📝 Publications

🗡️🛡️ Jailbreak Attacks and Defenses

ACL 2025

Hierarchical Safety Realignment: Lightweight Restoration of Safety in Pruned Large Vision-Language Models \ Yue Li^*, Xin Yi^*, Dongsheng Shi, Gerard de Melo, Xiaoling Wang and Linlin Wang^†.

Arxiv | Project | ACL Anthology

The current pruning methods will lead to a significant degradation of the model’s safety at a higher sparsity.
The HSR (Hierarchical Safety Realignment) method we proposed can achieve safety realignment for the pruned model by restoring only a very small number of neurons. HSR is effective for both LLM and LVLM.

ESWA 2026 Latent-space adversarial training with post-aware calibration for defending large language models against jailbreak attacks, Xin Yi, Yue Li, Dongsheng Shi, Linlin Wang^†, Xiaoling Wang and Liang He.
Preprint Unified defense for large language models against jailbreak and fine-tuning attacks in education, Xin Yi, Yue Li, Dongsheng Shi, Linlin Wang^†, Xiaoling Wang and Liang He.

📄🔍 Model Watermarks and Fingerprints

SIGKDD 2026

AGMark: Attention-Guided Dynamic Watermarking for Large Vision-Language Models \ Yue Li^*, Xin Yi^*, Dongsheng Shi, Yongyi Cui, Gerard de Melo and Linlin Wang^†.

Arxiv | Project

We propose AGmark, a watermarking method for LVLMs that follows the red–green token partitioning paradigm.
At each step of autoregressive generation, AGmark dynamically identifies candidate token weights and adaptively determines the size of the protected token set, effectively mitigating the trade-off between text quality and watermark detectability.

Preprint From Construction to Injection: Edit-Based Fingerprints for Large Language Models, Yue Li^*, Xin Yi^*, Dongsheng Shi, Yongyi Cui, Gerard de Melo and Linlin Wang^†.
KBS 2025 Unified Attacks to Large Language Model Watermarks: Spoofing and Scrubbing in Unauthorized Knowledge Distillation, Xin Yi, Yue Li, Shunfan Zheng, Linlin Wang^†, Xiaoling Wang and Liang He.

🤖🎯 Agents

Preprint MAE: Continuous Learning–based Medical Multi-Agent System via Experience Mining and Reuse, Dongsheng Shi, Xin Yi, Yue Li, Linlin Wang^†.
Preprint SURGENT: A Surgical Multi-Agent Assistance System Across the Perioperative Workflow, Dongsheng Shi, Yue Li, Xin Yi, Huawei Feng, Linlin Wang^†.

🗃️📊 Benchmarks

ESWA 2026 Benchmarking Large Language Models for End-to-End Clinical Support in Traditional Chinese Medicine, Dongsheng Shi, Xin Yi, Yue Li, Linlin Wang^†.

👻💭 Model Hallucinations