Miao Lu

About Me Welcome to my homepage! I am a third-year Ph.D. candidate in Operations Research from the Department of Management Science and Engineering (MS&E) at Stanford University, advised by Prof. Jose Blanchet. Prior to that, I obtained my Bachelor's degree in Probability and Statistics from the University of Science and Technology of China (USTC). Research Interests My research centers around reinforcement learning (RL). Recently, I am excited by the power of RL in advancing LLM agents. With backgrounds in probability and statistics, my past research includes mathematical theory and algorithm design of provably sample-efficient RL, training dynamics and generalization properties in deep learning loss landscapes, and RL for operations and economics. Contact Information Email: miaolu at stanford dot edu [Google Scholar] [GitHub] [LinkedIn] [CV]

Industrial Experiences

ByteDance Seed, San Jose, USA. Jun. 2025 - Sep.2025

• Research Scientist Intern in Foundations Models

Ubiquant Investment, Shanghai, China. Jun. 2022 - Sep. 2022

• Quantitative Research Intern in the AI Department

Selected Publications

• Scaling Multi-Turn LLM RL via End-to-End Summarization-based Context Management
  with Weiwei Sun, Weihua Du, Zhan Ling, Xuesong Yao, Kang Liu, Jiecao Chen
  Submitted, Sep, 2025
  

• Provably Mitigating Overoptimization in RLHF: Your SFT Loss is Implicitly an Adversarial Regularizer
  with Zhihan Liu, Shenao Zhang, Boyi Liu, Hongyi Guo, Yingxiang Yang, Jose Blanchet, Zhaoran Wang
  Neural Information Processing Systems (NeurIPS) 2024
  

• Benign Oscillation of Stochastic Gradient Descent with Large Learning Rates
  with Beining Wu, Xiaodong Yang, Difan Zou
  International Conference on Learning Representations (ICLR) 2024
  

• Maximize to Explore: One Objective Function Fusing Estimation, Planning, and Exploration
  with Zhihan Liu, Wei Xiong, Han Zhong, Hao Hu, Shenao Zhang, Sirui Zheng, Zhuoran Yang, Zhaoran Wang
  Neural Information Processing Systems (NeurIPS) 2023
  

• Double Pessimism is Provably Efficient for Distributionally Robust Offline Reinforcement Learning: Generic Algorithm and Robust Partial Coverage
  with Jose Blanchet, Tong Zhang, Han Zhong
  Neural Information Processing Systems (NeurIPS) 2023