Computer Science · Artificial Intelligence
Structure composing for situation assessment
See on the mapAbstract
Situation assessment is the integration of internal information and external evidence with the aim of understanding a situation of interest. Situation assessment is a major component of almost all intelligent behavior and includes such activities as medical diagnosis, market research, and multi-target tracking. This research aims at exploring concepts and techniques for automating and supporting situation assessment processes with computer-based systems. In particular, the research focuses on solving situation assessment problems which have significant uncertainties by dynamically creating and exercising of probabilistic models. Concepts and techniques from three fields are integrated in this research: artificial intelligence, probabilistic influence diagrams, and multi-target tracking. The result of this research is the Probabilistic Situation Assessment Architecture (PSAA). PSAA provides a representation system for beliefs, and reasoning tools for making inferences with the represented beliefs. The methodology of PSAA originates in estimation theory. Models of a situation of interest are build from general models of behavior through a process called structure composing. Evidence is then used to update the model of the situation and conclusions are derived from the model through probabilistic inference and data association. PSAA consist of four major components: a declarative representation of belief, a probabilistic inference mechanism, a data association methodology and a focus of attention mechanism called structure composing. Concepts and techniques from logic and influence diagrams are integrated to arrive at a graphical, probabilistic belief representation system. The representation system has two basic representations, the model and the situation. Current inference mechanisms for influence diagrams work with symmetric probability distributions. A representation for asymmetric distributions and an inference mechanism for reasoning with these asymmetric distributions is derived. In some situations, evidence from sensors are unassociated (i.e. not labeled as to identity or class) with the particular object or set of objects from which the evidence originates. Before such evidence is processed, hypotheses about the causal source for the evidence must be found. A data association algorithm which works within the PSAA belief representation system is developed. This research has developed a technique called structure composing which supports the ability to control what beliefs are used to represent a situation. (Abstract shortened with permission of author.)