Machine Learning for Image Classification on a Trapped Ion Quantum Computer

Machine learning has recently emerged as one of the most promising areas for applications of near-term quantum computers. The ability of quantum computers to apply nonclassical feature maps to classical data opens new possibilities for enhancing part or all of existing ML models. In supervised learning, image classification has stood out as one of the benchmark problems, and has been one of the most popular areas for testing QML algorithms. To date, most of the work in this direction has focused on classification of a subset of the relatively simple MNIST images. Furthermore, there has been a relative scarcity of implementations of these algorithms on quantum hardware. In this work, we report results of applying a hybrid quantum convolutional neural network model for classification of a complex real world dataset of road sign images. We will discuss our methods of classical processing and quantum encoding of the data, as well as the quantum circuit design and optimization methods. We will present results of training a full end-to-end classification model on up to 8 qubits on a trapped ion quantum computer as well as performing inference on the QPU backend with 16 qubits to classify up to 10 categories of images. We will also discuss projections of future performance of this model.