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

Diffusion Beats Autoregressive in Data-Constrained Settings

blog.ml.cmu.edu

Post reply on HN
Print view View original

11–16 of 16 posts

Re: Diffusion Beats Autoregressive in Data-Constrained Settings

#11

Post by astrange »

kevinwang wrote: post #10
thesz wrote: post #8

> This paper addresses the challenge by asking: how can we trade off more compute for less data? Autoregressive models are not matched by compute and this is the major drawback. There is evidence that training RNN models that compute several steps with same input and coefficients (but different state) lead to better performance. It was shown in a followup to [1] that performed ablation study. [1] https://arxiv.org/ab…

Not sure if you meant this because it doesn't cite the paper you mention, but it's a similar work: "An Investigation of Model-Free Planning", Guez et Al. (Deepmind) 2019 https://arxiv.org/abs/1901.03559

Speaking of not citing, that one could go a bit further back.

https://cdn.aaai.org/AAAI/1987/AAAI87-048.pdf

Re: Diffusion Beats Autoregressive in Data-Constrained Settings

#12

Post by tripplyons »

yorwba wrote: post #9
tripplyons wrote: post #7

Earlier quoted context omitted.

There are definitely parallels between diffusion and reasoning models, mostly being able to spend longer to get a better solution by using a more precise ODE solver for diffusion or using more tokens for reasoning. However, due to how diffusion models are trained, they never see their own predictions as input, so they cannot learn to store information across steps. This is the complete opposite for reasoning models.

You can train a diffusion model using its own predictions as input, no problem at all.

At that point it is not following a diffusion training objective. I am aware of papers that do this, but I have not seen one that shows it as a better pretraining objective than something like v-prediction or flow matching.

Re: Diffusion Beats Autoregressive in Data-Constrained Settings

#13

Post by imtringued »

thesz wrote: post #8

> This paper addresses the challenge by asking: how can we trade off more compute for less data? Autoregressive models are not matched by compute and this is the major drawback. There is evidence that training RNN models that compute several steps with same input and coefficients (but different state) lead to better performance. It was shown in a followup to [1] that performed ablation study. [1] https://arxiv.org/ab…

It has already been proven that deep equilibrium models with a single layer are equivalent to models with a finite number of layers and the converse as well. That you can get the performance of a DEQ using a finite number of layers.

The fixed point nature of DEQs means that they inherently have a concept of self assessment how close they are to the solution. If they are at the solution, they will simply stop changing it. If not, they will keep performing calculations.

Re: Diffusion Beats Autoregressive in Data-Constrained Settings

#14

Post by imtringued »

smokel wrote: post #4

I fail to understand why we would lack data. Sure, there is limited (historical) text, but if we just open up all available video, and send out interactive robots into the world, we'll drown in data. Then there is simulated data, and tons of sensors that can capture vast amounts of even more data. Edit: from the source [1], this quote pretty much sums it all up: "Our 2022 paper predicted that high-quality text data w…

Robotics data isn't labeled and if you build a robot, there ain't anyone who has collected data for your particular robot.

There is also the problem that on-device learning is not yet practical.

Re: Diffusion Beats Autoregressive in Data-Constrained Settings

#15

Post by blurbleblurble »

tripplyons wrote: post #7
blurbleblurble wrote: post #3

I have a feeling this technique might make waves: https://openreview.net/forum?id=c05qIG1Z2B#discussion

There are definitely parallels between diffusion and reasoning models, mostly being able to spend longer to get a better solution by using a more precise ODE solver for diffusion or using more tokens for reasoning. However, due to how diffusion models are trained, they never see their own predictions as input, so they cannot learn to store information across steps. This is the complete opposite for reasoning models.

I'm probably not understanding your point but did you look at the paper? This explicitly does diffusion in an autoencoded latent space of the autoregressive prediction itself. The starting point is that prediction, but depending on how much noise is used, the diffusion model itself directly contributes to the prediction process to some degree or another.

It should be trivial to make an encoder that has some memory of at least part of the prompt (say the tailing part) and do a diffusion step there too.

Re: Diffusion Beats Autoregressive in Data-Constrained Settings

#16

Post by mxwsn »

tripplyons wrote: post #12
yorwba wrote: post #9

Earlier quoted context omitted.

You can train a diffusion model using its own predictions as input, no problem at all.

At that point it is not following a diffusion training objective. I am aware of papers that do this, but I have not seen one that shows it as a better pretraining objective than something like v-prediction or flow matching.

Why is not the diffusion training objective? The technique is known as self-conditioning right? Is it an issue with conditional Tweedie's?
Post reply on HN