A Stochastic Trust Region Method for Unconstrained Optimization Problems

Figure 1

Convergence of the stochastic trust region method and SGD for function (35). Relative distance to opotimality is shown with respect to the number of the stochastic function processed. For our method, the number of iterations required to achieve a certain accuracy is smaller than the corresponding number for SGD.

(a) 1

(b) 2