Collective Computation With Continuous Variables

Abstract

The model on which collective computation in neural networks was originally shown to produce a content addressable memory was based on neurons with two discrete states (1) (or levels of activity) which might be thought of as "0" and "1". That model involved an asynchronous updating of the state of each neuron, and could be closely related to aspects of magnetic Ising systems. The role of exchange in the magnetic system is played by the connections between the neurons in the biological system.

Additional details