An accelerated directional derivative method for smooth stochastic convex optimization

We consider smooth stochastic convex optimization problems in the context of algorithms which are based on directional derivatives objective function. This can be considered as an intermediate one between derivative-free and gradient-based optimization. assume that at any given point for direction, a approximation derivative function this direction is available with some additive noise. The noise assumed to unknown nature, but bounded absolute value. underline we opposed coordinate descent methods use only directions. For setting, propose non-accelerated accelerated method provide their complexity bounds. Our algorithm has bound similar algorithm, is, without dimension-dependent factor. coincides up factor square root problem dimension. extend these results strongly problems.