Machine learning methods for robust quantum optimal control

Khalid, Muhammad 2024. Machine learning methods for robust quantum optimal control. PhD Thesis, Cardiff University. Item availability restricted.

Preview	PDF (Muhammed Irtaza Khalid, PhD, Thesis) - Accepted Post-Print Version Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (11MB) | Preview
	PDF (Cardiff University Electronic Thesis and Dissertation Form) - Supplemental Material Restricted to Repository staff only Download (140kB)

Abstract

Quantum technologies have the potential to revolutionize many classical tasks, particularly including sensing and simulation applications. Yet their full potential is limited by the presence of noise, amongst other issues. This thesis addresses the problem of quantum optimal control of a controllable system with noisy dynamics and an uncertain theoretical description. Towards this goal, this thesis makes two contributions. Firstly, it develops a novel robustness measure called the Robustness Infidelity Measure (RIM) for certification of robustness of optimal control schemes, agnostic of the acquisition method. The RIM is a statistical measure and it can be used to compare the robustness of different schemes. Secondly, this thesis develops novel optimization techniques based on Reinforcement Learning (RL) for robust optimization of noisy quantum dynamics with model uncertainties. In particular, a model-based RL algorithm is proposed that is able to improve over direct applications of model-free RL algorithms in terms of experimental resource consumption. This is done via incorporation of partial knowledge of the uncertain model whilst the rest is learned using experimental data. Our approach highlights the potential of extending pure model-free methods towards model-based approaches, with a learnable model, for noisy optimization tasks and brings RL algorithms closer to deployment on near-term quantum devices. We evaluate the RIM and various model-free RL algorithms on a number of benchmark problems. Our results show that the RIM is a valuable tool for assessing the robustness of quantum control schemes. Moreover, we demonstrate that RL algorithms are able to generate robust control schemes which outperform schemes generated using other methods. We also show how learned models of noisy quantum dynamics can be leveraged to increase the optimality of quantum control schemes found by RL algorithms whilst retaining their robustness performance.

Item Type:	Thesis (PhD)
Date Type:	Acceptance
Status:	Unpublished
Schools:	Computer Science & Informatics
Uncontrolled Keywords:	Theoretical computer science
Date of First Compliant Deposit:	17 July 2024
Date of Acceptance:	17 July 2024
Last Modified:	17 Jul 2024 15:06
URI:	https://orca.cardiff.ac.uk/id/eprint/170642

Actions (repository staff only)

Edit Item