Hongxin Wei

Assistant Professor, SDS, Southern University of Science and Technology

I am an Assistant Professor (Principal investigator, PhD supervisor) in the Department of Statistics and Data Science at SUSTech (China, Shenzhen). I obtained my Ph.D. degree at the School of Computer Science and Engineering, Nanyang Technological University, supervised by Prof. Bo An. During my Ph.D, I was fortunate to work as a visiting scholar in the group of Prof. Sharon Yixuan Li at the University of Wisconsin Madison in 2022. Previously I spent a wonderful year as a research assistant in the Institute for Interdisciplinary Information Sciences at Tsinghua University. Prior to that, I received my B.E. in Software Engineering from Huazhong University of Science and Technology in 2016.

My research studies reliable LLMs from both statistical and empirical perspectives, aiming for large language models and LLM-driven agents that are aware of what they do not know and accountable for what they produce. Statistically, we develop algorithms that make reliability provable rather than assumed, building on conformal prediction, distribution-free risk control, and the theory of data selection. Empirically, we turn these principles into working systems along the model lifecycle, from the data and post-training stage, through inference and agentic deployment, to auditing what a deployed model actually is.

Open reliability questions across the LLM lifecycle
Data

LLM Auto-labeling

When can a model label be trusted?

Training

LLM Post-training

What supervision is worth learning from?

Inference

Agent Inference

When should an agent doubt itself?

Deployment

Search Agents

Which evidence deserves belief?

Auditing

Model Provenance

Where did this model come from?

LLM Reliability Theory underpins every stage

Turning these questions into guarantees — conformal prediction and selection, distribution-free risk control, and the statistical theory of data selection

Join us — none of these questions is settled, and that is the fun part

Openings for Postdocs, PhD students and RA/interns are available all year round. Tell me which question keeps you curious.

Email me

News

2026-05
Four papers are accepted by ICML 2026. Congratulations to Zhenlong, Sai, and Junxian!
2026-01
We won the second place in the STP Open Challenge.
2025-10
I will be serving as a reviewer for JASA.
2025-09
We will host NeurIPS 2025 Seminar @ Shenzhen during Nov 22-23, Click here to sign up.
2025-09
I will be serving as a Area Chair for ICLR 2026.
2025-05
Three papers are accepted by ICML 2025. Congratulations to Shuoyuan and Hao Zeng!
2025-02
I will be serving as a Area Chair for NeurIPS 2025.
2025-01
Three papers are accepted by ICLR 2025. Congratulations to Wenyu, and Hengxiang!
2025-01
I will be serving as a SPC for IJCAI 2025.
2024-11
I will be serving as a Area Chair for ICML 2025.
2024-09
Two papers are accepted by NeurIPS 2024. Congratulations to Hongfu Gao!
2024-07
We release a Chinese Content Moderation Benchmark for LLMs: ChineseSafe-Benchmark.
2024-05
I am accepting Mphil and PhD applications (2025 fall). I am always looking for highly-motivated research interns, RAs and PostDocs to join our research (refer to this page).
2024-05
Four papers are accepted by ICML 2024.
2024-03
I will be serving as a Area Chair for NeurIPS 2024.
2024-01
Three papers are accepted by ICLR 2024 (Two are Spotlights).
2023-12
We release a Python toolbox for conformal prediction research TorchCP.
2022-10
I am honored to be recognized as Top Reviewers in NeurIPS 2022.
2022-05
Two papers are accepted by ICML 2022 (Accept rate: 21.9%).
2021-10
I am honored to receive NeurIPS 2021 Outstanding Reviewer Award (top 8% of reviewers).

Academic Service

  • Area Chair: NeurIPS, ICML, ICLR
  • Senior PC: IJCAI
  • Conference Reviewer: CVPR, ICCV, KDD, ACL
  • Journal Reviewer: JASA, JMLR, TPAMI, IJCV, ACM Computing Surveys