Social Programming: Investigations in Grammatical Swarm

Created by W.Langdon from gp-bibliography.bib Revision:1.4504

  title =        "Social Programming: Investigations in Grammatical
  author =       "Finbar Leahy",
  school =       "University of Limerick",
  year =         "2005",
  type =         "Master of Science in Software Engineering",
  address =      "University of Limerick, Ireland",
  month =        "16 " # oct,
  keywords =     "genetic algorithms, genetic programming, grammatical
                 evolution, grammatical swarm",
  URL =          "",
  size =         "129 pages",
  language =     "en",
  abstract =     "This study details a series of investigations
                 examining a recently introduced form of automatic
                 programming called Social Programming. The Grammatical
                 Swarm algorithm is a form of Social Programming as it
                 uses Particle Swarm Optimisation, a social swarm
                 algorithm, for the automatic construction of computer
                 programs for the optimisation of continuous, non-linear
                 problems. An investigation into the performance effects
                 of two different quality Pseudo-Random Number
                 Generators (PRNG) on the Grammatical Swarm algorithm
                 was examined. The results demonstrate that the choice
                 of PRNG does, in fact, have a small effect of the
                 performance of the Grammatical Swarm, with the more
                 sophisticated PRNG producing better results on two of
                 the four problems analysed.

                 An investigations was conducted into the effects of
                 increasing the size of the particle representations of
                 the Grammatical Swarm algorithm, such that the
                 hard-length vector constraint of all particles in the
                 swarm was doubled from 100 to 200. The results
                 demonstrated that this leads to a significant gain in

                 This thesis also introduces a new variable-length form
                 of the Grammatical Swarm algorithm. Thus, this can be
                 considered a proof of concept study. It examines the
                 possibility of constructing programs using a particles
                 representations which are variable in length and it is
                 referred to as the Variable-Length Grammatical Swarm.
                 This newly developed algorithm extends earlier work on
                 the fixed-length incarnation of Grammatical Swarm,
                 where each individual represents choices of program
                 construction rules, where these rules are specified
                 using a Backus-Naur Form grammar. The results
                 demonstrate that is possible to successfully generate
                 programs programs using a variable-length Particle
                 Swarm Algorithm. This investigation also examines the
                 performance effects of increasing the initialisation
                 size of the variable-length particles. The results
                 demonstrate that the performance of the Variable-Length
                 Grammatical Swarm can be increased by doubling the
                 potential size of the particle representations.
                 Furthermore, the evolution of size in the particle
                 representations is examined. This investigation was
                 conduced in an effort to determine if the the variable-
                 length particles suffered from bloat, which is a common
                 problem in other Evolutionary Algorithms that use
                 variable-length vector representations. No evidence of
                 bloat was found.

                 Based on an overall comparative review of the both the
                 fixed-length and variable-length forms of Grammatical
                 Swarm it is recommended that the simpler fixed-length
                 Grammatical Swarm with particle representation sizes of
                 200 codons in length be adopted.",
  notes =        "Supervisor: Dr. Michael O'Neill",

Genetic Programming entries for Finbar Leahy