Autonomy for Seeking, Understanding, and Presenting Information

There is an enormous amount of amount of data available on the Web. The richness of this data has made it a priority to develop intelligent systems that can mine and make sense of data in order to assist analysis in predicting and explaining events around the world. In this project, we propose to develop a scalable system to intelligently assist analysts by autonomously collecting, analyzing and presenting information. To address the challenge of determining what data is worthwhile for the system to collect, and what is most valuable for the analyst to see, we propose an architecture based on decision-theoretic search control. In this paradigm, the system estimates the utility of alternative options, enabling it to make rational choices, when faced with uncertainty, and a complex set of alternatives. Moreover, machine learning algorithms make it possible to estimate the utility of different alternatives based on experience. A second challenge we address is how to organize a complex system that can collect, interpret and mine large amounts of heterogeneous, multimodal data. We propose to employ a joint inference approach that allow components to flexibly communicate their findings, and in effect, work together to produce optimal results.