Differentially Private Regularized Logistic Regression
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 <DOI:10.1007/11681878_14>), 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.