Federated domain generalization via data-centric flatness optimization

doi:10.1016/j.patcog.2025.113023

Federated domain generalization via data-centric flatness optimization

2025-12-31

0

PRE

AI

DOI：10.1016/j.patcog.2025.113023

Original

Original request for help

Save

Abstract

En 中文

• Presents a data-centric method to find global flat minima in federated learning. • Generates surrogate global data through model-guided adversarial augmentation. • Theoretically upper-bounds robust risk, linking flatness to unseen-domain accuracy. • Outperforms earlier methods on several benchmarks.

Federated domain generalization via data-centric flatness optimization

Abstract

Journal

Organization