Towards Generating Essence Kernels Using Genetic Algorithms

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

@Article{Sedano:2015:PCS,
  author =       "Todd Sedano and Cecile Peraire and Jason Lohn",
  title =        "Towards Generating Essence Kernels Using Genetic
                 Algorithms",
  journal =      "Procedia Computer Science",
  volume =       "62",
  pages =        "55--64",
  year =         "2015",
  note =         "Proceedings of the 2015 International Conference on
                 Soft Computing and Software Engineering (SCSE'15)",
  ISSN =         "1877-0509",
  DOI =          "doi:10.1016/j.procs.2015.08.410",
  URL =          "http://www.sciencedirect.com/science/article/pii/S1877050915025454",
  abstract =     "The Software Engineering Method and Theory (SEMAT)
                 community created the Essence kernel as a unifying
                 framework for describing and analysing software
                 engineering endeavours. The Essence kernel is based
                 upon human experience and judgement, not empirical
                 data. Background At Carnegie Mellon University in
                 Silicon Valley, we have collected data from masters of
                 science in software engineering students as they
                 complete a team-based project course as their capstone
                 or practicum project using the Essence kernel. Each
                 week, the team recorded their progress in an Essence
                 Reflection meeting. This data serves as training data
                 for evaluating the Essence kernel and alternative
                 candidate kernels. Objective Generate candidate
                 replacement kernels by using a fitness function based
                 on empirical data. Method Using genetic programming,
                 the kernel genotype is represented as a collection of
                 linear state machines each with a collection of unique
                 check-list items. Operations to evolve the genotypes
                 include randomly moving check list items, splitting
                 states, and deleting states by moving their checklist
                 items to other states. Results Genetic programming
                 created random candidate essence kernels that scored
                 higher fitness scores than the original essence kernel.
                 The purpose of this exploratory work is to demonstrate
                 one way to generate a candidate Essence kernel directly
                 from empirical data, not to recommend a replacement for
                 the original Essence kernel. Reducing the Essence
                 kernel from seven alphas to one alpha results in higher
                 fitness scores. Limitations Given the limited amount of
                 data, the generated kernels may be over-optimized.
                 Additional empirical data is required before
                 recommending replacing the original kernel with a
                 candidate kernel that fits the data. Conclusion The
                 original Essence kernel is highly structured around
                 human notions of order. Genetic algorithms can generate
                 candidate kernels that humans might not normally
                 consider. Based on the analysis of the fitness
                 function, a kernel with a fundamentally different
                 structure might more effectively recommend next steps
                 for a team during Essence Reflection Meetings.",
  keywords =     "genetic algorithms, genetic programming, Essence
                 Kernel, Empirical Research",
}

Genetic Programming entries for Todd Sedano Cecile Peraire Jason Lohn

Citations