This is the code that led to the results presented in "Real-time calibration of coherent-state receivers: learning by trial and error". Marek's objective is to learn the optimal discrimination strategy over an unknown quantum-classical channel; we frame this as a reinforcement learning problem.
The goal is to calibrate the following receiver:
departing from complete ignorance of any experimental details. As explained in the paper, the model-free learning of such a receiver allows optimal success rate over noisy channels, in which dark counts, phase flips or energy shifts may occur.
For instance, this kind of learning curves are obtained:
