In this talk, I will introduce a class of compositional stochastic optimization involving conditional expectations. This class of problems lies in between the classical stochastic optimization and multistage stochastic programming, and finds a wide spectrum of applications particularly in reinforcement learning and robust learning. We establish the sample complexities of a modified Sample Average Approximation (SAA) and a biased Stochastic Approximation (SA) for such problems, under various structural assumptions such as smoothness and quadratic growth conditions. When only limited samples are available from the conditional distribution, we also present a sample efficient saddle point approach to address the problem and provide numerical results on several applications in reinforcement learning.

Speaker Bio:

Niao He is an assistant professor in the Department of Industrial and Enterprise Systems Engineering and Coordinated Science Laboratory at the University of Illinois at Urbana-Champaign. Before joining Illinois, she received her Ph.D. degree in Operations Research from Georgia Institute of Technology in 2015 and B.S. degree in Mathematics from University of Science and Technology of China in 2010. Her research interests are in large-scale optimization and machine learning, with a primary focus in bridging modern optimization theory and algorithms with core machine learning topics, like Bayesian inference, reinforcement learning, and adversarial learning. She is also a recipient of the Best Paper Award at AISTATS 2016, the NSF CISE Research Initiation Initiative (CRII) award, and the NCSA Faculty Fellowship.