Inverting resolution with conceptual graphs

Abstract

Methods for performing inductive inference have become very important in Artificial Intelligence, especially in the area of Machine Learning. One technique capable of performing induction is based on inverting the resolution process (with the help of an oracle). This is known as inverse resolution.

In this paper we investigate how inverse resolution can be performed using conceptual graphs. This is done by showing how the individual inverse resolution operators can be implemented using conceptual graphs. We show that the processes involved can actually be viewed as inverses of beta (and alpha) rules. Also, the operations can be seen as analogues to inverse resolution operators suggested, in the literature, for predicate calculus (e.g., absorption, identification, etc.).

The advantage of this approach is that it develops a technique for performing induction using conceptual graphs. In particular, two of the operators are capable of performing constructive induction through the introduction of new relations not present in the original graphs. We also claim that the use of conceptual graphs provides a natural way of performing these operations and that this leads to a better understanding of the processes involved.