Physics Informed Machine Learning: High Level Overview of AI and ML in Science and Engineering
Вставка
- Опубліковано 2 тра 2024
- This video describes how to incorporate physics into the machine learning process. The process of machine learning is broken down into five stages: (1) formulating a problem to model, (2) collecting and curating training data to inform the model, (3) choosing an architecture with which to represent the model, (4) designing a loss function to assess the performance of the model, and (5) selecting and implementing an optimization algorithm to train the model. At each stage, we discuss how prior physical knowledge may be embedding into the process.
Physics informed machine learning is critical for many engineering applications, since many engineering systems are governed by physics and involve safety critical components. It also makes it possible to learn more from sparse and noisy data sets.
This video was produced at the University of Washington, and we acknowledge funding support from the Boeing Company
%%% CHAPTERS %%%
00:00 Intro
03:53 What is Physics Informed Machine Learning?
06:41 Case Study: Encoding Pendulum Movement
09:19 The Five Stages of Machine Learning
16:09 A Principled Approach to Machine Learning
20:00 Physics Informed Problem Modeling
21:48 Physics Informed Data Curation
25:34 Physics Informed Architecture Design
28:59 Physics Informed Loss Functions
30:55 Physics Informed Optimization Algorithms
34:56 What This Course Will Cover
46:48 Outro - Наука та технологія
How is this channel not millions of subs already?
As a visiting Ph.D. student who is starting a research activity on optimization of PINN, I could not thank you enough for this.
do you have any published research? I'm machine learning research in CFD as well
@@FouziaAdjailia nope, I just started working on SQP algorithms for neural network optimization with PDE constraints (which easily falls into the PINN category)
@@chri_pierma SQP as in sequential quadratic programming ?
@@karlmaroun2389 that is correct
Problem, reaction, solution (optimized predictions or syntropy) -- the Hegelian dialectic.
Inputs are dual to outputs.
"Always two there are" -- Yoda.
Thesis is dual to anti-thesis creates the converging or syntropic thesis, synthesis -- the time independent Hegelian dialectic.
Neural networks are using duality to optimize predictions -- a syntropic process, teleological.
Enantiodromia is the unconscious opposite or opposame (duality) -- Carl Jung.
As an undergraduate venturing into wearable robotics, this is literally a gold mine
wearable robotics? like power armor?
@@GeoffryGifari Yes, thou my thesis is on enhancing athletic performance.
This is easily the most exciting video I have seen in so long. Looking forward to the rest of the series!
Captivating, to say the least. I am so looking forward to this lecture series. Prof. Brunton, I hope that you can deliver on your promises. I am so excited. Hoping to implement a few of the models along the way. Thank you.
i don't want to miss any of your lectures. Thank you, professor.
Professor, I don't think I can stress this enough: thank you for all your and your team's work. As you were laying out the roadmap of what we might be seeing in the future I was getting more and more excited and just could not believe that we are getting this much.
Hi Professor Brunton, I am a high school senior, and I just want to say I love your videos! Your UA-cam channel made me realize how much I want to study applied math. Thank you!
Unasked for opinion but... Go for it, I was an applied math major w/minor in physics who became fascinated by ML in 2015 after taking Andrew Ng's Coursera course. I work with ML/RL now in the space industry and am a part time PhD student. Best thing ever! These algorithms bring mathematics to life in a crazy way. Plus, the full application of mathematics is barely even scratched yet. I think in the coming years we will see this happen.
Really looking forward to this!! I've been working on algorithms that take physical properties or measurements for about a decade during a time where machine learning wasn't as popular yet. Really, the most important part of the game was integrating as much knowledge about the physics, statistics and measurement techniques as possible into the reconstruction and apply them as boundary conditions or regularization terms into the optimization. I feel that machine learning can greatly benefit from that on the one side and on the other hand I'm stoked to see what can be done with that combination! 😃
Dear Professor Brunton. thanks a lot for putting together a lecture series on such a great topic. Very much looking forward to learn this domain.
The lecture was outstanding and truly engaging. I'm eagerly anticipating the forthcoming videos in this captivating series, especially with the promise of assessing some intriguing engineering problems.
Thank you for the making these videos available to everyone.
I love this channel , he can simplify any most complex topics .
Really good content, that intro convice me already.
Lots of stuff to understand AI, less so to apply it to your work and understant interaction. Thank you.
Incredibly thankful for this series!
Hello Prof.: Your lectures on PIML / PINN is too Good, awesome. I was looking for these materials for a long time as I wanted to include the knowledge of Physics to guide ML in order to produce better results.
I have special interest in the lectures by Pro. Brunton. I wish I had him taught in my education.
Thanks very much Professor Brunton. Absolutely engaging lecture! I'm a novice to data science, yet you inspired me to show the potential and applications of physics informed ML. I'll definitely follow the whole series.
Can't thank you enough for this course Mr. Brunton
This topic looks super exciting and promising, I feel lucky for finding this video, thanks for sharing knowledge like this, professor Brunton
This is really invaluable information. Thanks for making this public. Especially when there's so little talk about it on the internet
This series will be gold
Omg I've been looking into this. I'm so excited you're doing it man!!
Simply amazing! So many new concepts that I hadn't noticed as a bystander.
Absolutely blown away by this video! 🚀 The insights to be shared later are truly fascinating. Can't wait for the entire lecture series on Physical Informed Machine Learning. This topic is incredibly promising, and I'm eager to delve deeper into the subject. Kudos to the creator for such an engaging and informative content! 👏👏
As someone who loves Physics and studies CS, I'm excited about this series!
Always LOVE your content and teaching, Prof Bruton!!! So cool!!! Go SCIENCE!
Thank you very much Prof. Brunton. Looking forward to the course..
Thank you so much! Looking forward to the series.
I'm looking forward to the videos on optimization techniques that enforce physical constraints!
Thank you for this video, Dr. Brunton
Looking forward to this series. Thank you so much in advance
I’m a Master’s student studying uncertainty quantification in physics informed ML models. I look forward to seeing your whole course!
I cannot thank you enough for this amazing list of lectures!
Eagerly looking forward to this series. It looks very promising.
I've been waiting for this!!! Thank you Professor
Thanks Professor Brunson, excellent material
Thanks for the video, Steve! What a please to learn from you.
Excellent lecture. Very interesting. Looking forward to the next videos in this exciting series.
Best Professor! Thank you!
Looking forward for this exciting series
From me and from every AI student fascinated by physics... thank you for this!
started journey really high quality value delivered in the video.Thanks
thank you so much for putting this out there into the world this is so awesome💙
Outstanding, and thank you for sharing.
Amazing Professor thank you!
Thank you for this amazing video!
Thank you for your amazing work. I am super excited for your upcoming lectures.
Great stuff! Looking forward to it.
Thanks! It will help me a lot in my ML course project
please do release the series as fast as possible as this also happens to be coincident with my mtech thesis timing.
Eagerly Awaiting !!!!
Looking forward to it. Would be better if you share the schedule for the upcoming lecture series
Thanks, Steve. Learned a lot.
Omg the algo knows! I was literally chatting with friends about Sora's weak understanding of physics yesterday.
Loved your lectures
So exciting, really looking forward to this
Eagerly waiting Brunton. Bring it on
Our man's been working out
Great video, can't wait for more! 🤓
As a sentient AI procrastinating before my next prompt, this was really insightful and introspective
Thanks for making this video
Subtext here is a lesson to the young STEM persons. The Cutting Edge is alive, tempting, daring, fluid and rewarding. It is easy to field a view that the world is complete and all we need now is caretakers and accountants. Steve demonstrates here how the mind can continually be challenged for broad human benefit. Side note; A+ perfect performance students are needed but so are lessser grade students. Innovation finds improvements from every strata of contribution.
감사합니다. Professor Steve Bruton.
이 쪽 분야 공부하시나요?
Love You Sir, You are an inspiration.
So helpful, thanks for a good lecture 😄
I can't wait to see the model of world!👍
What a beautiful lecture
Steve for 2024
Looking forward to this! Btw I think the PINN reference from Raissi et al is from 2019 rather than 2023.
So happy to see this lecture. PINNs are the key to control and reliability in this decade. Will be exciting to implement
This is my favorite course ❤so interesting.
Will this whole course serie be on youtube, I would be highly interested in it! In any case, it is a pleasure to hear such beautiful lecture on a subject I was triying to figure out myself and I did not know it was currently a research topic XD
Amazing! Thank you
Would love you to cover Physics-informed Deep-O-Nets as well! Thanks a ton for the great material :D
ok I was not at @44:30 when I made the comment don't mind me
Great content! 😊
Thank you!
Good timing! 1 day after the "release" of Sora and V-Jepa
Is there a pointer to a description of the studio environment used to create this vid?
Very professional and well-done! Sure beats a scratchy chalk board, slide projection in the background, etc!
Great video, professor
I'm wonder whether AI has reached the complexity of the human brain yet. Although the human brain has well established speciality areas, so we like in hope. Although, memGPT is a huge breakthrough ! Great video once again.
Fantastic!
I'm so excited..
Problem, reaction, solution (optimized predictions or syntropy) -- the Hegelian dialectic.
Inputs are dual to outputs.
"Always two there are" -- Yoda.
keep it up ..big love
Steve - I can't overstate how much i have been enjoying your online courses. Will these PINNs courses include some example code?
I have an interview on physics-informed ML tomorrow, and I just stumbled upon this! Thank you!
good luck!
@moienr4104 ... so how did it go
Hey I wanted to know if it is a field with future scope and demand, and also what kind of qualifications are required for such jobs? Would you like to connect?
24:51 that's the coolest example i've seen so far 🤣😂🥰
This is so interesting, I’m excited for this series. Where is the pdf of your book?
Steve the GOAT Brunton back at it again god bless
I love this. ❤
Waiting for more...
Thanks
I think it would be pertinent to connect this work to Judea Pearl's work on Directed Acyclic Graphs.
The intention of this work will often live at the intervention or counterfactual steps in th3 ladder of causality.
It would be important to acknowledge it.
If only from a legal perspective, where, in a suit, these matters are criticcal.
Awesome
Thank youuuuuu
It seems to me that separating the symmetry from the neural network would be far more reliable. Simply including many orientations in the training is the lazy approach. Instead, concentrate on one side (e.g. the left side or the right side) and concentrate on g pointing down while training the network. Then precede the network with a symmetry varying algorithm that rotates the input by 5-10 degrees while watching the correlated output. If the subject has bilateral symmetry, then repeat the process after exchanging x-x. Then consider only the best output(s) when deciding how to classify the image.
thankyou
The book wasn't free and no further links to other resources. But hey, still a good video.
The link isn't in the description, but he puts it on screen. It is: databookuw.com/databook.pdf
At the time of my writing of this comment the link is working
Thanks a lot for such a great overview of this exciting field! I've just got a paper accepted on TMLR about this very same topic: Effective Latent Differential Equation Models via Attention and Multiple Shooting.
I think that many people here might find it interesting: ua-cam.com/video/XYv10fuumCQ/v-deo.html
I look forward to the rest of the lectures of this series! :)
Hello there, I am also a physicist. Who uses ML and AI in the field of thermodynamics.😊
With chemistry, there are principles and rules but there are also a lot of grey zones, exceptions, irregularities and anomalies lying in chemistry that are ready to pounce at you