Systems and Means of Informatics

2026, Volume 36, Issue 1, pp 91-103

MODELING DISCRETE DISTRIBUTIONS IN EDUCATIONAL CONTENT GENERATION TASKS

A. V. Bosov
A. V. Ivanov

Abstract

The article presents the final result of a study on the application of generative adversarial neural networks (GANs) for the creation of educational content. Wasserstein GANs (WGANs) are selected as an alternative to standard deep convolutional networks, which are characterized by high training complexity and impose restrictions on the kinds of generated content. A review of practical applications highlights several advantages of this type of network. In the present study, a WGAN with gradient penalty is applied to the problem of generating a vector of parameters for examination tickets. The neural network was trained and evaluated using a set of mathematical problems for the final exams in the course \Theory of functions of a complex variables" previously developed and labeled by experts. In practice, this generation technique demonstrated advantages in training stability and achieved higher output quality. Furthermore, quality metrics for the generated educational content based on the Kullback{Leibler divergence are proposed.

[+] References (22)

Bosov, A. V., and A. V. Ivanov. 2024. O primenenii generativnykh modeley v sisteme elektronnogo obucheniya matematicheskim distsiplinam [On the application of generative models in an electronic learning system for mathematical disciplines]. Informatika i ee Primeneniya - Inform. Appl. 18(2):72-81. doi: 10.14357/19922264240210. EDN: UWKQLN.
Bosov, A. V., and A. V. Ivanov. 2026. One approach to mathematical e-learning systems content generation. Intelligent systems. Part 1. Eds. A. Diveev, V. Fomichev, A. Ilin, I. Zelinka, and E. Sofronova. Communications in computer and information science ser. Cham: Springer. 2603:130-141. doi: 10.1007/978-3-032-04758-8-10.
Bosov, A. V., N. L. Selivanova, and Yu. Yu. Pustylnik. 2024. Tsifrovoy obrazovatel'nyy kontent dlya sovremennogo uroka: tekhnologiya ekspertizy kachestva [Digital educational content for a modern lesson: Quality assessment technology]. Obrazo- vanie i samorazvitie [Education and Self-Development] 19(3):98-113. doi: 10.26907/ esd.19.3.08. EDN: MZIPFX.
Goodfellow, I. 2017. NIPS 2016 tutorial: Generative adversarial networks. Cornell University. 57 p. Available at: https://arxiv.org/pdf/1701.00160 (accessed March 12, 2026).
Arjovsky, M., and L. Bottou. 2017. Towards principled methods for training generative adversarial networks. Cornell University. 17 p. Available at: https://arxiv. org/pdf/1701.04862 (accessed March 12, 2026).
Yu, L., W. Zhang, J. Wang, and Y. Yu. 2017. SeqGAN: Sequence generative adversarial nets with policy gradient. 31st Conference on Artificial Intelligence Proceedings. San Francisco, CA: AAAI Press. 2852-2858. doi: 10.5555/3298483.3298649.
Jang, E., S. Gu, and B. Poole. 2016. Categorical reparameterization with Gumbel- Softmax. Cornell University. 13 p. Available at: https://arxiv.org/pdf/1611.01144 (accessed March 12, 2026).
Arjovsky, M., S. Chintala, and L. Bottou. 2017. Wasserstein generative adversarial networks. 34th Conference (International) on Machine Learning Proceedings. 70:214{ 223. doi: 10.5555/3305381.3305404.
Gulrajani, I., F. Ahmed, M. Arjovsky, V. Dumoulin, and A. Courville. 2017. Improved training of Wasserstein GANs. 31st Conference (International) on Neural Information Processing Systems Proceedings. Red Hook, NY: Curran Associates. 5769^779. doi: 10.5555/3295222.3295327.
Anaya-Sanchez, H., L. Altamirano-Robles, R. Diaz-Hernandez, and S. Zapotecas- Martinez. 2025. WGAN-GP for synthetic retinal image generation: Enhancing sensor- based medical imaging for classification models. Sensors - Basel 25(1): 167. 20 p. doi: 10.3390/s25010167.
Yuda, E., T. Ando, L. Kaneko, Y. Yoshida, and D. Hirahara. 2024. Comprehensive data augmentation approach using WGAN-GP and UMAP for enhancing Alzheimer's disease diagnosis. Electronics 13(18):3671. 10 p. doi: 10.3390/electronics13183671.
Park, C., J. W. Kang, D. E. Lee, W. Son, S. M. Lee, C. Park, and M. W. Kim. 2024. W-DRAG: A joint framework of WGAN with data random augmentation optimized for generative networks for bone marrow edema detection in dual energy CT. Comput. Med. Imag. Grap. 115:102387. 13 p. doi: 10.1016/j.compmedimag.2024.102387.
Lei, K., M. Mardani, J. Pauly, and S. Vasanawala. 2021. Wasserstein GANs for MR imaging: From paired to unpaired training. IEEE T. Med. Imaging 40(1):105{115. doi: 10.1109/TML2020.3022968.
Fang, R., R. Guo, M. Zhao, and M. Yao. 2023. Low-count PET image reconstruction algorithm based on WGAN-GP. 3rd Conference (International) on Bioinformatics and Intelligent Computing Proceedings. 60{64. doi: 10.1145/3592686.3592698.
Tirel, L., A. Mohamed, and A. Hashim. 2024. Novel hybrid integrated Pix2Pix and WGAN model with gradient penalty for binary images denoising. Systems Soft Computing 6:200122. 11 p. doi: 10.1016/j.sasc.2024.200122.
Heng, Y., M. Yinghua, F. G. Khan, A. Khan, F. Ali, A. A. AlZubi, and Z. Hui. 2025. Survey: Application and analysis of generative adversarial networks in medical images. Artif. Intell. Rev. 58(2):39. 81 p. doi: 10.1007/s10462-024-10992-z.
Panwar, S., P. Rad, T.-P. Jung, and Y. Huang. 2020. Modeling EEG data distribution with a Wasserstein generative adversarial network to predict RSVP events. IEEE T. Neur. Sys. Reh. 28(8):1720{1730. doi: 10.1109/TNSRE.2020.3006180.
Venugopal, A., and D.R. Faria. 2024. Boosting EEG and ECG classification with synthetic biophysical data generated via generative adversarial networks. Appl. Sci. - Basel 14(23): 10818. 21 p. doi: 10.3390/app142310818.
Yang, L., D. Zhang, and G. E. Karniadakis. 2020. Physics-informed generative adversarial networks for stochastic differential equations. SIAM J. Sci. Comput. 42(1):A292{A317. doi: 10.1137/18M122540.
Walia, M.S., B. Tierney, and S. McKeever. 2020. Synthesising tabular data using Wasserstein conditional GANs with gradient penalty. CEUR Workshop Procee. 2771:325{336. doi: 10.21427/e6wa-sz92.
Salimans, T., I. Goodfellow, W. Zaremba, V. Cheung, A. Radford, and X. Chen. 2016. Improved techniques for training GANs. 30th Conference (International) on Neural Information Processing Systems Proceedings. Red Hook, NY: Curran Associates. 2234{ 2242. doi: 10.5555/3157096.3157346.
Kullback, S., and R. A. Leibler. 1951. On information and sufficiency. Ann. Math. Stat. 22(1):79{86. doi: 10.1214/aoms/1177729694.

[+] About this article