ERROR BACK-PROPAGATION LEARNING METHOD OF AN ENTIRE DIRECTION NEURON CIRCUIT NETWORK USING AN OPTIMUM LEARNING RATE

摘要

PURPOSE: An error back-propagation learning method of an entire direction neuron circuit network is provided to use an optimum learning rate. CONSTITUTION: The error back-propagation learning method comprises the steps of: calculating an optimum learning rate of an output layer weight for minimizing an error of an output layer in a state of fixing a value of an interlayer neuron; changing the output layer weight on the basis of the optimum learning rate calculated in the step of calculating; calculating an optimum learning rate according to learning pattern for changing the interlayer weight in a state of fixing the changed output layer weight while calculating an optimum learning rate according to the interlayer neuron; and changing an interlayer weight by use of the calculated learning rate.