A Distributed Learning Scheme for Variational Quantum Algorithms

Variational quantum algorithms (VQAs) are prime contenders to gain computational advantages over classical algorithms using near-term quantum machines. As such, many endeavors have been made to accelerate the optimization of modern VQAs in past years. To further improve the capability of VQAs, here, we propose a quantum distributed optimization scheme (dubbed as QUDIO), whose back ends support both real quantum devices and various quantum simulators. Unlike traditional VQAs subsuming a single quantum chip or simulator, QUDIO collaborates with multiple quantum machines or simulators to complete learning tasks. In doing so, the required wall-clock time for optimization can be continuously reduced by increasing the accessible computational resources when ignoring the communication and synchronization time. Moreover, through the lens of optimization theory, we unveil the potential factors that could affect the convergence of QUDIO. In addition, we systematically understand the ability of QUDIO to reduce wall-clock time via two standard benchmarks, which are hand-written image classification and the ground energy estimation of the dihydrogen. Our proposal facilitates the development of advanced VQAs to narrow the gap between the state of the art and applications with the quantum advantage.