knockoff
A variable selection framework that constructs knockoff variables correlated with the original variables and compares their statistics, achieving exact false discovery rate control under finite samples without depending on the true model specification.
Timeline (last 90 days)
SelectPFN generates knockoff statistics through column-permutation equivariance combined with random interleaving of original columns and knockoff columns, achieving exact finite-sample false discovery rate control, and this guarantee holds even when the meta-prior is misspecified.
The knockoff method constructs a counterfeit variable for each original variable and requires the two to satisfy exchange symmetry in the joint distribution, thereby using the difference in statistics between original variables and knockoffs to control the FDR; column-permutation equivariance is a key prerequisite for constructing valid knockoff statistics.
All Facts (2)
SelectPFN generates knockoff statistics through column-permutation equivariance combined with random interleaving of original columns and knockoff columns, achieving exact finite-sample false discovery rate control, and this guarantee holds even when the meta-prior is misspecified.
50%UnverifiedThe knockoff method constructs a counterfeit variable for each original variable and requires the two to satisfy exchange symmetry in the joint distribution, thereby using the difference in statistics between original variables and knockoffs to control the FDR; column-permutation equivariance is a key prerequisite for constructing valid knockoff statistics.
50%