CV

Education

• M.S. in Data Science, École Polytechnique Fédérale de Lausanne (EPFL), 2021-2024
  • Thesis (6.0/6.0): “HyperINF: Scaling-up Accurate Approximation of Influence Function by the Hyperpower Method”
  • Advisor: Prof. Martin Jaggi
  • Selective Courses:
    • Deep Learning (5.75/6.0)
    • Graph Theory (5.5/6.0)
    • Linear Models (5.75/6.0)
    • Time Series (5.25/6.0)
    • Applied Data Analysis (5.5/6.0)
    • Machine Learning (5.25/6.0)
    • Foundations of Data Science (5.5/6.0)
• B.S. in Computer Science and Applied Mathematics, University of Electronic Science and Technology of China, 2017-2021
  • Thesis: “Non-Parametric Bayesian Optimization”
  • Major GPA: 4.0/4.0

Research experiences

• Feb. 2024 - Jun. 2024. Master Thesis (Machine Learning and Optimization Lab – EPFL)
  • We propose HyperINF as an accurate approximation method based on a hyperpower method, i.e. Schulz’s iterative algorithm (which enjoys a rigorous convergence guarantee) and Generalized Fisher Information Matrix (GFIM).
  • We demonstrate the superior accuracy and stability of HyperINF on matrix inversion through a synthetic convergence test.
  • We further validate the efficacy of HyperINF through extensive real-world data attribution problems, including mislabeled data detection, data selection for LLM finetuning, and multimodal instruct-tuning data selection for VLM pretraining.
• Jun. 2023 - Jun. 20244. Remote Research Assistant (HKUST & Mila)
  • We propose LoGAH, with an improved low-rank decoder, that is more scalable and can predict parameters of large networks without copying while having fewer trainable parameters and a lower training cost.
  • We create a new dataset of small ViT and GPT-2 architectures, allowing GHNs to be trained on Transformers for both vision and language domains. LoGAH shows excellent generalized capability on larger models.
  • We outperform GHN-3 as an initialization approach in multiple vision and language tasks by predicting more diverse and performant parameters.
• Oct.2023 - Feb.2024. Research Assistant (NLP Lab – EPFL)
  • Main goal: interpret the multi-modal models including ViLT, CLIP, and BLIP.
  • We try different methods to understand how the image interacts with the text, such as the Second-Gradient, Cross-Attention map,…
• Jul. 2022 – Dec. 2022. Research Assistant (Machine Learning and Optimization Lab – EPFL)
  • We propose a two-stage model SimSum for document-to-document simplification tasks, combining text simplification and summarization tasks innovatively.
  • We analyse and pre-process two document-level simplification datasets, and make the resulting datasets available for reproducibility.
  • Paper was accepted to ACL 2023 main conference.

Industry experiences

• Feb.2023 - Aug.2023. NLP Research Intern (AXA Group Operation Switzerland)
  • Main Task: Assess of large language models and its reasoning capabilities.
  • I Explore prompts for ChatGPT to generate different insurance claims for model’s performance testing.
  • I Deploy two Fake-Text-Detection models (MPU and DetectGPT) on Synthetic Text Detection subtask.

Academic Services

• Reviewer for Data-centric Machine Learning Research (DMLR) Workshop, ICML 2024

Technical Skills

• Programming Languages: Python, C++, MATLAB
• Machine Learning: PyTorch, HuggingFace
• Language Proficiency: GRE: 328, IELTS: 7.5