The acquisition of strategic knowledge
This research focuses on the problem of acquiring strategic knowledge-knowledge used by an agent to decide what action to perform next. Strategic knowledge is especially difficult to acquire from experts by conventional methods, and it is typically implemented with low-level primitives by a knowledge engineer. This dissertation presents a method for partially automating the acquisition of strategic knowledge from experts. The method consists of a representation for strategic knowledge, a technique for eliciting strategy from experts, and a learning procedure for transforming the information elicited from experts into operational and general form. The knowledge representation is formulated as strategy rules that associate strategic situations with equivalence classes of appropriate actions. The elicitation technique is based on a language of justifications with which the expert explains why a knowledge system should have chosen a particular action in a specific strategic situation. The learning procedure generates strategy rules from expert justifications in training cases, and generalizes newly-formed rules using syntactic induction operators. The knowledge acquisition procedure is embodied in an interactive program called ASK, which actively elicits justifications and new terms from the expert and generates operational strategy rules. ASK has been used by physicians to extend the strategic knowledge for a chest pain diagnosis application and by knowledge engineers to build a general strategy for the task of prospective diagnosis. A major conclusion is that expressive power can be traded for acquirability. By restricting the form of the representation of strategic knowledge, ASK can present a comprehensible knowledge elicitation medium to the expert and employ well-understood syntactic generalization operations to learn from the expert's explanations.
- Research Organization:
- Massachusetts Univ., Amherst, MA (USA)
- OSTI ID:
- 6506580
- Resource Relation:
- Other Information: Thesis (Ph. D.)
- Country of Publication:
- United States
- Language:
- English
Similar Records
Apprenticeship learning techniques for knowledge-based systems
The EPRI knowledge acquisition workshop handbook: Final report
Related Subjects
KNOWLEDGE BASE
DATA ACQUISITION
RETRIEVAL SYSTEMS
ARTIFICIAL INTELLIGENCE
COMPUTERIZED SIMULATION
DATA ANALYSIS
EXPERT SYSTEMS
INFORMATION RETRIEVAL
INFORMATION SYSTEMS
SYSTEMS ANALYSIS
SIMULATION
990200* - Mathematics & Computers
990300 - Information Handling