LOGICAL SPECIFICATION OF NEURAL NETWORKS

Motivation: Though neural networks are extensively used to tackle the problems associated with bioinformatics, they are still a “black box ” the outcome of which cannot be interpreted. To solve the problem, a neural network specification should be created that could serve to interpret the result from the standpoint of a problem area. Results: Based on MofN rules proposed by Shavlik, a logical representation of a neural network is constructed with the subsequent creation of neural network specification as a logical program. To realize the possibilities of the specification obtained, comparison of neural networks and semantic probability inference is made that forms the basis of compute