Optimization by iteration of orthogonal design neighborhood search

Orthogonal design local search (ODLS) solves optimization problems with hundreds of variables, a noisy objective function, and many local optima. ODLS iterates evaluation of watching points around the current point at the same random distance, selected by orthogonal design of experiments, determination of improvement direction based on the averages of plus- and minus direction at each variable, and binary search toward the improvement direction to find the new current point. Numerical experiments clarified that ODLS outperforms simulated annealing when the problem has more than hundreds of variables.