Johan Östman: The reality of federated learning: from life sciences to finance and beyond

Federated learning is a machine learning paradigm that promises to revolutionize how organizations collaborate on sensitive tasks. At its core, this technique allows multiple participants to collaboratively train a shared model while keeping their data localized, thereby avoiding data centralization. This approach is particularly relevant in today’s data-sensitive world, offering both improved data utilization and a gateway to novel applications that respect individual privacy concerns. The vast potential of federated learning is highlighted by its applicability across multiple industries. In the realm of life sciences, for example, it allows for the fusion of insights from disparate healthcare databases while preserving the privacy of individual patient data. Similarly, the finance sector can harness its power to collaboratively devise fraud detection models, without compromising transactional details of their clients.

However, as with any emerging technology, federated learning is not without its challenges and recent studies hint at underlying vulnerabilities. Examples include, data inference attacks, where individual data points are identified even without direct access to the data source, uneven benefits among clients, making the business use case especially intricate, and its susceptibility to malicious attacks, where the trained model may be manipulated for malicious objectives.

In this presentation, I will discuss the potential of federated learning from the perspective of different sectors and dissect the challenges that stand in its path. I will also discuss potential strategies and remedies to navigate these hurdles.

Johan Östman received his PhD in Information Theory in 2020 from Chalmers University of Technology. He has been active within the autonomous driving industry since 2020 and has been engaging in privacy-preserving machine learning since 2021. He is currently leading the research initiatives at AI Sweden within privacy-preserving machine learning, closely collaborating with organisations such as Chalmers, ESA, AstraZeneca, and RISE. He is the project initiator and principal investigator in a project on federated learning to combat anti-money laundering together with two of the largest banks in Sweden.