2025
Konen, Wolfgang
Radial basis functions for constrained numerical global optimization including the challenges of equality constraints Technical Report
2025, (Invited talk presented at conference MOPTA'2025 in Ponta Delgado, Azores).
Abstract | Links | BibTeX | Tags: equality constraint, optimization, SACOBRA, surrogate models
@techreport{Konen2025,
title = {Radial basis functions for constrained numerical global optimization including the challenges of equality constraints},
author = {Wolfgang Konen},
url = {https://www.gm.fh-koeln.de/ciopwebpub/Konen25a.d/MOPTA2025-pres.pdf},
year = {2025},
date = {2025-10-01},
urldate = {2025-10-01},
school = {Research Center CIOP (Computational Intelligence, Optimization and Data Mining)},
abstract = {The efficient solution of black box constrained optimization problems (COPs) is a difficult task. Strong derivative-free heuristics often demand an unaffordable number of function evaluations. The surrogate-assisted solver SACOBRA aims to reduce the required function evaluations by modeling objective and constraint functions by radial basis functions (RBF).
RBF surrogates were introduced as a versatile tool for constrained global optimization in 2014 by Regis and others (COBRA). By iteratively refining the surrogates when new infill points become available, it is possible to perform global optimization with relatively few function evaluations. Our new approach SACOBRA (self-adjusting COBRA) aims at overcoming an issue with COBRA that no fixed set of hyperparameters can solve all benchmark problems in the well-known G-function benchmark suite. SACOBRA works by collecting knowledge about the problem while optimizing. This allows for appropriate online tuning of parameters or rescaling of functions.
In this talk, we present results how SACOBRA improves over COBRA and highlight some reasons why this is the case. Furthermore, we demonstrate how the specific challenges posed by equality constraints can often be overcome using newly developed refinement mechanisms.
SACOBRA was originally written in R. In this talk, we report on the progress made in developing a new Python Version, which is now available on GitHub.},
note = {Invited talk presented at conference MOPTA'2025 in Ponta Delgado, Azores},
keywords = {equality constraint, optimization, SACOBRA, surrogate models},
pubstate = {published},
tppubtype = {techreport}
}
The efficient solution of black box constrained optimization problems (COPs) is a difficult task. Strong derivative-free heuristics often demand an unaffordable number of function evaluations. The surrogate-assisted solver SACOBRA aims to reduce the required function evaluations by modeling objective and constraint functions by radial basis functions (RBF).
RBF surrogates were introduced as a versatile tool for constrained global optimization in 2014 by Regis and others (COBRA). By iteratively refining the surrogates when new infill points become available, it is possible to perform global optimization with relatively few function evaluations. Our new approach SACOBRA (self-adjusting COBRA) aims at overcoming an issue with COBRA that no fixed set of hyperparameters can solve all benchmark problems in the well-known G-function benchmark suite. SACOBRA works by collecting knowledge about the problem while optimizing. This allows for appropriate online tuning of parameters or rescaling of functions.
In this talk, we present results how SACOBRA improves over COBRA and highlight some reasons why this is the case. Furthermore, we demonstrate how the specific challenges posed by equality constraints can often be overcome using newly developed refinement mechanisms.
SACOBRA was originally written in R. In this talk, we report on the progress made in developing a new Python Version, which is now available on GitHub.
RBF surrogates were introduced as a versatile tool for constrained global optimization in 2014 by Regis and others (COBRA). By iteratively refining the surrogates when new infill points become available, it is possible to perform global optimization with relatively few function evaluations. Our new approach SACOBRA (self-adjusting COBRA) aims at overcoming an issue with COBRA that no fixed set of hyperparameters can solve all benchmark problems in the well-known G-function benchmark suite. SACOBRA works by collecting knowledge about the problem while optimizing. This allows for appropriate online tuning of parameters or rescaling of functions.
In this talk, we present results how SACOBRA improves over COBRA and highlight some reasons why this is the case. Furthermore, we demonstrate how the specific challenges posed by equality constraints can often be overcome using newly developed refinement mechanisms.
SACOBRA was originally written in R. In this talk, we report on the progress made in developing a new Python Version, which is now available on GitHub.
2018
Bagheri, Samineh; Konen, Wolfgang; Bäck, Thomas
How to Solve the Dilemma of Margin-Based Equality Handling Methods Proceedings Article
In: Hoffmann, Frank; Hüllermeier, Eyke; Mikut, Ralf (Ed.): Proceedings - 28. Workshop Computational Intelligence, Dortmund, 29. - 30. November 2018, pp. 257-270, KIT Scientific Publishing, Karlsruhe, 2018, ISBN: 978-3-7315-0845-8, (**Young Author Award**).
Links | BibTeX | Tags: equality constraint, optimization, SACOBRA, surrogate models
@inproceedings{Bagheri2018,
title = {How to Solve the Dilemma of Margin-Based Equality Handling Methods},
author = {Samineh Bagheri and Wolfgang Konen and Thomas Bäck},
editor = {Frank Hoffmann and Eyke Hüllermeier and Ralf Mikut},
url = {https://blogs.gm.fh-koeln.de/ciop/files/2018/12/GMA2018.pdf},
doi = {10.5445/KSP/1000085935},
isbn = {978-3-7315-0845-8},
year = {2018},
date = {2018-11-29},
booktitle = {Proceedings - 28. Workshop Computational Intelligence, Dortmund, 29. - 30. November 2018},
pages = {257-270},
publisher = {KIT Scientific Publishing, Karlsruhe},
note = {**Young Author Award**},
keywords = {equality constraint, optimization, SACOBRA, surrogate models},
pubstate = {published},
tppubtype = {inproceedings}
}
Search field
1.
Konen, Wolfgang
Radial basis functions for constrained numerical global optimization including the challenges of equality constraints Technical Report
2025, (Invited talk presented at conference MOPTA'2025 in Ponta Delgado, Azores).
@techreport{Konen2025,
title = {Radial basis functions for constrained numerical global optimization including the challenges of equality constraints},
author = {Wolfgang Konen},
url = {https://www.gm.fh-koeln.de/ciopwebpub/Konen25a.d/MOPTA2025-pres.pdf},
year = {2025},
date = {2025-10-01},
urldate = {2025-10-01},
school = {Research Center CIOP (Computational Intelligence, Optimization and Data Mining)},
abstract = {The efficient solution of black box constrained optimization problems (COPs) is a difficult task. Strong derivative-free heuristics often demand an unaffordable number of function evaluations. The surrogate-assisted solver SACOBRA aims to reduce the required function evaluations by modeling objective and constraint functions by radial basis functions (RBF).
RBF surrogates were introduced as a versatile tool for constrained global optimization in 2014 by Regis and others (COBRA). By iteratively refining the surrogates when new infill points become available, it is possible to perform global optimization with relatively few function evaluations. Our new approach SACOBRA (self-adjusting COBRA) aims at overcoming an issue with COBRA that no fixed set of hyperparameters can solve all benchmark problems in the well-known G-function benchmark suite. SACOBRA works by collecting knowledge about the problem while optimizing. This allows for appropriate online tuning of parameters or rescaling of functions.
In this talk, we present results how SACOBRA improves over COBRA and highlight some reasons why this is the case. Furthermore, we demonstrate how the specific challenges posed by equality constraints can often be overcome using newly developed refinement mechanisms.
SACOBRA was originally written in R. In this talk, we report on the progress made in developing a new Python Version, which is now available on GitHub.},
note = {Invited talk presented at conference MOPTA'2025 in Ponta Delgado, Azores},
keywords = {},
pubstate = {published},
tppubtype = {techreport}
}
The efficient solution of black box constrained optimization problems (COPs) is a difficult task. Strong derivative-free heuristics often demand an unaffordable number of function evaluations. The surrogate-assisted solver SACOBRA aims to reduce the required function evaluations by modeling objective and constraint functions by radial basis functions (RBF).
RBF surrogates were introduced as a versatile tool for constrained global optimization in 2014 by Regis and others (COBRA). By iteratively refining the surrogates when new infill points become available, it is possible to perform global optimization with relatively few function evaluations. Our new approach SACOBRA (self-adjusting COBRA) aims at overcoming an issue with COBRA that no fixed set of hyperparameters can solve all benchmark problems in the well-known G-function benchmark suite. SACOBRA works by collecting knowledge about the problem while optimizing. This allows for appropriate online tuning of parameters or rescaling of functions.
In this talk, we present results how SACOBRA improves over COBRA and highlight some reasons why this is the case. Furthermore, we demonstrate how the specific challenges posed by equality constraints can often be overcome using newly developed refinement mechanisms.
SACOBRA was originally written in R. In this talk, we report on the progress made in developing a new Python Version, which is now available on GitHub.
RBF surrogates were introduced as a versatile tool for constrained global optimization in 2014 by Regis and others (COBRA). By iteratively refining the surrogates when new infill points become available, it is possible to perform global optimization with relatively few function evaluations. Our new approach SACOBRA (self-adjusting COBRA) aims at overcoming an issue with COBRA that no fixed set of hyperparameters can solve all benchmark problems in the well-known G-function benchmark suite. SACOBRA works by collecting knowledge about the problem while optimizing. This allows for appropriate online tuning of parameters or rescaling of functions.
In this talk, we present results how SACOBRA improves over COBRA and highlight some reasons why this is the case. Furthermore, we demonstrate how the specific challenges posed by equality constraints can often be overcome using newly developed refinement mechanisms.
SACOBRA was originally written in R. In this talk, we report on the progress made in developing a new Python Version, which is now available on GitHub.
2.
Bagheri, Samineh; Konen, Wolfgang; Bäck, Thomas
How to Solve the Dilemma of Margin-Based Equality Handling Methods Proceedings Article
In: Hoffmann, Frank; Hüllermeier, Eyke; Mikut, Ralf (Ed.): Proceedings - 28. Workshop Computational Intelligence, Dortmund, 29. - 30. November 2018, pp. 257-270, KIT Scientific Publishing, Karlsruhe, 2018, ISBN: 978-3-7315-0845-8, (**Young Author Award**).
@inproceedings{Bagheri2018,
title = {How to Solve the Dilemma of Margin-Based Equality Handling Methods},
author = {Samineh Bagheri and Wolfgang Konen and Thomas Bäck},
editor = {Frank Hoffmann and Eyke Hüllermeier and Ralf Mikut},
url = {https://blogs.gm.fh-koeln.de/ciop/files/2018/12/GMA2018.pdf},
doi = {10.5445/KSP/1000085935},
isbn = {978-3-7315-0845-8},
year = {2018},
date = {2018-11-29},
booktitle = {Proceedings - 28. Workshop Computational Intelligence, Dortmund, 29. - 30. November 2018},
pages = {257-270},
publisher = {KIT Scientific Publishing, Karlsruhe},
note = {**Young Author Award**},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
