Kunihiko Fukushima, then at NHK Broadcasting Science Research Laboratories in Tokyo, published “Neocognitron: A Self-organizing Neural Network Model for a Mechanism of Pattern Recognition Unaffected by Shift in Position” in the journal Biological Cybernetics, volume 36, pages 193 to 202, in 1980. Only the year is firmly documented for the precise date.
The paper proposed a network organized into layers of simple “S-cells” and complex “C-cells,” inspired by the visual cortex models of Hubel and Wiesel. The key property was that the network could recognize patterns based on their shape regardless of where they appeared in the input, learning “without a teacher” through repeated exposure to example patterns.
This shift-invariant, hierarchical design is widely seen as the direct conceptual ancestor of the convolutional neural network. The architectural ideas in the Neocognitron flowed into Yann LeCun’s later work on handwritten digit recognition and ultimately into the convolutional networks that power modern computer vision.