Professor of Econometrics and Statistics and James S. Kemper Faculty Scholar
My work brings together statistics and machine learning to analyze and develop tools for learning from large datasets.
I have specialized in Bayesian approaches which provide a structured framework for uncertainty quantification via hierarchical modeling.
My research aims at extending the reach of Bayesian methods to large complex data by developing scalable algorithms and providing inferential theory. For example, I have created several approaches for Bayesian variable selection (EMVS, Spike-and-Slab LASSO, Factor Rotations to Sparsity) and provided R package implementations for public use. In another line of research, I have focused on establishing inferential validity of widely used machine learning (ML) tools (trees/forests and deep learning). This work was motivated by the unmet demand for in-depth theoretical understanding of Bayesian tree-based regression (such as Bayesian CART and BART). More recently, my research agenda has grown to include Bayesian likelihood-free inference. This research incorporates aspects of machine learning (such as classification or generative adversarial networks) into Bayesian posterior sampling algorithms (ABC or Metropolis-Hastings). The prevailing areas of my applied research thus far have been public health and biomedical applications.