Log-linear convergence and divergence of the scale-invariant \((1+1)\)-ES in noisy environments (Q633840)

This paper investigates a certain class of noisy problems which use the so-called multiplicative noise, where the noiseless objective function is perturbed by the multiplication with a noise term. The objective of this paper is to address the question of how the properties of the noise distribution relate to convergence or divergence of the \((1+1)\)-ES (evolution strategies). The authors provide a mathematical analysis of the convergence of a \((1+1)\)-ES on unimodal spherical objective functions in the presence of multiplicative noise, prove that for a positive expected value of the noisy objective function, convergence or divergence happens depending on the infimum of the support of the noise. Moreover, they investigate convergence rates and show that log-linear convergence is preserved in presence of noise. These results are an extension of the results presented in [\textit{D. V. Arnold} and \textit{H.-G. Beyer}, Comput. Optim. Appl. 24, No. 1, 135--159 (2003; Zbl 1035.90110)].