Michael W. Trosset. On the use of direct search methods for stochastic optimization. Technical report, Department of Computational and Applied Mathematics, Rice University, 2000.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... 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 algorithm: p = 0.99, r = 0.1, c = 0.5, # = 0.8, q = 0.99, s t = 0.001. The test results for each benchmark function are shown ....

Michael W. Trosset. On the use of direct search methods for stochastic optimization. Technical report, Department of Computational and Applied Mathematics, Rice University, 2000. 23

....in our problem may be superimposed with noises and hence the accurate derivative estimation cannot be obtained. Direct search methods, which do not use any derivative information of objective functions, are widely used in practice and often recommended for optimization in the presence of noise[5, 12]. However, like other local search algorithms, the performance of traditional direct search methods, such as Nelder Mead simplex method[13] and pattern search[14] are still susceptive to the effect of noises and may be degraded in higher dimension problems[15, 6] Therefore, we have used a new ....

Michael W. Trosset. On the use of direct search methods for stochastic optimization. Technical report, Department of Computational and Applied Mathematics, Rice University, 2000.

....0.99 Figure 3: Convergence curve of random sampling with probability 0.99 be obtained. Direct search methods, such as Nelder Mead simplex method[11] and pattern search[12] do not exploit the derivative of the objective functions, are often recommended for optimization in the presence of noise[13, 14]. However, like other local search algorithms, they are still susceptible to the effect of noise and their performance may greatly degrade in high dimensional problems with many parameters[15, 4] Therefore, we propose a new direct search approach in our algorithm which is based on the high ....

.... 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 the search algorithms to Rastrigin function with its dimension n varying from 20 to 200, repeat each test with randomly selected starting points for 50 times and take the average of the results from these ....

Michael W. Trosset. On the use of direct search methods for stochastic optimization. Technical report, Department of Computational and Applied Mathematics, Rice University, 2000.

No context found.

Michael W. Trosset. On the use of direct search methods for stochastic optimization. Technical report, Department of Computational and Applied Mathematics, Rice University, 2000.

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