How to make AdaBoost.M1 work for weak base classifiers by changing only one line of the code

Günther Eibl; Karl Peter Pfeiffer

How to make AdaBoost.M1 work for weak base classifiers by changing only one line of the code

Günther Eibl, Karl Peter Pfeiffer

University of Innsbruck

Research output: Contribution to conference › Paper › peer-review

Abstract

If one has a multiclass classification problem and wants to boost a multiclass base classifier AdaBoost.M1 is a well known and widely applicated boosting algorithm. However AdaBoost.M1 does not work, if the base classifier is too weak. We show, that with a modification of only one line of AdaBoost.M1 one can make it usable for weak base classifiers, too. The resulting classifier AdaBoost.M1Wis guaranteed to minimize an upper bound for a performance measure, called the guessing error, as long as the base classifier is better than random guessing. The usability of AdaBoost.M1W could be clearly demonstrated experimentally.

Original language	English
Pages	72-83
Publication status	Published - 2003
Event	European Conference on Machine Learning - Helsinki, Finland Duration: 12 Aug 2002 → …

Conference

Conference	European Conference on Machine Learning
Country/Territory	Finland
Period	12/08/02 → …

Classification according to Österreichische Systematik der Wissenschaftszweige (ÖFOS 2012)

102019 Machine learning

Applied Research Level (ARL)

ARL Level 2 - Description of the application of a principle

Research focus/foci

Not applicable

Cite this

@conference{6b6da693a88043b2a40ffc033b076fd1,

title = "How to make AdaBoost.M1 work for weak base classifiers by changing only one line of the code",

abstract = "If one has a multiclass classification problem and wants to boost a multiclass base classifier AdaBoost.M1 is a well known and widely applicated boosting algorithm. However AdaBoost.M1 does not work, if the base classifier is too weak. We show, that with a modification of only one line of AdaBoost.M1 one can make it usable for weak base classifiers, too. The resulting classifier AdaBoost.M1Wis guaranteed to minimize an upper bound for a performance measure, called the guessing error, as long as the base classifier is better than random guessing. The usability of AdaBoost.M1W could be clearly demonstrated experimentally.",

keywords = "boosting, machine learning, artificial intelligence",

author = "G{\"u}nther Eibl and Pfeiffer, \{Karl Peter\}",

year = "2003",

language = "English",

pages = "72--83",

note = "European Conference on Machine Learning ; Conference date: 12-08-2002",

}

TY - CONF

T1 - How to make AdaBoost.M1 work for weak base classifiers by changing only one line of the code

AU - Eibl, Günther

AU - Pfeiffer, Karl Peter

PY - 2003

Y1 - 2003

N2 - If one has a multiclass classification problem and wants to boost a multiclass base classifier AdaBoost.M1 is a well known and widely applicated boosting algorithm. However AdaBoost.M1 does not work, if the base classifier is too weak. We show, that with a modification of only one line of AdaBoost.M1 one can make it usable for weak base classifiers, too. The resulting classifier AdaBoost.M1Wis guaranteed to minimize an upper bound for a performance measure, called the guessing error, as long as the base classifier is better than random guessing. The usability of AdaBoost.M1W could be clearly demonstrated experimentally.

AB - If one has a multiclass classification problem and wants to boost a multiclass base classifier AdaBoost.M1 is a well known and widely applicated boosting algorithm. However AdaBoost.M1 does not work, if the base classifier is too weak. We show, that with a modification of only one line of AdaBoost.M1 one can make it usable for weak base classifiers, too. The resulting classifier AdaBoost.M1Wis guaranteed to minimize an upper bound for a performance measure, called the guessing error, as long as the base classifier is better than random guessing. The usability of AdaBoost.M1W could be clearly demonstrated experimentally.

KW - boosting

KW - machine learning

KW - artificial intelligence

M3 - Paper

SP - 72

EP - 83

T2 - European Conference on Machine Learning

Y2 - 12 August 2002

ER -