About Me
Hi, I’m Hsun-Yu (訓佑). I am currently a master’s student in Computer Science at École Polytechnique Fédérale de Lausanne (EPFL). My research interests lie in understanding why deep learning models can generalise to unseen, out-of-distribution tasks and how to make them more robust and generalisable. In particular, I am fascinated by using Causality to explore these questions.
My recent work focuses on how to better utilise synthetic data—especially that generated by large language models (LLMs)—for model training. I study how to identify and prioritise informative data points using data weighting strategies to boost performance in tasks like text classification. Looking forward, I’m also interested in exploring the impact of data ordering, as well as extending these ideas to broader settings such as interactive environments and self-supervised learning, where understanding what data matters—and when—could further improve learning efficiency and robustness.
Previously, I was advised by Prof. Wei-Yun Ma at Academia Sinica and Prof. Pu-Jen Cheng at National Taiwan University, where I worked on various large language model (LLM) and natural language processing (NLP) topics, including retrieval-augmented generation (RAG), knowledge graphs, and text classification.
🗞️ Recent News
[06/2025] I’m excited to start my research internship with the Robust Machine Learning Group at the Max Planck Institute for Intelligent Systems, Tübingen, under the supervision of Wieland Brendel and Patrik Reizinger.
[02/2025] Our paper, ‘Not All LLM-Generated Data Are Equal: Rethinking Data Weighting in Text Classification,’ has been accepted to ICLR 2025 and selected as spotlight! 🌟