My research operates at the bidirectional frontier of Theoretical Physics and Artificial Intelligence. I develop AI architectures rooted in physical laws to solve fundamental problems in quantum field theory, while simultaneously using statistical mechanics to decode the “black box” of deep learning.

Generative Models for Lattice Field Theories

We design physics-conditioned generative models that respect the symmetries of quantum field theory. By establishing a formal bridge between AI and physics, we enable high-precision sampling that was previously computationally prohibitive.

• Diffusion as Stochastic Quantization: We proved that Diffusion Models are mathematically equivalent to Stochastic Quantization in field theory, providing a rigorous foundation for AI-based sampling.
• Neural Path Integrals: Developed Fourier-flow models and autoregressive networks to generate Feynman paths, capturing complex phase transitions (e.g., Kosterlitz–Thouless).

• Q Zhu, G Aarts, W Wang, K Zhou, L Wang*, Physics-Conditioned Diffusion Models for Lattice Gauge Theory, J. High Energ. Phys. 2026, 060 (2026).
• L. Wang, G. Aarts, and K. Zhou, Diffusion models as stochastic quantization in lattice field theory, J. High Energ. Phys. 2024, 060 (2024).
• S Chen, O Savchuk, S Zheng, B Chen, H Stoecker, L Wang*, K Zhou, Fourier-flow model generating Feynman paths, Phys. Rev. D 107, 056001 (2023).
• L Wang, Y Jiang, L He, K Zhou, Continuous-mixture autoregressive networks learning the kosterlitz-thouless transition, Chinese Physics Letters 39 (12), 120502 (2022).

Inverse Problems in Physics

Inverse problems in physics (like spectral reconstruction) are notoriously ill-posed. We utilize Automatic Differentiation (AD) and physics-driven neural networks to regularize these problems and extract precise physical information.

• AD-based Reconstruction: Pioneered the use of AD to reconstruct spectral functions and Neutron Star Equations of State (EoS) from lattice/observational data.
• Hadron Physics: Leveraging deep learning to decode hadron-hadron interactions and emitting sources, bridging the gap between theory and experiment.

• G. Aarts et al., Physics-Driven Learning for Inverse Problems in Quantum Chromodynamics, Nat. Rev. Phys. 7, 154 (2025).
• L. Wang, S. Shi, and K. Zhou, Reconstructing spectral functions via automatic differentiation, Phys. Rev. D 106, L051502 (2022).
• L. Wang and J. Zhao, Learning hadron emitting sources with deep neural networks, Commun. Phys. 9, 253 (2026).

AI for Complex Systems

We generalize the methodologies developed in theoretical physics to tackle broader scientific challenges, from climate patterns to epidemiological dynamics.

• Atmospheric Science: Developed all-weather atmospheric retrieval models and remote sensing frameworks that outperform traditional statistical methods.
• Epidemiology & Spatiotemporal Modeling: Applying Physics-Informed Neural Networks (PINNs) to model the spreading of diseases with spatial constraints.

• H Xiao, F Zhang, L Wang, B Pan, Y Zhu, M Wang, W Li, B Guo, J Li, High-resolution ensemble retrieval of cloud properties for all-day based on geostationary satellite, npj Climate and Atmospheric Science 8, no. 1 (2025): 386.
• L Wang, T Xu, H Stoecker, H Stoecker, Y Jiang, K Zhou, Machine learning spatio-temporal epidemiological model to evaluate Germany-county-level COVID-19 risk, Machine Learning: Science and Technology 2, no. 3 (2021): 035031.

Others

*Corresponding author