Symmetry-Based Disentangled Representation Learning Requires Interaction with Environments

Abstract

Finding a generally accepted formal definition of a disentangled representation in the context of an agent behaving in an environment is an important challenge towards the construction of data-efficient autonomous agents. Higgins et al. recently proposed Symmetry-Based Disentangled Representation Learning, a definition based on a characterization of symmetries in the environment using group theory. We build on their work and make observations, theoretical and empirical, that lead us to argue that Symmetry-Based Disentangled Representation Learning cannot only be based on static observations: agents should interact with the environment to discover its symmetries. Our experiments can be reproduced in Colab and the code is available on GitHub.

Cite

Text

Caselles-Dupré et al. "Symmetry-Based Disentangled Representation Learning Requires Interaction with Environments." Neural Information Processing Systems, 2019.

Markdown

[Caselles-Dupré et al. "Symmetry-Based Disentangled Representation Learning Requires Interaction with Environments." Neural Information Processing Systems, 2019.](https://mlanthology.org/neurips/2019/casellesdupre2019neurips-symmetrybased/)

BibTeX

@inproceedings{casellesdupre2019neurips-symmetrybased,
  title     = {{Symmetry-Based Disentangled Representation Learning Requires Interaction with Environments}},
  author    = {Caselles-Dupré, Hugo and Ortiz, Michael Garcia and Filliat, David},
  booktitle = {Neural Information Processing Systems},
  year      = {2019},
  pages     = {4606-4615},
  url       = {https://mlanthology.org/neurips/2019/casellesdupre2019neurips-symmetrybased/}
}