Hebbian covariance synapse: a self-tuning neural device for reinforcement learning

We propose a novel neural computation paradigm inspired by Hebbian covariance adaptation, a preeminent model of learning and memory in the brain. We show that this computational algorithm affords stable self-tuning adaptive optimal control of general nonlinear dynamical systems with unknown disturbances. This biologically-inspired adaptive control paradigm offers a new approach to the modeling of physiological control systems as well as the design of intelligent systems with potential applications in robotics, chemical control and other biomedical and industrial control problems