Covariant Compositional Networks For Learning Graphs

Risi Kondor, Hy Truong Son, Horace Pan, Brandon Anderson, Shubhendu Trivedi
1/7/2018
cs.LG

Official code: HyTruongSon/GraphFlow (56 stars, GPL-3.0, last commit 2021-09-13).

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.

View on arXivView PDF

Code Implementations(1)

HyTruongSon/GraphFlowOfficial100%
GPL-3.0

Deep Learning framework in C++/CUDA that supports symbolic/automatic differentiation, dynamic computation graphs, tensor/matrix operations accelerated by GPU and implementations of various state-of-the-art graph neural networks and other Machine Learning models including Covariant Compositional Networks For Learning Graphs [Risi et al]

5612C++, CudaDec 25, 20175 years agoGPL-3.0
computation-graphdeep-learning-framework

Similar Papers

Invariant Layers for Graphs with Nodes of Different Types

85% match2023

E(n) Equivariant Graph Neural Networks

83% match1 repo5502021

Approximately Equivariant Graph Networks

83% match1 repo22023

Graph Metanetworks for Processing Diverse Neural Architectures

83% match2023

P-tensors: a General Formalism for Constructing Higher Order Message Passing Networks

82% match2023

Geom-GCN: Geometric Graph Convolutional Networks

82% match1 repo3112020

Cite this paper

@article{kondor2018covariant,
  title  = {Covariant Compositional Networks For Learning Graphs},
  author = {Risi Kondor and Hy Truong Son and Horace Pan and Brandon Anderson and Shubhendu Trivedi},
  year   = {2018},
  eprint = {1801.02144},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1801.02144}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1801.02144.svg)](https://codeofpaper.com/paper/1801.02144)

Discussion