Les missions du poste
Many MOEAs feature various parameters, such as the population size and the pool of operators that they use, but due to their general-purpose nature of MOEAs, it is often unclear how to set these parameters properly. The theoretical investigation of MOEAs aims at providing useful suggestions
for such choices, based on the rigorous analysis of these algorithms. This field is part of the mathematical runtime analysis of randomized search heuristics [AD11, DN20], which has successfully aided in improved algorithm guarantees and design over two decades. Recently, the theoretical analysis of MOEAs garnered a lot of interest [ZLD22, ZD22, BQ22, ZD24b, DQ23a, DQ23b, DQ23c, DOSS23b, DOSS23a, CDH+23, WD23, ZD24c, ZLDD24]. Despite this sudden surge, current theoretical results remain limited to discrete problems although MOEAs also see heavy use for continuous problems.
Le profil recherché
Bienvenue chez Doctorat.Gouv.Fr
are naturally suited for such multi-objective optimization problems and have been extensively and with great success used for this purpose [ZQL+11].
Multi-objective evolutionary algorithms (MOEAs) are general-purpose solvers that are applicable to virtually any domain as long as the quality of a solution can be properly assessed. In particular, this covers both continuous and discrete domains, which are two very distinct important domains. Most popular MOEAs, such as the NSGA-II [DPAM02], NSGA-III [DJ14], and SMS-EMOA [BNE07], were proposed with the continuous domain in mind, for which they see most of their use.
Publiée le 18/05/2026 - Réf : eb5a77226d23ea7491241c7eac307738
