About Me

I am a Senior Algorithm Researcher at Horizon Robotics, focusing on World Models, Visual Tokenizers, and Large-scale Training. Previously, I worked at Tencent AI Lab and Tencent Hunyuan, contributing to LLM Alignment, Game AI, and Multimodal Learning. I received my M.S. from Hangzhou Dianzi University in 2024, advised by Prof. Yuyu Yin.

Research Interests

Large Language Models LLM Alignment RLHF / DPO World Models Visual Tokenizers Multimodal Learning Diffusion Models Reinforcement Learning

News

  • 📝 [2026.03] One paper submitted to ECCV 2026.
  • 🎉 [2026.01] One paper accepted to ICLR 2026.
  • 🎉 [2025.03] One paper accepted to ACL Findings 2025.
  • 🔥 [2025.01] SimPER adopted by LG AI Research as core training algorithm for EXAONE Deep.
  • 🎉 [2025.01] Two papers accepted to ICLR 2025.
  • 🎉 [2024.12] One paper accepted to NeurIPS 2024.
  • 🎉 [2024.10] One paper accepted to EMNLP 2024 Main.
  • 🎓 [2024.06] Graduated from Hangzhou Dianzi University with Outstanding Graduate of Zhejiang Province.

Selected Publications

ICLR 2025

SimPER: A Minimalist Approach to Preference Alignment without Hyperparameters

Tianhao Xiao, Yiyang Yuan, Zhuoan Chen, Mingxiao Li, Shuang Liang, Zongqiang Ren, Vasant G. Honavar
ICLR 2025
Paper
ICLR 2025

On a Connection Between Imitation Learning and RLHF

Tianhao Xiao, Yiyang Yuan, Mingxiao Li, Zhuoan Chen, Vasant G. Honavar
ICLR 2025
Paper
EMNLP 2024

How to Leverage Demonstration Data in Alignment for Large Language Model? A Self-Imitation Learning Perspective

Tianhao Xiao, Mingxiao Li*, Yiyang Yuan, Huazheng Zhu, Chengyu Cui, Vasant G. Honavar (*Equal contribution)
EMNLP 2024 Main
Paper
NeurIPS 2024

Cal-DPO: Calibrated Direct Preference Optimization for Language Model Alignment

Tianhao Xiao, Yiyang Yuan, Huazheng Zhu, Mingxiao Li, Vasant G. Honavar
NeurIPS 2024
Paper
ACL 2025

Advancing General Multimodal Capability of Vision-language Models with Pyramid-descent Visual Position Encoding

Zhuoan Chen, Mingxiao Li, Zhiwei Chen, Nan Du, Xiang Li, Yifei Zou
ACL Findings 2025
Paper
ECCV 2026

DINO-Tok: Adapting DINO for Visual Tokenizers

Mingjie Jia*, Mingxiao Li*, Lei Fan, Tianyu Shi, Jiajun Guo, Zongqing Li, Xiaoyan Guo, Xuxu Long, Qiang Zhang, Ping Tan, Wenguang Yin (*Equal contribution)
ECCV 2026 (Submitted)
Paper Code
ICLR 2026

RLVMR: Reinforcement Learning with Verifiable Meta-Reasoning Rewards for Robust Long-Horizon Agents

Zhe Zhang, Zhuoan Chen, Mingxiao Li, Zhaopeng Tu, Xiang Li
ICLR 2026
Paper
arXiv

VISTA: Enhancing Vision-Text Alignment in MLLMs via Cross-Modal Mutual Information Maximization

Mingxiao Li, Na Su, Fei Qu, Zhiwei Zhong, Zhuoan Chen, Yiming Li, Zong-Pei Tu, Xiang Li
arXiv preprint
Paper Code
arXiv

VDEP: Establishing Equivalence Between Image and Text Token Through Autoregressive Pre-training in MLLMs

Mingxiao Li, Fei Qu, Zhuoan Chen, Na Su, Zhiwei Zhong, Zhiwei Chen, Nan Du, Xiang Li
arXiv preprint
Paper

Experience

  • Jul 2025 - Present

    Horizon Robotics

    Senior Algorithm Researcher | World Model Team
    Working with Wei Yin. Core contributor to EponaV2 World Model and DINOTok visual tokenizer.
  • Dec 2024 - Jul 2025

    Tencent Hunyuan Digital Human

    Algorithm Researcher
    Working with Zhaopeng Tu. Core contributor to FPS Game AI Bot project, deployed in production.
  • Oct 2022 - Dec 2024

    Tencent AI Lab

    Algorithm Researcher (Intern → Full-time)
    Working with Nan Du and Ziyang Chen. Worked on LLM Alignment, Game AI Commentary, and CodeLLM. Team awarded 2022 Tencent Business Breakthrough Award.

Education

  • Sep 2021 - Jun 2024

    Hangzhou Dianzi University

    M.S. in Computer Science | Outstanding Graduate of Zhejiang Province
    Advised by Prof. Yuyu Yin.
  • Sep 2017 - Jun 2021

    Hainan University

    B.S. in Computer Science

Honors & Awards

  • [2024] Outstanding Graduate of Zhejiang Province
  • [2022] Tencent Business Breakthrough Award
  • [2020] First Prize, Pan-Pearl River Delta Computer Works Competition (Hainan Province)
  • [2020] First Prize, AI Competition Hainan Division
  • [2020] Second Prize, Network Technology Challenge (South China Region)