SRPO: Enhancing Multimodal LLM Reasoning via Reflection-Aware Reinforcement Learning

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wan, Zhongwei, Dou, Zhihao, Liu, Che, Zhang, Yu, Cui, Dongfei, Zhao, Qinjian, Shen, Hui, Xiong, Jing, Xin, Yi, Jiang, Yifan, Tao, Chaofan, He, Yangfan, Zhang, Mi, Yan, Shen
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916989506682880
author Wan, Zhongwei
Dou, Zhihao
Liu, Che
Zhang, Yu
Cui, Dongfei
Zhao, Qinjian
Shen, Hui
Xiong, Jing
Xin, Yi
Jiang, Yifan
Tao, Chaofan
He, Yangfan
Zhang, Mi
Yan, Shen
author_facet Wan, Zhongwei
Dou, Zhihao
Liu, Che
Zhang, Yu
Cui, Dongfei
Zhao, Qinjian
Shen, Hui
Xiong, Jing
Xin, Yi
Jiang, Yifan
Tao, Chaofan
He, Yangfan
Zhang, Mi
Yan, Shen
contents Multimodal large language models (MLLMs) have shown promising capabilities in reasoning tasks, yet still struggle with complex problems requiring explicit self-reflection and self-correction, especially compared to their unimodal text-based counterparts. Existing reflection methods are simplistic and struggle to generate meaningful and instructive feedback, as the reasoning ability and knowledge limits of pre-trained models are largely fixed during initial training. To overcome these challenges, we propose Multimodal Self-Reflection enhanced reasoning with Group Relative Policy Optimization (SRPO), a two-stage reflection-aware reinforcement learning (RL) framework explicitly designed to enhance multimodal LLM reasoning. In the first stage, we construct a high-quality, reflection-focused dataset under the guidance of an advanced MLLM, which generates reflections based on initial responses to help the policy model learn both reasoning and self-reflection. In the second stage, we introduce a novel reward mechanism within the GRPO framework that encourages concise and cognitively meaningful reflection while avoiding redundancy. Extensive experiments across multiple multimodal reasoning benchmarks, including MathVista, MathVision, MathVerse, and MMMU-Pro, using Qwen-2.5-VL-7B and Qwen-2.5-VL-32B demonstrate that SRPO significantly outperforms state-of-the-art models, achieving notable improvements in both reasoning accuracy and reflection quality.
format Preprint
id arxiv_https___arxiv_org_abs_2506_01713
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SRPO: Enhancing Multimodal LLM Reasoning via Reflection-Aware Reinforcement Learning
Wan, Zhongwei
Dou, Zhihao
Liu, Che
Zhang, Yu
Cui, Dongfei
Zhao, Qinjian
Shen, Hui
Xiong, Jing
Xin, Yi
Jiang, Yifan
Tao, Chaofan
He, Yangfan
Zhang, Mi
Yan, Shen
Computation and Language
Multimodal large language models (MLLMs) have shown promising capabilities in reasoning tasks, yet still struggle with complex problems requiring explicit self-reflection and self-correction, especially compared to their unimodal text-based counterparts. Existing reflection methods are simplistic and struggle to generate meaningful and instructive feedback, as the reasoning ability and knowledge limits of pre-trained models are largely fixed during initial training. To overcome these challenges, we propose Multimodal Self-Reflection enhanced reasoning with Group Relative Policy Optimization (SRPO), a two-stage reflection-aware reinforcement learning (RL) framework explicitly designed to enhance multimodal LLM reasoning. In the first stage, we construct a high-quality, reflection-focused dataset under the guidance of an advanced MLLM, which generates reflections based on initial responses to help the policy model learn both reasoning and self-reflection. In the second stage, we introduce a novel reward mechanism within the GRPO framework that encourages concise and cognitively meaningful reflection while avoiding redundancy. Extensive experiments across multiple multimodal reasoning benchmarks, including MathVista, MathVision, MathVerse, and MMMU-Pro, using Qwen-2.5-VL-7B and Qwen-2.5-VL-32B demonstrate that SRPO significantly outperforms state-of-the-art models, achieving notable improvements in both reasoning accuracy and reflection quality.
title SRPO: Enhancing Multimodal LLM Reasoning via Reflection-Aware Reinforcement Learning
topic Computation and Language
url https://arxiv.org/abs/2506.01713