| Abstract: | The Learning method of inductive logic programming evaluates the the quality of inductive hypotheses commonlyby how well the hypotheses cover training examples extensionally. Ths kind of evalution will be completely inapplica-ble to the situation where there are non-simple or noise examples. Based on Limited Negative semantics, This paperextends induction of simple examples to that of generalized ones and presents 3 criteria on accepting inductive hypothe-ses in order to fit for different goals of induction. The relation between inductive hypotheses and background knowledgeis also established in this Paper. The conclusions in this paper are helpful to implement learning algorithms and to ex-tend application fields of ILP. |
