Memorandum of Understanding between the University of Patras, GREECE and the Institutefor Data Science, Engineering and Analytics (IDE+A) Technische Hochschule Köln, GERMANY

The University of Patras and Institute for Data Science, Engineering, and Analytics (TH Koeln), wishing to
promote co-operation between the two institutions in education and in academic research, agree to explore:
a. the development of joint research activities;
b. and facilitate, university staff exchanges or mutual visits to both institutions;
c. doctoral student training and development;
d. student exchange and/or visiting programmes;
e. the exchange of information, including the results of teaching and research collaboration;
f. any other activities viewed to be mutually beneficial.

This agreement is based on the more than twenty-year collaboration between Prof. Dr. Thomas Bartz-Beielstein and Prof. Dr. Michael N. Vrahatis. In most of them, a third author, Prof. Dr. K. E. Parsopoulos, is involved. Konstantinos Parsopoulos is professor at the Department of Computer Science and Engineering, University of Ioannina, Greece.

The first joint works date back to the year 2002:
In dem Paper “Tuning PSO Parameters Through Sensitivity Analysis” [BPV02] a first analysis of the Particle Swarm Optimization (PSO) method’s parameters, using Design of Experiments (DOE) techniques, is performed, and important settings as well as interactions among the parameters, are investigated (screening). The 2004 papers “Designing particle swarm optimization with regression trees” [BVPV04] and “Analysis of Particle Swarm Optimization Using Computational Statistics” [BPV04] can be considered as milestones in the development of the Hyperparameter-tuning method, because they introduce the use of regression trees for the design of optimization algorithms and the use of computational statistics for the analysis of optimization algorithms. It allows the tuning of categorical hyperparameters as well as standard numerical ones, because the regression trees are able to cope with both of them. This approach is still used in the current research on hyperparameter tuning.

The focus of the paper “Tuning algorithms for black-box optimization: State of the art and future perspectives” [Bart19g] lies on automatic and interactive tuning methods for stochastic optimization algorithms, e.g., evolutionary algorithms. Algorithm tuning is important because it helps to avoid wrong parameter settings, to improve the existing algorithms, to select the best algorithm for working with a real-world problem, to show the value of a novel algorithm, to evaluate the performance of an optimization algorithm when different option settings are used, and to obtain an algorithm instance that is robust to changes in problem specification. This chapter discusses strategical issues and defines eight key topics for tuning, namely, optimization algorithms, test problems, experimental setup, performance metrics, reporting, parallelization, tuning methods, and software. Features of established tuning software packages such as IRACE, SPOT, SMAC, and ParamILS are compared.


[BPV02] Beielstein, T., Parsopoulos, K. E., and Vrahatis, M. N. Tuning PSO parameters through sensitivity analysis. Tech. rep., Universität Dortmund, Sonderforschungsbereich (SFB) 531, 01 2002.

[BLMS03b] Bartz-Beielstein, T., Mehnen, J., Schmitt, K., Parsopoulos, K. E., and Vrahatis, M. N. Particle Swarm Optimizers for Pareto Optimization with Enhanced Archiving Techniques. In Proceedings 2003 Congress on Evolutionary Computation (CEC’03), Canberra (Piscataway NJ, Dec 2003), R. Sarker et al., Eds., vol. 3, IEEE, pp. 1780–1787.

[BVPV04] Bartz-Beielstein, T., de Vegt, M., Parsopoulos, K. E., and Vrahatis, M. N. Designing Particle Swarm Optimization with Regression Trees. Tech. rep., 05 2004.

[BPV04] Bartz-Beielstein, T., Parsopoulos, K. E., and Vrahatis, M. N. Analysis of Particle Swarm Optimization Using Computational Statistics. In Proceedings International Conference Numerical Analysis and Applied Mathematics (ICNAAM) (Weinheim, Germany, 2004), T. E. Simos and C. Tsitouras, Eds., Wiley-VCH, pp. 34–37.

[BPV04b] Bartz-Beielstein, T., Parsopoulos, K. E., and Vrahatis, M. N. Design and analysis of optimization algorithms using computational statistics. Applied Numerical Analysis and Computational Mathematics (ANACM) 1, 2 (Dec 2004), 413–433.

[Bart19g] Bartz-Beielstein, T., Rehbach, F., and Rebolledo, M. Tuning algorithms for black-box optimization: State of the art and future perspectives. In Black Box Optimization, Machine Learning and No-Free Lunch Theorems, P. Pardalos, V. Rasskazova, and M. Vrahatis, Eds., no. 170 in Springer Optimization and Its Applications. Springer, 2021.