A general explanation and interrogation system for neural networks

Summary form only given, as follows. The information in a trained neural network is stored as numerical weights in the neural elements and the connectivity pattern of the network. For many applications, it is desirable to have this neural network information converted into symbolic knowledge form for communication with human or machine experts. Techniques are presented for converting the information in a trained network into symbolic form as a set of rules and for obtaining explanations from the network for specific inputs. These two techniques provide the neurocomputer with one advantage of expert systems while retaining the learning and generalization capability of the neural network.<>