FedHUG: Federated Heterogeneous Unsupervised Generalization for Remote Physiological Measurements

2025-10-14

AI Summary

Remote physiological measurement faces two major challenges: privacy-sensitive data collection and poor model generalizability due to unlabeled, heterogeneous client data. This paper proposes Federated Unsupervised Domain Generalization (FUDG), a novel framework that— for the first time—enables robust model training and cross-modal (RGB video + millimeter-wave radar) physiological signal estimation across multi-source, unlabeled, non-IID clients. Methodologically, FUDG introduces three key innovations: a minimum-bias aggregation mechanism, a global distribution-aware controller, and a dynamic weight adjustment strategy—jointly addressing label scarcity, domain shift, and long-tailed data distributions. Extensive experiments on diverse real-world scenarios demonstrate that FUDG significantly outperforms existing methods in both accuracy and cross-domain generalizability. The implementation code will be publicly released.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Computer Vision: Multi-modal Vision

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Security and Privacy: Large-scale security measurements
• Systems and Infrastructure for Web, Mobile and WoT: Federated Web and WoT systems, including distributed, federated and edge-based data processing

Abstract

Remote physiological measurement gained wide attention, while it requires collecting users' privacy-sensitive information, and existing contactless measurements still rely on labeled client data. This presents challenges when we want to further update real-world deployed models with numerous user data lacking labels. To resolve these challenges, we instantiate a new protocol called Federated Unsupervised Domain Generalization (FUDG) in this work. Subsequently, the extbf{Fed}erated extbf{H}eterogeneous extbf{U}nsupervised extbf{G}eneralization ( extbf{FedHUG}) framework is proposed and consists of: (1) Minimal Bias Aggregation module dynamically adjusts aggregation weights based on prior-driven bias evaluation to cope with heterogeneous non-IID features from multiple domains. (2) The Global Distribution-aware Learning Controller parameterizes the label distribution and dynamically manipulates client-specific training strategies, thereby mitigating the server-client label distribution skew and long-tail issue. The proposal shows superior performance across state-of-the-art techniques in estimation with either RGB video or mmWave radar. The code will be released.

Problem

• Federated learning addresses privacy-sensitive remote physiological measurements
• Handles unlabeled heterogeneous client data across multiple domains
• Mitigates server-client label distribution skew and long-tail issues

Innovation

• Dynamic aggregation weights for heterogeneous non-IID features
• Parameterized label distribution to mitigate skew issues
• Federated unsupervised learning for remote physiological measurements