Department of Statistics, The Ohio State University
Statistics and Biostatistics Colloquium Series

Problems of model search in regression with very large numbers of candidate models raise challenges for both specification and computation. Model/prior assumptions that encourage (or enforce) sparsity are desirable, if not necessary, in order that currently known model search methods -- stochastic or deterministic -- scale to even modest dimensions. However even under these assumptions of sparsity, the interesting regions of the model space are too large to search using standard search algorithms, and so novel search methods are needed for the rapid identification of promising models. Our work with large-scale regressions provides some examples of how coherent Bayesian models can be developed and applied in problems in high dimensions. We describe a distributed computational, "shotgun stochastic search" approach to regression model search, and address issues of model averaging for prediction over these large spaces.

Meet the speaker in Room 212 Cockins Hall at 4:30 p.m. Refreshments will be served.