Multi-Level Variational Autoencoder: Learning Disentangled Representations from Grouped Observations

Abstract

We would like to learn a representation of the data which decomposes an observation into factors of variation which we can independently control. Specifically, we want to use minimal supervision to learn a latent representation that reflects the semantics behind a specific grouping of the data, where within a group the samples share a common factor of variation. For example, consider a collection of face images grouped by identity. We wish to anchor the semantics of the grouping into a relevant and disentangled representation that we can easily exploit. However, existing deep probabilistic models often assume that the observations are independent and identically distributed. We present the Multi-Level Variational Autoencoder (ML-VAE), a new deep probabilistic model for learning a disentangled representation of a set of grouped observations. The ML-VAE separates the latent representation into semantically meaningful parts by working both at the group level and the observation level, while retaining efficient test-time inference. Quantitative and qualitative evaluations show that the ML-VAE model (i) learns a semantically meaningful disentanglement of grouped data, (ii) enables manipulation of the latent representation, and (iii) generalises to unseen groups.

Cited by in corpus (62)

• Deep Learning and Knowledge-Based Methods for Computer Aided Molecular Design -- Toward a Unified Approach: State-of-the-Art and Future Directions
• Are Disentangled Representations Helpful for Abstract Visual Reasoning?
• Neural Style Transfer: A Review
• Weakly-Supervised Disentanglement Without Compromises
• DIVA: Domain Invariant Variational Autoencoders
• Disentangling Hate in Online Memes
• Revisiting Training Strategies and Generalization Performance in Deep Metric Learning
• Factorizable Graph Convolutional Networks
• Weakly Supervised Disentanglement with Guarantees
• Disentangling Disentanglement in Variational Autoencoders
• Disentangling Factors of Variation Using Few Labels
• On Disentangled Representations Learned From Correlated Data
• Deep Music Analogy Via Latent Representation Disentanglement
• 3DMolNet: A Generative Network for Molecular Structures
• Sill-Net: Feature Augmentation with Separated Illumination Representation
• Copulas as High-Dimensional Generative Models: Vine Copula Autoencoders
• Robustly Disentangled Causal Mechanisms: Validating Deep Representations for Interventional Robustness
• Structural-analogy from a Single Image Pair
• Learning Disentangled Representations with Reference-Based Variational Autoencoders
• Desiderata for Representation Learning: A Causal Perspective
• Generative causal explanations of black-box classifiers
• Semi-Supervised StyleGAN for Disentanglement Learning
• Dual Swap Disentangling
• Disentangling Latent Hands for Image Synthesis and Pose Estimation
• An Identifiable Double VAE For Disentangled Representations
• Adversarial Learning of Deepfakes in Accounting
• Uncertainty in Multitask Transfer Learning
• Demystifying Inter-Class Disentanglement
• Autoencoder Image Interpolation by Shaping the Latent Space
• Product of Orthogonal Spheres Parameterization for Disentangled Representation Learning
• On the Fairness of Disentangled Representations
• Weakly Supervised Disentanglement by Pairwise Similarities
• Visual analytics of set data for knowledge discovery and member selection support
• Robust Ordinal VAE: Employing Noisy Pairwise Comparisons for Disentanglement
• Measuring the Biases and Effectiveness of Content-Style Disentanglement
• On the Interpretability and Evaluation of Graph Representation Learning
• An Image is Worth More Than a Thousand Words: Towards Disentanglement in the Wild
• Revisiting Classical Bagging with Modern Transfer Learning for On-the-fly Disaster Damage Detector
• DynamicVAE: Decoupling Reconstruction Error and Disentangled Representation Learning
• dMelodies: A Music Dataset for Disentanglement Learning
• S2SD: Simultaneous Similarity-based Self-Distillation for Deep Metric Learning
• Training Interpretable Convolutional Neural Networks by Differentiating Class-specific Filters
• Video Reenactment as Inductive Bias for Content-Motion Disentanglement
• Improving Style-Content Disentanglement in Image-to-Image Translation
• Latent Variable Modeling for Generative Concept Representations and Deep Generative Models
• Deep Anomaly Detection by Residual Adaptation
• WeLa-VAE: Learning Alternative Disentangled Representations Using Weak Labels
• Learning to Manipulate Individual Objects in an Image
• Learning Controllable Disentangled Representations with Decorrelation Regularization
• Revisiting Factorizing Aggregated Posterior in Learning Disentangled Representations
• Generative Model without Prior Distribution Matching
• Variationally Inferred Sampling Through a Refined Bound for Probabilistic Programs
• Inverse Learning of Symmetries
• Mixture Representation Learning with Coupled Autoencoders
• Rethinking Content and Style: Exploring Bias for Unsupervised Disentanglement
• Learning Conditional Invariance through Cycle Consistency
• Group-disentangled Representation Learning with Weakly-Supervised Regularization
• Adversarial Learning of Poisson Factorisation Model for Gauging Brand Sentiment in User Reviews
• Group-based Learning of Disentangled Representations with Generalizability for Novel Contents
• DAReN: A Collaborative Approach Towards Reasoning And Disentangling
• Domain Agnostic Learning for Unbiased Authentication
• Recursively Conditional Gaussian for Ordinal Unsupervised Domain Adaptation