Strategies for learning search control rules: an explanation-based approach
Steven Minton, Jaime Carbonell · International Joint Conference on Artificial Intelligence · 1987
Previous work in explanation-based learning has primarily focused on developing problem solvers that learn by observing solutions. However, learning from solutions is only one strategy for improving performance. This paper describes how the PRODIGY system uses explanation-based specialization to learn from a variety of phenomena, including solutions, failures, and goal-interactions. Explicit target concepts describe these phenomena, and each target concept is associated with a strategy for dynamically improving the performance of the problem solver. Explanations are formulated using a theory describing the domain and the PRODIGY problem solver. Both the target concepts and the theory are declaratively specified and extensible.