A Cognitive Architecture Focusing on Situation Estimation: Merging Perceptual and Cognitive Processes

摘要

Situation estimation is a key capability for all high-level animals that requires feedback between sensory and cognitive processing. The hard, interacting problems that must be solved in estimating the situation in an image of a scene are the same problems that must be solved when doing intelligence analysis, and monitoring tasks. Recognition of objects or situations in incomplete, noisy data is known to be an inverse problem. This implies that there is no general way to determine whether a pixel in an image is part of an object without a model of the object. The different types of models needed to perform situation estimation are discussed. There are computational advantages and psychological evidence for organizing these models into many local contexts, called Mental Spaces. Evidence is provided to show that the situation estimation process involves cycles of situation update and revision. We have developed a cognitive model to explain how humans organize and use their models while performing cycles of situation estimation while addressing perceptual or analytic problems. An example of this architecture has been developed and applied to finding groups within a set of people.