What is it about?
Optimizing a function, which is very expensive to evaluate requires the use of unusual methods, as gradient descents or genetic algorithms lead to a lot of calls of the objective function. Bayesian optimization is a state of the art method, using probabilistic metamodels instead of the true function. This article handles with the way to solve multi-objective, constrained black-box functions with an application to a concrete aircraft design. The main concepts of the algorithm are detailed, such as the obtained results, the computation time and the recommendations to use the method.
Featured Image
Photo by Sebastian Grochowicz on Unsplash
Why is it important?
Most of the real life engineering problems considers not only one expensive objective to compute, but many of them, with respect to constraints. Even if the computation power increases, aircrafts models tends to be more accurate and still stay very long to compute. Bayesian optimization is for the same reason useful for nowadays neural networks optimization.
Perspectives
Thanks to the proposed regularization method of infill criterions, new problems may be solve, such as those on which usual Bayesian optimization methods lacked of exploration or exploitation, leading to a poor Pareto optimal set of solutions. Many-objective problems are one of the main target as the regularization will give importance to all of the functions to optimize.
Robin Grapin
Groupe ISAE
Read the Original
This page is a summary of: Constrained Multi-Objective Bayesian Optimization with Application to Aircraft Design, June 2022, American Institute of Aeronautics and Astronautics (AIAA),
DOI: 10.2514/6.2022-4053.
You can read the full text:
Contributors
The following have contributed to this page
