John J. Grefenstette. A system for learning control strategies with genetic algorithms. pages 183--190, George Mason University, 1989. Naval Research Laboratory.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....of the conflicting rules, with an appropriate tiebreaker if necessary. A drawback of this fourth scheme is that identical or nearly identical rules can gang up on a better but less frequent rule. Another drawback is that further conflict resolution schemes are likely to be needed to break ties. Grefenstette (1989) addresses these problems by first selecting, for each possible classification, the fittest rule suggesting that classification, and then Financial Forecasting Using Genetic Algorithms 553 making a probabilistic choice, weighted according to fitness, from among those previously chosen. Other ....

....a problem similar to a conflict arises. As before, one can choose a class at random 554 S. Mahfoud and G. Mani from the outputs in the training set. A slightly better method is to choose the most frequent class indicated by the training data. Smarter strategies use partial matching (Grefenstette, 1989; Liepins Wang, 1991; Wilson, 1987) Greene and Smith (1993, 1994) avoid the problem of uncovered examples by covering each training example upon initialization; an elitist GA ensures that all newly generated populations also cover all training examples. The authors do not mention any strategy ....

Grefenstette, J. J. 1989. A system for learning control strategies with genetic algorithms. In Proceedings of the third international conference on genetic algorithms, 183190.

....less efficient than evaluation at the level of the system s principal components. Of course, there may be situations where a component interaction more complex than simple linear summation requires that the whole system be evaluated as a unit. An example may be recent work on a control problem [25] in which each population member is a set of rules. But in that case, too, the evolution times were long, on the order of one million trials. An emergent system is one that produces a global result through interaction of strictly local computations. Given an interaction scheme, design of such a ....

J. J. Grefenstette, A system for learning control strategies with genetic algorithms, in: Proc. Third Internat. Conf. on Genetic Algorithms, J. D. Schaffer, ed. (Morgan Kaufmann, San Mateo, CA, 1989).

....[111, 117, 147, 149, 151, 201] Germay, Noel, 202] 14 Gerys, D. 249] Glasmacher, Klaus, 41] Glesner, M. 203] Glover, David E. 39] Gold, Sonke Sonnich, 60] Gonzalez, Carlos, 184] Gorrini, V. 148] Goulter, I. C. 31] Greenwood, Garrison W. 118] Grefenstette, John J. [92] Gruau, Fr ed eric C. 204] Gubbi, Ananda V. 288] Gupta, Ajay, 118] Gupta, Mahesh C. 219, 225] Gupta, Yash P. 219, 225] Hasegawa, Yoshishige, 258] Hashimoto, Y. 80] Haupt, M. 101, 83] Hawaleshka, O. 77] Hegde, Shailesh U. 33, 34] Hegde, U. 40] Heinzmann, F. 99] ....

.... 257, 167, 180, 217, 60, 122, 128, 150, 162] parallel GA workstation network, 53] parallel processing, 169] permutation crossover, 27] permutation problems, 161] permutations, 171] planning assembly , 15] electronics, 13] population size, 225] 12, 126] 400, 21] process control, [91, 92, 96, 89, 93, 95] process planning, 98, 97] production planning, 99, 100, 102] PROGENITOR, 214, 215, 218] project management, 132] review process control, 89] robotics scheduling, 148] routing, 81] rules, 91] SAGA, 30] scheduling, 192, 221, 222, 187, 224, 246, 247, 261, 178, 203, 207, 213, 249, ....

[Article contains additional citation context not shown here]

John J. Grefenstette. A system for learning control strategies with genetic algorithms. In Schaffer [300], pages 183--190. ga:Grefenstette89a.

....genotype, traditionally as a bit string, but more generally as any convenient data structure. Strings have been preferred for simplicity and for the analogy with natural chromosomes, but CHAPTER 2. REVIEW OF GENETIC ALGORITHMS 11 other structures such as matrices, trees, and even lists of rules (Grefenstette, 1989), are also used. The distinction between the phenotype and its genotypic representation is important and will be adhered to here, because it facilitates the extension of the approach to novel classes of problems. However, it is not necessarily a rigid one: a genotype and the corresponding ....

Grefenstette, J. J. (1989). A system for learning control strategies with genetic algorithms. In (Scha#er, 1989), pages 183--190.

....closely as does GA, it does offer the opportunity to directly evolve programs of unusual complexity, without having to define the structure or the size of the program or genetic material in advance. Others have described GA approaches that operate on variable length or program structured genomes [5,14]. These approaches typically require more constraints on the form of the final solution than does GP. The work described here uses GP as defined by Koza in [9, 11] as well as a further elaboration of GP developed by Craig Reynolds called Steady State GP (SSGP) 13] Reynolds developed SSGP through ....

J. J. Grefenstette, "A System for Learning Control Strategies with Genetic Algorithms," in Proceedings of the 3rd International Conference on Genetic Algorithms, J. D. Schaffer, Ed. San Mateo, CA: Morgan Kaufmann, 1989.

....linear representations of genetic material, and operate on this material with fitness proportionate selection, reproduction, crossover, and mutation [De Jong 1987, Goldberg 1989] A number of researchers have experimented with GA s applied to variable length genetic representations. Grefenstette [Grefenstette 1989] and Schaffer [Schaffer 1984] contain notable examples. Bickel has applied GA s to tree structured genetic material [Bickel 1987] John Koza has developed Genetic Programming (GP) using analogues of GA genetic operators to directly modify and evolve tree structured programs (typically LISP ....

Grefenstette, J. J. (1989) "A System for Learning Control Strategies with Genetic Algorithms," in Proceedings of the 3rd International Conference on Genetic Algorithms, J. D.

....where each agent is codified by a neural network (NN) and Genetic Programming (GP) Koza 1992) where each individual in the population is a program in a high level language, can be viewed as paradigms belonging to this approach. The Pitt approach to Classifier Systems (Smith 1983; De Jong 1991; Grefenstette 1988, 1989, 1991) may be considered as an intermediate solution between GC and GP. In Classifier Systems (CS) a set of rules codifies the behavioral strategy of an agent. As Booker et al. 1990) point out, with respect to a NN, a rule based system allows more complex knowledge structures to emerge from ....

.... all kinds of systems where either an evolutionary or genetic algorithm is used to learn sets of rules to solve a given task, both when the rules are simple strings on a small alphabet (Smith 1983; De Jong 1991) and when the rules are codified in a higher level language, like Grefenstette s SAMUEL (Grefenstette 1989, 1991) This paper presents PANIC (Parallelism And Neural networks In Classifier systems) a parallel ERBS which uses a Genetic Algorithm (GA) Goldberg 1989) to evolve a population of sets of rules. The fitness of each genotype in the population is evaluated as the performance of the ....

[Article contains additional citation context not shown here]

Grefenstette, J. J. (1989). A system for learning control strategies with genetic algorithms. In Proceedings of the 3rd International Conference on Genetic Algorithms, San Mateo, CA: Morgan Kaufmann.

....corresponds to a rule of the KB and the GA searches the space of rules. The genetic search stresses the competition rather than the cooperation among rules, i.e. even the rules in the same set compete among each other to avoid being replaced by the GA. In a CS based on the Pitt approach, Pitt CS, [DeJ91, Gre89, Smi83], instead, a genotype codifies the whole KB of the CS. Thus, the competition among genotypes implicitly promotes the cooperation among the rules of the same KB. Our work investigates the role of rule cooperation in CSs and, in particular, the relation between rule cooperation and cooperation among ....

Grefenstette, J. J. (1989). A System for Learning Control Strategies with Genetic Algorithms. In Proceedings of the 3rd International Conference on Genetic Algorithms, San Mateo, CA: Morgan Kaufmann.

....examples of this is the application of GAs to rule learning. Evolving rule sets of varying length and complexity doesn t map neatly into the traditional GA paradigm, resulting in a variety of extensions including Holland s classifier system, Smith s LS system, and Grefenstette s Samuel system [11, 14, 7]. In this paper we present an extension of the traditional GA model which appears to have considerable potential for representing and solving more complex problems by explicitly Published in The Third Parallel Problem Solving From Nature, Jerusalem, Israel, pp. 249 257, SpringerVerlag, 1994 ....

J.J. Grefenstette. A system for learning control strategies with genetic algorithms. In J.D. Schaffer, editor, Proceedings of the Third International Conference on Genetic Algorithms, pages 183--190. Morgan Kaufmann, 1989.

....techniques more powerful by generalizing them: instead of looking for a single point solution to a specific instance of a problem, GP attempts to evolve a program capable of computing the solutions for any instance of a problem. While there are other non string based, variable length GA approaches [2, 11] which evolve programs for solving problems, and not solutions to individual problem instances, these programs are limited to simple if then rules. The paper proceeds as follows: Section 2 briefly reviews the GP approach. In Section 3 we explain some of the choices that must be made when ....

Grefenstette, J. J. A System for Learning Control Strategies with Genetic Algorithms. Proc. of 3rd Int`l Conf. on Genetic Algorithms, Morgan Kaufman, San Mateo, CA.,1989.

....CSs traditionally identifies two approaches, depending on how the GA is used as the discovery algorithm. Each genotype in the GA population may codify either a single rule, Michigan approach (Holland, 1986; Booker et al. 1990) or the whole KB, Pitt approach (Smith, 1980; DeJong and Spears, 1991; Grefenstette, 1989). In the first case, the GA acts on a population that includes all the rules in the KB. The fitness of each rule is a value, updated by the credit assignment system, that reflects its utility in the past behavior of the system. On the other hand, the Pitt approach maintains a population of KBs, ....

.... functions (Wilson, 1995) and developing behavioral strategies for a real robot (Dorigo and Colombetti, 1994; Dorigo, 1995) Pitt CSs (PCSs) have proved their effectiveness in producing non trivial rule sets for very different tasks, such as poker playing (Smith, 1983) multiagent interactions (Grefenstette, 1989; Grefenstette, 1991) evolving complex behavior in a simulated robot domain (Potter et al. 1995) and the control of gait in a wall climbing quadrupedal robot (Bull et al. 1995) 2.1.3 Credit Assignment In addition to rule discovery, a CS uses a second learning mechanism that allocates a kind ....

[Article contains additional citation context not shown here]

Grefenstette, J. J. (1989). A system for learning control strategies with genetic algorithms.

....by the current family of evolutionary algorithms. These approaches in turn have inspired the development of additional evolutionary algorithms such as classifier systems (Holland, 1986) the LS systems (Smith, 1983) adaptive operator systems (Davis, 1989) GENITOR (Whitley, 1989) SAMUEL (Grefenstette, 1989), genetic programming (de Garis, 1990; Koza, 1991) messy GAs (Goldberg, 1991) and the CHC approach (Eshelman, 1991) We will not attempt to survey this broad range of activities here. The interested reader is encouraged to peruse the recent literature for more details (e.g. Belew and ....

....algorithm (Eshelman, 1991) A second active area of application of EAs is in the design of robust rule learn ing systems. Holland s (1986) classifier systems were some of the early examples, followed by the LS systems of Smith (1983) More recent examples include the SAMUEL system developed by Grefenstette (1989), the GABIL system of De Jong and Spears (1991) and the GIL system of Janikow (1991) In each case, significant adaptations to the basic EAs have been made in order to effectively represent, evaluate, and evolve appropriate rule sets as defined by the environment. One of the most fascinating ....

Grefenstette, John J. (1989) A system for learning control strategies with genetic algorithms. Proceedings of the Third International Conference on Genetic Algorithms, 183-190. Fairfax, VA: Morgan Kaufmann.

....Learning the Persistence of Actions in Reactive Control Rules Helen G. Cobb and John J. Grefenstette Navy Center for Applied Research in Arti ficial Intelligence Naval Research Laboratory, Code 5514 Washington, DC 20375 5000 Abstract This paper explores the effect of explicitly searching for the persistence of each decision in a time dependent sequential decision task. In prior studies, ....

John J. Grefenstette. (1989) A system for learning control strategies with genetic algorithms. In J. David Schaffer (ed.), Proceedings of the Third International Conference on Genetic Algorithms and Their Applications. San Mateo, CA: Morgan Kaufmann.

....same offspring. The idea is to promote the inheritance of 4 A rule s strength increases as a function of the mean of the expected payoff and decreases with the variance of the expected payoff, so that high strength indicates both high utility and high confidence in the rule [8]. 5 The bid bias was inspired by a similar mechanism in Riolo s classifier system CFS C [14] and is similar in effect to the notion of using bidding noise based on a classifier s variance [4] 6 In all previously reported results with SAMUEL, the bid bias was set to 1. 7 We prefer to rely on ....

....the tracker has sensors that operate at a greater distance than the prey s sensors. The object is to keep the prey within range of 9 In the experiments described here, all sensors are structured attributes. 10 This environment differs from the EM problem in previous papers [8, 9, 13, 16]. In this paper, we introduce noise into both the agent s sensors and actions, and we vary both the initial state and the maneuverability characteristics of the adversary. As a result, the task is more realistic and more challenging. the tracker s sensors, without being detected by the prey. If ....

[Article contains additional citation context not shown here]

Grefenstette, J. J. (1989). A system for learning control strategies with genetic algorithms. Proceedings of the Third International Conference on Genetic Algorithms. Fairfax, VA: Morgan Kauf - mann (pp. 183-190)

No context found.

John J. Grefenstette. A system for learning control strategies with genetic algorithms. pages 183--190, George Mason University, 1989. Naval Research Laboratory.

No context found.

John J. Grefenstette. A System for Learning Control Strategies with Genetic Algorithms. In Schaer [351], pages 183-190.

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