Speaker
Description
The field of scientific machine learning (SciML) seeks to fuse traditional mathematical modeling with advances in machine learning to balance mechanist equations with data-driven inference, resulting in computational models that preserve scientific knowledge while readily adapting to the unknown through data-driven discovery. These advancements are setting the foundation for which SciML surrogates provide novel diagnostics that decompose global and local behavior inherent to high fidelity stochastic models and real-world data. This presentation will introduce neural network (NN) function approximations of model-form error to close the gap between ordinary differential equations (ODE) to an epidemiological agent-based model (ABM). This universal differential equations approximation to the ABM allows us to preserve the foundational ODE that represents the global disease dynamics and couples it with the NN for function approximations of nonlinear state transition dynamics. Equipped with an approximation to the ABM, we further our investigation to account for variability in contact patterns in real-world data by analyzing county level disease dynamics within the state of New Mexico U.S. and compare the result to trends observed in aggregation for the total state dynamics. Ultimately, we will introduce a novel diagnostic to elucidate the regional and global epidemiological trends to fundamentally understanding the impact of regional versus global trends on disease transmission.
