Perception and recognition in a neural network theory in which neurons exhibit hysteresis

TL;DR

This paper proposes a neural network theory for visual perception and recognition that involves neurons with hysteresis, allowing rapid recognition without synaptic changes, emphasizing a bidirectional information flow between retina and brain.

Contribution

It introduces a novel neural network model where neurons exhibit hysteresis, explaining rapid recognition and perception without synaptic plasticity.

Findings

Neurons with hysteresis can account for quick recognition.

Bidirectional information flow models perception and memory.

Recognition of faces occurs rapidly without synaptic changes.

Abstract

A neural network theory of visual perception and recognition is presented. Information flows both from the retina to the brain and from the brain to the retina. A report that when a scene is perceived 50 retinal cells are much more active than any of the other retinal cells is ascribed significance in the theory. The theory involves neurons that exhibit hysteresis, without the need for any changes in synaptic connection strengths during learning. The fact that the brain is able to recognize faces and other objects very rapidly is discussed in the context of the theory. The theory can be epitomized as "We see with our eyes and remember with our brains".