Dynamic Data Release Based on Differential Privacy

Data has immeasurable value but faces serious privacy concerns. Differential privacy provides a solution for privacy protection of data. Current research mainly applies differential privacy to static datasets, while data in real applications are often dynamic. Each release of dynamic data consumes a certain privacy budget. With the growth of time, the privacy budget decreases exponentially, and even the problem of privacy budget depletion also occurs. This paper proposes two approaches, LDPM and CDDP, to solve the dynamic data releasing problem with local differential privacy and center differential privacy, respectively. LDPM takes a memoization approach to cache possible truthful responses and perturbed responses locally so that the allocation of the privacy budget is only related to the number of possible truthful responses but not to the number of data updates. CDDP relaxes differential privacy into discounted differential privacy so that the privacy budget allocated per data release is constant or reduced by a small amount. Experiments on privacy and data availability are conducted on the Adult dataset. the results show that the LDPM and CDDP proposed in this paper are more effective compared to other methods.