Reinforcement learning in a spiking neural network with memristive plasticity

The reinforcement learning paradigm is for the first time presented for spiking neural network architecture with memristor-based local dynamic plasticity. The models of two kinds of such plasticity are used in the simulation study of the Cartpole task. Applying the Gaussian receptive field time-encoding scheme and simple reinforcing current pulses determined by the sign of reward change, the successful learning is demonstrated for both types of memristive plasticity.