Second learning of generalized multilayer perceptrons

Abstract

A method of pruning away redundant hidden neurons is given and discussed. Then the necessary and sufficient condition of the generalized value of change during the pruning process is given, as well as some relevant theorems, which leads to the formula of the generalized value of change. Finally, the steps of learning the patterns in the testing set which is called the second learning is addressed, and the change of generalization ability during second learning is proved.<>