[Paper Review] A Brief Review of Quantum Machine Learning for Financial Services

Mina Doosti, Petros Wallden+4 more|arXiv (Cornell University)|Jul 17, 2024

Blockchain Technology Applications and Security4 citations

TL;DR

This paper reviews quantum machine learning (QML) techniques—particularly Quantum Variational Classifiers, Quantum Kernel Estimation, and Quantum Graph Neural Networks—for financial applications such as credit scoring, risk management, fraud detection, and stock prediction. It evaluates near-term feasibility on NISQ devices and identifies QML as a promising but still emerging field with potential for long-term innovation in finance.

ABSTRACT

This review paper examines state-of-the-art algorithms and techniques in quantum machine learning with potential applications in finance. We discuss QML techniques in supervised learning tasks, such as Quantum Variational Classifiers, Quantum Kernel Estimation, and Quantum Neural Networks (QNNs), along with quantum generative AI techniques like Quantum Transformers and Quantum Graph Neural Networks (QGNNs). The financial applications considered include risk management, credit scoring, fraud detection, and stock price prediction. We also provide an overview of the challenges, potential, and limitations of QML, both in these specific areas and more broadly across the field. We hope that this can serve as a quick guide for data scientists, professionals in the financial sector, and enthusiasts in this area to understand why quantum computing and QML in particular could be interesting to explore in their field of expertise.

Motivation & Objective

• To evaluate the current state of quantum machine learning (QML) techniques applicable to financial services.
• To identify which QML algorithms are most viable for deployment on Noisy Intermediate-Scale Quantum (NISQ) devices in the near term.
• To assess the potential of QML for enhancing financial tasks such as credit scoring, risk management, fraud detection, and stock price prediction.
• To outline the challenges, limitations, and future prospects of QML in the financial sector, including data loading and scalability issues.
• To provide a practical guide for financial data scientists and professionals on the realistic opportunities and risks of adopting QML technologies.

Proposed method

• Focuses on QML techniques for classical data on quantum hardware, particularly hybrid quantum-classical algorithms.
• Reviews Quantum Variational Classifiers (QVC) and Quantum Kernel Estimation as near-term candidates for supervised learning in finance.
• Examines Quantum Neural Networks (QNNs), Quantum Transformers, and Quantum Graph Neural Networks (QGNNs) for generative and graph-based financial modeling.
• Analyzes the egograph-based QGNN (egoQGNN) architecture, which uses hierarchical processing and k-hop neighborhood decomposition for improved graph classification.
• Compares QML performance with classical counterparts through simulations and theoretical frameworks, emphasizing data mapping and classification capability.
• Uses insights from classical machine learning in finance (e.g., gradient-boosted trees, GNNs) to guide selection and evaluation of quantum analogs.

Experimental results

Research questions

• RQ1 Which QML algorithms show near-term feasibility for practical deployment in financial services on current NISQ hardware?
• RQ2 How do quantum-enhanced models like Quantum Variational Classifiers and Quantum Kernel Estimation compare to classical models in credit scoring and risk assessment tasks?
• RQ3 To what extent can Quantum Graph Neural Networks (QGNNs) improve fraud detection and stock price prediction compared to classical graph-based models?
• RQ4 What are the key challenges—particularly in data loading, training, and scalability—that limit the real-world adoption of QML in finance?
• RQ5 What long-term potential do QNNs and Quantum Transformers hold for transforming financial forecasting and decision-making systems?

Key findings

• Quantum Variational Classifiers and Quantum Kernel Estimation show near-term promise for financial applications like credit scoring and risk management on NISQ devices.
• The egograph-based QGNN (egoQGNN) architecture provides a theoretical framework for graph classification and demonstrates improved handling of real-world datasets through hierarchical k-hop neighborhood processing.
• While some QML algorithms achieve high classification success rates in simulation, real hardware validation remains essential to confirm performance gains over classical methods.
• Quantum Neural Networks (QNNs) and Quantum Transformers show strong long-term potential across supervised learning, generative AI, and graph-based problems, though they remain out of reach for current quantum hardware.
• Quantum Graph Neural Networks (QGNNs) hold significant promise for fraud detection and stock prediction, but scalability and data mapping challenges remain unresolved.
• Despite the absence of proven exponential quantum advantage in financial tasks, hybrid QML approaches offer a viable path for incremental improvements in precision and efficiency, especially with further experimental validation.