Join ExRAIL Lab
Recent Lab Highlights
ICML 2026 Oral Rui Xu — Information Flow Reveals When to Trust LMs (168 / 23918 papers, 0.7%)
ICML 2026 Spotlight Yazheng Liu — Explainability of Temporal Graph Networks (536 / 23918 papers, 2.2%)
ICML 2026 Rufeng Chen — PSG-Nav: Probabilistic Scene Graph Navigation
ECCV 2026 Yue Chang — RAG-3DSG: Enhancing 3D Scene Graphs with Retrieval-Augmented Generation
IJCAI 2026 Zhaofan Zhang — Perturbation-Resilient Navigation with Distributionally Robust RL
IROS 2026 Zhaofan Zhang — Adaptive-Critical: Risk-Sensitive Navigation
AAAI 2026 Yi Wang — Efficient LLM Fine-Tuning
ACL 2025 Xiaqiang Tang — Steering LLMs with Activation Vectors
About the Lab
The ExRAIL Lab (Trustworthy AI Exploration Lab) at HKUST(GZ), directed by Prof. Sihong Xie,
conducts research on trustworthy machine learning — explainability, uncertainty quantification, fairness, robustness,
and reliability — with applications in graph learning, language models, multi-agent systems, and robotics.
Prof. Xie is a tenured Associate Professor, recipient of the NSF CAREER Award (2022) and
国家自然科学基金优秀青年科学基金项目（海外） (2024). He is the Executive Deputy Director of the
HKUST(GZ)-Guangdong Unicom Joint Computing Lab, and serves on the executive committees of ACM SIGSPATIAL China, CCF-AI, and CCF-BigData.
He has published 130+ papers (NeurIPS, ICML, ICLR, AAAI, IJCAI, KDD, ACL, CVPR, CIKM) with 3,100+ citations and an h-index of 24.
Research Areas
Trustworthy AI
Explainability, uncertainty quantification, fairness, robustness, reliability, and graph-based trustworthy ML (interpretability, temporal GNNs).
LLM & Foundation Models
LLM interpretability (information flow, steering), faithfulness, VLM reasoning, and knowledge-intensive LLM.
RL for Robotics
Safe RL, distributionally robust RL, scene graph navigation, and autonomous path-planning.
3D Vision + Embodied AI
3D scene graphs, VLA (vision-language-action), retrieval-augmented 3D understanding, and efficient scene representation.
Working With Students
What you can expect as a member of ExRAIL:
- Publish at top venues — Lab members regularly publish at ICML, NeurIPS, ICLR, AAAI, ACL, ECCV, IJCAI, IROS. Multiple papers achieved Oral/Spotlight distinction (ICML 2026 Oral at 0.7% acceptance rate).
- Structured mentorship — Weekly 1:1 meetings with Prof. Xie, weekly group meetings for paper reading and systematic ML training. Senior PhD students serve as “senior collaborators” to provide day-to-day guidance.
- Independent research training — develop skills in problem formulation, experiment design, proposal writing, and project management. Students are encouraged to pursue their own research direction within the lab’s scope.
- Collaboration & networking — collaborate with co-advisors from HKUST, industry (Tencent, Huawei, Pengcheng Lab), and international partners. Long-term visits to HKUST Clear Water Bay campus are available.
- Career development — alumni have been admitted to top PhD/Master programs (NYU, CMU, UW-Madison) and placed in leading companies (Meta, Amazon).
Who We’re Looking For
We welcome students with strong backgrounds in ML, math, and programming. Here’s how your interests map to our current work:
| Your Background |
Research Topics |
Current Members |
| LLM / NLP |
LLM interpretability, steering, faithfulness, VLM reasoning |
Rui Xu, Yazheng Liu, Xiaqiang Tang |
| RL / Robotics |
Safe RL, distributionally robust RL, navigation, autonomous driving |
Zhaofan Zhang, Rufeng Chen |
| CV / 3D Vision |
3D scene graphs, VLA, embodied AI, multi-modal learning |
Yue Chang |
Open Positions
- Ph.D. positions starting 2026 Fall or 2027 Spring/Fall. Full scholarship (HKUST(GZ) standard) + competitive stipend. Research in any of the areas above.
- Research Assistant / Internship positions — paid according to experience and progress. Opportunities for co-authored publications. Outstanding RAs receive priority Ph.D. admission.
- Visiting students — welcome from partner universities for 3–12 month research visits.
Requirements
- Curiosity-driven — strong interest in the underlying principles of AI, not just applying off-the-shelf tools.
- Results-oriented — grounded in code, data, and reproducible experiments.
- Solid foundation — strong in calculus, probability & statistics, linear algebra, and programming (Python, PyTorch). Publication or open-source experience is a plus but not required.
- Language — English fluency meeting HKUST-GZ postgraduate admission standards.
How to Apply
Email the following to sihongxie@hkust-gz.edu.cn with subject line: Prospective Student – Your Name – Expected Start (e.g., 2026 Fall)
- CV / Resume
- Published or submitted papers; open-source projects (GitHub, demo, etc.)
- Research experience summary
- Research statement (1–2 pages)
- Academic transcripts
Official application portal: HKUST(GZ) Fok Ying Tung Graduate School