Dynamically weighted majority voting for incremental learning and comparison of three boosting based approaches

Aliasgar Gangardiwala; Robi Polikar

doi:10.1109/IJCNN.2005.1556012

Dynamically weighted majority voting for incremental learning and comparison of three boosting based approaches

Aliasgar Gangardiwala, Robi Polikar

Electrical Engineering

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

18 Scopus citations

Abstract

We have previously introduced Learn++, an ensemble based incremental learning algorithm for acquiring new knowledge from data that later become available, even when such data introduce new classes. In this paper, we describe a modification to this algorithm, where the voting weights of the classifiers are updated dynamically based on the location of the test input in the feature space. The new algorithm provides improved performance, stronger immunity to catastrophic forgetting and finer balance to the stability-plasticity dilemma than its predecessor, particularly when new classes are introduced. The modified algorithm and its performance, as compared to Adaboost.M1 and the original Learn++, on real and benchmark datasets are presented.

Original language	English (US)
Title of host publication	Proceedings of the International Joint Conference on Neural Networks, IJCNN 2005
Pages	1131-1136
Number of pages	6
DOIs	https://doi.org/10.1109/IJCNN.2005.1556012
State	Published - Dec 1 2005
Event	International Joint Conference on Neural Networks, IJCNN 2005 - Montreal, QC, Canada Duration: Jul 31 2005 → Aug 4 2005

Publication series

Name	Proceedings of the International Joint Conference on Neural Networks
Volume	2

Other

Other	International Joint Conference on Neural Networks, IJCNN 2005
Country	Canada
City	Montreal, QC
Period	7/31/05 → 8/4/05

All Science Journal Classification (ASJC) codes

Software
Artificial Intelligence

Access to Document

10.1109/IJCNN.2005.1556012

Fingerprint Dive into the research topics of 'Dynamically weighted majority voting for incremental learning and comparison of three boosting based approaches'. Together they form a unique fingerprint.

Cite this

Gangardiwala, A., & Polikar, R. (2005). Dynamically weighted majority voting for incremental learning and comparison of three boosting based approaches. In Proceedings of the International Joint Conference on Neural Networks, IJCNN 2005 (pp. 1131-1136). [1556012] (Proceedings of the International Joint Conference on Neural Networks; Vol. 2). https://doi.org/10.1109/IJCNN.2005.1556012

@inproceedings{62bf30d70b9d4c71a8d0c6fc465928db,

title = "Dynamically weighted majority voting for incremental learning and comparison of three boosting based approaches",

abstract = "We have previously introduced Learn++, an ensemble based incremental learning algorithm for acquiring new knowledge from data that later become available, even when such data introduce new classes. In this paper, we describe a modification to this algorithm, where the voting weights of the classifiers are updated dynamically based on the location of the test input in the feature space. The new algorithm provides improved performance, stronger immunity to catastrophic forgetting and finer balance to the stability-plasticity dilemma than its predecessor, particularly when new classes are introduced. The modified algorithm and its performance, as compared to Adaboost.M1 and the original Learn++, on real and benchmark datasets are presented.",

author = "Aliasgar Gangardiwala and Robi Polikar",

year = "2005",

month = dec,

day = "1",

doi = "10.1109/IJCNN.2005.1556012",

language = "English (US)",

isbn = "0780390482",

series = "Proceedings of the International Joint Conference on Neural Networks",

pages = "1131--1136",

booktitle = "Proceedings of the International Joint Conference on Neural Networks, IJCNN 2005",

note = "International Joint Conference on Neural Networks, IJCNN 2005 ; Conference date: 31-07-2005 Through 04-08-2005",

}

Gangardiwala, A & Polikar, R 2005, Dynamically weighted majority voting for incremental learning and comparison of three boosting based approaches. in Proceedings of the International Joint Conference on Neural Networks, IJCNN 2005., 1556012, Proceedings of the International Joint Conference on Neural Networks, vol. 2, pp. 1131-1136, International Joint Conference on Neural Networks, IJCNN 2005, Montreal, QC, Canada, 7/31/05. https://doi.org/10.1109/IJCNN.2005.1556012

Dynamically weighted majority voting for incremental learning and comparison of three boosting based approaches. / Gangardiwala, Aliasgar; Polikar, Robi.

Proceedings of the International Joint Conference on Neural Networks, IJCNN 2005. 2005. p. 1131-1136 1556012 (Proceedings of the International Joint Conference on Neural Networks; Vol. 2).

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

TY - GEN

T1 - Dynamically weighted majority voting for incremental learning and comparison of three boosting based approaches

AU - Gangardiwala, Aliasgar

AU - Polikar, Robi

PY - 2005/12/1

Y1 - 2005/12/1

N2 - We have previously introduced Learn++, an ensemble based incremental learning algorithm for acquiring new knowledge from data that later become available, even when such data introduce new classes. In this paper, we describe a modification to this algorithm, where the voting weights of the classifiers are updated dynamically based on the location of the test input in the feature space. The new algorithm provides improved performance, stronger immunity to catastrophic forgetting and finer balance to the stability-plasticity dilemma than its predecessor, particularly when new classes are introduced. The modified algorithm and its performance, as compared to Adaboost.M1 and the original Learn++, on real and benchmark datasets are presented.

AB - We have previously introduced Learn++, an ensemble based incremental learning algorithm for acquiring new knowledge from data that later become available, even when such data introduce new classes. In this paper, we describe a modification to this algorithm, where the voting weights of the classifiers are updated dynamically based on the location of the test input in the feature space. The new algorithm provides improved performance, stronger immunity to catastrophic forgetting and finer balance to the stability-plasticity dilemma than its predecessor, particularly when new classes are introduced. The modified algorithm and its performance, as compared to Adaboost.M1 and the original Learn++, on real and benchmark datasets are presented.

UR - http://www.scopus.com/inward/record.url?scp=33745950074&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=33745950074&partnerID=8YFLogxK

U2 - 10.1109/IJCNN.2005.1556012

DO - 10.1109/IJCNN.2005.1556012

M3 - Conference contribution

AN - SCOPUS:33745950074

SN - 0780390482

SN - 9780780390485

T3 - Proceedings of the International Joint Conference on Neural Networks

SP - 1131

EP - 1136

BT - Proceedings of the International Joint Conference on Neural Networks, IJCNN 2005

T2 - International Joint Conference on Neural Networks, IJCNN 2005

Y2 - 31 July 2005 through 4 August 2005

ER -