A Parallel Processor System for the Simulation of the Neuron Network
Takashi Omori · Systems and Computers in Japan · 1989
Abstract This paper investigates the recognition and intelligent system based on the neuron‐like network. A problem in the study is the computational complexity in the simulation of the neuron network model. A dedicated parallel processor system was developed from such a viewpoint, and a high‐speed experiment was realized. This paper describes first the background of the system, where the neuron model is represented essentially by a system of differential equations with local interaction terms. A hardware composed of 64 microprocessors for the parallel computation, as well as the parallel processing description monitor system with interaction, which helps the realization, are described. The monitor discriminates clearly the physical parallel processors and the computation process to be realized logically on the system. The data needed for the interaction terms of the equation are obtained by interprocessor communication. Thus, the description of the parallel processing is made possible with the overhead comparable to the address computation in the usual computer. The parallel computer system was actually constructed, and it was verified that the system is effective for the correspondence problem for binocular vision.