Learning Domain Theories via Analogical Transfer

Abstract

Learning domain theories is an important challenge for qualitative reasoning. We describe a method for learning new domain theories by analogy. We use analogies between pairs of problems and worked solutions to create a mapping between the familiar and the new domains, and use this mapping to conjecture general knowledge about the new domain. After some knowledge has been learned about the new domain, another analogy is made between the domain theories themselves providing conjectures about the new domain. An experiment is described where the system learns to solve rotational kinematics problems by analogy with translational kinematics problems, outperforming a version of the system that is incrementally given the correct domain theory.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Jan 01, 2007
Accession Number
ADA470404

Entities

People

  • Ken Forbus
  • Matthew Klenk

Organizations

  • Northwestern University

Tags

Communities of Interest

  • Advanced Electronics
  • Air Platforms

DTIC Thesaurus Topics

  • Analogies
  • Angular Acceleration
  • Artificial Intelligence
  • Automobiles
  • Cognitive Science
  • Cross Domain
  • Dynamics
  • Equations
  • Expert Systems
  • Kinematics
  • Language
  • Learning
  • Mechanics
  • Models
  • Physics
  • Reasoning
  • Textbooks

Readers

  • Artificial Intelligence
  • Computer Vision.