260 / 2024-04-23 22:02:23

A Tree-Based Continuous Simulation Optimization Procedure

simulation,optimization

In this work, tree-based procedures are proposed to solve continuous optimization via simulation (COvS) problems. The procedures search for the optimal solution by adaptively partitioning the design space and allocating more sampling efforts to the area where the optimal solution tends to lie. We establish different lower bounds on the minimax convergence rate for the optimization error when the objective function satisfies different local smoothness assumptions. Then we show that our procedures can solve these problems without necessarily knowing the objective function’s smoothness condition and achieve the optimal minimax convergence rates. Numerical results show the procedures are efficient.