Genetic Algorithms - IEEE Xplore

2 downloads 0 Views 163KB Size Report
Hans-Paul Schwefel ... place in Anchorage, Alaska, in 1998. ... Algorithms (EA) and methodologies at different places and independently from each other:.
Special Track on Computational Intelligence -Genetic Algorithms Hans-Paul Schwefel University of Dortmund, Dept. of Computer Science 4422 1 Dortmund, Germany Phone+492319700952 schwefel @LS11.informatik.uni-dortmund.de Evolutionary Computation, though existing since the early Sixties, has become part of the method set called Computational Intelligence (CI), a term that has become popular since the first World Congress on Computational Intelligence (WCCI) was held in Orlando, Florida, in 1994. (The second WCCI has already been announced to take place in Anchorage, Alaska, in 1998.) The other two of three CI paradigms are Neural and Fuzzy Computation, and for all three of them there exist or are planned handbooks of substantial size that best document the current state-of-the-fields [ 1-31.

Since the sixties there have been three still distinguishable subgroups that created their own Evolutionary Algorithms (EA) and methodologies at different places and independently from each other:

0

Evolutionary Programming (EP) [4-71, Genetic Algorithms (GA) [8-111, Evolution Strategies (ES) [12-151.

The most widespread, though not oldest, EA are the GA, created in Ann Arbor, Michigan, and proposed by John Holland as a model for studying adaptive processes at the level of genomes, represented by bitstrings, on which the genetic operators recombination (crossover), mutation, and selection work. Mutations are rare random bit flips, crossover means mixing two parental bitstrings by choosing a few random crossover sites to create an offspring, and (mating) selection lets better parents have more children, on average. Holland’s scholars later on used that tool mainly for solving combinatorial optimization problems, and since 1985 the biennial GA conferences have spread this seed around the world. EP, the oldest EA branch, was created in San Diego, California, and models the evolutionary process at the level of species. Lawrence Fogel devised this kind of EA to evolve finite automata able to predict sequences of symbols by learning from previously received time series. There were several kinds of mutations changing the size and connections of the automata, by which as many offspring were created as there were ancestors. Both of them went into a series of tournaments, and those who scored best became parents of the next generation. Recombination was and still is not incorporated since by definition species do not mix. Current EP versions, according to David Fogel, are in many aspects similar to ES - except for the selection scheme and the missing recombination. Since 1992, the EP group has held its own series of annual conferences, mostly in San Diego. The ES concept due to Rechenberg and Schwefel was created in Berlin as a set of rules for experimental optimization with discrete variables. Later, the method was transposed to numerical optimization, especially for finding best real valued parameters within simulation models. The evolving units are individuals represented by a vector of (phenotypic) object variables and additional parameters controlling the mutability of the object variables. There are different forms of recombination, like averaging two parental characters (intermediary recombination) or randomly drawing specific characters out of the whole gene pool (discrete recombination) to create offspring. Both mutation and recombination also work on the set of strategy parameters, i.e., the variances and covariances of the mutation operator, thus enabling on-line self-adaptation of some kind of “internal model” of the environmental conditions. Mating is strictly random so that parents have the same number of offspring, on average. Different from GA and EP, a surplus of offspring is created, and (environmental) selection deterministically cuts down that surplus so that only the best individuals become parents of the next generation. In 1990 a series of biennial conferences called Parallel Problem Solving from Nature (PPSN) [16,17] was set up with a broader scope but strong focus on Evolutionary Computation, including GA, EP, and ES. Since that time the common denominators Evolutionary Algorithms (EA) and Evolutionary Computation (EC) have become internationally agreed upon terms.

622

References E. Fiesler and R. Beale (eds.), Handbook of Neural Computation, Institute of Physics Publ. and Oxford University Press, New York, 1996. T.Black, D.B. Fogel, and Z. Michalewicz (eds.), Handbook of Evolutionary Computation, Institute of Physics Publ. and Oxford University Press, New York, 1997. E. Ruspini, P. Bonissone, and W. Pedrycz (eds.), Handbook of Fuzzy Computation, Institute of Physics Publ. and Oxford University Press, New York, 1997 (forthcoming). L.J. Fogel, A. J. Owens, and M.J. Walsh, Artificial Intelligence through Simulated Evolution, Wiley, New York, 1966. D.B. Fogel, Evolutionary Computation-Toward Piscataway NJ, 1995.

a New Philosophy of Machine Intelligence, IEEE Press,

D.B. Fogel and J.W. Atmar (eds.), Proc. First Annual Conf. Evolutionary Programming, San Diego CA, February 21-22, 1992, Evolutionary Programming Society, La Jolla CA, 1992. J.R. McDonnell, R.G. Reynolds, and L.J. Fogel (eds.), Evolutionary Programming IV, Proc. Fourth Annual Conf. Evolutionary Programming, San Diego CA, March 1-3, 1995, The MIT Press, Cambridge MA, 1995. J.H. Holland, Adaptation in Natural and Artificial Systems, University of Michigan Press, Ann Arbor MI, 1975, 2nd edition: The MIT Press, Cambridge MA, 1992. J. Koza, Genetic Programming, The MIT Press, Cambridge MA, 1992. J.J. Grefenstette (ed.), Proc. First Int’l Conf. Genetic Algorithms, Carnegie-Mellon University, Pittsburgh PA, July 24-26, 1985, Lawrence Erlbaum, Hillsdale NJ, 1985. L.J. Eshelman (ed.), Proc. Sixth Int’l Conf. Genetic Algorithms, University of Pittsburgh PA, July 15-19, 1995, Morgan Kaufmann, San Francisco CA, 1995. I. Rechenberg, Evolutionsstrategidptimierung technischer Systeme nach Prinzipien der biologischen Evolution, Frommann-Holzboog, Stuttgart, 1973. I. Rechenberg, Evolutionsstrategie’94, Frommann-Holzboog, Stuttgart, 1994. H.-P. Schwefel, Numerische Optimierung von Computer-Modellen mittels der Evolutionsstrategie, Interdisciplinary Systems Research Vol. 26, Birkhauser, Basel, Switzerland, 1977, English translation: Numerical Optimization of Computer Models, Wiley, Chichester, 1981. H.-P. Schwefel, Evolution and Optimum Seeking, Sixth-Generation Computer Technology Series, Wiley Interscience, New York, 1995. H.-P. Schwefel and R. Manner (eds.), Parallel Problem Solving from Nature, Proc. PPSN I Workshop, Dortmund, October 1-3, 1990, Lecture Notes in Computer Science Vol. 496, Springer, Berlin, 1991. H.-M. Voigt, W. Ebeling, I. Rechenberg, and H.-P. Schwefel (eds.), Parallel Problem Solving from Nature PPSN V, Proc. of the Int’l Conf. Evolutionary Computation, Berlin, September 22-26, 1996, Lecture Notes in Computer Science Vol. 1141, Springer, Berlin, 1996.

623