Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

The universal weight subspace hypothesis

arxiv.org

Post reply on HN
Print view View original

131–140 of 146 posts

Re: The universal weight subspace hypothesis

#131

Post by Atlas667 »

tim333 wrote: post #119
Atlas667 wrote: post #79

Imagine collectively trying to recreate a human brain with semiconductors so capitalists can save money by not having to employ as many people

There are other reasons beyond the employment thing. Understanding how the mind works maybe.

Understand the mind to then exploit it.

Why else would they put so much money into something if not to try and get more out of it?

Capitalists' morals are driven by their social position. To them this is right becauae its rewarding. To us its an akin abomination we create that destroys us

But the problem isnt inherently tech. Its how society is structured around it that allows it to be used against us.

Re: The universal weight subspace hypothesis

#134

Post by TheOtherHobbes »

energy123 wrote: post #59
mlpro wrote: post #36

Earlier quoted context omitted.

Not really. If the models are trained on different dataset - like one ViT trained on satellite images and another on medical X-rays - one would expect their parameters, which were randomly initialized to be completely different or even orthogonal.

Every vision task needs edge/contrast/color detectors and these should be mostly the same across ViTs, needing only a rotation and scaling in the subspace. Likewise with language tasks and encoding the basic rules of language which are the same regardless of application. So it is no surprise to see intra-modality shared variation. The surprising thing is inter-modality shared variation. I wouldn't have bet against it…

It hints there may be common higher-level abstraction and compression processes in human consciousness.

The "human" part of that matters. This is all human-made data, collected from human technology, which was created to assist human thinking and experience.

So I wonder if this isn't so much about universals or Platonic ideals. More that we're starting to see the outlines of the shapes that define - perhaps constrict - our own minds.

Re: The universal weight subspace hypothesis

#135

Post by kroaton »

EvanAnderson wrote: post #125
api wrote: post #117

Earlier quoted context omitted.

Find some images of the entire huge scale structure of the universe. It looks a bit like… a brain. What does this mean? Probably not nothing, but probably not “the cosmos is the mind of god.” It probably means that we live in a universe that tends to produce repeating nested patterns at different scales. But maybe that’s part of what makes it possible to evolve or engineer brains that can understand it. If it had no…

Similar feeling here re: "mind of God". I interpret these patterns as a very simple property of mathematics producing complex-looking patterns and evolution exploiting that complexity. Evolution is the ultimate procedural content generation machine.

I thought that was Houdini.

Re: The universal weight subspace hypothesis

#136

Post by farhanhubble »

mapontosevenths wrote: post #122
markisus wrote: post #115

Earlier quoted context omitted.

Each fine tune drags the model weights away from the base model in a certain direction. Given 500 fine tune datasets, we could expect the 500 drag directions to span a 500 dimensional space. After all, 500 random vectors in a high dimensional space are likely to be mutually orthogonal. The paper shows, however, that the 500 drag directions live in a ~40 dimensional subspace. Another way to say it is that you can comp…

Agreed. What's surprising here to me isn't that the fine tunes are compressible, it's the degree to which they're compressible. It seems like very little useful new information is being added by the fine-tune. They're using SVD to throw away almost all of the "new information" and apparently getting solid results anyhow. Which of course raises interesting questions if replicable. The code doesn't seem to have been re…

Yeah but it also made me think if deep down neural networks are curated random basis vectors, like in random projections.

Re: The universal weight subspace hypothesis

#137

Post by tim333 »

Atlas667 wrote: post #131
tim333 wrote: post #119

Earlier quoted context omitted.

There are other reasons beyond the employment thing. Understanding how the mind works maybe.

Understand the mind to then exploit it. Why else would they put so much money into something if not to try and get more out of it? Capitalists' morals are driven by their social position. To them this is right becauae its rewarding. To us its an akin abomination we create that destroys us But the problem isnt inherently tech. Its how society is structured around it that allows it to be used against us.

A lot of the basic research including this paper is out of academia rather than business.

Re: The universal weight subspace hypothesis

#138

Post by yorwba »

moi2388 wrote: post #108
woopsn wrote: post #76

Earlier quoted context omitted.

It's a little arbitrary. Look at the graph on page 6, there's no steep gap in the spectrum there. 16 just about the balance point

But there is a steep gap in the spectrum at 16 on page 7

That's the spectrum of LoRAs, which are LoW RAnk by design.

Re: The universal weight subspace hypothesis

#140

Post by imtringued »

farhanhubble wrote: post #16

Would you see a lower rank subspace if the learned weights were just random vectors?

This is a good point, but I think this only works for D*A, where D=Sigma is a diagonal matrix with learnable parameters. It probably doesn't work for a full singular value decomposition (SVD) UDV^T.

Basically, what if we're not actually "training" the model, but rather the model was randomly initialized and the learning algorithm is just selecting the vectors that happen to point into the right direction? A left multiplication of the form D*A with a diagonal matrix is equivalent to multiplying each row in A with the corresponding diagonal element. Low values mean the vector in question was a lottery blank and unnecessary. High values means that this turns out to be correct vector, yay!

But this trivial explanation doesn't work for the full SVD, because you now have a right multiplication U*D. This means each column gets multiplied against the corresponding diagonal element. Both the column in U and row vector in V^T have to perfectly coincide to make the "selection" theory work, which is unlikely to be true for small models, which happen to work just fine.*

Post reply on HN