Ziheng Qin

About Me

Hi, I'm Ziheng Qin, a PhD student in Computer Science at the National University of Singapore (NUS). My research interests include deep learning efficiency, focusing on data, optimizers, and large-scale training. I have a strong background in computer science and extensive experience in both academic and industry settings.

Education

• National University of Singapore (NUS), Singapore - PhD student in Computer Science, 2021 - Present
• University of Southern California (USC), CA, USA - Master in Computer Science, 2020 - 2021
• University of Michigan, Ann Arbor (UM), MI, USA - B.S.E in Computer Science, 2017 - 2019
• Shanghai Jiao Tong University (SJTU), Shanghai, China - B.S.E in Electronic and Computer Engineering (ECE), 2015 - 2019

Experience

• ECCV 2024: Dataset Growth (2023-2024)
  • Efficient online algorithm for data cleaning and selection.
  • Scalable and efficient for both single-modal and multi-modal tasks.
• ICLR 2024 Oral Presentation (2022-2023)
  • Developed InfoBatch for lossless training speed up via unbiased dynamic data pruning.
  • Save 20% ~ 40% overall cost on classification, segmentation, detection, image generation and instruction tuning tasks.
• ICCV 2023 (2022-2023)
  • Preventing zero-shot transfer degradation in continual learning of vision-language models
  • We proposed a novel method ZSCL to prevent zero-shot transfer degradation in the continual learning of vision-language models in both feature and parameter space. We proposed a more challenging Multi-domain Task Incremental Learning (MTIL) benchmark to evaluate different methods. ZSCL outperforms other methods in the traditional class-incremental learning setting and the MTIL by 9.7% average score
• AAAI 2023 Outstanding Paper (2021-2022)
  • Contributed to reducing CTR prediction model training time from 12 hours to 10 minutes on 1 GPU.
• Google LLC, California, USA - Summer Intern (2021)
  • Developed VM-based distributed training solution.
• Shanghai Jiao Tong University & Intel, Shanghai, China - Project Member (2019)
  • Developed edge cloud gaming architecture using StarlingX and Kubernetes.
• University of Michigan - Student Research Project Leader (2018-2019)
  • Developed table-embeddings for predicting table column names from content.

Skills

• Programming Languages: Python, C/C++, Java
• Platform: Linux
• Development: Machine Learning, Reinforcement Learning, Deep Learning, Data Mining, Distributed Training
• Other: MySQL, Spark, XGBoost, CSS, JavaScript, HTML, Hadoop, MATLAB, Mathematica, PPT, LaTeX

Contact

Email: e0823059@u.nus.edu

LinkedIn: Ziheng Qin

GitHub: henryqin1997

Google Scholar: Ziheng Qin