Federated Learning in Banking and Finance

Design and management of a secure federated learning platform for the banking and finance sector, including model assessment and concrete financial use cases.

The project at a glance

Start date:

01 Jan 2022
Duration in months:

36
Funding:

LuxHub
Principal Investigator(s):

Radu State

About

This project, launched in partnership with LuxHub, addresses a challenging problem from both a research and an operational standpoint: designing an effective and secure federated learning platform for financial institutions in Luxembourg. Anti-money laundering regulations, which are mandatory in the financial world, can leverage powerful machine learning models, capable to process large quantities of real-time data. However, the quality of the results depends directly on the input data used to train the algorithms. This input data can be biased, or insufficient to lead to accurate models. A promising approach to solve this challenge, and thus generate better models, consists in federating multiple actors and sharing the input data among them to train the algorithms. However, this is not as simple as it sounds. Data sharing among financial entities is impossible, due to the sensitivity and business relevance of the data. Nevertheless, a model can be trained by a federation of nodes using piecewise localized and privacy-preserving data processing. There are multiple research questions related to the overall performance of such a solution, as well as practical questions on its governance, and engineering challenges on how to render it operational. The project team expects multiple societal and break-through results thanks to the great potential of federating efforts, expertise and data among all participants.