Federated Learning for Nonparametric Function Estimation - Framework and Optimality

Explore federated learning for nonparametric function estimation in this comprehensive conference talk. Delve into the challenges of data governance and privacy in machine learning, focusing on how organizations can collaboratively train shared global statistical models without compromising raw data. Examine the statistical optimality of federated learning in nonparametric regression, considering heterogeneous settings with varying sample sizes and differential privacy constraints across servers. Investigate both global and pointwise estimation, uncovering optimal convergence rates over Besov spaces. Learn about proposed distributed privacy-preserving estimation procedures and their theoretical properties. Gain insights into the balance between accuracy and privacy preservation, understanding the trade-offs in terms of privacy budget and data distribution within the privacy framework. Discover how sample size affects privacy retention and explore the differences between pointwise and global estimation under distributed privacy constraints.