IARPA: Mercury

§  Development of methods to automatically sample SIGINT data to detect group-level behavior changes in anticipation of, and response to, events of interest.

§  Development of SIGINT data extraction techniques that focus on volume, rather than depth, by identifying shallow features of SIGINT data (keywords, geotags, timestamps, etc.) that correlate with group behavior.

§  Development of multivariate time-series models robust to non-stationary, heteroscedastic, and heterogeneous data to reveal patterns that precede events.

§  Development of innovative uses of statistical methods to fuse combinations of time series for generating probabilistic warnings of events.

§  Development of capabilities that account for geographical variations in dialect, communication preferences, rate, technology adoption and political openness.