@phdthesis{oai:oist.repo.nii.ac.jp:00002911,
 author = {Metz, Friederike},
 month = {2023-03-23, 2023-03-23},
 note = {In this thesis, I consider the three main paradigms of machine learning – supervised, unsupervised, and reinforcement learning – and explore how each can be employed as a tool to study or control quantum systems. To this end, I adopt classical machine learning methods, but also illustrate how present-day quantum devices and concepts from condensed matter physics can be harnessed to adapt the machine learning models to the physical system being studied. In the first project, I use supervised learning techniques from classical object detection to locate quantum vortices in rotating BoseEinstein condensates. The machine learning model achieves high accuracies even in the presence of noise, which makes it especially suitable for experimental settings. I then move on to the field of unsupervised learning and introduce a quantum anomaly detection framework based on parameterized quantum circuits to map out phase diagrams of quantum many-body systems. The proposed algorithm allows quantum systems to be directly analyzed on a quantum computer without any prior knowledge about its phases. Lastly, I consider two reinforcement learning applications for quantum control. In the first example, I use Q-learning to maximize the entanglement in discrete-time quantum walks. In the final study, I introduce a novel approach for controlling quantum many-body systems by leveraging matrix product states as a trainable machine learning ansatz for the reinforcement learning agent. This framework enables us to reach far larger system sizes than conventional neural network-based approaches.},
 school = {Okinawa Institute of Science and Technology Graduate University},
 title = {量子系とその制御における機械学習の応用},
 year = {}
}