Genie: A Database Generator for Testing Inference Detection Tools

This paper describes a system called Genie, which generates databases suitable for testing inference detection tools. In order to provide the inter-relationships that must exist among data instances if the database is actually to have inferences, Genie uses a simulator to mimic “real world” activity and captures data from the simulator. Since the data is based on a simulation, it will have the necessary inter-relationships. When simulator-based data cohesiveness is not required, Genie provides a means to generate instances that are not related to the simulator. It also provides a means to associate external semantics with the data by renaming data to associate it with desired “real-world” objects and activities. The paper describes the database that is currently generated by Genie and then shows how a set of inferences that have been identified by the AERIE inference research project can be supported by the database. These inferences are organized in terms of the inference targets specified by the AERIE inference model. The paper describes a language called FGL (Fact Generation Language), which can be used to program Genie to generate various databases, including the one presented in this paper. It then presents a description of the Genie architecture. Finally, the paper concludes with observations of our experience to date in using Genie to support the development of inference detection tools. This work was supported under Maryland Procurement Office Contract No. MDA 904-92-C-5146.