# Cognitron **Domain:** Neural Networks / Computer Vision **Doc Type:** Historical System Node **Maturity:** Developed ## Definition The **Cognitron** was a self-organizing multilayer neural network built and simulated by [[wiki/Kunihiko Fukushima|Kunihiko Fukushima]] in the 1970s. ## Architecture Repeated exposure changed connection strengths without a teacher assigning desired outputs to individual units. Deeper layers developed broader receptive fields and selective responses to learned patterns. ## Historical Significance The Cognitron was a demonstrated learning system, not a conceptual sketch. It supplied part of the architectural path to the [[wiki/Neocognitron|neocognitron]] and later [[wiki/Convolutional Neural Network|convolutional networks]]. ## Sources / Provenance - Kunihiko Fukushima, “Cognitron: A Self-Organizing Multilayered Neural Network,” 1975.