Off-campus UMass Amherst users: To download dissertations, please use the following link to log into our proxy server with your UMass Amherst user name and password.
Non-UMass Amherst users, please click the view more button below to purchase a copy of this dissertation from Proquest.
(Some titles may also be available free of charge in our Open Access Dissertation Collection, so please check there first.)
Learning plan schemas from cases
Abstract
This thesis addresses the problem of learning in the context of case-based reasoning (CBR). More specifically, we examine the limitations of and possible uses for generalization in the process of solving problems using cases. In some sense, every CBR system that acquires new cases dynamically can be said to learn, since the new cases enable the system to improve its performance over time. We usually, however, think of learning as involving some sort of change in representation. Most CBR work that deals with constructing new representations is in the area of indexing, i.e., learning descriptions for cases that enable them to be retrieved in appropriate situations. Our interest in learning includes learning generalized case descriptions, but there are certain techniques for such generalization that we feel are inappropriate. Specifically, there is a body of work that advocates using dynamically generated discrimination networks to index cases. We claim that, if one assumes a parallel retrieval mechanism, it is unnecessary to use explicit generalizations of this sort to index cases. In addition to learning generalized case descriptions, we are interested in generalizations of the cases themselves. We demonstrate that such generalizations can be of help both in the process of completing past cases to fit current situations and in the analysis of failed solution attempts. We present a technique for creating such generalizations that is well suited to case-based problem solving. The technique, called analytic concept creation, can be used in generalizing both cases and case descriptions. In order to carry out our explorations, we have developed a model for case-based problem solving, called RECODER. RECODER is a model that attempts to integrate the various phases of case-based problem solving, including retrieval of relevant cases, completion of past cases to fit current situations, debugging when errors occur, and reorganization (adding new cases and generalizing). The model has been implemented in a system called TA, which writes and debugs small LISP programs.
Subject Area
Computer science|Artificial intelligence
Recommended Citation
Williams, Robert Stuart, "Learning plan schemas from cases" (1990). Doctoral Dissertations Available from Proquest. AAI9110231.
https://scholarworks.umass.edu/dissertations/AAI9110231