Federated Domain Generalization with Latent Space Inversion

2025-12-10

AI Summary

Federated Domain Generalization (FedDG) addresses distribution shifts across non-IID clients while preserving data privacy. Existing approaches improve generalization by sharing statistical information, yet incur inherent privacy risks. This paper proposes a privacy-safe FedDG framework: first, it introduces a latent-space inversion mechanism to enhance domain-invariant representation learning at each client; second, it designs a gradient-importance-based parameter selection strategy for aggregation, uploading only the subset of model parameters that significantly contribute to local predictions. By avoiding the transmission of raw data or statistics, the method ensures stronger privacy guarantees. Extensive experiments on multiple benchmark datasets demonstrate that our approach consistently outperforms state-of-the-art FedDG methods in generalization accuracy, while reducing communication overhead by approximately 30%.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Natural Language Processing: Safety and Robustness
• Application Domains: Security

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Security and Privacy: Security and privacy of machine learning and AI applications
• Responsible Web: Data and user privacy-enhancing technologies for the Web

Abstract

Federated domain generalization (FedDG) addresses distribution shifts among clients in a federated learning framework. FedDG methods aggregate the parameters of locally trained client models to form a global model that generalizes to unseen clients while preserving data privacy. While improving the generalization capability of the global model, many existing approaches in FedDG jeopardize privacy by sharing statistics of client data between themselves. Our solution addresses this problem by contributing new ways to perform local client training and model aggregation. To improve local client training, we enforce (domain) invariance across local models with the help of a novel technique, extbf{latent space inversion}, which enables better client privacy. When clients are not emph{i.i.d}, aggregating their local models may discard certain local adaptations. To overcome this, we propose an extbf{important weight} aggregation strategy to prioritize parameters that significantly influence predictions of local models during aggregation. Our extensive experiments show that our approach achieves superior results over state-of-the-art methods with less communication overhead.

Problem

• Addresses distribution shifts among clients in federated learning.
• Improves generalization while preserving client data privacy.
• Aggregates local models effectively for non-i.i.d. client data.

Innovation

• Latent space inversion ensures client privacy
• Important weight aggregation prioritizes key parameters
• Reduces communication overhead while improving generalization