Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Federated Learning Priorities Under the European Union Artificial Intelligence Act

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

TL;DR

This work analyzes how the EU AI Act reshapes Federated Learning (FL) by foregrounding data governance, energy efficiency, and robustness requirements. It adopts an interdisciplinary approach that blends legal analysis with ML experimentation, including DP/SMPC/HE privacy techniques and a LoRA-based fine-tuning of a $110M$-parameter BERT in a high-risk text-classification scenario. Key findings show a pronounced privacy-energy trade-off in FL, with FL requiring significantly more energy than centralized training to reach comparable accuracy, though privacy-preserving methods and parameter-efficient fine-tuning can mitigate some costs. The study argues that, with targeted research priorities and regulatory-aligned development, FL could become the preferred, regulation-compliant pathway for responsible AI deployment in high-stakes settings, guiding both the FL community and policymakers toward practical, scalable solutions.

Abstract

The age of AI regulation is upon us, with the European Union Artificial Intelligence Act (AI Act) leading the way. Our key inquiry is how this will affect Federated Learning (FL), whose starting point of prioritizing data privacy while performing ML fundamentally differs from that of centralized learning. We believe the AI Act and future regulations could be the missing catalyst that pushes FL toward mainstream adoption. However, this can only occur if the FL community reprioritizes its research focus. In our position paper, we perform a first-of-its-kind interdisciplinary analysis (legal and ML) of the impact the AI Act may have on FL and make a series of observations supporting our primary position through quantitative and qualitative analysis. We explore data governance issues and the concern for privacy. We establish new challenges regarding performance and energy efficiency within lifecycle monitoring. Taken together, our analysis suggests there is a sizable opportunity for FL to become a crucial component of AI Act-compliant ML systems and for the new regulation to drive the adoption of FL techniques in general. Most noteworthy are the opportunities to defend against data bias and enhance private and secure computation

Federated Learning Priorities Under the European Union Artificial Intelligence Act

TL;DR

This work analyzes how the EU AI Act reshapes Federated Learning (FL) by foregrounding data governance, energy efficiency, and robustness requirements. It adopts an interdisciplinary approach that blends legal analysis with ML experimentation, including DP/SMPC/HE privacy techniques and a LoRA-based fine-tuning of a -parameter BERT in a high-risk text-classification scenario. Key findings show a pronounced privacy-energy trade-off in FL, with FL requiring significantly more energy than centralized training to reach comparable accuracy, though privacy-preserving methods and parameter-efficient fine-tuning can mitigate some costs. The study argues that, with targeted research priorities and regulatory-aligned development, FL could become the preferred, regulation-compliant pathway for responsible AI deployment in high-stakes settings, guiding both the FL community and policymakers toward practical, scalable solutions.

Abstract

The age of AI regulation is upon us, with the European Union Artificial Intelligence Act (AI Act) leading the way. Our key inquiry is how this will affect Federated Learning (FL), whose starting point of prioritizing data privacy while performing ML fundamentally differs from that of centralized learning. We believe the AI Act and future regulations could be the missing catalyst that pushes FL toward mainstream adoption. However, this can only occur if the FL community reprioritizes its research focus. In our position paper, we perform a first-of-its-kind interdisciplinary analysis (legal and ML) of the impact the AI Act may have on FL and make a series of observations supporting our primary position through quantitative and qualitative analysis. We explore data governance issues and the concern for privacy. We establish new challenges regarding performance and energy efficiency within lifecycle monitoring. Taken together, our analysis suggests there is a sizable opportunity for FL to become a crucial component of AI Act-compliant ML systems and for the new regulation to drive the adoption of FL techniques in general. Most noteworthy are the opportunities to defend against data bias and enhance private and secure computation
Paper Structure (24 sections, 2 equations, 4 figures, 4 tables, 2 algorithms)

This paper contains 24 sections, 2 equations, 4 figures, 4 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. The EU Artificial Intelligence Act
  3. Methodology
  4. Evaluation criteria
  5. Quantitative Analysis
  6. Qualitative Analysis
  7. Analysis
  8. Data Governance
  9. Energy Efficiency
  10. Robustness & Quality Management
  11. Qualitative Analysis
  12. Future Research Priorities
  13. Conclusion
  14. Details on the AI Act
  15. Article vs. Recitals in the AI Act
  16. ...and 9 more sections

Figures (4)

  • Figure 1: To achieve high privacy guarantees in small systems, we require high $z$ that come at significant model performance and efficiency costs. Training stability also diminishes with increasing $z$. $\epsilon$ is calculated based on $\delta = \frac{1}{16,000}$.
  • Figure 2: Baseline Experiments. We identify major causes of energy efficiencies in FL systems. The projected energy costs in the EU, especially CO$_2$ pricing, require us to focus on improving the energy efficiency of FL.
  • Figure 3: Visualization of client subsets for all of our experiments.
  • Figure 4: FL system design depicting the network topology for an aggregation round in FL between clients and the aggregation server. Every communication point consumes energy per transmitted bit, which must be accounted for.