Home Heaviside function as an activation function
Article
Licensed
Unlicensed Requires Authentication

Heaviside function as an activation function

  • Naoya Hatano , Masahiro Ikeda , Isao Ishikawa and Yoshihiro Sawano EMAIL logo
Published/Copyright: October 26, 2022

Abstract

The uniform closure of the neural networks with Heaviside activation is specified.

MSC 2010: 26A33

Award Identifier / Grant number: 19K14581

Award Identifier / Grant number: 19K03546

Funding statement: This work was supported by a JST CREST Grant (Number JPMJCR1913, Japan). This work was also supported by the RIKEN Junior Research Associate Program. The second author is supported by a Grant-in-Aid for Young Scientists Research (No. 19K14581), Japan Society for the Promotion of Science. The fourth author is supported by a Grant-in-Aid for Scientific Research (C) (No. 19K03546), Japan Society for the Promotion of Science and People’s Friendship University of Russia.

Acknowledgements

The authors are grateful to the anonymous referee for his careful reading of this paper.

References

[1] G. Cybenko, Approximation by superpositions of a sigmoidal function, Math. Control Signals Systems 2 (1989), no. 4, 303–314. 10.1007/BF02551274Search in Google Scholar

[2] B. Hanin, Universal function approximation by deep neural nets with bounded width and ReLU activations, Mathematics 7 (2019), no. 10, Paper No. 992. 10.3390/math7100992Search in Google Scholar

[3] N. Hatano, M. Ikeda, M. Ishikawa and Y. Sawano, A global universality of two-layer neural networks with ReLU activations, J. Funct. Spaces 2021 (2021), Article ID 6637220. 10.1155/2021/6637220Search in Google Scholar

[4] P. Kainen, V. Kůrková and A. Vogt, Best approximation by Heaviside perceptron networks, Neural Networks 13 (2000), no. 7, 695–697. 10.1016/S0893-6080(00)00056-3Search in Google Scholar

[5] A. Pinkus, Approximation theory of the MLP model in neural networks, Acta Numer. 8 (1999), 143–195. 10.1017/S0962492900002919Search in Google Scholar

[6] S. Sonoda and N. Murata, Neural network with unbounded activation functions is universal approximator, Appl. Comput. Harmon. Anal. 43 (2017), no. 2, 233–268. 10.1016/j.acha.2015.12.005Search in Google Scholar

Received: 2021-08-21
Revised: 2021-11-02
Accepted: 2021-12-14
Published Online: 2022-10-26
Published in Print: 2023-06-01

© 2022 Walter de Gruyter GmbH, Berlin/Boston

Downloaded on 18.9.2025 from https://www.degruyterbrill.com/document/doi/10.1515/jaa-2022-2012/html
Scroll to top button