E(n) Equivariant Graph Neural Networks

TL;DR

This paper presents E(n)-Equivariant Graph Neural Networks (EGNNs), a scalable model that maintains equivariance to geometric transformations without complex intermediate representations, outperforming existing methods in various applications.

Contribution

The work introduces a novel, efficient EGNN model that is scalable to higher dimensions and does not rely on costly higher-order features, improving over prior equivariant GNNs.

Findings

Effective in dynamical systems modeling

Improves molecular property prediction

Achieves competitive or better performance than existing methods

Abstract

This paper introduces a new model to learn graph neural networks equivariant to rotations, translations, reflections and permutations called E(n)-Equivariant Graph Neural Networks (EGNNs). In contrast with existing methods, our work does not require computationally expensive higher-order representations in intermediate layers while it still achieves competitive or better performance. In addition, whereas existing methods are limited to equivariance on 3 dimensional spaces, our model is easily scaled to higher-dimensional spaces. We demonstrate the effectiveness of our method on dynamical systems modelling, representation learning in graph autoencoders and predicting molecular properties.