclassification method estimation; linear threshold elements; recognition system tuning; three layer neural network; variable structure; Bayesian methods; Frequency estimation; Histograms; Hydrogen; Network synthesis; Neural networks; Pattern recognition; Probability density function; Random variables; Vectors;
Conference_Titel :
Neural Networks, 1993. IJCNN '93-Nagoya. Proceedings of 1993 International Joint Conference on
Print_ISBN :
0-7803-1421-2
DOI :
10.1109/IJCNN.1993.713939
