AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

RoSeViT: Role-Separated Vision Transformers for ARC Visual Reasoning

Zhiguang Liu, Yi Shang

NeurIPS 2026 · Architectures and efficiency › Attention and transformers · approach: Hybrid and interpretable ViT designs · 2 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes RoSeViT, a role-separated vision transformer for ARC visual reasoning that separates local visual state from global transformation logic.

Problem

ARC tasks require inferring a high-level transformation rule from a few demonstrations and applying it consistently across a grid; dense ViTs entangle local visual state and global transformation logic in shared tokens, which can hinder consistent reasoning.

Approach

Separates token representations by role (local visual state vs. global transformation logic) within a vision transformer rather than using a shared token representation for both.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace dense VARC-ViT backbone with one-step RoSeViT d-1VARC protocol · ARC-1pass@2 accuracy: 54.5 -> 56.6Improved
widen dense VARC-ViT to 28M parameters (parameter-matched control) (ablation)VARC-ViT · ARC-1pass@2 accuracy: 54.5 -> 53.6Hurt
scale dense VARC-ViT to 66M (ablation)VARC-ViT · ARC-1pass@2 accuracy: 54.5 -> 53.0Hurt
add recurrent refinement (c-2-TTT2) to role-separated backbone (ablation)RoSeViT · ARC-1pass@2 accuracy: 56.6 -> 58.0Improved
staged training with four recurrent steps and TTT3 (c-d-4-TTT3) (ablation)RoSeViT · ARC-1pass@2 accuracy: 58.0 -> 58.8Improved
use RoSeViT-Ensemble instead of VARC-Ensemblefull ARC system · ARC-1pass@2 accuracy: 60.4 -> 62.6Improved
stronger test-time adaptation (TTT1 to TTT3) on d-1 (ablation)RoSeViT d-1 · ARC-1 vs ARC-2pass@2 accuracy: ARC-1 56.6 -> 55.5; ARC-2 6.9 -> 8.3Mixed
replace VARC-ViT with RoSeViT d-1 under default adaptationVARC protocol · ARC-2pass@2 accuracy: 8.3 -> 6.9Hurt
use pure controller-bottleneck blocks (variants a/b) without dense visual pass (ablation)RoSeViT · ARC offline trainingtraining/validation curvesHurt
add local 3x3 workspace neighborhood (d over c) (ablation)RoSeViT · ARC offline trainingtraining/validation curvesImproved
zero controller embeddings at inference after TTT (ablation)RoSeViT d-1 · ARC-1pass@2 accuracy: 56.3 / 56.3 -> 43.3 / 43.0 (-13.1)Hurt
zero task token of dense control after TTT (ablation)Dense 28M · ARC-1pass@2 accuracy: 52.9 / 54.9 -> 50.3 / 52.5 (-2.5)No clear effect

Experimental setup

Baselines: dense Vision Transformers

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