Kriging is well-suited to parallelize optimization

Beyond both estalished frameworks of derivative-based descent and stochastic search algorithms, the rise of expensive optimization problems creates the need for new specific approaches and procedures. The word ”expensive” —which refers to price and/or time issues— implies severely restricted budgets in terms of objective function evaluations. Such limitations contrast with the computational burden typically associated with stochastic search techniques, like genetic algorithms. Furthermore, the latter evaluations provide no differential information in a majority of expensive optimization problems, whether the objective function originate from physical or from simulated experiments. Hence there exists a strong motivation for developing derivative-free algorithms, with a particular focus on their optimization performances in a drastically limited number of evaluations. Investigating and implementing adequate strategies constitute a contemporary challenge at the interface between Applied Mathematics and Computational Intelligence, especially when it comes to reducing optimization durations by efficiently taking advantage of parallel computation facilities.