A Thermodynamic Model of the Action-Perception Cycle in Brain Dynamics

abstract

Computational models of brain dynamics fall short of performance in speed and robustness of pattern recognition, especially in detecting minute but highly significant pattern fragments. A novel model uses the properties of thermodynamic systems operating far from equilibrium. Such systems construct order by dissipating energy. Conditioned stimuli in reinforcement learning establish categories of patterns in cortical connectivity that create phase domains. Retrieval of a selected category of stored knowledge is by phase transition that is induced by a conditioned stimulus. The key property is cortical background noise, which is simulated by band pass filtering brown noise (1/f²) in the beta (12–30 Hz) or gamma (30–80 Hz) ranges. The noise displays aperiodic null spikes at which analytic amplitude approaches zero. These events resemble vortices. Phase transitions in recall occur at null spikes owing to high signal/noise ratio in the presence of even very weak activity evoked by conditioned stimuli.