Quantum Machine Learning for Recognizing Gaussian Discord

Abstract

The quantum correlations measured by quantum discord have been paying significant attention since they are an essential resource for quantum information. This thesis focuses on the quantum discord in continuous-variable (CV) quantum computing systems, especially the quantum discord of the bipartite Gaussian states. We recognize the Gaussian quantum discord using quantum machine learning. We review in detail the CV quantum computing systems that implement hybrid quantum-classical machine learning algorithms. Both Gaussian and non-Gaussian transformations are necessary to construct universal quantum computing systems. The structure of quantum discord can be studied using Gaussian states. The analytical solutions of Gaussian quantum discord are the labels of Gaussian states data set used for training and evaluating machine learning models. We presented the classical machine learning optimization algorithm back-propagating (BP) of the neural network to realize quantum Gaussian discord. We proposed the supervised hybrid quantum-classical optimization performed on the variational quantum circuits for the Gaussian discord classification tasks. Moreover, we implemented a hybrid quantum-classical machine learning algorithm: Quantum Kitchen Sinks (QKS) for noisy intermediate-scale quantum (NISQ) devices. QKS uses the parametric variational quantum circuits to achieve non-linearly transformation from classical inputs to higher-dimensional feature vectors. Simulating the QKS on classical computers with the help of PennyLane, we demonstrated that the variational quantum circuits provide more excellent performance than the classical linear classification algorithm, successfully improving the classification accuracy from 70.12% up to 98.64%.