DETERMINING THE INFORMATION-FLOW OF DYNAMICAL-SYSTEMS FROM CONTINUOUSPROBABILITY-DISTRIBUTIONS

We present a technique to estimate the information flow in dynamical s ystems. The information Bow characterizes the underlying dynamics whic h can be stochastic, deterministic chaotic, or deterministic nonchaoti c. For chaos our approach offers a method for calculating the Kolmogor ov-Sinai (KS) entropy from continuous probability density functions. T hese density functions are approximated by neural networks. For stocha stic processes (noisy chaos, autoregressive models) our approach yield s the rate of memory loss of the dynamics.