FedGCA: Global Consistent Augmentation Based Single-Source Federated Domain Generalization

TL;DR

This work addresses the practical single-source Federated Domain Generalization (sFedDG) problem, where clients operate within a single domain and lack cross-domain interaction. The authors introduce FedGCA, which combines a style-complement augmentation module with global guided semantic consistency and class consistency losses to reduce semantic drift and improve domain generalization to unseen targets. The approach demonstrates strong empirical gains over multiple baselines on Digits and PACS, with ablations confirming the contributions of the global and fine-grained semantic constraints and the class-regularization mechanism. Overall, FedGCA offers a scalable solution to generalize from a single domain in federated settings, potentially benefiting privacy-preserving applications with restricted cross-domain data exchange.

Abstract

Federated Domain Generalization (FedDG) aims to train the global model for generalization ability to unseen domains with multi-domain training samples. However, clients in federated learning networks are often confined to a single, non-IID domain due to inherent sampling and temporal limitations. The lack of cross-domain interaction and the in-domain divergence impede the learning of domain-common features and limit the effectiveness of existing FedDG, referred to as the single-source FedDG (sFedDG) problem. To address this, we introduce the Federated Global Consistent Augmentation (FedGCA) method, which incorporates a style-complement module to augment data samples with diverse domain styles. To ensure the effective integration of augmented samples, FedGCA employs both global guided semantic consistency and class consistency, mitigating inconsistencies from local semantics within individual clients and classes across multiple clients. The conducted extensive experiments demonstrate the superiority of FedGCA.

Figures (4)

• Figure 1: Description of the sFedDG problem.
• Figure 2: Overview of the FedGCA framework. Left: Local training for the client. Right: Global update on the server.
• Figure 3: Sensitivity result for different values of $\alpha$ and $\beta$
• Figure 4: Visualization for CAMs of different methods.