Beyond Local Sharpness: Communication-Efficient Global Sharpness-aware Minimization for Federated Learning

Federated learning (FL) enables collaborative model training with privacy preservation. Data heterogeneity across edge devices (clients) can cause models to converge to sharp minima, negatively impacting generalization and robustness. Recent approaches use client-side sharpnessaware minimization (SAM) to encourage flatter minima, but the discrepancy between local and global loss landscapes often undermines their effectiveness, as optimizing for local sharpness does not ensure global flatness. This work introduces FEDGLOSS (Federated Global Serverside Sharpness), a novel FL approach that prioritizes the optimization of global sharpness on the server, using SAM. To reduce communication overhead, FEDGLOSS cleverly approximates sharpness using the previous global gradient, eliminating the need for additional client communication. Our extensive evaluations demonstrate that FED-GLOSS consistently achieves flatter minima and better performance compared to state-of-the-art FL methods in various federated vision benchmarks. Code available at github.com/pietrocagnasso/fedgloss.