Thanks a lot. This is an exciting and promising direction for NN's evolution. Maybe I'm wrong, but the formula for the Dirichlet principle should contain the squared gradient of u(x,..) (it could be obtained by multiplying diff. equation by u(x,..) and integrating by parts)
@@prakhars962 The nanoHUB tool "A Hands-on Introduction to Physics-Informed Neural Networks" is used in this hands-on tutorial and can be found on nanoHUB.org at: nanohub.org/tools/handsonpinns
Nice and clear presentation. The fourier features in the last network class worked excellent for my work. Can I somehow apply the same technique for images??
That is just one random function he took. I mean the equation was d psi/dx =f(x,psi). So he has taken some f(x, psi) to solve for psi to prove if his results are matching with the theoretical prediction.
UPDATE: it's working now. Need to put energy_tensor[j, 0] = 0.5 * (torch.sum(F**2) - 2.0) - torch.log((torch.det(F))) + 50.0*torch.log((torch.det(F)))**2 I'm getting RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn, for the Example 2 in this notebook. Is anyone getting this too? It seems like energy_tensor.requires_grad is False so can't actually do l.backward().
when you run into errors, you can submit a ticket through the nanoHUB ticket system so that the tool authors are notified. (They will likely not see comments posted here.)
![Physics Informed Neural Networks (PINNs) [Physics Informed Machine Learning]](http://i.ytimg.com/vi/-zrY7P2dVC4/mqdefault.jpg)
Excellent presentation. Thanks for sharing it. The only issue is the bad audio quality.
Thanks a lot. This is an exciting and promising direction for NN's evolution. Maybe I'm wrong, but the formula for the Dirichlet principle should contain the squared gradient of u(x,..) (it could be obtained by multiplying diff. equation by u(x,..) and integrating by parts)
You are right! Thanks for finding the typo!
What an effective presentation. Is it possible to download the Jupyter notebook?
Me too. Can't find a way to download the Ipython code.
@@prakhars962 The nanoHUB tool "A Hands-on Introduction to Physics-Informed Neural Networks" is used in this hands-on tutorial and can be found on nanoHUB.org at: nanohub.org/tools/handsonpinns
@@tanyafaltens5967 thx
Nice and clear presentation. The fourier features in the last network class worked excellent for my work. Can I somehow apply the same technique for images??
How did you get the formula at 11:17? exp(-x / 5.0) *cos(x) - Psi / 5.0. Thank you for your help
That is just one random function he took. I mean the equation was d psi/dx =f(x,psi). So he has taken some f(x, psi) to solve for psi to prove if his results are matching with the theoretical prediction.
Are there publicly available codes for these examples?
Yes - you can see the code and use the Jupyter notebook in nanoHUB at this link: nanohub.org/tools/handsonpinns
@@tanyafaltens5967 Thank you!
anyone got the codes? I don't seem to find it.
I cant find the notebook, I just see the video and presentation
Can you specify the ref [Raise 2019] ?
sir , you should have good mic for clear voice. At least , subtitles should be there. thanku
kindly apply this..must include subtitles in video so that anyone can easily get your point which is most important for us
How can I get the codes? Please help
The code is available here: nanohub.org/tools/handsonpinns
@@tanyafaltens5967 no its not
@@bigh8438 Did you launch the tool or download the source code- that link is right under the launch tool button.
I see- there is an error now when launching the tool. I submitted a help ticket for that. You can still download the tarball.
@@bigh8438 Things are fixed now (as of earlier this morning). Please try again!
Nicccccceeeee, fast and clear.
Could someone suggest some more info (a book, maybe a course) to dive in in the field?
1) research papers, 2) practice
@@csanadtemesvari9251 Could you suggest a good paper to feel the power of the method?
ruclips.net/p/PLCAl7tjCwWyGjdzOOnlbGnVNZk0kB8VSa
Nice!
This how FB maps it’s website.
UPDATE: it's working now. Need to put energy_tensor[j, 0] = 0.5 * (torch.sum(F**2) - 2.0) - torch.log((torch.det(F))) + 50.0*torch.log((torch.det(F)))**2
I'm getting RuntimeError: element 0 of tensors does not require grad and does not have a grad_fn, for the Example 2 in this notebook.
Is anyone getting this too?
It seems like energy_tensor.requires_grad is False so can't actually do l.backward().
when you run into errors, you can submit a ticket through the nanoHUB ticket system so that the tool authors are notified. (They will likely not see comments posted here.)