Pedro H. S. Soares, Renato Candido, Magno T. M. Silva. Comparações entre a rede Kolmogorov-Arnold e a rede perceptron multicamada. Simpósio Brasileiro de Telecomunicações – SBrT’2025, 2025, Natal, Anais do SBrT 2025, 2025, pp. 1–2. (in portuguese)

Abstract

In this work, classical multilayer perceptron (MLP) networks are compared with the newly proposed Kolmogorov-Arnold networks (KAN) in a regression problem and a binary classification problem. KAN, based on adjustable splines, offers greater interpretability at the cost of higher computational expense and slower convergence.

Keywords

machine learning, neural networks, Kolmogorov-Arnold theorem, splines, interpretability.

Downloads

• PDF