Abstract
Learning by explaining failures and avoiding similar ones thereafter is an attractive way to speed up problem solving. However, previous methods for explanation-based learning from failure can take too long to detect failures, explain them, or test the learned rules. This expense is especially critical for adaptive search, in which control knowledge acquired while solving an individual problem instance must be learned quickly enough to speed up its solution.
We present an adaptive search technique that speeds up state-space search by learning heuristic censors while searching. The censors speed up search by pruning away more and more of the space until a solution is found in the pruned space. Censors are learned by explaining dead ends and other search failures. To learn quickly, the technique overgeneralizes by assuming that certain constraints are preservable, i.e., remain true along at least one solution path. A recovery mechanism detects violations of this assumption and selectively relaxes learned censors. The technique, implemented in an adaptive problem solver named FAILSAFE-2, learns useful heuristics that cannot be learned by other reported methods.
We present experimental evidence that FAILSAFE-2 is effective (learns useful rules, even in recursive domains where PRODIGY and STATIC do not), adaptive (learns fast enough to pay off even within a single problem), and general (speeds up diverse problem solvers, even initially strong ones).
Article PDF
Similar content being viewed by others
Avoid common mistakes on your manuscript.
References
Bhatnagar, N. (1992). On-line learning from search failures. Doctoral dissertation, Rutgers University Computer Science Department, New Brunswick, NJ.
Bhatnagar, N. & Mostow, J. (1990). Adaptive search by explanation-based learning of heuristic censors. In Proceedings of the Eighth National Conference on Artificial Intelligence (AAAI90). Boston, MA: AAAI. (Available as Rutgers AI/Design Project Working Paper Number 169.)
Chase, M., Zweben, M., Piazza, R., Burger, J., Maglio, P., & Hirsh, H. (1989). Approximating learned search control knowledge. In Proceedings of the Sixth International Workshop on Machine Learning. Ithaca, NY: Morgan Kaufmann.
Dejong, G., & Mooney, R. (1986). Explanation-based generalizaton: an alternative view. Machine Learning, 1 (2), 145–176.
Etzioni, O. (1990). Why Prodigy/EBL works. In Proceedings of the Eighth National Conference on Artificial Intelligence (AAAI90). Boston, MA: AAAI.
Etzioni, O. (1990). A structural theory of explanation-based learning. Doctoral dissertation, Carnegie-Mellon University, Pittsburgh, PA.
Etzioni, O. (1991). STATIC: A problem-space compiler for PRODIGY. In Proceedings of the Eighth National Conference on Artificial Intelligence (AAAI90). Boston, MA: AAAI.
Gupta, A. (1987). Explanation-based failure recovery. AAAI87. Seattle, WA: AAAI.
Hammond, K.J. (1986). Learning to anticipate and avoid planning problems through the explanation of failures. AAAI86. Philadelphia, PA: American Association for Artificial Intelligence.
Iba, G.A. (1989). A heuristic approach to the discovery of macro-operators. Machine Learning, 3 (4), 285–317.
Kibler, D. & Morris, P. (1983). Don't be stupid. IJCAI83. Karlsruhe, Germany.
Knoblock, C.A. (1990). Learning abstraction hierarchies for problem solving. In Proceedings of the Eight National Conference on Artificial Intelligence (AAAI90). Boston, MA: AAAI.
Laird, J.E., (August 1988). Recovery from incorrect knowledge in SOAR. In Proceedings of the Seventh National Conference on Artificial Intelligence (AAAI88). St. Paul, MN: AAAI.
Laird, J.E. Rosenbloom, P.S., & Newell, A. (1986). Chunking in Soar: the anatomy of a general learning mechanism. Machine Learning, 1 (1), 11–46.
Laird, J.E., Rosenbloom, P.S., & Newell, A. (1986). Overgeneralization during knowledge compilation in Soar. Workshop on Knowledge Compilation. Oregon State University.
Laird, J.E., Newell, A., & Rosenbloom, P.S. (1987). Soar: An architecture for general intelligence. Artificial Intelligence, 33 (1), 1–64.
Mahadevan, S. (1989). An apprentice-based approach to learning problem-solving knowledge. Doctoral dissertation, Rutgers University Computer Science Department, New Brunswick, NJ. Rutgers Computer Science Technical Report Number ML-TR-30.
Minton, S. (1985). Selectively generalizing plans for problem-solving. Proceedings IJCAI85. Los Angeles, CA.
Minton, S. (1988). Learning effective search control knowledge: An explanation-based approach. Doctoral dissertation, Carnegie-Mellon University Computer Science Department, Pittsburgh, PA.
Minton, S. (1988). Quantitative results concerning the utility of explanation-based learning. In Proceedings of the Seventh National Conference on Artificial Intelligence (AAAI88). St. Paul, MN: AAAI.
Minton, S. (March 1990). Quantitative results concerning the utility of explanation-based learning. Artificial Intelligence, 42 (2–3), 363–391. (Revised version of AAAI88 paper.)
Mitchell, T.M., Keller, R.M., & Kedar-Cabelli, S.T. (1986). Explanation-based generalization: a unifying view. Machine Learning, 1 (1), 47–80.
Mostow, D.J. (1983). Machine transformation of advice into a heuristic search procedure. In J.G. Carbonell, R.S. Michalski, & T.M. Mitchell (Eds.), Machine learning. Palo Alto, CA: Tioga.
Mostow, J. (1985). Toward better models of the design process. AI Magazine, 6 (1), 44–57.
Mostow, J. (1989). Design by derivational analogy: issues in the automated replay of design plans. Artificial Intelligence, 40 (1–3), 119–184. In J. Carbonell (Ed.), Special Volume on Machine Learning. Reprinted as Machine learning: Paradigms and methods, Cambridge, MA: MIT Press, 1990.
Mostow, J. & Bhatnagar, (1987). Failsafe—a floor planner that uses EBG to learn from its failures. In Proceedings of the Tenth International Joint Conference on Artificial Intelligence (IJCA187). Milan, Italy: Morgan Kaufmann.
Mostow, J., & Prieditis, A.E. (1989). Discovering admissible heuristics by abstracting and optimizing: a transformational approach. In Proceedings of the Eleventh Joint International Conference on Artificial Intelligence. Detroit, MI.
Mostow, J., Barley, M., & Weinrich, T. (1989). Automated reuse of design plans, International Journal for Artificial Intelligence in Engineering, 4 (4), 181–196.
Nilsson, N. (1980). Principles of artificial intelligence. Los Altos, CA: Morgan Kaufmann.
Norton, S., & Kelly, K. (1988). Learning preference rules for a VLSI design problem-solver. The 4th IEEE Conference on AI Applications. San Diego, CA: IEEE (Available as Rutgers AI/VLSI Project Working Paper No. 66.)
Prieditis, A. (1990). Discovering admissible heuristics by abstraction and speedup: A transformational approach. Doctoral dissertation, Rutgers University Computer Science Department, New Brunswick, NJ.
Rosenbloom, P.S., & Laird, J.E. (1986). Mapping explanation-base generalization onto Soar. Proceedings AAAI86 Philadelphia, PA: AAAI.
Segre, A., Elkan, C., & Russell, A. (1991). A critical look at experimental evaluations of EBL. Machine Learning, 6 (4).
Siklossy, L., & Dowson, C. (1977). The role of preprocessing in problem solving systems. IJCAI-5. Cambridge, MA.
Smith, Gary. (1992). Statistical reasoning. Boston, MA: Allyn and Bacon, Inc.
Stallman, R., & Sussman, G. (1977). Forward reasoning and dependency-directed backtracking in a system for computer-aided circuit analysis. Artificial Intelligence, 9, 135–196.
Steier, D.M. (1987). Cypress-Soar: A case study in search and learning in algorithm design. In Proceedings IJCAI-87. Milan, Italy.
Sussman, G.J. (1973). A computational model of skill acquisition. Doctoral dissertation, Massachusetts Institute of Technology, Cambridge, MA.
Tong, C. (July 1987). Toward an engineering science of knowledge-based design. International Journal of Artificial Intelligence in Engineering, 2 (3). In special issue on AI in Engineering Design. (Also available as Rutgers AI/VLSI Project Working Paper Number 49.)
Tong, C. (1987). Learning justifiable design evaluation functions: A decomposable, multi-agent learning problem (Rutgers AI/VLSI Project Working Paper No. 48). Rutgers University, New Brunswick, NJ.
Tong, C. (1988). Knowledge-based circuit design. Doctoral dissertation, Stanford University Computer Science Department, Stanford, CA.
Tong, C. (1990). Knowledge-based design as an engineering science: The Rutgers AI/Design Project. In Proceedings of the AI in Engineering Conference. Boston, MA. Invited paper. (Also appears as Rutgers AI/Design Project Working Paper Number 167).
Tong, C. & Franklin, P. (1989). Tuning a knowledge base of refinement rules to create good circuit designs. In Proceedings of the Eleventh International Joint Conference on Artificial Intelligence. Detroit, MI.
Unruh, A., & Rosenbloom, P.S. (1989). Abstraction in problem solving and learning. In Proceedings of the Eleventh Joint International Conference on Artificial Intelligence, Detroit, MI.
Voigt, K., & Tong, C. (1989). Automating the construction of patchers that satisfy global constraints. In Proceedings of the Eleventh International Joint Conference on Artificial Intelligence. Detroit, MI.
Author information
Authors and Affiliations
Rights and permissions
About this article
Cite this article
Bhatnagar, N., Mostow, J. On-Line Learning from Search Failures. Machine Learning 15, 69–117 (1994). https://doi.org/10.1023/A:1022613220324
Issue Date:
DOI: https://doi.org/10.1023/A:1022613220324