The Importance of semantic context in tree based GP and its application in defining a less destructive, context aware crossover for GP

Created by W.Langdon from gp-bibliography.bib Revision:1.3949

@PhdThesis{Majeed:thesis,
  author =       "Hammad Majeed",
  title =        "The Importance of semantic context in tree based GP
                 and its application in defining a less destructive,
                 context aware crossover for GP",
  school =       "University of Limerick",
  year =         "2007",
  address =      "Ireland",
  month =        "20 " # nov,
  URL =          "http://www.cs.ucl.ac.uk/staff/W.Langdon/ftp/papers/Majeed_thesis.pdf",
  URL =          "https://docs.google.com/file/d/0B1TtmH1V-wKmNFJKRFRDRkNPQzQ/edit?usp=sharing",
  keywords =     "genetic algorithms, genetic programming, context aware
                 crossover, destructive crossover",
  abstract =     "This thesis gives an empirical proof of the existence
                 of competitive building blocks in Grammatical Evolution
                 (GE), a grammar based program evolving algorithm. It
                 shows that in GE, rooted and non-rooted building blocks
                 exist and over the period of time rooted building
                 blocks compete with each other to grow in size, while
                 non-rooted building blocks help them to accomplish
                 that. This is an offline study and done in a
                 retrospective manner.

                 We also present a comprehensive study of the importance
                 of semantic context of a sub-tree in tree based systems
                 and introduce a novel context aware evaluation
                 technique for evaluating sub-trees in context. The
                 usefulness of this technique is demonstrated on a
                 benchmark problem.

                 In this work, we introduce a new constructive and
                 context aware crossover for GP, Context-Aware
                 crossover, which works by placing the selected sub-
                 trees in their best possible context in any tree. This
                 is a greedy approach and results in an improved
                 performance. It is tested on a wide range of problems
                 and showed better performance on all the problems
                 except the Uni-Variate and Bi-Variate Polynomial
                 Symbolic Regression problems. Furthermore, the results
                 show that it generates very compact form of trees
                 without adversely affecting their fitness.

                 Finally, we show the usefulness of the context aware
                 evaluation technique in encapsulating useful trees at
                 the end of a run and using them to create a module
                 repository. This repository is later used to improve
                 the performance of the second cascaded run. For the
                 second run, a variant of context-aware crossover is
                 introduced, Context-Aware mutation which works on
                 module repository. The effectiveness of this setup is
                 demonstrated by re-examining a real world blood flow
                 problem and improving the previously published
                 results.",
  notes =        "Supervisor Conor Ryan",
}

Genetic Programming entries for Hammad Majeed

Citations