Evaluating Accuracy and Adversarial Robustness of Quanvolutional Neural Networks

• Published in: 2021 International Conference on Computational Science and Computational Intelligence (CSCI) pp. 152 - 157
• Main Authors: Sooksatra, Korn, Rivas, Pablo, Orduz, Javier
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.12.2021
• Subjects: Computers; convolutional neural networks; Deep learning; Machine learning algorithms; Neural networks; Quantum computing; quantum neural networks; quanvolutional neural networks; Robustness; Scientific computing
• DOI: 10.1109/CSCI54926.2021.00097

Machine learning can thrust technological advances and benefit different application areas. Further, with the rise of quantum computing, machine learning algorithms have begun to be implemented in a quantum environment; this is now referred to as quantum machine learning. There are several attempts to implement deep learning in quantum computers. Nevertheless, they were not entirely successful. Then, a convolutional neural network (CNN) combined with an additional quanvolutional layer was discovered and called a quanvolutional neural network (QNN). A QNN has shown a higher performance over a classical CNN. As a result, QNNs could achieve better accuracy and loss values than the classical ones and show their robustness against adversarial examples generated from their classical versions. This work aims to evaluate the accuracy, loss values, and adversarial robustness of QNNs compared to CNNs.