Knowledge intensive empirical learning using multiple levels of background knowledge

The author describes a substructure discovery system, PLAND, that combines empirical learning methods with knowledge-intense learning algorithms. Unlike other systems which combine similarity-difference-based and explanation-based learning techniques at a single level, the PLAND system uses knowledge to direct the learning process on three distinct levels. This multileveled approach to learning allows a system to be more flexible and adaptive to the current learning task than with a single-level approach. An example run of PLAND is presented.<
>