Symmetry-Based Disentangled Representation Learning requires Interaction with Environments
This addresses a foundational challenge in AI for building more efficient agents, though it is incremental as it builds on prior work.
The paper argues that symmetry-based disentangled representation learning, which aims to create data-efficient autonomous agents, cannot rely solely on static observations and requires agent interaction with the environment to discover symmetries, supported by theoretical and empirical observations.
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.