H.-G. Beyer and D. V. Arnold. Fitness noise and localization errors of the optimum in general quadratic fitness models. In Proceedings of the Genetic and Evolutionary Computation Conference (GECCO), pages 817--824, San Francisco, CA, 1999. Morgan Kaufmann.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....noise into account in order to decrease the gap between simulated and real world robots[11] In such cases, Evolutionary Algorithms (EA) 2, 6, 8, 10, 19] work well even in the presence of noise. In fact, many EA approaches have been applied to optimizing noisy objective functions effectively[1, 3, 4, 9, 15]. Back and Hammel[3, 9] observed the evolutionary process of ES on the sphere function and it s convergence reliability on Rastrigin s function and Ackley s function. They demonstrated that noise does not influence the performance of ES as long as the noise level is small compared to the function ....

....the performance of point based methods, e.g. Threshold Accepting and Pattern Search, and population based methods, e.g. ES and GA. Based on their experimental results, they concluded that, in the presence of noise, population based methods were preferable to point based methods. Beyer and Arnold[4] observed that noise reduces convergence velocity in both GA and ES. Angeline[1] compared the performance of the log normal update rule and the Gaussian update rule in EP. He demonstrated that the Gaussian update rule outperforms the log normal rule under some models of noise. This paper attempts ....

H.-G. Beyer and D. Arnold (1999), "Fitness Noise and Localization Errors of the Optimum in General Quadratic Fitness Models", Proc. of Genetic and Evolutionary Computation Conference (GECCO'99), pp. 817-824, Morgan Kaufmann.

....strength # is chosen. That is, such a system cannot be an optimizer in a classical sense. Even though we have considered the quadratic sphere model here, the e#ect can be observed qualitatively in all EA 11 systems with fixed and # and constant fitness noise (including GA, see Beyer and Arnold [44]) When considering the results on # # one notices that progress toward the optimum is a result of two opposite tendencies: a positive gain part and a negative loss part. The main e#ect of recombination is due to the reduction of the loss part by a factor of 1 compared to the (1, #) ES. The ....

H.-G. Beyer and D. V. Arnold. Fitness noise and localization errors of the optimum in general quadratic fitness models. In Proceedings of the Genetic and Evolutionary Computation Conference (GECCO), pages 817--824, San Francisco, CA, 1999. Morgan Kaufmann.

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