Hannover 2020 – wissenschaftliches Programm
Die DPG-Frühjahrstagung in Hannover musste abgesagt werden! Lesen Sie mehr ...
Q 33.3: Vortrag
Mittwoch, 11. März 2020, 14:45–15:00, e001
Neural Network Heuristics for Adaptive Bayesian Quantum Estimation — Lukas J. Fiderer1, Jonas Schuff1,2, and •Daniel Braun1 — 1Institute for Theoretical Physics, University of Tübingen, Tübingen, Germany — 2Department of Materials, University of Oxford, Oxford, United Kingdom
Adaptive experiment design is crucial in order to exploit the benefits of Bayesian quantum estimation. We propose and demonstrate a general method for creating fast and strong experiment design heuristics based on neural networks. Training of the neural networks relies on a combination of imitation and reinforcement learning. Based on the well-studied example of frequency estimation with a qubit which suffers from T2 relaxation, we demonstrate that neural networks trained with reinforcement learning are tailored to the properties of the estimation problem and take into account the availability of resources such as time or the number of measurements. The simultaneous estimation of the frequency and the relaxation rate is considered as well. We find that the neural network heuristics are able to outperform well-established heuristics in all examples.