arXiv:2404.08715v1 Announce Kind: new
Summary: This text introduces differentially personal log-location-scale (DP-LLS) regression fashions, which incorporate differential privateness into LLS regression by way of the useful mechanism. The proposed fashions are established by injecting noise into the log-likelihood perform of LLS regression for perturbed parameter estimation. We are going to derive the sensitivities utilized to find out the magnitude of the injected noise and show that the proposed DP-LLS fashions fulfill $epsilon$-differential privateness. As well as, we’ll conduct simulations and case research to guage the efficiency of the proposed fashions. The findings recommend that predictor dimension, coaching pattern measurement, and privateness funds are three key elements impacting the efficiency of the proposed DP-LLS regression fashions. Furthermore, the outcomes point out {that a} sufficiently giant coaching dataset is required to concurrently guarantee respectable efficiency of the proposed fashions and obtain a passable degree of privateness safety.
Supply hyperlink