Mirrored variants of the (1,4)-CMA-ES compared on the noiseless BBOB-2010 testbed

Anne Auger, Dimo Brockhoff, Nikolaus Hansen
2010 Proceedings of the 12th annual conference comp on Genetic and evolutionary computation - GECCO '10  
Derandomization by means of mirrored samples has been recently introduced to enhance the performances of (1, λ)-Evolution-Strategies (ESs) with the aim of designing fast and robust stochastic local search algorithms. This paper compares on the BBOB-2010 noiseless benchmark testbed two variants of the (1,4)-CMA-ES where the mirrored samples are used. Independent restarts are conducted up to a total budget of 10 4 D function evaluations, where D is the dimension of the search space. The results
more » ... ow that the improved variants are significantly faster than the baseline (1,4)-CMA-ES on 4 functions in 20D (respectively 7 when using sequential selection in addition) by a factor of up to 3 (on the attractive sector function). In no case, the (1,4)-CMA-ES is significantly faster on any tested target function value in 5D and 20D. Moreover, the algorithm employing both mirroring and sequential selection is significantly better than the algorithm without sequentialism on five functions in 20D with expected running times that are about 20% smaller. General Terms Algorithms Covariance matrix and step-size are updated using the selected steps [9, 1]. Independent Restarts. Similar to [3], we independently restarted all algorithms as long as function evaluations were left, where maximally 10 4 · D function evaluations have been used.
doi:10.1145/1830761.1830773 dblp:conf/gecco/AugerBH10b fatcat:g44cmhzlzjftvi3lcvexlxolne