Privacy-Preserving Quantile Treatment Effect Estimation for Randomized Controlled Trials

In accordance with the precept of “knowledge minimization,” many web corporations are opting to document much less knowledge. Nevertheless, that is typically at odds with A/B testing efficacy. For experiments with models with a number of observations, one fashionable data-minimizing approach is to mixture knowledge for every unit. Nevertheless, actual quantile estimation requires the total observation-level knowledge. On this paper, we develop a technique for approximate Quantile Remedy Impact (QTE) evaluation utilizing histogram aggregation. As well as, we will additionally obtain formal privateness ensures utilizing differential privateness.