Erlangen 2018 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

Q: Fachverband Quantenoptik und Photonik

Q 16: Quantum Information and Simulation

Q 16.1: Talk

Monday, March 5, 2018, 14:00–14:15, K 1.020

Improving the consistency of a quantum experiment with reinforcement learning — •Sabine Wölk and Hans Jürgen Briegel — Institute for Theoretical Physics, University of Innsbruck, 6020 Innsbruck, Austria

In quantum experiments, expectation values of observables are determined by repeating their measurement many times. Meaningful results can only be obtained if the conditions under which the experiment is performed can be kept constant for all the measurements. For setups with unstable conditions, e.g. frequency drift, this may require calibration measurements during data acquisition, which however increases the amount of resources, e.g., time, number of qubits, or general equipment. The problem of finding an optimal calibration strategy is in general highly non-trivial since the only available information is probabilistic.

We show that a learning agent using projective simulation [1] is able to find good solutions based solely on the experimental data. In this way, we also demonstrate that projective simulation is not limited to deterministic rewards but can also learn from probabilistic ones.

[1] H. J. Briegel and G. De las Cuevas, Sci. Rep. 2, 400 (2012)