Online Optimization Meets Federated Learning - Tutorial

Explore the intersection of online optimization and federated learning in this comprehensive 2-hour and 22-minute tutorial from the Uncertainty in Artificial Intelligence conference. Delve into state-of-the-art theoretical results in online and bandit convex optimization, federated/distributed optimization, and emerging findings at their convergence. Begin with an in-depth look at the Online Optimization setting, focusing on the adversarial model, regret notion, and various feedback models. Analyze performance guarantees of online gradient descent-based algorithms. Next, examine the Distributed/Federated Stochastic Optimization model, discussing data heterogeneity assumptions, local update algorithms, and min-max optimal algorithms. Highlight the scarcity of results beyond the stochastic setting, particularly in adaptive adversaries. Conclude by exploring the emerging field of Distributed Online Optimization, introducing a distributed notion of regret and recent developments in first and zeroth-order feedback. Gain insights into numerous open questions and practical applications of this framework, presented by experts Aadirupa Saha and Kumar Kshitij Patel.

UAI 2023 Tutorial: Online Optimization Meets Federated Learning