AI Paper Atlas

Research map / Architectures and efficiency / RNNs and normalization

One Step Forward and K Steps Back: Better Reasoning with Denoising Recursion Models

Chris Cameron, Wangzheng Wang, Nikita Ivanov, Ashmita Bhattacharyya, Didier Chételat, Yingxue Zhang

ICML 2026 · Architectures and efficiency › RNNs and normalization · approach: Scalable approximation and equivariant compute · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
train with denoising recursion (corrupt target, recover over k steps) instead of TRM backward-trainingTRM 7M · ARC-Easypass@2 accuracy (%): 45.7 -> 50.5Improved
use DRM training instead of TRMTRM 7M, no pretraining · ARC2-Evalpass@2 accuracy (%): 6.3 -> 9.6Improved
use DRM training instead of TRM, NVARC pretrainingTRM 7M and 14M · ARC2-Eval (NVARC pretraining)pass@2 accuracy (%): 12.5 -> 16.7 (7M); 21.8 -> 24.9 (14M)Improved
add deep supervision (intermediate TBPTT losses) to TRM (ablation)TRM 7M · ARC-Easypass@2 accuracy (%): 34.7 -> 45.7Improved
add recurrence (TRM) versus a standard non-recursive transformer (ablation)transformer · ARC-Easypass@2 accuracy (%): 25.5 -> 45.7Improved
use standard one-step masked diffusion (no recursion backprop), even stacked to 70M (ablation)masked diffusion · ARC-Easypass@2 accuracy (%): 0.0 (7M), 0.7 (70M)Hurt
scale DRM from 7M/4 grad steps to 14M/6 grad steps (ablation)DRM · ARC2-Train pretrainingpass@2 accuracy (%): +3-4%Improved
add NVARC to pretraining data (ablation)DRM 7M and 14M · ARC2-Evalpass@2 accuracy (%): about +7% (7M), +11.5% (14M)Improved
inject noise into latent states at TBPTT boundaries (SPRM) on TRMTRM 7M · ARC2-Eval (no pretraining)pass@2 accuracy (%): 6.3 -> 10.1Improved
SPRM state perturbation with reARC pretrainingTRM 7M · ARC2-Eval (reARC pretraining)pass@2 accuracy (%): 12.4 -> 11.8Hurt
DRM (14M, NVARC) versus NVARC 4B LLM trained on same dataDRM 14M · ARC2-Eval (NVARC pretraining)pass@2 accuracy (%): 22.6 -> 24.9Improved
use cosine instead of linear/sigmoid masking schedule (ablation)DRM · ARCtraining convergenceImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy