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

121–130 of 146 posts

Re: The universal weight subspace hypothesis

#121

Post by markisus »

bigbuppo wrote: post #29
kacesensitive wrote: post #4

interesting.. this could make training much faster if there’s a universal low dimensional space that models naturally converge into, since you could initialize or constrain training inside that space instead of spending massive compute rediscovering it from scratch every time

Wouldn't this also mean that there's an inherent limit to that sort of model?

On the contrary, I think it demonstrates an inherent limit to the kind of tasks / datasets that human beings care about.

It's known that large neural networks can even memorize random data. The number of random datasets is unfathomably large, and the weight space of neural networks trained on random data would probably not live in a low dimensional subspace.

It's only the interesting-to-human datasets, as far as I know, that drive the neural network weights to a low dimensional subspace.

Re: The universal weight subspace hypothesis

#122

Post by mapontosevenths »

markisus wrote: post #115
modeless wrote: post #58

This seems confusingly phrased. When they say things like "500 Vision Transformers", what they mean is 500 finetunes of the same base model, downloaded from the huggingface accounts of anonymous randos. These spaces are only "universal" to a single pretrained base model AFAICT. Is it really that surprising that finetunes would be extremely similar to each other? Especially LoRAs? I visited one of the models they refe…

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 released yet though.

Re: The universal weight subspace hypothesis

#123

Post by IAmBroom »

zkmon wrote: post #114

So, while the standard models are like herbivores grazing on the internet data, they built a model that is a carnivore or a predator species trained on other models? Sounds like an evolution of the species.

If I can understand your metaphor, it's probably not sophisticated enough to be relevant.

- I know what I do not know.

-- I do not know AI.

Re: The universal weight subspace hypothesis

#124

Post by IAmBroom »

pagekicker wrote: post #43
100721 wrote: post #21

Earlier quoted context omitted.

Acting as a pass-through for LLMs is logically equivalent to wiring up a bot account.

No, it's not, unless you can argue that the bot would have thought of asking the same question I did, which is unlikely.

"I asked [AI] and it said..." is not the path to social acceptance in this herd.

Re: The universal weight subspace hypothesis

#125

Post by EvanAnderson »

api wrote: post #117
EvanAnderson wrote: post #13

Earlier quoted context omitted.

On a hike this weekend my daughter and I talked about the similarities of the branching and bifurcating patterns in the melting ice on a pond, the branches of trees, still photos of lightning, the circulatory system, and the filaments in fractals.

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.

Re: The universal weight subspace hypothesis

#126

Post by sigbottle »

augment_me wrote: post #95

I think the paper in general completely oversells the idea of "universality". For CNNs, the 'Universal Subspace' is simply the strong inductive bias (locality) forcing filters into standard signal processing shapes (Laplacian/Gabor) regardless of the data. Since CNNs are just a constrained subset of operations, this convergence is not that surprising. For Transformers, which lack these local constraints, the authors…

For me at least, I wasn't even under the impression that this was a possible research angle to begin with. Crazy stuff that people are trying, and very cool too!

Re: The universal weight subspace hypothesis

#127

Post by CGMthrowaway »

mapontosevenths wrote: post #67
CGMthrowaway wrote: post #2

They compressed the compression? Or identified an embedding that can "bootstrap" training with a headstart ? Not a technical person just trying to put it in other words.

To use an analogy: Imagine a spreadsheet with 500 smoothie recipes one in each row, each with a dozen ingredients as the columns. Now imagine you discover that all 500 are really just the same 11 base ingredients plus something extra. What they've done here is use SVD, (which is normally used for image compression and noise reduction), to find that "base recipe". Now we can reproduce those other recipes by only recor…

Yeah that's pretty much how I understood it. Good analogy. We are finding the French Mother Sauce. Reading the comments it seems everyone is still clear on the practical implications of that.

Re: The universal weight subspace hypothesis

#128

Post by mlpro »

augment_me wrote: post #93
daemonologist wrote: post #86

Earlier quoted context omitted.

I agree - the results on the finetunes are not very surprising. The trained-from-scratch ResNets (Figure 2 and Section 3.2.1) are definitely more interesting, though somewhat limited in scope. In any case, my impression is that this is not immediately more useful than a LoRA (and is probably not intended to be), but is maybe an avenue for further research.

I don't think its that surprising actually. And I think the paper in general completely oversells the idea. The ResNet results hold from scratch because strict local constraints (e.g., 3x3 convolutions) force the emergence of fundamental signal-processing features (Gabor/Laplacian filters) regardless of the dataset. The architecture itself enforces the subspace. The Transformer/ViT results rely on fine-tunes because…

I think its very surprising, although I would like the paper to show more experiments (they already have a lot, i know).

The ViT models are never really trained from scratch - they are always finetuned as they require large amounts of data to converge nicely. The pretraining just provides a nice initialization. Why would one expect two ViT's finetuned on two different things - image and text classification end up in the same subspace as they show? I think this is groundbreaking.

I don't really agree with the drift far from the parent model idea. I think they drift pretty far in terms of their norms. Even the small LoRA adapters drift pretty far from the base model.

Re: The universal weight subspace hypothesis

#129

Post by mlpro »

modeless wrote: post #83
mlpro wrote: post #77

Earlier quoted context omitted.

Why would they be similar if they are trained on very different data? Also, trained from scratch models are also analyzed, imo.

They are trained on exactly the same data in the same order with the same optimizer because they are literally the same base model. With a little fine tuning added on top. I see now that they did one experiment with trained from scratch models. They trained five Resnet-50s on five disjoint datasets of natural images, most quite small. And IIUC they were able to, without further training, combine them into one "univer…

They are not trained on the same data. Even a skim of the paper shows very disjoint data.

The LLMs are finetuned on very disjoint data. I checked some are on Chinese and other are for Math. The pretrained model provides a good initialization. I'm convinced.

Re: The universal weight subspace hypothesis

#130

Post by tech_ken »

modeless wrote: post #58

This seems confusingly phrased. When they say things like "500 Vision Transformers", what they mean is 500 finetunes of the same base model, downloaded from the huggingface accounts of anonymous randos. These spaces are only "universal" to a single pretrained base model AFAICT. Is it really that surprising that finetunes would be extremely similar to each other? Especially LoRAs? I visited one of the models they refe…

This is an important clarification; from the abstract and title I was super confused how they identified a "subspace" that could be consistently identified across model structures (I was assuming they meant that they saw stability in the dimension of the weight subspace or something), but if they're just referring to one model class that clears things up substantially. It's definitely also a much weaker result IMO, basically just confirming that the model's loss function has a well-posed minima, which...duh? I mean I guess I'm glad someone checked that, but called it "the universal weight subspace hypothesis" seems a bit dramatic.
Post reply on HN