Neural three-valued-logic networks

Abstract

Summary form only given. The authors propose a novel neural network called neural-logic network which attempts to incorporate both the pattern processing and the logical reasoning capabilities within a single frame of neural network environment. This class of neural networks can model classical 2-valued Boolean logic effectively. In addition, it is an ideal model to study 'human logic' which is multivalued, fuzzy, and biased. A neural application of neural-logic networks is the study of connectionist expert systems.