If At First You Don’t Succeed, Try, Try Again. But please consider using another algorithm.

In many domains of optimization, heuristics are the state-of-the-art solution method. However, which one will work best on any given problem instance is often unclear up front. The research community has developed (and is still developing) approaches such as automated algorithm selection, instance-based algorithm configuration and dynamic algorithm selection to extract the most performance from a set of base heuristics on a given set of test problems.

Despite the capabilities of these AI systems, recent research has shown that even a very simple, greedily built restart schedule can close much of the performance gap in a portfolio of algorithms for continuous black-box optimization. [1]

In this thesis, your task is to take this approach "up to 11": More algorithms and more sophisticated scheduling techniques to create a better-performing static algorithm / restart schedule. Further, benchmarking can be extended to more comprehensive problem sets than the one used in [1], leading to additional insights about the generalizability and robustness of algorithm schedules.

[1] Schäpermeier, L. (2025). Greedy Restart Schedules: A Baseline for Dynamic Algorithm Selection on Numerical Black-box Optimization Problems. In Proceedings of the Genetic and Evolutionary Computation Conference (pp. 1199-1207). https://doi.org/10.1145/3712256.3726408