I would like to try transfer learning using Modulus.
Is it possible to do transfer learning like the following?
- Learn the laws of physics using enough simulation data for some different shapes.
- Learn about one of shapes using a few real experimental data for a shape.
It certainly is possible! Mixing between data / physics-based losses is an excellent area to explore. We dont have any examples doing this right now, but there is no limitation.
Which Modulus APIs or websites should we refer to design about basic transfer learning?
Hi @neutral, check the transfer learning example for the “limerock” example case. This may help.
Thank you for your reply.
Thank you for providing an answer to my question.
My pleasure! I hope it was useful!