puerta@gollum.UUCP (Angel R. Puerta) (09/06/89)
A few months ago, I posted an inquiry about Machine Learning applications or reseach for user interfaces. Let me first apologize for not having posted a summary, but in truth, 99% of the responses that I got were from people who were interested in getting the summary. So I guess the conclusion is that this subarea has not been explored to any major length. One of the few works on the area that I have come across was published in the Proceedings of CHI 89: "Inducing Programs in a Direct-Manipulation Environment" by David L. Maulsby and Ian H. Witten of the University of Calgary. As a matter of fact, I believe that there is a group working on this problem at U. Calgary. At the risk of making a self-serving offer, I have a couple of publications which are preliminary results of my dissertation work on this area. The abstracts will follow. Since some of you have inquired repeatedly about the summary I can mail a copy of these papers to those interested and hopefully this would serve as a starting point. Please include a regular mail address and specify the papers that you want. I will try to get them to you in a timely manner. Send requests to: Angel R. Puerta puerta@gollum.columbia.ncr.com puerta@ece.scarolina.edu ABSTRACTS ____________________________________________________________________ Annual Conference International Association of Knowledge Engineers (IAKE 89) University of Maryland, College Park, Maryland June 26-28, 1989 MACHINE LEARNING IN USER INTERFACES: OBJECTIVES AND STRATEGIES Angel R. Puerta and Ronald D. Bonnell Center for Machine Intelligence University of South Carolina puerta@ece.scarolina.edu bonnell@ece.scarolina.edu Researchers and user interface developers have expressed for many years the desire to add learning capabilities to the interface so the system itself would be able to modify its knowledge base and improve its performance over time. However, it is not until recent years that advances in several fields, including artificial intelligence and cognitive science, have allowed developers to construct interface architectures which can be expanded to include learning systems. In this paper, the issue of introducing a learning system into a user interface is addressed. In particular, it is determined why a learning system is necessary and what are the changes to the design and behavior of the interface that learning causes. Furthermore, a taxonomy of learning objectives is presented and a criteria for selecting appropriate learning strategies is examined. ________________________________________________________ XV LatinAmerican Conference on Informatics Santiago, Chile July-10-14, 1989 A BLACKBOARD MODEL FOR A SELF-IMPROVING INTELLIGENT USER INTERFACE Angel R. Puerta and Ronald D. Bonnell Center for Machine Intelligence University of South Carolina Columbia, South Carolina puerta@ece.scarolina.edu bonnell@ece.scarolina.edu Multidisciplinary advances encompassing the fields of artificial intelligence, cognitive science, and multiple media hardware are allowing researchers to start developing knowledge-based user interfaces which support multiple channels of communication between man and machine. The highly dynamic nature of the human-computer interaction process dictates that either the interface be designed for very specific tasks and users or that the knowledge base be unmanageably large to cope with all the possible interaction scenarios. Thus, there is a compelling need for systems that are able to modify its knowledge base therefore correcting mistakes and accounting for changes in the user's needs or preferences. This paper proposes and studies a blackboard model of a learning system for an intelligent user interface. The knowledge sources required are identified and the functionality of each subsystem is described emphasizing the self-improving nature of the system. The system uses a multilayer blackboard architecture in which knowledge sources in the lower layer provide interaction knowledge while those in the top layer contain learning knowledge. The top layer modifies the bottom layer to improve the performance of the interface over time. The possible learning objectives under this model are identified and a functional description of how the goals can be reached using the model is provided.