Secure Semantic Databases: Integrating Knowledge Management and Encrypted Machine Learning for Intelligent Data Systems

Abstract

The combination of mass data management and artificial intelligence has resulted in the desperate need of intelligent data system to be semantically, and highly-secure. Traditional databases are very good at storing systematized data but per se they lack enlightenment of convoluted connections and importance (semantics). Instead, Knowledge Management (KM) systems, such as knowledge graphs, provide such semantic layer permitting to perform strong reasoning and question. However, feeding machine learning (ML) using this semantic data would be tend to reveal sensitive information, which can be a severe privacy and security hazard.

The paper proposes a novel model of Secure Semantic Databases (S2DB), which would integrate ideas of Knowledge Management and the privacy-awareness of Encrypted Machine Learning (EML). The S2DB system is a structure that stores and processes data on a semantic knowledge graph (data structure) to make it (intelligently) queryable and allow relation finding. Most importantly, they enable the training and inference of ML models based on this encrypted semantic information using methods like Homomorphic or Secure Multi-Party Computation (S). This is to make certain that the sensitive data contained within the knowledge base remains confidential throughout the lifecycle of ML, even by the owners of the database themselves.

The outputs of this work are an architectural version of S2DB, a prototype implementation of the work that demonstrates its viability, and a performance test on benchmark applications. The gap between utility and security of data and open the door to a new generation of smart data system will rush to fill the gap in the potential for the implementation of safety in the privacy-sensitive setting of healthcare, finance and confidential business intelligence.