A hysteresis micro ensemble as a basic element of an adaptive neural net

The paper offers a model of an ensemble of strongly connected neurons. As compared with a single neuron, such an aggregation is shown to reveal new properties: it has three stable states (inactive, semi-active, and active states) and demonstrates the hysteresis effect. These properties may enable the use of time inter-image correlations, competitive learning, image selection in recognition and higher noise robustness. These possibilities are regarded from the viewpoint of using the neural ensemble as a basic element of an adaptive neural net.