Whitley, D. (1989). The GENITOR algorithm and selective pressure. Proceedings of the Third International Conference on Genetic Algorithms (ICGA-89), pp. 116-121. Califor- nia: Morgan Kaufmann.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....the mutation rate 2000 1000 Figure 1 is a pseudo code description of the model of evolutionary versatility. In this model, a genome is a string of bits. The model is a steady state genetic algorithm (as opposed to a generational genetic algorithm) in which children are born one at a time [28, 29, 39, 40]. In a generational genetic algorithm, the whole population is updated simulta neously, resulting in a sequence of distinct generations. Parents are selected using tourna ment selection [9, 10] In tournament selection, the population is randomly sampled and the two fittest individuals in the ....

Whitley, D. (1989). The GENITOR algorithm and selective pressure. Proceedings of the Third International Conference on Genetic Algorithms (ICGA-89), pp. 116-121. Califor- nia: Morgan Kaufmann.

....of string sequences not found in the parents, or in other words, new regions in the domain are visited. Other techniques have been suggested. We can establish a threshold heuristic so that the best individual is always saved from generation to generation (elitism) as is used in the Genitor model [Whitley, 1989]. This approach assures that evaluation values will never decrease from one generation to the next and assures that crossover and mutation do not lead to a degradation. Another strategy can be to consider the best individual to be a permanent mate and the other to be selected on spinning the ....

Whitley, D.: "The GENITOR Algorithm and Selective Pressure", pages 116 121. Morgan Kaufmann Publishers, Inc., San Mateo, USA (1989).

....governing Extractor s behavior is shown in Table 17 along with a brief explanation of the exact role each plays in the process described above. In the released version of Extractor, the twelve parameters are set to values determined using a commercially available genetic algorithm named Genitor [WHI89]. A genetic algorithm is a learning system for numerical optimization that works with populations of individuals, usually represented as binary vectors. Initially the vectors are set randomly. On each cycle (or generation) individuals are scored using a fitness function and the highest scoring ....

....it with the fitness function, and then using this individual to replace the least fit individual in the original population. This variation apparently produces a slightly higher selection pressure on the population than the original algorithm, causing it to converge more quickly to a steady state [WHI89]. The fitness function for the Extractor parameters was defined by treating keyphrase extraction as a retrieval task on a fixed corpus of documents with keyphrases already assigned. Thus Extractor was trained to match human choices of keyphrases as closely as possible. Each document was viewed as ....

D. Whitley. The GENITOR Algorithm and Selective Pressure. In Proc. ICGA-89. 1989. 116-121.

....clear as I describe the model. Figure 1 is a pseudo code description of the model of evolutionary versatility. In this model, a genome is a string of bits. The model is a steady state genetic algorithm (as opposed to a generational genetic algorithm) in which children are born one at a time [28, 29, 39, 40]. In a generational genetic algorithm, the whole population is updated simultaneously, resulting in a sequence of distinct generations. Parents are selected using tournament selection [9, 10] In tournament selection, the population is randomly sampled and the two fittest individuals in the ....

Whitley, D. (1989). The GENITOR algorithm and selective pressure. Proceedings of the Third International Conference on Genetic Algorithms (ICGA-89), pp. 116-121. California: Morgan Kaufmann.

....Canada K1A 0R6 peter.turney iit.nrc.ca 1 We assume here that fitness is measured directly from the individual s phenotype. We do not mean fitness as measured by the long term production of descendents. 2 A MODEL OF EVOLVABILITY Our model of evolvability was implemented by modifying Whitley s (1989) GENITOR software. 2 GENITOR is a steady state genetic algorithm (as opposed to a generational genetic algorithm) in which children are born one ata time. A new child replaces the least fit member of the current population. We fixed the population size at 2000 individuals. The initial ....

....of 50 bits. The bits in the target string are initially set randomly. Once every 8000 children, the target string is modified by randomly mutating 10 of its bits. We call this interval of 8000 children an era. Parents are randomly selected, with a bias of 2. 0 in favour of fitter individuals (see Whitley (1989) for details) We use single point crossover, with the constraint that the crossover point cannot occur within a pair; it must occur between two pairs. 3 Mutation is randomly applied to 50 of the children, after crossover. If a child is chosen for mutation, then the odd bits (evolvability bits) ....

Whitley, D. (1989). The GENITOR algorithm and selective pressure. Proceedings of the Third International Conference on Genetic Algorithms (ICGA-89), pp. 116-121.

....that a specialized algorithm, developed specifically for learning to extract keyphrases from text, might achieve better results than a general purpose learning algorithm, such as C4.5. Section 8 introduces the GenEx algorithm. GenEx is a hybrid of the Genitor steady state genetic algorithm (Whitley, 1989) and the Extractor parameterized keyphrase extraction algorithm (NRC, patent pending) Extractor works by assigning a numerical score to the phrases in the input document. The final output of Extractor is essentially a list of the highest scoring phrases. The behaviour of the scoring function is ....

....we came to believe that a tailor made algorithm for learning to extract keyphrases might be able to achieve better precision than a general purpose learning algorithm such as C4.5. This motivated us to develop the GenEx algorithm. GenEx has two components, the Genitor genetic algorithm (Whitley, 1989) and the Extractor keyphrase extraction algorithm (NRC, patent pending) Extractor takes a document as input and produces a list of keyphrases as output. Extractor has twelve parameters that determine how it processes the input text. In GenEx, the parameters of Extractor are tuned by the Genitor ....

[Article contains additional citation context not shown here]

Whitley, D. (1989). The GENITOR algorithm and selective pressure. Proceedings of the Third International Conference on Genetic Algorithms (ICGA-89), pp. 116-121. California: Morgan Kaufmann.

....Figure 7: A simple parameter representation of weights for a neural network. The tness of the policy is the payo when the agent uses the corresponding neural net as its decision policy. optimization can be used to optimize the weights of the neural network (Belew, McInerney, Schraudolph, 1991; Whitley, Dominic, Das, Anderson, 1993; Yamauchi Beer, 1993) This representation thus requires the least modi cation of the standard EA. We now turn to distributed representations of policies in EARL systems. 5.2 Distributed Representation of Policies In the previous section we outlined EARL ....

....on both simulated robots and on real robots. Because it exploits both human design and EARL methods to optimize system performance, this method shows much promise for scaling up to realistic tasks. Evolutionary Algorithms for Reinforcement Learning 10.3 Genitor Genitor (Whitley Kauth, 1988; Whitley, 1989) is an aggressive, general purpose genetic algorithm that has been shown e ective when specialized for use on reinforcement learning problems. Whitley et al. 1993) demonstrated how Genitor can e ciently evolve decision policies represented as neural networks using only limited reinforcement from ....

Whitley, D. (1989). The GENITOR algorithm and selective pressure. In Proceedings of the Third International Conference on Genetic Algorithms, pp. 116-121 San Mateo, CA. Morgan Kaufman.

....In the simplest case (see Figure 7) a neural network for the agent s decision policy is represented as a sequence of real valued connection weights. A straightforward EA for parameter optimization can be used to optimize the weights of the neural network (Belew, McInerney, Schraudolph, 1991; Whitley, Dominic, Das, Anderson, 1993; Yamauchi Beer, 1993) This representation thus requires the least modification of the standard EA. We now turn to distributed representations of policies in EARL systems. 5.2 Distributed Representation of Policies In the previous section we outlined EARL ....

....1998) The approach has been implemented and tested on both simulated robots and on real robots. Because it exploits both human design and EARL methods to optimize system performance, this method shows much promise for scaling up to realistic tasks. 10.3 Genitor Genitor (Whitley Kauth, 1988; Whitley, 1989) is an aggressive, general purpose genetic algorithm that has been shown effective when specialized for use on reinforcement learning problems. Whitley et al. 1993) demonstrated how Genitor can efficiently evolve decision policies represented as neural networks using only limited reinforcement ....

Whitley, D. (1989). The GENITOR algorithm and selective pressure. In Proceedings of the Third International Conference on Genetic Algorithms, pp. 116--121 San Mateo, CA.

....and ffl Evolution Strategies [Rec73, BSM93, Sch95] Whether Genetic Algorithms should be part of this list is a topic of discussion. According to DeJong GA s are no function optimizers [Jon93] On the other hand there are many successful applications of GA s to numerical optimization problems [Whi89, WGM94, PJ94, EvKK95] In all of these applications additional features are added to the GA, such that these GA s do not correspond to the pure definition of the GA any more. In this paper we are going to focus on evolution based optimization methods. Evolution can be used to learn about a ....

D. Whitley. The GENITOR algorithm and selective pressure. In Third International Conference on Genetic Algorithms, pages 116--121, 1989.

....to the actual domain. SAMUEL has been shown effective in several small problems including the evasive maneuvers problem (Grefenstette et al. 1990) and the game of cat and mouse (Grefenstette 1992) but has not yet been applied to largerscale real world domains. GENITOR (Whitley and Kauth 1988; Whitley 1989) is an aggressive search genetic algorithm that has been shown to be quite effective as a reinforcement learning tool for neural networks. GENITOR is considered aggressive because it uses small populations, large mutation rates, and rank based selection to create greater variance in solution ....

Whitley, D. (1989). The GENITOR algorithm and selective pressure. In Proceedings of the Third International Conference on Genetic Algorithms. Morgan Kaufman.

.... fixed for every run, and a random number, using the following equation: rank = int 0 Size of P opulation bias Gamma q bias 2 Gamma 4(bias Gamma 1)rand 2(bias Gamma 1) 1 A (2:1) 1 This scheme is original of PGA c fl Peter Ross and Geoff Ballinger, 1993 originally based on GENITOR [Whitley 89] MSc. Information Technology: Knowledge Based Systems September Evolutionary Divide and Conquer Luis F. Gonz alez Chapter 2. Background 12 where bias is a parameter determined at the beginning of the run and rand is a random number uniformly distributed in the interval [0,1) Tournament ....

D. Whitley. The GENITOR algorithm and selective pressure. In J.D. Schaffer, editor, Proceedings of the Third International Conference on Genetic Algorithms, pages 116--121. Morgan Kaufmann, 1989. MSc. Information Technology: Knowledge Based Systems September Evolutionary Divide and Conquer Luis F. Gonz'alez Bibliography 77

....Steady State algorithms use a biased selection scheme for reproduction, while unbiased (uniform) selection is used for the reduction phase. During the reduction phase it is decided which individual will be removed from the population to make room for new individuals. In the description of Genitor [Whi89] it is suggested that the application of a bias in the reduction phase too, can result in a faster search. We go even one step further. We use an unbiased selection of parents for reproduction, so the biased selection for reduction is our prime guide during the evolution. This modified scheme ....

D. Whitley. The genitor algorithm and selective pressure. In Third International Conference on Genetic Algorithms, pages 116--121, 1989.

....MORIARTY AND R. MIIKKULAINEN evolves neural networks. The interpolative ability of neural networks should allow SANE to learn tasks quicker than SAMUEL, however, it is easier to incorporate pre existing knowledge of the task into the initial population of SAMUEL. GENITOR (Whitley and Kauth, 1988; Whitley, 1989) is an aggressive search genetic algorithm that has been shown to be quite effective as a reinforcement learning tool for neural networks. GENITOR is considered aggressive because it uses small populations, large mutation rates, and rank based selection to create greater variance in solution ....

Whitley, D. (1989). The GENITOR algorithm and selective pressure. In Proceedings of the Third International Conference on Genetic Algorithms. San Mateo, CA: Morgan Kaufman.

....the influence of multi point crossover [1] and Syswerda defined and tested the uniform crossover [5] which can be tuned using a biased coin. The generalized n ary crossover operators [2, 3] adjusts the number of parents in order to tune disruptiveness. In this paper we focus on Steady State GA s [8, 6] using a bit representation. This GA will be applied to function optimization problems, as these problems are often used as a test suite in GA related papers. However, we think that most statements in this paper are of a more general nature: they can be applied to other kinds of problems and other ....

....bad performance. Given the second definition of P d , we see that a raising value for P I [H q ] which corresponds to a schemata of relatively high fitness, will result in the recombination operator getting less disruptive for schemata H. 4 Set up of the Experiments We consider Steady State GA s [8, 6]. During a single cycle of such a GA only a small fraction of the total population is replaced. In our experiments one child is produced, and replaces an individual of the population. A single cycle consist of selection, production and reduction. According to Syswerda [6] a Steady State algorithm ....

D. Whitley. The GENITOR algorithm and selective pressure. In Third International Conference on Genetic Algorithms, pages 116--121, 1989.

....p mutation , of being replaced by a uniformly distributed random number. The mutation range and the creep variance are larger for the locations on the chromosome which describe the x and y coordinate of the feature detectors.The selection operator used is Whitley s rank order selection [7]. Table 1 shows the parameters used for the experiment. The number of trial networks is equivalent to the population size for the MTA, but not necessarily so for the other two algorithms. For a fair FIGURE 2. Noisy digit images s n y j 1 J 1 y j j j = ....

D. Whitley, "The GENITOR algorithm and selective pressure", in Proceedings of the Third International Conference on Genetic Algorithms, J.D. Schaffer Ed. San Mateo, CA: Morgan Kaufmann, pp. 116-121, 1989.

....used [Bak85, GD91] In a ranking scheme the individuals are sorted and ranked on fitness. The probability of being selected for reproduction is coupled to the rank. By adjusting the probability of survival the selective pressure can be set. Ranking is often combined with a steady state algorithm [Sys91, Whi89]. Such a steady state algorithm replaces just a small part of the population during each iteration. Hence the rank will be recalculated often and a more aggressive search is obtained. As the search space is very large and it is assumed to contain many good solutions, an aggressive search method ....

D. Whitley. The genitor algorithm and selective pressure. In Third International Conference on Genetic Algorithms, pages 116--121, 1989.

....On the other hand, it is easier to incorporate preexisting domain knowledge into a set of decision rules than accross the weights of a neural network. The bottom line, however, is that both provide effective generalization of the decision policy. 7.1. 2 GENITOR GENITOR (Whitley and Kauth 1988; Whitley 1989) is an aggressive search genetic algorithm that has been shown effective in reinforcement learning problems. Whitley et al. 1993) demonstrated how GENITOR can efficiently evolve decision making neural networks using only limited reinforcement from the domain. Since GENITOR uses neural networks to ....

Whitley, D. (1989). The GENITOR algorithm and selective pressure. In Proceedings of the Third International Conference on Genetic Algorithms, 116--121. San Mateo, CA: Morgan Kaufman.

....population. In the canonical genetic algorithm, fitness is defined by: f i = f where f i is the evaluation associated with string i and f is the average evaluation of all the strings in the population. Fitness can also be assigned based on a string s rank in the population (Baker, 1985; Whitley, 1989) or by sampling methods, such as tournament selection (Goldberg, 1990) String 1 String 2 String 3 String 4 String 1 String 2 String 2 String 4 (Duplication) Crossover) Next Generation t 1 Intermediate Generation t Generation t Current Selection Recombination Offspring A (1 X 2) Offspring B ....

....strings for generating an expression for computing all terms of the form P(Z,t 1) 6. 1 A Generalized Form Based on Equation Generators The 3 bit equations were generated by hand; however, a general pattern exists which allows the automatic generation of equations for arbitrary problems (Whitley, Das and Crabb, 1992) The number of terms in the equations is greater than the number of strings in the search space. Therefore it is only practical to develop exact equations for problems with approximately 15 bits in the encoding. The development of a general form for these equations is ....

[Article contains additional citation context not shown here]

Whitley, D. (1989) The GENITOR Algorithm and Selective Pressure. Proc 3rd International Conf on Genetic Algorithms, Morgan-Kaufmann, pp 116-121.

....population. In the canonical genetic algorithm, fitness is defined by: f i = f where f i is the evaluation associated with string i and f is the average evaluation of all the strings in the population. Fitness can also be assigned based on a string s rank in the population (Baker, 1985; Whitley, 1989) or by sampling methods, such as tournament selection (Goldberg, 1990) It is helpful to view the execution of the genetic algorithm as a two stage process. It starts with the current population. Selection is applied to the current population to create an intermediate population. Then ....

....given an appropriate population size. But a simple static count of the number of schemata available for processing fails to consider the dynamic behavior of the genetic algorithm. As discussed later in this tutorial, dynamic models of the genetic algorithm now exist (Vose and Liepins, 1991; Whitley, Das and Crabb 1992) There has not yet, however, been any real attempt to use these models to look at complex interactions between large numbers of hyperplane competitions. It is obvious in some vacuous sense that knowing the distribution of the initial population as well as the fitnesses of these ....

[Article contains additional citation context not shown here]

Whitley, D. (1989) The GENITOR Algorithm and Selective Pressure. Proc 3rd International Conf on Genetic Algorithms, Morgan-Kaufmann, pp 116-121.

No context found.

Whitley, D., "The GENITOR Algorithm and Selective Pressure", Proceedings of the Third International Conference on Genetic Algorithms, pages 116-121. Morgan Kaufmann Publishers, Inc., San Mateo, USA (1989).

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