Situation awareness using heterogeneous models

Maintaining situation awareness is a fundamental yet challenging task. In many domains, it´s not the lack of data that complicates this task. Instead, the challenge is that relevant information is oftentimes obscured by the shear volume of available data, data that in some cases is incomplete or incorrect. For such domains, it may be difficult to quickly identify and adjudicate relevant information from the volume of available data. Helping analysts sift through the noise to find relevant data and to then help them interpret the implications of these data is one of the goals of the research described in this paper. In this paper, we describe two complimentary modeling approaches aimed at supporting situation awareness. Specifically, normalcy models are used to automatically identify anomalous situations that may merit analyst investigation (for example, detection of unusual airline flight patterns). Suitable for domains with uncertain data, the second modeling approach involves the discovery and use of fuzzy event-based models supporting predictive course-of-action (COA) identification and assessment. Using these models, observation data is automatically fused against COA models, allowing analysts to quickly identify and interpret unfolding complex activities. This paper presents an overview of these two types of models and briefly describes how machine learning technology is used to support rapid development and adaptation of these models.