Implementing Knowledge Database in Neural Networks
Knowledge database coded as triples: Agent, Relationship, Patient may be easily implemented in neural network. In the described program all the arguments of relations are coded in patterns - orthogonal sequences of bits. After giving patterns of Agent and Relationship to the input of the network, there is a pattern of the proper Patient in the output of the network. BAM network was used in the system. The learning rule of this network is so easy, that controlling of the network can be done also by a neural network. In the described problem BAM network obtained better results, when signal was being sent between the layers only one time. The possibility of using Hopfield network to filter noises introduced by the layer network has been discussed.