Figure 4
From: Investigating molecular transport in the human brain from MRI with physics-informed neural networks
Influence of PINN hyperparameters on the diffusion coefficient estimated from clinical data. (a) Diffusion coefficient during training for different PDE weights \(w_r\) and exponentially decaying learning rate from \(10^{-4}\) to \(10^{-5}\). Dashed lines indicate result with residual based adaptive refinement (RAR). (b) Estimated diffusion coefficient with \(p=2\) for different PDE weights \(w_r\) as a function of the \(\ell ^1\)-norm of the residual after training. The values for FEM and the green horizontal bars indicating an error estimate are taken from Valnes et al.28.
