Physics Informed Neural Networks (PINNs) [Physics Informed Machine Learning]

Steve Brunton
Steve Brunton
49,080
0
2024-05-29に共有
This video introduces PINNs, or Physics Informed Neural Networks. PINNs are a simple modification of a neural network that adds a PDE in the loss function to promote solutions that satisfy known physics. For example, if we wish to model a fluid flow field and we know it is incompressible, we can add the divergence of the field in the loss function to drive it towards zero. This approach relies on the automatic differentiability in neural networks (i.e., backpropagation) to compute partial derivatives used in the PDE loss function.

Original PINNs paper: www.sciencedirect.com/science/article/abs/pii/S002…

Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations
M. Raissi P. Perdikaris, G.E. Karniadakis
Journal of Computational Physics
Volume 378: 686-707, 2019

This video was produced at the University of Washington, and we acknowledge funding support from the Boeing Company

%%% CHAPTERS %%%
00:00 Intro
01:54 PINNs: Central Concept
06:38 Advantages and Disadvantages
11:39 PINNs and Inference
15:23 Recommended Resources
19:33 Extending PINNs: Fractional PINNs
21:40 Extending PINNs: Delta PINNs
25:33 Failure Modes
29:40 PINNs & Pareto Fronts
31:57 Outro
コメント (21)
  • @rehankhan-gn2jr
    The way of teaching is highly beneficial and outstanding. Thank you, Steven!
  • @alessandrobeatini1882
    This is hands down one of the best videos I've seen on YouTube. Great work, keep it up!
  • @jiaminxu7275
    Hi Prof. Brunton, I am a Ph.D. student from UT Austin majoring in Mechanical Engineering with specification of dynamical system and control. Your vedios has been helping me by either giving me a deeper understanding of foundamental knowledge or openning my horizon, ever since I begin my Ph.D. Just want to express my great gratitude to you again and hope I can meet you in certain conferences so that I can say thank you to you in person.
  • @code2compass
    Steve your videos are always helpful, clear and concise. Thank you so much for such amazing content. You are my hero
  • @markseagraves5486
    Very helpful Steven. I work in consciousness studies and find too often the math is written off as too complicated. On the other side, many computational scientists may write off consciousness studies as too ethereal to be of much value. Bridging these two worlds with insight and rigor, I feel advances our understanding of both artificial and human intelligence. You have contributed to this effort here. Thank you.
  • @THEPAGMAN
    This is really helpful, if only you posted this sooner! Thanks <3
  • @mostafasayahkarajy508
    Thank you very much for the lecture. I am looking forward for your next lecture on this topic.
  • @ryansoklaski8242
    I would love to see a video on Universal ODEs (which leverages auto-diff through diffEQ solvers). Chris Rackauckas' work in the Julia language on these methods has been striking - would love to see your take on it.
  • @blacklabelmansociety
    Hi Professor Steve. I’d love to see a series on Transformers. Thanks for your content, greetings from Brazil.
  • @drozdchannel8707
    Great video! it may be useful to do another video about Neural Operators. It is more stable and faster in many physical tasks as i know.