Bayesian-Optimized Physics-Informed Neural Networks for the FitzHugh-Nagumo Model

Physics-Informed Neural Networks (PINNs) offer a promising bridge between deep learning and biophysical modeling by embedding differential equations directly into the learning process. This paper explores an automated framework using Bayesian Optimization (BO) and PINNs in order to model electrophysiological processes. The FitzHugh-Nagumo (FHN) model is used as a fundamental system in excitable media research to test this approach. Our study uses BO to automatically tune the structural hyperparameters of the network, specifically the number of layers and neurons. This demonstrates the potential of BO-PINNs to simplify the model selection process for time-dependent dynamics. This paper was developed within the framework of the STRATIFYHF project.