東華大學圖書館 |

Language: English

Help

回圖書館首頁

手機版館藏查詢

Back

Switch To: Labeled | MARC Mode | ISBD

Diversity in neural network ensembles.

Schmidt, Konrad.

Linked to FindBook

Google Book

Amazon

博客來

Diversity in neural network ensembles.

Record Type:	Electronic resources : Monograph/item
Title/Author:	Diversity in neural network ensembles./
Author:	Schmidt, Konrad.
Description:	120 p.
Notes:	Source: Dissertation Abstracts International, Volume: 65-08, Section: B, page: 4116.
Contained By:	Dissertation Abstracts International65-08B.
Subject:	Computer Science. -
Online resource:	http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3117015
ISBN:	0496025864

Diversity in neural network ensembles.
Schmidt, Konrad.

Diversity in neural network ensembles. - 120 p.

Source: Dissertation Abstracts International, Volume: 65-08, Section: B, page: 4116.

Thesis (D.CS.)--Colorado Technical University, 2004.

It is recognized that neural network ensembles generally provide better predictive accuracy than do individual classifiers. A key component of an effective neural network ensemble is diversity among the members of the ensemble. After all, if every member of the ensemble gave the same response to a given input, there is nothing to be gained by using the ensemble. Previous research has explored various methods by which diversity in a neural network ensemble can be measured. All of the previously proposed methods measure diversity based upon the result of new inputs presented to the ensemble. The major difficulty with the previous methods is that they rely on luck or skill to select the proper inputs. This dissertation introduces two new methods by which diversity can be measured, neither of which requires any additional inputs to, nor outputs from, the neural network ensemble. Instead, the methods rely solely on the architecture (nodes, connections, weights) of the component networks, irrespective of the individual inputs or outputs.

ISBN: 0496025864Subjects--Topical Terms:

626642
Computer Science.

Diversity in neural network ensembles.
LDR:02513nmm 2200277 4500 001 1840389
005 20050721103005.5
008 130614s2004 eng d
020 $a 0496025864
035 $a (UnM)AAI3117015
035 $a AAI3117015
040 $a UnM $c UnM
100 1 $a Schmidt, Konrad. $3 1928730
245 1 0 $a Diversity in neural network ensembles.
300 $a 120 p.
500 $a Source: Dissertation Abstracts International, Volume: 65-08, Section: B, page: 4116.
500 $a Chair: Bo I. Sanden.
502 $a Thesis (D.CS.)--Colorado Technical University, 2004.
520 $a It is recognized that neural network ensembles generally provide better predictive accuracy than do individual classifiers. A key component of an effective neural network ensemble is diversity among the members of the ensemble. After all, if every member of the ensemble gave the same response to a given input, there is nothing to be gained by using the ensemble. Previous research has explored various methods by which diversity in a neural network ensemble can be measured. All of the previously proposed methods measure diversity based upon the result of new inputs presented to the ensemble. The major difficulty with the previous methods is that they rely on luck or skill to select the proper inputs. This dissertation introduces two new methods by which diversity can be measured, neither of which requires any additional inputs to, nor outputs from, the neural network ensemble. Instead, the methods rely solely on the architecture (nodes, connections, weights) of the component networks, irrespective of the individual inputs or outputs.
520 $a The hypothesis of this dissertation is that there is a positive correlation between ensemble diversity and accuracy. To prove the hypothesis, many neural networks are trained and randomly grouped into a number of ensembles. A genetic algorithm is employed for one of the methods in order to minimize the diversity calculation. The diversity of the ensembles is measured using both methods, and a chart of diversity vs. accuracy is produced for each method. Finally, a statistical analysis is performed to prove that there exists a positive correlation between accuracy and diversity for each of the newly introduced methods.
590 $a School code: 1271.
650 4 $a Computer Science. $3 626642
690 $a 0984
710 2 0 $a Colorado Technical University. $3 1031996
773 0 $t Dissertation Abstracts International $g 65-08B.
790 1 0 $a Sanden, Bo I., $e advisor
790 $a 1271
791 $a D.CS.
792 $a 2004
856 4 0 $u http://pqdd.sinica.edu.tw/twdaoapp/servlet/advanced?query=3117015