[bsenftner]

Graphics programming has this one very, very useful aspect, exponentially more valuable today: the matrix algebra pipelines, and then the requirement to 'think in matrix transforms' is a wonderful and visually engaging way to get your foundation for machine learning math.

[eichin]

I don't really see this with modern graphics programming, but I was highly amused that my 1980s-1990s graphics skills (in particular, coordinate transform math) were very useful when I started working in robotics in the 2010s-2020s (because forward and inverse kinematics are exactly the same thing as 2d/3d projections.)

The trick there is that they both have related physical analogs, and machine learning math doesn't really (in that while you can visualize them spatially, it doesn't seem to help solve any problems in that space.)

[mathisfun123]

This is like saying being a cashier prepares you for a job in high-finance because both involve arithmetic on dollars and cents.

I've been in ML for ~5 years in multiple FAANGs and I have never seen a rotation matrix.

[bsenftner]

A rotation matrix is but one of dozens and dozens of different types of basic transforms. It gets really fun with jacobian 12x12 matrix operations, and free form deformations. Which maps to ML far better than most imagine.

[jplusequalt]

TBF, I bet any graphics programmer would be a boon for a ML shop for their GPU/performance optimization knowledge alone.

[bsenftner]

One of the largest, uncomprehendingly huge, omissions in AI video models is the lack of anyone from film, film production, animation, 3D graphics, or VFX on their video AI teams. Why would that be needed? Well, so the freaking models understand when anyone from any of these backgrounds prompts them with their industry standard language and terminology to describe a set, a camera move, an type of environmental lights, or any of the thousands of production terms used globally for the creation of media. The latest models are just now starting to understand basic terms, but describing a sequence with a motion camera moving for a narrative reason with action and dialogue in front is like describing calculus to a toddler. That's not to mention how much optimization knowledge anyone from a digital production developer background brings to the table...

[KellyCriterion]

AFAIR, the guys at DeepSeek with their PTX-assembly approach didnt have any graphics programming experience

[cognoboffin]

Doesn’t RoPE use 2D rotation matrices ?

[mathisfun123]

congrats you've found literally the only example ("the exception that proves the rule").

[bsenftner]

Your fixation on the rotation matrix == 3D graphics is not right

[cognoboffin]

SVD, and PCA are also examples.

[mathisfun123]

there is absolutely no sense in which the SVD/PCA decomposition is just a rotation matrix. you should probably review your linear algebra textbook (hint: scaling is extremely important).

[moregrist]

PCA is an orthogonal transformation of the covariance matrix, so like all orthogonal transformations, it’s _literally a rotation_ in N-dimensional space.

SVD is more complex but ultimately it’s just another useful decomposition of a matrix.

I’m not sure why you’re both negative and dismissive. Transformation matrices in graphics are a good and approachable way to get used to linear transformations, which turn out to be useful pretty much everywhere.

Whether or not that helps you with ML depends more on what you’re doing in ML. FAANG doesn’t have a monopoly on ML or on interesting work in ML.

[srean]

Cognoboffin is exactly right. SVD decomposed matrix into a sequence of rotation, scaling and unrotation matrices.

If anyone needs a review it's not cognoboffin.

You led with the claim you have never seen a rotation matrix in ML. I am having doubts about whether you have the ability to recognise one.

I suspect new hires get a free pass as long as they can talk a storm about backpropagation these days.

[cognoboffin]

SVD is the decomposition of a matrix into two rotation matrices and a scaling matrix, by definition:

https://en.wikipedia.org/wiki/Singular_value_decomposition

[srean]

... and I have been both situations for longer and have seen tons and tons of them (*)... So?

Not so hypotheticals -- Heck the inputs that you want labelled could be rotation matrices. The desired output could be a rotation matrix. Generating more convenient features could be via a rotation matrix. Dimensionality reduction could be through a reduction matrix. Sparsity could be encouraged by proper use of rotation matrices. Shows up if you want to build in group theoretic invariance in your predictive model.

(*) If you consider Householders then even more

[kilpikaarna]

I mainly learned linear algebra via hands-on 3D graphics, and have a hard time thinking about a matrix as anything other than 4x4 and representing a linear transform...

How much do you even think about explicit matrix math when doing high-level ML?

[bsenftner]

3D graphics is so much more than the basic transforms. Add in all the deformation systems blending together, and those often being physics driven off the animation. You all have seen modern VFX, right? That is not basic 4x4 transforms.

[skydhash]

I’ve not done high level ML, but I’ve done introductory ML and the truth is while the input space and the output space can have N and M dimensions, there’s not a lot of constraints involved. The matrix there are more randoms.

The whole ML field is basically about starting from random points and trying to find useful shapes and constraints. Basically like trying to get object likeness in clouds

[rustystump]

5y multiple fang? Dont you know it is MANGOs now and has been for a while. Are you sure it is FAANGs?

[pascahousut]

Is the linear algebra of machine learning more complicated than that of graphics?

[dahart]

Both are large fields with many varied applications of linear algebra (and non-linear math too), and many people trying a lot of interesting & complicated ideas. The question is way too vague to answer meaningfully, it depends on what you mean by ‘graphics’ and by ‘machine learning’ and by ‘linear algebra’ and by ‘more complicated’. ;)

The linear algebra used in the basic raster pipeline to manage drawing a 3d unshaded mesh is pretty simple, and you can get by knowing just a little bit of linear algebra, like dot products and how to multiply matrices, and maybe what homogeneous coordinates are. But that is by no means the extent of linear algebra in all of graphics.

The linear algebra used in a basic neural network is also pretty simple, and you can get by knowing dot products and matrix multiply if you’re writing your own inference, and maybe just a tiny bit of derivative calculus if you’re writing your own backprop, but otherwise you don’t need anything else.

Students in both fields have to learn some basic linear algebra, but most people working in ML & graphics generally don’t use any linear algebra day to day, because most people aren’t writing inference/backprop and most people aren’t writing the graphics pipeline.

BTW, matrices and linear algebra are pure convenience for neural networks, and maybe for the graphics raster pipeline too. You can do both of these things without using matrices per se (though you might re-invent something equivalent and/or less efficient by avoiding matrices).

[ekholm_e]

I'm a data scientist and not a graphics programmer, but my guess is that it's just abstracted away more. If you're using ML/DL libraries, you're mostly just calling APIs that handle the linear algebra and calculus for you. Unless you're actively contributing to those libraries, you likely don't ever need to "touch" any of the underlying operations. Up to a point, it's useful to understand how things work under the hood, but where that point is kinda depends on your job. For instance, I could write code to do 'naive' matrix multiplication, but I couldn't, like, contribute to BLAS.

[molybd3num]

i think so