Automatic acquisition of search guiding heuristics

  • Christian Suttner
  • Wolfgang Ertel
Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 449)


Our approach to improve the search of a theorem prover employs empirical knowledge gained from former proofs. A connectionist network is used to learn heuristics which order the choices at nondeterministic branch-points. This is done by estimating their relative chance for leading to a shortest proof. Using the method it was possible to reduce the search effort required by a high speed theorem prover. Several experiments are presented showing the attained improvements.


Automated theorem proving model elimination heuristics evaluation functions features learning connectionism back-propagation 


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. [1]
    W. Bibel. Matings in Matrices. In Comm. of the ACM, 26 1983.Google Scholar
  2. [2]
    K. Bläsius, N. Eisinger, J. Siekmann, G. Smolka, A. Herold, and C. Walther. The Markgraf Karl refutation proof procedure. In Proceedings of the Seventh International Joint Conference on Artificial Intelligence, Vancouver, 1981.Google Scholar
  3. [3]
    W. Ertel. Back-Propagation with Temperature Parameter and Random Pattern Presentation. Techn. Report FKI-122-90, Technische Universität München, 1990.Google Scholar
  4. [4]
    W. Ertel, J. Schumann, and C.B. Suttner. Learning Heuristics for a Theorem Prover using Back Propagation. In Proceedings of the 5. ÖGAI-Conference, Igls, Austria, 1989. Springer.Google Scholar
  5. [5]
    K.F. Lee and S. Mohajan. A Pattern Classification Approach to Evaluation Function Learning. Artificial Intelligence, 36, 1988.Google Scholar
  6. [6]
    R. Letz, S. Bayerl, J. Schumann, and W. Bibel. SETHEO: A High-Performance Theorem Prover. to appear in: Journal of Automated Reasoning.Google Scholar
  7. [7]
    D.W. Loveland. Mechanical theorem-proving by model elimination. JACM, 15(2), 1968.Google Scholar
  8. [8]
    R. Overbeek, J. McCharen, and L. Wos. Complexity and Related Enhancements for Automated Theorem-Proving Programs. Comp. & Maths. with Appls., 2(1-A), 1976.Google Scholar
  9. [9]
    F. J. Pelletier, Seventy-five problems for testing Automated Theorem Provers. Journal of Automated Reasoning, 2:191–216, 1986.Google Scholar
  10. [10]
    D.E. Rumelhart, G.E. Hinton, and R.J. Williams. Learning Internal Representations by Error Propagation. In Parallel Distributed Processing, 1986.Google Scholar
  11. [11]
    A.L. Samuel. Some Studies in Machine Learning Using the Game of Checkers. IBM Journal, 1(3), 1959.Google Scholar
  12. [12]
    A.L. Samuel. Some Studies in Machine Learning Using the Game of Checkers, II. IBM Journal, 11(6), 1967.Google Scholar
  13. [13]
    J.R. Slagle and C.D. Farrell. Experiments in Automatic Learning for a Multipurpose Heuristic Program. Comm. of the ACM, 14(2), 1971.Google Scholar
  14. [14]
    M.E. Stickel. A prolog technology theorem prover. New Generation Computing, 2(4):371–383, 1984.Google Scholar
  15. [15]
    M.E. Stickel. Schubert's steamroller problem: Formulations and solutions. Journal of Automated Reasoning, 2:89–101, 1986.Google Scholar
  16. [16]
    C.B. Suttner. Learning Heuristics for Automated Theorem Proving. Diploma Thesis, Technical University Munich, May 1989.Google Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 1990

Authors and Affiliations

  • Christian Suttner
    • 1
  • Wolfgang Ertel
    • 1
  1. 1.FG Künstliche IntelligenzTechnische UniversitätMünchen 2West-Germany

Personalised recommendations