Auxiliary Deep Generative Models
Abstract
Deep generative models parameterized by neural networks have recently achieved state-of-the-art performance in unsupervised and semi-supervised learning. 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.
Cite
@article{arxiv.1602.05473,
title = {Auxiliary Deep Generative Models},
author = {Lars Maaløe and Casper Kaae Sønderby and Søren Kaae Sønderby and Ole Winther},
journal= {arXiv preprint arXiv:1602.05473},
year = {2016}
}
Comments
Proceedings of the 33rd International Conference on Machine Learning, New York, NY, USA, 2016, JMLR: Workshop and Conference Proceedings volume 48, Proceedings of the 33rd International Conference on Machine Learning, New York, NY, USA, 2016