TITLE: Nonparametric Graph Estimation

SPEAKER: Han Liu, Johns Hopkins University

ABSTRACT:

Graphical models have proven to be an extremely useful abstraction in statistical machine learning. The starting point is the graph of a distribution. While usually the graph is assumed given, we are interested in estimating the graph from data. In this talk we present
several new nonparametric and semiparametric methods for graph estimation. One approach is named "the nonparanormal," which uses copula methods to transform the variables by monotonic functions, relaxing the fully parametric assumptions made by the Gaussian
graphical model. Another approach is to restrict the family of allowed graphs to forest graphs, enabling the use of fully nonparametric density estimation. Finally, we introduce a new framework named "graph-valued regression" where the graph is estimated conditioned on
a recursive partition of a set of covariates. The resulting methods are easy to understand and compute, theoretically well supported, and effective for modeling and exploring high dimensional data. Joint work with Fang Han, John Lafferty, Larry Wasserman, and Tuo Zhao.

Bio: Han Liu graduated from the joint Ph.D. program in Machine Learning and Statistics at Carnegie Mellon University in 2011.  Since then he became an assistant professor of Biostatistics and Computer Science at Johns Hopkins University. His research lies at the intersection of Modern Statistics and Computer Science. Especially, he is interested in large-scale nonparametric methods, which directly conduct inference in infinite-dimensional
spaces and are more flexible to capture the subtleties in modern applications. He is the PI of a NSF grant and has won several academic awards, including the J. Tinsley Oden faculty fellowship from UT Austin, the best overall paper runner up award at the 26th
international conference of Machine Learning and a PHD fellowship from Google.