Generating knockoffs via conditional independence

DOI10.1214/23-EJS2198MaRDI QIDQ6200873zbMATH OpenFDO

Authors Emanuela Dreassi, Fabrizio Leisen, Luca Pratelli, Pietro Rigo

Publication date 25 March 2024

Published in Electronic Journal of Statistics (Search for Journal in Brave)

Copyright license Creative Commons Attribution 4.0 International

Full work available at URL https://arxiv.org/abs/2212.09398, https://projecteuclid.org/journals/electronic-journal-of-statistics/volume-18/issue-1/Generating-knockoffs-via-conditional-independence/10.1214/23-EJS2198.full

zbMATH Keywords

approximationconditional independencehigh-dimensional regressionknockoffsmultivariate dependencepartial exchangeabilityvariable selection

Mathematics Subject Classification ID

Probability distributions: general theory (60E05) Characterization and structure theory of statistical distributions (62E10) Characterization and structure theory for multivariate probability distributions; copulas (62H05) General nonlinear regression (62J02)

Abstract: Let $X$ be a $p$-variate random vector and $widetildeX$ a knockoff copy of $X$ (in the sense of cite{CFJL18}). A new approach for constructing $widetildeX$ (henceforth, NA) has been introduced in cite{JSPI}. NA has essentially three advantages: (i) To build $widetildeX$ is straightforward; (ii) The joint distribution of $(X,widetildeX)$ can be written in closed form; (iii) $widetildeX$ is often optimal under various criteria, including mean absolute correlation and reconstructability. However, for NA to apply, the distribution of $X$ needs to be of the form ${\displaystyle }$(*)quadquadquadquadquadquad P(X_1in A_1,ldots,X_pin A_p)=EBigl{prod_{i=1}^pP(X_iin A_imid Z)Bigr}${\displaystyle }$ for some random element $Z$. Our first result is that any probability measure $mu$ on $mathbb{R}^p$ can be approximated by a probability measure $m{u}_0$ which makes condition (*) true. If $mu$ is absolutely continuous, the approximation holds in total variation distance. In applications, regarding $mu$ as the distribution of $X$, this result suggests using the knockoffs based on $m{u}_0$ instead of those based on $mu$ (which are generally unknown). Our second result is a characterization of the pairs $(X,widetildeX)$ where $widetildeX$ is obtained via NA. It turns out that $(X,widetildeX)$ is of this type if and only if it can be extended to an infinite sequence so as to satisfy certain invariance conditions. The basic tool for proving this fact is de Finetti's theorem for partially exchangeable sequences.

Cites work

• Bayesian Inference for Logistic Models Using Pólya–Gamma Latent Variables
• Controlling the false discovery rate via knockoffs
• De Finetti's theorem for Markov chains
• Deep knockoffs
• Gene hunting with hidden Markov model knockoffs
• scientific article; zbMATH DE number 3875591 (Why is no real title available?)
• scientific article; zbMATH DE number 3896009 (Why is no real title available?)
• scientific article; zbMATH DE number 720689 (Why is no real title available?)
• Introduction to Graphical Modelling
• Limit theorems for a class of identically distributed random variables.
• Metropolized Knockoff Sampling
• New perspectives on knockoffs construction
• Normal variance mixtures: distribution, density and parameter estimation
• Panning for Gold: ‘Model-X’ Knockoffs for High Dimensional Controlled Variable Selection
• Powerful knockoffs via minimizing reconstructability
• Robust inference with knockoffs
• Testing goodness-of-fit and conditional independence with approximate co-sufficient sampling

Cited in

(1)

• Knockoffs for partially exchangeable categorical covariates

This page was built for publication: Generating knockoffs via conditional independence

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6200873)