Ph.D. student · Texas State University

I build systems that make machine learning scale.

I am a Computer Science Ph.D. student advised by Dr. Chul-Ho Lee. My work sits at the intersection of machine learning systems, graph neural networks, distributed training, and high-performance computing.

GitHub LinkedIn X Email
Current focus

A scalable temporal graph neural network training framework

Hiding data movement behind useful GPU work

What I am exploring now

Research

Communication, memory systems, and performance optimization for large-scale learning.

Work in progress Current

A Scalable Temporal Graph Neural Network Training Framework

A Planner–Syncer design that prepares data placement and asynchronously multicasts next-batch features while the current forward pass is still running.

  • Hides dynamic-feature transfer latency during otherwise idle NCCL bandwidth windows.
  • Uses P2P transport for high-dimensional features and shared memory for low-dimensional features.
  • Combines caching, zero-copy communication, multiprocessing, and multithreading.
  • PyTorch Distributed
  • DGL
  • NCCL
  • Gloo
  • Shared memory
  • Zero-copy

Selected writing

Publications

Research across segmentation, visual attention, and image quality.

  1. 01
    2025Multimedia Tools and Applications

    A Pavement Crack Segmentation Method Based on Deformable Convolution and Enhanced Perceive Network

    Lei Zhao, Longsheng Wei, Zhi Ma, and Zhiheng Liu.

    Read paper
  2. 02
    2023JACIII

    Single Human Parsing Based on Visual Attention and Feature Enhancement

    Zhi Ma, Lei Zhao, and Longsheng Wei.

    Read paper
  3. 03
    2021Applied Sciences

    Reduced Reference Quality Assessment for Image Retargeting by Earth Mover's Distance

    Longsheng Wei, Lei Zhao, and Jian Peng.

    Read paper

Things I have built

Projects

From bare-metal GPU infrastructure to applied deep learning systems.

A laboratory GPU computing cluster
Infrastructure · 2022–2023

Low-Cost, High-Performance GPU Computing Servers

Built and benchmarked a lab GPU cluster, then configured CUDA, Docker, NFS, remote access, and reusable environments.

CUDADockerNFSLinux
Deep learning · 2022

Real-Time Driver Fatigue Detection

A CNN-based system that fuses EEG and EOG features for real-time fatigue detection.

View on GitHub
Computer vision · 2022–2023

Guangzhou Metro Malfunction Detection

An end-to-end detection service using YOLOv5, Fast R-CNN, Flask, and Docker Compose.

YOLOv5FlaskDocker

The path so far

Experience & background

Research, engineering, and a long-running habit of building useful things.

Education

  1. 2024—Present Texas State University

    Ph.D. Student in Computer Science

  2. 2021—2024 China University of Geosciences

    M.Eng. in Electronic and Information

  3. 2008—2012 Hubei University of Technology

    B.Eng. in Electrical Engineering and Automation

Industry

  1. 2016—2020 Wuhan Stack Tech Co., Ltd.

    Founder and Developer

  2. 2012—2016 Ericsson

    Radio Access Network Engineer

  3. 2010—2011 HGDonline.net Coding Club

    Web Developer

Toolbox

Systems thinking, from model to metal.

PyTorchCUDAC/C++PythonDDPDeepSpeedLinuxDockerSlurmNetwork analysis

Let us connect

Interested in ML systems, GNNs, or distributed training?

I am always happy to talk about research ideas, systems problems, and collaborations.

Email me