Könik, T., and Laird, J. 2004. Learning Goal Hierarchies from Structured Observations and Expert Annotations. In 14th International Conference on Inductive Logic Programming (ILP-2004), Lecture Notes in AI 3194. Springer

[DOWNLOAD PDF]

Abstract.

We describe a framework for generating agent programs that model expert task performance in complex dynamic domains, using expert behavior observations and goal annotations as the primary source. We map the problem of learning an agent program on to multiple learning problems that can be represented in a “supervised concept learning” setting. The acquired procedural knowledge is partitioned into a hierarchy of goals and it is represented with first order rules. Using an inductive logic programming (ILP) learning component allows us to use structured goal annotations, structured background knowledge and structured behavior observations. We have developed an efficient mechanism for storing and retrieving structured behavior data. We have tested our system using artificially created examples and behavior observation traces generated by AI agents. We evaluate the learned rules by comparing them to hand-coded rules.