First-order and online optimization methods

In this class we will study first-order optimization methods for constrained and unconstrained optimization methods. In addition, a major part of the lecture will be devoted to aspects of online convex optimization, which is a combination of convex optimization, statistical learning, and game theory. Online optimization is motivated from practical applications in which the environment is so complex that it is difficult to design robust optimization models and apply classic algorithmic theory. In the online optimization framework, the optimization is instead considered as a process that learns from experience as one goes along and more aspects of the problem are observed. In the exercise class (on demand) a practical application to recommender systems will be considered.

Date and time info
Lectures: Tuesdays 11:00-12:30, Exercises (biweekly): Tuesdays 14:00-15:00

Keywords
online convex optimization, optimization on manifolds, multi-armed bandit, games and saddle point problems

Prerequisites
Basics of linear algebra, analysis, and probability