Yingru LI

Yingru LI

Member of Technical Staff

xAI

About me

I am a Member of Technical Staff at xAI. I earned my Ph.D. in Computer Science in 2025 from The Chinese University of Hong Kong (CUHK), where I had the privilege of being advised by Prof. Zhi-Quan (Tom) Luo, with Prof. Benjamin Van Roy on my thesis committee.

Prior to xAI, I was a Research Scientist at ByteDance. I also had the valuable opportunity to collaborate with Prof. Tong Zhang and Prof. John Hopcroft.

Research Vision

My research aims to develop intelligent agents capable of reliably interacting with complex environments. By bridging foundational theory with scalable algorithms, I advance reinforcement learning, optimization, and large language model (LLM) reasoning to create systems for trustworthy decision-making.

📄 View my CV/resumé     🐦 Follow me on X for updates.

Interests
  • Reinforcement Learning
  • Large-Scale Optimization
  • LLM Reasoning and Agents
Education

  • Ph.D. in Computer Science, 2025

    The Chinese University of Hong Kong

  • M.S. & B.E. with Honors, 2020

    Huazhong University of Science and Technology

Recent Posts

The Optimal Token Baseline
RL training for LLMs frequently suffers from training collapse due to exploding gradient variance in long-horizon tasks. We derive the Optimal Token Baseline (OTB) from first principles, proving that updates should be weighted inversely to their accumulated uncertainty (Realized Energy). Our computationally free Logit-Gradient Proxy eliminates training collapse, matches N=32 performance with just N=4, and reduces token consumption by 62-66%.
Yingru LI, Jiawei Xu, Ziniu Li, Jiacai Liu, Yuxuan Tong, Wei Liu, Longtao Zheng, Zhenghai Xue, Yaxiang Zhang, Tianle Cai, Ge Zhang, Qian Liu, Baoxiang Wang
Dec 20, 2025 1 min read Research, Theory
The Stability Gap: Why Top-K Routing Breaks RL Optimization
A rigorous mathematical analysis showing that Top-K expert routing in Mixture of Experts creates two fundamental pathologies: gradient blackout (zero gradients almost everywhere) and first-order approximation failure (discontinuous policy mapping), explaining why MoE-RL training can be unstable.
Yingru LI
Dec 7, 2025 11 min read Research, Theory
Scalable Exploration via Ensemble++
Ensemble++ achieves Thompson Sampling-level exploration with only O(d log T) ensemble directions, enabling scalable uncertainty quantification for neural bandits and beyond.
Yingru LI
Nov 29, 2025 4 min read Research
Language as a Universal Interface for Reinforcement Learning Agents
This post establishes a formal mathematical framework for language agents, deriving fundamental challenges from first principles and providing concrete design guidelines with real-world examples from SWE-Bench.
Yingru LI
Nov 7, 2025 22 min read Research, Theory, Engineering
Mathematical Formulations of Rollout Correction Methods
Definitive mathematical formulations for rollout correction methods in VeRL, progressing from REINFORCE to PPO to Decoupled PPO. Handles policy mismatch, temporal lag, replay buffers, and off-policy algorithms with importance sampling and rejection sampling techniques.
Yingru LI
Nov 4, 2025 1 min read Research, Theory, Documentation
See all posts

Contact

szrlee [at] gmail [dot] com