This research continues work on knowledge-intensive systems for machine learning. The approach taken, Explanation Based Learning, is expanded from small "clean" problems here into more complex domains. In such complex domains, an automated learning system will inevitably acquire incorrect concepts. Inference will not be logically complete, but only plausible. Research is targeted toward detecting and refining such concepts when they cause performance failures. The significance of this work is that it expands machine learning capabilities attack more realistic problems, and offers a promising avenue of solution to the problem of knowledge acquisition in expert systems, since an EBL system derives its own "rules" rather than needing them to be specified and programmed by a human intermediary.
