Classifying 2D Input with Quantum Reservoir Computing

Abstract

Though promising, current quantum computing is still nascent and in development. The processors available to us are noisy and quite small in scale, making it impossible to test any of the famous quantum algorithms that have been developed and to demonstrate any significant quantum speedups compared to classical computing. However, quantum machine learning has emerged as a potential use for the current quantum processors at our disposal. Recently, a technique known as quantum reservoir computing has emerged as a particularly interesting method of implementing machine learning algorithms due to its relatively low resource costs and efficient training algorithms compared to classical neural networks. In this thesis, we describe our current understanding of the quantum reservoir computing framework before demonstrating it in action with a simple binary classification problem for a 2D input referred to as the noisy spiral dataset. We show how to train a model using a quantum circuit reservoir by deriving its features. We also show numerically that increasing the number of layers in the circuit reservoir expands the frequency spectrum available to the quantum model, making it better at learning more complex functions and bettering our understanding of the quantum reservoir computing framework.