Date：2025/9/17 10:30-11:40
Location：R102, CSIE
Speakers：Shih-Hsin Wang
Host：林軒田教授
Abstract:
Mathematics and engineering together can drive more trustworthy, efficient, and cost-effective innovation in AI. In this talk, I will share how mathematical thinking helps tackle real-world challenges in machine learning. With a background in pure math, I approach deep learning by reformulating complex problems into precise mathematical terms and developing principled solutions. I will highlight how I apply mathematical tools in areas such as geometric deep learning and flow matching, showing how they reveal models’ effectiveness and limitations while guiding improvements or new solutions. Finally, I will discuss how collaboration with mathematics can open new opportunities for advancing AI research and applications.
Biography:
B.S. Math (Honors), NTU → Ph.D. Math, Univ. of Utah
Publications:
- 7 papers in ML: 4 published in ICLR or ICML, 3 under review
- 2 papers in Algebraic Geometry
- 1 paper in other fields