Fonseca, M.C., Fleming, P.J.: Multi-objective genetic algorithms made easy: Selection, sharing and mating restrictions. In: Proceedings of the 1st International Conference on Genetic Algorithms in Engineering Systems: Innovations and Application. (1995) 45--52  Home/Search   Document Not in Database   Summary   Related Articles   Check

....Figure 1. The Pareto optimal range of solutions for some multiobjective functions can include almost all allowable solutions to the problem. Hence, for many problems, the set of solutions deemed acceptable by a user will be a small sub set of the set of Pareto optimal solutions to the problems [4]. Manually choosing an acceptable solution can be a laborious task, which would be avoided if the GA could be directed by a ranking method to converge only on acceptable solutions. For this work, an acceptable solution (or champion solution) is defined: Definition 2. A solution is an acceptable ....

....with a range dependent aggregation method such as sum of weighted objectives (just increase the weights) with a range independent method such as non dominated sorting , specifying importance is more complex. Fonseca forces a kind of importance with his preference articulation method [4], but this requires detailed knowledge of the ranges of the functions themselves, and is not a continuous guide to evolution. Thus, alternative methods of ranking multiobjective solutions are required, that are ideally range independent and allow the easy specification of importance, to enable ....

[Article contains additional citation context not shown here]

Fonseca, C. M, & Fleming, P. J., 1995b, Multiobjective Genetic Algorithms Made Easy: Selection, Sharing and Mating Restriction. Genetic Algorithms in Engineering Systems: Innovations and Applications, Sheffield, 45-52.

....are made clear. In order to carry out Pareto rank multiobjective optimization, it is necessary for multiple solutions to be compared with each other. Algorithms in which a pool of solutions exists concurrently, e.g. genetic algorithms and PRSA algorithms, excel at Pareto rank optimization [123]. Synthesis of Low Power Heterogeneous Distributed Systems In this chapter, we present MOGAC, an adaptive multiobjective genetic algorithm for hardware software co synthesis of distributed embedded systems. This algorithm, and its associated software implementation, solves the hardware software ....

C. M. Fonseca and P. J. Fleming, "Multiobjective genetic algorithms made easy: Selection, sharing and mating restrictions," in Proc. of Genetic Algorithms in Engineering Systems: Innovations and Applications, pp. 45--52, Sept. 1995.

....a b, if fi(a) fi(b) A 1, m : The solution a is said to be indifferent to a solution c, if neither solution is dominating the other one. When no a priori preference is defined among the objectives, dominance is the only way to determine, if one solution performs better than the other [16]. The best solutions to a multi objective problem are the Pareto ideal solutions, which represent the nondominated subset among all feasible solutions. In other words, starting from a Pareto solution, one objective can only be improved at the expense of at least one other objective. 2.2 ....

Fortseca, M.C., Fleming, P.J.: Multi-objective genetic algorithms made easy: Selection, sharing and mating restrictions. In: Proceedings of the 1st International Conference on Genetic Algorithms in Engineering Systems: Innovations and Application. (1995) 45-52

....and Thiele [99] In Paper [IV] a discussion of some of the most common algorithms is presented. Here just the multiobjective GA (MOGA) is described, since it is one of the cornerstones of the new multiobjective genetic algorithm being proposed. In the MOGA presented by Foseca and Fleming [24] and [25], each individual is ranked according to their degree of dominance. The more population members that dominate an individual, the higher the ranking for the individual. An individual s ranking equals the number of individuals that it is dominated by plus one (see Figure 13) Individuals on the ....

FONSECA C. M. AND FLEMING P. J., "Multiobjective genetic algorithms made easy: Selection, sharing and mating restriction," in Proceedings of 1st IEE/IEEE International Conference on Genetic Algorithms in Engineering Systems, Sheffield, England, 1995.

....lattice [thin lines] A random simplex of adjacent neurons is created [bold line] Within the simplex a uniformly distributed random point [plus symbol] is generated. 2. 3 Experimental Results The performance of the SOM MOEA is analyzed for the two objective test func tion of Fonseca and Fleming [4]: fu2 i exp ( xk 4 v T) 2 ) 5) with x. n [ 1, 1] The exact Pareto front is obtained for x. n = t, V t Vff n. The number of design variables is set to n = 10. An optimization run is started with the SOM MOEA and a population size of 60 individuals. A simple selection operator ....

Fortseca, M.C., Fleming, P.J.: Multi-objective genetic algorithms made easy: Selection, sharing and mating restrictions. Proceedings of the 1st International Conference on Genetic Algorithms in Engineering Systems: Innovations and Applications, London, UK, (1995) 45-52

.... on Pareto based methods because of their acknowledged advantages over aggregation and population based methods [5] Different techniques inferring a scalar value from the partially ordered objective vectors include the dominance level (or nondominated sorting [6, 15] the dominance grade or rank [4] and the strength measure [20] Some algorithms use a further means to reach a better or more uniform distribution Archiving Selection Elitism Primary Fitness Secondary Fitness population no archive only from population from population and archive only from archive dominance ....

....to bias the sampling rates accordingly. The relevance of density estimation in the context of (multi objective) evolutionary algorithms has been put forward by [5] where the authors noted that the standard fitness sharing concept is essentially the application of a kernel density estimator. In [4], existing results from kernel density estimation were used to derive guidelines for the fitness sharing parameters. Many advanced multi objective evolutionary algorithms use some form of density dependent selection. Furthermore, nearly all techniques can be expressed in terms of density ....

C. M. Fonseca and P. J. Fleming. Multiobjective genetic algorithms made easy: Selection, sharing and mating restrictions. In First Int'l Conf. on Genetic Algorithms in Engineering Systems: Innovations and Applications (GALESIA 95), pages 45--52, London, UK, 1995. The Institution of Electrical Engineers.

....de S sup erieures a 2, les individus de dernier rang peuvent avoir des probabilit es de s election n egatives. Dans ce cas, ils ne sont jamais s electionn es. ffl NDS (Non Dominated Sorting) dans cette m ethode, le rang d un individu est le nombre de solutions dominant l individu plus un [24]. Consid erons par exemple un individu c i a la g en eration t, qui est domin e par p t i individus dans la population courante. Son rang dans la population est donn e par : rang(c i ; t) 1 p t i Un individu non domin e de la population poss ede donc le rang 1 (fig.12) 25] Les rangs ....

....individus de la population. Elle retourne 1 si les deux individus sont identiques, 0 s ils sont diff erents ( a partir d un seuil donn e) et une valeur interm ediaire pour des niveaux de similarit e interm ediaires. La distance dist peut etre dans l espace de d ecision ou dans l espace objectif [24], et d epend g en eralement du probl eme trait e. Le maintien de la diversit e dans l espace objectif peut ne pas g en erer une diversit e dans l espace de d ecision, qui peut etre importante pour le d ecideur. L algorithme peut ne pas trouver de multiples solutions pour des probl emes o u ....

C. M. Fonseca and P. J. Fleming. Multiobjective genetic algorithms made easy: selection, sharing and mating restrictions. In IEEE Int. Conf. on Genetic Algorithms in Engineering Systems: Innovations and Applications, pages 45--52, Sheffield, UK, 1995.

....evaluation function becomes mono objective, using ranking methods to sort the population according to the definition of Pareto dominance. The GA handle diversity using sharing, and elitism is used to speedup the search. We use the ranking function that is proposed by Fonseca and Fleming in [FF95a]. An individual i of the population, dominated by k individuals, obtains the rank k 1. The ranking is based on the three following objective functions: minimize the number of sites, minimize the overall interference, and minimize the traffic loss. To maintain diversity along the Pareto frontier, ....

C.M. Fonseca and P.J. Fleming. Multiobjective genetic algorithms made easy: selection, sharing and mating restrictions. In IEEE Int. Conf. on Genetic Algorithms in Engineering Systems: Innovations and Applications, pages 45-- 52, Sheffield, UK, 1995.

....pressure and to the sharing pressure applied. Small values of the tournament size (close to 1 2 of the population size) results in too many dominated individuals (i.e. a very fuzzy front) while higher values (more than 20 ) result in premature convergence to a small portion of the front [6]. Finally, a small percentage of random individuals were introduced in each generation to make the GA more sensible to new zones [6] These individuals are assigned to zones not yet represented by any of the features that we seek to identify. In our formulation, uptrends were defined by means of ....

.... in too many dominated individuals (i.e. a very fuzzy front) while higher values (more than 20 ) result in premature convergence to a small portion of the front [6] Finally, a small percentage of random individuals were introduced in each generation to make the GA more sensible to new zones [6] . These individuals are assigned to zones not yet represented by any of the features that we seek to identify. In our formulation, uptrends were defined by means of logical expressions based on comparison of successive peaks and valleys in the time series. The definition of uptrend is a soft ....

C. Fonseca and P. Fleming. Multiobjective genetic algorithms made easy: Selection, sharing and mating restriction. In Proc. the First IEE/IEEE Intl. Conf. on Genetic Algorithms in Engineering Systems, pages 44--52, 1995.

....its one decision variable implies a large search space should be used when testing an MOEA. MOP2 is also an unconstrained two objective MOP, having the additional advantage of arbitrarily adding decision variables without changing PF true s structure; P true is also given in closed form [4]. Figures 1 and 2 shows MOP2 s Pareto optimal set and front. Finally, we propose MOP3 [20] This MOP has three objective functions, and its Pareto front appears to be a k dimensional curve following a convoluted path through objective space. This characteristic should challenge an MOEA s ability ....

Fonseca, Carlos M. and Peter J. Fleming. "Multiobjective Genetic Algorithms Made Easy: Selection, Sharing, and Mating Restriction." Proceedings of the 1st International Conference on Genetic Algorithms in Engineering Systems: Innovations and Applications. Number 414. 45--52. September: IEE, 1995.

....assignments can be made on an exponential rather than a linear ranking scale. The general implementation would be to assign fitnesses of 1; s j ; s j 1 ; from best to worst, where s 1:0 and j is some positive integer. Both s and j may be varied to obtain the desired level of pressure [27, 41]. Ranking methods employed in this research use such a scale, with s = P Gamma1 P and j = 3. Roulette Wheel and Tournament Selection Stochastic selection mechanisms following either a proportionate fitness assignment or a ranking scheme appear in several variations in the literature. The ....

....of objective B is less than some B niche . These tolerances may be static or dynamic according to latest population statistics. In either case, they are usually based on a calculated optimal separation distance, determined by dividing the assumed frontier surface area by the size of the population [27]. Note that if X is in the niche of Y then the converse is true, and that as the number of objectives increases, the time averaged niche density decreases for a given population size. Niching should be a concern in any Pareto GA, not only those employing tournament selection. Niche counts may be ....

C. M. Fonseca and P. J. Fleming. Multiobjective genetic algorithms made easy: Selection, sharing and mating restriction. In Proceedings of the First International Conference on Genetic Algorithms in Engineering Systems: Innovations and Applications, Sheffield UK, September 1995.

....algorithm used by CORDS to optimize PE allocations, communication resource allocations, task assignments, and communication resource connectivities. This algorithm shares some properties with parallel recombinative simulated annealing algorithms [11] and multiobjective genetic algorithms [12]. CORDS maintains a pool of architectures. A generation is a discrete unit of time. In every generation, architectures reproduce. The new architectures mutate and trade information with each other. The architectures are then ranked, relative to each other. Poor quality architectures are eliminated ....

....depend on the lost PE such that none of the tasks or communication resources depend on the lost PE. Communication resource allocation mutation is analogous to PE allocation mutation. Information trading: CORDS uses an evolutionary algorithm which is based on two types of genetic algorithm [11] [12]. However, each of these algorithms has problems dealing with the optimization of multi dimensional information. Below, we describe these problems and explain how CORDS avoids them. In a genetic algorithm, each architecture is represented by a string, i.e. a linear array, of values. Genetic ....

[Article contains additional citation context not shown here]

C. M. Fonseca and P. J. Fleming, "Multiobjective genetic algorithms made easy: Selection, sharing and mating restrictions, " in Proc. Genetic Algorithms in Engineering Systems: Innovations and Applications, pp. 45--52, Sept. 1995.

.... capable of true multiobjective optimization, exploring the set of solutions which can only be improved in one way by being degraded in another (the Pareto optimal set) instead of collapsing all costs into one with a weighted sum, as is the case for most other probabilistic optimization algorithms [24] [26] Saha s exploratory work demonstrates that genetic algorithms can be applied to the hardware software partitioning problem [27] A number of simplifying assumptions are made in Saha s system, however. Only one software processor is allowed, there are no provisions for synthesizing systems ....

C. M. Fonseca and P. J. Fleming, "Multiobjective genetic algorithms made easy: Selection, sharing and mating restrictions," in Proc. Genetic Algorithms in Engineering Systems: Innovations and Applications, pp. 45--52, Sept. 1995.

....algorithms are especially useful for simultaneously optimizing more than one cost. Conventional iterative improvement and simulated annealing algorithms maintain only one solution at a time. Most single solution optimization algorithms collapse all costs into a single value with a weighted sum [21], 22] Genetic algorithms maintain a pool of solutions which evolve in parallel. This allows solutions to be ranked relative to each other. Genetic algorithms are capable of true multiobjective optimization, exploring the Pareto optimal set of solutions, i.e. those solutions which are better ....

C. M. Fonseca and P. J. Fleming, "Multiobjective genetic algorithms made easy: Selection, sharing and mating restrictions," in Proc. Genetic Algorithms in Engineering Systems: Innovations and Applications, pp. 45--52, Sept. 1995.

....of the trade off surface contribute to the quality more than convex parts and may, as a consequence, dominate the performance assessment. Finally, the distribution as well as the extent of the nondominated front is not considered. Another interesting way of performance assessment was proposed by Fonseca and Fleming (1996) Given a set X 0 X of nondominated points, a boundary function divides the search space into two regions: the points not dominated by or equal to members of X 0 and the points covered by X 0 . They call this particular function, which can also be seen as the locus of the family of tightest goal vectors ....

Fonseca, C. M. and P. J. Fleming (1995a). Multiobjective genetic algorithms made easy: Selection, sharing and mating restrictions. In First International Conference on Genetic Algorithms in Engineering Systems: Innovations and Applications (GALESIA 95), pp. 45--52. The Institution of Electrical Engineers.

....on control engineering applications of GAs, including those discussed in Chapter 2. The full bibliography appeared in volume 7, number 18 (July 1, 1993) of the electronicmail digest GA List aic.nrl.navy.mil. A survey and discussion of evolutionary algorithms in multiobjective optimization (Fonseca and Fleming, 1995d) Here, current evolutionary approaches to multiobjective optimization are critically reviewed, and some of the issues raised by multiobjective optimization in the context of evolutionary search are identified. A unified, scale independent preference articulation framework for multiobjective ....

....approaches to multiobjective optimization are critically reviewed, and some of the issues raised by multiobjective optimization in the context of evolutionary search are identified. A unified, scale independent preference articulation framework for multiobjective and constrained optimization (Fonseca and Fleming, 1995b) Combining goal and priority information with the notion of Pareto dominance results in a transitive relational operator (preferability) of which Paretodominance is a particular case. This operator allows sets of candidate solutions to be ranked according to (absolute) goal information and any ....

[Article contains additional citation context not shown here]

Fonseca, C. M. and Fleming, P. J. (1995a). Multiobjective genetic algorithms made easy: Selection, sharing and mating restriction. In (GALESIA, 1995), pages 45--52.

....(f 1 (x) fn (x) The problem has usually no unique, perfect solution, but a set of equally efficient, or non inferior, alternative solutions, known as the Pareto optimal set. Multiobjective optimization using genetic algorithms has been investigated by many authors in recent years e.g. (Fonseca Flemming, 1995a; Shaffer, 1985; Valenzuela Uresti, 1997; Veldhuizen, 1999; Zitzler Thiele, 1998) among others. However in some realworld optimization problems the application of genetic algorithms is limited by the requirement of finding an adequate codification of the problem. Additionally the parameters ....

....codification of the problem. Additionally the parameters associated with the evolutionary strategies, such as: weight factors used in function aggregation, priorities used in lexicographic ordering and Pareto ranking, niche size in niche induction techniques, tournament size, etc. e.g. see (Fonseca, 1995a) need to be established. Finding the appropriate values for the parameters in evolutionary algorithms is most of the time as difficult as the main problem solution. In Operations Research, many techniques (more that 20 (Coello, 1999) have been developed over the years to try to deal with ....

[Article contains additional citation context not shown here]

Fonseca, Carlos M. and Flemming Peter J. (1995b) "Multiobjective Genetic Algorithms Made Easy: Selection, Sharing, and mating Restriction." Proceedings of the 1st International Conference on Genetic Algorithms in Engineering Systems: Innovations and Applications.pp. 45-52. September: IEEE, 1995.

....in some meaningful way. This means predefining priorities and preferences, adjusting weights within the function to accurately reflect the decision making process, and even creating a search space which may not effectively represent the problem. The multi objective genetic algorithm presented by Fonseca and Fleming (1995) is attractive as it avoids this problem. The technique uses Pareto optimisation to search from a selection of possible solutions rather than working in a single search direction. This allows us to optimise schedules according to multiple criteria, and allows a (human) decision maker to pick out ....

....have a choice of lines and without this measure, it is possible that a large number of products might be made on the same line. Implementation of the MOGA for the scheduling problem The implementation of the MOGA scheme applied to the problem is based on Pareto optimality, as suggested by Fonseca and Fleming, 1995). When presented with a population, it evaluates each cost separately for each individual, and picks out the non dominated solutions solutions such that the set of costs found each cannot have any cost contained within them improved without adversely affecting another cost as possible ....

[Article contains additional citation context not shown here]

Fonseca, C. M. & Fleming, P.J., (1995). Multiobjective genetic algorithms made easy: Selection, sharing and mating restriction, Proceedings of the First IEE/IEEE International Conference on Genetic Algorithms in Engineering Systems: Innovations and Applications, pp. 44 - 52, The Institution of Electrical Engineers.

No context found.

Fonseca, M.C., Fleming, P.J.: Multi-objective genetic algorithms made easy: Selection, sharing and mating restrictions. In: Proceedings of the 1st International Conference on Genetic Algorithms in Engineering Systems: Innovations and Application. (1995) 45--52

No context found.

M. C. Fonseca and P. J. Fleming, "Multi-objective genetic algorithms made easy: Selection, sharing and mating restrictions", Proceedings of the 1st International Conference on Genetic Algorithms in Engineering Systems: Innovations and Applications, No. 414, pp. 45-52. IEEE, 1995

No context found.

Fonseca C. M. and J. F. P., "Multiobjective genetic algorithms made easy: Selection, sharing and mating restriction," presented at 1st IEE/IEEE International Conference on Genetic Algorithms in Engineering Systems, Sheffield, England, 1995.

No context found.

Fonseca, C. M, & Fleming, P. J. (1995b). Multiobjective Genetic Algorithms Made Easy: Selection, Sharing and Mating Restriction. Genetic Algorithms in Engineering Systems: Innovations and Applications (GALESIA 95), Sheffield (pp. 45-52).

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