Title :
Cyclic stochastic optimization with an application to structure from motion and convergence for separable loss functions
Author :
Hernndez-Cuevas, Karla
Author_Institution :
Whiting Sch. of Eng., Johns Hopkins Univ., Baltimore, MD, USA
Abstract :
In this paper we present the motivation behind cyclic stochastic optimization and give an example of how it can be applied to a structure from motion problem. We give sufficient conditions for convergence of cyclic simultaneous perturbation stochastic approximation as well as for cyclic stochastic gradient in the case of a linearly separable loss function.
Keywords :
gradient methods; motion estimation; optimisation; stochastic programming; cyclic simultaneous perturbation stochastic approximation; cyclic stochastic gradient; cyclic stochastic optimization; separable loss functions; structure from motion; Convergence; Equations; Loss measurement; Noise measurement; Optimization; Stochastic processes; Vectors;
Conference_Titel :
Information Sciences and Systems (CISS), 2013 47th Annual Conference on
Conference_Location :
Baltimore, MD
Print_ISBN :
978-1-4673-5237-6
Electronic_ISBN :
978-1-4673-5238-3
DOI :
10.1109/CISS.2013.6552333
