RPE-BAM networks: Bidirectional heteroassociation in neural networks of recursive nodes with rich dynamics

This paper addresses networks of recursive processing elements (RPEs) that exhibit, even at the single node level, rich dynamics and switching between ordered, erratic (chaotic) and diverse periodic trajectories. These networks are considered here for the implementation of bidirectional heteroassociation. These newly proposed architectures are named RPE-BAM. Dynamic mixture of erratic and ordered dynamics is explored, during the episodes of: a) input prompting; b) search for the embedded heteroassociations compatible with the input pattern; c) the production of an heteroassociation pair. Concepts and design methods on RPE-BAMs and on parametric coupling of recursive nodes are discussed, and numerical experiments are analyzed, showing robust operation of the RPE-BAM architecture.