GuangChen Li

pic.jpg

Ann Arbor, MI

gcli@umich.edu

Hi, I’m GuangChen Li.

I am an M.S. student in Data Science at the University of Michigan, Ann Arbor. Before coming to Michigan, I received my B.S. in Physics from Renmin University of China. My research lies at the intersection of sequential modeling, temporal point processes, and LLM agent systems.

Research

  • Graph-aware monitoring and anomaly detection in LLM agent systems
    I am currently working on a graph-aware monitoring framework for LLM-based agentic systems. The project studies how to use intermediate telemetry signals and terminal outcomes to detect anomalies in multi-step agentic workflows. Our current approach combines graph-structured proxy representations with sequential detection methods such as CUSUM and EWMA, with the goal of providing early warning for workflow instability and performance degradation.

  • Generative adaptation of temporal point processes
    In this project, we proposed GA-TPP, a generative adaptation framework for temporal point processes that enables zero-shot adaptation to unseen conditions. The method combines condition-specific LoRA adapters with a conditional diffusion model, allowing a pretrained TPP to adapt dynamically based on sequence-level side information. We evaluated the framework on multiple real-world datasets and observed strong out-of-distribution improvements over baseline TPP models.

  • Deep learning for particle-in-cell simulation
    My bachelor’s thesis explored replacing the Poisson solver in a 1D particle-in-cell (PIC) framework with neural networks. I built a supervised learning pipeline from phase-space density matrices to grid-level electric fields using data generated by a Fourier-based PIC solver, and studied whether neural surrogates could reproduce plasma two-stream instability while maintaining acceptable physical fidelity.

Publication

  • Generative Adaptation of Temporal Point Processes for Generalizable Event Sequence Prediction and Generation
    Qingmei Wang, GuangChen Li, Yitian Wang*, Tianyu Huang, Junchi Yan, Hongteng Xu.
    Under submission to KDD 2026.

CV

You can download my CV here.