Information system for image generation with the possibility of federated learning and further training of generative models
DOI:
https://doi.org/10.34185/1562-9945-5-162-2026-16Keywords:
machine learning, diffusion models, neural networks, generative adversarial networks, testing, variational autoencoders, federated learningAbstract
The paper discusses the development and research of an information system for image generation based on modern generative artificial intelligence models with support for feder-ated learning and retraining mechanisms. The proposed system is focused on ensuring effec-tive generation of visual content while maintaining the confidentiality of user data, which is especially relevant in conditions of limited access to centralized data sets.
The research analyzes the architecture of the information system, the principles of inte-gration of generative models, as well as approaches to organizing federated learning, in which model parameters are updated on local nodes without transmitting the output data to the central server. Particular attention is paid to methods of retraining models, which allow the system to adapt to new types of images, styles and user requirements during operation.
The performance and efficiency of the proposed system are evaluated in terms of the quality of generated images, learning speed and resistance to changes in input data. The re-sults obtained confirm the feasibility of using a federated approach and further training of generative models to create scalable, adaptive, and secure image generation information sys-tems.
In the future, it is planned to expand the functionality of the system, including adding full registration and authorization, the ability to use multiple LoRA adapters simultaneously, increasing the number of models available for training and generation, and implementing ad-ditional algorithms for federated learning.
References
Goodfellow I. Pouget-Abadie J. Mirza M. et al. Generative Adver- sarial Networks. Elec-tronic resource] // arXiv.org. 2014 року. URL:https://arxiv.org/pdf/1406.2661
Ho J., Jain A., Abbeel P. Denoising Diffusion Probabilistic Модель. [Electronic resource] / / arXiv.org. 2020. URL: https://arxiv.org/pdf/2006.11239
Rombach R., Blattmann A., Lorenz D., et al. High-Resolution Image Synthesis with Latent Diffusion Models. [Electronic resource] // arXiv.org. 2022. URL: https://arxiv.org/pdf/2112.10752
Xiang C., Wang L., Zhou M. Comparative Analysis of Generative Models: Enhancing Im-age Synthesis with VAEs, GANs, and Stable Дифузія. [Electronic resource] // arXiv.org. 2023. URL: https://arxiv.org/pdf/2408.08751
Stable Дифузія. [Electronic resource] URL: https://hugging-face.co/blog/stable_diffusion
Hu EJ, Shen Y., Wallis P., et al. LoRA: Low-Rank Adaptation of Large Language Models. [Electronic resource] // arXiv.org. 2021. URL:https://arxiv.org/pdf/2106.09685
Yao Y., Gao T., Li W., et al. FedEx-LoRA: Exact Aggregation for Federated and Efficient Fine-Tuning of Foundation Models. [Electronic resource] // arXiv.org. 2025. URL:https://arxiv.org/pdf/2410.09432
Kairouz P., McMahan HB, Avent B., та інші. Advances and Open Problems in Federated Learning. [Electronic resource] // arXiv.org. 2021. URL: https://arxiv.org/pdf/1912.04977
Ostrovska K., Borysiuk V. «IMAGE GENERATION WITH THE ABILITY TO ENABLE FEDERATED TRAINING AND FURTHER TRAINING OF GENERATIVE MODELS» // the 3rd International Scientific and Practical Conference. International Scientific Unity. No-vember 12-14, 2025. Lisbon, Portugal. Р 203-206.
Downloads
Published
Issue
Section
License
Copyright (c) 2026 System technologies
This work is licensed under a Creative Commons Attribution 4.0 International License.
