Adapting Bagging and Boosting to Learning Classifier Systems

Created by W.Langdon from gp-bibliography.bib Revision:1.4549

  author =       "Yi Liu2 and Will N. Browne and Bing Xue",
  title =        "Adapting Bagging and Boosting to Learning Classifier
  booktitle =    "21st International Conference on the Applications of
                 Evolutionary Computation, EvoIASP 2018",
  year =         "2018",
  editor =       "Stefano Cagnoni and Mengjie Zhang",
  series =       "LNCS",
  volume =       "10784",
  publisher =    "Springer",
  pages =        "405–-420",
  address =      "Parma, Italy",
  month =        "4-6 " # apr,
  organisation = "Species",
  keywords =     "genetic algorithms, genetic programming, Learning
                 classifier systems, Multiple domain learning, Ensemble
  isbn13 =       "978-3-319-77537-1",
  DOI =          "doi:10.1007/978-3-319-77538-8_28",
  abstract =     "Learning Classifier Systems (LCSs) have demonstrated
                 their classification capability by employing a
                 population of polymorphic rules in addressing numerous
                 benchmark problems. However, although the produced
                 solution is often accurate, the alternative ways to
                 represent the data in a single population obscure the
                 underlying patterns of a problem. Moreover, once a
                 population is dominated by over-general rules, the
                 system will sink into the local optimal trap. To grant
                 a problem's patterns more transparency, the redundant
                 rules and optimal rules need to be distinguished.
                 Therefore, the bagging method is introduced to LCSs
                 with the aim to reduce the variance associated with
                 redundant rules. A novel rule reduction method is
                 proposed to reduce the rules' polymorphism in a
                 problem. This is tested with complex binary problems
                 with typical epistatic, over-lapping niches,
                 niche-imbalance, and specific-addiction properties at
                 various scales. The results show the successful
                 highlighting of the patterns for all the tested
                 problems, which have been addressed successfully.
                 Moreover, by combining the boosting method with LCSs,
                 the hybrid system could adjust previously defective
                 solutions such that they now represent the correct
                 classification of data.",
  notes =        "EvoApplications2018 held in conjunction with
                 EuroGP'2018 EvoCOP2018 and EvoMusArt2018

Genetic Programming entries for Yi Liu2 Will N Browne Bing Xue