Seminar, Hosted by MIT Center for Computational Science and Engineering (CCSE) Distinguished Seminar
Generative sampling of high-dimensional probability distributions for lattice field theory studies of the structure of matter
MIT Department of Physics
Abstract
Lattice quantum field theory is a demanding setting for physics-informed machine learning: algorithms must sample extremely high-dimensional distributions at scale while respecting exact mathematical and physical constraints. Phiala Shanahan describes recent advances in generative modelling that emerged from the challenge of exact sampling from known probability distributions in lattice calculations for particle and nuclear physics, focusing on flow-based generative models, the importance of guarantees of exactness, and how complex symmetries such as gauge symmetry can be built directly into model architectures. She connects these developments to broader questions in scalable sampling, structure-preserving computation and scientific machine learning.
Topics
- Lattice quantum field theory, broader topic: quantum field theory
- generative modelling
- flow-based generative models
- exact sampling, broader topic: sampling
- high-dimensional probability distributions, broader topic: probability distributions
- gauge symmetry
- physics-informed machine learning, broader topic: machine learning
- structure-preserving computation, search World Wide