Targeting Well-Balanced Solutions in Multi-Objective Bayesian Optimization under a Restricted Budget

Autor: Victor Picheny, Vincent Herbert, David Gaudrie, Rodolphe Le Riche, Benoit Enaux
Přispěvatelé: École des Mines de Saint-Étienne (Mines Saint-Étienne MSE), Institut Mines-Télécom [Paris] (IMT), PSA Peugeot Citroën, PSA Peugeot Citroën (PSA), Laboratoire d'Informatique, de Modélisation et d'Optimisation des Systèmes (LIMOS), Ecole Nationale Supérieure des Mines de St Etienne-Université Clermont Auvergne (UCA)-Centre National de la Recherche Scientifique (CNRS), Institut Henri Fayol (FAYOL-ENSMSE), Institut Mines-Télécom [Paris] (IMT)-Institut Mines-Télécom [Paris] (IMT), Département Génie mathématique et industriel (FAYOL-ENSMSE), Ecole Nationale Supérieure des Mines de St Etienne-Institut Henri Fayol, Centre National de la Recherche Scientifique (CNRS), Unité de Mathématiques et Informatique Appliquées de Toulouse (MIAT INRA), Institut National de la Recherche Agronomique (INRA), Battiti R., Brunato M., Kotsireas I., Pardalos P., Ecole Nationale Supérieure des Mines de St Etienne-Centre National de la Recherche Scientifique (CNRS)-Université Clermont Auvergne [2017-2020] (UCA [2017-2020]), PSA Peugeot - Citroën (PSA), Ecole Nationale Supérieure des Mines de St Etienne-Université Clermont Auvergne [2017-2020] (UCA [2017-2020])-Centre National de la Recherche Scientifique (CNRS), Ecole Nationale Supérieure des Mines de St Etienne (ENSM ST-ETIENNE)-Université Clermont Auvergne [2017-2020] (UCA [2017-2020])-Centre National de la Recherche Scientifique (CNRS), Ecole Nationale Supérieure des Mines de St Etienne (ENSM ST-ETIENNE)-Institut Henri Fayol, Ecole Nationale Supérieure des Mines de St Etienne, École des Mines de Saint-Étienne ( Mines Saint-Étienne MSE ), Institut Mines-Télécom [Paris], PSA Peugeot - Citroën ( PSA ), Peugeot-Citroën, Laboratoire d'Informatique, de Modélisation et d'Optimisation des Systèmes ( LIMOS ), Sigma CLERMONT ( Sigma CLERMONT ) -Université Clermont Auvergne ( UCA ) -Centre National de la Recherche Scientifique ( CNRS ), Unité de Mathématiques et Informatique Appliquées de Toulouse ( MIAT INRA ), Institut National de la Recherche Agronomique ( INRA ), Ecole Nationale Supérieure des Mines de St Etienne (ENSM ST-ETIENNE)
Jazyk: angličtina
Rok vydání: 2018
Předmět:
[ MATH.MATH-OC ] Mathematics [math]/Optimization and Control [math.OC]
Mathematical optimization
Computer science
Computer experiments
Gaussian processes
Context (language use)
02 engineering and technology
[STAT.OT]Statistics [stat]/Other Statistics [stat.ML]
Multi-objective optimization
Set (abstract data type)
symbols.namesake
[SPI]Engineering Sciences [physics]
[MATH.MATH-ST]Mathematics [math]/Statistics [math.ST]
Convergence (routing)
0202 electrical engineering
electronic engineering
information engineering
[ SPI ] Engineering Sciences [physics]
Parsimonious optimization
[INFO]Computer Science [cs]
[ MATH.MATH-ST ] Mathematics [math]/Statistics [math.ST]
[MATH]Mathematics [math]
Gaussian process
ComputingMilieux_MISCELLANEOUS
Bayesian optimization
Pareto principle
Preference-based optimization
Computer experiment
020202 computer hardware & architecture
symbols
020201 artificial intelligence & image processing
[MATH.MATH-OC]Mathematics [math]/Optimization and Control [math.OC]
Zdroj: Lecture Notes in Computer Science
12th International Conference on Learning and Intelligent Optimization
12th International Conference on Learning and Intelligent Optimization, Jun 2018, Kalamata, Greece. pp 175-179, ⟨10.1007/978-3-030-05348-2_15⟩
PGMO Days 2017
PGMO Days 2017, Nov 2017, Saclay, France
Journées du GdR Mascot-Num 2018
Journées du GdR Mascot-Num 2018, Mar 2018, Nantes, France
Journées Oquaido 2017
Journées Oquaido 2017, Nov 2017, Orléans, France. 2017
International Conference on Learning and Intelligent Optimization (LION'18)
International Conference on Learning and Intelligent Optimization (LION'18), Jun 2018, Kalamata, Greece
Lecture Notes in Computer Science ISBN: 9783030053475
LION
Popis: International audience; Multi-objective optimization aims at finding the Pareto set composed of all the best trade-off solutions between several objectives. When dealing with expensive-to-evaluate black box functions, surrogate-based approaches, in the vein of EGO have proven their effectiveness. However, for extremely narrow budgets, and/or when the number of objectives is large, uncovering the entire Pareto set becomes out of reach even for these approaches. In this work, we restrict our search to well-chosen targeted parts of the Pareto set. This accelerates the search as only a subset of the objective space is considered. As an end-user would typically choose solutions with equilibrated trade-offs between objectives rather than ones favoring a single objective over the others, we will focus on the central part of the Pareto Front, which corresponds to the most well-balanced solutions. First, we discuss how to define and estimate the center of the Pareto Front. That estimated point has to fairly represent the topology of the front, in spite of the parsimonious knowledge of the objective space. Then, three infill criteria which will guide the optimization, includiing the Expected Hypervolume Improvement are studied and tailored through some of their hyperparameters to enable them to target.To assess performance, a benchmark built from real-world airfoil aerodynamic data, with variable dimension and number of objectives is used. Compared with standard techniques, the proposed methodology leads to a faster and a more precise convergence towards the Pareto Frontin the targeted region.
Databáze: OpenAIRE
Externí odkaz: