Carnegie Mellon University
file.pdf (382.49 kB)

Design of representation-changing algorithms

Download (382.49 kB)
journal contribution
posted on 1996-07-01, 00:00 authored by Eugene Fink
We explore methods for improving the performance of AI problem-solvers by automatically changing problem representations. The performance of all problem-solving systems depends crucially on problem representation. The same problem may be easy or difficult to solve depending on the way we describe it. Researchers have designed a variety of learning algorithms that deduce important information from the description of the problem domain and use the deduced information to improve the representation. Examples of these representation improvements include generating abstraction hierarchies, replacing operators with macros, and decomposing complex problems into subproblems. There has, however, been little research on the common principles underlying representation-improving algorithms and the notion of useful representation changes has remained at an informal level. We present preliminary results on a systematic approach to the design of algorithms for automatically improving representations. We identify the main desirable properties of such algorithms, present a framework for formally specifying these properties, and show how to implement a representation-improving algorithm based on the specification of its properties. We illustrate the use of this approach by developing novel algorithms that improve problem representations.


Publisher Statement

©1996 IEEE. Personal use of this material is permitted. However, permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE



Usage metrics


    Ref. manager