FedSDAF: Leveraging Source Domain Awareness for Enhanced Federated Domain Generalization

TL;DR

This work tackles Federated Domain Generalization under strict data isolation by challenging the emphasis on domain-invariant representations and showing that source-domain-aware features can generalize better to unseen targets. It introduces FedSDAF, a dual-adapter architecture comprising a global Domain-Invariant Adapter and a local Domain-Aware Adapter, coupled with Bidirectional Knowledge Distillation and MHSA-based integration to fuse local expertise with global consensus. The method achieves state-of-the-art results across PACS, VLCS, OfficeHome, and DomainNet, including strong improvements on the Sketch domain, and demonstrates favorable convergence and communication efficiency in privacy-preserving federated settings. These findings highlight the practical value of leveraging source-domain knowledge to improve generalization in distributed, privacy-conscious scenarios, with potential extension to other modalities and larger-scale deployments.

Abstract

Traditional Federated Domain Generalization (FedDG) methods focus on learning domain-invariant features or adapting to unseen target domains, often overlooking the unique knowledge embedded within the source domain, especially in strictly isolated federated learning environments. Through experimentation, we discovered a counterintuitive phenomenon: features learned from a complete source domain have superior generalization capabilities compared to those learned directly from the target domain. This insight leads us to propose the Federated Source Domain Awareness Framework (FedSDAF), the first systematic approach to enhance FedDG by leveraging source domain-aware features. FedSDAF employs a dual-adapter architecture that decouples "local expertise" from "global generalization consensus." A Domain-Aware Adapter, retained locally, extracts and protects the unique discriminative knowledge of each source domain, while a Domain-Invariant Adapter, shared across clients, builds a robust global consensus. To enable knowledge exchange, we introduce a Bidirectional Knowledge Distillation mechanism that facilitates efficient dialogue between the adapters. Extensive experiments on four benchmark datasets (OfficeHome, PACS, VLCS, and DomainNet) show that FedSDAF significantly outperforms existing FedDG methods. The source code is available at https://github.com/pizzareapers/FedSDAF.

Figures (10)

• Figure 1: Illustration of domain generalization.
• Figure 2: FedSDAF architecture design.
• Figure 3: Optimization process.
• Figure 4: Convergence analysis on different clients in domain benchmark.
• Figure 5: Effects of $\alpha$.