|
Activity Number:
|
292
- Inferential Thinking in a Machine Learning World
|
|
Type:
|
Invited
|
|
Date/Time:
|
Wednesday, August 11, 2021 : 3:30 PM to 5:20 PM
|
|
Sponsor:
|
Section on Statistical Learning and Data Science
|
|
Abstract #314471
|
|
|
Title:
|
Recent Advances in Applying Floodgate to High-Dimensional Inference
|
|
Author(s):
|
Lucas Janson* and Lu Zhang
|
|
Companies:
|
Harvard University and Harvard University
|
|
Keywords:
|
high-dimensional inference;
floodgate;
variable importance;
heterogeneity;
model-X
|
|
Abstract:
|
A recently introduced method called floodgate can provide asymptotic inference in the form of a lower confidence bound for the importance of a covariate in a (possibly high-dimensional) regression. The measure of importance that serves as the inferential target is interpretable yet completely model-free, capturing arbitrary nonlinearities and interactions in the conditional relationship between a covariate and the response given the other covariates. The floodgate method is based on the novel idea of a floodgate function, which gives a flexible deterministic yet unobservable lower-bound for the inferential target, but is much easier to provide inference for than the original target. This talk will discuss recent advances in the application of the floodgate framework to address fundamental challenges in high-dimensional inference.
|
Authors who are presenting talks have a * after their name.
