Skip to content
Book Open access

FedGAT: Federated Geometry-Aware and Topological Condensation

Nov 2026 · Proceedings of the 34th ACM International Conference on Multimedia · 0 citations · 55 references

Abstract

Federated learning has emerged as a promising paradigm for privacy-preserving collaborative learning across distributed clients. However, the data owned by different clients are usually highly heterogeneous and severely Non-Independent and Identically Distributed (Non-IID), which leads to client drift and unsatisfactory global model performance. Although existing federated learning methods have attempted to alleviate the Non-IID problem, most of them mainly operate at the model level and still rely on frequent communication. In this paper, we propose a novel data-centric federated learning framework called FedGAT, to address severe data heterogeneity under limited communication budgets. On the client side, we improve the quality and diversity of condensed knowledge by preserving local geometric structure and prioritizing insufficiently learned yet non-redundant samples. On the server side, we further enhance global discrimination under heterogeneous data distributions by adaptively aligning class prototypes. Extensive experiments on eight medical and natural image datasets against ten state-of-the-art methods demonstrate that FedGAT achieves superior accuracy, outperforming the strongest baseline by up to 16.48 percentage points.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.