|
Activity Number:
|
601
- Recent Advances in Variable Selection for Linear and Nonlinear Models
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Thursday, August 1, 2019 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Biometrics Section
|
|
Abstract #306384
|
Presentation
|
|
Title:
|
Metropolized Knockoff Sampling
|
|
Author(s):
|
Stephen Bates* and Emmanuel Candes and Lucas Janson and Wenshuo Wang
|
|
Companies:
|
Stanford and Stanford University and Harvard University and Harvard University
|
|
Keywords:
|
false discovery rate;
Metropolis-Hastings;
Ising model;
junction tree;
treewidth;
Markov chain
|
|
Abstract:
|
Model-X knockoffs is a method for transforming nearly any feature importance measure into a variable selection procedure that provably controls the false discovery rate in finite samples. A remaining challenge when applying this method is to construct the knockoff variables, a synthetic set of variables obeying a crucial exchangeability property with the explanatory variables under study. We introduce techniques for knockoff generation in great generality; we provide a sequential characterization of all possible knockoff distributions, which leads to an MCMC formulation of exact knockoff sampling. We then show how to use conditional independence structure to speed up computations and prove that our algorithms achieve near-optimal computational complexity in certain cases. The techniques we develop are sufficiently rich to enable knockoff sampling in challenging models including cases where the covariates are continuous and heavy-tailed and where they follow an Ising model.
|
Authors who are presenting talks have a * after their name.
