Stick-Breaking Variational Autoencoders
This work addresses the need for Bayesian nonparametric latent representations with stochastic dimensionality in machine learning, offering an incremental improvement over existing variational autoencoders.
The authors tackled the problem of performing posterior inference for the weights of Stick-Breaking processes by extending Stochastic Gradient Variational Bayes, resulting in a Stick-Breaking Variational Autoencoder (SB-VAE) that learns highly discriminative latent representations, often outperforming Gaussian VAEs.
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.