Extended Lock-in Feedback Applicable on Higher Dimensional Function Maximization

Abstract

This thesis introduces two novel versions of the existing Lockin Feedback algorithm. This algorithm is a means of performing stochastic optimization. The novel versions include alterations that make them applicable on higher dimensional function maximization problems. By running several simulation tests and examining the cumulative regret returned by each method, this thesis shows that the proposed extensions prove to be performing well on function maximization problems of two dimensions. Both versions are also applied on a function containing multiple maxima, in order to test their ability to deal with more complex maximization problems. By making adjustments to the Lock-in Feedback algorithm inspired on the Artificial Bee Colony algorithm, it was made sure that the method would uncover global in stead of local maxima.