Implementing Quantum-Classical Machine Learning Architectures to Optimize Convolutional Neural Networks

Overview

Larger sets of data and information in computational systems are oftentimes misinterpreted or tedious to analyze because of human error. Classical machine learning (ML) can be sufficient in categorizing data, but as dimensionality and features increase, it becomes more difficult to find patterns and relationships in the data. By implementing quantum computing, the performance of ML can improve performance, scalability, and enable the solution of problems that are currently intractable for classical computers. This research studies the effects of quantum-classical machine learning architectures on convolutional neural networks (CNNs). I created a total of 4 different Quantum-Classical CNNs. The first consisted of a parametrized quantum circuit that worked via the implementation of a single qubit y-rotational value. The second worked via a four-qubit y-rotational value, the third through entanglement, and the fourth via the quantum approximation optimization algorithm. Each circuit was run 3 times under 5,10, 15, and 20 training iterations (epochs) and with 100, 150, and 200 samples each to find the average result in accuracy in data predictions. 93 The accuracy of test data performance improved as each algorithm had an increasing amount of training iterations as well as when there were more data samples to learn from. Not only did accuracy improve, but the behavior of each model improved with the implementation of quantum protocols by quickly processing the data and having high accuracy in predictions. Overall, by combining classical and quantum computing in machine learning, quantum algorithms have the ability to perform faster calculations.