An article titled "Learnable activation functions in physics-informed neural networks for solving partial differential equations," authored by Computational Science and Engineering PhD candidate Afrah Farea, has been published in the journal Computer Physics Communications.

In this study, the use of learnable activation functions is investigated to address key challenges in Physics-Informed Neural Networks (PINNs), such as spectral bias and convergence instability. Through Neural Tangent Kernel and Hessian analyses, various network architectures and partial differential equation (PDE) types are examined.

The published article can be accessed via the link below, and the related codes have been made available as open-source on GitHub for researchers.

🔗 Article link: https://doi.org/10.1016/j.cpc.2025.109139

Related Github repository can be found at  https://github.com/afrah/pinn_learnable_activation