Inducing Interpretable Representations with Variational Autoencoders
Machine Learning
2016-11-23 v1 Computer Vision and Pattern Recognition
Machine Learning
Abstract
We develop a framework for incorporating structured graphical models in the \emph{encoders} of variational autoencoders (VAEs) that allows us to induce interpretable representations through approximate variational inference. This allows us to both perform reasoning (e.g. classification) under the structural constraints of a given graphical model, and use deep generative models to deal with messy, high-dimensional domains where it is often difficult to model all the variation. Learning in this framework is carried out end-to-end with a variational objective, applying to both unsupervised and semi-supervised schemes.
Cite
@article{arxiv.1611.07492,
title = {Inducing Interpretable Representations with Variational Autoencoders},
author = {N. Siddharth and Brooks Paige and Alban Desmaison and Jan-Willem Van de Meent and Frank Wood and Noah D. Goodman and Pushmeet Kohli and Philip H. S. Torr},
journal= {arXiv preprint arXiv:1611.07492},
year = {2016}
}
Comments
Presented at NIPS 2016 Workshop on Interpretable Machine Learning in Complex Systems