K. Deb and D. E. Goldberg. Analyzing deception in trap functions. IlliGAL Report No. 91009, University of Illinois at Urbana-Champaign, Illinois Genetic Algorithms Laboratory, Urbana, IL, 1991.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....Evolutionary Computation (EC) are widely used to tackle hard optimization problems. More recently we can highlight new population based models that rely on probability distributions that sample the search space [1] or that take into account the variable dependencies like Bayesian networks [6, 7]. In this paper we show that another kind of bio inspired algorithm, novel ant based algorithm, can be used to perform probabilistic search with the help of Hidden Markov Models (HMMs) Ant based methods share a common feature with EC: a population of solutions is maintained. The population used ....

M. Pelikan, D.E. Goldberg, and E. Cantd-Paz. Link- age problem, distribution estimation, and bayesian networks. IlliGAL Report 98013, Illinois Genetic Algorithms Laboratory, University of Illinois, 1998.

....of the problem, other probabilistic models, able to cover higher order interactions, appear in the literature. For this purpose, Bayesian networks, graphical representations able to cover these higher order interactions, can be used. Thus, EBNA (Etxeberria and Larra naga [16] BOA (Pelikan et al. [49]) are algorithms which use different implementations of Bayesian networks for estimating the joint distribution of promising solutions. In this way, multivariate interactions among variables can be covered. In Soto et al. 56] a Bayesian network with a polytree structure is proposed: the proposed ....

Pelikan, M., Goldberg, D.E., and Cant'u-Paz, E., BOA: The Bayesian Optimization Algorithm, IlliGAL Report 99003, Urbana: University of Illinois at Urbana-Champaign, Illinois Genetic Algorithms Laboratory, 1999.

.... system will have in order to solve the problem successfully, or the number of possible recombinations that might have been done by the genetic algorithm to find new and (hopefully) better classifiers an equation has been derived in order to 51 estimate the quality of convergence of a GA (see [Cant u Paz E. Miller 96] for more information about this) The population sizes used for these experiments were: 50, 100, 200, 300 and 400 classifiers. 5.2.2 Different Random Seeds Used. For the purpose of comparing the different behaviours achieved, different random seeds were used, using the same values for the ....

G. Goldberg D.E. Cant'u-Paz E., Harik and B.L. Miller. The gambler's ruin problem, genetic algorithms and the sizing of populations. Technical report, Illinois Genetic Algorithms Laboratory, 1996.

....of the problem, other probabilistic models, able to cover higher order interactions, appear in the literature. For this purpose, Bayesian networks, graphical representations able to cover these higher order interactions, can be used. Thus, EBNA (Etxeberria and Larra naga [16] BOA (Pelikan et al. [53]) are algorithms which use different implementations of Bayesian networks for estimating the joint distribution of promising solutions. In this way, multivariate interactions among variables can be covered. In Soto et al. 60] a Bayesian network with a polytree structure is proposed: the proposed ....

M. Pelikan, D.E. Goldberg and E. Cant'u-Paz, BOA: The Bayesian Optimization Algorithm, IlliGAL Report 99003, Urbana: University of Illinois at Urbana-Champaign, Illinois Genetic Algorithms Laboratory, 1999.

....of the problem, other probabilistic models, able to cover higher order interactions, appear in the literature. For this purpose, Bayesian networks, graphical representations able to cover these higher order interactions, can be used. Thus, EBNA (Etxeberria and Larra naga [16] BOA (Pelikan et al. [53]) are algorithms which use di erent implementations of Bayesian networks for estimating the joint distribution of promising solutions. In this way, multivariate interactions among variables can be covered. In Soto et al. 60] a Bayesian network with a polytree structure is proposed: the proposed ....

M. Pelikan, D.E. Goldberg and E. Cantu-Paz, BOA: The Bayesian Optimization Algorithm, IlliGAL Report 99003, Urbana: University of Illinois at Urbana-Champaign, Illinois Genetic Algorithms Laboratory, 1999.

....of the problem which is often not available. Without the need of this extra information about the decomposition and factorization of the problem, Bayesian networks are graphical representations which cover higher order interactions. EBNA (Etxeberria and Larra naga [29] BOA (Pelikan et al. [71]) are algorithms which use Bayesian networks for estimating the joint distribution of promising solutions. In this way multivariate interactions among problem variables can be covered. Based on the EBNA work of Etxeberria and Larra naga [29] we propose the use of Bayesian networks as the models ....

M. Pelikan, D.E. Goldberg, E. Cant'u-Paz, BOA: The Bayesian Optimization Algorithm, IlliGAL Report 99003, Urbana: University of Illinois at Urbana-Champaign, Illinois Genetic Algorithms Laboratory, 1999.

....the aim was to obtain several points from the Pareto set. Therefore, some kind of mechanism is required in order to maintain diversity in population. The most obvious way would be to use the fitness value sharing. It has been shown that this approach fails to preserve the diversity in population [16]. Therefore, a modified algorithm is proposed. 4 M AKINEN, PERIAUX TOIVANEN 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 f2 f1 f=4 f=3 f=2 f=1 FIG. 1. An example of how the fitness values are assigned. Instead of using some previously considered method, we develop a new way to ....

C. K. Oei, D. E. Goldberg and S.-J. Chang, Tournament selection, niching, and the preservation of diversity, IlliGAL report no. 91011, University of Illinois at Urbana-Champaign, Illinois Genetic Algorithms Laboratory, 1991.

....GAs are particularly resilient in sampled objective functions. Grefenstette and Fitzpatrick s (1985) sampled image registration problem reinforced this view, but for a long time there was little analytical support to this largely empirical observation. More recently, other works (Miller, 1997; Miller and Goldberg, 1996a) have analyzed a related question, the optimal level of sampling in problems with additive gaussian noise, and these calculations appear to be useful in understanding sampled objective functions. There has been, however, only limited experimentation undertaken (Miller, 1997) to understand the ....

....showed results for which sampling improves performance when the number of generations was fixed and when it varied according to the sample size. Other works by Aizawa and Wah (1994a, 1994b) have focused on sampling strategies for GAs in the case of a fixed population size. More recently, Miller and Goldberg (1996a) have investigated the optimum sampling size for noisy fitness functions. The work of Miller and Goldberg provided a domain independent lower bound and for the optimal sample size that was experimentally verified using the OneMax domain (problem) A domain dependent upper bound for the optimal ....

[Article contains additional citation context not shown here]

Miller, B. L. and Goldberg, D. E. 1996a. Optimal Sampling for Genetic Algorithms. Illinois Genetic Algorithms Laboratory (IlliGAL) Report No. 96005. University of Illinois at Urbana-Champaign General Engineering Department. August 29, 1996.

....for sampling, as we did for selection. New sample members are chosen to have maximal competitive shared fitness within the current sample. This is equivalent to choosing the sample from the population via truncation selection using competitive fitness sharing with continuously updated sharing [69]. In this way, each successive member of the sample is chosen to be one that competed well against individuals that the other members of the current sample did not compete well against. This technique will be called shared sampling, and is specified in detail by the algorithm in Figure III.1. ....

....the hall of fame sample is filled out with random individuals as necessary for these first few generations. The use of standard fitness sharing with tournament selection can lead to fluctuations in niche sizes. The method of continuously updated sharing has been developed to overcome this problem [69], This method involves calculating shared fitness within the next generation s population, currently being constructed, rather than in the original population. This resembles the method of shared sampling used here. Continuously updated sharing was tried here with competitive fitness sharing, 2 ....

[Article contains additional citation context not shown here]

C.K. Oei, D.E. Goldberg, and S.J. Chang. Tournament selection, niching, and the preservation of diversity. Technical Report IlliGAL Report 91011, Urbana: University of Illinois, Illinois Genetic Algorithms Laboratory, 1991.

No context found.

K. Deb and D. E. Goldberg. Analyzing deception in trap functions. IlliGAL Report No. 91009, University of Illinois at Urbana-Champaign, Illinois Genetic Algorithms Laboratory, Urbana, IL, 1991.

No context found.

D. E. Goldberg, D. Kalyanmoy, and J. Horn. Massive multimodality, deception, and genetic algorithms. Technical Report 92005, Illinois Genetic Algorithms Laboratory, University of Illinois at Urbana-Champaign, 1992.

No context found.

D. E. Goldberg, D. Kalyanmoy, and J. Horn. Massive multimodality, deception, and genetic algorithms. Technical Report 92005, Illinois Genetic Algorithms Laboratory, University of Illinois at Urbana-Champaign, 1992.

....it is very dicult to decide which model is the best one for our purpose. Too simple model may not cover all important interactions. Too complex model may not bring enough variation in the optimization process. Both too simple and too complex models may thus result in inferior performance. See Pelikan and Goldberg (2000a) and Pelikan, Goldberg, and Sastry (2000) for a discussion on this topic. Subsequent paragraphs describe basic principles of the algorithms that use probabilistic models of promising solutions to guide their search. For a more detailed overview of PMBGAs, see Pelikan et al. 2000) Probably the ....

....See Pelikan and Goldberg (2000a) and Pelikan, Goldberg, and Sastry (2000) for a discussion on this topic. Subsequent paragraphs describe basic principles of the algorithms that use probabilistic models of promising solutions to guide their search. For a more detailed overview of PMBGAs, see Pelikan et al. 2000) Probably the simplest way to estimate the distribution of good solutions is to assume that the variables in a problem are independent. New solutions can be generated by only preserving the proportions of the values of all variables independently of the context. This is the basic principle of ....

[Article contains additional citation context not shown here]

Pelikan, M., Goldberg, D. E., & Cantu-Paz, E. (2000a). Bayesian optimization algorithm, population sizing, and time to convergence (IlliGAL Report No. 2000001). Urbana, IL: University of Illinois at Urbana-Champaign, Illinois Genetic Algorithms Laboratory.

....it is very dicult to decide which model is the best one for our purpose. Too simple model may not cover all important interactions. Too complex model may not bring enough variation in the optimization process. Both too simple and too complex models may thus result in inferior performance. See Pelikan and Goldberg (2000a) and Pelikan, Goldberg, and Sastry (2000) for a discussion on this topic. Subsequent paragraphs describe basic principles of the algorithms that use probabilistic models of promising solutions to guide their search. For a more detailed overview of PMBGAs, see Pelikan et al. 2000) Probably the ....

....See Pelikan and Goldberg (2000a) and Pelikan, Goldberg, and Sastry (2000) for a discussion on this topic. Subsequent paragraphs describe basic principles of the algorithms that use probabilistic models of promising solutions to guide their search. For a more detailed overview of PMBGAs, see Pelikan et al. 2000) Probably the simplest way to estimate the distribution of good solutions is to assume that the variables in a problem are independent. New solutions can be generated by only preserving the proportions of the values of all variables independently of the context. This is the basic principle of ....

[Article contains additional citation context not shown here]

Pelikan, M., & Goldberg, D. E. (2000b). Genetic algorithms, clustering, and the breaking of symmetry (IlliGAL Report No. 2000013). Urbana, IL: University of Illinois at Urbana-Champaign, Illinois Genetic Algorithms Laboratory.

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