Thu, 12-Mar-2020 / 4:30pm / Packard 101
Policy gradients methods are perhaps the most widely used class of reinforcement learning algorithms. These methods apply to complex, poorly understood, control problems by performing stochastic gradient descent over a parameterized class of polices. Unfortunately, due to the multi-period nature of the objective, policy gradient algorithms face non-convex optimization problems and can get stuck in suboptimal local minima even for extremely simple problems. This talk with discus structural properties – shared by several canonical control problems – that guarantee the policy gradient objective function has no suboptimal stationary points despite being non-convex. Time permitting, I’ll also discuss (1) convergence rates that follow as a consequence of this theory and (2) consequences of this theory for policy gradient performed with highly expressive policy classes.
This talk is based on ongoing joint work with Jalaj Bhandari.
Daniel Russo is an assistant professor in the decision, risk, and operations division of the Columbia Business School. His research lies at the intersection of statistical machine learning and sequential decision-making, and contributes to the fields of online optimization, reinforcement learning, and sequential design of experiments. He joined after spending a year as an assistant professor at Northwestern’s Kellogg School of Management and one year at Microsoft Research in New England as a postdoctoral researcher. He received his Ph.D. from Stanford University in 2015 and his B.S. from the University of Michigan in 2011.
Packard Building
Electrical Engineering Department
350 Jane Stanford Way
Stanford, CA 94305-9515
Thursdays, 4pm, Packard 202
