Bezier curve as an activation function of artificial neural networks

Authors

Khachumov V. ,  Khachumov M.

Annotation

It is proposed to use the Bezier curve as an activation function of artificial neural networks. A variant of the modified Bezier interpolating curve as a basis function is considered. A relationship has been established between the parameters of the sigmoidal activation function and the curve. The basic requirements for the activation function are given and it is shown that the Bezier curve corresponds to them. The prospects of using the curve in artificial neural networks are outlined.

External links

DOI: 10.14357/20790279260302

Download the article (PDF) from the journal's website (in Russian): http://www.isa.ru/proceedings/images/documents/2026-76-3/9-13.pdf

Download the article (PDF) from eLibrary (in Russian, registration required): https://www.elibrary.ru/item.asp?id=93335279

Reference link

Khachumov, M., Khachumov, V. Bezier curve as an activation function of artificial neural networks // Proceeding of the Institute for Systems Analysis of the Russian Academy of Science, 2026, volume 76, № 3, pp. 9–13.