Institutional Repository
Technical University of Crete
Technical University of Crete
EN
|
EL
| URI: | http://purl.tuc.gr/dl/dias/B8CC4085-C8DC-4308-B12D-EF32A61FA4AD | ||
| Year | 2023 | ||
| Type of Item | Diploma Work | ||
| License |
|
||
| Bibliographic Citation | Dimitrios Komninos, "Quantum computing for generative modeling and applications", Diploma Work, School of Electrical and Computer Engineering, Technical University of Crete, Chania, Greece, 2023 https://doi.org/10.26233/heallink.tuc.98643 | ||
| Appears in Collections |
This thesis dives into the intersection of quantum computing and generative modeling by exploring their relationship and potential applications across various domains, with a primary focus on finance. The journey begins with a comprehensive analysis of the mathematical framework behind quantum mechanics. Then, classical generative modeling techniques are presented, specifically restricted Boltzmann machines (RBMs) for data reconstruction and denoising and generative adversarial networks (GANs) for the generation of synthetic data, using the popular MNIST dataset as a benchmark. Building on this foundational knowledge, we transition into the realm of quantum machine learning. The struggles of implementing a fault-tolerant quantum computer for learning tasks is presented and how we can approach such pieces of work through different angles with currently available technology. We introduce parameterized quantum circuits (PQCs) and quantum circuit Born machines (QCBMs), two essential components of quantum computing for generative modeling and machine learning tasks in general. A key highlight of this section is the training of various topologies of Born machines on a simple dataset, showcasing the ability to effectively learn the underlying data distribution through quantum processes. We then discuss how the above classical and quantum approaches can be used in the financial sector. Leveraging the power of generative modeling, a Wasserstein GAN with gradient penalty is employed to generate realistic financial time series data, using the S&P 500 index closing values as a benchmark. This marks a critical step towards synthesizing financial data for various analytical and predictive purposes. At last, we introduce and study a quantum Wasserstein GAN (QWGAN) in the financial domain. Here, the traditional WGAN generator is replaced by a parameterized quantum circuit featuring diverse architectures. This novel approach not only has the potential to enrich the generative capabilities, but also harnesses the inherent quantum advantages for more, possibly, efficient and accurate data generation.
Copyright © DIAS 2013
