Zexue He

I'm a Postdoc Fellow at Human-Centered AI Institude (HAI) in Stanford University, working with Alex 'Sandy' Pentland , Yejin Choi , and Erik Brynjolfsson. I recieved my Ph.D. in Computer Science and Engineering department at University of California, San Diego, advised by Prof. Julian McAuley, and was honored to be awarded the IBM PhD Fellowship in 2022.

I previously spent two years as a Research Scientist at the MIT-IBM Lab , responsible for the safety and alignment training of IBM's multimodal foundation models (Granite Vision Series). I also worked as an intern at the Microsoft Research and Google.

Contact: zexueh [AT] stanford.edu | zehe [AT] ucsd.edu
Full Publications: Google Scholar

My research advances human-centered Large Language Models and Vision-Language Models that can remember, reason, and act reliably in complex and failure-prone decision-making settings:

• When models must be safe and aligned with humans: I work on LLM safety and alignment, building loyal agents and personalized AI systems that reason under uncertainty while respecting diverse human preferences.
• When the modeling horizon is long: I develop memory-augmented LLMs and agentic architectures that support persistent state tracking, structured long-context reasoning, and multi-session decision-making.
• When applications are high-stakes: : I design and evaluate the reliability and societal impact LLMs for human-centered domains such as healthcare, low-resource settings, education, and employment.

• [3/2026]: I will present at Gates Foundation Workshop on Future of Personal AI: Global Portable Memory Workshop in Seattle.
• [2/2026]: I'm co-organizing the first workshop on Emerging Directions in Data for Multimodal Foundation Models at CVPR 2026.
• [10/2025]: I'm organizing the first workshop on Memory and Vision at ICCV 2025.
• [7/2025]: I'm organizing the second workshop on Long-Context Foundation Models at ICML 2025.
• [4/2025]: Invited panlist on New Frontiers in Associative Memories Workshop at ICLR 2025.
• [4/2025]: Invited talk on Safe AI and NLP for High-Stakes Human-Centric Tasks at Yale.
• [3/2025]: Invited talk on Trustworthy AI for High-Stakes Human-Centric Tasks at UMiami.
• [1/2025]: Invited talk on Responsible AI for High-Stakes Human-Centric Domains at Tufts.
• [1/2025]: Invited talk on Responsible and Trustworthy AI at UCLA.
• [11/2024] : Invited talk on Responsible AI for Human-centered Challenges at MIT.
• [10/2024]: Talk on Trustworthy GenAI at UCSD CSE.