General mathematical runtime analysis of evolutionary multitask multi-objective optimization

Establish a general mathematical runtime analysis for evolutionary multitask multi-objective optimization, beyond the specific two-task bi-objective benchmark pairs analyzed for the MO-MFEA.

Background

Evolutionary multitask optimization has received theoretical runtime analyses primarily in the single-objective setting, while the multi-objective setting introduces additional complications. In particular, the MO-MFEA uses the NSGA-II-based non-dominated sorting and crowding-distance order, supports general population sizes rather than the constant population size studied in earlier work, and involves dynamically changing distributions of individuals across tasks together with cross-task knowledge transfer.

The paper develops initial runtime results for the MO-MFEA on two specific pairs of bi-objective problems, namely (OMM, OJZJ) and (LOTZ, DLTB). However, these results constitute only a first step toward the broader unresolved task of establishing mathematical runtime theory for evolutionary multitask multi-objective optimization in general.

References

How to establish the mathematical runtime analysis for the evolutionary multitask multi-objective optimization remains an important open problem.

— First Mathematical Runtime Analyses of Evolutionary Multitask Multi-Objective Optimization  (2610.21311 - Zhang et al., 15 Oct 2026) in Section 1, Introduction