A0875
Title: Bayesian closure modeling for dynamic systems
Authors: Toryn Schafer - Texas A&M University (United States) [presenting]
Abstract: Closure modeling is a key challenge in the simulation of dynamical systems, where unresolved processes must be represented accurately to ensure predictive fidelity. A Bayesian approach to closure modeling is explored by casting the problem as variable selection over a latent derivative process governed by an ordinary differential equation. Building on recent work in dynamic discovery, the focus is on scalable computational strategies for posterior inference, including implementations that target GPU architectures. This framework enables rigorous uncertainty quantification while maintaining computational tractability, even in the presence of sparse or noisy observations. The method is demonstrated on an ecological case study, and broader implications are discussed for dynamic model discovery in scientific domains.
