In many real-world optimization problems, there isn't a single objective that, when optimized, automatically yields the desired result in practice. Rather, these problems have multiple targets (e.g., tour duration, fuel costs, and environmental impact in a package delivery problem) that require to find a trade-off between them. The most prominent tools to solve these multi-objective (MO) problems are optimization heuristics such as evolutionary algorithms.

While the default settings for MO heuristics usually deliver reasonable results, they often benefit from parameter tuning; some heuristics may even deteriorate into random search or suffer from early termination if not set up correctly for the problem(s) at hand. Recent advances in benchmarking MO heuristics, such as evolutionary algorithms, on artificial test problems have greatly expanded the realm of well-understood test problems for continuous decision spaces. In particular, a novel collection of well-understood problems is presented in the BONO-Bench test suite [1,2].

The tasks in this thesis are to investigate the tuning sensitivity and potential performance gains of standard MO evolutionary algorithms (NSGA-II, SMS-EMOA, ...) on the overall BONO-Bench suite as well as on individual problem categories within the test suite. You should compare different MO heuristics (e.g., from the pymoo package), configuration spaces and algorithm configuration libraries (e.g., irace, smac) to investigate the most successful setups, resulting in novel insights into biases in wide-spread heuristics, performance tuning guidelines and a general wider set of benchmarked algorithms.

[1] Schäpermeier, L. & Kerschke, P. (2026). BONO-Bench: A Comprehensive Test Suite for Bi-objective Numerical Optimization with Traceable Pareto Sets. Accepted in ACM Transactions on Evolutionary Learning and Optimization. https://doi.org/10.1145/3795775

[2] Schäpermeier, L. (2026). bonobench [Python package]. GitHub. https://github.com/schaepermeier/bonobench