I am a Master's student at UPenn (Applied Math & Computational Science) and a graduate of Bowdoin College (Math & Economics), originally from Shanghai, China.
"Scale is hitting diminishing returns. The next breakthrough lies in efficiency, biological inspiration, and emergent behavior."
I believe the future of AI will not be defined by monolithic models, but by the efficient coordination of specialized agent populations. My research operationalizes biological dynamics to create systems that are not just smarter, but drastically more efficient and adaptable than scaling alone allows.
Distributed Combinatorial Optimization for Generative Agents
To realize this vision, I treat Multi-Agent Systems not as a prompt engineering task, but as a Combinatorial Assignment Problem. The goal is to optimize the bipartite matching between N agents and R tasks without central control.
Instead of manual prompting or standard RL, I develop population-based metaheuristics (specifically, competitive exclusion dynamics) to search the discrete strategy space of Large Language Models. This approach functions as a decentralized solver, empirically achieving near-optimal resource allocation—effectively solving the Maximum Weight Matching problem in a gradient-free setting.
Key Results:
- 🚀 Efficiency: Converges to the theoretical performance ceiling (Oracle Accuracy).
- 🧠 Emergence: Specialization arises from local competition, not central design.
- 📉 Resource Economy: Achieves these results with 99% less memory than standard MARL baselines.
