High Performance Evolutionary Computing

Created by W.Langdon from gp-bibliography.bib Revision:1.3973

@InProceedings{Nunez:2006:HPCMP,
  author =       "E. Nunez and E. R. Banks and P. Agarwal and 
                 M. McBride and R. Liedel",
  title =        "High Performance Evolutionary Computing",
  booktitle =    "HPCMP Users Group Conference, 2006",
  year =         "2006",
  pages =        "354--359",
  address =      "Denver, USA",
  month =        jun,
  publisher =    "IEEE",
  keywords =     "genetic algorithms, genetic programming",
  ISBN =         "0-7695-2797-3",
  DOI =          "doi:10.1109/HPCMP-UGC.2006.31",
  abstract =     "Evolutionary computing (EC) comprises a family of
                 global optimisation techniques that start with a random
                 population of potential solutions and then evolve more
                 fit solutions over many generations. To accomplish this
                 increase in fitness, EC uses basic operations like
                 selection, recombination, and mutation. Because of its
                 compute- intensive nature, EC research is an obvious
                 candidate for hosting on HPC clusters or systems. EC
                 requires high performance computers (HPC) because the
                 selection process needs to evaluate the fitness of each
                 member of a population of solutions, so the more fit
                 individuals may propagate their characteristics to the
                 next generation of solutions. This requirement becomes
                 even more acute because the evaluation process must be
                 iterated over a very large number of generations. In
                 this paper, we provide a general overview of EC, its
                 applicability to a broad range of problems. In
                 particular, we focus on some subclasses of EC known as
                 genetic programming (GP), genetic algorithms (GA),
                 hybrids, and other EC forms. This paper also discusses
                 the architectural issues of hosting EC on a HPC
                 cluster, and the related issue of population
                 management. Two possible EC architectures are
                 presented: (1) a single chromosome evaluator that
                 treats a pool of cluster nodes as evaluators for an
                 individual solution, and (2) a parallel evolver that
                 manages a sub-population of solutions at each node.
                 Advantages and disadvantages of each approach will be
                 discussed. EC may be applied to a wide variety of
                 problems. Applications of EC include schedule
                 optimisation, robotic navigation, image
                 enhancement/processing, discrimination of buried
                 unexploded ordnance, discovery of innovative electronic
                 filter and controller designs, lens design
                 optimization, radar response modelling, and many more.
                 EC excels at solving high-dimensional and nonlinear
                 problems. HPC resources have enabled the broader
                 application of EC optimisation techniques. However, at
                 present, EC is underused in the- HPC environment. This
                 paper raises awareness of EC's general applicability
                 and its power when coupled with HPC resources",
  notes =        "INSPEC Accession Number: 9398906

                 USASMDC Adv. Res. Center, COLSA Corp., Huntsville,
                 AL;",
}

Genetic Programming entries for Edwin Nunez Edwin Roger Banks Paul Agarwal Marshall McBride Ronald Liedel

Citations