Analysis of types of generative competitive networks

Keywords: змагальна мережа, аналіз, алгоритм, нейронна мережа, дискримінатор, генератор, дані


The article analyzes the types of generative competitive networks RNNGAN, WGAN, BiGAN. Their structure and components are disclosed. It is noted that in modern research in the field of artificial intelligence and signal processing, an approach based on the use of recurrent neural networks (RNN) and generative adversarial networks (GAN) is widely used. One of the innovative concepts in this area is a model known as Recurrent Neural Network GAN. The architecture of the network is proposed and the loss function of the generator and discriminator is presented mathematically. The Recurrent Conditional GAN model is presented, which was created for the purpose of generating medical data, which is an urgent task in the modern fields of medical research and diagnostics. This approach uses a combination of recurrent neural networks (RNNs) and conditional generative adversarial networks (cGANs). The architecture of the network is proposed and the loss function of the generator and discriminator is presented mathematically. The principles of the TimeGAN network are disclosed. It is emphasized that the TimeGAN algorithm includes a framework that uses elements of conventional unsupervised GAN training approaches as well as a tutored approach. A complete architecture of the TimeGAN model and a description of the mathematical functions are offered. Bidirectional Generative Adversarial Networks (BiGAN) are analyzed, which are a type of generative adversarial networks that includes an encoder, in addition to the usual generator components and a discriminator, which transforms real data into a latent space into which the generator is input, actually performing the inverse function compared to the generator. It is noted that training generative-competitive networks (GAN) is a rather difficult task. There is a possibility that the models may not converge to the optimal state. The presented analysis highlights the promising prospects and diversity of the described approaches, which can contribute to the further development of the fields where they are applied, from medicine to art and engineering


1. Ісаєнков Я. О., Мокін О. Б. Аналіз генеративних моделей глибокого навчання та особливостей їх реалізації на прикладі WGAN. Вісник Вінницького політехнічного інституту. 2022. № 1. С. 82-94.
2. Сулема Є. С., Топчієв Б. С. Інтелектуальна колоризація зображень за допомогою генеративних змагальних мереж. «Системні технології» «System technologies», 2019. № 5 (124). С. 94-103.
3. Аналіз математичних моделей протидії банківським кібершахрайствам / Кузьменко О. В., Яровенко Г. М., Скринька Л.О. // Вісник СумДУ. Серія «Економіка», 2022. № 2’. С. 111-120.
4. Сеніва К. Р. Способи використання нейронних мереж та машинного навчання в комп’ютерних іграх. Вісник Хмельницького національного університету, 2021. №2 (295). С. 97-100.
5. T. Karras, S. Laine, M. Aittala, J. Hellsten, J. Lehtinen, and T. Aila, “Analyzing and Improving the Image Quality of StyleGAN,” 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020, pp. 8107-8116.

Abstract views: 0
PDF Downloads: 0
How to Cite
Prozur , V. (2023). Analysis of types of generative competitive networks. COMPUTER-INTEGRATED TECHNOLOGIES: EDUCATION, SCIENCE, PRODUCTION, (52), 104-110.
Computer science and computer engineering