Google Inc. the source texts of the library “” with the implementation of methods , allowing for statistical operations on datasets to be performed with sufficiently high accuracy without the possibility of identifying individual records in them. The library's code is written in C++ and under the Apache 2.0 license.
Analysis using differential privacy methods enables organizations to create analytical samples from statistical databases without allowing data partitioning and extracting specific individuals' data from the overall information. For example, to identify differences in patient care, researchers can be provided with information that allows them to compare the average length of patient stays in hospitals while maintaining patient confidentiality and preventing the disclosure of their details.
The proposed library includes the implementation of several algorithms for generating aggregated statistics based on sets of numerical data that include confidential information. To verify the correctness of the algorithms, a is provided. The algorithms allow for operations on the data such as summation, counting, calculating averages, standard deviation, variance, and order statistics, including determining the minimum, maximum, and median. It also includes the implementation of , which can be used for calculations not covered by predefined algorithms.
The library uses a modular architecture that allows for expanding existing functionality and adding additional mechanisms, aggregate functions, and privacy control tools.
Based on the library for PostgreSQL database management system 11 an extension with a set of anonymous aggregate functions utilizing differential privacy methods — ANON_COUNT, ANON_SUM, ANON_AVG, ANON_VAR, ANON_STDDEV, and ANON_NTILE.
Source: opennet.ru
