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 (2)

• Deep Convolutional Networks on Graph-Structured Data
• Interaction Networks for Learning about Objects, Relations and Physics

Cited by in corpus (10)

• Understanding and mitigating gradient pathologies in physics-informed neural networks
• Learning Multimodal Graph-to-Graph Translation for Molecular Optimization
• On the Universality of Rotation Equivariant Point Cloud Networks
• On the Universality of Invariant Networks
• Stability and Generalization of Graph Convolutional Neural Networks
• Molecule Property Prediction and Classification with Graph Hypernetworks
• Reducing the dilution: An analysis of the information sensitiveness of capsule network with a practical improvement method
• Steerable Wavelet Scattering for 3D Atomic Systems with Application to Li-Si Energy Prediction
• Human Action Recognition with Multi-Laplacian Graph Convolutional Networks
• Dilated Convolutional Neural Networks for Sequential Manifold-valued Data