Physics Informed Neural Networks (PINNs) (Udemy.com)

Thank you so much for the content, Mohammed. I really enjoyed the course and learned a lot of new skills. Even though I’m still at a beginner level in programming, I was able to follow along by using Copilot chat in Visual Studio Code, which made the learning process much smoother.

I definitely suggest this course for anyone wanting to learn about PINNs! The professor presents the topic of PINNs with many examples, not only with pytorch but also a nice library called deepxde that makes the lives of people easier. There are detailed explanations on how to set up the models, that go step by step, without leaps, so that people can easily follow. Also, to understand from a practical point of view the advantage of PINNs, the first sections present the 'traditional' way of solving 1D and 2D problems. I am putting 4.5 and not 5 stars, just because maybe the physical modelling (types of BCs etc) was chosen on purpose to be simple, for the sake of presentation, but sometimes not accurate. Also, a suggestion for the future could be to add a section in this course for the case of building PINNs with different BCs at the same dataset. The problem could be a simple 1D problem, to keep things as simple as possible. Overall, for me this course is more than just worthwhile to follow!

I think the content covered was great. I did not give 5 stars because I think the flow could be improved. Specifically, it would be better (in my opinion) if the same problem sets are solved using the 3 different methodologies. The first part of the course consisted in solving the 1D heat equation and 2D Burgers equation using finite difference. However, the PINNs solutions were done for different equation sets and the solutions using Deepxde were done with yet a different set. That means that we can't directly compare each method side-by-side because the 3 methods solved different problems. Finally, the codes provided in each section should be commented for users to understand why certain libraries are used, etc.

This was a good course on PINNs and a good review on calculus concepts I haven't seen in a while. At first it was a bit hard to follow the instructor coding through the problems as not everything was explained thoroughly, however the various examples were similar in nature and so by the end of the course the concepts were more clear. I do wish the instructor would offer downloads of the slides, as they contain pertinent information and are seldom on-screen.

Its a decent introductory course. - It would have been better if the instructor had used the same example for 1D and 2D for PDF, PINNs (torch coding) and DeepXDE. Would have maintained a good flow. - I would have like to look at how PINNs can be trained for a more general purpose use, such as 1D/2D flow/heat-dissipation for a range of geometry and boundary conditions. And see how much faster PINNs are as compared to CFD. I understand the training dataset and time required would have been much longer, but that's the primary use-case of PINNs. I hope the instructor takes my review in positive light for his future courses. Overall, it was a decent course and worth the price :)

A lot of the times, I am wishing for more explanations of the "why" behind the examples. What is Burger's Equation modelling? What would be a closed form solution to see if my PINN works as intended? Why design a Neural Network in a certain way? Also, if they are using Images from academic literature, I highly appreciate this, but recommending that exact literature would have been super interesting to me as well!

The course content is good. As I came to learn the basics of PINNs and their implementation, I'm glad I was introduced to DeepXDE. But the way the codes are explained is pretty bad to a newcomer. The "on the fly" reading of the code makes it difficult to grasp that line of code, and I had to ask ChatGPT to understand it, where at some places it's still unclear. Maybe if the instructor re-records with an explanation of each line of code with syntax and why that line is needed, the content might be delivered better.

I thoroughly enjoyed the course on PINNs. Despite having some prior knowledge, I found the detailed explanations highly informative. Thank you for offering such a valuable learning experience. One suggestion for improvement would be to include sections that compare and visually illustrate the differences between numerical solutions and PINN models.

Good course and good explanation of the topic. only question is how are you defining the boundary conditions. I mean at the centre there is higher value of temperature and at the it is zero. this is defined as initial condition. then what is exactly defined in the boundary condition. Is it like temperature is zero at the boundary and it is defined as zero using boundary condition ? whats is physical meaning ??

Mohammad's style of teaching is really affective. He has provided example code, which he runs through explaining the basic theory and than demonstrates by coded examples, which I can run on my side. I'm really enjoying the course and feel like my knowledge, and ability to implement PINN's has increased significantly.