Incorporating Common Sense into a Machine Learning System ( Invited Paper )

The knowledge that must be acquired by machine learning systems which try to mimic common sense, as exhibited by humans, is inherently incomplete, redundant or even contradictory. Thus, the main characteristics of common sense is nonmonotonicity, which is introduced by exceptions to general rules, redundancy, which is introduced by continuous belief revisions and ambiguity, which is introduced by conflicting information. Incorporating common sense into artificial intelligent systems plays an important role in their development. In all cases presented in the literature, common sense is investigated within the context of artificial reasoning systems. In this paper, we examine common sense within the context of a machine learning system which, of course, is also capable for reasoning. More specifically, we present how the proposed NRL system can handle nonmonotonicity, redundancy and ambiguity, in a consistent manner. NRL is based on a hybrid machine learning system capable of acquiring symbolic knowledge of a domain, refining it using a set of classified examples along with Connectionist learning techniques and, finally, extracting comprehensible symbolic information. Moreover, we present how NRL can be used in classification, as a data mining task. Copyright c © 20042005 Yang’s Scientific Research Institute, LLC. All rights reserved.