Dissertation
From Benchmarking Optimization Heuristics to Dynamic Algorithm Configuration
For optimization problems, it is often unclear how to choose the most appropriate optimization algorithm. As such, rigorous benchmarking practices are critical to ensure we can gain as much insight into the strengths and weaknesses of these types of algorithms.
- Author
- D.L. Vermetten
- Date
- 13 February 2025
- Links
- Thesis in Leiden Repository
While this can be done by looking at performance, we show that a broader view of benchmarking, facilitated by our IOHprofiler toolbox, enables us to address a wider range of research questions.
When building on the results from these benchmarking studies, we observe large complementarity between different algorithms. To exploit this complementarity, we design modular algorithm frameworks which are subsequently tuned to specific types of problems.
While algorithm configuration manages to exploit algorithm complementarity on a per-problem level, we can use the performance data to design algorithms which benefit from performance differences within a single problem. This results in dynamic algorithm selection and configuration techniques, where the algorithm or its settings can be changed during the optimization process.
While dynamic algorithm selection and configuration have a lot of potential, accurately testing these techniques requires a broader range of benchmark problems to evaluate them on. As such, we end the thesis by introducing a problem generator for this kind of analysis.