Optimal Tuning of Quantum Generative Adversarial Networks for Multivariate Distribution Loading

Abstract

Loading data efficiently from classical memories to quantum computers is a key challenge of noisy intermediate-scale quantum computers. Such a problem can be addressed through quantum generative adversarial networks (qGANs), which are noise tolerant and agnostic with respect to data. Tuning a qGAN to balance accuracy and training time is a hard task that becomes paramount when target distributions are multivariate. Thanks to our tuning of the hyper-parameters and of the optimizer, the training of qGAN reduces, on average, the Kolmogorov–Smirnov statistic of 43–64% with respect to the state of the art. The ability to reach optima is non-trivially affected by the starting point of the search algorithm. A gap arises between the optimal and sub-optimal training accuracy. We also point out that the simultaneous perturbation stochastic approximation (SPSA) optimizer does not achieve the same accuracy as the Adam optimizer in our conditions, thus calling for new advancements to support the scaling capability of qGANs.