IAAIDenoising Recursion Models
One Step Forward and K Steps Back: Better Reasoning with Denoising Recursion Models
Corrupts target solutions with noise like diffusion models but trains a looped refiner to undo the corruption over multiple recursive steps rather than one, giving a curriculum of intermediate states that better matches test-time refinement.
Inside the method
Noisy solution → [ Shared block ] × K steps → clean solution
Simplified conceptual schematic. Consult the paper for the complete architecture.
- Recurrence family
- Core recurrence
- Depth control
- Multi-step recursive denoising
- KV / state strategy
- Not applicable (non-autoregressive solver)
Reading note
Outperforms the Tiny Recursion Model on ARC-AGI, encouraging non-greedy, forward-looking generation.
Sources checked 2026-10-28. This catalog does not imply independent reproduction.
Related works
Cite this work
@misc{denoisingrecursionmodels2026,
title = {One Step Forward and K Steps Back: Better Reasoning with Denoising Recursion Models},
author = {Chris Cameron and Wangzheng Wang and Nikita Ivanov and Ashmita Bhattacharyya and Didier Chételat and Yingxue Zhang},
year = {2026},
eprint = {2604.18839},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2604.18839}
}