Investigating the Impact of Sequential Selection in the (1,2)-CMA-ES on the Noisy BBOB-2010 Testbed

Anne Auger (1) , Dimo Brockhoff (1) , Nikolaus Hansen (1)

1 TAO - Machine Learning and Optimisation

Sequential selection was introduced for Evolution Strategies (ESs) with the aim of accelerating their convergence---performing the evaluations of the different offspring sequentially and concluding an iteration immediately if one offspring is better than the parent. This paper investigates the impact of the application of sequential selection to the (1,2)-CMA-ES on the BBOB-2010 noisy benchmark testbed. The performance of the (1,2$^s$)-CMA-ES, where sequential selection is implemented, is compared to the baseline algorithm (1,2)-CMA-ES. Independent restarts for the two algorithms are conducted up to a maximum number of $10^{4} D$ function evaluations, where $D$ is the dimension of the search space. The results show a slight improvement of the (1,2$^s$)-CMA-ES over the baseline (1,2)-CMA-ES on the sphere function with Cauchy noise and a stronger decline on the sphere function with moderate uniform noise. Overall, the (1,2$^s$)-CMA-ES seems slighly less reliable and we conclude that for the (1,2)-CMA-ES, sequential selection is no improvement on noisy functions.

Domains: Neural and Evolutionary Computing [cs.NE]

Anne Auger, Dimo Brockhoff, Nikolaus Hansen. Investigating the Impact of Sequential Selection in the (1,2)-CMA-ES on the Noisy BBOB-2010 Testbed. GECCO workshop on Black-Box Optimization Benchmarking (BBOB'2010), Jul 2010, Portland, OR, United States. pp.1605--1610, ⟨10.1145/1830761.1830779⟩. ⟨inria-00502432⟩