Incremental learning of new classes from unbalanced data

Gregory Ditzler, Gail Rosen, Robi Polikar

Research output: Chapter in Book/Report/Conference proceedingConference contribution

5 Citations (Scopus)

Abstract

Multiple classifier systems tend to suffer from outvoting when new concept classes need to be learned incrementally. Out-voting is primarily due to existing classifiers being unable to recognize the new class until there is a sufficient number of new classifiers that can influence the ensemble decision. This problem of learning new classes was explicitly addressed in Learn ++.NC, our previous work, where ensemble members dynamically adjust their own weights by consulting with each other based on their individual and collective confidence in classifying each concept class. Learn++.NC works remarkably well for learning new concept classes while requiring few ensemble members to do so. Learn++.NC cannot cope with the class imbalance problem, however, as it was not designed to do so. Yet, class imbalance is a common and important problem in machine learning, made even more challenging in an incremental learning setting. In this paper, we extend Learn++.NC so that it can incrementally learn new concept classes even if their instances are drawn from severely imbalanced class distributions. We show that the proposed algorithm is quite robust compared to other state-of-the-art algorithms.

Original languageEnglish (US)
Title of host publication2013 International Joint Conference on Neural Networks, IJCNN 2013
DOIs
StatePublished - Dec 1 2013
Externally publishedYes
Event2013 International Joint Conference on Neural Networks, IJCNN 2013 - Dallas, TX, United States
Duration: Aug 4 2013Aug 9 2013

Publication series

NameProceedings of the International Joint Conference on Neural Networks

Other

Other2013 International Joint Conference on Neural Networks, IJCNN 2013
CountryUnited States
CityDallas, TX
Period8/4/138/9/13

Fingerprint

Classifiers
Learning systems

Keywords

  • incremental learning
  • multiple classifier systems
  • unbalanced data

ASJC Scopus subject areas

  • Software
  • Artificial Intelligence

Cite this

Ditzler, G., Rosen, G., & Polikar, R. (2013). Incremental learning of new classes from unbalanced data. In 2013 International Joint Conference on Neural Networks, IJCNN 2013 [6706770] (Proceedings of the International Joint Conference on Neural Networks). https://doi.org/10.1109/IJCNN.2013.6706770

Incremental learning of new classes from unbalanced data. / Ditzler, Gregory; Rosen, Gail; Polikar, Robi.

2013 International Joint Conference on Neural Networks, IJCNN 2013. 2013. 6706770 (Proceedings of the International Joint Conference on Neural Networks).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Ditzler, G, Rosen, G & Polikar, R 2013, Incremental learning of new classes from unbalanced data. in 2013 International Joint Conference on Neural Networks, IJCNN 2013., 6706770, Proceedings of the International Joint Conference on Neural Networks, 2013 International Joint Conference on Neural Networks, IJCNN 2013, Dallas, TX, United States, 8/4/13. https://doi.org/10.1109/IJCNN.2013.6706770
Ditzler G, Rosen G, Polikar R. Incremental learning of new classes from unbalanced data. In 2013 International Joint Conference on Neural Networks, IJCNN 2013. 2013. 6706770. (Proceedings of the International Joint Conference on Neural Networks). https://doi.org/10.1109/IJCNN.2013.6706770
Ditzler, Gregory ; Rosen, Gail ; Polikar, Robi. / Incremental learning of new classes from unbalanced data. 2013 International Joint Conference on Neural Networks, IJCNN 2013. 2013. (Proceedings of the International Joint Conference on Neural Networks).
@inproceedings{332456421d974c8f9ca6754400a2cb7f,
title = "Incremental learning of new classes from unbalanced data",
abstract = "Multiple classifier systems tend to suffer from outvoting when new concept classes need to be learned incrementally. Out-voting is primarily due to existing classifiers being unable to recognize the new class until there is a sufficient number of new classifiers that can influence the ensemble decision. This problem of learning new classes was explicitly addressed in Learn ++.NC, our previous work, where ensemble members dynamically adjust their own weights by consulting with each other based on their individual and collective confidence in classifying each concept class. Learn++.NC works remarkably well for learning new concept classes while requiring few ensemble members to do so. Learn++.NC cannot cope with the class imbalance problem, however, as it was not designed to do so. Yet, class imbalance is a common and important problem in machine learning, made even more challenging in an incremental learning setting. In this paper, we extend Learn++.NC so that it can incrementally learn new concept classes even if their instances are drawn from severely imbalanced class distributions. We show that the proposed algorithm is quite robust compared to other state-of-the-art algorithms.",
keywords = "incremental learning, multiple classifier systems, unbalanced data",
author = "Gregory Ditzler and Gail Rosen and Robi Polikar",
year = "2013",
month = "12",
day = "1",
doi = "10.1109/IJCNN.2013.6706770",
language = "English (US)",
isbn = "9781467361293",
series = "Proceedings of the International Joint Conference on Neural Networks",
booktitle = "2013 International Joint Conference on Neural Networks, IJCNN 2013",

}

TY - GEN

T1 - Incremental learning of new classes from unbalanced data

AU - Ditzler, Gregory

AU - Rosen, Gail

AU - Polikar, Robi

PY - 2013/12/1

Y1 - 2013/12/1

N2 - Multiple classifier systems tend to suffer from outvoting when new concept classes need to be learned incrementally. Out-voting is primarily due to existing classifiers being unable to recognize the new class until there is a sufficient number of new classifiers that can influence the ensemble decision. This problem of learning new classes was explicitly addressed in Learn ++.NC, our previous work, where ensemble members dynamically adjust their own weights by consulting with each other based on their individual and collective confidence in classifying each concept class. Learn++.NC works remarkably well for learning new concept classes while requiring few ensemble members to do so. Learn++.NC cannot cope with the class imbalance problem, however, as it was not designed to do so. Yet, class imbalance is a common and important problem in machine learning, made even more challenging in an incremental learning setting. In this paper, we extend Learn++.NC so that it can incrementally learn new concept classes even if their instances are drawn from severely imbalanced class distributions. We show that the proposed algorithm is quite robust compared to other state-of-the-art algorithms.

AB - Multiple classifier systems tend to suffer from outvoting when new concept classes need to be learned incrementally. Out-voting is primarily due to existing classifiers being unable to recognize the new class until there is a sufficient number of new classifiers that can influence the ensemble decision. This problem of learning new classes was explicitly addressed in Learn ++.NC, our previous work, where ensemble members dynamically adjust their own weights by consulting with each other based on their individual and collective confidence in classifying each concept class. Learn++.NC works remarkably well for learning new concept classes while requiring few ensemble members to do so. Learn++.NC cannot cope with the class imbalance problem, however, as it was not designed to do so. Yet, class imbalance is a common and important problem in machine learning, made even more challenging in an incremental learning setting. In this paper, we extend Learn++.NC so that it can incrementally learn new concept classes even if their instances are drawn from severely imbalanced class distributions. We show that the proposed algorithm is quite robust compared to other state-of-the-art algorithms.

KW - incremental learning

KW - multiple classifier systems

KW - unbalanced data

UR - http://www.scopus.com/inward/record.url?scp=84893536538&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84893536538&partnerID=8YFLogxK

U2 - 10.1109/IJCNN.2013.6706770

DO - 10.1109/IJCNN.2013.6706770

M3 - Conference contribution

AN - SCOPUS:84893536538

SN - 9781467361293

T3 - Proceedings of the International Joint Conference on Neural Networks

BT - 2013 International Joint Conference on Neural Networks, IJCNN 2013

ER -