A Factorial Mixture Prior for Compositional Deep Generative Models
arXiv:1812.07480
Abstract
We assume that a high-dimensional datum, like an image, is a compositional expression of a set of properties, with a complicated non-linear relationship between the datum and its properties. This paper proposes a factorial mixture prior for capturing latent properties, thereby adding structured compositionality to deep generative models. The prior treats a latent vector as belonging to Cartesian product of subspaces, each of which is quantized separately with a Gaussian mixture model. Some mixture components can be set to represent properties as observed random variables whenever labeled properties are present. Through a combination of stochastic variational inference and gradient descent, a method for learning how to infer discrete properties in an unsupervised or semi-supervised way is outlined and empirically evaluated.
arXiv comment16 pagers, 10 figures
References in corpus (10)
- Auto-Encoding Variational Bayes
- Neural Discrete Representation Learning
- Disentangling by Factorising
- Inferring Parameters and Structure of Latent Variable Models by Variational Bayes
- Deep Unsupervised Clustering with Gaussian Mixture Variational Autoencoders
- Variational Lossy Autoencoder
- Isolating Sources of Disentanglement in Variational Autoencoders
- Stick-Breaking Variational Autoencoders
- Associative Compression Networks for Representation Learning
- On the Convergence of Stochastic Variational Inference in Bayesian Networks