Search the Dagstuhl Website
Looking for information on the websites of the individual seminars? - Then please:
Not found what you are looking for? - Some of our services have separate websites, each with its own search option. Please check the following list:
Schloss Dagstuhl - LZI - Logo
Schloss Dagstuhl Services
Within this website:
External resources:
  • DOOR (for registering your stay at Dagstuhl)
  • DOSA (for proposing future Dagstuhl Seminars or Dagstuhl Perspectives Workshops)
Within this website:
External resources:
Within this website:
External resources:
  • the dblp Computer Science Bibliography

Dagstuhl Seminar 23332

Synergizing Theory and Practice of Automated Algorithm Design for Optimization

( Aug 13 – Aug 18, 2023 )

(Click in the middle of the image to enlarge)

Please use the following short url to reference this page:



Shared Documents


The design of optimization algorithms for tackling real-world problems often requires making tens to hundreds of design decisions that are reflected in the choice of algorithmic components and their hyperparameter values. Due to being a complex, tedious, error-prone, and time-consuming task, there has been major research efforts in the last decade towards developing automatic methods for algorithm design, including algorithm configuration, algorithm selection, and automated machine learning. The general problem of automated algorithm design (AAD) can be formulated as a meta-optimization problem that is mixed-variable, black-box, and stochastic, which requires the use of meta-heuristics.

Despite its relevance, AAD is not sufficiently well understood from a mathematical and theoretical point of view. However, in recent years, theoretical analyses, especially runtime guarantees, have made major contributions to relevant AAD subproblems. The next step is to advance this initial research to the more general problem of AAD. To this end, it has proven useful in the past to support theoretical analyses by empirical research, for example, by searching for optimal parameter values that are unproven yet, or by providing evidence for conjectures regarding the relationship between parameters, problem features, and algorithm performance. So far, this effort has led to a handful of publications that aim to bridge the gap between theory and practice, even including a PhD thesis and one paper winning a best-paper award at GECCO 2022. Still, the potential of such collaborations is exploited quite rarely.

This Dagstuhl Seminar aims at gaining more fundamental insights into AAD by combining theoretical and empirical research and at initiating collaborations between these two areas. To this end, we want to bring together researchers working on the theory of meta-heuristics or on empirical methods and applications of AAD, providing them with the opportunity to improve the mutual understanding between the two fields and to develop ideas about how theory and practice can help each other derive mutually useful results. The participants will work together to tackle various problems in the intersection of both domains, such as: What are the key differences between AAD and other optimization scenarios? What properties of current AAD approaches should come with mathematical guarantees? What characteristics of empirical benchmarks should carry over to benchmarks used for theoretical analysis? How can the empirical toolbox be made more appealing for theoretical research such that it can help uncover unproven, yet interesting hypotheses? How can a unified and consistent language between researchers focusing on theory or empirical results be established? We envision a dynamic seminar where questions and insights arise organically from dynamic discussion groups. This seminar will give participants the chance to meet new people and start new collaborations as well as to shape the future direction of the intersection of theory and practice for AAD.

Copyright Martin S. Krejca, Marius Lindauer, Manuel López-Ibáñez, and Katherine M. Malan


  • Machine Learning
  • Neural and Evolutionary Computing

  • automated algorithm design
  • hyper-parameter tuning
  • parameter control
  • heuristic optimization
  • black-box optimization