Abstract:Symmetry is present throughout nature and continues to play an increasingly central role in physics and machine learning. Fundamental symmetries, such as Poincaré invariance, allow physical laws discovered in laboratories on Earth to be extrapolated to the farthest reaches of the universe. Symmetry is essential to achieving this extrapolatory power in machine learning applications. For example, translation invariance in image classification allows models with fewer parameters, such as convolutional neural networks, to be trained on smaller data sets and achieve state-of-the-art performance. In this paper, we provide a unifying theoretical and methodological framework for incorporating symmetry into machine learning models in three ways: 1. enforcing known symmetry when training a model; 2. discovering unknown symmetries of a given model or data set; and 3. promoting symmetry during training by learning a model that breaks symmetries within a user-specified group of candidates when there is sufficient evidence in the data. We show that these tasks can be cast within a common mathematical framework whose central object is the Lie derivative associated with fiber-linear Lie group actions on vector bundles. We extend and unify several existing results by showing that enforcing and discovering symmetry are linear-algebraic tasks that are dual with respect to the bilinear structure of the Lie derivative. We also propose a novel way to promote symmetry by introducing a class of convex regularization functions based on the Lie derivative and nuclear norm relaxation to penalize symmetry breaking during training of machine learning models. We explain how these ideas can be applied to a wide range of machine learning models including basis function regression, dynamical systems discovery, neural networks, and neural operators acting on fields.

Symmetry-aware Graph Matching.

Coupled Structure-from-motion and 3D Symmetry Detection for Urban Facades

3D Symmetry Detection by a Single Image and Geometric Transformation

Multiresolution Rotational Symmetry Detection Via Radius-Based Frieze-Expansion.

Symmetry-based 3-D Reconstruction from Perspective Images

Analysis on Image Symmetry Detection Algorithms

Skeleton-Intrinsic Symmetrization of Shapes

Large-Baseline Matching And Reconstruction From Symmetry Cells

Symmetrical Singular Value Decomposition Representation for Pattern Recognition

Symmetry Strikes Back: From Single-Image Symmetry Detection to 3D Generation

On Symmetry and Multiple-View Geometry: Structure, Pose, and Calibration from a Single Image

Symmetry Discovery Beyond Affine Transformations

A Unified Framework to Enforce, Discover, and Promote Symmetry in Machine Learning

Symmetry Weight-sharing for Patch-based Stereo Matching

Beyond Planar Symmetry: Modeling Human Perception of Reflection and Rotation Symmetries in the Wild

Finding Mirror Symmetry via Registration

2017 ICCV Challenge: Detecting Symmetry in the Wild.

Reflection symmetry detection using locally affine invariant edge correspondence.

Estimating True Symmetry In Scale-Space

Symmetry Feature In Content-Based Image Retrieval

Symmetry and asymmetry analysis and its implications to computer-aided diagnosis: A review of the literature