Neural network test method e.g. for air conditioner, by testing for error between network output signal and teacher output signal for each input signal sub-space

摘要

The method involves dividing an m-dimensional input signal space (X) at an input layer into a number of input signal sub-spaces (XP), while one region of each input signal is defined for input to the input layer as a one-dimensional input signal space. The neural network is tested by testing for an error (E) between the network output signal and the teacher output signal for each input signal sub-space. The method may also involve calculating an interval error representing an achievable region of the error between the teacher output signal and the network output signal. Independent claims are included for a data carrier in which a program for performing the test method is recorded, a test apparatus, and a method of manufacturing an air conditioner.