Google has launched a system for analyzing datasets without compromising privacy.

Google Inc. introduced cryptographic protocol for confidential multiparty computation Private Join and Compute, enabling analysis and computations on encrypted datasets from multiple participants while preserving each participant's data privacy (each participant cannot obtain information about other participants' data but can perform aggregate computations on them without decryption). The implementation code of the protocol is open under the Apache 2.0 license.

Private Join and Compute allows for the transfer of a private dataset to a third party that can analyze it and assess differences with its own dataset in an aggregated manner while not being able to learn the specific values of individual records. For example, it is possible to obtain information such as the number of matching identifiers with its own dataset and the sums of values of records with matching identifiers in the encrypted dataset. However, it is impossible to know which specific values and identifiers are present in the dataset.

The Private Join and Compute protocol, also known as Private Intersection-Sum, established is based on a combination of the protocol random oblivious transfer (Random Oblivious Transfer), encrypted Bloom filters and double masking Pollard–Hellman.

The proposed system may be useful, for example, when one medical institution has information about patients' health conditions while another has information about the prescription of a new preventive drug. The "Private Join and Compute" protocol allows for the merging of encrypted datasets without revealing information and generating overall statistics that can determine whether the prescribed medication reduces morbidity. Another example, based on accident data from the traffic police and data on the use of advanced safety features in cars, can assess whether the introduction of these features affects the number of accidents.

Another example is when, based on the employee database of one company and purchase data from another, it is possible to determine how many employees from the first company made purchases at the second and for what amount. In the context of advertising networks, similar calculations can be made to assess the effectiveness of advertising campaigns, operating with lists of users who were shown the ad (or who clicked on the link) and who made purchases in the online store.

Source: opennet.ru