English

Spacetime $E(n)$-Transformer: Equivariant Attention for Spatio-temporal Graphs

Machine Learning 2024-08-13 v1 Artificial Intelligence

Abstract

We introduce an E(n)E(n)-equivariant Transformer architecture for spatio-temporal graph data. By imposing rotation, translation, and permutation equivariance inductive biases in both space and time, we show that the Spacetime E(n)E(n)-Transformer (SET) outperforms purely spatial and temporal models without symmetry-preserving properties. We benchmark SET against said models on the charged NN-body problem, a simple physical system with complex dynamics. While existing spatio-temporal graph neural networks focus on sequential modeling, we empirically demonstrate that leveraging underlying domain symmetries yields considerable improvements for modeling dynamical systems on graphs.

Keywords

Cite

@article{arxiv.2408.06039,
  title  = {Spacetime $E(n)$-Transformer: Equivariant Attention for Spatio-temporal Graphs},
  author = {Sergio G. Charles},
  journal= {arXiv preprint arXiv:2408.06039},
  year   = {2024}
}
R2 v1 2026-06-28T18:10:16.409Z