Group Equivariant Convolutional Networks

Abstract

We introduce Group equivariant Convolutional Neural Networks (G-CNNs), a natural generalization of convolutional neural networks that reduces sample complexity by exploiting symmetries. G-CNNs use G-convolutions, a new type of layer that enjoys a substantially higher degree of weight sharing than regular convolution layers. G-convolutions increase the expressive capacity of the network without increasing the number of parameters. Group convolution layers are easy to use and can be implemented with negligible computational overhead for discrete groups generated by translations, reflections and rotations. G-CNNs achieve state of the art results on CIFAR10 and rotated MNIST.

References in corpus (5)

• Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs)
• Wide Residual Networks
• Rotation-invariant convolutional neural networks for galaxy morphology prediction
• Exploiting Cyclic Symmetry in Convolutional Neural Networks
• Learning Invariant Representations with Local Transformations

Cited by in corpus (258)

• Shortcut Learning in Deep Neural Networks
• Alias-Free Generative Adversarial Networks
• Informed Machine Learning -- A Taxonomy and Survey of Integrating Knowledge into Learning Systems
• Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges
• On the Generalization of Equivariance and Convolution in Neural Networks to the Action of Compact Groups
• Rotation equivariant vector field networks
• Gauge Equivariant Convolutional Networks and the Icosahedral CNN
• Why do deep convolutional networks generalize so poorly to small image transformations?
• Exploiting Cyclic Symmetry in Convolutional Neural Networks
• Universal Invariant and Equivariant Graph Neural Networks
• Perceiver: General Perception with Iterative Attention
• E(n) Equivariant Graph Neural Networks
• Invariant Representations without Adversarial Training
• Learning Disentangled Representations in the Imaging Domain
• CLIPort: What and Where Pathways for Robotic Manipulation
• Covariant Compositional Networks For Learning Graphs
• Incorporating Symmetry into Deep Dynamics Models for Improved Generalization
• Compositional Generalization in Semantic Parsing: Pre-training vs. Specialized Architectures
• Lorentz Group Equivariant Neural Network for Particle Physics
• Generalizing Convolutional Neural Networks for Equivariance to Lie Groups on Arbitrary Continuous Data
• Adversarial Attack and Defense on Point Sets
• DC3: A learning method for optimization with hard constraints
• A posteriori learning for quasi-geostrophic turbulence parametrization
• Hierarchical, rotation-equivariant neural networks to select structural models of protein complexes
• Machine-learning hidden symmetries
• A Survey of Deep Learning for Scientific Discovery
• CoMIR: Contrastive Multimodal Image Representation for Registration
• ReduNet: A White-box Deep Network from the Principle of Maximizing Rate Reduction
• On Learning Sets of Symmetric Elements
• MDP Homomorphic Networks: Group Symmetries in Reinforcement Learning
• Universal approximations of permutation invariant/equivariant functions by deep neural networks
• Gauge Equivariant Mesh CNNs: Anisotropic convolutions on geometric graphs
• Adversarial Examples - A Complete Characterisation of the Phenomenon
• Deep Scale-spaces: Equivariance Over Scale
• Scale-Equivariant Steerable Networks
• VV-Net: Voxel VAE Net with Group Convolutions for Point Cloud Segmentation
• Transporter Networks: Rearranging the Visual World for Robotic Manipulation
• ReDet: A Rotation-equivariant Detector for Aerial Object Detection
• Using Machine Learning Safely in Automotive Software: An Assessment and Adaption of Software Process Requirements in ISO 26262
• Relational Pooling for Graph Representations
• A General Theory of Equivariant CNNs on Homogeneous Spaces
• LieTransformer: Equivariant self-attention for Lie Groups
• Deformable Kernels: Adapting Effective Receptive Fields for Object Deformation
• Stacked Capsule Autoencoders
• Natural Graph Networks
• CyCNN: A Rotation Invariant CNN using Polar Mapping and Cylindrical Convolution Layers
• Equivariant Transformer Networks
• Scale-covariant and scale-invariant Gaussian derivative networks
• Attentive Group Equivariant Convolutional Networks
• Transformation Consistent Self-ensembling Model for Semi-supervised Medical Image Segmentation
• Data-driven emergence of convolutional structure in neural networks
• Toward Decoding the Relationship between Domain Structure and Functionality in Ferroelectrics via Hidden Latent Variables
• Standardised convolutional filtering for radiomics
• Neural Networks Enforcing Physical Symmetries in Nonlinear Dynamical Lattices: The Case Example of the Ablowitz-Ladik Model
• A method to challenge symmetries in data with self-supervised learning
• Scrambling and decoding the charged quantum information
• Group Convolutional Neural Networks Improve Quantum State Accuracy
• Theoretical Aspects of Group Equivariant Neural Networks
• SE(3)-equivariant prediction of molecular wavefunctions and electronic densities
• A Practical Method for Constructing Equivariant Multilayer Perceptrons for Arbitrary Matrix Groups
• Burst Photography for Learning to Enhance Extremely Dark Images
• Warped Convolutions: Efficient Invariance to Spatial Transformations
• Meta-Learning Symmetries by Reparameterization
• Provably scale-covariant continuous hierarchical networks based on scale-normalized differential expressions coupled in cascade
• Recent advances in deep learning theory
• Constrained Learning with Non-Convex Losses
• Selecting Data Augmentation for Simulating Interventions
• A Wigner-Eckart Theorem for Group Equivariant Convolution Kernels
• Automatic Symmetry Discovery with Lie Algebra Convolutional Network
• Scale-invariant scale-channel networks: Deep networks that generalise to previously unseen scales
• Is novelty predictable?
• Hexagonal Image Processing in the Context of Machine Learning: Conception of a Biologically Inspired Hexagonal Deep Learning Framework
• You Only Hypothesize Once: Point Cloud Registration with Rotation-equivariant Descriptors
• Co-Attentive Equivariant Neural Networks: Focusing Equivariance On Transformations Co-Occurring In Data
• Meta-Learning Stationary Stochastic Process Prediction with Convolutional Neural Processes
• On the Benefits of Invariance in Neural Networks
• Neural Ordinary Differential Equations for Semantic Segmentation of Individual Colon Glands
• Machine Learning Statistical Gravity from Multi-Region Entanglement Entropy
• Set2Graph: Learning Graphs From Sets
• DeepSphere: towards an equivariant graph-based spherical CNN
• A Data and Compute Efficient Design for Limited-Resources Deep Learning
• Rotation-Equivariant Deep Learning for Diffusion MRI
• Counting and Segmenting Sorghum Heads
• Homogeneous vector bundles and $G$-equivariant convolutional neural networks
• On the Universality of Invariant Networks
• Heterogeneous Molecular Graph Neural Networks for Predicting Molecule Properties
• A Simple Proof of the Universality of Invariant/Equivariant Graph Neural Networks
• Temporal-Clustering Invariance in Irregular Healthcare Time Series
• B-Spline CNNs on Lie Groups
• Equivariant Subgraph Aggregation Networks
• On the energy landscape of deep networks
• Orchestrating the Development Lifecycle of Machine Learning-Based IoT Applications: A Taxonomy and Survey
• Ternary Compression for Communication-Efficient Federated Learning
• Universal Equivariant Multilayer Perceptrons
• Covariance in Physics and Convolutional Neural Networks
• Rotation Equivariant Feature Image Pyramid Network for Object Detection in Optical Remote Sensing Imagery
• Geometric and Physical Quantities Improve E(3) Equivariant Message Passing
• Differentiable Physics: A Position Piece
• Adversarial Examples on Object Recognition: A Comprehensive Survey
• Scaling-Translation-Equivariant Networks with Decomposed Convolutional Filters
• Quaternion Equivariant Capsule Networks for 3D Point Clouds
• Coordinate Independent Convolutional Networks -- Isometry and Gauge Equivariant Convolutions on Riemannian Manifolds
• Principled Simplicial Neural Networks for Trajectory Prediction
• ESP: Exploiting Symmetry Prior for Multi-Agent Reinforcement Learning
• Spin-Weighted Spherical CNNs
• Provably Strict Generalisation Benefit for Equivariant Models
• Improving Transformation Invariance in Contrastive Representation Learning
• Group Invariant Dictionary Learning
• Invariance-inducing regularization using worst-case transformations suffices to boost accuracy and spatial robustness
• Some open questions on morphological operators and representations in the deep learning era
• Reducing the dilution: An analysis of the information sensitiveness of capsule network with a practical improvement method
• Building Deep, Equivariant Capsule Networks
• Quantum algorithms for group convolution, cross-correlation, and equivariant transformations
• Group Equivariant Generative Adversarial Networks
• Convex Potential Flows: Universal Probability Distributions with Optimal Transport and Convex Optimization
• DDNet: Cartesian-polar Dual-domain Network for the Joint Optic Disc and Cup Segmentation
• Learning Compositional Structures for Deep Learning: Why Routing-by-agreement is Necessary
• Towards Learning Affine-Invariant Representations via Data-Efficient CNNs
• Shift Invariance Can Reduce Adversarial Robustness
• ChaRRNets: Channel Robust Representation Networks for RF Fingerprinting
• Generalization in Deep RL for TSP Problems via Equivariance and Local Search
• A contribution to Optimal Transport on incomparable spaces
• Orientation-Disentangled Unsupervised Representation Learning for Computational Pathology
• Topographic VAEs learn Equivariant Capsules
• Deep Rotation Equivariant Network
• Neural Epitome Search for Architecture-Agnostic Network Compression
• Visual Representation Learning Does Not Generalize Strongly Within the Same Domain
• Probabilistic Numeric Convolutional Neural Networks
• E(n) Equivariant Normalizing Flows
• Learning Equivariant Energy Based Models with Equivariant Stein Variational Gradient Descent
• Disentangling by Subspace Diffusion
• ${\mathbb{Z}}_2\times {\mathbb{Z}}_2$ Equivariant Quantum Neural Networks: Benchmarking against Classical Neural Networks
• Automated Machine Learning with Monte-Carlo Tree Search
• AVT: Unsupervised Learning of Transformation Equivariant Representations by Autoencoding Variational Transformations
• Trajectory Prediction using Equivariant Continuous Convolution
• Steerable Wavelet Scattering for 3D Atomic Systems with Application to Li-Si Energy Prediction
• Deep learning in bioinformatics: introduction, application, and perspective in big data era
• Geometric Prediction: Moving Beyond Scalars
• Equivariant Learning of Stochastic Fields: Gaussian Processes and Steerable Conditional Neural Processes
• Voint Cloud: Multi-View Point Cloud Representation for 3D Understanding
• A Computationally Efficient Neural Network Invariant to the Action of Symmetry Subgroups
• GraphTER: Unsupervised Learning of Graph Transformation Equivariant Representations via Auto-Encoding Node-wise Transformations
• Improved Generalization Bounds of Group Invariant / Equivariant Deep Networks via Quotient Feature Spaces
• Discrete Rotation Equivariance for Point Cloud Recognition
• Equivariant $Q$ Learning in Spatial Action Spaces
• Model-Based Robust Deep Learning: Generalizing to Natural, Out-of-Distribution Data
• Deep Positional and Relational Feature Learning for Rotation-Invariant Point Cloud Analysis
• Geometric Wavelet Scattering Networks on Compact Riemannian Manifolds
• Jet Single Shot Detection
• Group Equivariant Stand-Alone Self-Attention For Vision
• Inverse reinforcement learning for autonomous navigation via differentiable semantic mapping and planning
• Learnable Gabor modulated complex-valued networks for orientation robustness
• The Convolution Exponential and Generalized Sylvester Flows
• A Cyclically-Trained Adversarial Network for Invariant Representation Learning
• Sampling Equivariant Self-attention Networks for Object Detection in Aerial Images
• Learning Compressed Transforms with Low Displacement Rank
• Isometric Transformation Invariant and Equivariant Graph Convolutional Networks
• Affine Variational Autoencoders: An Efficient Approach for Improving Generalization and Robustness to Distribution Shift
• A simple equivariant machine learning method for dynamics based on scalars
• NeRV: Neural Representations for Videos
• Intelligence plays dice: Stochasticity is essential for machine learning
• Inability of spatial transformations of CNN feature maps to support invariant recognition
• ExplainFix: Explainable Spatially Fixed Deep Networks
• Capsule network with shortcut routing
• Understanding the Generalization Benefit of Model Invariance from a Data Perspective
• Robust Equivariant Imaging: a fully unsupervised framework for learning to image from noisy and partial measurements
• Data augmentation and image understanding
• Grounding inductive biases in natural images:invariance stems from variations in data
• Wasserstein Routed Capsule Networks
• Group Equivariant Deep Reinforcement Learning
• Explainable 3D Convolutional Neural Networks by Learning Temporal Transformations
• Fast and Accurate: Structure Coherence Component for Face Alignment
• Competitive Mirror Descent
• Supervised Whole DAG Causal Discovery
• Contemplating real-world object classification
• Group Equivariant Conditional Neural Processes
• Kähler Geometry of Quiver Varieties and Machine Learning
• Occlusion-Invariant Rotation-Equivariant Semi-Supervised Depth Based Cross-View Gait Pose Estimation
• Rotaflip: A New CNN Layer for Regularization and Rotational Invariance in Medical Images
• Circular-Symmetric Correlation Layer based on FFT
• Symmetry constrained neural networks for detection and localization of damage in metal plates
• RoI Tanh-polar Transformer Network for Face Parsing in the Wild
• Spherical Transformer: Adapting Spherical Signal to CNNs
• Restore from Restored: Single Image Denoising with Pseudo Clean Image
• Transformationally Identical and Invariant Convolutional Neural Networks by Combining Symmetric Operations or Input Vectors
• Scale Equivariant Neural Networks with Morphological Scale-Spaces
• Resampling and super-resolution of hexagonally sampled images using deep learning
• Deformation Robust Roto-Scale-Translation Equivariant CNNs
• Can we have it all? On the Trade-off between Spatial and Adversarial Robustness of Neural Networks
• Using convolutional neural networks for the classification of breast cancer images
• A New Neural Network Architecture Invariant to the Action of Symmetry Subgroups
• Disentangling images with Lie group transformations and sparse coding
• The Devils in the Point Clouds: Studying the Robustness of Point Cloud Convolutions
• Rotation Equivariant Deforestation Segmentation and Driver Classification
• Generating Unrestricted Adversarial Examples via Three Parameters
• Implicit Equivariance in Convolutional Networks
• Euclidean Invariant Recognition of 2D Shapes Using Histograms of Magnitudes of Local Fourier-Mellin Descriptors
• Capacity of Group-invariant Linear Readouts from Equivariant Representations: How Many Objects can be Linearly Classified Under All Possible Views?
• TraceCaps: A Capsule-based Neural Network for Semantic Segmentation
• Equivariant neural networks for inverse problems
• Measure, Manifold, Learning, and Optimization: A Theory Of Neural Networks
• No Representation without Transformation
• Equivariant Deep Dynamical Model for Motion Prediction
• Beyond permutation equivariance in graph networks
• PADDIT: Probabilistic Augmentation of Data using Diffeomorphic Image Transformation
• Neural Networks for Learning Counterfactual G-Invariances from Single Environments
• Equivariant Networks for Pixelized Spheres
• Collaborative Inference for Efficient Remote Monitoring
• Learning Non-Parametric Invariances from Data with Permanent Random Connectomes
• Transformationally Identical and Invariant Convolutional Neural Networks through Symmetric Element Operators
• Invariant Integration in Deep Convolutional Feature Space
• 3D Solid Spherical Bispectrum CNNs for Biomedical Texture Analysis
• Do CNNs Encode Data Augmentations?
• Group equivariant neural posterior estimation
• C-SURE: Shrinkage Estimator and Prototype Classifier for Complex-Valued Deep Learning
• RISA-Net: Rotation-Invariant Structure-Aware Network for Fine-Grained 3D Shape Retrieval
• Finite Group Equivariant Neural Networks for Games
• Invariance-based Multi-Clustering of Latent Space Embeddings for Equivariant Learning
• Learning Augmentation Distributions using Transformed Risk Minimization
• On Universalized Adversarial and Invariant Perturbations
• Rotational Rectification Network: Enabling Pedestrian Detection for Mobile Vision
• Cascade Network with Guided Loss and Hybrid Attention for Two-view Geometry
• On Equivariant and Invariant Learning of Object Landmark Representations
• A de Finetti-type representation of joint hierarchically exchangeable arrays on directed acyclic graphs
• Geometric Data Augmentation Based on Feature Map Ensemble
• Towards glass-box CNNs
• Distribution-Based Invariant Deep Networks for Learning Meta-Features
• Autoequivariant Network Search via Group Decomposition
• Provably Strict Generalisation Benefit for Invariance in Kernel Methods
• Learning Identity-Preserving Transformations on Data Manifolds
• Equivariant Manifold Flows
• Implicit Bias of Linear Equivariant Networks
• Equivariant Contrastive Learning
• Burst Denoising of Dark Images
• ChebLieNet: Invariant Spectral Graph NNs Turned Equivariant by Riemannian Geometry on Lie Groups
• Rotation Equivariant 3D Hand Mesh Generation from a Single RGB Image
• Symmetry-driven graph neural networks
• Equivariance-bridged SO(2)-Invariant Representation Learning using Graph Convolutional Network
• Rotating spiders and reflecting dogs: a class conditional approach to learning data augmentation distributions
• Group Equivariant Subsampling
• Training or Architecture? How to Incorporate Invariance in Neural Networks
• Automatic size and pose homogenization with spatial transformer network to improve and accelerate pediatric segmentation
• Deep Autoencoders: From Understanding to Generalization Guarantees
• Physics-Augmented Learning: A New Paradigm Beyond Physics-Informed Learning
• Unsupervised Learning Framework of Interest Point Via Properties Optimization
• Learning Canonical Transformations
• Intrinsic Image Decomposition using Paradigms
• Learning Equivariant Representations
• More Is More -- Narrowing the Generalization Gap by Adding Classification Heads
• Self-Supervised Adaptation for Video Super-Resolution
• Differential Similarity in Higher Dimensional Spaces: Theory and Applications
• Fast Jacobian-Vector Product for Deep Networks
• Fast Markov Chain Monte Carlo Algorithms via Lie Groups
• Self-Refining Deep Symmetry Enhanced Network for Rain Removal
• Equivariant Point Network for 3D Point Cloud Analysis
• GEM: Group Enhanced Model for Learning Dynamical Control Systems
• Understanding (Non-)Robust Feature Disentanglement and the Relationship Between Low- and High-Dimensional Adversarial Attacks
• Feature Lenses: Plug-and-play Neural Modules for Transformation-Invariant Visual Representations