1: \begin{abstract}

2: 	Permutation equivariant neural networks are typically used to learn

3: 	from data that lives on a graph.  However, for any graph $G$ that has

4: 	$n$ vertices, using the symmetric group $S_n$ as its group of

5: 	symmetries does not take into account the relations that exist between

6: 	the vertices.  Given that the actual group of symmetries is the

7: 	automorphism group $\Aut(G)$, we show how to construct neural networks

8: 	that are equivariant to $\Aut(G)$ by obtaining a full characterisation

9: 	of the learnable, linear, $\Aut(G)$-equivariant functions between

10: 	layers that are some tensor power of $\mathbb{R}^{n}$.  In particular,

11: 	we find a spanning set of matrices for these layer functions in the

12: 	standard basis of $\mathbb{R}^{n}$.  This result has important

13: 	consequences for learning from data whose group of symmetries is a

14: 	finite group because a theorem by \citet{frucht} showed that any finite

15: 	group is isomorphic to the automorphism group of a graph. 

16: \end{abstract}

17: