NEURAL NETWORK AND ADAPTIVE INFERENCE SYSTEM

PURPOSE: To solve a logic function which cannot linearly be separated with two layers by transforming an input value and forming the product. CONSTITUTION: An input signal is linearly transformed in a preprocessor and the outputs are connected as the product of linear transformation. Then, the product is examined based on a threshold. In such a case, the constitution of a processing element executes non-linear transformation between input data and an output result and it forms Boolean logic function which can linearly be separated and cannot linearly be separated. The neural node element 60 contains power series transformation units 62 containing product units 64 and the output signals of the processing element s64 are fed back through a learning algorithm 66. The element 60 executes a calculation operation and outputs all the logic functions as much as possible. Thus, the two layer neural network to which judgment is requested and which solves a arbitrary problem can be provided.