GECCO Papers

At the Genetic and Evolutionary Computation Conference 2026 in San Jose, several research papers were presented that deal with current challenges and innovative solutions in the field of evolutionary optimization and robust planning under uncertainty.

Title: "Evolving Robust Schedules: Advantages of Genetic Algorithms for FJSSP-W under Uncertainties"

Authors: David Hutter, Thomas Steinberger, Michael Hellwig

Content: When factories operate under uncertain conditions - for example, when people work at different speeds or are absent at short notice - they need particularly robust scheduling methods. The paper compares two approaches: classical mathematical methods and so-called genetic algorithms. While traditional methods deliver very good plans under stable conditions, they lose a lot of quality as soon as uncertainty comes into play. Genetic algorithms can take these uncertainties into account directly and therefore deliver significantly more robust results. The study shows: For realistic, "turbulent" factory environments, genetic algorithms have a clear advantage. 

Title: "On the Use of the "Mutate Large, But Inherit Small Principle" in Global and Noisy Optimization" 

Authors: Lisa Schönenberger, Hans-Georg Beyer

Content: Optimization in continuous, highly multimodal landscapes is one of the greatest challenges of modern numerical methods. An illustrative example can be found in robotics: When controlling a robotic arm with many joints, an optimal trajectory must be found that is both energy-efficient and collision-free. However, the underlying objective function has countless possible configurations, many of which are only apparently optimal. Therefore, with increasing dimensionality, not only the number of potential extrema grows, but also the difficulty of capturing global structures of the objective function. Classical gradient methods quickly reach their limits here, as they typically get stuck in nearby local minima and do not have an overview of the entire landscape. 

Evolution strategies have already proven themselves in practice for solving such high-dimensional problems. However, here too, successful localization of the optimum depends heavily on several parameter settings. These include the relationship between step size and mutation strength. This influence is already known in practice and the optimal ratio has been investigated in experiments. This paper offers a first mathematical approach on how to determine the optimal ratio between step size and mutation strength. The resulting findings can contribute to the development of better and more stable algorithms in the long term.