STRUCTURE DEVICE FOR NEURAL NETWORK AND PATTERN RECOGNITION DEVICE USING NEURAL NETWORK

摘要

PURPOSE:To make the constitution of a neural network arithmetic means irreducible and necessary by simple arithmetic by providing a means which optimizes the neural network according to sets of tutor data and specific conditions which are given previously. CONSTITUTION:A storage means 103 and a neural network system optimum design means 105 are provided. The neural network arithmetic means 102 processes an output signal corresponding to an input signal according to the constitution of the multi-layered neural network. Combinations of desired outputs to inputs (tutor data) are stored in the storage means 103 and the neural network system optimum design means 105 structures the neural network which satisfies the conditions according to the specified conditions and teacher data. Consequently, the neural network can be structured analytically and constitution which hardly causes misrecognition is obtained.