Systematic Experimentation with Deductive Learning: Satisficing vs. Optimizing Search

Shaul Markovitch and Irit Rosdeutscher. Systematic Experimentation with Deductive Learning: Satisficing vs. Optimizing Search. In Proceedings of the Knowledge Compilation and Speedup Learning Workshop, Aberdeen, Scotland, 1992.

Abstract

Most of the research conducted in the area of deductive learning is experimental. However, many of the experiments reported are far from being systematic and thorough. There are many parameters that are embedded in the system's architecture and it is not clear how they affect the utility of the learned knowledge. In this paper we describe an attempt to perform systematic experiments in the domain of deductive learning. The part described here explores how the search strategy employed during problem solving and during learning affects the utility of learning process. It was concluded that the utility of deductive learning is negative when the learned knowledge is applied in optimizing search procedure during problem solving, but becomes positive in satisficing search. It was also concluded that for off-line learning it is more beneficial to use optimizing search during the learning process. Knowledge acquired in this method improves both the efficiency and the quality of the problem solving.

Keywords: Macro Learning, Utility Problem

Secondary Keywords:

Online version:

Bibtex entry:

 @inproceedings{Markovitch:1992:SEDa,
  Author = {Shaul Markovitch and Irit Rosdeutscher},
  Title = {Systematic Experimentation with Deductive Learning: Satisficing vs. Optimizing Search},
  Year = {1992},
  Booktitle = {Proceedings of the Knowledge Compilation and Speedup Learning Workshop},
  Address = {Aberdeen, Scotland},
  Url = {http://www.cs.technion.ac.il/~shaulm/papers/pdf/Markovitch-Rosdeutscher-kcsl1992.pdf},
  Keywords = {Macro Learning, Utility Problem},
  Secondary-keywords = {Deductive Learning},
  Abstract = {
    Most of the research conducted in the area of deductive learning
    is experimental. However, many of the experiments reported are far
    from being systematic and thorough. There are many parameters that
    are embedded in the system's architecture and it is not clear how
    they affect the utility of the learned knowledge. In this paper we
    describe an attempt to perform systematic experiments in the
    domain of deductive learning. The part described here explores how
    the search strategy employed during problem solving and during
    learning affects the utility of learning process. It was concluded
    that the utility of deductive learning is negative when the
    learned knowledge is applied in optimizing search procedure during
    problem solving, but becomes positive in satisficing search. It
    was also concluded that for off-line learning it is more
    beneficial to use optimizing search during the learning process.
    Knowledge acquired in this method improves both the efficiency and
    the quality of the problem solving.
  }

  }