On convergence of the stochastic subgradient method with on-line stepsize rules (Q1083369)

The stochastic subgradient method for solving convex programming problems is considered in the paper. More precisely the authors deal with the problem of determining the stepsize coefficients in the classical stochastic subgradient methods under the assumptions that neither the values nor the subgradients of the optimized function are available. First, the authors follow some former papers on this topic. Further, they analyse the properties of the method with special stepsize coefficients. A convergence theorem is introduced and proved in the paper.