Exploration of the attractor space of small networks of reciprocally connected processing elements

The properties of a type of a dynamical network are investigated. Three properties were selected: fast memory access, content addressable access, and large memory capacity. While these properties are useful, in order to bring these networks to practical applications a learning algorithm needs to be developed which will determine the set of network weights for a given signal to be stored. It is noted that, in addition to content-addressable memory, dynamical networks have exciting long-term potential as the technology base of a new approach to developing machine intelligence where cognitive skills are self-organized directly in, and by, the dynamic architecture