Incremental Multiple Concept Learning Using Experiments

The learning method presented here is a general data-driven method that learns multiple discriminant disjunctive descriptions incrementally from experiments assuming perfect classifications. New points are selected for classification by the environment based on the current concept descriptions. Unlike previous methods for acquiring concepts, attributes with finite unordered and infinite totally-ordered domains are integrated into a uniform framework in which concept descriptions are not only constrained by negative examples, but, more importantly, by the current descriptions of other the classes.