AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Seeing is Solving: Unlocking Efficient Multimodal RL via View Alignment

Qinsi Wang, Jing Shi, Kun Wan, Handong Zhao, Hancheng Ye, Zishan Shao, Jinghan Ke, Yudong Liu, Daniel Miranda, Purvak Lapsiya, Yiran Chen, Wentian Zhao

ICML 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Reinforcement learning for visual reasoning

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 view-alignment-based efficient RL fine-tuning for vision-language models to reduce the training cost of multimodal RLFT.

Problem

RL fine-tuning of VLMs for multimodal reasoning is prohibitively expensive, and existing methods port LLM RLFT techniques without accounting for dynamic text-vision alignment.

Approach

Exploits the dynamic text-vision alignment property of multimodal models to make multimodal RL fine-tuning more efficient, rather than directly porting LLM RLFT techniques. Full method details are not available from the abstract.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add PVA-based difficulty-aware curriculum data processing (ablation)vanilla GRPO · six reasoning benchmarks, Qwen2.5-VL-7Baverage accuracy: 42.7 -> 44.7Improved
add RVA-based CoT reward on top of data processing (ablation)FOCUS-RL (Data) · six reasoning benchmarks, Qwen2.5-VL-7Baverage accuracy: 44.7 -> 45.3Improved
use PVA curriculum instead of GainRL difficulty estimateGRPO + GainRL · six reasoning benchmarks, Qwen2.5-VL-7Baverage accuracy: 43.8 -> 44.7Improved
use PVA curriculum instead of AdaRFTGRPO + AdaRFT · six reasoning benchmarks, Qwen2.5-VL-7Baverage accuracy: 43.1 -> 44.7Improved
add FOCUS-RL (Data+Policy) to vanilla GRPOvanilla GRPO · LogicVista, Qwen2.5-VL-7Baccuracy: 44.2 -> 49.1Improved
add FOCUS-RL to vanilla GRPOvanilla GRPO · six reasoning benchmarks, Qwen2.5-VL-3Baverage accuracy: 33.2 -> 35.3Improved
add FOCUS-RL to vanilla GRPOvanilla GRPO · six reasoning benchmarks, InternVL3-2Baverage accuracy: 30.9 -> 33.6Improved
add FOCUS-RL to vanilla GRPOvanilla GRPO · six reasoning benchmarks, InternVL3-8Baverage accuracy: 40.7 -> 42.4Improved
add FOCUS-RL to GRPO training for speedvanilla GRPO · MathVista, Qwen2.5-VL-7Biterations to peak accuracy: 2.3x speedupImproved
add FOCUS-RL to GRPO training for speedvanilla GRPO · MathVista, InternVL3-8Biterations to peak accuracy: 4x speedupImproved
add FOCUS-RL (curriculum from easy to hard) to GRPOvanilla GRPO · Qwen2.5-VL-7B trainingcompletion length stabilityImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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