About Me

Hi, I’m Zida Wu! I am a Ph.D. candidate in the Department of Electrical and Computer Engineering at UCLA, where I am advised by Prof. Ankur Mehta. Prior to joining UCLA, I earned my Master’s degree from Shanghai Jiao Tong University and my Bachelor’s degree from Xidian University. 

My research focuses on scalable decision-making and estimation in large-scale multi-agent systems, at the intersection of reinforcement learning (RL), game theory, and robotics. I aim to develop learning-based algorithms with rigorous theoretical guarantees that can operate reliably in complex, real-world environments involving hundreds or thousands of interacting agents.

My current work includes three main directions:

1. Learning Nash Equilibria in Large-Scale Multi-Agent Systems

I developed RL methods to solve large-population decision-making problems modeled as Mean Field Games. The RL-based approach enables the learning of Nash equilibrium (NE) policies without the need to solve high-dimensional partial differential equations (PDEs). My work facilitates scalable NE learning from arbitrary initial distributions and is adaptable to continuous environments by integrating population-aware policy learning, online mirror descent, and generative modeling of agent dynamics.

2. Curvature-Aware Reinforcement Learning for Multi-Agent Coverage

I developed learning-based approaches to continuous-space coverage and coordination in robotics. My research proposes a curvature-aware imitation-to-reinforcement learning framework. By combining submodular optimization with policy learning, this framework enables near-optimal multi-agent coverage strategies with theoretical guarantees in continuous environments, notably obviating the need for greedy searches via environment discretization.

3. Decentralized State and Input Estimation in Sensor Networks

I designed decentralized estimation algorithms for distributed sensing and robotic systems operating under communication and uncertainty constraints. This work includes joint state and unknown-input estimation, as well as scalable Kalman filtering methods that remain robust despite intermittent communication and heterogeneous sensing capabilities.

Industrial Impact

At TikTok (ByteDance), I deployed discovery-based recommendation models that increased GMV by 1.45% in production. 

At Tencent RoboticsX Lab, I developed localization systems for data center robots, achieving sub-millimeter localization accuracy.

At Agency for Science, Technology and Research (A*STAR) of Singapore, I investigated a SLAM-based docking method for dynamically moving objects in a manufacturing plant.

Email: zdwu[at]ucla[dot]edu

GitHub; LinkedIn; Google Scholar