Data Heterogeneity
Data heterogeneity, the variability in data distributions across different sources in federated learning, significantly hinders model accuracy and convergence. Current research focuses on mitigating this challenge through various techniques, including personalized model architectures (e.g., using adaptors or subnetworks), robust aggregation methods (e.g., weighted averaging based on client performance or data characteristics), and innovative training strategies (e.g., warmup phases, loss decomposition, and adversarial training). Addressing data heterogeneity is crucial for advancing federated learning's applicability in diverse real-world scenarios, particularly in healthcare, industrial IoT, and other domains with decentralized, privacy-sensitive data.
Papers
Addressing Data Heterogeneity in Federated Learning of Cox Proportional Hazards Models
Navid Seidi, Satyaki Roy, Sajal K. Das, Ardhendu Tripathy
FedDM: Enhancing Communication Efficiency and Handling Data Heterogeneity in Federated Diffusion Models
Jayneel Vora, Nader Bouacida, Aditya Krishnan, Prasant Mohapatra
Synthetic Data Aided Federated Learning Using Foundation Models
Fatima Abacha, Sin G. Teo, Lucas C. Cordeiro, Mustafa A. Mustafa
FedTSA: A Cluster-based Two-Stage Aggregation Method for Model-heterogeneous Federated Learning
Boyu Fan, Chenrui Wu, Xiang Su, Pan Hui
Beyond the Federation: Topology-aware Federated Learning for Generalization to Unseen Clients
Mengmeng Ma, Tang Li, Xi Peng