Can an AI Learn Physics from Data? (Parameter Identification)

[@monsterhunter8595]

I love your content! You talk about subjects that aren't much talked about, love it!

[@JoelRosenfeld]

I’m glad you like it!

[@BrunoJedynak]

Thank you. happy new year Joel!

[@JoelRosenfeld]

Happy New Year, Bruno!

[@DistortedV12]

Lastly, would love to see it in action

[@JoelRosenfeld]

I have an older playlist called Data Driven Methods in Dynamical Systems where I’ve done a lot of this from different perspectives

[@commonwombat-h6r]

thank you very much! Your videos are gold

[@numoru]

Sorry but yeah man Isaac Amidror is awsome. Would love to see some of the methods he goes over implementation in neutral layers. Note some of his minimization methods such as halftones (use in newspaper prints usually). Then correlate that to a case like Sphaerophoria who was having trouble with a image to ASCII AI that worked better when the initial image was darker kzbin.info/www/bejne/o4WxgKl6ZsR9pdksi=Es0QTn56OfybuZDH which I suppose could have been due to " Digital halftoning based on some modern image processing tools such as nonlinear diffusion and stochastic flipping” preventing over fitting due to what im calling basis matching of the ai's layers with prominently repetitive features which to it seem like a moire pattern from the ai's 'perspective' (the grids lined up to well like a camera and strip shirt. Further, what is going on is the human interaction doesn't perceive it this way but it can still be perceived that way by other 'entities' (AI). This correlates to Multistable perception (or bistable perception) is a perceptual phenomenon in which an observer experiences an unpredictable sequence of spontaneous subjective changes. While usually associated with visual perception (a form of optical illusion), multistable perception can also be experienced with auditory and olfactory percepts. Or even if you e seen the philosophical question posed rather every one sees the same color say "green" as you do. In the AI's cases I fundamentally believe it will always be a heck-no it doesn't perceive the same way, and it won't until it's tapping into our biological sensory input IN THEE EXACT SAME WAY our minds are connected to the rest of our body. At which point we begin to have A Pauli Exclusion Principle Perception and cognition between bio and wires, lol weird. But further what if we tapped in using some type of entangle ment to softly sense the sensory input similar to the theorems that would allow us to test a sensitive bomb without it going off. (Note it probably won't work with our brains but smaller bio produced organisms I could definitely see someone reaching for a can of nanobot infused organelles with a pin hole in it to project an image onto the organelles eye in order for the AI to better process images and give more realistic feed back. Weird. I should write scifi, I would read that shit. Oh God flesh 😢

[@numoru]

Aye bro 12:11 a lower left or upper right upper bound be T_1 baby not T_N. Correct me if Im wrong been looking at this video all day bro. Dope stuff, felt like feynman tricks but applied numerically. And that pseudo-inverse is clutch you said Penrose right? Yo you should check out "The Theory of the Moiré Phenomenon Volume II: Aperiodic Layers” by Isaac Amidror of EPLF. I truly believe this is what neural networks are doing under the hood and what generates the emergent behavior. We can learn much from it by drawing analogies to our work. I hope to start a company that utilizes many of these mechanisms and apply to analog computing, but im just a poor black kid staying with his parents right now with a dumb physics bs and no money to go back for research or to find a decent mentor. Keep up the good work and inspiration. Love that you brought the kids into it, I try but my baby mother acts like its condescending when I teach my son advanced concepts or allow him to experiment. (Psychology major). Well best❤❤

[@DistortedV12]

7:19 could stochastic calculus be used as well by treating it as an Itô process?

[@JoelRosenfeld]

Yes! Absolutely

[@soyoltoi]

I think you need more pauses or scene changes. The music changes but nothing new is happening on screen

[@JoelRosenfeld]

Ok I’ll work on it

[@rammsund]

I see that you've linked the Data Driven Methods in Science and Engineering by prof. Brunton and prof. Kutz, but the legends as they are, they have also published the book for free. Link to download the pdf in Prof Brunton's video kzbin.info/www/bejne/pXa7g2x7o6ano5Y

[@DistortedV12]

I don't understand the reason for integrating the signal and then the time derivative instead of operating on the raw features/trajectories?

[@JoelRosenfeld]

Can you give me the time stamp?

[@DistortedV12]

@@JoelRosenfeld I think it starts here 4:37, and then you say we can take a derivative as a sample of that function we are trying to estimate as a naive approach, then later on you do numerical differentiation instead and was wondering why one needed to take integral/derivative and can't learn the function from the raw data.

[@dariosilva85]

Was this about AI?

[@JoelRosenfeld]

Yeah, this is more or less how machine learning and AI systems are trained. You have a loss function and then you optimize for the correct weights. Neural networks are a collection of basis function for which you need to select weights based on the data and loss functions. For instance, this could be seen as a shallow neural network, and a deeper neural network would have more parameters but also integrating in back propagation (which is essentially the chain rule). AI encompasses a broad range of approximation techniques.

[@dariosilva85]

@@JoelRosenfeld Thank you.