A. Juels and M. Wattenberg. Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Technical Report Technical Report CSD-94-834. Computers Science Department, University of California at Berkeley, USA, 1995.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... pattern search is chosen because multistart type algorithms are always one of the most popular methods in practice[4] and have been demonstrated to work very well and outperform many more sophisti12 cated algorithms, such as genetic algorithm and simulated annealing, in many practical problems[16, 17, 18]. Pattern search[14] is one of direct search techniques which are usually recommended for black box optimization problems[19] In the tests, the search algorithms are executed on each function with the function dimension varying from 20 to 2000. We have used the following parameters for the RRS ....

A. Juels and M. Wattenberg. Stochastic hillclimbing as a baseline method for evaluating generic algorithms. In D. S. Touretzky, M. C. Mozer, and M. E. Hasselmo, editors, Advances in Neural Information Processing Systems, volume 8, pages 430--436. 1996.

.... work very well for many practical problems, for example, it produced excellent solutions on practical computer vision tasks[16] outperformed simulated annealing on the traveling salesman problem (TSP) 17] and outperformed genetic algorithms and genetic programming on several large scale testbeds[18]. Our version of multi start hillclimbing uses pattern search[12] as its local search method since it is one of direct search techniques which are usually recommended for blackbox optimization problems[14] 13 We first test the scalability of our algorithm to high dimensional problems. We apply ....

A. Juels and M. Wattenberg. Stochastic hillclimbing as a baseline method for evaluating generic algorithms. In D. S. Touretzky, M. C. Mozer, and M. E. Hasselmo, editors, Advances in Neural Information Processing Systems, volume 8, pages 430--436. 1996.

....to establish the conditions in which an optimization method surpasses another [8] 11] 4] and led to Wolpert MacReady s NFL Theorem [12] This theorem states that is not possible to establish the superiority of a method over any other when they are averaged over all the possible functions. In [6][5] 10] can be found many real world problems in which a simple Stochastic Hill Climbing surpasses other methods as the Genetic Algorithms [2] Taboo Search or Simulated Annealing. This work follows that research line in the multiobjective optimization field. 2. ADAPTATION OF THE METHODS TO ....

....via a weighted sum. The weights are chosen in a random way in each generation. In that manner the direction of the search will vary during the evolution, leading to a greater distribution of the obtained Pareto Set. 2. 2 Stochastic Hill Climbing The Stochastic (SHC) is a very simple method [5][6][10] It starts from an initial random solution. Later a set of iterations is made, in each one of them an unary mutation operator is applied over the current solution. If the new solution is better than the previous one, then the new solution will be taken as the current solution for the next ....

[Article contains additional citation context not shown here]

Juels, A., Watenberg, M: "Stochastic Hill-Climbing as a Baseline Method for Evaluating Genetic Algorithms", Tech. Report, University of California at Berkeley, 1994.

....because almost everybody thinks that this simple algorithm is not very useful. In general, it is hard to identify a priori which algorithm will be the best for a specific task. We hope that this paper will induce the authors of EA papers to use SHC as a baseline method in order to show EA power [Juels and Wattenberg, 1994]. In this paper, we also compare many Evolutionary Algorithms in the context of Graph Drawing (Section 3) Graph Drawing addresses the problem of finding a representation of a graph that satisfies a given aesthetic objective. The representation of a graph is often given by an embedding of its ....

....of evaluations are done. Some versions of this simple algorithm have been developed. For instance, Multiple SHC is a version of SHC that restarts the search after a given number of fitness evaluations are completed. An extended discussion of versions of SHC is in [Baluja, 1995; Jones, 1995; Juels and Wattenberg, 1994; Yuret, 1994] Two basic version of SHC are: Next Ascent SHC (NA SHC) It works as it was explained in the previous paragraph. NA SHC explores the neighborhood until a solution equal or better than the current one is found. Then, the algorithm takes this new solution as the current one. ....

[Article contains additional citation context not shown here]

Juels, A., Wattenberg, M.: Stochastic Hill-climbing as a Baseline Method for Evaluating Genetic Algorithms, Technical Report, University of California at Berkeley, 1994.

....look at the performance of random search and simple hill climbing techniques on all problems being examined. There is no need to even consider the use of more complex search strategies (such as genetic algorithms) if the given problem is effectively solved by random search or a simple hill climber (Juels Wattenberg, 1994; Lang, 1994; Baluja, 1995; McIlhagga, Husbands, Ives, 1996b; Zitzler Thiele, 1998; Jakob, Gorges Schleuter, Sieber, 1998; Langdon Poli, 1998) 1.3 The contributions of the thesis As a whole the thesis is an advocation of the new approach to ESA research described above. As such, the ....

....papers in which the basic story is that hill climbing has been shown to outperform an evolutionary search algorithm of a more complex nature. We shall look at a few good examples. Lang (1994) shows how Genetic programming is outperformed by stochastic hill climbing when designing boolean circuits. Juels and Wattenberg (1994) compares various different search algorithms on 4 test problems and argues that because stochastic hill climbing often performs well compared to more complex algorithms, such as GAs, that it should be used as a base line method for evaluating the effectiveness of ESAs. Only if the ESA outperforms ....

[Article contains additional citation context not shown here]

Juels, A., & Wattenberg, M. (1994). Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Tech. rep., University of California at Berkeley.

.... 387, 388] Swiss Federal Institute of Technology (ETH) 218] Technische Universit at der Berlin, 211, 353, 354, 397] The University of Texas at Austin, 394] Tulane University, 52] Universidad de Granada, 183, 243, 244, 377] University of Bristol, 251] University of California at Berkley, [19] University of Cambridge, 72] University of Durham, 396] University of Edinburgh, 42] University of Granada, 221, 239, 240, 242] University of Illinois at Urbana Champaign, 189, 191, 217, 223, 235, 362, 363, 364, 383, 389] University of Nebraska Lincoln, 108, 322] Patents 13 University ....

....Heikki, 199] Hy otyniemi, Heikki, 257, 304] Ikonen, Ilkka, 305] Iwamoto, Takashi, 348] Jain, L. C. 84] Jakobsson, Matti, 162] Janikow, Cesary Z. 18] J aske, Harri, 258] Johnson, R. P. 84] Johnsson, Mika, 311] Jokinen, Hannu, 70] Jong, Kenneth A. De, 28, 38, 51, 87] Juels, Ari, [19] Juhola, Martti, 326] Julstrom, Bryant A. 259, 277, 306] Kalganova, Tatiana, 307] Kampen, Antoine H. C. van, 260] Kampen, W. van, 172] Kargupta, Hillol, 226] Kasik, David J. 71] Kaski, Kimmo, 213] Kaukoranta, Timo, 302] Kazarlis, Spyros A. 151, 200] Keskinen, Kari I. 184] ....

[Article contains additional citation context not shown here]

Ari Juels and Martin Wettenberg. Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Technical Report CSD-94-834, University of California at Berkley, Department of Computer Science, 1994. (also as [?]) Key: ga94aJuels.

.... Fe Institute, 168, 315, 342, 343] Swiss Federal Institute of Technology (ETH) 178] Technische Universitat der Berlin, 171, 308, 309, 352] The University of Texas at Austin, 349] Universidad de Granada, 142, 203, 204, 332] University of Bristol, 211] University of California at Berkley, [18] University of Cambridge, 53] University of Durham, 351] University of Granada, 181, 199, 200, 202] University of Illinois at Urbana Champaign, 148, 150, 177, 183, 195, 317, 318, 319, 338, 344] University of Nebraska Lincoln, 85, 282] University of Strathclyde, 345, 346, 347] University ....

....Heikki, 158] Hyotyniemi, Heikki, 217, 264] Ikonen, Ilkka, 265] Iwamoto, Takashi, 303] Jain, L. C. 64] Jakobsson, Matti, 120] Janikow, Cesary Z. 17] Jaske, Harri, 218] Johnson, R. P. 64] Johnsson, Mika, 271] Jokinen, Hannu, 51] Jong, Kenneth A. De, 25, 30, 40, 66] Juels, Ari, [18] Juhola, Martti, 285] Julstrom, Bryant A. 219, 237, 266] Kalganova, Tatiana, 267] Kampen, Antoine H. C. van, 220] Kampen, W. van, 131] Kargupta, Hillol, 186] Kasik, David J. 52] Kaski, Kimmo, 173] Kaukoranta, Timo, 262] Kazarlis, Spyros A. 109, 159] Keskinen, Kari I. 143] ....

[Article contains additional citation context not shown here]

Ari Juels and Martin Wettenberg. Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Technical Report CSD-94-834, University of California at Berkley, Department of Computer Science, 1994. (also as [?]) Key: ga94aJuels.

....3. y : x 4. Flip each bit in y independently with probability p(n) 5. If f(y) # f(x) set x : y. 6. Continue at 3. It is a common experience that simple hill climbers are often able to find solutions that are at least comparable to those of more sophisticated evolutionary algorithms [8]. Setting p(n) 1 n implies that during one mutation step on average one bit flips. Thus, the (1 1) EA may be regarded as a kind of randomized hill climber. In fact, the most recommended fixed choice for the mutation probability p(n) is 1 n [1, 9] For linear functions choosing p(n) #(1 n) can ....

Juels, A. and Wattenberg, M. (1994). Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Tech. Report CSD-94-834 Univ. of California.

....cases, and the task of a search algorithm is to find good enough objects within a reasonable time. There are numerous well documented search algorithms, such as genetic algorithms (Goldberg, 1989) evolution strategies (Schwefel Rudolph, 1995) simulated annealing (Davis, 1987) and hill climbing (Juels Wattenberg, 1994). The difference between each is the way in which they decide where in the search space to look next: i.e. how they decide which of the billions of possible objects to spend precious time evaluating (figure 3) Each possible way of choosing the next object to evaluate is called a search bias and ....

....A Offspring before mutation Fitness Figure 8: Cross over implements the search bias of assuming that better individuals will exist between the encodings of two good individuals. But this will not always be the case; in fact often much simpler search methods compare favourably (Baluja, 1995; Juels Wattenberg, 1994). Whether or not time is a concern depends on the motivation for the use of search. In general the choice of search algorithm should depend only on the role that search will play within a research project. More will be said on this later. in summary So, putting it all together, most ....

[Article contains additional citation context not shown here]

Juels, A., & Wattenberg, M. (1994). Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Tech. rep., University of California at Berkeley.

....techniques, if applicable, will out perform the best GA. Neither are GAs necessarily more precise or efficient than other non deterministic methods, such as simulated annealing. In fact, their performance in certain test cases has been shown inferior even to simple stochastic hill climbing [52]. They are not well suited for applications that require guaranteed response times (such as such as realtime control systems) their response time variance is in fact quite high in relation to other stochastic methods [62] 3 Rather, the distinguishing characteristics of GAs are their ....

A. Juels and M. Wattenberg. Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Technical Report CSD-94-834, University of California at Berkely, September 1994.

.... Algo MT MT MT rithme 6 Theta 6 10 Theta 10 20 Theta 5 Carlier and Pinson [CP89] B B 55 930 1165 Nakano and Yamada [NY91] EA 55 965 1215 Nakano and Yamada [NY92] EA 55 930 1184 Dorndorf and Pesh [DP92] 55 938 1178 Fang, Ross and Corne [FRC93] EA 949 1189 Juels and Wattenberg [JW94] EA 937 1174 Soares [Soa94] EA 58 997 Kobayashi et al. [KOY95] EA 930 Our results EA 55 953 1180 Optimum 55 930 1165 Table 5. This table displays the results obtained by different authors on the Muth and Thompson problems. The first column gives the references to the work. The ....

....to the work. The second column indicates the kind of algorithm that have been used. The three subsequent columns give the results that are reported (best found makespan) for the 3 problems. NY91] uses an indirect representation. NY92] uses a direct representation with the GA GT operator. [JW94] uses the representation 2. They use an other recombination operator. Using an AE, only Nakano and Yamada [NY92] and [KOY95] have obtained the optimum of the MT10 Theta 10. No one has obtained the optimum of the MT20 Theta 5. It is noteworthy that for the big problems, the research space is ....

Ari Juels and Martin Wattenberg. Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Technical Report csd-94-834, University of California, 1994.

.... Algo MT MT MT rithm 6 Theta 6 10 Theta 10 20 Theta 5 Carlier and Pinson [CP89] B B 55 930 1165 Nakano and Yamada [NY91] EA 55 965 1215 Nakano and Yamada [NY92] EA 55 930 1184 Dorndorf and Pesh [DP92] 55 938 1178 Fang, Ross and Corne [FRC93] EA 949 1189 Juels and Wattenberg [JW94] EA 937 1174 Soares [Soa94] EA 58 997 Kobayashi et al. [KOY95] EA 930 Our results (best obtained results) EA 55 937 1178 Optimum 55 930 1165 Table 2. This table displays the results obtained by different authors on the Muth and Thompson instances. The first column gives the ....

....the references to the work. The second column indicates the kind of algorithm that have been used. The three subsequent columns give the results that are reported (best found makespan) for the 3 instances. NY91] uses an indirect encoding. NY92] uses a direct encoding with the GA GT operator. [JW94] uses the indirect encoding. They use an other recombination operator. The Branch and Bound (B B) is an exact method. Carlier and Pinson [CP89] have obtained the optimal solutions using this method. Using an EA with an indirect representation, Nakano et Yamada [NY91] have obtained the optimal ....

Ari Juels and Martin Wattenberg. Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Technical Report csd-94-834, University of California, 1994.

....to perform selection, crossover, mutation, and replacement as well as the choice of the concrete representation of the individuals o ers a great variety of di erent EAs. The analysis here concentrates on the most simple variant of an EA that is still of theoretical and practical interest [9,16]. We restrict the size of the population to just one individual and do not use crossover. Since we assume the objective function to have Boolean inputs we represent the current individual as a bit string. This is the usual choice for Genetic Algorithms. We use a bitwise mutation operator that ips ....

A. Juels and M. Wattenberg, Stochastic hillclimbing as a baseline method for evaluating Genetic Algorithms, (Technical Report, University of California, Computer Science Department, CSD-04-834, 1994).

....process is that 2 as far as we understand this natural process, and we are able to simulate it Reference MT10 Theta 10 MT20 Theta 5 best number of sampled points pop size prob. success best number of sampled points pop size prob. success [NY92] 930 1184 [DP92] 938 1178 [FRC93] 949 1189 [JW94] 937 1174 [Soa94] 997 20000 100 [DPT95] ours) 953 500000 250 1180 500000 250 Table 1: Comparing results from different authors is not easy According to what we have found in several articles, we display all found characteristics. We have not provided the probability of success for our results ....

Ari Juels and Martin Wattenberg. Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Technical Report csd-94-834, University of California, 1994.

....than ours this often is the only reasonable thing to do. Stochastic Hill Climbing (SHC) SHC may be the simplest elitist algorithm using direct search in policy space. It should be mentioned, however, that despite its simplicity SHC often outperforms more complex elitist methods such as GAs (Juels Wattenberg 1996). Anyway, SHC and more complex elitist algorithms such as Genetic Algorithms and Evolution strategies are equally affected by the central question of this paper: how many trials should be spent on the evaluation of a given policy We implement SHC as follows: 1. Initialize policy p to 0.5, and ....

Juels, A., and Wattenberg, M. 1996. Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. In Touretzky, D. S.; Mozer, M. C.; and Hasselmo, M. E., eds., Advances in Neural Information Processing Systems, volume 8, 430--436. The MIT Press, Cambridge, MA.

....and efficient way of solving this particular problem. Stochastic Hill Climbing (SHC) SHC is one of the simplest elitist algorithms using direct search in policy space. It should be mentioned, however, that despite its simplicity SHC often outperforms more complex elitist methods such as GAs [3]. We implement SHC as follows: 1. Initialize policy p to 0.5, and real valued variables BestP olicy and BestResult to p and 0, respectively. 2. If there have been more than 30000 Gamma T rialLength pulls then exit (T rialLength is an integer constant) Otherwise evaluate p by measuring the ....

A. Juels and M. Wattenberg. Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. In David S. Touretzky, Michael C. Mozer, and Michael E. Hasselmo, editors, Advances in Neural Information Processing Systems, volume 8, pages 430--436. The MIT Press, Cambridge, MA, 1996.

....methods were also compared against special case greedy heuristics. The results suggest that although it is possible to get good results with genetic algorithms, a similar performance can be achieved faster by using stochastic algorithms with simpler structures, which supports the observations in [14, 9, 8]. Nevertheless, genetic algorithms appear to be quite consistent in the sense that variance in the quality of the results obtained is very small. The results of this work are currently being exploited in a major industrial software project. Acknowledgments This research has been supported by the ....

A. Juels and M. Wattenberg. Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Technical Report CSD-94-834, Department of Computer Science, University of California at Berkeley, 1994.

.... The aim of this paper is to compare two standard genetic algorithms with simpler methods of optimization: multiple restart stochastic hillclimbing (MRSH) and population based incremental learning (PBIL) Previous comparisons between forms of MRSH and GAs can be found in [Ackley, 1994] Juels Wattenberg, 1994], Forrest Mitchell, 1992] Mitchell Holland, 1994] and [Davis, 1991] to name a few. A comparison between GAs and PBIL has been made in [Baluja, 1994] Baluja Caruana, 1995] This paper provides a large scale empirical comparison of these algorithms on problems commonly found in GA ....

....the report and suggests some areas for future studies. page 5 2. MULTIPLE RESTART STOCHASTIC HILLCLIMBING Multiple restart stochastic hillclimbing (MRSH) is a method of iterative optimization of static functions. It is the simplest of the optimization procedures explored in this paper. [Wattenberg and Juels, 1994] have compared one version of stochastic hillclimbing with GAs on several problems commonly used for gauging genetic algorithms and genetic programming, and have achieved very promising results. The basic stochastic hillclimbing algorithm is shown in Figure 1. Three variants of this algorithm are ....

[Article contains additional citation context not shown here]

Wattenberg, M. & Juels, A. (1994) "Stochastic Hillclimbing as a Baseline Method for Evaluating Genetic Algorithms," University of California - Berkeley, Technical Report. CSD-94-834.

.... Reference Algo MT MT MT rithm 6 Theta 6 10 Theta 10 20 Theta 5 Carlier and Pinson [CP89] B B 55 930 1165 Nakano and Yamada [NY91] EA 55 965 1215 Nakano and Yamada [NY92] EA 55 930 1184 Dorndorf and Pesh [DP92] 55 938 1178 Fang, Ross and Corne [FRC93] EA 949 1189 Juels and Wattenberg [JW94] EA 937 1174 Soares [Soa94] EA 58 997 Kobayashi et al. [KOY95] EA 930 Our results (best obtained results) EA 55 937 1178 Optimum 55 930 1165 Table 2. This table displays the results obtained by different authors on the Muth and Thompson instances. The first column gives the references ....

....the references to the work. The second column indicates the kind of algorithm that have been used. The three subsequent columns give the results that are reported (best found makespan) for the 3 instances. NY91] uses an indirect encoding. NY92] uses a direct encoding with the GA GT operator. [JW94] uses the indirect encoding. They use an other recombination operator. The Branch and Bound (B B) is an exact method. Carlier and Pinson [CP89] have obtained the optimal solutions using this method. Using an EA with an indirect representation, Nakano et Yamada [NY91] have obtained the optimal ....

Ari Juels and Martin Wattenberg. Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Technical Report csd-94-834, University of California, 1994.

....that although it is possible to get good results with genetic algorithms, a similar performance can be achieved more quickly by using stochastic algorithms with simpler structures. Similar results has been Lahtinen et al.: Empirical comparison of stochastic algorithms . 59 obtained earlier in [12, 7, 6]. In the future, we will continue to investigate this question and extend our ongoing experimentation in different problem domains. Acknowledgments This research has been supported by the Technology Development Center (TEKES) and Nokia Research Center. ....

A. Juels and M. Wattenberg. Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Technical Report CSD-94-834, Department of Computer Science, University of California at Berkeley, 1994.

....GAucsd1.4 initially written by Grefenstette (GENESIS) and modified by Schraudolph. This software was written to be be used for research into Genetic Algorithms. This software was written to support binary encoding and was extended to real encoding. GAucsd1.4 has been used by Juels and Wattenberg [19], Wright [31] 2.11 Travelling salesperson problem TSP The travelling salesperson problem is a scheduling problem in which there are N towns and the salesperson must visit each town only once and return to the starting town. The cost distance of travelling between each pair of towns is given. The ....

Ari Juels and Martin Wattenberg. Stochastic hill climbing as a baseline method for evaluating genetic algorithms. 1994.

....uniformly better performance over the GA results presented here. Figure 11: Venus, femur and pelvis models page 15 The third method explored is next ascent hillclimbing (NAH) described in section 3.2. This was able to uniformly do better than any of the genetic algorithms attempted. Recently, [Juels and Wattenberg, 1994] have compared stochastic hillclimbing methods with GAs and have found similar results. For the runs reported in this paper, NAH was restarted multiple times until the total number of evaluations equalled those used by the GA and PBIL methods. The best evaluation found, over the multiple restarts, ....

Juels, A. and Wattenberg, M. (1994). Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Technical Report CSD-94-834, University of California - Berkeley.

....has better or equal fitness. The algorithm always maintains the fittest solution of all candidates examined in addition to a current solution. Comparative investigations of this sort of algorithm with GAs or even GP specifically have been previously conducted (e.g. O Reilly and Oppacher 1994a; Juels and Wattenberg 1994; Jones 1995; Lang 1995] The crossover (stochastic iterated) hill climbing algorithm of this chapter uses GP crossover as the operator which generates the candidate solution. At the outset, a mate and the current solution are randomly generated. For a specified number of attempts, parameter ....

Juels, A. and Wattenberg, M. (1994). Stochastic hill climbing as baseline methods for evaluating genetic algorithms. Technical report, Computer Science Dept., University of California, Berkely.

....containing the feasible designs. One can check whether the fitness function for the satellite docking problem exhibits this property by performing a large number of statistical hillclimbing runs. In the experiments for this problem, only 8 out of 480 single start statistical hill climbing runs [6] converged to a feasible solution that is approximately 1.7 . All the other runs got stuck in an infeasible local maximum. Although 1.7 is not extremely small, it does indicate that it is best to use an algorithm that can avoid local maxima for instance, the agent based genetic algorithm. In ....

....cost for finding a feasible solution. 8 Performance Analysis This section compares the performance of the agentbased genetic algorithm with multiple restart statistical hill climbing (MRSH) MRSH has been suggested as an adequate benchmark for evaluating the performance of genetic algorithms [1, 6]. The MRSH algorithm is implemented as 24 single start statistical hill climbing algorithms running in parallel on 24 Sparc workstations. On the other hand, the agent based genetic algorithm a total of 46 agents: 20 creation agents, 3 crossover agents, 3 MutateModule agents, 3 AddDeleteModule ....

Juels, A., and Wattenberg, M. 1994. Stochastic Hillclimbing as a Baseline Method for Evaluating Genetic Algorithms. Technical Report CSD--94--834. Computers Science Department, University of California at Berkeley.

.... XXX YYY, 1998, http: www.icsi.berkeley.edu jagota NCS 31 other interesting methods used for optimization include dynamic hill climbing (DHC) which seems to give similar convergence as the simulated quenching approach [24] GA compared to the multistart gradient methods are rarely more efficient [147, 148]. RasID algorithm has been used so far only for one real problem (gasoline blending problem [48] outperforming slightly backpropagation with momentum and adaptive learning rates. The NOVEL results were compared [20] with a number of other minimization methods in application to the twospiral and a ....

A. Juels and M. Wattenberg, Stochastic hillclimbing as a baseline method for evaluating genetic algorithm. Advances in Neural Information Processing Systems, MIT Press, Vol 8: 430-436, 1996.

.... optimisers; also the quality of results achieved by other mutation only techniques such as simulated annealing support the claim that mutation alone can form the basis of an effective optimisation technique other studies confirm this view [Schaffer et al. 89] Interestingly, a recent study [Juels Wattenberg 94] found that simple stochastic hill climbing methods were able to achieve results comparable, or superior, to GAs for certain problems. At the extreme, Fogel [Fogel Atmar 90] argues that crossover has no general advantage over mutation. The opposing camp point to empirical studies [Schaffer ....

Ari Juels and Martin Wattenberg. Stochastic Hillclimbing as a Baseline Method for Evaluating Genetic Algorithms. Technical report, UC Berkeley, 1994.

No context found.

A. Juels and M. Wattenberg. Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Technical Report Technical Report CSD-94-834. Computers Science Department, University of California at Berkeley, USA, 1995.

No context found.

Juels, A. and Wattenberg, M. (1995). Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Technical Report Technical Report CSD-94-834. Computers Science Department, University of California at Berkeley, USA.

No context found.

A. Juels and M. Wattenberg. Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Technical Report Technical Report CSD-94-834. Computers Science Department, University of California at Berkeley, USA, 1995.

No context found.

A. Juels and M. Wattenberg. Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Technical Report CSD-94-834, Department of Computer Science, University of California at Berkeley, USA, 18 July 1995.

No context found.

A. Juels and M. Wattenberg. Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Technical Report Technical Report CSD-94-834. Computers Science Department, University of California at Berkeley, USA, 1995.

No context found.

Ari Juels and Martin Wattenberg. Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Technical Report CSD-94-834, Department of Computer Science, University of California at Berkeley, USA, 18 1994.

No context found.

A. Juels, M. Wattenberg (1994) Stochastic Hillclimbing as a Baseline Method for Evaluating Genetic Algorithms. Technical Report, UC Berkeley

No context found.

A. Juels and M. Wattenberg. Stochastic hillclimbing as a baseline method for evaluating generic algorithms. In D. S. Touretzky, M. C. Mozer, and M. E. Hasselmo, editors, Advances in Neural Information Processing Systems, volume 8, pages 430--436. 1996.

No context found.

Ari Juels and Martin Wattenberg. Stochastic hillclimbing as a baseline method for evaluating genetic algorithms. Technical Report CSD-94-834, University of California, Berkeley, Berkeley, CA 94720, 1994.

No context found.

Juels, A., Wattenberg, M., \Stochastic Hillclimbing as a Baseline Method for Evaluating Genetic Algorithms," in Touretzky, D.S., et al., ed. Advances in NIPS 8, MIT Press, 1996. pp. 430-436.

No context found.

A. Juels, M. Wattenberg (1994) Stochastic Hillclimbing as a Baseline Method for Evaluating Genetic Algorithms. Technical Report, UC Berkeley

No context found.

Juels, A., and Wattenberg, M. 1994. Stochastic Hillclimbing as a Baseline Method for Evaluating Genetic Algorithms. Technical Report CSD--94--834. Computers Science Department, University of California at Berkeley.

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