The Impact of Parameterized Activation Functions in Physics-Oriented Neural Networks

This project was undertaken as coursework for the graduate course in Mathematics, MAT1510: Deep Learning Theory and Data Science, at the University of Toronto during my undergraduate studies.

This project aimed to study parameterized functions and their roles in learning PDEs in neural networks designed for physics. This was motivated by the PINNsFormer paper and its introduction of the Wavelet activation function. I explored an expanded version of the ablation study provided by the authors in studying how different parameterized and non-parameterized activation functions affect the learning of PDEs in neural networks.

Full details about the project can be found in the project report.

Link to the project report.