As they gain expertise in problem solving, humans increasingly rely on patterns and spatially-oriented reasoning. The goals of this project are to learn pattern associations and spatially-oriented reasoning, and then to integrate the knowledge to make decisions in conjunction with other high-level reasoning. This research develops an associative visual pattern classifier and algorithms for the automated acquisition of new, spatially-oriented reasoning agents. These components are readily integrated into a multi-agent decision making program for broad classes of related problems. The pre-existing program robustly combines agents with conflicting perspectives and withstands incomplete and erroneous evidence for the early stages of learning. It makes decisions with a hierarchy of modules representing individual, limitedly rational, heuristic agents. Each agent may draw upon a variety of knowledge bases generated and maintained by various learning strategies. The new visual pattern classifier learns meaningful patterns from experience. In addition, domain-specific spatial concepts are generalized from these patterns to create new heuristic agents.
