Using differential privacy instead of deteriorating data quality to develop trustworthy AI

Data protection should always come first when storing and processing data. At the same time, data analysis is essential in research as well as in countless business processes. In order to make sensitive or personal data usable, they are usually anonymised. But anonymised data is either not secure or not useable for research. Rather, a secure, quantifiable and thus verifiable method is required that ensures the data is protected: differential privacy. The DQ0 platform for trustworthy AI uses this mathematical concept to enable unparalleled insights from sensitive data. Customers use DQ0 to halt data leaks, to create data marketplaces where external analysts can work safely with the data, or to use external data sources safely without having to trust another entity. The system provides one-click privacy. The data does not move to the analysis; the analysis moves to the data. The DQ0 Data Shield will be installed on premise right in front of the data warehouse solution or as a trusted cloud enclave with always-on end-to-end encryption. Either way the data is protected and at the same time securely available for advanced analytics for (external) data science teams. Use cases range from securely process sensitive medical data, personal user information or secret machine data. On top of these strongest privacy guarantees the DQ0 platform offers sophisticated tools to support the data science work. The integrated studi0 solution can manage data sets and models with complete versioning and runs machine learning experiments with accessible dashboards on an unlimited scale. Data officers grant access only to a selected few (data owners). Everybody else will access the datasets through the DQ0 platform (data analysts). DQ0 is a mathematically proven and certified platform for secure machine learning. It can reduce business risks and IT management complexity, create data marketplaces or make otherwise lost data sets available to researchers and data science teams.