David Hunter

Assistant Professor of Statistics
Ph.D., University of Michigan, 1999

Summary of research interests

Dr. Hunter's research concerns computational methods for optimizing high-dimensional functions which are becoming more and more commonplace in statistics. When solutions of such functions cannot be determined analytically, numerical methods become necessary to find approximate solutions.

One such numerical method is the EM algorithm for obtaining maximum likelihood estimates in the context of incomplete data. Dr. Hunter's publications deal with a kind of generalization of the EM algorithm called optimization transfer in which incomplete data are not required. Like the EM itself, these algorithms tend to require many simple iterations, as compared with few difficult iterations for a method such as Newton-Raphson. For problems in which the dimension of the parameter space is large, the tradeoff tends to be worth it in terms of total computation.

Representative publications

Hunter, D. R., Lange, K. (2000) "An Optimization Transfer Algorithm for Quantile Regression". Journal of Computational and Graphical Statistics, to appear.