RECALL of multilayer perceptron

A method, called RECALL, which provides a way for directly perceiving a multilayer perceptron´s learning is discussed. After the learning process of networks has been accomplished, this method is used to recall what the network has learned, or what underlying rule it has had. The recall result will reflect the network´s internal character because multi layer perceptron discernment mainly corresponds to it. By this method it was found that in the multilayer perceptron with one hidden layer the relevant information in the internal representation is significantly reflected upon the recall result. The different desired value would attract the network´s attention to different places. With this it is believed that it is more reasonable to apply internal representation which implies the relationship among training samples. This RECALL method can be used to analyze other learning problems and provides some implication for similarity between the connectionist model and human beings