E-Gamma Divergence: Its Properties and Applications in Differential Privacy and Mixing Times

We investigate the strong data processing inequalities of contractive Markov Kernels under a specific f-divergence, namely the E-gamma-divergence. More specifically, we characterize an upper bound on the E-gamma-divergence between PK and QK, the output distributions of contractive Markov kernel K, in terms of the E-gamma-divergence between the corresponding input distributions P and Q. Interestingly, the tightest such upper bound turns out to have a non-multiplicative form. We apply our results to derive new bounds for the local differential privacy guarantees offered by the sequential application of a privacy mechanism to data and we demonstrate that our framework unifies the analysis of mixing times for contractive Markov kernels.