ABSTRACT
A model of an oscillatory associative memory is described which is an ideal candidate for implementation as a self-organising analogue information-processing system. The equations that represent a fixed point associative memory are modified to introduce oscillations and create a system that is capable of reproducing the input segmentation behaviour exhibited by similar oscillatory systems. Within this system phase locked oscillations are employed to represent correlations that can be reinforced via a Hebbian learning rule. Consideration of small systems demonstrates that these phase locked oscillations are based on cooperative interactions between nodes. The system’s ability to segment a mixed input into its constituent components is enhanced by variations between nodes, which will be unavoidable in any analogue system.
