Empirical analysis of stochastic methods of linear algebra

Abstract

In this paper we present the results of an empirical study of stochastic projection and stochastic gradient descent methods as means of obtaining approximate inverses and preconditioners for iterative methods. Results of numerical experiments are used to analyse scalability and overall suitability of the selected methods as practical tools for treatment of large linear systems of equations. The results are preliminary due to the code being not yet fully optimized.