Active Explanation Reduction: An Approach to the Multiple Explanations Problem.
Abstract
The multiple explanations problem is central to explanation-based learning from imperfect theories. In this paper, we present a new approach called active explanation reduction to deal with this problem. Active explanation reduction involves the purposeful alteration of the world to generate new information. This new information with cause some of the explanations to become inconsistent with reality, thereby eliminating them from further consideration. Active explanation reduction may also be viewed as experiment design. This paper presents a theory of experiment design which is based on the principle of refutation. The theory describes three strategies for designing experiments - elaboration, discrimination and transformation. The theory and an experiment engine - an implementation of the theory - are illustrated using a detailed example which involves constructing explanations from intractable theories. The relation of the multiple explanations problem to the imperfect theory problems is also described. Finally, active explanation reduction is evaluated based on four criteria - completeness, efficiency, tolerance of unavailable data and feasibility. Keywords: Machine learning.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 1988
- Accession Number
- ADA194127
Entities
People
- Gerald F. Dejong
- Shankar A. Rajamoney
Organizations
- University of Illinois Urbana–Champaign