The research team from the Institute of Automation and Computer Science conducts fundamental research in the field of benchmarking algorithms for black-box (simulation-based, derivative-free) optimization.

Benchmarking is a field that is mostly concerned with methodologies for the comparison of optimization algorithms – from the careful selection of unbiased test instances and representative algorithms to the use of appropriate statistical techniques. The end goals of benchmarking are closely tied to efficient (and possibly automated) algorithm selection based on certain characteristics (both high-level and numerical-based such as the Exploratory Landscape Analysis features) of the given black-box optimization problem. These research findings can offer valuable insights for practitioners and researchers, enabling them to tackle diverse optimization problems effectively.

Our contributions to this field are related to identifying structural biases in benchmark sets and optimization algorithms (such as in https://doi.org/10.1038/s42256-022-00579-0 and https://doi.org/10.1145/3583133.3590653), in bringing together the two communities developing stochastic and deterministic methods for black-box optimization problems (such as in https://doi.org/10.1109/TEVC.2024.3379756), and expanding the benchmarking sets towards real-world problem instances (such as in https://doi.org/10.1007/978-3-031-30229-9_24).

The research team regularly attends top-tier conferences (core A or B) where current research on benchmarking black-box optimization algorithms is discussed, such as Genetic and Evolutionary Computation Conference (GECCO), Parallel Problem Solving From Nature (PPSN), International Conference on the Applications of Evolutionary Computation (EvoApplications), or IEEE Congress on Evolutionary Computation (IEEE CEC). Members of the research team also serve in the program committees for some these conferences.

On this research topic, we have established international cooperation with a research groups from Vilnius University (Lithuania) and Jönköping University (Sweden). Members of the research team also participate in the COST Action “CA22137 – Randomised Optimisation Algorithms Research Network (ROAR-NET)” through which more international connection are expected to be established.

Media
https://vedavyzkum.cz/z-domova/z-domova/vedec-objevil-systemovou-chybu-v-oboru-jeho-clanek-ted-otiskl-nature (in Czech)

Contact
Ing. Jakub Kůdela, Ph.D.