Freiburg 2019 – scientific programme

Sessions | Days | Selection | Search | Updates | Downloads | Help

FM: Fall Meeting

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

FM 23.4: Talk

Monday, September 23, 2019, 17:45–18:00, 3043

Improving the dynamics of quantum sensors with reinforcement learning — Jonas Schuff, •Lukas Fiderer, and Daniel Braun — Eberhard-Karls-University Tuebingen

Quantum sensors so far have been based almost exclusively on integrable systems, such as precessing spins or harmonic oscillators (e.g., modes of an electro-magnetic field). Non-classical initial states promise large enhancements in measurement precision but are experimentally very difficult to prepare and protect against decoherence.

We recently proposed a new approach that achieves quantum enhancements by rendering the dynamics of the quantum sensor chaotic while using classical initial states that are easy to prepare. Starting from an integrable sensor, the dynamics can be rendered chaotic by applying nonlinear kicks during the parameter-encoding transformation. In this work we deal with the following question: Given the possibility of applying non-linear kicks, what the best strategy to choose the position and strength of these kicks?

This a difficult optimization problem which we tackle with reinforcement learning. As a reward for the learning agent we calculate the quantum Fisher information. At the example of a spin subjected to superradiant damping, we demonstrate how the agent is able to find new strategies. Most strikingly, it is able to adopt to the superradiance decoherence model: quantum Fisher information can be increased further even when it would decay to zero for sensor dynamics without kicks.