Ph.D. Candidate Nanjing University School of Intelligence Science and Technology Suzhou, Jiangsu Province, P.R.China Email: x.song.matt[at]gmail.com; x.song[at]smail.nju.edu.cn [CN]

Biography

I am now pursuing my Ph.D. studies in the PR-Lab at Nanjing University under the supervision of Prof. Caifeng Shan and Assoc. Prof. Yuqi Fang. Prior to this, I served as a Research Specialist at the Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, working under the direction of Prof. Ruxin Wang. In 2023, I completed my Master's degree at Beijing University of Technology, where I was fortunate to be advised by Assoc. Prof. Xiaodan Zhang and co-advised by Prof. Junzhong Ji.

My research interests lie in the Multi-modal (CV, NLP, etc.) Learning and Causality in Biomedical Informatics. Currently, I also possess an interest in the Trustworthy Multi-modal Large Language Models and their applications for Healthcare.

News

Publications | Google Scholar

* equal contribution

Manuscripts

• Towards Unbiased Multimodal Hallucination Evaluation: Evidence-Grounded Benchmark Auditing via Causal Intervention.
  Xiao Song, Chenhao Guan, Weijia Li, Yanbo Fan, Yuqi Fang, Caifeng Shan.
  
• Where Medical Responses First Go Wrong: Logic-Threaded Semantic Tree for Hallucination Detection and Correction.
  Haonan Qin*, Xiao Song*, Ruiying Chen, Caifeng Shan, Yuqi Fang.
  
• Möbius: Bridging Mutual Supervision and Compensation between LVLMs and SAMs for Trustworthy Medical Vision-Language Systems.
  Xiao Song*, Haonan Qin*, Jiaxin Liu, Yang Bai, Weijia Li, Yuqi Fang, Caifeng Shan.
  
• Beyond Scaling Monoliths: Structural Auditability via Cognitive Decomposition for Trustworthy Medical Vision--Language Models.
  Ningyi Zhang, Yuan Gao, Xin Wang, Xiao Song, Hang Qu, Kaixuan Ren, Yue Gu, Yi Zhou, Steven Schalekamp, Chantong Lam, Caifeng Shan, Sio-Kei Im, Yue Sun, Tao Tan.
  
• Discovering Explicit Rules for MLLMs in Industrial Anomaly Detection.
  Weijia Li, Haolin Wang, Zhao Wang, Jiaju Jiang, Xin Liu, Xiao Song, Caifeng Shan, Fang Zhao.
  

Conference & Journal Papers

• MedF2Route: Evidence-Routed Graph Verification for Hallucination Detection in Clinical LVLMs.
  Xiao Song*, Ruiying Chen*, Haonan Qin, Jianyuan Huang, Yuqi Fang, Caifeng Shan.
  EMNLP, 2026.
• Detecting Clinical Hallucinations in LVLMs via Counterfactual Visual Grounding Uncertainty.
  Xiao Song*, Haonan Qin*, Zhaoxu Zhang, Jiong Zhang, Yuqi Fang, Caifeng Shan.
  MICCAI, 2026.
• Rethinking Radiology Report Generation via Causal Inspired Counterfactual Augmentation.
  Xiao Song, Jiafan Liu, Yun Li, Yan Liu, Wenbin Lei, Ruxin Wang.
  ACM BCB (Oral), 2024, Article No.5, 1–10. [paper]
• Multi-scale Superpixel based Hierarchical Attention Model for Brain CT Classification.
  Xiao Song, Xiaodan Zhang, Junzhong Ji, Ying Liu.
  J. Vis. Commun. Image R. (JVCIR), 2023, 91:103773. [paper]
• Cross-modal Contrastive Attention Model for Medical Report Generation.
  Xiao Song, Xiaodan Zhang, Junzhong Ji, Ying Liu, Pengxu Wei.
  COLING (Oral), 2022:2388–2397. [paper]

Patents

• 一种多模态推理大模型和分割大模型的互监督优化方法. CN121789001B
• 基于反事实数据增强的放射学报告生成方法. 202311704996.X
• 一种基于跨模态对比注意力机制的医学报告自动生成方法. CN115394397B

Teaching

2025 Spring	TA	Natural Language Processing	@ NJU
2026 Spring	TA	Natural Language Processing	@ NJU