AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

CR³: Boosting Compositional Reasoning in MLLMs Through Rule-Based Reinforcement Learning

Shun Qian, Bingquan Liu, Chengjie Sun, Peijin Xie, Zhen Xu, Baoxun Wang

AAAI 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Chain-of-thought multimodal reasoning · 0 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 CR³, a rule-based reinforcement learning approach to boost compositional reasoning in multimodal large language models (MLLMs).

Problem

Compositional reasoning (structured combinations of objects, attributes, and relations) has been studied mainly for VLMs like CLIP and SigLIP, with limited exploration in MLLMs.

Approach

Applies rule-based reinforcement learning to MLLMs to improve compositional reasoning; the abstract is truncated, so the specific reward design and method details are not stated.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
apply CR3 rule-based RLQwen2.5-VL-7B · Compositional avgaverage score: 47.2 -> 57.2Improved
apply CR3 rather than SFT on same dataQwen2.5-VL-3B · Compositional avgaverage score: 29.9 -> 44.6Improved
apply CR3InternVL3-2B · Compositional avgaverage score: 24.9 -> 36.4Improved
apply SFT on compositional dataInternVL3-8B · Compositional avgaverage score: 48.8 -> 50.2Improved
apply SFT on compositional data (out-of-domain)InternVL3-2B · MMMUaccuracy: 44.7 -> 42.7Hurt
apply CR3 (out-of-domain)Qwen2.5-VL-7B · MMMUaccuracy: 46.7 -> 52.0Improved
add dynamic task mixing instead of uniform mixing (ablation)CR3 on Qwen2.5-VL-3B · MMVPaccuracy: 38.0 -> 44.7Improved
add dynamic task mixing instead of fixed-ratio mixing (ablation)CR3 on Qwen2.5-VL-3B · Winogroundgroup score: 20.5 -> 27.0Improved
train on all three tasks instead of TG-VCR only (ablation)CR3 on Qwen2.5-VL-3B · Winogroundgroup score: 20.5 -> 27.0Improved
train on CITM only (ablation)Qwen2.5-VL-3B · MMVPaccuracy: 26.0 -> 39.3Improved
train on VG-TCR only (ablation)Qwen2.5-VL-3B · Winogroundtext score: 61.8 -> 65.8Improved
set format reward weight to 1.0 instead of 0.0 (ablation)CR3 on Qwen2.5-VL-3B · Winogroundgroup score: 19.5 -> 27.0Improved

Limitations

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