Hyderabad: The Machine Learning Lab (MLL) at the International Institute of Information Technology Hyderabad (IIITH) is now spearheading advancements in federated learning (FL) and offering innovative solutions for financial services. Led by Prof. Sujit Gujar, the lab will be tackling challenges related to privacy, fairness and incentivisation, which play a critical role in adapting artificial intelligence to the financial sector.
In simple words, FL allows companies to develop powerful machine learning models without directly accessing client data, an approach vital for industries like finance, where data privacy is paramount. “For example, a company might want to create a service to predict loan defaulters by gathering the data from banks. However, banks may be reluctant to share their sensitive customer data due to privacy concerns. FL resolves this issue by enabling model training on decentralised data, ensuring banks keep control over their data while still contributing to model improvements,” Prof Gujar explains.
Despite these benefits, he notes that FL in the financial sector presents its own challenges. One concern is the risk of model inversion attacks, where a model could indirectly reveal sensitive information about individuals. There is also the issue of fairness, as these models need to perform well across diverse demographic groups. Moreover, banks might be tempted to “free-ride,” benefiting from the shared model without contributing their data unless incentivised.
The MLL has addressed these concerns by combining machine learning and game theory. “To mitigate privacy risks, we have developed techniques that allow optimal learning without compromising sensitive data. Additionally, by employing incentive engineering, they ensure fair compensation for all participants, preventing free-riding,” he explained. Finally, their research includes developing methods to ensure the final model is fair across various demographics, an essential factor in financial services where bias could lead to significant problems. While these innovations are motivated by financial sector applications, Prof. Gujar says that the techniques can be adapted to other industries as well.
