Coordinate Independent Convolutional Networks -- Isometry and Gauge Equivariant Convolutions on Riemannian Manifolds

Abstract

Motivated by the vast success of deep convolutional networks, there is a great interest in generalizing convolutions to non-Euclidean manifolds. A major complication in comparison to flat spaces is that it is unclear in which alignment a convolution kernel should be applied on a manifold. The underlying reason for this ambiguity is that general manifolds do not come with a canonical choice of reference frames (gauge). Kernels and features therefore have to be expressed relative to arbitrary coordinates. We argue that the particular choice of coordinatization should not affect a network's inference -- it should be coordinate independent. A simultaneous demand for coordinate independence and weight sharing is shown to result in a requirement on the network to be equivariant under local gauge transformations (changes of local reference frames). The ambiguity of reference frames depends thereby on the G-structure of the manifold, such that the necessary level of gauge equivariance is prescribed by the corresponding structure group G. Coordinate independent convolutions are proven to be equivariant w.r.t. those isometries that are symmetries of the G-structure. The resulting theory is formulated in a coordinate free fashion in terms of fiber bundles. To exemplify the design of coordinate independent convolutions, we implement a convolutional network on the Möbius strip. The generality of our differential geometric formulation of convolutional networks is demonstrated by an extensive literature review which explains a large number of Euclidean CNNs, spherical CNNs and CNNs on general surfaces as specific instances of coordinate independent convolutions.

arXiv commentThe implementation of orientation independent Möbius convolutions is publicly available at https://github.com/mauriceweiler/MobiusCNNs

References in corpus (61)

• Semi-Supervised Classification with Graph Convolutional Networks
• Geometric deep learning: going beyond Euclidean data
• Spectral Networks and Locally Connected Networks on Graphs
• MeshCNN: A Network with an Edge
• Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges
• Group Equivariant Convolutional Networks
• Tensor field networks: Rotation- and translation-equivariant neural networks for 3D point clouds
• Making Convolutional Networks Shift-Invariant Again
• Rigid-Motion Scattering for Texture Classification
• Rotation equivariant vector field networks
• DeepSphere: Efficient spherical Convolutional Neural Network with HEALPix sampling for cosmological applications
• Why do deep convolutional networks generalize so poorly to small image transformations?
• Exploiting Cyclic Symmetry in Convolutional Neural Networks
• Nonlocal and quasi-local field theories
• Learning Spherical Convolution for Fast Features from 360° Imagery
• A Riemannian Network for SPD Matrix Learning
• Equivariance Through Parameter-Sharing
• Hyperspherical Variational Auto-Encoders
• General $E(2)$-Equivariant Steerable CNNs
• Group Equivariant Capsule Networks
• Incorporating Symmetry into Deep Dynamics Models for Improved Generalization
• Hyperbolic Graph Neural Networks
• 3D G-CNNs for Pulmonary Nodule Detection
• Lorentz Group Equivariant Neural Network for Particle Physics
• Generalizing Convolutional Neural Networks for Equivariance to Lie Groups on Arbitrary Continuous Data
• Spherical CNNs on Unstructured Grids
• N-body Networks: a Covariant Hierarchical Neural Network Architecture for Learning Atomic Potentials
• PDE-based Group Equivariant Convolutional Neural Networks
• Explorations in Homeomorphic Variational Auto-Encoding
• Hyperbolic Attention Networks
• Relevance of Rotationally Equivariant Convolutions for Predicting Molecular Properties
• Building Deep Networks on Grassmann Manifolds
• Universal approximations of permutation invariant/equivariant functions by deep neural networks
• MDP Homomorphic Networks: Group Symmetries in Reinforcement Learning
• Fanaroff-Riley classification of radio galaxies using group-equivariant convolutional neural networks
• Scale equivariance in CNNs with vector fields
• Scale Steerable Filters for Locally Scale-Invariant Convolutional Neural Networks
• LieTransformer: Equivariant self-attention for Lie Groups
• Intertwiners between Induced Representations (with Applications to the Theory of Equivariant Neural Networks)
• How should spin-weighted spherical functions be defined?
• Attentive Group Equivariant Convolutional Networks
• Theoretical Aspects of Group Equivariant Neural Networks
• ZerNet: Convolutional Neural Networks on Arbitrary Surfaces via Zernike Local Tangent Space Estimation
• On the Universality of Rotation Equivariant Point Cloud Networks
• Einstein's vierbein field theory of curved space
• Parallel Transport Convolution: A New Tool for Convolutional Neural Networks on Manifolds
• ManifoldNet: A Deep Network Framework for Manifold-valued Data
• 3D-Rotation-Equivariant Quaternion Neural Networks
• Universal Equivariant Multilayer Perceptrons
• Hyperbolic Deep Neural Networks: A Survey
• Quaternion Equivariant Capsule Networks for 3D Point Clouds
• Spin-Weighted Spherical CNNs
• Chart Auto-Encoders for Manifold Structured Data
• Universal Approximation Theorem for Equivariant Maps by Group CNNs
• Probabilistic Numeric Convolutional Neural Networks
• Disentangling by Subspace Diffusion
• Topological Constraints on Homeomorphic Auto-Encoding
• Gauge Theories and Fiber Bundles: Definitions, Pictures, and Results
• DV-ConvNet: Fully Convolutional Deep Learning on Point Clouds with Dynamic Voxelization and 3D Group Convolution
• Isometric Transformation Invariant Graph-based Deep Neural Network
• Group Equivariant Deep Reinforcement Learning

Cited by in corpus (6)

• PDE-based Group Equivariant Convolutional Neural Networks
• Gauge Equivariant Mesh CNNs: Anisotropic convolutions on geometric graphs
• Steerable Partial Differential Operators for Equivariant Neural Networks
• A simple equivariant machine learning method for dynamics based on scalars
• Capacity of Group-invariant Linear Readouts from Equivariant Representations: How Many Objects can be Linearly Classified Under All Possible Views?
• ChebLieNet: Invariant Spectral Graph NNs Turned Equivariant by Riemannian Geometry on Lie Groups