Zhonglin Xie

prof_pic.jpg

Welcome to my personal homepage! I am a researcher at ByteDance Seed. I received my Ph.D. in June 2026 from the Beijing International Center for Mathematical Research, Peking University, under the guidance of Prof. Zaiwen Wen, where I was a member of the Elite Ph.D. Program in Applied Mathematics.

My work centers around Reinforcement Learning for LLMs: developing more efficient RL algorithms and improving post-training accuracy to enhance model performance; Theoretical Analysis and Algorithm Design: leveraging Ordinary Differential Equations (ODEs) to analyze and design optimization algorithms; and Learning to Optimize (L2O): exploring how machine learning techniques can significantly improve algorithm performance for specific problem classes, particularly in industrial settings.

I received my bachelor’s degree in 2021 from the School of Mathematical Sciences at Peking University, majoring in computational mathematics. During my undergraduate studies, I was also a member of the Elite Program of Applied Mathematics and Statistics for Undergraduates. Additionally, I earned a second bachelor’s degree in economics from the National School of Development between 2018 and 2021.

selected publications

* denotes equal contribution
  1. OptMATH: A Scalable Bidirectional Data Synthesis Framework for Optimization Modeling
    Hongliang Lu*Zhonglin Xie*, Yaoyu Wu, Can Ren, Yuxuan Chen, and Zaiwen Wen
    Forty-Second International Conference on Machine Learning (ICML), 2025
  2. ODE-based Learning to Optimize
    Zhonglin XieWotao Yin, and Zaiwen Wen
    Mathematical Programming, 2025

news

Jul 06, 2026 “EdgeBench: Unveiling Scaling Laws of Learning from Real-World Environments” is now available on arXiv.
Jun 30, 2026 I defended my Ph.D. thesis and graduated from BICMR, Peking University. I have joined ByteDance Seed as a researcher.
Apr 09, 2026 Our paper “DRAW: Domain Weight Randomization with Bayesian Updating for LLM Pre-Training” has been accepted by TMLR.
Nov 24, 2025 “ODE-based Learning to Optimize” is now published in Mathematical Programming.
Sep 18, 2025 Our paper “Accelerating Optimization via Differentiable Stopping Time” has been accepted as a spotlight at NeurIPS 2025!
Jun 04, 2025 I will present “OptMATH: A Scalable Bidirectional Data Synthesis Framework for Optimization Modeling” at ICCOPT 2025 in Los Angeles, University of Southern California (pending US visa approval).
Jun 04, 2025 Our paper “OptMATH: A Scalable Bidirectional Data Synthesis Framework for Optimization Modeling” has been accepted as a poster presentation at ICML 2025! I will attend the conference (pending Canadian visa approval). Find us at Vancouver!
Apr 03, 2025 I will give two talks titled ‘‘OptMATH: A Scalable Bidirectional Data Synthesis Framework for Optimization Modeling ‘’ and ‘‘Accelerating Optimization via Differentiable Stopping Time’’ at MOS2025.

latest posts