In this paper, we introduce Artificial Nonmonotonic Neural Networks (ANNNs)
, a kind of hybrid learning systems that are capable of nonmonotonic reason
ing. Nonmonotonic reasoning plays an important role in the development of a
rtificial intelligent systems that try to mimic common sense reasoning, as
exhibited by humans. On the other hand, a hybrid learning system provides a
n explanation capability to trained Neural Networks through acquiring symbo
lic knowledge of a domain, refining it using a set of classified examples a
long with Connectionist learning techniques and, finally, extracting compre
hensible symbolic information. Artificial Nonmonotonic Neural Networks acqu
ire knowledge represented by a multiple inheritance scheme with exceptions,
such as nonmonotonic inheritance networks, and then can extract the refine
d knowledge in the same scheme. The key idea is to use a special cell opera
tion during training in order to preserve the symbolic meaning of the initi
al inheritance scheme. Methods for knowledge initialization, knowledge refi
nement and knowledge extraction are introduced. We, also, prove that these
methods address perfectly the constraints imposed by nonmonotonicity. Final
ly, performance of ANNNs is compared to other well-known hybrid systems, th
rough extensive empirical tests. (C) 2001 Elsevier Science B.V. All rights
reserved.