Autoencoder node saliency: Selecting relevant latent representations
Abstract
The autoencoder is an artificial neural network that performs nonlinear dimension reduction and 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 eigenvectors. In this work, we propose a novel autoencoder node saliency method that examines whether the features constructed by autoencoders exhibit properties related to known class labels. The supervised node saliency ranks the nodes based on their capability of performing a learning task. It is coupled with the normalized entropy difference (NED). We establish a property for NED values to verify classifying behaviors among the top ranked nodes. Lastly, by applying our methods to real datasets, we demonstrate their ability to provide indications on the performing nodes and explain the learned tasks in autoencoders.
- Authors:
-
- Lawrence Livermore National Lab. (LLNL), Livermore, CA (United States)
- Publication Date:
- Research Org.:
- Lawrence Livermore National Lab. (LLNL), Livermore, CA (United States)
- Sponsoring Org.:
- USDOE National Nuclear Security Administration (NNSA)
- OSTI Identifier:
- 1491664
- Alternate Identifier(s):
- OSTI ID: 1636754
- Report Number(s):
- LLNL-JRNL-741590
Journal ID: ISSN 0031-3203; 896098
- Grant/Contract Number:
- AC52-07NA27344
- Resource Type:
- Accepted Manuscript
- Journal Name:
- Pattern Recognition
- Additional Journal Information:
- Journal Volume: 88; Journal Issue: C; Journal ID: ISSN 0031-3203
- Publisher:
- Elsevier
- Country of Publication:
- United States
- Language:
- English
- Subject:
- 97 MATHEMATICS AND COMPUTING; Autoencoder; Latent representations; Unsupervised learning; Neural networks; Node selection; Model interpretation
Citation Formats
Fan, Ya Ju. Autoencoder node saliency: Selecting relevant latent representations. United States: N. p., 2018.
Web. doi:10.1016/j.patcog.2018.12.015.
Fan, Ya Ju. Autoencoder node saliency: Selecting relevant latent representations. United States. https://doi.org/10.1016/j.patcog.2018.12.015
Fan, Ya Ju. Mon .
"Autoencoder node saliency: Selecting relevant latent representations". United States. https://doi.org/10.1016/j.patcog.2018.12.015. https://www.osti.gov/servlets/purl/1491664.
@article{osti_1491664,
title = {Autoencoder node saliency: Selecting relevant latent representations},
author = {Fan, Ya Ju},
abstractNote = {The autoencoder is an artificial neural network that performs nonlinear dimension reduction and 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 eigenvectors. In this work, we propose a novel autoencoder node saliency method that examines whether the features constructed by autoencoders exhibit properties related to known class labels. The supervised node saliency ranks the nodes based on their capability of performing a learning task. It is coupled with the normalized entropy difference (NED). We establish a property for NED values to verify classifying behaviors among the top ranked nodes. Lastly, by applying our methods to real datasets, we demonstrate their ability to provide indications on the performing nodes and explain the learned tasks in autoencoders.},
doi = {10.1016/j.patcog.2018.12.015},
journal = {Pattern Recognition},
number = C,
volume = 88,
place = {United States},
year = {Mon Dec 17 00:00:00 EST 2018},
month = {Mon Dec 17 00:00:00 EST 2018}
}
Web of Science
Works referenced in this record:
Unsupervised feature extraction with autoencoder trees
journal, October 2017
- İrsoy, Ozan; Alpaydın, Ethem
- Neurocomputing, Vol. 258
Reducing the Dimensionality of Data with Neural Networks
journal, July 2006
- Hinton, G. E.
- Science, Vol. 313, Issue 5786
A Connection Between Score Matching and Denoising Autoencoders
journal, July 2011
- Vincent, Pascal
- Neural Computation, Vol. 23, Issue 7
Multi-Site Diagnostic Classification of Schizophrenia Using Discriminant Deep Learning with Functional Connectivity MRI
journal, April 2018
- Zeng, Ling-Li; Wang, Huaning; Hu, Panpan
- EBioMedicine, Vol. 30
Neural networks and principal component analysis: Learning from examples without local minima
journal, January 1989
- Baldi, Pierre; Hornik, Kurt
- Neural Networks, Vol. 2, Issue 1
Consensus self-organized models for fault detection (COSMO)
journal, August 2011
- Byttner, S.; Rögnvaldsson, T.; Svensson, M.
- Engineering Applications of Artificial Intelligence, Vol. 24, Issue 5
Self-monitoring for maintenance of vehicle fleets
journal, August 2017
- Rögnvaldsson, Thorsteinn; Nowaczyk, Sławomir; Byttner, Stefan
- Data Mining and Knowledge Discovery, Vol. 32, Issue 2
A survey on feature selection methods
journal, January 2014
- Chandrashekar, Girish; Sahin, Ferat
- Computers & Electrical Engineering, Vol. 40, Issue 1
A Survey on Feature Selection
journal, January 2016
- Miao, Jianyu; Niu, Lingfeng
- Procedia Computer Science, Vol. 91
An efficient semi-supervised representatives feature selection algorithm based on information theory
journal, January 2017
- Wang, Yintong; Wang, Jiandong; Liao, Hao
- Pattern Recognition, Vol. 61
Connectionist learning procedures
journal, September 1989
- Hinton, Geoffrey E.
- Artificial Intelligence, Vol. 40, Issue 1-3
Gradient-based learning applied to document recognition
journal, January 1998
- Lecun, Y.; Bottou, L.; Bengio, Y.
- Proceedings of the IEEE, Vol. 86, Issue 11
A trainable feature extractor for handwritten digit recognition
journal, June 2007
- Lauer, Fabien; Suen, Ching Y.; Bloch, Gérard
- Pattern Recognition, Vol. 40, Issue 6
A novel hybrid CNN–SVM classifier for recognizing handwritten digits
journal, April 2012
- Niu, Xiao-Xiao; Suen, Ching Y.
- Pattern Recognition, Vol. 45, Issue 4
The genomic and transcriptomic architecture of 2,000 breast tumours reveals novel subgroups
journal, April 2012
- Curtis, Christina; Shah, Sohrab P.; Chin, Suet-Feung
- Nature, Vol. 486, Issue 7403
Extremely randomized trees
journal, March 2006
- Geurts, Pierre; Ernst, Damien; Wehenkel, Louis
- Machine Learning, Vol. 63, Issue 1
Identification of bursts in spike trains
journal, January 1992
- Cocatre-Zilgien, J. H.; Delcomyn, F.
- Journal of Neuroscience Methods, Vol. 41, Issue 1
Works referencing / citing this record:
Assessment of Autoencoder Architectures for Data Representation
book, October 2019
- Pawar, Karishma; Attar, Vahida Z.; Pedrycz, Witold
- Deep Learning: Concepts and Architectures, p. 101-132
A Robust Deep Learning Approach for Spatiotemporal Estimation of Satellite AOD and PM2.5
journal, January 2020
- Li, Lianfa
- Remote Sensing, Vol. 12, Issue 2
Discriminative stacked autoencoder for feature representation and classification
journal, January 2020
- Gao, Yiping; Li, Xinyu; Gao, Liang
- Science China Information Sciences, Vol. 63, Issue 2