Investigation of a simulator-trained machine discovery system

An approach is proposed to the problem of discovering, adapting, and maintaining knowledge bases in the absence of experts or human trainers, for domains where large-scale simulators exist. That is, rather than having access to human trainers or a real device, the discovery system has access to a software simulator of a real device or situation. This approach attempts to integrate and extend several known learning algorithms (ID3, bucket brigade, and EURISKO´s), in order to exploit their proven capabilities to the largest extent possible. Two initial investigations and preliminary findings are presented and discussed, along with the rudiments of a theory of parameters