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.
Hugo Caselles-Dupré (Flowers Laboratory (ENSTA ParisTech & INRIA) & Softbank Robotics Europe)
PhD candidate working on Reinforcement Learning and Developmental Robotics.
Michael Garcia Ortiz (SoftBank Robotics Europe)
David Filliat (ENSTA)
More from the Same Authors
2019 Poster: Unsupervised Emergence of Egocentric Spatial Structure from Sensorimotor Prediction »
Alban Laflaquière · Michael Garcia Ortiz