Disentangling by Subspace Diffusion

Abstract

We present a novel nonparametric algorithm for symmetry-based disentangling of data manifolds, the Geometric Manifold Component Estimator (GEOMANCER). GEOMANCER provides a partial answer to the question posed by Higgins et al. (2018): is it possible to learn how to factorize a Lie group solely from observations of the orbit of an object it acts on? We show that fully unsupervised factorization of a data manifold is possible if the true metric of the manifold is known and each factor manifold has nontrivial holonomy -- for example, rotation in 3D. Our algorithm works by estimating the subspaces that are invariant under random walk diffusion, giving an approximation to the de Rham decomposition from differential geometry. We demonstrate the efficacy of GEOMANCER on several complex synthetic manifolds. Our work reduces the question of whether unsupervised disentangling is possible to the question of whether unsupervised metric learning is possible, providing a unifying insight into the geometric nature of representation learning.

arXiv commentCamera-ready version for NeurIPS 2020

References in corpus (20)

• Efficient Estimation of Word Representations in Vector Space
• Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representations
• Disentangling by Factorising
• Group Equivariant Convolutional Networks
• Towards a Definition of Disentangled Representations
• 3D Steerable CNNs: Learning Rotationally Equivariant Features in Volumetric Data
• Spherical CNNs
• Universal Invariant and Equivariant Graph Neural Networks
• Learning Deep Disentangled Embeddings with the F-Statistic Loss
• Structured Disentangled Representations
• Angles between subspaces and their tangents
• Life-Long Disentangled Representation Learning with Cross-Domain Latent Homologies
• Generalizing Convolutional Neural Networks for Equivariance to Lie Groups on Arbitrary Continuous Data
• Learning the Irreducible Representations of Commutative Lie Groups
• Unsupervised Model Selection for Variational Disentangled Representation Learning
• Robust Multiple Manifolds Structure Learning
• A Spectral Regularizer for Unsupervised Disentanglement
• An Unsupervised Algorithm For Learning Lie Group Transformations
• Learning Group Structure and Disentangled Representations of Dynamical Environments
• Spectral Inference Networks: Unifying Deep and Spectral Learning

Cited by in corpus (5)

• Overcoming Failures of Imagination in AI Infused System Development and Deployment
• Coordinate Independent Convolutional Networks -- Isometry and Gauge Equivariant Convolutions on Riemannian Manifolds
• GeomCA: Geometric Evaluation of Data Representations
• Linear Disentangled Representations and Unsupervised Action Estimation
• Learning to Transfer: A Foliated Theory