New publication on "Demonstrating Advantages of Neuromorphic Computation"
Our paper "Demonstrating Advantages of Neuromorphic Computation: A Pilot Study" has been published: we implemented a on-chip closed-loop reinforcement learning experiment utilizing a Reward-modulated Spike-Timing-Dependent Plasticity (R-STDP) rule. The embedded processor simulates the Pong video game, a shallow two-layer neural network receives the current ball position as input and steers the paddle using its "motor-action" neurons. We compare the experiment to a NEST-based simulation, evaluate the influence of noise on parameter space exploration, and the effects of learning on fixed-pattern variation. See https://doi.org/10.3389/fnins.2019.00260 for details.