Learning Invariant Features Using Subspace Restricted Boltzmann Machine

Jakub M. Tomczak; Adam Gonczarek

doi:10.1007/s11063-016-9519-9

Learning Invariant Features Using Subspace Restricted Boltzmann Machine

Jakub M. Tomczak^*, Adam Gonczarek

^*Corresponding author for this work

Research output: Contribution to Journal › Article › Academic › peer-review

Abstract

The subspace restricted Boltzmann machine (subspaceRBM) is a third-order Boltzmann machine where multiplicative interactions are between one visible and two hidden units. There are two kinds of hidden units, namely, gate units and subspace units. The subspace units reflect variations of a pattern in data and the gate unit is responsible for activating the subspace units. Additionally, the gate unit can be seen as a pooling feature. We evaluate the behavior of subspaceRBM through experiments with MNIST digit recognition task and Caltech 101 Silhouettes image corpora, measuring cross-entropy reconstruction error and classification error.

Original language	English
Pages (from-to)	173-182
Number of pages	10
Journal	Neural Processing Letters
Volume	45
Issue number	1
DOIs	https://doi.org/10.1007/s11063-016-9519-9
Publication status	Published - 1 Feb 2017
Externally published	Yes

Funding

The research conducted by the authors has been partially co-financed by the Ministry of Science and Higher Education, Republic of Poland, namely, Jakub M. Tomczak: Grant No. B50106W8/K3, Adam Gonczarek: Grant No. B50137W8/K3.

Funders	Funder number
Ministerstwo Edukacji i Nauki	B50106W8/K3, B50137W8/K3

Keywords

Deep model
Feature learning
Invariant features
Subspace features
Unsupervised learning

Access to Document

10.1007/s11063-016-9519-9

Persistent URL (handle)

Copy link address

Cite this

@article{f3bbf967fa134eb29910a9b9c6ce24f3,

title = "Learning Invariant Features Using Subspace Restricted Boltzmann Machine",

abstract = "The subspace restricted Boltzmann machine (subspaceRBM) is a third-order Boltzmann machine where multiplicative interactions are between one visible and two hidden units. There are two kinds of hidden units, namely, gate units and subspace units. The subspace units reflect variations of a pattern in data and the gate unit is responsible for activating the subspace units. Additionally, the gate unit can be seen as a pooling feature. We evaluate the behavior of subspaceRBM through experiments with MNIST digit recognition task and Caltech 101 Silhouettes image corpora, measuring cross-entropy reconstruction error and classification error.",

keywords = "Deep model, Feature learning, Invariant features, Subspace features, Unsupervised learning",

author = "Tomczak, {Jakub M.} and Adam Gonczarek",

year = "2017",

month = feb,

day = "1",

doi = "10.1007/s11063-016-9519-9",

language = "English",

volume = "45",

pages = "173--182",

journal = "Neural Processing Letters",

issn = "1370-4621",

publisher = "Springer Netherlands",

number = "1",

}

TY - JOUR

T1 - Learning Invariant Features Using Subspace Restricted Boltzmann Machine

AU - Tomczak, Jakub M.

AU - Gonczarek, Adam

PY - 2017/2/1

Y1 - 2017/2/1

N2 - The subspace restricted Boltzmann machine (subspaceRBM) is a third-order Boltzmann machine where multiplicative interactions are between one visible and two hidden units. There are two kinds of hidden units, namely, gate units and subspace units. The subspace units reflect variations of a pattern in data and the gate unit is responsible for activating the subspace units. Additionally, the gate unit can be seen as a pooling feature. We evaluate the behavior of subspaceRBM through experiments with MNIST digit recognition task and Caltech 101 Silhouettes image corpora, measuring cross-entropy reconstruction error and classification error.

AB - The subspace restricted Boltzmann machine (subspaceRBM) is a third-order Boltzmann machine where multiplicative interactions are between one visible and two hidden units. There are two kinds of hidden units, namely, gate units and subspace units. The subspace units reflect variations of a pattern in data and the gate unit is responsible for activating the subspace units. Additionally, the gate unit can be seen as a pooling feature. We evaluate the behavior of subspaceRBM through experiments with MNIST digit recognition task and Caltech 101 Silhouettes image corpora, measuring cross-entropy reconstruction error and classification error.

KW - Deep model

KW - Feature learning

KW - Invariant features

KW - Subspace features

KW - Unsupervised learning

UR - http://www.scopus.com/inward/record.url?scp=84964057082&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84964057082&partnerID=8YFLogxK

U2 - 10.1007/s11063-016-9519-9

DO - 10.1007/s11063-016-9519-9

M3 - Article

AN - SCOPUS:84964057082

SN - 1370-4621

VL - 45

SP - 173

EP - 182

JO - Neural Processing Letters

JF - Neural Processing Letters

IS - 1

ER -

Learning Invariant Features Using Subspace Restricted Boltzmann Machine

Abstract

Funding

Keywords

Access to Document

Persistent URL (handle)

Other files and links

Fingerprint

Cite this