Abstract

Current Explanation-Based Learning (EBL) systems take a training example in a non-operational form and transform it into a general operational concept which describes the goal to be achieved. However, the form in which knowledge is embedded in such learned concepts may restrict them from being usefully applied to later situations. This restriction arises from three possible causes: the learned concept's knowledge is too specific, the learned concept lacks domain knowledge lost in the generalization process, the learned concept is unable to adapt to differences found in similar situations encountered later. This results in the learned concept often not being useful in complex domains. Case-Based Learning (CBL) systems retain all the specific knowledge in the form of training examples. This restrains CBL's deduction capabilities, as the mapping of knowledge from a pre-stored example to a current situation limits the flexibility to explain slightly different cases. Pre-stored training exa...

Citations