This study examines the structure and characteristics of multilayer autoencoders (MAEs) used in detecting computer attacks. The potential of MAEs for improving detection capabilities in cybersecurity is analyzed, with a focus on their role in reducing the dimensionality of large datasets involved in identifying computer attacks. The study explores the use of different neuron activation functions within the network and the most commonly applied loss functions that define reconstruction quality of the original data. Additionally, an optimization algorithm for autoencoder parameters is considered, designed to accelerate model training, reduce the likelihood of overfitting, and minimize the loss function.

Keywords: neural networks, layers, neurons, loss function, activation function, mobile applications, attacks, hyperparameters, optimization, machine learning