Cognitron: A self-organizing multilayered neural network

摘要: A new hypothesis for the organization of synapses between neurons is proposed: "The synapse from neuron x to y reinforced when fires provided that no in vicinity firing stronger than y". By introducing this hypothesis, a algorithm with which multilayered neural network effectively organized can be deduced. self-organizing network, named "cognitron", constructed following algorithm, and simulated on digital computer. Unlike usual brain models such as three-layered perceptron, self-organization cognitron progresses favorably without having "teacher" instructs all particulars how individual cells respond. After repetitive presentations several stimulus patterns, self-organized way receptive fields become relatively larger deeper layer. Each cell final layer integrates information whole parts first selectively responds specific pattern or feature.