HEP-PHLGHEP-EXDec 25, 2024

Learning Broken Symmetries with Approximate Invariance

arXiv:2412.18773v24 citationsh-index: 93Physical Review D
Originality Incremental advance
AI Analysis

This addresses the challenge of handling broken symmetries in neural network training for domains like particle physics, where detector asymmetries or varying resolutions break ideal symmetries, offering a solution that is incremental in improving upon existing constrained methods.

The paper tackled the problem of broken symmetries in real-world data, where exact symmetries are idealized and standard methods like data augmentation or equivariant networks fail, by proposing a dual-subnet model that balances constrained and unconstrained learning; in a toy example demonstrating Lorentz invariance violation, the model learned as rapidly as symmetry-constrained networks while overcoming their performance limitations.

Recognizing symmetries in data allows for significant boosts in neural network training, which is especially important where training data are limited. In many cases, however, the exact underlying symmetry is present only in an idealized dataset, and is broken in actual data, due to asymmetries in the detector, or varying response resolution as a function of particle momentum. Standard approaches, such as data augmentation or equivariant networks fail to represent the nature of the full, broken symmetry, effectively overconstraining the response of the neural network. We propose a learning model which balances the generality and asymptotic performance of unconstrained networks with the rapid learning of constrained networks. This is achieved through a dual-subnet structure, where one network is constrained by the symmetry and the other is not, along with a learned symmetry factor. In a simplified toy example that demonstrates violation of Lorentz invariance, our model learns as rapidly as symmetry-constrained networks but escapes its performance limitations.

Code Implementations1 repo
Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes