Structured Disentangled Representations

Abstract

Deep latent-variable models learn representations of high-dimensional data in an unsupervised manner. A number of recent efforts have focused on learning representations that disentangle statistically independent axes of variation by introducing modifications to the standard objective function. These approaches generally assume a simple diagonal Gaussian prior and as a result are not able to reliably disentangle discrete factors of variation. We propose a two-level hierarchical objective to control relative degree of statistical independence between blocks of variables and individual variables within blocks. We derive this objective as a generalization of the evidence lower bound, which allows us to explicitly represent the trade-offs between mutual information between data and representation, KL divergence between representation and prior, and coverage of the support of the empirical data distribution. Experiments on a variety of datasets demonstrate that our objective can not only disentangle discrete variables, but that doing so also improves disentanglement of other variables and, importantly, generalization even to unseen combinations of factors.

Cited by in corpus (34)

• Are Disentangled Representations Helpful for Abstract Visual Reasoning?
• Learning Disentangled Representations in the Imaging Domain
• Unsupervised Object Segmentation by Redrawing
• On the Binding Problem in Artificial Neural Networks
• Counterfactual Data Augmentation using Locally Factored Dynamics
• InfoCatVAE: Representation Learning with Categorical Variational Autoencoders
• Disentangling Influence: Using Disentangled Representations to Audit Model Predictions
• Disentangled Representations from Non-Disentangled Models
• On the Fairness of Disentangled Representations
• Disentangling and Learning Robust Representations with Natural Clustering
• Fully-hierarchical fine-grained prosody modeling for interpretable speech synthesis
• Improving VAEs' Robustness to Adversarial Attack
• Benchmarks, Algorithms, and Metrics for Hierarchical Disentanglement
• An Improved Semi-Supervised VAE for Learning Disentangled Representations
• Generalization and Robustness Implications in Object-Centric Learning
• A survey on Variational Autoencoders from a GreenAI perspective
• Interpretable Neuron Structuring with Graph Spectral Regularization
• Disentangling by Subspace Diffusion
• OOGAN: Disentangling GAN with One-Hot Sampling and Orthogonal Regularization
• Disentangling to Cluster: Gaussian Mixture Variational Ladder Autoencoders
• Discond-VAE: Disentangling Continuous Factors from the Discrete
• Semi-supervised Disentanglement with Independent Vector Variational Autoencoders
• Adversarial Disentanglement with Grouped Observations
• Differentially Private Variational Autoencoders with Term-wise Gradient Aggregation
• PRI-VAE: Principle-of-Relevant-Information Variational Autoencoders
• Decomposing Normal and Abnormal Features of Medical Images into Discrete Latent Codes for Content-Based Image Retrieval
• Learning Bijective Feature Maps for Linear ICA
• Symmetric Wasserstein Autoencoders
• Semi-Supervised Disentanglement of Class-Related and Class-Independent Factors in VAE
• Bayesian neural networks and dimensionality reduction
• Training Machine Learning Models by Regularizing their Explanations
• On Controlled DeEntanglement for Natural Language Processing
• Deep Dive into Semi-Supervised ELBO for Improving Classification Performance
• Text Classification through Glyph-aware Disentangled Character Embedding and Semantic Sub-character Augmentation