Using Filters to Improve Efficiency of Game-playing Learning Procedures

Reuven Hasson, Shaul Markovitch and Yaron Sella. Using Filters to Improve Efficiency of Game-playing Learning Procedures. In Proceedings of Eleventh International Conference of the Chilean Computer Science Society, 125-137 Santiago, Chile, 1991.

Abstract

In this paper we describe a learning system which acquires knowledge in an attempt to improve the performance of a problem solver that plays the game of fives. We present experimental results indicating that the performance of the problem solver is improved at the beginning, but degrades afterwards down to almost its initial level. We introduce a selection mechanism that tests the system performance with and without the acquired knowledge and allows its addition to the knowledge base only if it proves to be beneficial. We introduce additional selection mechanisms to reduce the number of tests that need to be performed by the above filter and save learning resources. Experiments show that the addition of the filters eliminates the deterioration in performance and improves the learning outcome significantly.

Keywords: Learning in Games, Utility Problem, Active Learning, Selective Learning

Secondary Keywords:

Online version:

Bibtex entry:

 @inproceedings{Hasson:1991:UFI,
  Author = {Reuven Hasson and Shaul Markovitch and Yaron Sella},
  Title = {Using Filters to Improve Efficiency of Game-playing Learning Procedures},
  Year = {1991},
  Booktitle = {Proceedings of Eleventh International Conference of the Chilean Computer Science Society},
  Pages = {125--137},
  Address = {Santiago, Chile},
  Url = {http://www.cs.technion.ac.il/~shaulm/papers/pdf/Hasson-Markovitch-Sella-1991.pdf},
  Keywords = {Learning in Games, Utility Problem, Active Learning, Selective Learning},
  Secondary-keywords = {Adversary Search},
  Abstract = {
    In this paper we describe a learning system which acquires
    knowledge in an attempt to improve the performance of a problem
    solver that plays the game of fives. We present experimental
    results indicating that the performance of the problem solver is
    improved at the beginning, but degrades afterwards down to almost
    its initial level. We introduce a selection mechanism that tests
    the system performance with and without the acquired knowledge and
    allows its addition to the knowledge base only if it proves to be
    beneficial. We introduce additional selection mechanisms to reduce
    the number of tests that need to be performed by the above filter
    and save learning resources. Experiments show that the addition of
    the filters eliminates the deterioration in performance and
    improves the learning outcome significantly.
  }

  }