Conference paper · 2025 · Open access

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

Zhongwei Wan, Zhihao Dou, Che Liu, Yu Zhang, Dongfei Cui, Qinjian Zhao, Hui Shen, Jing Xiong, Yi Xin, Yifan Jiang, Chaofan Tao, Yangfan He and 2 more

neural information processing systems, 170566–170603 · Published 1 January 2025 · DOI 10.52202/085713-5139

Free PDFView at the publisherCite this paper

In brief

“SRPO: Enhancing Multimodal LLM Reasoning via Reflection-Aware Reinforcement Learning” is a 2025 conference paper by Zhongwei Wan and 13 co-authors, published in neural information processing systems. OpenAlex counts 67 citations as of 15 November 2026, which puts it in the top 1% of works published in 2025 by citations. It is open access: the full text is free to read at arXiv (Cornell University).

Citations
67
Most cited in
2026 (46)
Field-weighted impact
22.75
References
48
Authors
14

Abstract

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.

Cite this paper

Formatted from the catalogue record. Check names, capitals and page numbers against the paper before you submit.

APA 7

Wan, Z., Dou, Z., Liu, C., Zhang, Y., Cui, D., Zhao, Q., Shen, H., Xiong, J., Xin, Y., Jiang, Y., Tao, C., He, Y., Zhang, M., & Yan, S. (2025). SRPO: Enhancing Multimodal LLM Reasoning via Reflection-Aware Reinforcement Learning. neural information processing systems, 170566–170603. https://doi.org/10.52202/085713-5139

MLA 9

Wan, Zhongwei, et al. “SRPO: Enhancing Multimodal LLM Reasoning via Reflection-Aware Reinforcement Learning.” neural information processing systems, 2025, pp. 170566–170603. https://doi.org/10.52202/085713-5139.

Chicago

Wan, Zhongwei, Zhihao Dou, Che Liu, Yu Zhang, Dongfei Cui, Qinjian Zhao, Hui Shen, et al. 2025. “SRPO: Enhancing Multimodal LLM Reasoning via Reflection-Aware Reinforcement Learning.” neural information processing systems: 170566–170603. https://doi.org/10.52202/085713-5139.

Harvard

Wan, Z. et al. (2025) ‘SRPO: Enhancing Multimodal LLM Reasoning via Reflection-Aware Reinforcement Learning’, neural information processing systems, pp. 170566–170603. doi:10.52202/085713-5139.

IEEE

Z. Wan et al., “SRPO: Enhancing Multimodal LLM Reasoning via Reflection-Aware Reinforcement Learning,” neural information processing systems, pp. 170566–170603, 2025, doi: 10.52202/085713-5139.

BibTeX

@inproceedings{wan2025srpo,
  title     = {{SRPO: Enhancing Multimodal LLM Reasoning via Reflection-Aware Reinforcement Learning}},
  author    = {Wan, Zhongwei and Dou, Zhihao and Liu, Che and Zhang, Yu and Cui, Dongfei and Zhao, Qinjian and Shen, Hui and Xiong, Jing and Xin, Yi and Jiang, Yifan and Tao, Chaofan and He, Yangfan and Zhang, Mi and Yan, Shen},
  booktitle = {neural information processing systems},
  year      = {2025},
  pages     = {170566--170603},
  doi       = {10.52202/085713-5139},
  url       = {https://doi.org/10.52202/085713-5139}
}

Download for a reference manager: BibTeX (.bib) · RIS for Zotero, Mendeley and EndNote (.ris)

Topic and keywords

Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Multimodal Machine Learning Applications

  • multimodal large language models
  • self-reflection
  • self-correction
  • Group Relative Policy Optimization
  • multimodal reasoning
  • reward processing
  • mathematical reasoning

Work with this paper

Most-cited papers in Computer Science

All highly cited Computer Science papers

Questions about this paper

How do I cite this paper in APA style?

Wan, Z., Dou, Z., Liu, C., Zhang, Y., Cui, D., Zhao, Q., Shen, H., Xiong, J., Xin, Y., Jiang, Y., Tao, C., He, Y., Zhang, M., & Yan, S. (2025). SRPO: Enhancing Multimodal LLM Reasoning via Reflection-Aware Reinforcement Learning. neural information processing systems, 170566–170603. https://doi.org/10.52202/085713-5139

How many times has this paper been cited?

67 times, according to OpenAlex on 15 November 2026. Google Scholar and Scopus count different sets of sources, so their numbers differ.

Is this paper free to read?

Yes. It is open access: read the free full text at arXiv (Cornell University).

Who wrote this paper?

Zhongwei Wan, Zhihao Dou, Che Liu, Yu Zhang, Dongfei Cui, Qinjian Zhao, Hui Shen, Jing Xiong, Yi Xin, Yifan Jiang, Chaofan Tao, Yangfan He and 2 more.

What is the DOI of this paper?

10.52202/085713-5139. The permanent link is https://doi.org/10.52202/085713-5139.

Bibliographic data from OpenAlex (CC0), last checked 15 November 2026. ResearchGuru does not host this paper; the links go to the publisher or the repository that does.