English

On the finite representation of group equivariant operators via permutant measures

Group Theory 2022-03-11 v2 Machine Learning Representation Theory

Abstract

The study of GG-equivariant operators is of great interest to explain and understand the architecture of neural networks. In this paper we show that each linear GG-equivariant operator can be produced by a suitable permutant measure, provided that the group GG transitively acts on a finite signal domain XX. This result makes available a new method to build linear GG-equivariant operators in the finite setting.

Keywords

Cite

@article{arxiv.2008.06340,
  title  = {On the finite representation of group equivariant operators via permutant measures},
  author = {Giovanni Bocchi and Stefano Botteghi and Martina Brasini and Patrizio Frosini and Nicola Quercioli},
  journal= {arXiv preprint arXiv:2008.06340},
  year   = {2022}
}

Comments

We have extended the introduction, providing more context about our approach, and the discussion. We have also added Section 4, showing possible advantages of the use of GENEOs built by permutant measures. The experiment described in Section 4 has been made by Giovanni Bocchi. For this reason, he has been added to the list of authors of the paper