Meta Modeling for Combinatorial Catalyst Optimization
Résumé
Our aim is to find the best catalyst, the best combination of compounds, in order to optimize a chemical reaction. The chemists use mainly a heuristic algorithm, especially an evolutionary algorithm, to achieve the best combination. In this paper, we outline a variant of evolutionary optimization algorithm, says meta modeling. Our idea is to combine a statistical learning algorithm with the optimization process. The goal is a better use of the past experience, the labelled individuals, in the guidance of the search exploration of the optimal solution. The approach is especially useful in the combinatorial catalysis optimization because the fitness function is unknown and the labelled individual is obtained by real chemical reaction. This is highly costly and takes time. We show on a well-known chemists' benchmark that our process slightly the average performance of the standard evolutionary algorithms. But numerous problems remain opened. We try to inventory them in order to define our future work to improve the approach.
Origine : Fichiers produits par l'(les) auteur(s)
