I am a Ph.D. student in Statistics at UC Irvine, advised by Prof. Annie Qu and Prof. Rui Miao. I received my B.S. in Mathematics and Applied Mathematics from Jilin University in 2024.
I work on reinforcement learning and multi-task optimization: training one model on many tasks without the hardest ones falling behind, from small control policies to VLA models and LLMs. My NeurIPS 2026 paper, TOPPO, shows that in multi-task PPO the critic’s gradients are badly imbalanced across tasks, and balances them. I now study how tasks compete inside deep networks. More on my research.
News
- Received the Graduate Travel Award from the UC Irvine Department of Statistics.
- Released the TOPPO code and training logs.
- TOPPO was accepted at NeurIPS 2026 as a poster.
- TOPPO ranked #8 in Robotics on alphaXiv for May.
- Posted the TOPPO preprint on arXiv.
- Started my Ph.D. in Statistics at UC Irvine.
Outside research
Robots. I am learning robot learning hands-on: training locomotion policies for a Unitree Go2 quadruped in Isaac Lab and checking them sim-to-sim in MuJoCo, the last step before the real robot, and experimenting with humanoid motion tracking in simulation. I expect robotics to become part of my research.
Linux, networks and agent harnesses. I run my own small compute setup of headless Linux machines on a private network, and I enjoy building the harness around coding agents: skills, MCP servers, and the permission checks that let an agent use a shared cluster safely. Most of my projects started there.