Uncertainty Autoencoders: Learning Compressed Representations via Variational Information Maximization
A. Grover, and S. Ermon. Proceedings of Machine Learning Research, volume 89 of Proceedings of Machine Learning Research, page 2514--2524. PMLR, (16--18 Apr 2019)
Abstract
Compressed sensing techniques enable efficient acquisition and recovery of sparse, highdimensional data signals via low-dimensional projections. In this work, we propose Uncertainty Autoencoders, a learning framework for unsupervised representation learning inspired by compressed sensing. We treat the low-dimensional projections as noisy latent representations of an autoencoder and directly learn both the acquisition (i.e., encoding) and amortized recovery (i.e., decoding) procedures. Our learning objective optimizes for a tractable variational lower bound to the mutual information between the datapoints and the latent representations. We show how our framework provides a unified treatment to several lines of research in dimensionality reduction, compressed sensing, and generative modeling. Empirically, we demonstrate a 32% improvement on average over competing approaches for the task of statistical compressed sensing of high-dimensional datasets.
Description
Uncertainty Autoencoders: Learning Compressed Representations via Variational Information Maximization
%0 Conference Paper
%1 pmlr-v89-grover19a
%A Grover, Aditya
%A Ermon, Stefano
%B Proceedings of Machine Learning Research
%D 2019
%E Chaudhuri, Kamalika
%E Sugiyama, Masashi
%I PMLR
%K uncertainty variational
%P 2514--2524
%T Uncertainty Autoencoders: Learning Compressed Representations via Variational Information Maximization
%U http://proceedings.mlr.press/v89/grover19a.html
%V 89
%X Compressed sensing techniques enable efficient acquisition and recovery of sparse, highdimensional data signals via low-dimensional projections. In this work, we propose Uncertainty Autoencoders, a learning framework for unsupervised representation learning inspired by compressed sensing. We treat the low-dimensional projections as noisy latent representations of an autoencoder and directly learn both the acquisition (i.e., encoding) and amortized recovery (i.e., decoding) procedures. Our learning objective optimizes for a tractable variational lower bound to the mutual information between the datapoints and the latent representations. We show how our framework provides a unified treatment to several lines of research in dimensionality reduction, compressed sensing, and generative modeling. Empirically, we demonstrate a 32% improvement on average over competing approaches for the task of statistical compressed sensing of high-dimensional datasets.
@inproceedings{pmlr-v89-grover19a,
abstract = {Compressed sensing techniques enable efficient acquisition and recovery of sparse, highdimensional data signals via low-dimensional projections. In this work, we propose Uncertainty Autoencoders, a learning framework for unsupervised representation learning inspired by compressed sensing. We treat the low-dimensional projections as noisy latent representations of an autoencoder and directly learn both the acquisition (i.e., encoding) and amortized recovery (i.e., decoding) procedures. Our learning objective optimizes for a tractable variational lower bound to the mutual information between the datapoints and the latent representations. We show how our framework provides a unified treatment to several lines of research in dimensionality reduction, compressed sensing, and generative modeling. Empirically, we demonstrate a 32% improvement on average over competing approaches for the task of statistical compressed sensing of high-dimensional datasets.},
added-at = {2019-12-11T13:56:49.000+0100},
author = {Grover, Aditya and Ermon, Stefano},
biburl = {https://www.bibsonomy.org/bibtex/2b56be6ddbfd1852b2b5e30af855cd5d1/kirk86},
booktitle = {Proceedings of Machine Learning Research},
description = {Uncertainty Autoencoders: Learning Compressed Representations via Variational Information Maximization},
editor = {Chaudhuri, Kamalika and Sugiyama, Masashi},
interhash = {eec2bf81730946b1e5eb8f148672b2d0},
intrahash = {b56be6ddbfd1852b2b5e30af855cd5d1},
keywords = {uncertainty variational},
month = {16--18 Apr},
pages = {2514--2524},
pdf = {http://proceedings.mlr.press/v89/grover19a/grover19a.pdf},
publisher = {PMLR},
series = {Proceedings of Machine Learning Research},
timestamp = {2019-12-11T13:56:49.000+0100},
title = {Uncertainty Autoencoders: Learning Compressed Representations via Variational Information Maximization},
url = {http://proceedings.mlr.press/v89/grover19a.html},
volume = 89,
year = 2019
}