Master of Application Informatics
2021 - 2024
University of Göttingen, Germany
Master Thesis:
Improving the portability and interoperability of deep learning workloads using ONNX. Developed a supervised NLP classification model with the attention mechanism to automate the AI tasks in the GWDG ticket system. With a well-trained model, we cyclically update the model by retraining and inference with multiple programming languages on different devices. A cross-programming languages and devices framework has been established for federated learning and on-device learning with ONNX.
Research Intern
2023.10 - 2024.05
- Implemented a ResNet series deep learning framework using Golang with GPU utilization, optimizing the environment to eliminate dependency issues for execution within GWDG's cluster. Explored performance comparisons between this implementation and Python, as well as distributed learning implementations for this approach.
- Evaluating high-performance computing (HPC) benchmarks, including IO500, HPL, HPCG, and STREAM, to assess system performance. Developing and implementing the MiniBUDE benchmark across multiple parallel programming models—OpenMP, Julia, CUDA, OpenACC, and OpenCL—on both CPU and GPU architectures.
- I have gained practical experience in high-performance computing (HPC) through Linux system operations, cluster management, and parallel computing using MPI and CUDA. Designed and implemented a distributed learning system from the ground up utilizing Golang in combination with MPI for inter-process communication.