Most machine learning is focused on inductive generalization from
empirical data and does not explicitly exploit prior knowledge of the
domain. Explanation-based learning is a radically different approach
that uses existing declarative domain knowledge to "explain"
individual examples and uses this explanation to drive a
knowledge-based generalization of the example. It is therefore
capable of learning a very general concept from only a single training
example. Our work was some of the original research on this approach
and lead to our subsequent work on
theory
refinement and on
learning for planning and problem-solving.