ECE598: Interplay between Control and Machine Learning (Fall 2020)
Course Information
Course DescriptionAdvanced graduate course focuses on interplay between control and machine learning. The first half of the course focuses on tailoring control tools to study algorithms in large-scale machine learning. In the second half of the course, students will study how to combine reinforcement learning and model-based control methods for control design problems. The following topics will be covered: empirical risk minimization; first-order methods for large-scale machine learning; stochastic optimization; dissipation inequality; jump system theory; Lur’e-Postnikov type Lyapunov functions; integral quadratic constraints; KYP Lemma; graphical interpretations for optimization methods; implicit bias; neural tangent kernel and adaptive control; control-oriented analysis tools for temporal difference learning and Q-learning; reinforcement learning for linear quadratic regulator (LQR) problems; learning model predictive control for iterative tasks; zeroth-order optimization and evolutionary strategies; policy gradient for robust control (global convergence and implicit bias); adversarial reinforcement learning; data-driven control of large-scale switching systems; iterative learning control; imitation learning for control; regularization of model-free control via prior model-based design; constrained policy optimization.Required MaterialsThere is no required textbook for the class. All course material will be presented in class and/or provided online as notes. Links for relevant papers will be listed in the resourse section of the course website. PrerequisitesECE 515. ECE 534 and ECE 490 are recommended, but not required. Grading50% regular homework sets (3 sets of homework in total, 15%+20%+15%); 50% written research report (detailed guidelines for the final projects will be posted in the resource section). |