DPGen: Automated Program Synthesis for Differential Privacy

Abstract

Differential privacy has become a de facto standard for releasing data in a privacy-preserving way. Creating a differentially private algorithm is a process that often starts with a noise-free (nonprivate) algorithm. The designer then decides where to add noise, and how much of it to add. This can be a non-trivial process -if not done carefully, the algorithm might either violate differential privacy or have low utility.In this paper, we present DPGen, a program synthesizer that takes in non-private code (without any noise) and automatically synthesizes its differentially private version (with carefully calibrated noise). Under the hood, DPGen uses novel algorithms to automatically generate a sketch program with candidate locations for noise, and then optimize privacy proof and noise scales simultaneously on the sketch program. Moreover, DPGen can synthesize sophisticated mechanisms that adaptively process queries until a specified privacy budget is exhausted. When evaluated on standard benchmarks, DPGen is able to generate differentially private mechanisms that optimize simple utility functions within 120 seconds. It is also powerful enough to synthesize adaptive privacy mechanisms. CCS CONCEPTS• Security and privacy → Logic and verification; • Theory of computation → Program analysis.

PPA: Preference Profiling Attack Against Federated Learning

“…Dropout [43] and differential privacy [12], [1] are two widely used techniques in the fields of formal verification [48], program synthesis [49] and deep learning privacy protection [51], [46], [56]. Here we investigate the attack accuracy and model utility when they are adopted in FL that is under the attack of PPA.…”

PPA: Preference Profiling Attack Against Federated Learning

“…Dropout [41] and differential privacy [12] are two widely used techniques in the fields of formal verification [45], program synthesis [46] and deep learning privacy protection [44,48,53]. Here we investigate the attack accuracy and model utility when PPA is adopted in FL.…”

Privacy accounting εconomics: Improving differential privacy composition via a posteriori bounds

“…Later approaches [6,35,36,38,40] rely on the SMT solver Z3 [12], the MaxSMT solver νZ [8], or the probabilistic analysis tool PSI [24] to minimize the manual effort necessary to prove or disprove differential privacy. Wang et al's tool DPGen [37] can even transform programs violating differential privacy into differentially private ones. Recent work by Bichsel et al [7] uses machine learning to detect differential privacy violations.…”