Hebbian learning in the agglomeration of conducting particles

The Hebbian learning rule is a fundamental concept in the learning of a neu ronal net, when a frequently used connection of two neurons is continually reinforced. We study the properties of self-assembling connections of condu cting particles in a dielectric liquid, and find that the strength of the c onnection between different electrodes represents a memory for the history of the system. Optimal parameters and sequences of stimulation for effectiv e training are determined. We discuss a future application of our results f or the implementation of a nonvolatile neuronal network based on self-assem bling nanowires on a semiconductor surface. [S1063-651X(99)03603-X].