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 (22)

• Deep Learning and Knowledge-Based Methods for Computer Aided Molecular Design -- Toward a Unified Approach: State-of-the-Art and Future Directions
• Disentangling Hate in Online Memes
• Learning Disentangled Representations with Reference-Based Variational Autoencoders
• Adversarial Learning of Deepfakes in Accounting
• Product of Orthogonal Spheres Parameterization for Disentangled Representation Learning
• Robust Ordinal VAE: Employing Noisy Pairwise Comparisons for Disentanglement
• An Image is Worth More Than a Thousand Words: Towards Disentanglement in the Wild
• On the Interpretability and Evaluation of Graph Representation Learning
• 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
• Improving Style-Content Disentanglement in Image-to-Image Translation
• Latent Variable Modeling for Generative Concept Representations and Deep Generative Models
• Learning Controllable Disentangled Representations with Decorrelation Regularization
• Deep Anomaly Detection by Residual Adaptation
• Learning to Manipulate Individual Objects in an Image
• WeLa-VAE: Learning Alternative Disentangled Representations Using Weak Labels
• Generative Model without Prior Distribution Matching
• Group-disentangled Representation Learning with Weakly-Supervised Regularization
• Recursively Conditional Gaussian for Ordinal Unsupervised Domain Adaptation
• Domain Agnostic Learning for Unbiased Authentication
• Adversarial Learning of Poisson Factorisation Model for Gauging Brand Sentiment in User Reviews