Job Opportunities User Modeling Conventional information systems perform their functions based on hard-coded rules about user needs. Such systems lack an awareness of their users’ changing informational goals. Over several dozen projects, SET team members have developed a user modeling toolkit which can automatically develop models of users’ interests based solely on their information processing activities. Our user Modeling System (UMS) toolkit has been employed as part of several applications, including a context-driven search mechanism that outperforms COTS search engines such as Google. UMS is coded in Java, and provides several Web services. The system has a small memory footprint, performs rapidly, and is adaptable to most personalization applications. Information Integration To be fully aware of the intricacies of a complex situation, decision makers need to have comprehensive and rapid access to a variety of information sources. In reality, sources of interest are stove-piped and data of interest often resides in multiple repositories with no linkage among related items. SET has developed a novel approach to data integration which leverages schema and ontology matching techniques and also gathers insights from users’ querying activity to automatically learn the semantic relationship among the information sources. This approach allows the system to cope with newly-discovered information resources, providing decision makers with a powerful information management tool. Social Network Analysis Analysts currently depend on very simplistic tools beyond paper and pencil to manipulate and reason about social networks. While COTS tools allow analysts to visualize leadership networks and plot social connections, those tools do not provide the means to depict, visualize, and manipulate power relationships in an effective manner. SET is tackling these problems by creating a tool to visualize, manipulate, and reason about the power structures in a social network. One of the tool’s novel features is the ability to automatically create a quantitative model of power structure in social networks from intelligence data. Our technology will allow analysts to rapidly analyze power relationships to deal with rapidly emerging threats. Cyber Security Commanders and CEOs need to know the operational impact of cyber threats as well as cyber-security improvements, particularly the impact on the abilities of their most valuable asset, their users. For example, how does a cyber attack really affect a commander’s ability to complete a mission? Or, how does the latency induced by a cyber-security product affect overall productivity? Current tools and methods for emulating cyber users rely on static scripts of repeated semi-randomized activities with no adaptability to an attack or changing system/network conditions. To substantially improve user-based cyber-security test & evaluation, SET has a toolkit of methods and software that incorporate adaptive users models and improve on balanced emulation of users’ regularity and diversity. SET’s approach scales from modeling one user to modeling thousands of users and has been designed to accommodate the needs of the National Cyber Range. Virtual Environments for Intelligent Agents Learning is a key characteristic of all intelligent behavior. By observing and interacting with its environment, an intelligent agent can acquire a large repertoire of knowledge and skills. However, learning experiences are difficult and costly if they must take place in the physical world. So SET has developed software that allows agents to do their learning in virtual worlds instead. This software, called CASTLE, generates an agent’s sensory inputs and simulates the effects of its motor or muscle outputs so that it can efficiently acquire experiences that are realistically consistent and richly detailed. Because CASTLE also imposes biologically appropriate limits on agent inputs and outputs, it is especially well suited for agents, such as cognitive models, that are meant to replicate animal or human intelligence. CASTLE can populate the virtual world with additional entities by linking to military training and mission rehearsal simulations through industry standard interfaces. CASTLE is written in Java, and freely distributed under an open-source license. Threat Prediction Potential IED Threat System Plus Plus (PITS++) is an adaptive software system that quantitatively predicts the timing and location of IED emplacements by fusing together diverse data sources including geographical, human terrain, and attack-specific information. It combines SET Corporation's IED emplacement prediction algorithm with Lehigh University's case-based reasoning technologies. It is based on our previous PITS work under DTO funding that demonstrated statistically significant performance in the prediction of IED emplacement location. The original PITS system uses reinforcement learning technique to predict IED threats from geographic features. The PITS++ system enables warfighters to better predict the timing and location of IED emplacements. In addition, this system offers opportunities for preventing IED attacks by learning enemy’s preferences for IED attacks.