I realise that for the Physics informed part you use MSE Loss but for the pointwise solution error (supervised training) you use your custom LP Loss function. I understand that when comparing to zero you can't do a relative comparison. But why not use absolute for both, or MSE for both. Is there a reason to use a mixture of methods?
I realise that for the Physics informed part you use MSE Loss but for the pointwise solution error (supervised training) you use your custom LP Loss function. I understand that when comparing to zero you can't do a relative comparison. But why not use absolute for both, or MSE for both. Is there a reason to use a mixture of methods?