Event Date
Speaker: Peter Kramlinger, Visiting Assistant Professor, Statistics, UC Davis
Title: "Model Selection and Inference based on the Lasso"
Abstract: It is common practice in data analysis to perform inference in models obtained by data-driven variable selection. Ignoring the uncertainty introduced by the selection procedure leads to invalid inference. This talk discusses the intrinsic features of inference based on selection procedures through the example of the Lasso, which simultaneously performs selection and estimation.
In particular, we will present our findings on how to conduct uniformly valid inference in the presence of nuisance parameters. We construct confidence sets for the regression coefficients in Gaussian linear mixed models that are based on Lasso-type estimators. We show these sets are uniformly valid over the parameter spaces of both the regression coefficients and the covariance parameters. To this end, we also prove a novel result on uniform consistency of the restricted maximum likelihood (REML) estimators of the covariance parameters.
+ + + + + + + + + + + + + + + + + + + + + + + + + + + +
Seminar date/time: Thursday, February 9th, 2023, at 4:10pm
Location: MSB 1147 (Colloquium Room)
Refreshments: 3:30pm, MSB 1147