Knowledge Graph — Coursera Notes › Academic disciplines › Information Technology / Computer Science › Machine Learning › Federated Learning

Non-IID Data in Federated Learning

concept · part of Federated Learning

Non-IID data, where each device's local data distribution differs from the global distribution, poses a significant challenge in federated learning. This can lead to slow convergence, reduced model accuracy, or even divergence if not addressed. Techniques to mitigate non-IID effects include using weighted aggregation, personalized federated learning (e.g., multi-task learning or meta-learning), and strategies like FedProx or SCAFFOLD that introduce proximal terms or control variates to stabilize training.

This is the text view of an interactive 3D knowledge graph — open this page with JavaScript enabled to explore it visually.