Non-Stochastic Control Problem: Online Gradient Boosting

This lecture covers the non-stochastic control problem, focusing on online gradient boosting for dynamical systems. It discusses the challenges in control, the main results of efficient algorithms, and the ingredients for successful control. The instructor presents the concept of policy regret minimization, the use of weak predictors in boosting, and the removal of state via stability. The lecture also includes experiments demonstrating the effectiveness of the proposed methods in various scenarios, such as dealing with correlated noise and controlling an inverted pendulum.