ELEG 626 - Learning for Dynamical Control

Credit(s): 3

LEARNING FOR DYNAMICAL CONTROL
Component: Lecture
Modeling and controlling real-world dynamical systems requires understanding the underlying stochastic nature of observations and their operations. The underlying mechanics may be completely unknown or partially observable, and one has to learn how to act or control from incomplete information. This can involve model-based and model-free approaches. This course will cover the fundamentals of dynamical modeling and reinforcement learning. Key concepts and approaches covered include dynamic programming, stochastic function approximation, the exploration-exploitation trade-off, temporal difference learning, Q-learning, and actor-critic. Applications in contexts involving multiple agents, human-in the-loop, and human feedback will be explored along with issues of safety and alignment.
Repeatable for Credit: N Allowed Units: 3 Multiple Term Enrollment: N Grading Basis: Student Option
RESTRICTIONS: Previous course in machine learning or control theory recommended.

Course Typically Offered: Fall