Quantum Optical Projective Simulation for Reinforcement Learning

Variational quantum algorithms show significant potential in effectively operating on noisy intermediate-scale quantum (NISQ) devices. Using classical optimizers and parameterized quantum circuits, these hybrid quantum-classical algorithms can adapt to and mitigate noise-induced errors and imperfections in quantum hardware. A novel variational approach to reinforcement learning has been recently proposed, incorporating linear-optical interferometers and a classical learning model known as projective simulation (PS). PS is a decision-making tool for reinforcement learning and can be classically represented as a random walk on a graph that describes the agent’s memory. In its optical quantum version, this approach utilizes quantum walks of single photons on a mesh of tunable Mach-Zehnder interferometers to select actions. In this thesis, we physically implement this algorithm on a photonic chip provided by Quandela, using Perceval, a framework to simulate and interface with discrete-variable photonic quantum computers. Finally, we evaluate the algorithm on a use case of industrial interest, taking the first steps towards the application of interpretable quantum learning agents.