Speaker
Description
Physics-Informed Neural Networks (PINNs) have emerged as a powerful tool for solving forward and inverse problems by embedding physical laws directly into the loss function. However, PINNs often struggle with spectral bias, making it difficult to capture the high-frequency oscillations and stiff dynamics present in modeling excitable cells. Finite-Basis PINNs (FBPINNs) address these limitations by integrating domain decomposition methods with deep learning. By partitioning the global domain into smaller subdomains, FBPINNs enhance the local approximation capabilities of the network, effectively capturing sharp gradients and fast temporal scales. In this work, we demonstrate the efficacy of FBPINNs in modeling excitable cells, which are governed by coupled reaction-diffusion partial differential equations with stiff ordinary differential equations. We focus on the FitzHugh-Nagumo model, a fundamental model in computational neuroscience and cardiology. Our results show how the FBPINNs can overcome the training instabilities of traditional PINNs when dealing with stiff ionic dynamics. Finally, we discuss the strengths and limitations of this approach.