LaSDI: A Data-Driven Approach to Reduced Order Models
In this talk, I will discuss a novel data-driven technique for generating reduced order models (ROMs) for computational expensive parameter-dependent PDEs. LaSDI, Latent-Space Dynamic Identification, relies on finding a latent-space for existing full-order simulations, approximating their dynamics, and using this to generate new simulations within the parameter space. I will present the full method, how it can be implemented, and examples of both advection-dominated and diffusion-dominated problems. Ultimately, I will present results that show small relative reconstruction errors and impressive speedups.