Temporal infomax leads to almost deterministic dynamical systems

The well-known Kullback-Leibler divergence of a random field from its factorization quantifies spatial interdependences of the corresponding stochastic elements. We introduce a generalized measure called 'stochastic interaction' that captures also temporal interdependences. Maximization of stochastic interaction in the setting of Markov chains is shown analytically and by simulations to result in an almost deterministic global dynamics, but almost unpredictable single Unit activity. (C) 2003 Elsevier Science B.V. All rights reserved.