Federated Unlearning: a Perspective of Stability and Fairness

TL;DR

This paper analyzes federated unlearning's stability and fairness issues under data heterogeneity, introduces metrics and an optimization framework, and proposes mechanisms validated through empirical experiments.

Contribution

It provides a comprehensive theoretical analysis of trade-offs in federated unlearning and proposes mechanisms to manage these trade-offs considering data heterogeneity.

Findings

Proposed metrics for FU assessment including verification, stability, and fairness.

Formulated FU process with data heterogeneity via an optimization framework.

Empirical validation shows proposed mechanisms effectively balance trade-offs.

Abstract

This paper explores the multifaceted consequences of federated unlearning (FU) with data heterogeneity. We introduce key metrics for FU assessment, concentrating on verification, global stability, and local fairness, and investigate the inherent trade-offs. Furthermore, we formulate the unlearning process with data heterogeneity through an optimization framework. Our key contribution lies in a comprehensive theoretical analysis of the trade-offs in FU and provides insights into data heterogeneity's impacts on FU. Leveraging these insights, we propose FU mechanisms to manage the trade-offs, guiding further development for FU mechanisms. We empirically validate that our FU mechanisms effectively balance trade-offs, confirming insights derived from our theoretical analysis.