This paper reports on recent additions to Michalski’s work on machine learning. The new contribution is the use of inference and other techniques to determine relevant features that were not explicitly in the input. These features are used for grouping the objects into categories. There is background information on how this clustering is done. The authors explain the annotated predicated calculus used for representation, the use of goals, a model-driven method, and a data-driven method. These sections give a clear overview of previous work; no background is assumed.

There is nothing particularly original about this work. Stepp and Michalski have applied a lot of AI ideas to the task, but analysis of theoretical and design choices is lacking. While the authors have developed improved cluster analysis tools, they offer nothing of more general interest. Nor is there any demonstration of applicability to any real learning task.