Огляд існуючих генеративних нейронних мереж

Abstract

This work is devoted to a look at research generative neural networks, namely the principles of generative neural networks (GNN) capable of creating new data resembling their training sets. The training process involves data preprocessing, training on carefully prepared data, and ultimately generating new data. Various types of GNNs, such as Variational Autoencoders, Generative Adversarial Networks, Transformer Networks, and Diffusion Models, offer di verse applications in content creation and beyond. Challenges include understanding decision-making processes, biases, and potential misuse, but optimistic prospects involve refining algorithms and leveraging technological advancements.