A fast method for rule extraction in neural networks

Most methods for finding regularities or rule extraction based data start out from the idea that the representation of the data must evolve from a distributed representation of the information to a more localised representation which will represent the skeleton of the network. This idea involves the problem of needing long training times imposed by the back propagation algorithm, as well as the errors deriving from incorrect elimination of connections and units in order to extract the structure of the network. The method proposed here, which I call Direct Method for Structural Learning, allows the appropriate learning and pruning to be achieved in a very short time due to the fact that it starts with a non-skeletal local representation of the network.