Huanxin Sheng

Email: huanxin [dot] sheng [at] mbzuai [dot] ac [dot] ae

I am currently a STAT&DS PhD studnet at MBZUAI, where I was lucky to be advised by Dr. Jian Kang. I mastered out from my CS PhD career at University of Rochester since Fall 2024, where I was lucky to be advised by Dr. Jian Kang and Dr.Hangfeng He. My research interest lies in trustworthy AI system and model evaluation. Presently I am exploring agent evaluation and on-policy distillation.

News

[Aug 2026] I restart my PhD at MBZUAI!
[Jun 2026] A collaboration paper got accepted by ECCV 2026.
[Mar 2026] I begun my first AI internship at AntGroup as a Research Intern in AQ team.
[Dec 2025] I successfully passed the area exam and earned a MS degree in CS! Many thanks to Prof. Hangfeng He!
[Nov 2025] I was invited by NICE community to give a talk about my EMNLP paper! Here is the recording.
[Nov 2025] My EMNLP paper has officially received the EMNLP 2025 SAC Highlight honor (top 0.6%)! This is my first paper!
[Sep 2025] My EMNLP paper has been accepted for presentation at NeurIPS 2025 LLM Evaluation Workshop! San Diego here I come!
[Aug 2025] My first paper "Analyzing Uncertainty of LLM-as-a-Judge: Interval Evaluations with Conformal Prediction" has been accepted for presentation at the Main Conference by EMNLP 2025!
[Sep 2024] Started my PhD journey at the University of Rochester.

Blog Updates

[Nov 3, 2026] Count the Holdings, Not the Assets: How Much Risk a Constrained Portfolio Hides
[Nov 3, 2026] Looping a Decision Model: One Extra Pass Helps, More Does Not
[Oct 26, 2026] Generative Judges Before Jev, and a Meta-Harness for a 4B Judge
[Oct 24, 2026] What Can a Decision Model Do for Robot Manipulation?
[Oct 23, 2026] Accounting for Compute in a Looped Model
[Jul 20, 2026] On-Policy Distillation: Discussion of Loss Design

All posts →

Publications

Video-Based Reward Modeling for Computer-Use Agents Linxin Song, Jieyu Zhang, Huanxin Sheng, Taiwei Shi, Gupta Rahul, Yang Liu, Ranjay Krishna, Jian Kang, Jieyu Zhao
ECCV, 2026
[PDF] [Code]

1% of Tokens Can Be Enough: On Gradient Estimation in On-Policy Distillation Huanxin Sheng, Zhiling Ye, Haonan Wang, Jian Wang, Jinjie Gu, Jian Kang
arXiv preprint arXiv:2609.24432, 2026
[arXiv] [Code]

Analyzing Uncertainty of LLM-as-a-Judge: Interval Evaluations with Conformal Prediction Huanxin Sheng, Xinyi Liu, Hangfeng He, Jieyu Zhao, Jian Kang
EMNLP (poster, SAC highlights), 2025 | NeurIPS LLM Evaluation Workshop (poster), 2025
[PDF] [Code] [Talk]

Services

Teaching Assistant (@UR) CSC 440 Data Mining (2025 Spring), CSC 442 Artificial Intelligence (2025 Fall)
Reviewer NeurIPS 2026, LOG 2026, LOG 2025; Secondary Reviewer ACL 2025, EMNLP 2026; Workshop WWW2025, KDD2026.

Misc

✋ Before my PhD: I completed my Bachelor of Science degree in Statistics at Xiamen University, where I was lucky to be advised by Dr. Wei Zhong. I was a member of XMU MIREL Lab led by Dr. Yunfeng Wu. I was also a remote research assistant working on high-dimensional portfolio analysis for Dr. Qingliang Fan at the CUHK. The project finally got accepted by JASA.
💬 Languages: My native language is Hokkien (闽南语) .
🎵 Music: I love playing instruments (Guitar, Erhu) and singing.
🐾 Animals: I love cats and dogs, especially Corgis!