On learning in the presence of unspecified attribute values

Abstract

Presented is the relevant variable (RV) oracle that takes as input a subcube of {0, 1}n and returns a relevant variable of the target on this subcube, if one exists. It is then shown that a class is query learnable using MQs and RV. Next a lower bound is given for the number of UAV-MQs. Furthermore, it is shown that decision trees are learnable using UAV-MQs and UAV-EQs.