Stick-Breaking Variational Autoencoders
arXiv:1605.06197
Abstract
We extend Stochastic Gradient Variational Bayes to perform posterior inference for the weights of Stick-Breaking processes. This development allows us to define a Stick-Breaking Variational Autoencoder (SB-VAE), a Bayesian nonparametric version of the variational autoencoder that has a latent representation with stochastic dimensionality. We experimentally demonstrate that the SB-VAE, and a semi-supervised variant, learn highly discriminative latent representations that often outperform the Gaussian VAE's.
ICLR 2017, Conference Track
Cited by in corpus (15)
- A Tutorial on Deep Latent Variable Models of Natural Language
- Adversarial Neural Pruning with Latent Vulnerability Suppression
- Dirichlet Graph Variational Autoencoder
- Variational Autoencoders with Riemannian Brownian Motion Priors
- A Latent Morphology Model for Open-Vocabulary Neural Machine Translation
- Lifelong Bayesian Optimization
- Topic Modelling Meets Deep Neural Networks: A Survey
- Variational Autoencoder with Implicit Optimal Priors
- Bayesian Structure Adaptation for Continual Learning
- Improving Disentangled Representation Learning with the Beta Bernoulli Process
- Hyperprior Induced Unsupervised Disentanglement of Latent Representations
- Residual-Recursion Autoencoder for Shape Illustration Images
- ByPE-VAE: Bayesian Pseudocoresets Exemplar VAE
- A Factorial Mixture Prior for Compositional Deep Generative Models
- Analytical Probability Distributions and EM-Learning for Deep Generative Networks