We propose an artificial neural network architecture that utilizes A.G. Ivakhnenko’s group-based argument accounting method, aimed at sequentially increasing the degree of the Kolmogorov-Gabor polynomial and selecting the best options at each row. The method is embedded in the architecture of the Kolmogorov-Arnold artificial neural network, which is based on representing a function of many variables as a composition of functions of lower complexity in accordance with the superposition theorem. Neurons with “s-parabola” and “unit step” activation functions are used in the terminal layers. The resulting combined artificial neural network is called the Ivakhnenko-Kolmogorov-Arnold network. The experimental part of the work is concerned with solving a binary pattern recognition problem using invariant moments, using aircraft as an example. Upon completion of the network setup, when a vector of input features is fed to the input, the output value corresponds to the class number.
DOI: 10.1109/ICIEAM69213.2026.11549772
Download the article (PDF) or read online at IEEE Xplore (registration required): https://ieeexplore.ieee.org/document/11549772
At the Intelligent Control Laboratory of the Program Systems Institute of the RAS website: https://icontrol.psiras.ru/publications/
M. Khachumov, Y. Emelyanova and M. Shustova. Ivakhnenko-Kolmogorov-Arnold Artificial Neural Network and its Application in Pattern Recognition Problems // 2026 International Conference on Industrial Engineering, Applications and Manufacturing (ICIEAM), Sochi, Russian Federation, 2026, pp. 645–649.