Noise-aware Client Selection for carbon-efficient Federated Learning via Gradient Norm Thresholding

Researchers propose a new client selection method for carbon-efficient federated learning that filters out noisy data to improve model performance. The approach uses gradient norm thresholding to better identify quality clients while maintaining sustainability goals in distributed AI training across renewable energy-powered data centers.

• →Current client selection strategies for carbon-efficient federated learning tend to select clients with noisy data, degrading model performance.
• →A new gradient norm thresholding mechanism using probing rounds can more effectively detect and filter out noisy clients.
• →The method balances carbon budget constraints with model convergence requirements in federated learning systems.
• →Federated learning enables distributed AI training across geographically distributed data centers using renewable energy sources.
• →The approach addresses privacy-preserving challenges where data quality on client devices remains unknown.