Ph.D. student in Computer Science
School of Computing, University of Connecticut (UConn)
Advised by Prof. Chuxu Zhang
Actively seeking a research internship for Summer 2027.
I am a Ph.D. student in the School of Computing at the University of Connecticut, advised by Prof. Chuxu Zhang. Before that, I obtained my master's degree from Hunan University under the supervision of Prof. Youhuan Li, and my bachelor's degree from South China Normal University. I am deeply grateful to Prof. Youhuan Li, whose guidance allowed me to experience the fascinating world of scientific research.
I work on large language models, with a focus on LLM agents and post-training. I am interested in how agents decide when and how much to think, and how training can make such decisions reliable across environments. Before moving to LLMs, I worked on databases and graphs, including subgraph matching, graph query processing, and graph neural networks for database systems. If you are interested in collaboration, feel free to contact me.
Selected publications are listed below; . Full list also on Google Scholar. My name is in bold.
Neural Information Processing Systems (NeurIPS), 2026
The same uncertainty or difficulty signal can indicate opposite compute utility across environments and backbones. DIAL learns when additional test-time computation actually helps instead of assuming a fixed signal direction.
International Conference on Machine Learning (ICML), 2026
Graph structure is treated as a persistent structural substrate shared across modalities and tasks rather than rebuilt per task. A unified structural schema and interleaved role-based training outperform task-isolated and naive multi-task baselines.
Conference on Empirical Methods in Natural Language Processing (EMNLP), 2026
An LLM agent designs molecular dynamics pipelines from verbal rewards produced by expert debate, matching human specialists on binding free-energy benchmarks and discovering an experimentally validated high-affinity binder.
ACM International Conference on Web Search and Data Mining (WSDM), 2025
A new search framework for learned subgraph matching that uses reinforcement learning and GNNs to generate fine-grained search plans.
International Joint Conference on Artificial Intelligence (IJCAI), Survey Track, 2025
A structured and in-depth overview of how graph neural networks are applied inside database systems.
ACM SIGMOD International Conference on Management of Data (SIGMOD), 2025
Introduces the match cover of a subgraph query: a subset of answers whose vertices are exactly those covered by the entire answer set.
IEEE International Conference on Data Engineering (ICDE), 2024
A new search process for continuous subgraph matching with multi-expand, cache reuse, and adaptive check strategies.
IEEE International Conference on Data Engineering (ICDE), 2023
Vertex encodings that determine edge nonexistence so that no-result edge queries are never executed.
Research Intern
Training-data optimization for coding agents.
Research Intern, ML Platform Department, TEG · Mentor: Yang Li
LLM agents and AI4DB.
Ziming Li