Example Project Topics

• Application style projects:
  • Premise: Given the source code for recent model of GA/ GP establish the usefulness of the model in a different application context.
    • Do the properties that were originally reported as providing uniqueness or specific advantages over other machine learning domains still demonstrated?
    • How does the `fancy' GA/ GP model compare with other more `classical' examples of Machine Learning.
  • Specific examples:
    • Does the Symbiotic Bid-based model of GP provide unique opportunities under specialist domains?
    • How relevant are instances of Pareto based multi-objective optimization outside of the benchmarking data sets typically employed for establishing model superiority.
• Engineering style projects: Graphics Processing Units have attracted a lot of interest recently with their potential utility as a speedup for GP fitness evaluation. There are potential caveats however.
  • Specific examples:
    • What significance does the size of the data partition have on performance? GPU communication is through the PCI bus. What happens when all the data cannot be transferred to the GPU host for evaluation in a single step?
    • Does supporting subsampling of the training partition during evolution result in CPU performance catching up with GPU performance or conversely, does supporting subsampling result in better GPU performance?
  • Note:
    • Several open source GPU implementations are available. It is suggested that you begin with one of these and then make modifications. Specific examples including code from Bill Langdon or Denis Robilliard.
    • A summary of recent publications in the GPGPU field may provide additional sources.
    • Depending on the source code, expect to see requirements for specific (open source) development environments and potentially manufacturer specific GPU requirements.
• Verify style projects: Given a variant of canonical GA/ GP reverse engineer the proposed variant under a different GP representation and perform a through assessment under different problem domains.
  • Specific examples:
    • Code bloat and Selection Operators
      • Selection and search operators interact, resulting in desirable and undesirable traits
      • One proposal for attempting to minimize the code bloat effect has been to encourage children to behave differently from the parents.
      • Project would involve evaluating such a behaviour under one of the following:
        • Linear GP (e.g. modify the page-based code).
        • Tree GP (e.g. modify the lilgp code).
          P.W.H. Smith, K. Harries, Evolutionary Computation, 6(4), 1999.
    • Context sensitive search operators under Grammatical Evolution.
      • Crossover and mutation in GE is exceptionally disruptive.
      • CFG used to provide the geno- to phenotype mapping provides information to guide the search operators.
        R. Harper, A. Blair, IEEE Congress on Evolutionary Computation. Vol 1, 2005
    • Scaling GP to larger data sets and the class imbalance problem: Are any of the following models any better/ worse than any others on particular problem domains?
      • Simple sampling model
        Doucette J., Heywood M.I. (2008) GP Classification under Imbalanced Data Sets: Active Sub-sampling and AUC approximation European Conference on Genetic Programming (EuroGP). LNCS 4971
      • Host-Parasite model
        • McIntyre A., Heywood M.I., On Multi-Class Classification by Way of Niching, GECCO-04, Proceedings of the Genetic and Evolutionary Computation Conference, July 2004
      • Evaluation of Pareto based co-evolutionary GA under deceptive datasets.
        E.D. de Jong J.B. Pollack, Evolutionary Computation 12(2), 2004
    • Compact GA - population of 1
      • evaluate performance of Compact GA under a selection of deceptive problems
        G.R. Harik, F.G. Lobo, D.E. Goldberg. IEEE Transactions on Evolutionary Computation. 3(4) 1999
    • Applications of GP to games.
      • Co-evolution of players
      • Support for memory.
        Fogal, Proceedings of the IEEE, Special Issue on Computational Intelligence, 87(9), Sept 1999
        • Special Issue on EC in Games, IEEE Transactions on Evolutionary Computation, 9(6), 2005.
      • Example Evolutionary Gaming Competitions:
        • Ms Pac Man
        • Ant wars
        • Car racing
    Author - Malcolm Heywood