Shih-Hsin Wang
(王士欣)

About Me

I recently successfully defended my PhD in Mathematics at the University of Utah (advised by Bao Wang and Tommaso de Fernex). My research focuses on developing mathematically rigorous methods that minimize errors, reduce resource waste and cost, and maximize efficiency.

My research bridges rigorous theory and practical application across geometric deep learning and generative models (e.g., flow matching & diffusion models). AI for Science is one of my main goals; I work to translate complex mathematical insights into practical, reliable solutions for molecular and biological scientific applications.
Theory & Knowledge → Foundations
Building foundations through mathematical guarantees and the integration of domain knowledge to ensure reliability and efficiency.
↔
Practical Solution → Application
Translating theoretical foundations into efficient, controllable AI systems that deliver reliable performance in real-world settings.

Research Directions

  • Geometric Deep Learning Equivariant graph neural networks and expressive geometric graph representations.
  • Generative Models Diffusion models, flow matching, single-step models, and test-time steering.
  • AI for Science Modeling complex structures like molecules and proteins.

News & Updates

Mar 2026
Successfully defended my PhD dissertation.
Jan 2026
"RMFlow: Refined Mean Flow by a Noise-Injection Step for Multimodal Generation" accepted to ICLR 2026.
Sep 2025
"Towards Multiscale Graph-based Protein Learning with Geometric Secondary Structural Motifs" accepted to NeurIPS 2025.
May 2025
"Improving Flow Matching by Aligning Flow Divergence" accepted to ICML 2025.
Apr 2025
ICLR 2025 Oral Presentation: Presenting our paper on molecular graph representation. [Slides] [Poster]
Mar 2025
Started visiting UCLA under Andrea Bertozzi, working on flow matching and its applications in RNA/DNA 3D folding.

Publications

Machine Learning

RMFlow: Refined Mean Flow by a Noise-Injection Step for Multimodal Generation
Huang, Y., Wang, S. H., C., Bertozzi, A. L., Wang, B.
ICLR 2026
Towards Multiscale Graph-based Protein Learning with Geometric Secondary Structural Motifs
Wang, S. H., Huang, Y., Transue, T., Baker, J. M., Forstater, J., Strohmer, T., Wang, B.
NeurIPS 2025
Plug-and-Play Image Restoration with Flow Matching: A Continuous Viewpoint
Jia, F., Huang, Y., Wang, S. H., Garcia-Cardona, C., Bertozzi, A. L., Wang, B.
Under Review
Improving Flow Matching by Aligning Flow Divergence
Huang, Y., Transue, T., Wang, S. H., Feldman, W. M., Zhang, H., Wang, B.
ICML 2025
[Slides]
A Theoretically-Principled Sparse, Connected, and Rigid Graph Representation of Molecules
Wang, S. H.*, Huang, Y.*, Baker, J., Sun, Y. E., Tang, Q., Wang, B.
ICLR 2025 [Oral Presentation]
[Slides] [Poster] [Code]
Learning to Control the Smoothness of Graph Convolutional Network Features
Wang, S. H.*, Baker, J.*, Hauck, C. D., Wang, B.
Under Review
An Explicit Frame Construction for Normalizing 3D Point Clouds
Baker, J.*, Wang, S. H.*, de Fernex, T., Wang, B.
ICML 2024
Rethinking the Benefits of Steerable Features in 3D Equivariant Graph Neural Networks
Wang, S. H., Hsu, Y. C., Baker, J., Bertozzi, A. L., Xin, J., Wang, B.
ICLR 2024

Algebraic Geometry

Arcs on Rational Double Points in Arbitrary Characteristic
Wang, S. H., de Fernex, T.
Under Review
Families of Jets of Arc Type and Higher (Co)Dimensional Du Val Singularities
Wang, S. H., de Fernex, T.
C.R. Math. Acad. Sci. Paris, Special Volume in Memory of Jean-Pierre Demailly (2024)

Other Fields

GenFuzz: GPU-Accelerated Hardware Fuzzing Using Genetic Algorithm with Multiple Inputs
Lin, D. L., Zhang, Y., Ren, H., Khailany, B., Wang, S. H., Huang, T. W.
ACM/IEEE Design Automation Conference (DAC), 2023

Highlights & Experience

Recent Experience

  • Visiting Graduate Researcher, UCLA (Mar 2025 – June 2025)
    Initiating a pipeline for 3D RNA/DNA folding from secondary structure using flow-matching models, supervised by Andrea Bertozzi.
  • Research Intern, Los Alamos National Lab (May – Aug 2024)
    Developed a sparse, rigid, and hyperparameter-free graph representation for molecular structures, supervised by Qi Tang.

Invited Talks & Presentations

  • ICLR 2025 – Oral presentation on "A Theoretically-Principled Sparse, Connected, and Rigid Graph Representation of Molecules" (Singapore)
  • JMM 2025 – "Expanding the Mathematical Horizons of Machine Learning"
  • SIAM GL 2023 – "Leveraging Geometric Symmetries with GNNs"
  • NCTS Algebraic Geometry Seminar 2023 – "Families of Jets on Du Val Singularities"

Academic Service

  • Conference Reviewer: ICLR 2025-2026, ICML 2024–26, NeurIPS 2024–25, AISTATS 2025
  • Journal Reviewer: TMLR, SIAM J. on Applied Algebra and Geometry, ACM TOSN