The autoencoder is an artificial neural network model that learns hidden
representations of unlabeled data. With a linear transfer function it is
similar to the principal component analysis (PCA). While both methods use
weight vectors for linear transformations, the autoencoder does not come with
any indication similar to the eigenvalues in PCA that are paired with the
eigenvectors. We propose a novel supervised node saliency (SNS) method that
ranks the hidden nodes by comparing class distributions of latent
representations against a fixed reference distribution. The latent
representations of a hidden node can be described using a one-dimensional
histogram. We apply normalized entropy difference (NED) to measure the
"interestingness" of the histograms, and conclude a property for NED values to
identify a good classifying node. By applying our methods to real data sets, we
demonstrate the ability of SNS to explain what the trained autoencoders have
learned.
%0 Generic
%1 fan2017autoencoder
%A Fan, Ya Ju
%D 2017
%K autoencoder to_read unsupervised
%T Autoencoder Node Saliency: Selecting Relevant Latent Representations
%U http://arxiv.org/abs/1711.07871
%X The autoencoder is an artificial neural network model that learns hidden
representations of unlabeled data. With a linear transfer function it is
similar to the principal component analysis (PCA). While both methods use
weight vectors for linear transformations, the autoencoder does not come with
any indication similar to the eigenvalues in PCA that are paired with the
eigenvectors. We propose a novel supervised node saliency (SNS) method that
ranks the hidden nodes by comparing class distributions of latent
representations against a fixed reference distribution. The latent
representations of a hidden node can be described using a one-dimensional
histogram. We apply normalized entropy difference (NED) to measure the
"interestingness" of the histograms, and conclude a property for NED values to
identify a good classifying node. By applying our methods to real data sets, we
demonstrate the ability of SNS to explain what the trained autoencoders have
learned.
@misc{fan2017autoencoder,
abstract = {The autoencoder is an artificial neural network model that learns hidden
representations of unlabeled data. With a linear transfer function it is
similar to the principal component analysis (PCA). While both methods use
weight vectors for linear transformations, the autoencoder does not come with
any indication similar to the eigenvalues in PCA that are paired with the
eigenvectors. We propose a novel supervised node saliency (SNS) method that
ranks the hidden nodes by comparing class distributions of latent
representations against a fixed reference distribution. The latent
representations of a hidden node can be described using a one-dimensional
histogram. We apply normalized entropy difference (NED) to measure the
"interestingness" of the histograms, and conclude a property for NED values to
identify a good classifying node. By applying our methods to real data sets, we
demonstrate the ability of SNS to explain what the trained autoencoders have
learned.},
added-at = {2018-03-12T09:55:59.000+0100},
author = {Fan, Ya Ju},
biburl = {https://www.bibsonomy.org/bibtex/2fb18f7e0591ef4e0c76c33d3f7b41028/jk_itwm},
description = {Autoencoder Node Saliency: Selecting Relevant Latent Representations},
interhash = {f2a57f96ba709282641361767f428806},
intrahash = {fb18f7e0591ef4e0c76c33d3f7b41028},
keywords = {autoencoder to_read unsupervised},
note = {cite arxiv:1711.07871},
timestamp = {2018-03-12T09:55:59.000+0100},
title = {Autoencoder Node Saliency: Selecting Relevant Latent Representations},
url = {http://arxiv.org/abs/1711.07871},
year = 2017
}