Yan Yan - Colloquium Speaker
Machine learning has been widely used in many areas. How can we learn a predictive model that is more robust? In this presentation, Dr. Yan will talk about their recent research on developing more robust machine learning algorithms. First, Dr. Yan will talk about how to efficiently learn a deep neural network with better generalization performance by solving a non-convex minimization problem. Second, their efforts for solving large-scale min-max problems with applications to robust learning with imbalanced data will be presented. Finally, Dr. Yan will talk about how to reformulate the min-max problem for variance-based regularization into a min-min problem and how to solve the min-min problem more efficiently than solving the min-max problem.