Confident Region (CoRe) Approach for Variational Quantum Eigensolvers

by Shinichi Nakajima (Technische Universität Berlin)

Tuesday 13 Aug 2024, 13:00 → 14:00 Europe/Berlin

This talk introduces confident region (CoRe), our recently proposed notion that specifies the area in which a machine learning model has sufficient knowledge to accurately predict the function values.  With CoRe, we can complement the existing Bayesian optimization approaches which do not consider how accurately we would know about the function values after the next observations, leading to a novel acquisition function, expected maximum improvement over CoRe (EMICoRe).  We show advantages of our approach in variational quantum eigensolvers, where CoRe expands non-trivially when a proper kernel is used, and discuss possible other applications.