Performing Neuro-Like Computation Through Attractor Networks with Nodes with Rich Dynamics

This paper discusses concepts, methodologies and characterization experiments for neuro-like architectures composed of recursive nodes with diverse dynamical behavior, bifurcation and chaotic dynamics. These networks present collective attractors, which are used to represent and store information in associative architectures through the network´s long-term cycles. The nodes of the studied networks are mathematically described through recursive maps, which interact for the formation of collective spatio-temporal patterns through parametric coupling, i.e., through dynamic modulation of the bifurcation parameters. The relationships between strength of coupling between nodes, network size, network stability and network performance (both, in terms of precision on pattern recovery, and in terms of speed of operation), are addressed in the analyses and experiments, allowing thus the understanding of several of the network phenomena as well as the definition of procedures for architectural design aiming improved performance.