Estimation of generalization capability by combination of new information criterion and cross validation

The authors propose a novel method of selecting the optimal neural network structure with maximum generalization capability. By expanding Akaike's information criterion, they propose a new information criterion that can estimate generalization capability without the maximum likelihood estimator of synaptic weights. The cross validation method is used to calculate the new information criterion. Computer simulation shows that the proposed information criterion can accurately predict the generalization capability of multilayer perceptrons, and thus the optimal number of hidden units can be determined.<<ETX>>