ITR: Knowledge-Based Inference Techniques to Ensure the Security of Database Content

University Of California-Los Angeles, Los Angeles CA

Investigators

Abstract

The utility of databases could be much greater if they were generally accessible across a network. But such network-available databases face serious security challenges. Users can use inference techniques on information from multiple databases to obtain data that none of these databases would directly divulge. We propose to build a system to protect network-accessible databases form this threat by establishing an inference protection system at the directory (e.g., the standard resource description framework, RDF) site. The system will require that all requests, sent to sites that store the databases, be submitted through the RDF directory site. As a result, this site will be able to observe the information content of all requests. By keeping proper records on the requests submitted by each user, and by using knowledge about database schema, contents and patterns leading to security violations, this site will be able to detect when an individual user is attempting to use data mining techniques to infer information that he could not obtain directly. A test bed of the proposed security violation protection system will be constructed. A series of experiments with data traces from real applications will be used to evaluate the effectiveness of the proposed system.