Quantum machine learning is a popular topic these days, but its near-term applications for practical data science problems are unclear. At the same time, one of the fundamental problems of data science is developing new models, especially for data whose structure is nontrivial. Parametrised quantum circuits can provide a way to enhance models. For example, they can be used to define a similarity measure between two pieces of classical data, yielding a kernel matrix that can be plugged into a support vector machine or a kernel ridge regression algorithm. Another approach is to use parametrised quantum circuits as part of a neural network architecture. Both of these approaches let classical models work implicitly in quantum state space and enable data scientists to potentially gain some benefit from near-term quantum computers. What’s more, the integration of a parametrised quantum circuit into a data science workflow is straightforward, especially for kernel-based algorithms. In this talk, we will discuss the use of parametrised quantum circuits as part of both kernel algorithms and generative algorithms, show simple code examples for doing so using an applications development framework (Qiskit Machine Learning) and discuss the implications for developing new models.
Christa Zoufal is a Ph.D. candidate working in the Quantum Technology group of the Science & Technology department at IBM Research-Zurich. She received a B.Sc. as well as an M.Sc. degree in physics from ETH Zurich. During her studies, she focused on quantum information theory and computational methods. Her master thesis on “The quantum SWITCH and its causal simulations” focused on an indefinite causal quantum structure which leads to advantages in several information processing tasks. As part of the Quantum Finance & Optimization group her current research focuses on the exploitation of quantum information processing within the context of machine learning and optimization problems.
Join Zoom Meeting
Meeting ID: 929 1692 3767