Application of the growing neural GAS algorithm to optimize computational resources in neural network architecture
2025, vol.17 , no.2, pp. 25-32
Article [2025-02-03]
The paper is devoted to the study of methods for improving the efficiency of using and training neural networks, given their complex architectures and a large number of parameters. The author has developed a method for optimizing the number of trained parameters using the growing neural gas algorithm. Approbation results show that the proposed method, combined with convolutional layers, shows a degradation in accuracy compared to traditional convolutional networks, but still outperforms random guessing. Changing the criterion for adding new nodes based on the current error reduced the number of neurons by half. The conclusions indicate that the developed method does not compete with existing solutions, but its further improvement may lead to optimization of computational resources in neural networks.
neural network optimization, machine learning automation, growing neural gas algorithm
https://doi.org/10.59035/ORVP4379
N. V. Eliseeva, V. E. Petrov, Yu. V. Kaykova, I. V. Kaykova. Application of the growing neural GAS algorithm to optimize computational resources in neural network architecture. International Journal on Information Technologies and Security, vol.17 , no.2, 2025, pp. 25-32. https://doi.org/10.59035/ORVP4379