Computer Architectures for Understanding

We describe a new Bayesian architecture for knowledge-based information exploitation and situation assessment, the Cognitive Bayesian INformation EXploitation (CBNX) architecture. CBNX provides mechanisms for the effective transformation of data into information and the dynamic integration of information into higher level cognitive structures supporting situation assessment and dynamic decision making. CBNX combines three distinct forms of knowledge representation. At the top level, schemas encode cognitive "design patterns", i.e., patterns for the large scale organization and interpretation of information. Domain knowledge is encoded at the middle level into a version of first order probabilistic logic with recursion called Loopy Logic. Finally, this time-independent domain knowledge is dynamically compiled into Situation Specific Bayesian Models at the bottom level. CBNX builds on recent work in frame-based probabilistic models, probabilistic relational models, and probabilistic logic, but extends it with new schema-based meta-representations and algorithms. This architecture provides a powerful set of mechanisms for aggregating data and information into interpretive frames, filtering out data that is contextually irrelevant, and illuminating the potential significance of the relevant information in context. It also allows the opportunity for dynamic data management and dynamically configured displays that clearly exhibit the relationship of new information to situation or threat components.