Zirui He

About Me

I am a second-year Ph.D. student in Data Science at the New Jersey Institute of Technology (NJIT), advised by Dr. Mengnan Du. My current research focuses on interpretability for large language models (LLMs). Before joining NJIT, I received my M.S. degree in Computer Science from Stevens Institute of Technology and my B.S. degree in Computer Science from China University of Geosciences, Wuhan.

Research Interests

Explainable AI, Natural Language Processing, Interpretability, Reasoning and Memorization

News

• [Feb. 2026] We released a new preprint, Rep2Text: Decoding Full Text from a Single LLM Token Representation.
• [Nov. 2025] I passed my qualify exam and became a PhD Candidate.
• [Oct. 2025] Invited Talk @NJIT Data Sciences Seminar: Representation Space of LLMs.
• [Sep. 2025] We released a new preprint, MemLens: Uncovering Memorization in LLMs with Activation Trajectories.
• [Aug. 2025] Our paper is accepted to EMNLP 2025.
• [Feb. 2025] We released a new preprint, Saif: A sparse autoencoder framework for interpreting and steering instruction following of language models.
• [May. 2024] One paper is accepted to COLING 2024.

Publications

Teaching

• Deep Learning (DS677), Spring 2026, NJIT, Teaching Assistant
• Machine Learning (DS675), Fall 2025, NJIT, Teaching Assistant
• Natural Language Processing (DS680), Spring 2025, NJIT, Teaching Assistant

Services

Conference Reviewers

• International Conference on Artificial Intelligence and Statistics (AISTATS) 2026
• International Conference on Machine Learning (ICML) 2026

Journal Reviewers

• Applied AI Letters (AIL)

Powered by Jekyll and Minimal Light theme.