ConceptLEACE
LEACE
A linear concept erasure method that guarantees the failure of linear probes by projecting representations onto a subspace orthogonal to sensitive attributes, but cannot prevent nonlinear probes from exploiting residual information.
Timeline (last 90 days)
Oct 26
Linear concept erasure methods (such as LEACE) can guarantee that no linear classifier can predict the sensitive attribute after erasure, but they cannot prevent MLP probes from recovering the erased information.
Unverified50%