Quantum Machine Learning – successful implementation of generative models on Quantum Processors (trapped Ion Processor)
By way of background, in May 2019, in an article published in NPJ Quantum Information, we presented a framework for data-driven quantum circuit learning. We designed a generative machine learning model – The Quantum Circuit Born Machine (QCBM) that encodes a target probability distribution in the wave function of a set of qubits. We successfully trained the model with a simulator, on the canonical Bars-and-Stripes data set, and for Boltzmann distributions. As an application, we use the generative performance of the model to craft a benchmark for hybrid quantum-classical computers, the qBAS score, which we demonstrate on a trapped-ion quantum processing unit (QPU).
In a companion article, now published in Science Advances on 18 Oct 2019, we perform the full training procedure of the QCBM on a hybrid computer, with a trapped-ion QPU. We compare Bayesian and particle swarm optimizations for training. Bayesian optimization outperforms particle swarm in our setting, although the large number of parameters challenges both optimizers. We show that the ideal simulated system is not faster than the experimental system, indicating that the training bottleneck is the classical optimizer.