Switching in Self-Controlled Chaotic Neuromodules

The activity dynamics of recurrent neural networks can exhibit deterministic chaos due to the nonlinear transfer functions. Chaotic attractors are winded around infinitely many unstable periodic orbits. Each can be stabilized by a feedback control. In the present article we explore the flexibility of the chaotic dynamics of a recurrent neuromodule and construct a neural controller which is able to switch between several periodic patterns in two different ways: either deterministically by external inputs or spontaneously by dynamic noise. The chaotic attractor acts as an intermediate state between successively stabilized dynamic patterns. It has all the possible motions present, like an attentive state between different stimuli. ∗in: Proceedings of World Congress on Neural Networks, San Diego, California, Sept. 15 18, 1996, INNS Press and Lawrence Erlbaum, New Jersey, 1966, pp. 680 684.