Explanation-Based Generalization: A Unifying View
Machine Learning
| Hi-index | 0.01 |
In general, as building a problem solver, the expert in the problem-solving, not the knowledge-base, can easily give the system the basic-operators that transform the problem state formally. However, the condition-parts of these basic-operators that transform the problem state formally. However, the condition-parts of these basic-operators are usually too general. For example, considering a basic-operator: LHS = RHS -- LHS + A = RHS + B in the solving of equations, this operator is applicable to any equation and A, B are arbitrary formulas. Therefore, without any strategy knowledge that can control the applications of basic-operators, the search space explosively gets large and the problem solver can not solve the problem by itself. The strategy knowledge typically contains problem states in which each basic-operator should be applied. Unfortunately, this strategy knowledge is difficult for the expert to give to the problem solver. In general, an expert can solve the problem given to the problem solver, thus another input that an expert can easily give to a problem solver is a worked example which is the history of the expert's problem solving. Thus, if a learning system which can acquire the strategy knowledge from worked examples and basic-operators is added to the problem solver, the expert will be able to easily build the knowledge base for the problem solver. This method of knowledge acquisition that the learning system obtains, the expertise from expert's behavior is called Learning Apprentice [1] and some applications were reported [1,2].