PURPOSE: To efficiently perform learning by calculating an index for indicating the error of output data and learning output data as an evaluation function for evaluating the learning state of a neural network and adjusting connection weight in the neural network so as to reduce the evaluation function. CONSTITUTION: Input data X1-X3 are respectively inputted to the three neurons N1(1) -N3(1) of an input layer and signals transmitted from the first neuron N1(1) of the input layer to the first neuron N1(2) of an intermediate layer are W11(2) X1. 1/2 of the sum of the square error of the output data Yi and teaching data di is satisfactory as the evaluation function for evaluating the learning state of the neural network and the index for indicating the error of the output data Yi and the teaching data di is satisfactory as the evaluation function. Forth output data Y4 are calculated corresponding to Y4=g(Y1, Y2, Y3)=1-(Y1+Y2+Y3).
展开▼