SEPARATELY TRAINED SYSTEM AND METHOD USING TWO-LAYERED NEURAL NETWORK WITH TARGET VALUES OF HIDDEN NODES

摘要

A separately training system and a method using a two-layered neural network having target values of hidden nodes are provided to reduce the amount of calculation and to enhance reliability by performing a separately training process of an upper layer and a lower layer. An input layer(110) receives training data from a user and includes input nodes. A hidden layer(130) includes hidden nodes. A first connection unit(120) consisting of three modules connects the input layer and the hidden layer and performs a training process by changing a weight between the input layer and the hidden layer. An output layer(150) outputs completed training data. A second connection part(140) connects the input layer and the hidden layer and produces target values of the hidden nodes from present errors of the output nodes. A control part(160) performs a repetitive training process to the target values of the hidden nodes between the input nodes and the hidden nodes by fixing the first or the second connection part and performing the training process with the first or the second connection part.