Learning Disentangled Representations with Semi-Supervised Deep Generative Models

Abstract

Variational autoencoders (VAEs) learn representations of data by jointly training a probabilistic encoder and decoder network. Typically these models encode all features of the data into a single variable. Here we are interested in learning disentangled representations that encode distinct aspects of the data into separate variables. We propose to learn such representations using model architectures that generalise from standard VAEs, employing a general graphical model structure in the encoder and decoder. This allows us to train partially-specified models that make relatively strong assumptions about a subset of interpretable variables and rely on the flexibility of neural networks to learn representations for the remaining variables. We further define a general objective for semi-supervised learning in this model class, which can be approximated using an importance sampling procedure. We evaluate our framework's ability to learn disentangled representations, both by qualitative exploration of its generative capacity, and quantitative evaluation of its discriminative ability on a variety of models and datasets.

Accepted for publication at NIPS 2017

Cited by in corpus (27)

• Pyro: Deep Universal Probabilistic Programming
• Recent Advances in Autoencoder-Based Representation Learning
• Dynamical Variational Autoencoders: A Comprehensive Review
• Mining gold from implicit models to improve likelihood-free inference
• Isolating Sources of Disentanglement in Variational Autoencoders
• Learning Disentangled Representations in the Imaging Domain
• DIVA: Domain Invariant Variational Autoencoders
• Factorizable Graph Convolutional Networks
• Lifelong Teacher-Student Network Learning
• Zero-Resource Knowledge-Grounded Dialogue Generation
• MutexMatch: Semi-Supervised Learning with Mutex-Based Consistency Regularization
• IB-GAN: Disentangled Representation Learning with Information Bottleneck Generative Adversarial Networks
• ADEV: Sound Automatic Differentiation of Expected Values of Probabilistic Programs
• Learning Interpretable Deep Disentangled Neural Networks for Hyperspectral Unmixing
• Disentanglement Learning for Variational Autoencoders Applied to Audio-Visual Speech Enhancement
• Leveraging the Exact Likelihood of Deep Latent Variable Models
• Disentangling Controllable and Uncontrollable Factors of Variation by Interacting with the World
• ProBO: Versatile Bayesian Optimization Using Any Probabilistic Programming Language
• Deep Probabilistic Programming Languages: A Qualitative Study
• Hyperprior Induced Unsupervised Disentanglement of Latent Representations
• Pixyz: a Python library for developing deep generative models
• Cross-domain Transfer Learning and State Inference for Soft Robots via a Semi-supervised Sequential Variational Bayes Framework
• Semi-supervised Sequential Generative Models
• Disentanglement Learning via Topology
• Capturing Actionable Dynamics with Structured Latent Ordinary Differential Equations
• Disentangling Video with Independent Prediction
• A Deeper Look at the Unsupervised Learning of Disentangled Representations in $\beta$-VAE from the Perspective of Core Object Recognition