Generalization of the network

[Mengting009]

Hi Lu,

Thank you for sharing the library.

It seems that a trained network can only be used to predict the results of the same geometry with the same boundary condition and initial condition. Once any of the above-mentioned conditions is changed, the network has to be trained again for new predictions. I am wondering is there a possibility to generalize the approach such that once the network is trained, it can be used for predictions with different initial and boundary conditions? Thank you!

[lululxvi]

Yes, PINN is for specific IC/BC. If you want to train a network for different IC/BC, you can use these IC/BC as an extra network input, and train the network for different IC/BC. Another way is to use DeepONet, see https://doi.org/10.1038/s42256-021-00302-5

[alokwarey]

@lululxvi: Is there an example or sample code snippet on how to setup a problem for different IC/BC using PINNs with deepxde?

[lululxvi]

No good examples yet, but see #273