Covariant Compositional Networks For Learning Graphs

Abstract

Most existing neural networks for learning graphs address permutation invariance by conceiving of the network as a message passing scheme, where each node sums the feature vectors coming from its neighbors. We argue that this imposes a limitation on their representation power, and instead propose a new general architecture for representing objects consisting of a hierarchy of parts, which we call Covariant Compositional Networks (CCNs). Here, covariance means that the activation of each neuron must transform in a specific way under permutations, similarly to steerability in CNNs. We achieve covariance by making each activation transform according to a tensor representation of the permutation group, and derive the corresponding tensor aggregation rules that each neuron must implement. Experiments show that CCNs can outperform competing methods on standard graph learning benchmarks.

References in corpus (8)

• Molecular Graph Convolutions: Moving Beyond Fingerprints
• Quantum-Chemical Insights from Deep Tensor Neural Networks
• Deep Convolutional Networks on Graph-Structured Data
• Learning Convolutional Neural Networks for Graphs
• Interaction Networks for Learning about Objects, Relations and Physics
• Group Equivariant Convolutional Networks
• Gated Graph Sequence Neural Networks
• The Multiscale Laplacian Graph Kernel

Cited by in corpus (54)

• Improving Graph Neural Network Expressivity via Subgraph Isomorphism Counting
• 3D Steerable CNNs: Learning Rotationally Equivariant Features in Volumetric Data
• Understanding and mitigating gradient pathologies in physics-informed neural networks
• Graph Neural Networks: Taxonomy, Advances and Trends
• Learning Multimodal Graph-to-Graph Translation for Molecular Optimization
• Universal Invariant and Equivariant Graph Neural Networks
• Supervised Community Detection with Line Graph Neural Networks
• E(n) Equivariant Graph Neural Networks
• Distance Encoding: Design Provably More Powerful Neural Networks for Graph Representation Learning
• Graph Capsule Convolutional Neural Networks
• A Survey on The Expressive Power of Graph Neural Networks
• Self-learning Monte Carlo with Deep Neural Networks
• 3D G-CNNs for Pulmonary Nodule Detection
• N-body Networks: a Covariant Hierarchical Neural Network Architecture for Learning Atomic Potentials
• On Learning Sets of Symmetric Elements
• Learning metrics for persistence-based summaries and applications for graph classification
• Universal approximations of permutation invariant/equivariant functions by deep neural networks
• A General Theory of Equivariant CNNs on Homogeneous Spaces
• Natural Graph Networks
• Intertwiners between Induced Representations (with Applications to the Theory of Equivariant Neural Networks)
• Hierarchical Graph-to-Graph Translation for Molecules
• SE(3)-equivariant prediction of molecular wavefunctions and electronic densities
• Optimal Transport Graph Neural Networks
• On the Universality of Rotation Equivariant Point Cloud Networks
• Building powerful and equivariant graph neural networks with structural message-passing
• Provably Powerful Graph Networks
• Set2Graph: Learning Graphs From Sets
• On the Universality of Invariant Networks
• Equivariant Subgraph Aggregation Networks
• Multiresolution Graph Transformers and Wavelet Positional Encoding for Learning Hierarchical Structures
• Stability and Generalization of Graph Convolutional Neural Networks
• Universal Equivariant Multilayer Perceptrons
• Global Attention Improves Graph Networks Generalization
• Frame Averaging for Invariant and Equivariant Network Design
• Spin-Weighted Spherical CNNs
• Directional Graph Networks
• A smooth basis for atomistic machine learning
• Reducing the dilution: An analysis of the information sensitiveness of capsule network with a practical improvement method
• Graph Convolutional Neural Networks via Motif-based Attention
• Molecule Property Prediction and Classification with Graph Hypernetworks
• Geometric Scattering for Graph Data Analysis
• Equivariant Maps for Hierarchical Structures
• Steerable Wavelet Scattering for 3D Atomic Systems with Application to Li-Si Energy Prediction
• Equivariant Entity-Relationship Networks
• Human Action Recognition with Multi-Laplacian Graph Convolutional Networks
• Geometric Wavelet Scattering Networks on Compact Riemannian Manifolds
• Physics-Constrained Predictive Molecular Latent Space Discovery with Graph Scattering Variational Autoencoder
• Dilated Convolutional Neural Networks for Sequential Manifold-valued Data
• Rotation-Invariant Autoencoders for Signals on Spheres
• Message Passing Graph Kernels
• Discriminative Learning of Similarity and Group Equivariant Representations
• Equivariant and Invariant Reynolds Networks
• Equivariant Networks for Pixelized Spheres
• Capacity of Group-invariant Linear Readouts from Equivariant Representations: How Many Objects can be Linearly Classified Under All Possible Views?