Auxiliary deep generative models

Deep generative models parameterized by neural networks have recently achieved state-of-the-art performance in unsupervised and semi-supervised learning.
Lars Maaløe, Casper Kaae Sønderby, Søren Kaae Sønderby, Ole Winther

Explore our Research

We extend deep generative models with auxiliary variables, which improves the variational approximation. 

The auxiliary variables leave the generative model unchanged but make the variational distribution more expressive.

Inspired by the structure of the auxiliary variable, we also propose a model with two stochastic layers and skip connections. Our findings suggest that more expressive and properly specified deep generative models converge faster with better results.

We show state-of-the-art performance within semi-supervised learning on MNIST, SVHN, and NORB datasets.


More from Raffle