Towards Realistic Optimization Benchmarks: A Questionnaire on the Properties of Real-World Problems
CrossMind.ai logo

Towards Realistic Optimization Benchmarks: A Questionnaire on the Properties of Real-World Problems

Jan 18, 2021
|
38 views
|
Details
Abstract: Optimization algorithms are commonly compared on benchmarks to get insight into performance differences. However, it is not clear how well, and to which extent, benchmarks match the properties found in real-world problems. This work investigates the properties of real-world problems through a questionnaire, in order to enable the design of benchmark problems that more closely match those found in the real world. Initial results show that the vast majority of problems are constrained and require substantial computation times for both the objectives and the constraints. Further, known optimal solutions and analytical gradients are rarely available, limiting the options in guiding the optimization process. All of these are important considerations to be taken into account in the design of realistic benchmark problems. To further improve the understanding of real-world problems, the reader is encouraged to fill out the questionnaire: https://tinyurl.com/opt-survey Authors: Koen van der Blom, Timo M. Deist, Tea Tušar, Mariapia Marchi, Yusuke Nojima, Akira Oyama, Vanessa Volz and Boris Naujoks (Leiden University, Centrum Wiskunde & Informatica, Jozef Stefan Institute, ESTECO SpA, Osaka Prefecture University, Japan Aerospace Exploration Agency, modl.ai, TH Köln)

Comments
loading...