Quantum-Inspired Reinforcement Learning in Neuromorphic Systems

This chapter explores how quantum-inspired reinforcement learning (QIRL) can be combined with neuromorphic computing to mitigate the major constraints of other types of reinforcement learning, such as energy consumption, latency and scaling. Annealing-based heuristic and probabilistic sampling are quantum-inspired approaches to optimization and exploration, and are considered in the framework of spike-based and event-driven neuromorphic systems. It gives a systematic background of reinforcement learning, neuromorphic architectures and quantum-inspired computation and next coordinates systematic mapping of the components of reinforcement learning on spiking neural substrates using biologically plausible learning rules. It introduces a hybrid neuromorphically modeled QIRL using a spiking actor-critic model and reservoir computing and is tested. Comparative analysis shows enhanced convergence of rewards with respect to sample and energy efficiency in comparison with the classical and quantum-inspired methods that are realized on conventional hardware.