Hierarchical control of production systems
Allan James Terry · 1983
As expert systems grow in size and complexity, they require an increasing amount of guidance in order to solve the assigned task. A larger knowledge base means more pieces of knowledge that apply to more situations, often leading to combinatorial explosion. The problem of how the limited resources of the system should be allocated when there are many plausible choices is called the focus-of-attention problem. We develop a solution to this problem by utilizing the expert's strategic knowledge to guide the system's basic problem-solving activities. The expert's control heuristics are represented (as is the other knowledge in the system) as production rules, and are organized into what we call a hierarchical production system. Control in this architecture proceeds from the top down, through many levels of control down to object-level heuristics at the bottom of the hierarchy. Each level is a complete production system that examines the current situation and invokes one or more sets of rules at the next lower level. As control moves from general to very specific strategies, from a broad to very narrow view of the situation, focus of attention is achieved in a very clear and efficient manner. We demonstrate these ideas in an expert system called CRYSALIS. The domain is X-ray protein crystallography and the task is to interpret a three-dimensional image of the electron density cloud surrounding the molecule. This particular instance of a hierarchical production system is a data-driven system built using the blackboard model. The object-level knowledge sources are controlled by two level of strategy heuristics.