CIMA - AN INTERACTIVE CONCEPT-LEARNING SYSTEM FOR END-USER APPLICATIONS

Personalizable software agents will learn new tasks from their users. In many cases the most appropriate way for users to teach is to demons trate examples. Learning complex concepts from examples alone is hard, but agents can exploit other forms of instruction that users might gi ve, ranging from yes/no responses to ambiguous, incomplete hints. Agen ts can also exploit background knowledge customized for applications s uch as drawing, word processing, and form filling. The Cima system lea rns generalized rules for classifying, generating, and modifying data, given examples, hints, and background knowledge. It copes with the am biguity of user instructions by combining evidence from these sources. A dynamic bias manager generates candidate features (attribute values , functions, or relations)from which the learning algorithm selects re levant ones and forms appropriate rules. When tested on dialogs observ ed in a prior user study on a simulated interface agent, the system ac hieved 95% of the learning efficiency observed in that study.