CV
Contents are generated from _data/cv.yml. The PDF version is available above.
Basics
| Name | Zhonglin Xie |
| Label | Researcher |
| xzlmath@gmail.com | |
| Url | https://zhonglinxie.github.io |
| Summary | Researcher at ByteDance Seed. Ph.D. in Computational Mathematics from Peking University, working on reinforcement learning for large language models, optimization algorithm design from the ODE perspective, and learning to optimize. |
Education
-
2021.09 - 2026.06 Beijing, China
Ph.D.
Beijing International Center for Mathematical Research, Peking University
Computational Mathematics
- Member of The Elite Ph.D. Program in Applied Mathematics
- Advisor: Prof. Zaiwen Wen
-
2018.09 - 2021.07 Beijing, China
-
2017.09 - 2021.07 Beijing, China
B.Sc.
School of Mathematical Sciences, Peking University
Computational Mathematics
- Member of The Elite Program of Applied Mathematics and Statistics for Undergraduates
Work
-
2026.07 - Present Beijing, China
-
2020.08 - 2020.10 Beijing, China
Awards
- 2024
Presidential Scholarship
Peking University
- 2023
Third-Class Scholarship
Peking University
- 2023
Presidential Scholarship
Peking University
- 2022
Ubiquant Scholarship
Peking University
Publications
-
2026 DRAW: Domain Weight Randomization with Bayesian Updating for LLM Pre-Training
Transactions on Machine Learning Research (TMLR)
Randomizes pre-training domain weights and refines them by Bayesian updating.
-
2026 EdgeBench: Unveiling Scaling Laws of Learning from Real-World Environments
arXiv preprint, arXiv:2607.05155
A benchmark for studying the scaling laws of learning from real-world environments.
-
2025 ODE-based Learning to Optimize
Mathematical Programming
A novel learning to optimize framework based on the ODE viewpoint of optimization algorithms.
-
2025 OptMATH: A Scalable Bidirectional Data Synthesis Framework for Optimization Modeling
International Conference on Machine Learning (ICML)
A scalable bidirectional data synthesis framework for optimization modeling.
-
2025 Accelerated Natural Gradient Method for Parametric Manifold Optimization
arXiv preprint, arXiv:2504.05753
An accelerated natural gradient method for optimization over parametric manifolds.
-
2025 Accelerating Optimization via Differentiable Stopping Time
Advances in Neural Information Processing Systems (NeurIPS), Spotlight
Accelerates optimization algorithms by making the stopping time differentiable.
-
2023 LRSDP: Low-rank SDP for triple patterning lithography layout decomposition
ACM/IEEE Design Automation Conference (DAC)
Applies the SDPDAL solver to patterning lithography layout decomposition problem.
-
2021 Joint bandwidth allocation and path selection in WANs with path cardinality constraints
Journal of Communications and Information Networks
Proposes an algorithm based on ADMM that jointly optimizes the lantency and fairness under the resource constraint.