PrivateLR (version 1.2-22)
Differentially Private Regularized Logistic Regression
Description
Implements two differentially private algorithms for
estimating L2-regularized logistic regression coefficients. A randomized
algorithm F is epsilon-differentially private (C. Dwork, Differential
Privacy, ICALP 2006 ), if
|log(P(F(D) in S)) - log(P(F(D') in S))| <= epsilon
for any pair D, D' of datasets that differ in exactly one record, any
measurable set S, and the randomness is taken over the choices F makes.