ARTIFICIAL NEURAL-NETWORK SIMULATOR WITH INTEGRATED LEARNING SUPERVISION

In recent years, there has been a growing interest in the field of Art ificial Neural Networks (ANNs). But at present, there is no rule or fo rmula that can give adequate ANN design parameters for a given task. T o find these parameters, the developer has to rely on his expertise, o n simulation results and on analysis of the learning behaviour of diff erent ANN configurations. Current ANN simulators offer various tools t o assist the developer in analyzing the state of the ANN during or aft er training. The ANN simulator presented in this paper supervises dire ctly the learning behaviour of the ANN; as the human developer does. I t has the ability to detect critical situations during the training, a nd it gives meaningful results to help guide the developer in making t he proper design choices. The simulator is intended to be used in coll aboration with an expert system that will automatically choose the des ign parameters, in an attempt to automate the design process of ANNS. This article is intended to present this aspect of ANN simulator desig n in particular.