Adaptive Social Spider Algorithm for Training Neural Networks

Abstract: The social spider algorithm (SSA) is a novel population-based meta-heuristic algorithm for solving global optimization problems. It is inspired from the special group of spiders which are social, live in groups and interact with others. These spiders communicate with each other through vibrations on the net and detect the location of the prey. Vibration is the main concept that distinguish social spider algorithm from the other meta-heuristic algorithms. The algorithms performance is tested on global optimization problems and compared with the state-of-the-art optimization algorithms. The results showed that the performance of social spider algorithm is superior to the compared algorithms. The SSA algorithm is also used to train artificial neural networks (ANNs) by the authors and given good results. In this study, different from the basic social spider algorithm that uses fixed parameter scheme, adaptive parameter scheme is used for parameter tuning in order to improve the performance of the algorithm. The SSA algorithm has three user-controlled parameters affecting the quality of the solution which are ra, pc and pm. Among these parameters ra determines the rate of vibration attenuation when propagating over the spider web, pc manages the probability of the spiders changing their dimension mask in the random walk step and pm determines the probability of each value to be one in the dimension mask. By taking into consideration the iteration number, these parameters are computed adaptively in this study. Performance of the proposed algorithm is tested on some classification benchmark problems for training neural networks and compared with the basic SSA algorithm. Obtained results showed that the adaptive parameter tuning improves the capability of SSA algorithm for training neural networks.

Keywords: Social spider algorithm, Adaptive parameter tuning, Training neural networks

Publication details: Full-text conference paper presented at the International Symposium for Production Research 2017 (ISPR 2017), 13-15 September 2017, Vienna, Austria. Published in: Proceedings of the International Symposium for Production Research 2017 (ISPR 2017), pp. 345-350. Publisher: Çağlayan Kitabevi. ISBN: 978-975-436-100-1.

Acknowledgement: This study was supported by the Research Fund of Erciyes University (Project number: 6771).

How to cite / Kaynak gösterimi

APA 7: Gülmez, B., & Kulluk, S. (2017, September). Adaptive Social Spider Algorithm for Training Neural Networks [Conference paper]. In M. N. Durakbaşa & M. G. Gençyılmaz (Eds.), Proceedings of the International Symposium for Production Research 2017 (ISPR 2017) (pp. 345–350). Çağlayan Kitabevi. International Symposium for Production Research 2017 (ISPR 2017), Vienna, Austria. https://doi.org/10.5281/zenodo.23036857

IEEE: B. Gülmez and S. Kulluk, "Adaptive Social Spider Algorithm for Training Neural Networks," in Proceedings of the International Symposium for Production Research 2017 (ISPR 2017), Vienna, Austria, Sep. 2017, pp. 345–350, ISBN: 978-975-436-100-1, doi: 10.5281/zenodo.23036857.

Chicago: Gülmez, Burak, and Sinem Kulluk. 2017. "Adaptive Social Spider Algorithm for Training Neural Networks." In Proceedings of the International Symposium for Production Research 2017 (ISPR 2017), 345–350. Vienna, Austria: Çağlayan Kitabevi. https://doi.org/10.5281/zenodo.23036857.

MLA 9: Gülmez, Burak, and Sinem Kulluk. "Adaptive Social Spider Algorithm for Training Neural Networks." Proceedings of the International Symposium for Production Research 2017 (ISPR 2017), Çağlayan Kitabevi, 2017, pp. 345–350. https://doi.org/10.5281/zenodo.23036857.

Harvard: Gülmez, B. and Kulluk, S. (2017) 'Adaptive Social Spider Algorithm for Training Neural Networks', in Proceedings of the International Symposium for Production Research 2017 (ISPR 2017). Vienna, Austria: Çağlayan Kitabevi, pp. 345–350. doi: 10.5281/zenodo.23036857.

BibTeX:

@inproceedings{Gulmez2017adaptive,
  author    = {Gülmez, Burak and Kulluk, Sinem},
  title     = {Adaptive Social Spider Algorithm for Training Neural Networks},
  booktitle = {Proceedings of the International Symposium for Production Research 2017 (ISPR 2017)},
  year      = {2017},
  month     = {sep},
  pages     = {345--350},
  address   = {Vienna, Austria},
  publisher = {Çağlayan Kitabevi},
  isbn      = {978-975-436-100-1},
  doi       = {10.5281/zenodo.23036857},
  url       = {https://doi.org/10.5281/zenodo.23036857}
}