Modeling Unknown Stochastic Dynamical System via Autoencoder

Speaker: Zhongshu Xu (OSU)
Dates: 2024/10/03
Location: MW 154
Abstract: This talk presents a novel data-driven method for modeling unknown stochastic dynamical systems using autoencoders. The approach focuses on learning the flow map of an underlying system by identifying unobserved latent variables and enforcing their density to approach a standard normal distribution. The method is particularly effective for low-dimensional systems with complex stochastic behaviors, including those driven by non-Gaussian noise. Through a range of numerical experiments, we demonstrate that the proposed autoencoder-based framework achieves high accuracy and robust predictions. This method provides a computationally efficient alternative for predicting the behavior of unknown stochastic systems, with potential for future applications in high-dimensional and real-world problems.