CV

Education

Ph.D. in Computer Science Engineering, University of Texas at Arlington, 2017-2023
- Focus: Deep Learning Frameworks, GPU Computing
M.E. in Information, Production, and Systems, Waseda University, 2014-2016
- Focus: 3D Integrated Circuit Design
B.E. in Electronic Science and Engineering, Southeast University, 2011-2016
- Focus: Electronic Science

2023-Present: Assistant Professor
- Faculty of Data Science, City University of Macau
- Research focus: Deep Learning Systems, Performance Optimization
- Location: Macau, China

2021: Software Engineer Intern
- ProtagoLabs (Remote)
- Developed stable elastic distributed deep learning system for large-scale NLP model training
2020: Software Engineering Intern
- ByteDance Machine Learning System Lab, Seattle
- Integrated TVM graph runtime into mobile applications
- Optimized model inference performance using TVM
2015: Research Intern
- Hitachi Research Laboratory, Japan
- Data analysis for Ropits autonomous vehicle
- Safety scope determination through vehicle speed and tire rotation analysis
2014: Research Intern
- Institute of Electronics, Chinese Academy of Sciences, Beijing
- Implemented GPU-accelerated target detection algorithm

Large Model Training AI Infrastructure (2024-Present)
- Working with Nvidia Megatron-LM
- Developing AI infrastructure based on domestic Ascend 910 series
Brain-inspired Deep Learning (2023-Present)
- Algorithm optimization for Spiking Neural Networks in CNN model accuracy
Deep Learning Systems (2018-Present)
- Atom System for Large Language Model Training
  - Elastic fault-tolerant system for asynchronous training in decentralized environments
  - Memory swapping and sub-model preloading for improved resource utilization
- Data Loader and Data Reuse Acceleration
  - Novel optimization method for data reuse without accuracy degradation
- Deep Learning Compilation
  - TVM integration and automatic optimization for various models
- SwitchFlow Framework
  - Preemptive multitasking framework for deep learning
  - Best Paper Award at Middleware 2021 (1/107)
JVM HotSpot Heap Management (2017-2018)
- Side-channel attack research on parallel scavenge
- Investigation of time-stretching effects on heap size adjustment

2024: Guangdong Institute of Intelligent Science and Technology - FTBC: Forward Temporal Bias Correction
2024: Pengcheng Laboratory - Decentralized Distributed Deep Learning Training Framework
2023: University of Macau - Bridging the Resource Scheduling Gap in Deep Learning System
2023: Zhejiang Lab - Decentralized Distributed Deep Learning Training Framework
2023: Microsoft Research DeepSpeed Team
2021: Middleware’21 Conference - SwitchFlow: Preemptive Multitasking for Deep Learning
2018: HotCloud’18 Conference - HotSpot Side-channel Attack

Programming Languages: C/C++, Python, CUDA, Java
Frameworks & Tools: TensorFlow, PyTorch, Huggingface NLP, Keras, TVM, Docker, Ray, Bazel, CMake, gRPC
Hardware Development:
- Second Prize in National College Student Electronic Design Competition (2013)
- Second Prize in Southeast University Embedded System Design Competition (2014)
- Book Chapter: “National College Student Electronic Design Competition Excellent Works Collection” Book Chapter