Automated Extraction of Expert Domain Knowledge from Genetic Programming Synthesis Results

Created by W.Langdon from gp-bibliography.bib Revision:1.4524

  author =       "Trent McConaghy and Pieter Palmers and 
                 Georges Gielen and Michiel Steyaert",
  title =        "Automated Extraction of Expert Domain Knowledge from
                 Genetic Programming Synthesis Results",
  booktitle =    "Genetic Programming Theory and Practice {VI}",
  year =         "2008",
  editor =       "Rick L. Riolo and Terence Soule and Bill Worzel",
  series =       "Genetic and Evolutionary Computation",
  chapter =      "8",
  pages =        "111--125",
  address =      "Ann Arbor",
  month =        "15-17 " # may,
  publisher =    "Springer",
  keywords =     "genetic algorithms, genetic programming, synthesis,
                 domain knowledge, multi-objective, data mining, analog,
                 integrated circuits, age layered population structure",
  DOI =          "doi:10.1007/978-0-387-87623-8_8",
  URL =          "",
  size =         "14 pages",
  isbn13 =       "978-0-387-87622-1",
  abstract =     "Recent work in genetic programming shows how expert
                 domain knowledge can be input to a genetic programming
                 (GP) synthesis system, to speed it up by orders of
                 magnitude and give trustworthy results. On the flip
                 side, this paper shows how expert domain knowledge can
                 be output from the results of a synthesis run, in forms
                 that are immediately recognisable and transferable for
                 problem domain experts. Specifically, using the
                 application of analog circuit design, this paper
                 presents a methodology to automatically generate a
                 decision tree for navigating from performance
                 specifications to topology choice; a means to extract
                 the relative importances of topology and parameters on
                 performance; and to generate whitebox models that
                 capture tradeoffs among performances. The extraction
                 uses a combination of data-mining and genetic
                 programming technologies. This paper also presents
                 techniques to ensure that the GP-based synthesis system
                 can indeed create a richly-populated, high-performance
                 dataset, including: a parallel-computing,
                 multi-objective age-layered population structure (ALPS)
                 for fast and reliable convergence; average ranking on
                 Pareto fronts (ARF) to handle many objectives; and
                 generating good initial topology sizings via multigate
                 constraint satisfaction. Results are shown on
                 operational amplifier synthesis across thousands of
                 topologies that generated a database containing
                 thousands of Pareto-optimal designs across five
                 objectives and dozens of constraints.",
  notes =        "part of \cite{Riolo:2008:GPTP} published in 2009",

Genetic Programming entries for Trent McConaghy Pieter Palmers Georges G E Gielen Michiel Steyaert