Learning Group Structure and Disentangled Representations of Dynamical Environments

2020/02/17 by Robin Quessard, Quessard, Robin, Thomas D. Barrett +3 · 4 citations
Computer Science · Mathematics · Physics and Astronomy · #Computational Physics and Python Applications #FOS: Computer and information sciences #Generative Adversarial Networks and Image Synthesis #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Model Reduction and Neural Networks #Neural Networks and Applications #Time Series Analysis and Forecasting #cs.LG #stat.ML

paper · pdf · doi:10.48550/arxiv.2002.06991

Accepted to NeurIPS 2020

openalex publication_date 2020/02/17 · arxiv created 2020/10/25 · arxiv updated 2020/10/27 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Learning disentangled representations is a key step towards effectively discovering and modelling the underlying structure of environments. In the natural sciences, physics has found great success by describing the universe in terms of symmetry preserving transformations. Inspired by this formalism, we propose a framework, built upon the theory of group representation, for learning representations of a dynamical environment structured around the transformations that generate its evolution. Experimentally, we learn the structure of explicitly symmetric environments without supervision from observational data generated by sequential interactions. We further introduce an intuitive disentanglement regularisation to ensure the interpretability of the learnt representations. We show that our method enables accurate long-horizon predictions, and demonstrate a correlation between the quality of predictions and disentanglement in the latent space.

Citations

• Representation Learning: A Review and New Perspectives
• A Survey of Inductive Biases for Factorial Representation-Learning
• DARLA: Improving Zero-Shot Transfer in Reinforcement Learning
• Disentangling by Factorising
• Understanding disentangling in β-VAE
• Flatland: a Lightweight First-Person 2-D Environment for Reinforcement Learning
• Towards a Definition of Disentangled Representations
• Disentangled State Space Representations
• Learning Symmetries of Classical Integrable Systems
• InfoGAN: Interpretable Representation Learning by Information Maximizing Generative Adversarial Nets
• Hamiltonian Generative Networks
• Learning Compositional Koopman Operators for Model-Based Control
• Deep Variational Bayes Filters: Unsupervised Learning of State Space Models from Raw Data

Cited by

• Who Said Neural Networks Aren't Linear?
• Group Equivariant Conditional Neural Processes
• Learning Abstract World Models with a Group-Structured Latent Space
• Addressing the Topological Defects of Disentanglement via Distributed\n Operators

Related

• Symmetry-Based Disentangled Representation Learning requires Interaction with Environments
• Towards a Definition of Disentangled Representations
• Unsupervised Learning of Equivariant Structure from Sequences
• Disentangling Autoencoders (DAE)
• Learning Abstract World Models with a Group-Structured Latent Space