Liu Hengjie

I am a Ph.D. student at Kyushu University in Fukuoka, Japan. My research focuses on AI safety and the formal verification of Transformer models, particularly certified robustness, scalable bound propagation, and relational verification. I am also interested in counterexample analysis, explainability, and model repair.

Before joining Kyushu University, I studied low-resource neural machine translation at Waseda University, focusing on German–Upper Sorbian translation and case-based methods for neural machine translation. I completed my undergraduate education in Cyberspace Security at Sichuan University as a member of the Wu Yuzhang Honors College.

I also have industry experience developing LLM- and OCR-based document-understanding systems, data-generation pipelines, and retrieval-augmented question-answering systems.

Research interests

My research interests lie broadly in the formal verification of neural networks. I am particularly interested in the verification of Transformer models, robustness analysis, and the formal specification of verification tasks and desired model properties.

Learn more about my research outputs, research projects, and experience.