Mirrored Variants of the (1,2)-CMA-ES Compared on the Noiseless BBOB-2010 Testbed
Abstract
Derandomization by means of mirroring has been recently introduced to enhance the performances of $(1,\lambda)$-Evolution-Strategies (ESs) with the aim of designing fast robust local search stochastic algorithms. This paper compares on the BBOB-2010 noiseless benchmark testbed two variants of the (1,2)-CMA-ES where the mirroring method is implemented. Independent restarts are conducted till a total budget of $10^{4} D$ function evaluations per trial is reached, where $D$ is the dimension of the search space. The results show that the improved variants increase the success probability on 5 (respectively 7) out of 24 test functions in 20D and at the same time are significantly faster on 9 (10) functions in 20D by a factor of about 2--3 (2--4) for a target value of $10^{-7}$ while in no case, the baseline (1,2)-CMA-ES is significantly faster on any tested target function value in 5D and 20D.
Origin | Files produced by the author(s) |
---|