1: \begin{abstract}

2: This paper considers global optimization with a black-box unknown

3: objective function that can be non-convex and non-differentiable. Such a

4: difficult optimization problem arises in many real-world applications,

5: such as parameter tuning in machine learning, engineering design problem,

6: and planning with a complex physics simulator. This paper proposes a

7: new global optimization algorithm, called \emph{Locally Oriented Global

8: Optimization} (LOGO), to aim for both fast convergence in practice and

9: finite-time error bound in theory. The advantage and usage of the new

10: algorithm are illustrated via theoretical analysis and an experiment

11: conducted with $11$ benchmark test functions. Further, we modify the

12: LOGO algorithm to specifically solve a planning problem  via policy search with continuous

13: state/action space and long time horizon while maintaining its finite-time

14: error bound. We apply the proposed planning method to accident management

15: of a nuclear power plant. The result of the application study demonstrates

16: the practical utility of our method.

17: \end{abstract}

18: