This thesis analyzes and develops constraint handling techniques for stochastic black-box optimization methods. In the first part, evaluation techniques for constrained stochastic optimization are developed and the framework for comparing continuous optimization algorithms is extended with these methods. Several testbeds are formulated for the empirical evaluation of blackbox optimizers with constraint handling and known algorithms are evaluated on these testbeds. The second part is concerned with developing of improved or new constraint handling techniques. Penalization methods, multi-objective approaches and methods that specifically modify the sample distribution of a stochastic optimization algorithm will be considered. The techniques that are developed will be applicable to CMA-ES and should be able to handle a comparatively large number of constraints of different types.
Context
Numerical optimization problems are at the core of many present industrial design or development tasks and inherent to some of the decisive issues of our society. Numerical blackbox optimization methods, interpreting a problem as a black-box where the internal model is non-differentiable, non-convex, multi-modal, noisy, or too complex to be mathematically tractable. Simulation is a cornerstone for the analysis, design and evaluation of blackbox optimizers as efficient algorithms are often too intricate to be easily analyzed theoretically.
This thesis will be connected to the ANR project NumBBO (Oct 2012 - Oct 2016, coordinated by Anne Auger) whose aim is to analyze, improve, and evaluate numerical blackbox optimizers for single-objective (constrained, large-scale), multiobjective, and expensive optimization with a focus on stochastic optimization. A cornerstone of the project is the COmparing Continuous Optimizers (COCO) framework that allows to design better algorithms faster.
Objectives
The objectives are twofold:
develop performance evaluation techniques for constraint optimization with particular focus on stochastic blackbox optimizers. This part involves the formulation of performance criteria and of sound testbeds. The COCO framework will be extended for benchmarking constraint optimizers, implementing a testbed and performance criteria, where the postprocessing tools will be adapted to constraint optimization. The work is connected to multi-objective optimization (a constraint can be seen as additional objective) and done in collaboration with the NumBBO partners involved in the multi-objective part.
develop new algorithms for handling constraints in ranked-based stochastic optimizers. The objective is to propose preferably rank-based methods, applicable to CMA-ES, that can handle a comparatively large number of constraints. Approaches based on penalization and methods that modify the sampling distribution of a stochastic algorithm are considered.
Work program
Extra information
Prerequisite
- interest in optimization and simulation
- programming skills preferably in Python
Détails
Expected funding
Institutional funding
Status of funding
Expected
Candidates
Utilisateur
nikolaus.hansen
Créé
Lundi 17 juin 2013 14:39:20 CEST
dernière modif.
Lundi 17 juin 2013 14:39:20 CEST
Fichiers joints
filename
créé
hits
filesize
Aucun fichier joint à cette fiche
Connexion
Ecole Doctorale Informatique Paris-Sud
Directrice
Nicole Bidoit Assistante
Stéphanie Druetta Conseiller aux thèses
Dominique Gouyou-Beauchamps
ED 427 - Université Paris-Sud
UFR Sciences Orsay
Bat 650 - aile nord - 417
Tel : 01 69 15 63 19
Fax : 01 69 15 63 87
courriel: ed-info à lri.fr