Sitzungen | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

FM: Fall Meeting

FM 23: Quantum & Information Science: Neural Networks, Machine Learning, and Artificial Intelligence I

FM 23.5: Talk

Montag, 23. September 2019, 18:00–18:15, 3043

Photonic architecture for reinforcement learning — •Fulvio Flamini, Arne Hamann, Sofiène Jerbi, Lea M. Trenkwalder, Hendrik Poulsen Nautrup, and Hans J. Briegel — Institut für Theoretische Physik, Universität Innsbruck, Technikerstraße 25, 6020 Innsbruck, Austria

Artificial intelligence and photonic technologies are driving the limits of present computing devices. Motivated by the recent success in both fields, this work will bring together their state of the art within the framework of reinforcement learning (RL). Specifically, we present the blueprint for a photonic implementation of an active learning agent that can accommodate well-established RL algorithms, such as SARSA, Q-learning, and projective simulation. We numerically investigate its performance within typical task environments, demonstrating that the approach is effective at solving standard RL problems. The simulation is carried out considering imperfect experimental implementations, where we observe that realistic levels of noise can be tolerated or even be beneficial for the learning process. The proposed architecture, based on single-photon evolution on a mesh of tunable beamsplitters, is simple, scalable, and a first integration in portable systems appears to be within the reach of near-term technology.