Watch a 56-minute FLOW seminar presentation where Carnegie Mellon University's Gauri Joshi explores the optimization of distributed learning systems by leveraging spatial and temporal correlations. Dive into federated learning concepts, starting with training at the edge and progressing through federated optimization objectives and the FedAvg algorithm. Learn about distributed mean estimation techniques, spatial similarity leveraging, and Rand-k sparsification methods. Explore practical applications through case studies including distributed power iteration, K-means clustering, and logistic regression. Examine the impact of intermittent client participation on FedAvg convergence while gaining insights into measuring spatial correlation and implementing various spatial estimators. Originally presented on January 18th, 2023, as part of the Federated Learning One World Seminar series, this technical talk provides a comprehensive exploration of advanced distributed learning optimization strategies. Read more