Publications
Two questions run through most of my work: how much structure a language model needs in order to act reliably, and how much structure it can recover from unstructured text. The first is what I work on now, on agents and their security; the second is what led me there, through information extraction on scientific writing.
Research
Agents that use tools and catch their own mistakes
RIMRULE (ACL 2026) learns rules for tool-using agents under an MDL objective, so that what the agent learns stays compact enough to be inspected. Metacognitive Self-Correction for Multi-Agent Systems (ACL 2026) reconstructs what a multi-agent system was about to do next, and uses the mismatch as a signal for catching its own errors mid-execution.
AI agent security
This is my current focus at AWS: tool-use security, robustness, and anomaly detection for agents that take real actions in real systems. An agent that can call tools is an agent that can be made to call the wrong one — the reliability question and the security question turn out to be the same question.
Structured knowledge from text
Most of my Ph.D. work is about pulling entities, relations, and mentions out of scientific writing — and about making that extraction hold up when supervision is noisy or scarce. SciER (EMNLP 2024) is an entity and relation extraction dataset covering datasets, methods, and tasks in scientific documents; DMDD (TACL) and SciDMT (LREC-COLING 2024) target dataset and scientific-mention detection at scale. DynClean (NAACL 2025) uses training dynamics to clean labels for distantly-supervised NER, and Many-Shot In-Context Learning for NER (ACL 2026) asks how far in-context learning can substitute for annotation in the low-resource case.
Structure for scientific domains
A related thread applies this to specific scientific fields, where the vocabulary is the hard part. Taxonomy-Driven Knowledge Graph Construction (ACL 2025) builds domain knowledge graphs from a taxonomy rather than a flat schema; ClimateIE and Querying Climate Knowledge (both 2025) bring information extraction and semantic retrieval to climate science; FlowLearn (ECAI 2024) extends the question past text, evaluating how well vision-language models read flowcharts.
Publications
Underline marks equal contribution. A continuously updated list lives on Google Scholar.
2026
- Scaling Performance and Low-Resource Annotation with Many-Shot In-Context Learning for Named Entity Recognition Qi Zhang, Fangping Lan, Cornelia Caragea, Longin Jan Latecki, Eduard Dragut ACL 2026 · Findings
- RIMRULE: Improving Tool-Using Language Agents via MDL-Guided Rule Learning Xiang Gao, Yuguang Yao, Qi Zhang, Kun Dong, Avirup Baidya, Rui Guo, Hugo Hasson, Kanak Das ACL 2026 · Main
- Metacognitive Self-Correction for Multi-Agent Systems via Prototype-Guided Next-Execution Reconstruction Qi Zhang, Xu Shen, Song Wang, Zhen Tan, Xinyu Zhao, Laura Yao, Vaishnav Tadiparthi, Hossein Nourkhiz Mahjoub, Ehsan Moradi Pari, Kwonjoon Lee, Tianlong Chen ACL 2026 · Findings
- Making Revisions Understandable: A Survey of Edit Intentions, Methods, and Applications Fangping Lan, Qi Zhang, Eduard Dragut ACL 2026 · Findings
2025
- DynClean: Training Dynamics-based Label Cleaning for Distantly-Supervised Named Entity Recognition Qi Zhang, Huitong Pan, Zhijia Chen, Longin Jan Latecki, Cornelia Caragea, Eduard Dragut NAACL 2025 · Findings
- Taxonomy-Driven Knowledge Graph Construction for Domain-Specific Scientific Applications Huitong Pan, Qi Zhang, Eduard Dragut, Cornelia Caragea, Longin Jan Latecki ACL 2025 · Findings
- Querying Climate Knowledge: Semantic Retrieval for Scientific Discovery Mustapha Adamu, Qi Zhang, Huitong Pan, Eduard Dragut, Longin Jan Latecki SIGIR 2025 · MANILA Workshop
- ClimateIE: A Dataset for Climate Science Information Extraction Huitong Pan, Mustapha Adamu, Qi Zhang, Eduard Dragut, Longin Jan Latecki ACL 2025 · 2nd ClimateNLP Workshop
2024
- SciER: An Entity and Relation Extraction Dataset for Datasets, Methods, and Tasks in Scientific Documents Qi Zhang, Zhijia Chen, Huitong Pan, Cornelia Caragea, Longin Jan Latecki, Eduard Dragut EMNLP 2024 · Main (20.8% acceptance)
- FlowLearn: Evaluating Large Vision-Language Models on Flowchart Understanding Huitong Pan, Qi Zhang, Cornelia Caragea, Eduard Dragut, Longin Jan Latecki ECAI 2024 (23% acceptance)
- SciDMT: A Large-Scale Corpus for Detecting Scientific Mentions Huitong Pan, Qi Zhang, Cornelia Caragea, Eduard Dragut, Longin Jan Latecki LREC-COLING 2024
2023
- DMDD: A Large-Scale Dataset for Dataset Mention Detection Huitong Pan, Qi Zhang, Eduard Dragut, Cornelia Caragea, Longin Jan Latecki Transactions of the ACL (TACL)
2022
- A Novel Sleep Staging Network Based on Multi-Scale Dual Attention Huafeng Wang, Chongang Lu, Qi Zhang, Zhimin Hu, Xiaodong Yuan, Pingshu Zhang, Wanquan Liu Biomedical Signal Processing and Control