Federated Learning is Needed to Overcome Key Challenges Arising from the European Union AI Act

Herbert Woisetschläger, Alexander Erben, Bill Marino, Shiqiang Wang, Nicholas D. Lane, Ruben Mayer, Hans Arno Jacobsen

Keywords: federated learning, european union artificial intelligence act, technical and legal analysis

TL;DR: We provide an analysis of the EU AI Act from a technical perspective in support of our position that federated learning is needed to overcome key challenges arising from the AI Act.

Abstract:

The European Union AI Act (AI Act) introduces comprehensive requirements for AI systems regarding data governance, safety and security, and energy efficiency and sustainability, among others. High-risk AI applications, such as AI systems for medical data processing, face particularly stringent compliance requirements. We argue that _Federated Learning (FL) is needed to overcome key challenges arising from the AI Act_, especially with regard to data governance.
Through careful analysis of the AI Act from a technical perspective, we show that the distributed architecture of FL inherently addresses regulatory requirements around data privacy, consent-based processing, and computational resource allocation. We critically examine the current shortcomings of FL in the context of the AI Act and map out research priorities that are needed to on the path towards full regulatory compliance.

Submission Number: 362