Having your credit card stolen is a very frustrating experience, and sadly, one that many can relate to. In 2019 alone, fraudulent transactions involving credit cards issued in Europe accounted for €1.87 billion – a staggering amount, and one that is only expected to increase as online shopping grows. With its advanced capability to analyse large amounts of data and create forecasts, artificial intelligence (AI) could provide financial institutions with significant help in fighting fraud. However, using AI to detect financial crimes requires access to sensitive data, which raises concerns about data privacy and security. That’s where federated learning comes in.
“Federated learning allows the training of an AI model on your own data, without sharing the data with someone else; you just share updates to a global model. The data stays on each client’s premises – in the case of a bank, down to a single branch – to avoid any risks, but the system allows every participant to leverage their collective intelligence to train the global model. It’s a win-win: everyone is better off by collaborating, while ensuring data security and privacy at the same time,” said Prof. Radu State, head of the Services and Data Management (SEDAN) research group at SnT, and principal investigator of the research project in federated learning for PSD2-compliant data analytics, an initiative launched in February 2022 in partnership with LUXHUB.
In fact, SnT and LUXHUB have been working together to create added value, service-based financial data. The partnership is focusing on implementing groundbreaking technology in the field of artificial intelligence, while respecting the industry’s main concern: the safety and privacy of sensitive data. Experts from FinTech/ICT research and the financial sector have been working together on a federated learning model for the benefit of the entire financial service industry.
To make an example, the AI model could be trained on detecting a specific fraudulent pattern in bank transactions. Different financial institutions could then run their data through the model to find matches (and therefore frauds), while at the same time further training the model, refining its ability to detect that kind of fraud, learn from variations, and forecast the next kind of iteration. The AI would effectively act as a financial sleuth, building forecasts, testing its hypothesis, and preventing new fraud attempts. All of this would happen without any of the banks sharing sensitive data with anyone else – in fact, the data remains on their own premises during the entire process. The model only absorbs the learning it was exposed to, so that it can become more accurate, and therefore more effective for everyone.
The partnership consists of two research projects, aiming to develop a platform and service based on federated learning for business cases specific to LUXHUB. The first project focuses on the design and management of the platform, while the second project tests the models against concrete financial use cases, such as fraud detection, anti-money laundering, loan risk prediction, and transaction categorisation.
“Machine learning provides financial institutions the flexibility they need to dynamically detect novel fraud attempts, instead of blocking transactions that fall within certain static rules,” said Prof. State. “Federated learning, in addition to respecting data privacy, also allows the use of data from all over the world, in full respect of the different national jurisdictions, whereas sharing financial data across countries, even within the same bank, is not allowed,” he added.
“LUXHUB is all about fostering innovation and collaboration. That is what we have been doing since our very inception, being created by four major banks to mutualise their PSD2 compliance efforts. Collaboration is deeply rooted in the company’s DNA,” underlined LUXHUB’s Chief Operating Officer, Claude Meurisse. “Leveraging data and the knowledge of SnT researchers, this project with LUXHUB on federated machine learning is highly innovative, aiming at identifying illicit financial activity by enabling shared learning, but without any risk in sharing data. The project outcomes have tremendous potential allowing the design of solutions that leverage data across several stakeholders in a secure and compliant manner,” said Director of SnT, Björn Ottersten.
On SnT’s side, the project is led by State’s research group SEDAN, which has a long-standing experience in developing FinTech solutions, especially in the areas of anti-money laundering and know-your-customer technologies. They will also analyse the meaning and the outcomes of this project in relation to its potential for FinTech at large. “Developing a federated learning platform will also have spillover effects onto other areas of FinTech. Research is about opening doors, oftentimes some you didn’t even know existed,” concluded State.