Photonic architecture for reinforcement learning

Reinforcement learning algorithms are a powerful tool to manage the interaction between a system and its environment. Here we present an approach to apply these algorithms within modern-day photonic technologies. Numerical tests, performed on typical learning tasks, show that the architecture is robust against experimental noise, which can even be beneficial for the learning process. The proposed architecture, based on single-photon evolution on a tree structure of tunable beamsplitters, is simple, easy to implement and an integration in quantum optics applications appears to be within the reach of near-term technology.