FLEX: Robust client selection for dynamic federated learning environments

Mao Li

doi:10.1016/j.knosys.2026.116014

FLEX: Robust client selection for dynamic federated learning environments

2026-04-25

0

PRE

AI

DOI：10.1016/j.knosys.2026.116014

Original

Original request for help

Save

Abstract

En 中文

• FLEX: Novel client selection for dynamic FL via change detection and adaptive restart. • Page-Hinckley framework monitors client contributions with provable detection guarantees. • Adaptive threshold improves late-stage detection sensitivity by 50%–200% efficiently. • Achieves 94.2% performance retention and 2–3 × faster convergence than baselines.

Keywords:

Federated Learning

Client Selection

Change Detection

Adaptive Restart

Dynamic Environments

FLEX: Robust client selection for dynamic federated learning environments

Abstract

Journal

Organization