Shangyu Xing

I am now a Master student (3 year, course and research based) working with Prof. Xinyu Dai at Natural Language Processing Group, School of Artificial Intelligence, Nanjing University. I got my B.Sc. degree in Computer Science, School of Computer Science and Technology, Nanjing University in 2023.

My current research centers on the domain of multimodality and RL for LLMs, with the goal of advancing the capabilities and reliability of intelligent systems in complex, real-world scenarios.

Pretraining and preference alignment of Vision-Language models. This line of work investigates novel strategies for integrating visual and textual modalities during both pretraining and alignment stages. By improving VLMs’ perceptual abilities, cross-modal understanding, and factual consistency, I aim to enhance their robustness and generalization across diverse applications.
Reinforcement learning for reasoning Large Language Models. Leveraging advanced RL techniques such as GRPO, along with new paradigms like concept-based and latent language modeling, this research seeks to enhance LLMs’ reasoning capabilities, enabling them to perform more sophisticated, coherent, and reliable reasoning in tasks that require structured thought, logical inference, and long-term planning.

I am now actively seeking 2026 PhD opportunities. Here is my cv, or a pdf version here. Full publication list is available here or on Google Scholar. Looking forward to your interest and recommendation!

selected publications

ACMMM

DRIN: Dynamic Relation Interactive Network for Multimodal Entity Linking

Shangyu Xing^*, Fei Zhao^*, Zhen Wu, Chunhui Li, and 2 more authors

In Proceedings of the 31st ACM International Conference on Multimedia, 2023

arXiv Bib Code

@inproceedings{xing2023drin,
  title = {DRIN: Dynamic Relation Interactive Network for Multimodal Entity Linking},
  author = {Xing, Shangyu and Zhao, Fei and Wu, Zhen and Li, Chunhui and Zhang, Jianbing and Dai, Xinyu},
  booktitle = {Proceedings of the 31st ACM International Conference on Multimedia},
  pages = {3599--3608},
  year = {2023},
}

EMNLP

EFUF: Efficient Fine-grained Unlearning Framework for Mitigating Hallucinations in Multimodal Large Language Models

Shangyu Xing, Fei Zhao, Zhen Wu, Tuo An, and 4 more authors

EMNLP 2024, 2024

arXiv Bib Code

@article{xing2024efuf,
  title = {EFUF: Efficient Fine-grained Unlearning Framework for Mitigating Hallucinations in Multimodal Large Language Models},
  author = {Xing, Shangyu and Zhao, Fei and Wu, Zhen and An, Tuo and Chen, Weihao and Li, Chunhui and Zhang, Jianbing and Dai, Xinyu},
  journal = {EMNLP 2024},
  year = {2024},
}

ICLR

AnyPrefer: An Agentic Framework for Preference Data Synthesis

Yiyang Zhou, Zhaoyang Wang, Tianle Wang, Shangyu Xing, and 12 more authors

In International Conference on Learning Representations, 2025

arXiv Bib

@inproceedings{zhou2024anyprefer,
  title = {AnyPrefer: An Agentic Framework for Preference Data Synthesis},
  author = {Zhou, Yiyang and Wang, Zhaoyang and Wang, Tianle and Xing, Shangyu and Xia, Peng and Li, Bo and Zheng, Kaiyuan and Zhang, Zijian and Chen, Zhaorun and Zheng, Wenhao and Zhang, Xuchao and Bansal, Chetan and Zhang, Weitong and Wei, Ying and Bansal, Mohit and Yao, Huaxiu},
  booktitle = {International Conference on Learning Representations},
  year = {2025},
}

preprint

GePBench: Evaluating Fundamental Geometric Perception for Multimodal Large Language Models

Shangyu Xing, Changhao Xiang, Yuteng Han, Yifan Yue, and 5 more authors

submitted to NeurIPS 2025, 2025

arXiv Bib Slides

@article{xing2024gepbenchevaluatingfundamentalgeometric,
  title = {GePBench: Evaluating Fundamental Geometric Perception for Multimodal Large Language Models},
  author = {Xing, Shangyu and Xiang, Changhao and Han, Yuteng and Yue, Yifan and Wu, Zhen and Liu, Xinyu and Wu, Zhangtai and Zhao, Fei and Dai, Xinyu},
  year = {2025},
  journal = {submitted to NeurIPS 2025},
}