Stochastic incremental mirror descent algorithms with Nesterov smoothing (Q6145577)

This paper presents two incremental stochastic mirror descent algorithms for minimizing sums of finitely many non-smooth convex functions over convex sets. The Nesterov smoothing technique is used for the gradients of the smooth summands of the objective function instead of their subgradients. The functions are not required to be Lipschitz continuous or differentiable. Three algorithms are considered including an optimization problem in classifying images using support vector machines, a problem in tomography and a continuous location problem.