Comparison of K-means clustering initialization approaches with brute-force initialization

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F61989100%3A27240%2F17%3A10236274" target="_blank" >RIV/61989100:27240/17:10236274 - isvavai.cz</a>

Nalezeny alternativní kódy

RIV/61989100:27740/17:10236274

Výsledek na webu

<a href="https://link.springer.com/content/pdf/10.1007%2F978-981-10-3409-1_7.pdf" target="_blank" >https://link.springer.com/content/pdf/10.1007%2F978-981-10-3409-1_7.pdf</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1007/978-981-10-3409-1_7" target="_blank" >10.1007/978-981-10-3409-1_7</a>

Jazyk výsledku

angličtina

Název v původním jazyce

Comparison of K-means clustering initialization approaches with brute-force initialization

Popis výsledku v původním jazyce

Data clustering is a basic data mining discipline that has been in center of interest of many research groups. This paper describes the formulation of the basic NP-hard optimization problem in data clustering which is approximated by many heuristic methods. The famous k-means clustering algorithm and its initialization is of a particular interest in this paper. A summary of the k-means variants and various initialization strategies is presented. Many initialization heuristics tend to search only through a fraction of the initial centroid space. The final clustering result is usually compared only to some other heuristic strategy. In this paper we compare the result to the solution provided by a brute-force experiment. Many instances of the k-means can be executed in parallel on the high performance computing infrastructure, which makes brute-force search for the best initial centroids possible. Solutions obtained by exact solvers [2, 11] of the clustering problem are used for verification of the brute-force approach. We present progress of the function optimization during the experiment for several benchmark data sets, including sparse document-term matrices. © Springer Nature Singapore Pte Ltd. 2017.

Název v anglickém jazyce

Comparison of K-means clustering initialization approaches with brute-force initialization

Popis výsledku anglicky

Data clustering is a basic data mining discipline that has been in center of interest of many research groups. This paper describes the formulation of the basic NP-hard optimization problem in data clustering which is approximated by many heuristic methods. The famous k-means clustering algorithm and its initialization is of a particular interest in this paper. A summary of the k-means variants and various initialization strategies is presented. Many initialization heuristics tend to search only through a fraction of the initial centroid space. The final clustering result is usually compared only to some other heuristic strategy. In this paper we compare the result to the solution provided by a brute-force experiment. Many instances of the k-means can be executed in parallel on the high performance computing infrastructure, which makes brute-force search for the best initial centroids possible. Solutions obtained by exact solvers [2, 11] of the clustering problem are used for verification of the brute-force approach. We present progress of the function optimization during the experiment for several benchmark data sets, including sparse document-term matrices. © Springer Nature Singapore Pte Ltd. 2017.

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Projekt

<a href="/cs/project/LQ1602" target="_blank" >LQ1602: IT4Innovations excellence in science</a><br>

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)<br>S - Specificky vyzkum na vysokych skolach

Rok uplatnění

2017

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

Advances in Intelligent Systems and Computing. Volume 567

ISBN

978-981-10-3408-4

ISSN

2194-5357

e-ISSN

neuvedeno

Počet stran výsledku

12

Strana od-do

103-114

Název nakladatele

Springer

Místo vydání

Singapur

Místo konání akce

Kalkata

Datum konání akce

12. 8. 2016

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

—