An improved periodic activation for PINNs reconstructing convective flows

This paper investigates network architectures with complex exponential activation functions for Physics-Informed Neural Networks (PINNs), which generate paired sine and cosine outputs. The proposed method is evaluated against comparable sine-activated multi-layer perceptrons for the task of temperature reconstruction from sparse velocity data in convective flows. The study demonstrates that periodic activations provide benefits compared to monotonic counterparts in this fluid dynamics application.