Emergent complex patterns in autonomous distributed systems: mechanisms for attention recovery and relation to models of clinical epilepsy

Dynamical systems based on distributed elements can exhibit complex autonomous behavior. Simultaneous existence of separate stable dynamic states (attractors) and the transitions between them can model certain forms of epileptic discharge. Multi-stable systems have also been proposed for storage and retrieval of activation patterns. Here we consider systems with alternative types of collective behavior. In these systems emergent intermittency allows for autonomous switching between turbulent (chaotic) and laminar phases. We demonstrate that the distributions of the duration of various phases have distinctive statistical properties, different from those in multi-stable systems that are driven by stochastic processes. These properties are proposed to identify and classify mechanisms that may underlie paroxysmal activity as revealed in electrophysiological recordings of epileptiform activity. Unlike spontaneous stochastically-driven ictal transitions in multi-stable systems, certain features of intermittency-based transitions can, in principle, be forecasted and perhaps even ameliorated. We show that intermittency in a recurrent network does not require plastic connections. At the same time, we argue that an autonomous system with modifiable connections might require intermittent transition mechanisms in order to sustain proper connectivity and function. Networks showing intermittency avoid lockups and at the same time respond robustly and commensurably to dynamical input perturbation. They may thus provide a candidate mechanism for pattern recognition and attention recovery in biological and artificial systems.