Numerical-logical processing in Neural Networks for the decision support

Artificial Neural Networks (ANN) embed shallow knowledge through learning. Used in diagnosis and decision support, ANN are immediate computational models for effects and causes as from human experience but keep out from the deep knowledge of them. The paper presents a way of embedding logical processing over the numerical ones in ldquoneural logical sitesrdquo for the classical ANN paradigms, then proposes a way of structuring deep knowledge in the network for all types of abduction problems in a unified way, which is compared with similar attempt. The approach may be spread in any diagnosis and decision support applications involving deep and shallow knowledge.