Organized by: George S. Dulikravich, Florida International University, USA.

Description:
Pareto front in multi-objective optimization evolves iteratively and migrates in the objective function space until it converges to a relatively steady configuration. The process of determining the first few points on an intermediate (not converged) Pareto front takes many evaluations of the objective functions. Further iterative migration of the intermediate Pareto front to its final converged configuration takes many more evaluations of the objective functions. In practical cases where the evaluations of the objective functions is very costly, we typically use surrogate models/metamodels in the form of response hyper-surfaces. However, the overall cost of the entire optimization process could be significantly reduced if it would be possible to predict the iteratively evolving Pareto front configuration several iterations (generations) into the future by using the intermediate Pareto front configurations (and the corresponding values of design variables) from the current and the previous few iterations (generations). In other words, by modeling the Pareto front configuration evolution process its final (steady) configuration could be predicted based on the information available from the first few Pareto front configurations.