Symbolic Memory of Motion Patterns using Hierarchical Bifurcations of Attanctors in an Associative Memory Model

Abstract

We describe a memory system of motion patterns which forms symbolic representations and maintains specific-symbolic hierarchy based on stored memories. The system consists of auto-correlation based feature vectors and a nonmonotonic associative memory model. The feature vectors clarify the global structure of motion patterns as a cluster structure. The associative memory model encodes symbols and the hierarchy by bifurcations attractors naturally mirroring the inherent structure of the stored motion patterns.