Matterwave interferometry in a shaken optical lattice designed by reinforcement learning

We design an interferometer to measure acceleration in one dimension with high precision using ultracold atoms moving in an optical lattice. We utilize a branch of machine learning, reinforcement learning, to generate the shaking protocols needed to realize lattice-based analogs of elementary optical components, including a beam-splitter, a mirror, and a recombiner. The performance of these protocols is determined through fidelity measures that compare with ideal optical components. The interferometer's ability to measure acceleration is quantitatively evaluated using a Bayesian approach applied to measurements of the momentum distribution, and comparison is made with standard Bragg interferometers, demonstrating the potential for the application of reinforcement learning algorithms to these kinds of quantum sensing tasks.