I am a PhD student in Information and Communication Engineering at Nanjing University of Information Science and Technology. I received my bachelor's degree in Communication Engineering and my master's degree in Electronic Science and Technology, both from University of South China. From 2021 to 2022, I conducted research in the field of intelligent transportation at the Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences. I have participated in multiple national key R&D programs and projects funded by the National Natural Science Foundation of China, and have been awarded honors such as the National Scholarship for Doctoral Students.
My research focuses on spatiotemporal prediction and severe weather forecasting, with additional work in intelligent transportation and optimization algorithms. Notably, the lightning nowcasting model I developed is currently undergoing real-time operational testing in meteorological observatory warning systems.
Research Interests
Currently, I focus on the field of Multimodal Fusion and Spatio-Temporal Prediction.
■Multimodal Fusion: A wide array of meteorological observation instruments has collected massive volumes of meteorological variable data and remote sensing data. I am currently researching effective methods to fuse these multi-source data to achieve high-accuracy precipitation forecasting and thunderstorm warning.
■Spatio-Temporal Prediction: Spatiotemporal prediction tasks are highly complex. My research focuses on modeling spatiotemporal dependencies and applying these methods to tasks such as traffic forecasting and meteorological prediction.
Education
■
2024.9-present
Nanjing University of Information Science and Technology | Information and Communication Engineering / Environmental Perception and Intelligent Control | Doctor of Engineering
■
2014.9-2020.8
University of South China | Electronic science and technology | Master of Engineering
■
2010.9-2014.6
University of South China | Communications Engineering | Bachelor of Engineering
Publications
MissPred: A Robust Two-Stage Radar Echo Extrapolation Algorithm in Missing Pattern
Ziqi Zhao, Chunxu Duan, Lin Song, Qilin Zhang, Wenda Zhu, Yi Liu
2025-07 Our survey: AI Alignment: A Contemporary Survey has been accepted by ACM Computing Surveys
■
2025-07 Five papers (*Spotlight, ³Poster) are accepted by NeurIPS 2025
■
2025-07 Language Model Resist Alignment has been awarded the ACL 2025 Best Paper!
■
2024-01 MedAligner has been accepted to The Innovation
■
2025-05 Four papers are accepted by ACL 2025 Main
■
2025-05 SAE-V has been accepted as ICML 2025
Honors & Awards
■National Scholarship for Doctoral Students, Nanjing University of Information Science and Technology
■First-Class Scholarship for Doctoral Students of Nanjing University of Information Science and Technology
■Outstanding Student of Nanjing University of Information Science and Technology
Academic Service
■Reviewer for Atmospheric Research
■Reviewer for Remote Sensing
■Student Member of China Computer Federation (CCF)
■IEEE Student Member
■Student Member of the Chinese Association of Automation
AI Alignment Survey Accepted
Our survey "AI Alignment: A Contemporary Survey" has been officially accepted by ACM Computing Surveys (Impact Factor: 28.0, ranked 1/147 in Computer Science Theory & Methods). This survey systematically reviews the latest progress in AI alignment, covering core methods, theoretical challenges, and application scenarios.
5 Papers Accepted by NeurIPS 2025
Five of our papers have been accepted by NeurIPS 2025, including 1 Spotlight paper and 3 Poster papers. The accepted works cover topics such as large model alignment, physical reasoning, and safety evaluation.
ACL 2025 Best Paper Award
Our paper "Language Model Resist Alignment" has been awarded the Best Paper of ACL 2025. This work proposes a novel framework to analyze the resistance of large language models to alignment operations, providing new insights for AI safety research.
MedAligner Accepted to The Innovation
Our work "MedAligner" has been accepted by The Innovation (Impact Factor: 32.1). MedAligner is a medical large model alignment tool that improves the safety and accuracy of LMs in clinical application scenarios.
4 Papers in ACL 2025 Main Track
Four of our papers have been accepted by the main track of ACL 2025, covering areas such as multimodal alignment, low-resource NLP, and ethical NLP. This marks our continuous research progress in natural language processing.
SAE-V Accepted by ICML 2025
Our paper "SAE-V: Sparse Autoencoder for Alignment Verification" has been accepted by ICML 2025. This work proposes a sparse autoencoder-based method to verify the effectiveness of large model alignment, reducing the cost of alignment evaluation.