MEASURING OF GENERALIXATION CAPACITY FOR NEURAL NETWORKS

System model of recognition pattern process is presented. Conception of nondegenerated recognition scheme is introduced. Method of generalization capacity measuring for neural networks, which is suitable for small value of training set, is suggested. Methodology of the experiments is given. Results of the generalization capacity measuring for feed forward neural networks are shown.

Authors: A. Yu. Dorogov, V. C. Abaturov

Direction: Informatics, management and Computer Technology

Keywords: Neural network, generalization capacity, separating power, system model


View full article