Loading...
Quantum generative learning with trapped-ion quantum simulators
Stringer-Usdan, Aaron
Stringer-Usdan, Aaron
Citations
Altmetric:
Advisor
Editor
Date
Date Issued
2024
Date Submitted
Keywords
Collections
Research Projects
Organizational Units
Journal Issue
Embargo Expires
Abstract
Generative artificial intelligence has become an important area of research in the broader field of artificial intelligence and machine learning. Generative models have been applied to other areas of research, such as computer vision, natural language processing, and drug discovery. They have also seen a great deal of success in commercial applications derived from such research, with chat bots and AI art programs attracting widespread public attention. For all the success that generative AI has already achieved, the field is still relatively new, and there are many unsolved problems and avenues for improvement on existing techniques. One such avenue is quantum generative learning, the intersection between classical generative techniques and quantum computing. In this thesis, I explore several quantum generative models, with practical implementation on trapped-ion quantum simulators.
In Chapter 2, I examine a popular gate-based quantum generative model known as the quantum circuit Born machine (QCBM). I extend this model to a model involving continuous time evolution, which I call the quantum evolution Born machine (QEBM), and I compare the performance of the two models on several generative learning tasks. I find that the QEBM exhibits enhanced performance when all pairs of qubits are allowed to interact, compared to interactions between nearest neighbors only, which recommends trapped ion quantum simulators as a platform for its implementation. Additionally, I find that on the learning tasks under consideration, the QEBM outperforms two standard formulations of QCBM, except when certain modifications are made to the QCBM so that it is as similar as possible to the QEBM.
In Chapter 3, I examine a hybrid quantum/classical generative model, which combines the QCBM with the classical generative adversarial network (GAN), which has previously been implemented on a trapped ion quantum computer at a small scale. I investigate the performance of this hybrid technique when the QCBM is replaced with the aforementioned QEBM. I find some indications that the QEBM may improve performance over GAN alone, but not to the extent reported in a previous publication using QCBM. However, further investigation is needed to reveal any quantum advantage of this hybrid model.
In Chapter 4, I study a recently-introduced generative learning model called the quantum neuron Born machine (QNBM), which has also been implemented very recently on a trapped ion quantum computer. I am able to reproduce existing results based on this model. However, I identify a potential major challenge in training the model. One has to either post select an exponentially small set of measurement data or use a cost function that is intrinsically random and may contain large statistical fluctuation. I conclude that future work is needed to address this challenge before claiming any quantum advantage of the QNBM over classical models.
Associated Publications
Rights
Copyright of the original work is retained by the author.