Dvosmerni i dvokoračni ubrzani metodi za bezuslovnu optimizaciju

Abstract

This thesis is based on the originally published papers as well as on the results first introduced herein. Taking into account known gradient methods for non-conditional optimization, new accelerated models with specific double direction and double step size formulations for related problem solving are proposed and described in this paper. The emphasis was put on accelerated parameters of these models. The conducted convergence analysis and numerical experiments' results with new iterative patterns confirm theoretical importance of these new methods.