I don't think these are useful at all. If you implement a simple network that approximates 1D functions like sin or learn how image blurring works with kernels and then move into ML/AI that gave me a much better understanding.
So... I remember math including doing quite a bit of geometry by hand and with real tools, at least initially. "Math" is not just the symbol stuff written with a pencil, or with a keyboard.
Idk, it's fun. 20 years ago I made a cubic neural model in Flash that actually lit up cubes depending on how much they were being accessed. This was a case of binding logic way too tightly to display code, but it was a cool experiment.
They're likely of limited use for someone looking for introductory material to ML, but for someone having done some computer vision and used various types convolution layers, it can be useful to see a summary with visualizations.
Thank you for saying this. I often find this "glib" explains of ML stuff very frustrating as a human coming from an Applied Math background. It just makes me feel a bit crazy and alone to see what appears to be a certain kind of person saying "gosh" at various "explanations" when I just don't get it.
Obviously this is beautiful as art but it would also be useful to understand how exactly these visualizations are useful to people who think they are. Useful to me means you gain a new ability to extrapolate in task space (aka "understanding").
It says the input has 3 dimensions, two spatial dimensions and one feature dimension. So it would be a 2D grid of numbers. Like a grayscale photo. But at 00:38 it shows the numbers and it looks like each of the blocks positioned in 3D space holds a floating-point value. Which would make it a 4-dimensional input.
This is a fantastic educational resource! Visual animations like these make understanding complex ML concepts so much more intuitive than just reading equations.
The neural network visualization is particularly well done - seeing the forward and backward passes in action helps build the right mental model. Would be great to see more visualizations covering transformer architectures and attention mechanisms, which are often harder to grasp.
For anyone building educational tools or internal documentation for ML teams, this approach of animated explanations is really effective for knowledge transfer.
Nice! I made my own version of this many years ago, with a very basic manim animation
https://www.jerpint.io/blog/2021-03-18-cnn-cheatsheet/
Years back I worked on some animated ML articles, my favorites being: https://mlu-explain.github.io/neural-networks/ and https://mlu-explain.github.io/decision-tree/
not deep learning but this oldie is a goodie too (since we are sharing favorites): https://narrative-flow.github.io/exploratory-study-2/
I originally had it saved as [[ https://www.r2d3.us/visual-intro-to-machine-learning-part-1/ ]] but it seems that link is gone?
I don't think these are useful at all. If you implement a simple network that approximates 1D functions like sin or learn how image blurring works with kernels and then move into ML/AI that gave me a much better understanding.
Yes, especially if you ask someone why one is better than the other in a certain configuration.
Yup, I'd say you learn more by doing math by hand (shouldn't be that surprising).
So... I remember math including doing quite a bit of geometry by hand and with real tools, at least initially. "Math" is not just the symbol stuff written with a pencil, or with a keyboard.
The mechanical analog computers of old (e.g. https://youtu.be/IgF3OX8nT0w, or https://youtu.be/s1i-dnAH9Y4) are examples too that math is more than symbol manipulation.
Idk, it's fun. 20 years ago I made a cubic neural model in Flash that actually lit up cubes depending on how much they were being accessed. This was a case of binding logic way too tightly to display code, but it was a cool experiment.
They're likely of limited use for someone looking for introductory material to ML, but for someone having done some computer vision and used various types convolution layers, it can be useful to see a summary with visualizations.
Thank you for saying this. I often find this "glib" explains of ML stuff very frustrating as a human coming from an Applied Math background. It just makes me feel a bit crazy and alone to see what appears to be a certain kind of person saying "gosh" at various "explanations" when I just don't get it.
Obviously this is beautiful as art but it would also be useful to understand how exactly these visualizations are useful to people who think they are. Useful to me means you gain a new ability to extrapolate in task space (aka "understanding").
Learning first principles of something are always useful for beginners.
Everyone is a beginner at something.
I worked on something similar but specifically for transformer architecture: https://transformer.sujayk.me/
On Safari mobile it shows a modal that can’t be scrolled nor closed
Yeah, it's not mobile-friendly. didn't get a chance to look into it
Is there an error in the first video at 00:25?
https://www.youtube.com/watch?v=eMXuk97NeSI&t=25
It says the input has 3 dimensions, two spatial dimensions and one feature dimension. So it would be a 2D grid of numbers. Like a grayscale photo. But at 00:38 it shows the numbers and it looks like each of the blocks positioned in 3D space holds a floating-point value. Which would make it a 4-dimensional input.
Nice work. A while back, I learned convolutions using similar animations by Vincent Dumoulin and Francesco Visin's gifs
https://github.com/vdumoulin/conv_arithmetic
I feel like these are helpful, and I think the calculus oriented visualizations of convex surfaces and gradient descent help a lot as well.
here is the github link for anyone wanting to star the repo https://github.com/animatedai/animatedai
Shameless plug for my writeup about convolutions: https://jlebar.com/2023/9/11/convolutions.html
This is a fantastic educational resource! Visual animations like these make understanding complex ML concepts so much more intuitive than just reading equations.
The neural network visualization is particularly well done - seeing the forward and backward passes in action helps build the right mental model. Would be great to see more visualizations covering transformer architectures and attention mechanisms, which are often harder to grasp.
For anyone building educational tools or internal documentation for ML teams, this approach of animated explanations is really effective for knowledge transfer.
also look at https://poloclub.github.io/transformer-explainer/
You should add dilated conv and conv_transpose to the list.
amazing resource!
[dead]