AOAB ‒ Automated Optimization Algorithm Benchmarking

Authors

Abstract

In this paper we present AOAB, the Automated Optimization Algorithm Benchmarking system. AOAB can be used to automatically conduct experiments with numerical optimization algorithms by applying them to different benchmarks with different parameter settings. Based on the results, AOAB can automatically perform comparisons between different algorithms and settings. It can aid the researcher to identify trends for good parameter settings and to find which algorithms are suitable for which type of problem.
We introduce the system structure of AOAB (the server and the graphical client interface), define the way in which optimizers and benchmark functions can be implemented for the use in AOAB, and conduct an illustrative example experiment with our system: a comparison between Random Search and two Hill Climbers.

Keywords

Benchmark Problems, Real Vector-based Search Spaces

BibTeX

@inproceedings{WNT2010AOABAOAB,
  author                    = {Thomas Weise and Li Niu and K{\={e}} T{\'{a}}ng},
  title                     = {{AOAB {--} Automated Optimization Algorithm Benchmarking}},
  booktitle                 = {Black Box Optimization Benchmarking (BBOB'10), Companion Publication of the Genetic and Evolutionary Computation Conference (GECCO'10 Companion)},
  publisher                 = {{ACM Press: {New York, NY, USA}}},
  pages                     = {1479--1486},
  year                      = {2010},
  location                  = {{Portland Marriott Downtown Waterfront Hotel: {Portland, OR, USA}}},
  url                       = {http://home.ustc.edu.cn/~newly/newly.files/Academy/paper.pdf},
  doi                       = {10.1145/1830761.1830763},
  key                       = {WNT2010AOABAOAB},
},

Links