Figure 15.

The time evolution of the MSE loss for different models used in solving the 2D heat equation. Except the vanilla PINN (no error), all models were trained with data sampled from the initial timeslice corrupted with additive Gaussian noise with zero mean and $\sigma =1.0$. MSE error evaluated over 50k points chosen over the entire spatio-temporal domain for the different models is given in the accompanying table.