A hybrid classical-quantum approach for multi-class classification

• Published in: Quantum information processing Vol. 20; no. 3
• Main Authors: Chalumuri, Avinash, Kune, Raghavendra, Manoj, B. S.
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.03.2021; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Banknotes; Circuits; Classification; Classifiers; Computers; Data Structures and Information Theory; Datasets; Machine learning; Mathematical Physics; Mechanical properties; Model accuracy; Physics; Physics and Astronomy; Quantum computers; Quantum Computing; Quantum entanglement; Quantum Information Technology; Quantum mechanics; Quantum Physics; Qubits (quantum computing); Simulation; Spintronics; Variational quantum circuit; Classification; Quantum machine learning
• ISSN: 1570-0755; 1573-1332
• DOI: 10.1007/s11128-021-03029-9

Quantum machine learning recently gained prominence due to the computational ability of quantum computers in solving machine learning problems that are intractable on a classical computer. However, achieving a quantum advantage on present-day quantum computers remains an open challenge. In this work, we primarily focus on solving machine learning problems using a hybrid model based on both quantum and classical computers together for the classification task. We propose the quantum multi-class classifier (QMCC) as a variational circuit with a hybrid classical-quantum approach using quantum mechanical properties such as superposition and entanglement. A unitary operation on a single qubit for the state preparation is designed and also demonstrated using a real quantum computer on the IBMQX platform. The entire variational circuit for the classification task is implemented on a quantum simulator. We performed our quantum simulations on three benchmark datasets: Iris dataset , Banknote Authentication (BNA) dataset , and Wireless Indoor Localization (WIL) dataset for machine learning algorithms. Our simulation results show that the proposed QMCC model classified Iris dataset with an accuracy of 92.10%, BNA dataset with an accuracy of 89.50%, and WIL dataset with an accuracy of 91.73%. The proposed model can also be extended to multiple class classifiers.