Federated Learning Strategies Compared for Clinical Mortality Prediction

Researchers have benchmarked several federated learning strategies for predicting in-hospital mortality using the MIMIC-IV dataset. The study found that FedProx performed best in terms of AUC-ROC and AUC-PR, outperforming other strategies like FedAvg and FedAdagrad. However, no single strategy dominated all metrics, and a centralized baseline model still achieved superior performance. The research also highlighted that global models do not serve all individual care units equally, indicating a need for attention to client-specific heterogeneity in federated learning applications. AI

IMPACT Highlights the trade-offs between federated learning and centralized models for clinical prediction, informing best practices for privacy-preserving healthcare AI.

RANK_REASON Academic paper detailing a comparative benchmark of machine learning strategies. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• FedAdagrad
• FedAdam
• FedAvg
• FedCluster
• federated learning
• FedProx
• MIMIC-IV
• Rodrigo Tertulino

• paper
• model release
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →