FET: Small: Enhancing Machine Learning via Quantum Effects

Modern machine learning has made a great impact in our society, but it faces challenges such as the requirement for large amounts of computational cost and data. This project explores how the strange features of quantum physics called quantum correlations, can be used to design more efficient machine learning models based on quantum computing. The research aims to build quantum machine learning models enhanced by quantum correlations with less cost and complexity compared to classical models, including those widely used in industry. Meanwhile, the project will show that the enhancement is not only theoretical but also useful for real-world problems by establishing a foundation for why learning human-generated data like natural language can benefit from quantum correlations. Ultimately, the project aims to adapt these models to near-term quantum experiments, paving the way toward building practical quantum machine learning systems. In addition, the research will be integrated with the education and training of both graduate and undergraduate students, along with outreach activities connected to the quantum industry. This research investigates how to build potentially practical quantum machine learning systems through the following three steps. First, the investigator explores the connection between quantum contextuality (a typical form of quantum correlation) and tools from optimization theory, such as the Sum-of-Squares hierarchy, to demonstrate the enhanced expressive power of quantum models compared to classical ones and to pinpoint the origin of this advantage as quantum contextuality. Second, based on a Bayesian interpretation of these tools and supported by experimental results from cognitive science, the investigator aims to explain why quantum contextuality is useful for capturing certain correlations in human-generated data, making the first step not only theoretical but also practically relevant. Third, the investigator will design quantum machine learning models for real-world problems that can be naturally implemented in near-term quantum experiments, on either analog or digital quantum devices. This is possible because quantum contextuality is both commonly present in quantum devices and useful for machine learning as shown in the first two steps. The resulting models will be tested through numerical simulations and collaborations with experimental groups, laying the groundwork for practical quantum advantage in machine learning tasks. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.