An Integrated Framework for Empirical Discovery

  • Authors:
  • Bernd Nordhausen;Pat Langley

  • Affiliations:
  • Institute of Systems Science, National University of Singapore, Kent Ridge, Singapore 0511. BERND@ISS.NUS.SG;Learning Systems Department, Siemens Corporate Research, 755 College Road East, Princeton, NJ 08540 USA. LANGLEY@LEARNING.SCR.SIEMENS.COM

  • Venue:
  • Machine Learning
  • Year:
  • 1993

Quantified Score

Hi-index 0.00

Visualization

Abstract

In this article we present a framework that integrates three aspects of empirical discovery—the formation of taxonomies, the generation of qualitative laws, and the detection of numeric relations. We specify a control structure that integrates these component processes, embedding qualitative discovery within taxonomy formation, and embedding numeric discovery within both of these activities. We also describe the framework's basic representation and organization of knowledge, which combines elements from recent work in machine discovery and qualitative physics. In addition, we describe IDS, a running system that instantiates this framework, and report its behavior on problems from the history of science. Finally, we discuss some limitations of the system as revealed by experimental studies, and propose some directions for future research.