On the adaptation of noise level for stochastic optimization
Olivier Teytaud and Anne Auger
In: CEC 2007, 2007, Singapore.
Abstract— This paper deals with the optimization of noisy
ﬁtness functions, where the noise level can be reduced by
increasing the computational effort. We theoretically investigate
the question of the control of the noise level. We analyse two
different schemes for an adaptive control and prove sufﬁcient
conditions ensuring the existence of an homogeneous Markov
chain, which is the ﬁrst step to prove linear convergence when
dealing with non-noisy ﬁtness functions. We experimentally
validate the relevance of the homogeneity criterion. Large-scale
experiments conclude to the efﬁciency in a difﬁcult framework.