Застосування методів вищої математики в алгоритмах штучного інтелекту для автоматизації задач комп`ютерної інженерії

Abstract

This paper considers the application of mathematical methods to image classification using a multilayer perceptron. Digital images are represented as vectors in a high-dimensional space, which enables the use of linear algebra, differential, and integral calculus for neural network training. The loss function and the gradient descent method are analyzed, and the influence of training parameters on classification accuracy is investigated. The obtained results confirm the effectiveness of multilayer perceptrons for image classification tasks in computer engineering systems with limited computational resources.