We describe a framework for exact post-selection inference in regression problems. At the core of our framework is selective inference, which builds on earlier work of Benjamini and others. In this context, we define selective Type I and II errors for hypothesis tests and selective intervals. Time allowing, several examples will be discussed, including the LASSO, forward stepwise, and LASSO with a randomized response.
The talk will be an overview of joint work with many different collaborators.
Speaker(s)
Jonathan Taylor
Professor of Statistics, Stanford University
Jonathan Taylor is a professor of statistics at Stanford University. Jonathan's recent work is in the field of selective inference in high-dimensional regression problems. Selective inference allows researchers the flexibility of using data-generated hypotheses while maintaining statistical rigor. Jonathan has also worked extensively on multiple comparison problems with structured noise such as in neuroscience and astrophysics applications. Geometric methods are at the core of much of his research.