Physics-Driven Generative Modeling: Scalable and Interpretable Solutions for Lattice Systems in Theoretical Physics
The goal of this project is to develop scalable and robust generative samplers capable of providing unbiased estimations of physical and thermodynamic observables. Current state-of-the-art generative models struggle to match the scalability of MCMC simulations for lattice field theory (LFT) with scaling to large lattice sizes remaining a significant challenge. To address this, we propose renormalization group (RG)-based approaches where sampling is performed hierarchically, utilizing generative AI to enhance scalability. By overcoming these challenges, the project seeks to establish generative modeling as a practical and transferable tool for studying large-scale lattice systems.
Research Junior Group Lead
Research Junior Group Lead
Postdoctoral Researcher