Bayesian Optimization Priors for Efficient Variational Quantum Algorithms

Abstract

Quantum computers currently rely on a hybrid quantum-classical approach known as Variational Quantum Algorithms (VQAs) to solve problems. Still, there are several challenges with VQAs on the classical computing side: it corresponds to a black-box optimization problem that is generally non-convex, the observations from the quantum hardware are noisy, and the quantum computing time is expensive. The first point is inherent to the problem structure; as a result, it requires the classical part of VQAs to be solved using global optimization strategies. However, there is a trade-off between cost and accuracy; typically, quantum computers return a set of bit strings, where each bitstring is referred to as a shot. The probabilistic nature of quantum computing (QC) necessitates many shots to measure the circuit accurately. Since QC time is charged per shot, reducing the number of shots yields cheaper and less accurate observations. Recently, there has been increasing interest in using basic Bayesian optimization (BO) methods to globally optimize quantum circuit parameters. This work proposes two modifications to the basic BO framework to provide a shot-efficient optimization strategy for VQAs. Specifically, we provide a means to place a prior on the periodicity of the rotation angles and a framework to place a topological prior using few-shot quantum circuit observations. We demonstrate the effectiveness of our proposed approach through an ablation study, showing that using both proposed features statistically outperforms a standard BO implementation within VQAs for computational chemistry simulations.