Distributed Genetic Programming Models with Application to Logic Synthesis on FPGAs

Created by W.Langdon from gp-bibliography.bib Revision:1.7630

@PhdThesis{fernandez:thesis,

• author = "Francisco {Fernandez de Vega}",
• title = "Distributed Genetic Programming Models with Application to Logic Synthesis on FPGAs",
• school = "University of Extremadura",
• year = "2001",
• address = "Spain",
• email = "fcofdez@unex.es",
• keywords = "genetic algorithms, genetic programming, reconfigurable hardware, EHW, PADGP, IMGP",
• broken = "http://cum.unex.es/profes/profes/fcofdez/escritorio/investigacion/pgp/thesis/phd.html",
• URL = "http://www.researchgate.net/publication/256474009_Distributed_Genetic_Programming_Models_with_Application_to_Logic_Synthesis_on_FPGAs._PhD._Thesis._2001",
• size = "156 pages",
• notes = "For Spanish version see \cite{fernandez:thesis:espanol}

  CONCLUSIONS AND FINAL REMARKS

  We have presented a new implementation of GP - based on MPI - which allows us to make use of parallelism as well as experimenting with different communication topologies and GP parameters

  We have compared performances of this methodology ?PADGP ? with classic GP. The tool was applied to the study of two important parameters that affect convergence results on PADGP: the number and size of populations. By means of this study, we have observed the existence of a region of effort which defines the best number of individuals we must use when employing a given number of populations with PADGP.

  This region of effort has been detected both in benchmark problems and in ?real life? problems.

  We have also presented random topology as a way of improving convergence when using PADGP.

  We have used PADGP with random topology and compared it to classic GP. This comparison showed that the former gives better results.

  We have also compared random topology and grid topology and we have shown that results are similar. Nevertheless random topology requires a smaller amount of communication processes.

  We have presented a methodology that is based on PADGP, and which aids medical diagnosing. We used this problem to check the validity of results obtained in the benchmark problem, while we also proposed PADGP as an appropriate methodology for extracting medical knowledge.

  We have studied isolated subpopulations (IMGP) as a limit case of PADGP and we have experimentally seen that IMGP obtains similar convergence results than GP; sometimes results are even better if the total number of individuals is high.

  We have then dealt with an optimisation problem: the problem of placement and routing on FPGAs. We have developed a new methodology based on GP, and this allows us to represent circuits by means of GP trees. Furthermore, the methodology achieved the proposed goal: finding several ways of placing and routing circuits on reconfigurable hardware. The problem was later used for checking the conclusions which had been reached in the first part of this research. All statistical results obtained are in agreement with those obtained from benchmark problems.

  We think that the main goals we established at the beginning have been achieved: checking the usefulness of PADGP with random communications and developing a methodology for logic synthesis on FPGAs. In the researching process we discovered the concept of region of effort and we obtained interesting conclusions via the use of IMGP.

  Results we obtained during our research have been published in the main conferences and reviews that deal with the different topics addressed in this thesis (see References).",

• notes = "Gruau embryology. p113 Figure how recessive genes work in crossover between trees. Isolated multi-population genetic programming (IMGP)",

}

Genetic Programming entries for Francisco Fernandez de Vega

Citations