Equivariant Representations for Non-Free Group Actions

Abstract

We introduce a method for learning representations that are equivariant with respect to general group actions over data. Differently from existing equivariant representation learners, our method is suitable for actions that are not free i.e., that stabilize data via nontrivial symmetries. Our method is grounded in the orbit-stabilizer theorem from group theory, which guarantees that an ideal learner infers an isomorphic representation. Finally, we provide an empirical investigation on image datasets with rotational symmetries and show that taking stabilizers into account improves the quality of the representations.

PDF NeurIPSW OpenReview Semantic Scholar

Cite

Text

Rey et al. "Equivariant Representations for Non-Free Group Actions." NeurIPS 2022 Workshops: NeurReps, 2022.

Markdown

[Rey et al. "Equivariant Representations for Non-Free Group Actions." NeurIPS 2022 Workshops: NeurReps, 2022.](https://mlanthology.org/neuripsw/2022/rey2022neuripsw-equivariant/)

BibTeX

@inproceedings{rey2022neuripsw-equivariant,
  title     = {{Equivariant Representations for Non-Free Group Actions}},
  author    = {Rey, Luis Armando Pérez and Marchetti, Giovanni Luca and Kragic, Danica and Jarnikov, Dmitri and Holenderski, Mike},
  booktitle = {NeurIPS 2022 Workshops: NeurReps},
  year      = {2022},
  url       = {https://mlanthology.org/neuripsw/2022/rey2022neuripsw-equivariant/}
}