Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

QI: Fachverband Quanteninformation

QI 3: Quantum Machine Learning

QI 3.7: Vortrag

Montag, 6. März 2023, 12:45–13:00, B305

Quantum Gaussian Processes for Bayesian Optimization — •Frederic Rapp and Marco Roth — Fraunhofer IPA, Stuttgart 70569, Nobelstrasse 12

An important aspect of machine learning is finding the best possible hyperparameters for a given model. Bayesian optimization is one often used algorithm when tackling this task. It requires a surrogate model where Gaussian processes can be used. Gaussian processes are a method based on the evaluation of kernel matrices that serve as covariance functions. These matrices can be evaluated using a quantum computer by encoding the data into the quantum Hilbert space. We study Gaussian processes using quantum kernels based on parameterized quantum circuits, and their application to regression tasks, as well as their usage as a surrogate model for Bayesian optimization. We show that the method can solve a regression of a one-dimensional function under the influence of different quantum computing noise sources. We discuss the important aspects of the model and provide an example of the optimization of the method when solving a multi-dimensional regression task. Finally, we perform a hyperparameter tuning using Bayesian optimization based on quantum Gaussian process regression. We show that the quantum version of the algorithm is able to find suitable hyperparameter settings of a given problem that are comparable to applying the classical counterpart and even better than using a random search based algorithm.