# How to Pretrain Transformers Without Backpropagation Using Token-wise Activation Noise

- Link: https://qlabs.sh/research/dust
- Discussion: https://startupxo.com/en/833/
- Posted by startupxo at 2026-10-06T08:39:02.529Z
- 0 points, 0 comments, kind story, origin curated
- Written in ko, machine translated to en

## Summary

- Dust is a new algorithm that applies random perturbations to the neural activation values of each token and estimates gradients by measuring and averaging how much these perturbations reduce loss on a per-token basis.
- Traditional evolution strategies required a separate forward pass for each weight, but Dust independently applies perturbations at the token level, evaluating thousands of perturbations in a single forward pass, greatly improving computational efficiency.
- Experiments by the Q Labs research team showed that Dust improved efficiency by 1000 to 10,000 times compared to the existing weight-space evolution strategy EGGROLL when training on over 1 million tokens; this efficiency gain became more pronounced as model size increased.
- They also confirmed that gradients estimated by evolution strategies matched backpropagation results well up to scales of 1 billion tokens; however, currently, the computational efficiency is not yet sufficient to fully replace backpropagation.

## Sources

- Hacker News: https://news.ycombinator.com/item?id=49970871 (201 points, 49 comments, checked 2026-10-06T10:30:50Z)

## Comments (0)

No comments yet.

JSON: https://startupxo.com/api/v1/item.json?id=833&lang=en
