Benchmarking Multi-turn Medical Diagnosis: Hold, Lure, and Self-correction
About
I am an Assistant Professor in the Computer Science Department at the College of Staten Island, City University of New York.
Before joining the City University of New York, I was a postdoctoral researcher at the University of Illinois Urbana-Champaign, where I was advised by Prof. Heng Ji and Prof. Chengxiang Zhai. I also worked with Prof. H. Chad Lane.
My research focuses on the knowledge lifecycle of large language models, with the goal of building more reliable and intelligent AI systems.
Recent Updates
[Aug 2026] Openings: I am looking for motivated students interested in interpretable, reliable, and trustworthy AI.
[Aug 2026] Our recent papers Geometric-disentanglement Unlearning (GU) and MemGuard were accepted to EMNLP 2026 Main; Current Agents Fail to Leverage World Model as Tool for Foresight to ACL 2026 Main; EMCompress to Findings of ACL 2026; ShortageSim to AAAI 2026 (Oral); ModelingAgent and SafeSwitch to Findings of EMNLP 2025; and How Do Analytic Design Choices Shape MLLM-Based Detection of Non-Verbal Behaviors in 360° Classroom Video to the EDM 2026 Proceedings!
[Aug 2026] I am joining the College of Staten Island, City University of New York as an Assistant Professor of Computer Science!
[Mar 2026] Call for papers for our 4th Towards Knowledgeable Foundation Models Workshop at ACL 2026.
[Mar 2026] Invited talk at the University of Arizona — Knowledge is Power, But Power Casts Shadows?
[Jan 2026] Invited talk at the University of Edinburgh — Knowledge is Power, But Power Casts Shadows?
[Jan 2026] Invited talk at the University of Massachusetts — Knowledge is Power, But Power Casts Shadows?
[Jan 2026] Invited talk at the UIUC Large NLP Group Seminar — Knowledge Lifecycle of LLMs
[Aug 2025] Organized the 3rd Towards Knowledgeable Foundation Models Workshop at ACL 2025.
Research Interest
My research spans NLP, LLM interpretation, and trustworthy AI, with a focus on how LLMs acquire, represent, and use knowledge to improve reliability and performance.
More broadly, I am interested in questions at the intersection of knowledge and reasoning: Can failures in AI reasoning be traced to deficiencies in how knowledge is represented and constructed? Can highly reliable reasoning emerge from better underlying knowledge building, updating, and organization?
My current research directions include:
- HallucinationUnderstanding hallucination through the lens of knowledge representation and interaction, with the goal of interpreting, predicting, and preventing failures such as knowledge overshadowing.
- UpdatingDeveloping reliable knowledge updating methods that incorporate new knowledge while preserving existing knowledge and model robustness.
- AcquisitionImproving knowledge acquisition and organization to enhance model intelligence and support stronger reasoning.
Selected Publications
see Google Scholar for all
Atomic Reasoning for Scientific Table Claim Verification
The Law of Knowledge Overshadowing: Towards Understanding, Predicting, and Preventing LLM Hallucination
Knowledge Overshadowing Causes Amalgamated Hallucination in Large Language Models
Geometric-disentanglement Unlearning
MemGuard: Preventing Memory Contamination in Long-Term Memory-Augmented Large Language Models
ModelingAgent: Bridging LLMs and Mathematical Modeling for Real-World Challenges
📢 ShortageSim: Simulating Drug Shortages under Information Asymmetry
Invited Talks & Tutorials
-
Knowledge is Power, But Power Casts Shadows?
-
Knowledge is Power, But Power Casts Shadows?
-
The Knowledge Lifecycle of LLMs: Memorization, Editing, and Beyond
-
Knowledge is Power, But Power Casts Shadows?
-
Knowledge is Power, But Power Casts Shadows?
-
Towards Knowledgeable Foundation Models
-
Robust and Trustworthy Language Models
-
Knowledge Overshadowing Causes Amalgamated Hallucination in Large Language Models
-
Knowledge Overshadowing Causes Amalgamated Hallucination in Large Language Models
-
The Lifecycle of Knowledge in Large Language Models: Memorization, Editing, and Beyond
-
Knowledge Overshadowing Causes Amalgamated Hallucination in Large Language Models
Teaching
- Instructor · CSC 733: Natural Language Processing
- Guest Lecturer · CS 496: Agent AI
- Instructor · CS 591 BAI: Biologically Plausible Artificial Intelligence
- Instructor · CS 591 BAI: Biologically Plausible Artificial Intelligence
- Instructor · CS 591 BAI: Biologically Plausible Artificial Intelligence
- Instructor · CS 591 BAI: Biologically Plausible Artificial Intelligence
- Teaching Assistant · COMP 3134: Business Intelligence and CRM
- Teaching Assistant · COMP 5511: Artificial Intelligence Concepts
- Teaching Assistant · COMP 5511: Artificial Intelligence Concepts
- Teaching Assistant · COMP 4141: Crowdfunding and e-Finance
- Teaching Assistant · COMP 1433: Introduction to Data Analytics
- Teaching Assistant · COMP 1433: Introduction to Data Analytics
Professional Service
- Organizer · Workshop on Towards Knowledgeable Foundation Models
- Session Chair · Language Models and Interpretability
- Session Chair · Language Modeling
- Area Chair
- Conference Reviewer
Research Experience
University of Illinois Urbana-Champaign
Postdoctoral Researcher · 2024–2026
Postdoctoral Researcher · 2024–2026
University of Illinois Urbana-Champaign
Visiting Ph.D. Student · 2023–2024
Visiting Ph.D. Student · 2023–2024
The Hong Kong Polytechnic University
Ph.D. in Computing · 2020–2024
Ph.D. in Computing · 2020–2024
Dartmouth College
Research Intern · 2019–2020
Research Intern · 2019–2020
