M.S. in Computer Science, University of Illinois Urbana-Champaign
Advised by Prof. Fan Lai
I am an M.S. student in Computer Science at UIUC, working with Prof. Fan Lai on machine learning systems. I am broadly interested in how large models are trained and served under real-world constraints — compute, latency, memory, and energy — and in building the infrastructure that makes intelligence cheaper to deploy, from datacenter GPUs down to heterogeneous edge and embedded devices. Most recently, I have been working on fast serving of robot policies — making vision-language-action and diffusion policies cheap enough for real-time, multi-robot control.
My undergraduate background combines computer science and economics, which shapes how I approach systems problems: performance is a budget question as much as an engineering one, and the interesting solutions are the ones that move the whole cost–quality frontier rather than a single benchmark number.
Before UIUC I studied Computer Science at Sichuan University and transferred in 2024. Outside research I enjoy teaching — currently as a TA for CS 421.
Reflex: Fast Robot Policy Serving by Reusing and Correcting Past Actions
A backbone-free serving layer for robot policies that caches complete action chunks from past executions, with action-level residual correction and progress-guided policy refresh — 1.4×–11× lower latency and 2.0×–2.2× higher multi-robot throughput across five VLA, world-action, and diffusion policies and a real SO-101 arm, without modifying the policy.
Dora: QoE-Aware Hybrid Parallelism for Distributed Edge AI
Hybrid pipeline parallelism for LLM training across heterogeneous edge devices: a DP-based Pareto-optimal model partitioner under latency and energy constraints, plus a network-aware wavefront scheduler with online re-planning — 1.1×–6.3× end-to-end improvement over the state of the art.