AI/ML+Physics Part 1: Choosing what to model [Physics Informed Machine Learning]
Вставка
- Опубліковано 17 тра 2024
- This video discusses the first stage of the machine learning process: (1) formulating a problem to model. There are lots of opportunities to incorporate physics into this process, and learn new physics by applying ML to the right problem.
This video was produced at the University of Washington, and we acknowledge funding support from the Boeing Company
%%% CHAPTERS %%%
00:00 Intro
04:51 Deciding on the Problem
07:08 Why do you need an ML Model?
14:54 Case Study: Super Resolution
17:07 Case Study: Discovering New Physics
18:37 Case Study: Materials Discovery
19:12 Case Study: Computational Chemistry
20:50 Case Study: Digital Twins & Discrepancy Models
21:56 Case Study: Shape Optimization
25:13 The Digital Twin
29:16 Modeling the Math
33:31 Modeling the Chaos
34:18 Case Study: Climate Modeling
35:08 Benchmark Systems
35:47 Case Study: Turbulence Closure Modeling
39:16 When not to use Machine Learning
42:15 Outro - Наука та технологія
![AI/ML+Physics Part 2: Curating Training Data [Physics Informed Machine Learning]](http://i.ytimg.com/vi/g-S0m2zcKUg/mqdefault.jpg)
![AI/ML+Physics Part 2: Curating Training Data [Physics Informed Machine Learning]](/img/tr.png)
Thanks for posting these lessons. There isn’t enough good material about this out there.
Hi professor brunton. thank you for this lectures. i am really enjoying your videos. can't wait for the next one.
I already love this series! I honestly think that the choice of the problem to model is BY FAR the most important one. You can bake so much prior knowledge into that alone, it can totally make or break the entire endevour.
This course is one of the best learning tool on the internet. Thank you Mr Brunton
Thanks for sharing your knowledge with us all! I feel fortunate to be able to access this level of learning for free
This is excellent - cant wait to see the whole series
Very interesting, can't wait to see where you take this!
Thank you for the lectures, learned/got inspired a lot.
42:05 "... you don't want to be in the crystal energy group..."
Ah, those pesky condensed matter physicists!
Quality content is an understatement Waiting for more
Amazing! Thank you so much for this set of lectures!
Thank you for doing these excellent lectures Dr Brunton
Beautiful, just beautiful...Thank you
Thank you for this excellent lecture. Learned a lot.
Really amazed, Thank you Prof. Brunton.
Awesome lecture. God bless you for sharing this knowledge on youtube.
Super stoked to see our car in the presentation 😊
Great tutorial 😊 thanks so much
Excellent lecture! Many thanks professor!!!
Would have really appreciated some concrete examples and case studies. With concrete math and code.
I loved watching many of your videos from the databook series, because they were so unique- using math and code. And you are, always, a superb teacher, explainer.
Thank you for making this series. There's really a lack of good content in this area. I really am grateful, and appreciate you doing this.
Will wait for future videos. 😇😊
You are a legend, professor.
Thanks!
I really appreciated you for your efforts.
YES! THANK YOU BOSS!
This is amazing!
Im excited, are there groups/communities for the general public to join for this topic ?
Where can I find the text book ?
And thnx your explanation is what help me to really understand what I missed
@stevebrunton can you share a git repo with a basic project with problem description , setup env , ML model
There is a reference to "Rolles-Royce + Cambridge", is this a typo of "Rolls-Royce" in collaboration with University of Cambridge (UK)?
Thanks for deciding to do this series, I've recently discovered some of your other playlists and they're excellent.
I really can't thank you enough.
Thx for core steps
Thank you for posting this knowledge. I've been watching almost exclusively your content over the last year in 2023. I found super interesting the case studies you shared about the super alloy at Rolls-Royce and the predictive shimming at Boeing. It would be amazing if we could see more case studies like that. I am trying to wrap my head around on how to approach a ML model that will predict perceived color of different materials taking as input data about various processes of production for the respective materials.
So what would be the 'hello world!' tutorial/dataset of Physics Informed machine learning?
Some general 'hello world!'s in machine learning are MNIST(handwriting Digits identification), Iris Flowers Classification, or Cancer , Ham v Spam (email), etc.
The first two are notable as to how relatable they are that one could imagine making the dataset themselves, though really with a lower sample size due to the effort involved vs a real dataset.
One suggestion: on shape engineering, MIT made the toroidal propeller. Maybe do a case study on that? Like walk us through the process
Hi I student want to use artificial intelligence in aerospace aerodynamic can you show me the step by step how to start and wich book should read (the point start and to the end point)???
If you explain in the a clip is great
I never really undertood the need for the term multiphysics. There are certainly different length and time scales in complex phenomena like cloud formation, but those processes are, as far as I am aware of, governed by physics (not multiphysics). Do we also apply the same idea to math, when we refer to different fields of mathematics when solving a problem? Something like multimathematics?
I get the impression that it basically means multi-scale-physics.
If people find the term “multiphysics” more convenient than “multi-scale-physics” (or “a combination of physical models that model physics of things at different scales”), I don’t have a problem with that.
Multiphysics is just to highlight the need for multiple domain knowledge under the umbrella of "physics". For example, you can call transport phenomena and electromagnetic field theory are just "physics", as opposed to chemistry, biology, right?
But you can also say they are different physics --- physical mechanisms.
What happened to the 3rd lecture on architectures?
Computer just give you a answer because we just program it to answer. We still don't know enough about our brain process to make the computer stimulate our brain curious process, and the way we control that curious not gone wrong, we still don't research enough to make physical part(hardware) to stimulate structure to run that function
The second video is missing, where can I find it?
Ok is the video tuned with ML to my research?? I’m literally working on discovering new physics for plasmas in spherical tokamaks! Spooky…
Saludos desde Colombia.
I see Steve+AI, I click, I like
Repeated the part 0 from last week all over again without any additional values and depth.
You never put the links you mention.
So as for physics EMBEDDED machine learning (I am sticking with 'embedded' as opposed to 'informed' because it's closer to the design and 'informed' sounds a bit old-academic and is less 'tactile' to visualization and interpretation - which is very important). But it could be 'data science' or 'cryptographic' embedded machine learning right? And that's what we could be seeing demonstrated by algorithms like Q-star (differentiating between encrypted and plain text data-sets to crack encryption standards). I believe Sora is using Unreal Engine 5 for its training data (synthetic), and the power in the physics is evident when you have potentially infinite choice and combination of physical scenario, as syntetic data allows....accelerating numerical computations by taking a simulation at low-res then scaling up in resolution by way of augmented machine learning is simply MASSIVE! - just in-terms of the sheer affect on research and industry, chip design and manufacture (I would have kept silent with regards the cov. vaccine incidentally, and we won't have the long term vaccine-injury data on that for about another 20 years or so...that's a HOWLER I'm afraid 😞
There is a step zero. 0. Watch and thoroughly absorb this video.
The teacher (Yoda) is dual to the pupil (Luke Skywalker) -- The Hegelian dialectic.
Master (Lordship, client) is dual to the slave (bondsman, server) -- The Hegelian dialectic.
Problem, reaction, solution -- The Hegelian dialectic.
"Always two there are" -- Yoda.
You would think all ML models are "physics-informed" to function correctly...heck, even just to work (to be able to run).
where is the math?
why is the math?
25:00
I like the notions he has on astrology and ai powered products advertisement. 😂 . Please don't take a week to upload chapters. Upload all at once.
Too much blah blah. Would be more useful if we'd actually start solving problems with code and math. All this talk just comes in one eat and goes out the other without practice.
No, it is going in one ear and out of the other because you’re not taking notes like a good student who knows how to learn something from a lecture.
It’s also part of a series; here he’s covering a first stage of problem solving that comes before coding.
I suggest you resist the immature impulse to code before having done any intellectual work.
Can you elaborate on "just start solving problems with code and maths"
Coding without understanding is just wasting your time. If you can't understand what he is saying then I'd try another subject. He is a really very good teacher
If only he had made a video explaining the importance of understanding your problem before jumping into the math and coding...
This is not blah blah...this is the motivation to start the topic.