Spatial Transformer K-Means

Romain Cosentino; Randall Balestriero; Yanis Bahroun; Anirvan Sengupta; Richard Baraniuk; Behnaam Aazhang

doi:10.1109/IEEECONF56349.2022.10064695

Spatial Transformer K-Means

Romain Cosentino, Randall Balestriero, Yanis Bahroun, Anirvan Sengupta, Richard Baraniuk, Behnaam Aazhang

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

1 Scopus citations

Abstract

The K-means algorithm is one of the most employed centroid-based clustering algorithms. Unfortunately, it often requires intricate data embeddings for good performance, which comes at the cost of reduced theoretical guarantees and loss of interpretability. Instead, we propose to use the intrinsic data space and augment K-means with a similarity measure invariant to non-rigid transformations. This enables (i) the reduction of intrinsic nuisances associated with the data, making the clustering task simpler and improving performance, leading to state-of-theart results, (ii) clustering in the input space of the data, providing a fully interpretable clustering algorithm, and (iii) the benefit of convergence guarantees.

Original language	English (US)
Title of host publication	56th Asilomar Conference on Signals, Systems and Computers, ACSSC 2022
Editors	Michael B. Matthews
Publisher	IEEE Computer Society
Pages	1444-1448
Number of pages	5
ISBN (Electronic)	9781665459068
DOIs	https://doi.org/10.1109/IEEECONF56349.2022.10064695
State	Published - 2022
Externally published	Yes
Event	56th Asilomar Conference on Signals, Systems and Computers, ACSSC 2022 - Virtual, Online, United States Duration: Oct 31 2022 → Nov 2 2022

Publication series

Name	Conference Record - Asilomar Conference on Signals, Systems and Computers
Volume	2022-October
ISSN (Print)	1058-6393

Conference

Conference	56th Asilomar Conference on Signals, Systems and Computers, ACSSC 2022
Country/Territory	United States
City	Virtual, Online
Period	10/31/22 → 11/2/22

All Science Journal Classification (ASJC) codes

Signal Processing
Computer Networks and Communications

Keywords

K-means
Spatial transformer
Symmetry
Thin plate spline interpolation

Access to Document

10.1109/IEEECONF56349.2022.10064695

Cite this

Cosentino, R., Balestriero, R., Bahroun, Y., Sengupta, A., Baraniuk, R., & Aazhang, B. (2022). Spatial Transformer K-Means. In M. B. Matthews (Ed.), 56th Asilomar Conference on Signals, Systems and Computers, ACSSC 2022 (pp. 1444-1448). (Conference Record - Asilomar Conference on Signals, Systems and Computers; Vol. 2022-October). IEEE Computer Society. https://doi.org/10.1109/IEEECONF56349.2022.10064695

@inproceedings{ed361f1564ea4f259cfd5d5ccf8176f6,

title = "Spatial Transformer K-Means",

abstract = "The K-means algorithm is one of the most employed centroid-based clustering algorithms. Unfortunately, it often requires intricate data embeddings for good performance, which comes at the cost of reduced theoretical guarantees and loss of interpretability. Instead, we propose to use the intrinsic data space and augment K-means with a similarity measure invariant to non-rigid transformations. This enables (i) the reduction of intrinsic nuisances associated with the data, making the clustering task simpler and improving performance, leading to state-of-theart results, (ii) clustering in the input space of the data, providing a fully interpretable clustering algorithm, and (iii) the benefit of convergence guarantees.",

keywords = "K-means, Spatial transformer, Symmetry, Thin plate spline interpolation",

author = "Romain Cosentino and Randall Balestriero and Yanis Bahroun and Anirvan Sengupta and Richard Baraniuk and Behnaam Aazhang",

note = "Publisher Copyright: {\textcopyright} 2022 IEEE.; 56th Asilomar Conference on Signals, Systems and Computers, ACSSC 2022 ; Conference date: 31-10-2022 Through 02-11-2022",

year = "2022",

doi = "10.1109/IEEECONF56349.2022.10064695",

language = "English (US)",

series = "Conference Record - Asilomar Conference on Signals, Systems and Computers",

publisher = "IEEE Computer Society",

pages = "1444--1448",

editor = "Matthews, {Michael B.}",

booktitle = "56th Asilomar Conference on Signals, Systems and Computers, ACSSC 2022",

address = "United States",

}

Cosentino, R, Balestriero, R, Bahroun, Y, Sengupta, A, Baraniuk, R & Aazhang, B 2022, Spatial Transformer K-Means. in MB Matthews (ed.), 56th Asilomar Conference on Signals, Systems and Computers, ACSSC 2022. Conference Record - Asilomar Conference on Signals, Systems and Computers, vol. 2022-October, IEEE Computer Society, pp. 1444-1448, 56th Asilomar Conference on Signals, Systems and Computers, ACSSC 2022, Virtual, Online, United States, 10/31/22. https://doi.org/10.1109/IEEECONF56349.2022.10064695

Spatial Transformer K-Means. / Cosentino, Romain; Balestriero, Randall; Bahroun, Yanis et al.
56th Asilomar Conference on Signals, Systems and Computers, ACSSC 2022. ed. / Michael B. Matthews. IEEE Computer Society, 2022. p. 1444-1448 (Conference Record - Asilomar Conference on Signals, Systems and Computers; Vol. 2022-October).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - Spatial Transformer K-Means

AU - Cosentino, Romain

AU - Balestriero, Randall

AU - Bahroun, Yanis

AU - Sengupta, Anirvan

AU - Baraniuk, Richard

AU - Aazhang, Behnaam

PY - 2022

Y1 - 2022

N2 - The K-means algorithm is one of the most employed centroid-based clustering algorithms. Unfortunately, it often requires intricate data embeddings for good performance, which comes at the cost of reduced theoretical guarantees and loss of interpretability. Instead, we propose to use the intrinsic data space and augment K-means with a similarity measure invariant to non-rigid transformations. This enables (i) the reduction of intrinsic nuisances associated with the data, making the clustering task simpler and improving performance, leading to state-of-theart results, (ii) clustering in the input space of the data, providing a fully interpretable clustering algorithm, and (iii) the benefit of convergence guarantees.

AB - The K-means algorithm is one of the most employed centroid-based clustering algorithms. Unfortunately, it often requires intricate data embeddings for good performance, which comes at the cost of reduced theoretical guarantees and loss of interpretability. Instead, we propose to use the intrinsic data space and augment K-means with a similarity measure invariant to non-rigid transformations. This enables (i) the reduction of intrinsic nuisances associated with the data, making the clustering task simpler and improving performance, leading to state-of-theart results, (ii) clustering in the input space of the data, providing a fully interpretable clustering algorithm, and (iii) the benefit of convergence guarantees.

KW - K-means

KW - Spatial transformer

KW - Symmetry

KW - Thin plate spline interpolation

UR - http://www.scopus.com/inward/record.url?scp=85150214701&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85150214701&partnerID=8YFLogxK

U2 - 10.1109/IEEECONF56349.2022.10064695

DO - 10.1109/IEEECONF56349.2022.10064695

M3 - Conference contribution

AN - SCOPUS:85150214701

T3 - Conference Record - Asilomar Conference on Signals, Systems and Computers

SP - 1444

EP - 1448

BT - 56th Asilomar Conference on Signals, Systems and Computers, ACSSC 2022

A2 - Matthews, Michael B.

PB - IEEE Computer Society

T2 - 56th Asilomar Conference on Signals, Systems and Computers, ACSSC 2022

Y2 - 31 October 2022 through 2 November 2022

ER -

Spatial Transformer K-Means

Abstract

Publication series

Conference

All Science Journal Classification (ASJC) codes

Keywords

Access to Document

Other files and links

Fingerprint

Cite this