Shun Liu - CS Ph.D. Student at VCU

News

Sep. 2026 📝 One paper working on Privacy-Preserving Visual Generation submitted to ICLR'27!
Sep. 2026 🎓 Started my Ph.D. journey at Virginia Commonwealth University, Richmond, USA!
Jul. 2026 📝 One paper working on Interpretable Vision-Language Reasoning submitted to AAAI'27!
Apr. 2026 📝 New preprint on open‑ended surgical segmentation released on arXiv!
Jun. 2025 🎓 Completed Bachelor’s degree in Shanghai University of Finance and Economics (SHUFE)!
Nov. 2024 📝 One paper accepted at CIKM 2024!

Selected Publications

Attribute Retrieving for Open-Vocabulary Endoscopic Compositional Referring Segmentation

Shun Liu, Nan Xi, Yang Liu, Tianyu Luan, Xuan Gong, David Doermann

arXiv, 2026

📄 Paper 💻 Code (coming soon)

We introduce ReferEndoscopy, the first large-scale instruction-based endoscopic RIS benchmark, and propose the AR-ERIS framework for open-vocabulary compositional referring segmentation, w.r.t. the frequency properties of scene objects. Our method achieves state-of-the-art performance on both simulated and real-world endoscopic data.

MEDFuse: Multimodal EHR Data Fusion with Masked Lab-Test Modeling and Large Language Models

Thao Minh Nguyen Phan, Cong-Tinh Dao, Chenwei Wu, Jian-Zhe Wang, Shun Liu, Jun-En Ding, David Restrepo, Feng Liu, Fang-Ming Hung, Wen-Chih Peng

CIKM, 2024

🌐 Project Page 📄 Paper 💻 Code

We propose MEDFuse, a multimodal EHR fusion framework integrating masked lab-test modeling and fine-tuned LLMs to unify structured lab results and unstructured clinical notes. Evaluated on MIMIC-III and internal FEMH datasets, MEDFuse reaches over 90% F1 score on 10-disease multi-label classification for reliable clinical prediction.