Jiyan He

Jiyan He

Jiyan He

researcher

Research Interests

Large ModelGenerative AIAgent Systems
14
Publications
0
Invited Talks

About

Jiyan He is a faculty member at Zhongguancun Academy (ZGCA) and a researcher at the Zhongguancun Institute of Artificial Intelligence (ZGCI). His main research interests include large language models, agents, and AI safety. He received both his bachelor's and Ph.D. degrees from the University of Science and Technology of China (USTC) and previously conducted research on machine learning and scientific intelligence at Microsoft Research Asia. Dr. He has published multiple papers in top international journals and conferences such as Nature Machine Intelligence, ICLR, ICML, and NeurIPS. His research spans the fundamental theory of artificial intelligence, system design, and interdisciplinary applications. He formerly served as CEO of the Linux User Group and the Microsoft Student Club at USTC, and has received several honors including the National Scholarship and the Guo Yonghuai Scholarship. He has also won championships in multiple international and domestic algorithm, supercomputing, and cybersecurity competitions. His work focuses on large language models, generative AI, AI for Science, and privacy and security, aiming to advance the theoretical foundations and algorithmic design of next-generation AI models and agents. Related work: Google Scholar

Education

PhD in Cyberspace Security

School of Cyberspace Security, University of Science and Technology of China

Bachelor of Computer Science

School of Computer Science, University of Science and Technology of China

Publications

Predicting equilibrium distributions for molecular systems with deep learning

Nature Machine Intelligence2024140 citations

Benchmarking graphormer on large-scale molecular modeling datasets

arXiv preprint202288 citations

Exploring the limits of differentially private deep learning with group-wise clipping

ICLR 2023202258 citations

Aqualora: Toward white-box protection for customized stable diffusion models via watermark lora

ICML 2024202431 citations

Control risk for potential misuse of artificial intelligence in science

arXiv preprint202311 citations

Catch you everything everywhere: Guarding textual inversion via concept watermarking

arXiv preprint202311 citations

Masked molecule modeling: A new paradigm of molecular representation learning for chemistry understanding

arXiv preprint20227 citations

Physical Consistency Bridges Heterogeneous Data in Molecular Multi-Task Learning

NeurIPS 202420242 citations

RedChronos: A Large Language Model-Based Log Analysis System for Insider Threat Detection in Enterprises

arXiv preprint20251 citations

Sparse Autoencoders Reveal Interpretable Structure in Small Gene Language Models

arXiv preprint20250 citations

No Free Lunch: Rethinking Internal Feedback for LLM Reasoning

arXiv preprint20250 citations

Modeling Earth-Scale Human-Like Societies with One Billion Agents

arXiv preprint20250 citations

Towards Generalist Prompting for Large Language Models by Mental Models

arXiv preprint20240 citations

Tracing the Misuse of Personalized Textual Embeddings for Text-to-Image Models

ICLR 2025 Workshop on Navigating and Addressing Data Problems for Foundation Models20250 citations