
Certificate of Completion
THIS ACKNOWLEDGES THAT
HAS COMPLETED THE SUMMER 2025 DEEP LEARNING BOOT CAMP
Navid Bahadoran
Roman Holowinsky, PhD
AUGUST 15, 2025
DIRECTOR
DATE

TEAM
Robust Physics-Informed Neural Networks for Modeling Reaction-Diffusion Dynamics under Adversarial Perturbations
Navid Bahadoran

This project develops a hybrid modeling framework using Physics-Informed Neural Networks (PINNs) and Graph Neural Networks (GNNs) to forecast the spatial spread of COVID-19. PINNs solve reaction-diffusion PDEs by embedding physical laws into neural network training, enabling mesh-free, continuous solutions. GNNs model spatial dependencies via county adjacency graphs, learning directly from data. Using real COVID-19 case data, we preprocess with log-transform and normalization, then train both models and compare their accuracy. PINNs yield lower prediction error and capture physical structure, while GNNs scale better across time and are suited for real-time graph-based modeling. The results highlight the trade-offs between physics-guided and data-driven approaches.
