Dynamical encoding by networks of competing neuron groups: Winnerless competition - art. no. 068102

Following studies of olfactory processing in insects and fish, we investiga te neural networks whose dynamics in phase space is represented by orbits n ear the heteroclinic connections between saddle regions (fixed points or li mit cycles). These networks encode input information as trajectories along the heteroclinic connections. If there are N neurons in the network, the ca pacity is approximately e(N - 1)!, i.e., much larger than that of most trad itional network structures. We show that a small winnerless competition net work composed of FitzHugh-Nagumo spiking neurons efficiently transforms inp ut information into a spatiotemporal output.