J. Schaffer and L. Eshelman. On Crossover as an Evolutionary Viable Strategy. In R. Belew and L. Booker, editors, Proceedings of the 4th International Conference on Genetic Algorithms, pages 61--68. Morgan Kaufmann, 1991.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... by Holland [6] 11] 12] and subsequently studied by De Jong [13] 14] 15] 16] Goldberg [17] 18] 19] 20] 21] and others such as Davis [22] Eshelman [23] 24] Forrest [25] Grefenstette [26] 27] 28] 29] Koza [30] 31] Mitchell [32] Riolo [33] 34] Schaffer [35] [36], 37] to name only a few, have been originally proposed as a general model of adaptive processes, but by far the largest application of the techniques is in the domain of optimization [15] 16] Since this is true for all three of the main stream algorithms presented in this paper we will ....

J. D. Schaffer and L. J. Eshelman, "On crossover as an evolutionary viable strategy," In Belew and Booker [59], pp. 61--68.

....both sides of the equation we obtain the theorem. The response to selection equation for mutation contains no heritability. Instead there is an offset, defined by the difference of the probabilities of getting better or worse. The importance of s t and f t has been independently discovered by Schaffer Eshelman, 1991). They did not use the difference of the probabilities, but the quotient which they called the safety factor. F = s t f t In order to obtain an empirical law we have to estimate s t and f t . This can be done by using the results of (Muhlenbein, 1991) The estimation requires the average ....

Schaffer, J.D., & Eshelman, L.J.(1991). On crossover as an evolutionary viable strategy. In R. K. Belew & L. Booker (Eds.), Procedings of the Fourth International Conference on Genetic Algorithms, (pp. 61-68), San Mateo: Morgan Kaufmann.

....become more and more similar, causes crossover to become gradually less disruptive during a run. In other words: the gradual change from explorative to exploitive comes about automatically for crossover. A.E. Eiben and C.A. Schippers On evolutionary exploration and exploitation 9 reported in [43] show that mutation and selection are more powerful than previously believed. Nevertheless, it is observed that a GA with highly disruptive crossover outperforms a GA with mutation alone on problems with a low level of interactions between genes. It is also remarked that the power of an operator ....

....Additionally, crossover is useful for maximizing the accumulated payoff, while it can be harmful if optimality is sought. Based on a comparison of EP and GA in [23] it is argued that crossover does not have any competitive advantage over mutation. Critiques on this investigation, for instance in [43], initiated more extensive comparisons of GAs and EP on function optimization problems. The results in [24] show a clear advantage of EP and make the authors conclude that no consistent advantage accrues from representing real valued parameters as binary strings and allowing crossover to do ....

J.D. Schaffer and L.J. Eshelman. On crossover as an evolutionary viable strategy. In R.K. Belew and L.B. Booker, editors, Proceedings of the 4th International Conference on Genetic Algorithms, pages 61--68. Morgan Kaufmann, 1991.

....a GA. Traditionally, recombination has been considered as the primary operator of a GA and thought to be responsible for the generation and propagation of solutions. More recently, there have been many studies on the role played by traditional crossover operators, compared with mutation, in a GA (Schaffer Eshelman 1991, Spears 1993 and Wu, Lindsay Riolo 1997) Crossover operators have also been classified on their usefulness in terms of generating and propagating solutions (Eshelman Schaffer 1995) There are now many different ways of implementing recombination (Spears 1997) Some forms of recombination are ....

Schaffer, J. D. & Eshelman, L. J. (1991) On Crossover as an Evolutionary Viable Strategy. In In R. Belew and L. Booker (eds.), Proceedings of the Fourth International Conference on Genetic Algorithms, 61-68. Morgan Kaufmann.

....a GA. Traditionally, recombination has been considered as the primary operator of a GA and thought to be responsible for the generation and propagation of solutions. More recently, there have been many studies on the role played by traditional crossover operators, compared with mutation, in a GA (Schaffer Eshelman 1991, Spears 1993 and Wu, Lindsay Riolo 1997) Crossover operators have also been classified on their usefulness in terms of generating and propagating solutions (Eshelman Schaffer 1995) There are now many different ways of implementing recombination (Spears 1997) Some forms of recombination are ....

Schaffer, J. D. & Eshelman, L. J. (1991) On Crossover as an Evolutionary Viable Strategy. In In R. Belew and L.

....a GA. Traditionally, recombination has been considered as the primary operator of a GA and thought to be responsible for the generation and propagation of solutions. More recently, there have been many studies on the role played by traditional crossover operators, compared with mutation, in a GA (Schaffer Eshelman 1991, Spears 1993 and Wu, Lindsay Riolo 1997) Crossover operators have also been classified on their usefulness in terms of generating and propagating solutions (Eshelman Schaffer 1995) There are now many different ways of implementing recombination (Spears 1997) Some forms of recombination are ....

Schaffer, J. D. & Eshelman, L. J. (1991) On Crossover as an Evolutionary Viable Strategy. In In R. Belew and L.

....than that observed as the mutation pa 24800 24820 24840 24860 24880 24900 24920 0 0.2 0.4 0.6 0.8 1 Crossover parameter Figure 5: Crossover Probability rameter is varied. In fact it is suggested [SCED89] that the optimum probability for mutation is much more critical than that for crossover. In [SE91] this hypothesis was investigated further and it was found that crossover algorithms evolve much faster than mutation alone but mutation generally finds better solutions than algorithms that utilise only crossover. This is certainly true for the maintenance scheduling problem as seen above. ....

J. D. Schaffer and L. J. Eshelman. On crossover as an evolutionary viable strategy. In Proceedings of the Fourth International Conference on Genetic Algorithms, pages 61--68. Morgan Kaufmann, 1991.

....with MSE being higher than 0.2. However, when the cellular GA was used, none of the simulations was found to be trapped in these local optima. Therefore, it is more appropriate to use cellular GAs in this case. One may notice that cellular GAs use crossover extensively. It has been criticized [40] that the use of crossover can be detrimental to searching for a good solution in some circumstances. Fogel et al. 13] 14] also found that there was no advantage of using crossover in their experiments. To investigate the effectiveness of the crossover operator, we have removed it from the ....

J. D. Schaffer and L. J. Eshelman. On crossover as an evolutionary viable strategy. In Proceedings of the Fourth International Conference on Genetic Algorithms, pages 61--68, 1991.

....that sexual recombination is useful for low and medium epistasis (not very rugged landscapes) We found that on these very problems a generalization of 1 point 2 parent crossover (that they use) to n parent diagonal crossover increases GA performance. Also the results of Schaffer and Eshelman, [13], come into mind looking at our observations. They conclude that crossover is useful on mildly epistatic non deceptive problems. On such problems it is worth to use more parents (within the diagonal scheme) that is multi parent s niche is (at least) as big as usual crossover s niche, but comes ....

....it is worth to use more parents (within the diagonal scheme) that is multi parent s niche is (at least) as big as usual crossover s niche, but comes with higher performance. There is a growing number of studies on the usefulness of sexual recombination. Besides those within the GA paradigm [6, 13, 16], also comparisons of GAs and EP (where no recombination is used) have been performed. The results in [7, 8] indicate that EAs with mutation only can be better than EAs with crossover and mutation. Our observations on multi parent crossovers and the comparisons between asexual and sexual ....

J.D. Schaffer and L.J. Eshelman. On crossover as an evolutionary viable strategy. In Fourth International Conference on Genetic Algorithms, pages 61--68, 1991.

....are applied in could be established so far. Most probably this relation is (to some extent) problem dependent. Besides comparisons of crossover operators, the relative importance, i.e. search power, of sexual recombination and asexual, unary operators is investigated. The experiments reported in [28] show that mutation and selection are more powerful than previously believed. Nevertheless, it is observed that a GA with highly disruptive crossover outperforms a GA with mutation alone on problems with a low level of interactions between genes. It is also remarked that the power of an operator ....

....crossover is useful for maximizing the accumulated payoff, while it can be harmful if optimality is sought. Based on a comparison of EP and GA in [14] it is argued that crossover does not have any competitive advantage above mutation. Critiques on this investigation, for instance in [28], initiated more extensive comparisons of GAs and EP on function optimization problems. The results in [15] show a clear advatage of EP and make the authors conclude that no consistent advantage accrues from representing real valued parameters as binary strings and allowing crossover to do ....

J.D. Schaffer and L.J. Eshelman. On crossover as an evolutionary viable strategy. In Fourth International Conference on Genetic Algorithms, pages 61--68, 1991.

....can determine overall NN behaviour. Further different emphasis on different parts (such as links, nodes activation functions and weights) of an NN contributes to nonlinearity. This is avoided only when GA NN combinations are constrained to search of learning parameters or initial weights (eg, Schaffer and Eshelman 1991). Normally, there would be close to maximum epistasis no subset of genes is independent of any other genes. Effectively this implies a random objective function However, this is not necessarily the case for GA NN systems; the high level of epistasis can be conceptualised as exploring the ....

Schaffer, J D and Eshelman, L J (1991). "Crossover as an evolutionary viable strategy". In R K Belew and L B Booker (eds.), Fourth international conference on genetic algorithms (pp. 661-68). San Mateo, CA: Morgan Kaufmann.

....16 ganisms by itself, and biologists consider it as the main drive in the evolutionary process. There is evidence that suggests that although mutation only evolution is slower than combining it with crossover, in general it produces better quality solutions than a crossover only algorithm [Schaffer Eshelman 91] In traditional genetic algorithms that use a binary coding, mutation would just mean flipping a certain bit. The use of non binary alphabets forces the extension of this concept. Here, a random number up to the number of possible alleles for a gene would be generated, and the corresponding ....

J. D. Schaffer and L. J. Eshelman. On crossover as an evolutionary viable strategy. In R. K. Belew and L. B. Booker, editors, Proceedings of the Fourth International Conference on Genetic Algorithms, pages 61--68. Morgan Kaufman, 1991.

....(i.e. they implement a simplification of sexual reproduction) and mutation. There are several papers investigating the advantages and disadvantages of mutation with respect to crossover (Eshelman and Schaffer, 1993; Fogel and Atmar, 1990; Fogel and Stayton, 1994; Hordijk and Manderick, 1995; Schaffer and Eshelman, 1991; Spears, 1993) At the moment the question whether mutation or crossover is preferable (or rather, which one is preferable under certain circumstances) is still an open research issue. Technically, the question concerns the arity of the reproduction operators. Mutation and crossover have arity ....

Schaffer, J. and Eshelman, L. (1991). On crossover as an evolutionary viable strategy. In (Belew and Booker, 1991), pages 61--68.

....and techniques from two sub areas in evolutionary computation. One area is the study of reproduction operators. There has been a lot of research on the power of different crossover operators and on the power of crossover (binary reproduction operator) versus mutation (unary reproduction operator) [9, 10, 11, 17, 16]. Recently, n ary crossovers were introduced and shown to have increased performance for higher arities on some numerical optimization problems and Technical Report 97 11, Department of Computer Science, Leiden University. Available as ....

....and Technical Report 97 11, Department of Computer Science, Leiden University. Available as ftp: ftp.wi.leidenuniv.nl pub CS TechnicalReports 1997 tr97 11.ps. gz NK landscapes [6, 7, 8] Here we investigate fitness landscapes that were specifically designed for studying crossovers in [9, 16, 17]. Adaptivity and emergent properties in simulated evolutionary processes form the second area this research is based upon. Having several options for a certain GA parameter or component implies that a choice has to be made between them. Choosing the best crossover operator usually happens by ....

[Article contains additional citation context not shown here]

J.D. Schaffer and L.J. Eshelman. On crossover as an evolutionary viable strategy. In Belew and Booker [1], pages 61--68.

.... Delta 10 4 0.964 0.00164 U 1 2:373870 Delta 10 1 Uniform crossover, applied with a probability p c 1, seems to be important for the search. The concentration on this operator clarifies that highly disruptive operators can be helpful, which was already indicated by other researchers [21, 19]. The elitist property is invented in order to guarantee that advantageous information is not lost by too disruptive operators and weak selective pressure. Remaining fitness differences between similar individuals in table 1 can be explained by the fact that the genetic algorithm fitness ....

J. D. Schaffer and L. J. Eshelman. On crossover as an evolutionary viable strategy. In Belew and Booker [6], pages 61--68.

....was average online performance. This investigation is in the spirit of this paper. But the authors did not test whether the optimal parameters were contained in their set. Nevertheless, some of their results can be understood by our investigation. Recent studies have been made by Schaffer et al. [SE91] and Fogel et al. [FA90] In the new investigation Schaffer et al. used five discrete fitness functions of size 100 and a fixed population size of 512. The mutation rate was set to 0.0005. Compared to their earlier study, they made a much more detailed study of mutation and crossover. But ....

....functions The outcome of a comparison of mutation and crossover depends on the fitness landscape. Therefore a carefully chosen set of test functions is necessary. We will use test functions which we have theoretically analyzed in [Muh93] They are similar to the test functions used by Schaffer [SE91]. The test suite consists of ONEMAX PLATEAU(k,l) SYMBASIN(k,l) DECEPTION(k,l) The fitness of ONEMAX is given by the number of 1 s in the string. For the PLATEAU function k bits have to be flipped in order that the fitness increases by k. The DECEPTION function has been defined by Goldberg [GDK90] ....

[Article contains additional citation context not shown here]

J.D. Schaffer and L.J. Eshelman. On crossover as an evolutionary viable strategy. In R. K. Belew and L. Booker, editors, Procedings of the Fourth International Conference on Genetic Algorithms, pages 61--68, San Mateo, 1991. Morgan Kaufmann.

No context found.

J. Schaffer and L. Eshelman. On Crossover as an Evolutionary Viable Strategy. In R. Belew and L. Booker, editors, Proceedings of the 4th International Conference on Genetic Algorithms, pages 61--68. Morgan Kaufmann, 1991.

No context found.

J.D. Schaffer and L.J. Eshelman. On crossover as an evolutionary viable strategy. In R. K. Belew and L. Booker, editors, Procedings of the Fourth International Conference on Genetic Algorithms, pages 61--68, San Mateo, 1991. Morgan Kaufmann.

No context found.

J.D. Scha#er and L.J. Eshelman. On crossover as an evolutionary viable strategy. In R.K. Belew and L.B. Booker, editors, Proceedings of the 4th International Conference on Genetic Algorithms, pages 61--68, 1991.

No context found.

J.D. Schaffer and L.J. Eshelman. On crossover as an evolutionary viable strategy. In R. K. Belew and L. Booker, editors, Procedings of the Fourth International Conference on Genetic Algorithms, pages 61--68, San Mateo, 1991. Morgan Kaufmann.

No context found.

J.D. Schaffer and L.J. Eshelman (1991). On crossover as an evolutionary viable strategy. In R. K. Belew and L. Booker, editors, Procedings of the Fourth International Conference on Genetic Algorithms, pages 61--68, Morgan Kaufmann, San Mateo.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC