We cast the discrimination of two coherent states of light as a reinforcement learning problem, in which an agent has to choose among a large number of configurations of a receiver composed of simple linear optics elements, on/off photodetectors and feedback, all within reach of current technology. The agent, though completely ignorant about the receiver, is asked to find its optimal configuration by repeating the experiment a finite number of times, based only on the information obtained from the photodetectors and on the correctness of its guess. Despite the fact that the quantum signals are not perfectly distinguishable and therefore an optimal configuration may lead to an incorrect guess (no reward), we construct agents that can both discover near-optimal configurations and achieve high real-time success rate, even in the presence of several noise-sources.


Seminar, February 14, 2020, 12:00. ICFO’s Seminar Room

Hosted by Prof. Antonio Acín