Using Stitching for faster sampling (Prof. Mark Huber)

Title: Using Stitching for faster sampling

Speaker: Mark Huber, Department of Mathematics, Claremont McKenna College

Abstract: Point processes are used to model location data, such as the locations of trees in a forest, or cities in a plain.  Repulsive point processes modify the basic model in order to obtain points that are farther apart from each other than would be expected if they were placed uniformly at random.  In order to understand the behavior of these models, Monte Carlo methods are used, which draw samples from the probabilistic model.  In this talk, I’ll show how to draw from a particular example of a repulsive point process called the Strauss process for parameters that were never possible before.  The method is called stitching, and is a type of divide-and-conquer algorithm that is surprisingly effective for these types of problems.

Huber got his start in data science (before it was called that) at HMC (’94).  He then headed to Cornell and obtained his Ph.D. from the Operations Research and Industrial Engineering department.  After a postdoc at Stanford and a position at Duke, he returned to the West Coast and is now the Fletcher Jones Foundation Professor of Mathematics and Statistics and George R. Roberts Fellow, and the Program Director of Data Science and Computer Science at Claremont McKenna.  His third book, “Probability Adventures”, is now available.