Development of a Simulator for Generating Synthetic Data to Train Neural Networks under Uncertainty
V. V. Kovalev, А А Корнеева, Anastasija O. Fedorkova, E. A. Maslennikova, Denis M. Filatov, Danil P. Mikhailov · 2025
This paper presents the development of a simulator for generating synthetic data intended for training neural networks under conditions of uncertainty. The proposed approach enables the modeling of complex scenes with a high level of photorealism to enhance the generalization capability of machine learning algorithms when applied in real-world scenarios. The study explores methods for generating artificial data, their use in training neural networks, and approaches to accounting for uncertainty factors. Experiments have been conducted to demonstrate the effectiveness of synthetic data in improving model robustness to changing conditions.