Artificial neural networks-learning and generalization

Summary form only given. This presentation is intended to address issues that are related to learning and generalization capability of ANN. It is also intended to examine the state-of-the-art and, hopefully, stimulate discussions on where research should be directed. A survey on recent developments in supervised and unsupervised learning is given. Details of both learning strategies are elaborated with regard to some classes of ANN and their applications examined. The concept of selective learning is also discussed. Generalization capability of some classes of ANN is addressed, particularly, from the viewpoint of function realization. Special attention is focused on multilayer perceptrons. Other related questions such as "How large does a network have to be to perform a desired task?" are discussed.