Two-valued neural logic network

ABSTRACT We have introduced Neural Logic Networks that uses an ordered pair of numbers as its activation [Hsu 90],[Chan 89JjTeh 89]. The advantage is that it can deal withsituations in which a proposition can be TRUE, FALSE or UNKNOWN. The disadvantage is that computation is much more time consumming than networks whoseactivation consists of a single value. For situations where Boolean logic is sufficient, weintroduce the network described in this paper.Keywords: Boolean Neural-Logic Network, connectionists expert systems. 1. IntroductionIt has been shown that Artificial Neural Networks can be successfully used to perform logical reasoning, with and without probabilistic considerations [Hsu 90],[Chan 89],[Teh 89]. These papers proposed the use of three-value logic [Turner 84]. To cater for the complex structure of the three value logic which can have 39 different binary operators, the neural network had to be generalized so that both theactivation of the neurons and the connecting weights are not represented by simple