A recognition with a large-scale network is simulated on PDP-11/34 minicomputer and shown to have great capability for visual pattern recognition. The model consists of nine layers cells. authors demonstrate that the can be trained recognize handwritten Arabic numerals even considerable deformations in shape. learning-with-a-teacher process used reinforcement modifiable synapses new model, instead learning-without-a-teacher applied previous model. focus mechanism rather than self-organization.

A model of a neural network is presented that offers insight into the brain's complex mechanisms as well design principles for information processors. The has properties and abilities most modern computers pattern recognizers do not possess; recognition, selective attention, segmentation, associative recall. When composite stimulus consisting two or more patterns presented, pays attention to each one after other, segments from rest, recognizes it separately in contrast earlier models. This...

A pattern recognition system which works with the mechanism of neocognitron, a neural network model for deformation-invariant visual recognition, is discussed. The neocognition was developed by Fukushima (1980). has been trained to recognize 35 handwritten alphanumeric characters. ability deformed characters correctly depends strongly on choice training set. Some techniques selecting patterns useful large number are suggested.

A new type of visual feature extracting network has been synthesized, and the response simulated on a digital computer. This research done as first step towards realization recognizer handwritten characters. The design was suggested by biological systems, especially, systems cat monkey. is composed analog threshold elements connected in layers. Each element receives inputs from large number neighbouring layers performs its own special functions. It takes care one restricted part...

A neural network model of selective attention is discussed. When two patterns or more are presented simultaneously, the successively pays to each one, segmenting it from rest and recognizing separately. In presence noise defects, can recall complete pattern in which has been eliminated defects corrected. These operations be successfully performed regardless deformation input patterns. This an improved version earlier proposed by author: ability segmentation lateral inhibition.

An electronic model of the retina has been developed with a new technique in making interconnections between cells. The consists about 700 photoreceptors and same number output cells have concentric on-center type receptive fields show Mach-band Broca-Sulzer effects.

Deep convolutional neural networks (deep CNNs) show a large power for robust recognition of visual patterns. The neocognitron, which was first proposed by Fukushima (1979), is recognized as the origin deep CNNs. Its architecture suggested neurophysiological findings on systems mammals. It acquires ability to recognize patterns robustly through learning. Although neocognitron has long history, improvements network are still continuing. This article discusses recent focusing differences from...