FedOpt++ - Federation Beyond Uniform Client Selection, Optimization Beyond Simple Minimization
Watch a technical lecture exploring advanced concepts in federated learning and distributed optimization algorithms, focusing on two key limitations in current FL literature. Learn how cyclic client participation patterns affect FedAvg algorithm convergence, moving beyond traditional assumptions of full or uniform client sampling. Dive into minimax optimization applications in GANs, multi-agent games, and reinforcement learning, examining the convergence properties of Stochastic Gradient Descent Ascent (SGDA) in nonconvex settings. Discover novel analysis frameworks for SGDA-like methods in federated environments, with improved convergence and communication guarantees presented by Dr. Pranay Sharma from Carnegie Mellon University's Department of Electrical and Computer Engineering.