! The results of this competition will not be benchmarked automatically. Each submission will be reviewed manually by a jury of experts.
Quantum Federated Learning for Fraud Detection
Abstract
This use case focuses on Quantum Federated Learning (QFL) for fraud detection in financial systems, highlighting the importance of maintaining privacy while identifying fraudulent activities. Financial institutions often face challenges when handling sensitive data and ensuring privacy during model training. QFL decentralizes the training process, leveraging quantum techniques to enhance efficiency and safeguard privacy. The goal of this challenge is to design a QFL architecture specifically for fraud detection, tackling key technical obstacles such as data privacy, scalability, and accuracy. Participants must propose solutions that demonstrate practical implementation and address the unique complexities of quantum and federated learning integration.
Business Problem and Inspiration
Fraud detection is a critical issue in the financial sector, requiring sophisticated models to identify complex patterns indicative of fraudulent behavior. However, the privacy of customer data remains a significant concern, making it difficult for institutions to share information for model development. This challenge is inspired by the need for an approach that allows for decentralized learning without compromising privacy. By employing Quantum Federated Learning, institutions can train models across multiple datasets while keeping the data secure, improving collaboration without the need for direct data sharing.
Technical Problem Statement
Participants are tasked with developing a QFL architecture for fraud detection. The challenge lies in designing an architecture that enables decentralized learning from multiple financial institutions while maintaining privacy and high-performance metrics, such as accuracy, precision, recall, F1 score, etc. The solution should focus on practical implementation, ensuring the model can handle the complexities of fraud detection across distributed datasets.
Current Classical Approaches
Federated Learning (FL) is widely used for privacy-preserving machine learning across distributed datasets. However, this challenge specifically explores the integration of quantum computing with federated learning, focusing on how quantum technologies can be applied in this context to further enhance the ability to detect fraudulent behavior. The use of quantum techniques in this challenge aims to open new possibilities for handling sensitive financial data in fraud detection systems.
Quantum Hardware
Participants are encouraged to use quantum simulators. If they manage to utilize quantum hardware, it will be considered an additional advantage. Quantum hardware introduces considerations like qubit noise, connectivity, and coherence times, which must be taken into account during implementation. Simulators are acceptable for demonstrating the model, but actual hardware implementation will be viewed favorably.
Desired Outcomes
The aim is to develop a QFL-based architecture for fraud detection that demonstrates the potential of combining quantum and federated learning techniques. The architecture should ensure privacy, address scalability issues, and solve fraud detection challenges. The solution should also consider practical implementation aspects, including the challenges associated with quantum algorithms, and demonstrate how QFL can be applied effectively in a financial context.
Dataset
Participants are encouraged to use the following preferred datasets for their experiments:
- Dataset 1: Synthetic Financial Datasets for Fraud Detection (Link)
- Dataset 2: Fraud Detection Bank Dataset 20K Records Binary (Link)
However, they may also choose any publicly available financial fraud dataset, such as those on Kaggle. Please note that the complexity of the chosen dataset, including the number of features and data points, will be considered during the evaluation process.
Evaluation Criteria & Numerical Metrics
Submissions will be evaluated based on the following criteria:
-
Architecture Design (25%): Innovation and clarity in the proposed QFL architecture, focusing on privacy and fraud detection capabilities.
-
Implementation (30%): The effectiveness of the model's implementation, focusing on practical applicability and scalability.
-
Performance (20%): Demonstration of how the proposed model addresses fraud detection challenges within a decentralized learning environment, assessed through binary classification metrics such as accuracy, precision, recall, F1-score, and AUC-ROC.
Note: For complex datasets, a minimum accuracy threshold of x% is required. For simpler datasets, an accuracy of at least 90% will be expected to be considered competitive.
-
Privacy Consideration (15%): Robustness in preserving data privacy during the model’s decentralized training process.
-
Quantum Hardware/Simulator Utilization (10%): Effective use of quantum simulators or hardware, with a plus for successful real hardware implementation.
References
- Chen, S. Y. C., & Yoo, S. (2021). Federated quantum machine learning. Entropy, 23(4), 460.
- Li, W., Lu, S., & Deng, D. L. (2021). Quantum federated learning through blind quantum computing. Science China Physics, Mechanics & Astronomy, 64(10), 100312.
- Chehimi, M., & Saad, W. (2022, May). Quantum federated learning with quantum data. In ICASSP 2022-2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) (pp. 8617-8621). IEEE.
- Chehimi, M., Chen, S. Y. C., Saad, W., Towsley, D., & Debbah, M. (2023). Foundations of quantum federated learning over classical and quantum networks. IEEE Network.
- Innan, N., Khan, M. A. Z., Marchisio, A., Shafique, M., & Bennai, M. (2024). FedQNN: Federated Learning using Quantum Neural Networks. arXiv preprint arXiv:2403.10861.
- Innan, N., Marchisio, A., Shafique, M., & Bennai, M. (2024). QFNN-FFD: Quantum Federated Neural Network for Financial Fraud Detection. arXiv preprint arXiv:2404.02595.
- Innan, N., Sawaika, A., Dhor ... & Bennai, M. (2024). Financial fraud detection using quantum graph neural networks. Quantum Machine Intelligence, 6(1), 7.
- Innan, N., Khan, M. A. Z., & Bennai, M. (2024). Financial fraud detection: a comparative study of quantum machine learning models. International Journal of Quantum Information, 22(02), 2350044.
- Tudisco, A., Volpe, D., Ranieri, G., Curato, G., Ricossa, D., Graziano, M., & Corbelletto, D. (2024). Evaluating the computational advantages of the Variational Quantum Circuit model in Financial Fraud Detection. IEEE Access.
- Di Pierro, A., & Incudini, M. (2021). Quantum machine learning and fraud detection. In Protocols, Strands, and Logic: Essays Dedicated to Joshua Guttman on the Occasion of his 66.66th Birthday (pp. 139-155). Cham: Springer International Publishing.
Challenge Organizers:
Muhammad Shafique, Alberto Marchisio, Muhammad Kashif, Nouhaila Innan
*Supported by NYUAD Center for Quantum and Topological Systems (CQTS).
