Latent Space Dynamic Identification

Latent space dynamic identification (LaSDI) aims to create efficient reduced-order models (ROMs) for complex dynamical systems by learning simplified governing equations in a lower-dimensional latent space. Current research focuses on improving LaSDI's robustness to noise (e.g., through weak-form formulations) and efficiency (e.g., via greedy active learning and optimized neural network architectures like Taylor series-based networks), often incorporating physics-informed constraints for enhanced accuracy. These advancements significantly accelerate simulations of high-dimensional systems (e.g., fluid dynamics, plasma physics) while maintaining high fidelity, impacting fields requiring computationally intensive modeling.