☆ 4.8 Article

Machine Learning Hidden Symmetries

PHYSICAL REVIEW LETTERS (2022)

Enforcing Analytic Constraints in Neural Networks Emulating Physical Systems

Tom Beucler et al.

Summary: This study introduces a systematic way to enforce nonlinear analytic constraints in neural networks through constraints in the architecture or the loss function, which improves the accuracy of neural networks in modeling nonlinear physical systems, particularly in climate modeling applications.

PHYSICAL REVIEW LETTERS (2021)

添加到收藏夹

Article Physics, Fluids & Plasmas

Machine-learning nonconservative dynamics for new-physics detection

Ziming Liu et al.

Summary: This paper presents a method for data-driven new physics discovery by using neural networks to decompose the force field into conservative and nonconservative components. The new method successfully rediscovers friction, Neptune, and gravitational waves, and outperforms traditional neural networks in predicting the future of damped double pendulums.

PHYSICAL REVIEW E (2021)

添加到收藏夹

Article Astronomy & Astrophysics

Sampling using SU(N) gauge equivariant flows

Denis Boyda et al.

Summary: A flow-based sampling algorithm has been developed for SU(N) lattice gauge theories, ensuring gauge invariance by construction. This algorithm constructs flows on an SU(N) variable (or on a U(N) variable through a simple alternative) that respects matrix conjugation symmetry, and has been applied to sample distributions of single SU(N) variables and to construct flow-based samplers for SU(2) and SU(3) lattice gauge theory in two dimensions.

PHYSICAL REVIEW D (2021)

添加到收藏夹

Article Multidisciplinary Sciences