eDSL in R for Safe Variable Transformarion.
Arbitrary data transformations that work at the level of single individual data can be safely applied before applying a differentially private data analysis if an adversary only gets to observe the result of the differentially private analysis. The current version of the PSI prototype offers support for writing variable transformations as R programs that can be run on the data before running the other private statistics. Arbitrary R programs can allow for leakage of information beyond the output, and "side-channel attacks" where an adversary observes this additional leakage and thereby undermines the privacy guarantees. TransformeR is a subset of the R language that can serve as a domain specifc language useful to write the needed data transformations and at the same time be more maneageable for preventing security weaknesses and side-channel attacks.