Query learning of bounded-width OBDDs

Abstract

In this paper, we study exact learnability of bounded-width ordered binary decision diagrams (OBDDs), when no ordering of the variables is given and learning is conducted by way of equivalence queries and membership queries. We present a learning algorithm for width-2 OBDDs, an algorithm which uses O(n3) equivalence queries alone, where n is the number of variables. We also present a learning algorithm for width-2 OBDDs which uses O(n) proper equivalence queries and O(n+) membership queries. Further, we show a negative result: that there are no polynomial-time algorithms capable of learning width-3 OBDDs from proper equivalence queries alone.