Xin-Yang Liu

Focus Areas

My work sits at the overlap of generative AI, computational physics, and scientific machine learning. I am interested in models that do more than fit data: they should encode structure, respect dynamics, and generalize across regimes that matter in practice.

Recent topics include latent diffusion for spatiotemporal turbulence, physics-preserved neural architectures, operator-aware learning for PDE systems, and scalable reinforcement learning for expensive fluid-structure interaction environments.

Selected Publications

1. CoNFiLD: Conditional Neural Field Latent Diffusion Model Generating Spatiotemporal Turbulence
   Arxiv, 2024.
   Efficient turbulence generation with latent diffusion in conditional neural field encoded space.
2. Asynchronous Parallel Reinforcement Learning for Optimizing Propulsive Performance in Fin Ray Control
   Arxiv, 2024.
   Asynchronous parallel training strategy for online reinforcement learning with expensive fluid-structure interaction environments.
3. Multi-resolution Partial Differential Equations Preserved Learning Framework for Spatiotemporal Dynamics
   Communications Physics, 2024.
   Code
4. Predicting 3D Soft Tissue Dynamics from 2D Imaging Using Physics Informed Neural Networks
   Communications Biology, 2023.
5. Physics-informed Dyna-style Model-based Deep Reinforcement Learning for Dynamic Control
   Proceedings of the Royal Society A, 2021.
   Code