Intelligent Detection of Human-generated Threats (IDEHT)

We propose to continue Phase II development of the Intelligent Detection of Human-generated Threats (IDEHT) system, which will use artificial intelligence and machine learning capabilities to detect emerging human-generated physical threats to Army installations by monitoring open source and security-related data sets. Our innovative approach includes analyzing potentially threatening individuals and groups of individuals in the context of their overall networks of relationships across disparate data sources, using advanced artificial intelligence methods to learn to automatically detect patterns of threatening behavior in the data, and alerting emergency management personnel to identified threats with supporting evidence and analysis that is designed to facilitate immediate action to address the threat. IDEHT fuses data from open source digital media sources and from security-related data sources. Information on potential threats as well as their associations are gathered across data sources, creating a single unified view of the threat, presented in the context of their overall network of relationships. Deep machine learning methods are able to learn from user-supplied feedback, allowing IDEHT to continually improve threat detection performance. Threats are identified and assessed by evaluating several different threat factors associated with individuals and networks of individuals (e.g., ideological radicalization, propensity for violence, proximity to installation, etc.). Network analysis, semantic analysis, and intuitive visualizations are used to support EM personnel’s deeper understanding of the threat, so that a rapid and effective response may be carried out. Threats are presented to EM personnel via real-time alerts so that timely response is assured. In Phase I of this project we established initial use cases focusing on lone wolf shooters and population-based riots at a physical installation and used these to drive requirements analysis.  We established the overall system design, including how the various algorithmic pieces fit together and the system architecture.  Our proof of concept evaluation includes the development and application of advanced machine learning based text analysis to the classification of known labeled datasets, proving the effectiveness of this technical approach.  Our proof of concept demonstration shows how threats identified by IDEHT can be visualized in an alert-based dashboard with key findings and supporting evidence available for user review.  We also secured interest from some preliminary transition targets. Based on these strong accomplishments, we have crafted a plan for Phase II continuation. Our overall objectives in Phase II are: (1) Build a prototype IDEHT system; (2) demonstrate its power to detect emerging human-generated physical threats to Army installations by monitoring open source and security-related data sets; (3) integrate with PSIF; and (4) initiate transition to military end users.