Speaker
Description
In this work we propose an algorithm for a QBM using DualQITE for the Gibbs state preparation. The choice of the Hamiltonian, which defines model connectivity, and the ansatz, used for preparation of the Gibbs state, plays a crucial role in performance of QBMs. With an application to classification tasks, we conduct a study on semi-restricted and fully-connected QBMs to explore the influence of the Hamiltonian choice. We consider synthetic datasets which are hard to tackle with classical machine learning techniques. As the first step, we implemented a 3- and 4-qubit QBMs to validate the proposed algorithm. We use different data-encodings, including symmetry-inspired ones, to use the efficient number of tunable parameters depending on the properties of the dataset. For the considered small benchmark models the performance of QBM reaches more than 95\% accuracy with our approach. As the next step, we are working on scaling up the system size and application to the realistic datasets.
