search

sebastkm / hybrid-fem-nn

27 stars 17 forks source link

Considering the temporal coordinate (i.e., time) as an input? #7

Open jgim06932 opened 1 year ago

jgim06932 commented 1 year ago

Hi,

I have been reading your JCP paper on this hybrid FEM-NN and it is very interesting. But, it seems that all the examples described in the paper considers only the spatial coordinates, i.e., x, y, z, as inputs for the neural network.

So, I am wondering if this library can consider the temporal coordinate, i.e., time (t) as inputs in addition to the spatial ones?

Thank you for your great efforts on this again, and I look forward to hearing from you.

sebastkm commented 1 year ago

Hi, thanks for your interest in our work!

The temporal coordinate could also be an input, but we avoided this mainly due to the complications of extrapolating in time. A normal feed-forward neural network trained on the timeline [0, T), would likely not be able to predict anything meaningful for t > T.

If the purpose is to interpolate for t in the training interval [0, T), I think this can do fine.