1999 Volume 12 Issue 10 Pages 604-613
The dynamic processes of chaotic neural networks are investigated in the reinforcement adaptation scheme as a simple model of brain-like function. There appears spontaneous transition from a learning phase with slow adjustment to a retrieving phase with fast adjustment under switching inputs. Learning and retrieving can be considered as two aspects of the same dynamical process. In a reinforcement scenario, it is important that there should be some source of fluctuation in the network, so that the space of possible outputs can be explored until a correct output is found. This is well done by using chaotic neurons, not stochastic units.