About me

Hello there!

I am a PhD candidate at University of Bristol, supervised by Song Liu. I am interested in teaching machines to uncover the intrinsic nature of the world. I am particularly fascinated by the idea that, in order to generate complex patterns, a model must first learn something fundamental about how those patterns are formed.

My PhD research focuses on generative modeling [paper] and its application, including:

• Simulation-Based Inference [paper]
• Representation Learning [paper]
• Hypothesis Testing [paper]

Publications & Preprints

* denotes equal contribution; a full list can be found at my google scholar.

• (Preprint) Zero-Flow Two-Sample Tests. Wang, Y., Wang,L., Liu, S. & Suzuki, T.

• (ICML2026) Zero-Flow Encoders. Wang, Y.*, Wang,L.*, Liu, S. & Suzuki, T.

• (NeurIPS2025) Direct Fisher Score Estimation for Likelihood Maximization. Khoo, S., Wang, Y., Liu, S. & Beaumont, M.
  • Spotlight, Top 3%
• (UAI2025) Guiding Time-Varying Generative Models with Natural Gradients on Exponential Family Manifold. Liu, S., Wang, L. & Wang, Y.
  • Outstanding long paper, at Deep Generative Model in Machine Learning: Theory, Principle and Efficacy workshop at ICLR2025.