Refined Query Inference

Authors: Efim B. Kinber and Thomas Zeugmann^*

Source: “ Analogical and Inductive Inference, International Workshop All '89 Reinhardsbrunn Castle, GDR, October 1-6, 1989, Proceedings,” (K.P. Jantke Ed.), Lecture Notes in Artificial Intelligence 397, pp. 148 - 160, Springer-Verlag 1989.

Abstract. We have examined, in some more detail, inference via queries restricted to a fixed number of mind changes as well as inference via queries where the finally synthesized programs may have anomalies. Thereby we could show that even questions involving only one type of quantifier may aid the learning process in two directions. First, the number of mind changes can be reduced, and second, more function classes become inferrible. On the other hand, the Theorems 6 and 15 show that the learning potential of QIMs does mainly depend on the complexity of the allowed questions measured in the number of involved quantifiers.