Lekai Chen

Research Assistant | PhD Candidate
Boulder, Colorado, USA | lekai.chen@colorado.edu | GitHub

_{Last update: September 26, 2025}

Education

BEng in Computer Science and Technology, Chongqing University — CGPA 3.7 / 4.0 (senior years)
June 2023
Master of Science in Computer Science, University of Colorado Boulder
December 2025
PhD in Computer Science, University of Colorado Boulder
June 2028 (expected)

Reinforcement Learning R&D Engineer, SenseTime — Beijing, China
Sep 2021 – June 2023
- Design multi-agent reinforcement learning models and build scalable RL pipelines for real-world customers’ problems.
- Actively involved in DI-STAR and GoBigger projects; implement a large-scale framework for distributed reinforcement learning.
- Design and conduct RL experiments and compose research papers.
- Tags: Reinforcement Learning, MARL, Algorithm Design

RL-based game AI for the most complex Zerg vs. Zerg setting; solution outperformed AlphaStar in comparisons.

Designed new features and encoders to accelerate training and improve supervised-phase performance.
Streamlined the learning pipeline and model; implemented LR warmup to enable single-PC training.
Leveraged league learning; designed expert-simulation models to diversify the league and improve robustness. Tags: Multi-Agent RL, Game AI, Pipeline, Optimization

Platform to study swarm intelligence across multiple teams in an Agar-like game.

Designed reward shaping that encourages agents to clone and feed teammates.
Implemented multiple baseline single-agent and MARL algorithms trainable on a single GPU despite long trajectories.
Showed some MARL baselines do not improve collaboration in complex environments. Tags: Swarm Intelligence, MARL, Baselines, Simulation

Ming Zhang, Shenghan Zhang, Zhenjie Yang, Lekai Chen, Jinliang Zheng, Chao Yang, Chuming Li, Hang Zhou, Yazhe Niu, Yu Liu. GoBigger: A Scalable Platform For Multi-Agent Interactive Simulation. ICLR 2023. Code · Paper
Lekai Chen, Ashutosh Trivedi, Alvaro Velasquez. LLMs as Probabilistic Minimally Adequate Teachers. Under review.
Lekai Chen, Alvaro Velasquez, Ashutosh Trivedi. CEDAR: A Counter-Example Driven Agent with Regular Restriction. Under review.