← Back to the collection
Adaptive depth ·
Recurrent Denoiser
Diffusion as a Training Curriculum for Timestep-Free Iterative Reasoning
Diffusion model with recurrent memory that does more than state-based self-correction
Inside the method
See paper for architecture
Simplified conceptual schematic. Consult the paper for the complete architecture.
- Recurrence family
- Adaptive depth
- Depth control
- See paper
- KV / state strategy
- See paper
Reading note
Community-submitted work. Consult the linked paper for architecture details and experimental conditions.
Sources checked 2026-09-15. This catalog does not imply independent reproduction.
Cite this work
@misc{arxiv2609014492026,
title = {Diffusion as a Training Curriculum for Timestep-Free Iterative Reasoning},
author = {Mariia Drozdova and Aidan Sirbu and Pietro Miotti and Robert Obryk and Mayalen Etcheverry and Eyvind Niklasson and Blake Richards},
year = {2026},
eprint = {2609.01449},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.01449}
}