Local Information Privacy and Its Application to Privacy-Preserving Data Aggregation

In this article, we propose local information privacy (LIP), and design LIP based mechanisms for statistical aggregation while protecting users’ without relying on a trusted third party. The concept of context-awareness is incorporated in LIP, which can be viewed as exploiting data prior (both privatizing post-processing) to enhance utility. We present an optimization framework minimize the mean square error each user’s input or correlated latent variable by satisfying constraints. Then, study optimal under different scenarios considering uncertainty correlation with variable. Three types are studied including randomized response (RR), unary encoding (UE), hashing (LH), derive closed-form solutions perturbation parameters that prior-dependent. compare LIP-based those LDP, theoretically show former achieve enhanced then two applications: (weighted) summation histogram estimation, how proposed applied application. Finally, validate our analysis simulations using both synthetic real-world data. Results impact utility distributions, correlations, domain sizes. also provide better utility-privacy tradeoffs than LDP-based ones.