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.

arXiv comment33rd Conference on Neural Information Processing Systems (NeurIPS 2019), Vancouver, Canada

References in corpus (9)

• Scikit-learn: Machine Learning in Python
• Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations
• Recurrent World Models Facilitate Policy Evolution
• Towards a Definition of Disentangled Representations
• Are Disentangled Representations Helpful for Abstract Visual Reasoning?
• Independently Controllable Factors
• Independently Controllable Features
• Steps Towards a Theory of Visual Information: Active Perception, Signal-to-Symbol Conversion and the Interplay Between Sensing and Control
• Flatland: a Lightweight First-Person 2-D Environment for Reinforcement Learning

Cited by in corpus (4)

• Learning Disentangled Representations in the Imaging Domain
• A survey on intrinsic motivation in reinforcement learning
• Commutative Lie Group VAE for Disentanglement Learning
• No Representation without Transformation