Transformation invariant sparse coding

Morten Mørup; Mikkel Nørgaard Schmidt

doi:10.1109/MLSP.2011.6064547

Transformation invariant sparse coding

Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review

Abstract

Sparse coding is a well established principle for unsupervised learning. Traditionally, features are extracted in sparse coding in specific locations, however, often we would prefer invariant representation. This paper introduces a general transformation invariant sparse coding (TISC) model. The model decomposes images into features invariant to location and general transformation by a set of specified operators as well as a sparse coding matrix indicating where and to what degree in the original image these features are present. The TISC model is in general overcomplete and we therefore invoke sparse coding to estimate its parameters. We demonstrate how the model can correctly identify components of non-trivial artificial as well as real image data. Thus, the model is capable of reducing feature redundancies in terms of pre-specified transformations improving the component identification.

Original language	English
Title of host publication	2011 IEEE International Workshop on Machine Learning for Signal Processing (MLSP)
Publisher	IEEE
Publication date	2011
ISBN (Print)	978-1-4577-1621-8
ISBN (Electronic)	978-1-4577-1622-5
DOIs	https://doi.org/10.1109/MLSP.2011.6064547
Publication status	Published - 2011
Event	2011 IEEE International Workshop on Machine Learning for Signal Processing - Beijing, China Duration: 18 Sept 2011 → 21 Sept 2011 Conference number: 21 https://ieeexplore.ieee.org/xpl/conhome/6058570/proceeding

Conference

Conference	2011 IEEE International Workshop on Machine Learning for Signal Processing
Number	21
Country/Territory	China
City	Beijing
Period	18/09/2011 → 21/09/2011
Internet address	https://ieeexplore.ieee.org/xpl/conhome/6058570/proceeding

Series	Machine Learning for Signal Processing
ISSN	1551-2541

Access to Document

10.1109/MLSP.2011.6064547

http://mlsp2011.conwiz.dk/

OpenUrl availability

Full text

Cite this

@inproceedings{b5e2f6d2f0d14f07b200bda224842cd0,

title = "Transformation invariant sparse coding",

abstract = "Sparse coding is a well established principle for unsupervised learning. Traditionally, features are extracted in sparse coding in specific locations, however, often we would prefer invariant representation. This paper introduces a general transformation invariant sparse coding (TISC) model. The model decomposes images into features invariant to location and general transformation by a set of specified operators as well as a sparse coding matrix indicating where and to what degree in the original image these features are present. The TISC model is in general overcomplete and we therefore invoke sparse coding to estimate its parameters. We demonstrate how the model can correctly identify components of non-trivial artificial as well as real image data. Thus, the model is capable of reducing feature redundancies in terms of pre-specified transformations improving the component identification.",

author = "Morten M{\o}rup and Schmidt, {Mikkel N{\o}rgaard}",

year = "2011",

doi = "10.1109/MLSP.2011.6064547",

language = "English",

isbn = "978-1-4577-1621-8",

series = "Machine Learning for Signal Processing",

booktitle = "2011 IEEE International Workshop on Machine Learning for Signal Processing (MLSP)",

publisher = "IEEE",

address = "United States",

note = "2011 IEEE International Workshop on Machine Learning for Signal Processing, MLSP 2011 ; Conference date: 18-09-2011 Through 21-09-2011",

url = "https://ieeexplore.ieee.org/xpl/conhome/6058570/proceeding",

}

TY - GEN

T1 - Transformation invariant sparse coding

AU - Mørup, Morten

AU - Schmidt, Mikkel Nørgaard

N1 - Conference code: 21

PY - 2011

Y1 - 2011

N2 - Sparse coding is a well established principle for unsupervised learning. Traditionally, features are extracted in sparse coding in specific locations, however, often we would prefer invariant representation. This paper introduces a general transformation invariant sparse coding (TISC) model. The model decomposes images into features invariant to location and general transformation by a set of specified operators as well as a sparse coding matrix indicating where and to what degree in the original image these features are present. The TISC model is in general overcomplete and we therefore invoke sparse coding to estimate its parameters. We demonstrate how the model can correctly identify components of non-trivial artificial as well as real image data. Thus, the model is capable of reducing feature redundancies in terms of pre-specified transformations improving the component identification.

AB - Sparse coding is a well established principle for unsupervised learning. Traditionally, features are extracted in sparse coding in specific locations, however, often we would prefer invariant representation. This paper introduces a general transformation invariant sparse coding (TISC) model. The model decomposes images into features invariant to location and general transformation by a set of specified operators as well as a sparse coding matrix indicating where and to what degree in the original image these features are present. The TISC model is in general overcomplete and we therefore invoke sparse coding to estimate its parameters. We demonstrate how the model can correctly identify components of non-trivial artificial as well as real image data. Thus, the model is capable of reducing feature redundancies in terms of pre-specified transformations improving the component identification.

U2 - 10.1109/MLSP.2011.6064547

DO - 10.1109/MLSP.2011.6064547

M3 - Article in proceedings

SN - 978-1-4577-1621-8

T3 - Machine Learning for Signal Processing

BT - 2011 IEEE International Workshop on Machine Learning for Signal Processing (MLSP)

PB - IEEE

T2 - 2011 IEEE International Workshop on Machine Learning for Signal Processing

Y2 - 18 September 2011 through 21 September 2011

ER -

Transformation invariant sparse coding

Abstract

Conference

Access to Document

OpenUrl availability

Fingerprint

Cite this