Look Before You Decide: Prompting Active Deduction of MLLMs for Assumptive Reasoning

Yian Li¹,  Wentao Tian¹,  Yang Jiao¹,  Jingjing Chen^{1 ✉},  Tianwen Qian²,  Bin Zhu³, 
Na Zhao⁴,  Yu-Gang Jiang^1
1Shanghai Key Lab of Intell. Info. Processing, School of CS, Fudan University
²East China Normal University   ³Singapore Management University
⁴Singapore University of Technology and Design
✉ Corresponding Author

📣 News

• [2025-06-02] Qwen-AD paper is released on the arxiv and GitHub repo is created.

🏠 Abstract

Recently, Multimodal Large Language Models (MLLMs) have achieved significant success across multiple disciplines due to their exceptional instruction-following capabilities and extensive world knowledge. However, whether these MLLMs possess human-like compositional reasoning abilities remains an open problem. To unveil their reasoning behaviors, we first curate a Multimodal Assumptive Reasoning Benchmark (MARS-Bench) in this paper. Interestingly, we find that most prevalent MLLMs can be easily fooled by the introduction of a presupposition into the question, whereas such presuppositions appear naive to human reasoning. Besides, we also propose a simple yet effective method, Active Deduction (AD), a novel reinforcement learning paradigm to encourage the model to actively perform composite deduction before reaching a final decision. Equipped with the proposed AD method, a MLLM demonstrates significant improvements in assumptive reasoning abilities without compromising its general-purpose question-answering performance. We also provide extensive evaluations of both open-source and private MLLMs on MARS-Bench, along with experimental analyses of the AD method.

🔨 Install

Coming soon.

🎓 Training

Coming soon.

🤖 Inference

Coming soon.

🤗 Checkpoints

Coming soon.

📚 Datasets

Coming soon.

🔗 Citation

If you find this project useful in your research, please consider citing:

@article{li2024look,
  title={Look before you decide: Prompting active deduction of mllms for assumptive reasoning},
  author={Li, Yian and Tian, Wentao and Jiao, Yang and Chen, Jingjing and Qian, Tianwen and Zhu, Bin and Zhao, Na and Jiang, Yu-Gang},
  journal={arXiv preprint arXiv:2404.12966},
  year={2024}
}

👏 Acknowledgement

We are immensely grateful to the ms-swift, Qwen2.5-VL and VLMEvalKit projects for the inception of this repository. Thanks for their great work! 🙏