Privacy Guarantees in Posterior Sampling under Contamination

In recent years, differential privacy has been adopted by tech-companies and governmental agencies as the standard for measuring privacy in algorithms. We study the level of differential privacy in Bayesian posterior sampling setups. As opposed to the common privatization approach of injecting Laplace/Gaussian noise into the output, Huber's contamination model is considered, where we replace at random the data points with samples from a heavy-tailed distribution. We derived bounds for the differential privacy level (, ) for our approach while lifting the common restriction on assuming bounded observation and parameter space seen in the existing literature. We further consider the effect of sample size on privacy level and the convergence rate of (, ) to zero. Asymptotically, the contamination approach is fully private at no cost of information loss. We also provide some examples depicting inference models that our setup is applicable to with a theoretical estimation of convergence rate.