Automated knowledge acquisition using unsupervised learning

Previously developed methods for automated knowledge acquisition are based on decision trees, progressive rule generation and supervised neural networks. In some real world situations, supervised learning is not possible. Previous methods are not applicable in these situations. A method, based on neural networks, is presented which learns symbolic knowledge representations using unsupervised learning. It is illustrated that symbolic knowledge extraction can be successfully performed using unsupervised neural networks, where no target output vectors are available to the automated knowledge acquisition technique during training