Inducing logic programs with genetic algorithms: the Genetic Logic Programming System

Created by W.Langdon from gp-bibliography.bib Revision:1.4216

  author =       "Man Leung Wong and Kwong Sak Leung",
  title =        "Inducing logic programs with genetic algorithms: the
                 Genetic Logic Programming System",
  journal =      "IEEE Expert",
  year =         "1995",
  volume =       "10",
  number =       "5",
  pages =        "68--76",
  month =        oct,
  keywords =     "genetic algorithms, genetic programming, ILP, FOIL,
                 genetic Logic Programming System, evolutionary
                 processing, knowledge representation, learning, logic
                 program induction, knowledge representation, learning
                 (artificial intelligence), logic programming",
  DOI =          "doi:10.1109/64.464935",
  size =         "9 (22) pages",
  abstract =     "Inductive Logic Programming (ILP) integrates the
                 techniques from traditional machine learning and logic
                 programming to construct logic programs from training
                 examples. Most existing systems employ greedy search
                 strategies which may trap the systems in a local
                 maxima. This paper describes a system, called the
                 Genetic Logic Programming System (GLPS), that uses
                 Genetic Algorithms (GA) to search for the best program.
                 This novel framework combines the learning power of GA
                 and knowledge representation power of logic programming
                 to overcome the shortcomings of existing paradigms.

                 A new method is used to represent a logic program as a
                 number of tree structures. This representation
                 facilitates the generation of initial logic programs
                 and other genetic operators. Four applications are used
                 to demonstrate the ability of this approach in inducing
                 various logic programs including the recursive
                 factorial program. Recursive programs are difficult to
                 learn in Genetic Programming (GP). This experiment
                 shows the advantage of Genetic Logic Programming (GLP)
                 over GP.

                 Only a few existing learning systems can handle noisy
                 training examples, by avoiding overfitting the training
                 examples. However, some important patterns will be
                 ignored. The performance of GLPS on learning from noisy
                 examples is evaluated on the chess endgame domain. A
                 systematic method is used to introduce different
                 amounts of noise into the training examples. A detailed
                 comparison with FOIL has been performed and the
                 performance of GLPS is significantly better than that
                 of FOIL by at least 5 percent at the 99.995 percent
                 confidence interval at all noise levels. The largest
                 difference even reaches 24 percent. This encouraging
                 result demonstrates the advantages of our approach over
                 existing ones.",
  notes =        "IEEE Expert Special Track on Evolutionary Programming
                 (P. J. Angeline editor) \cite{angeline:1995:er}


Genetic Programming entries for Man Leung Wong Kwong-Sak Leung