Creating Robust Relation Extract and Anomaly Detect via Probabilistic Logic Based Reasoning and Learning

We consider a three-pronged approach to deep exploration and filtering of text. The first is development of a set of scalable state-of-the-art learning algorithms that are capable of learning generalized probabilistic logic rules from noisy, incomplete data. The second is a data management system that is widely accepted as the state-of-the art for knowledge base construction KBC and is highly scalable. The final direction is the design and adaptation of the scalable management and learning algorithms for the tasks of deep knowledge understanding such as knowledge-based population and anomaly detection. In this report, they organize and present their accomplishments the approaches and their intuitive, theoretical and empirical ramifications from the DEFT cooperative agreement into 3 main focus areas or research thrusts. Each of them is motivated and introduced separately in their respective sections.Descriptors: