Learning optimal classifier chains for real-time big data mining

Jie Xu, Cem Tekin, Mihaela Van Der Schaar

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

7 Citations (Scopus)

Abstract

A plethora of emerging Big Data applications require processing and analyzing streams of data to extract valuable information in real-time. For this, chains of classifiers which can detect various concepts need to be constructed in real-time. In this paper, we propose online distributed algorithms which can learn how to construct the optimal classifier chain in order to maximize the stream mining performance (i.e. mining accuracy minus cost) based on the dynamically-changing data characteristics. The proposed solution does not require the distributed local classifiers to exchange any information when learning at runtime. Moreover, our algorithm requires only limited feedback of the mining performance to enable the learning of the optimal classifier chain. We model the learning problem of the optimal classifier chain at run-time as a multi-player multi-armed bandit problem with limited feedback. To our best knowledge, this paper is the first that applies bandit techniques to stream mining problems. However, existing bandit algorithms are inefficient in the considered scenario due to the fact that each component classifier learns its optimal classification functions using only the aggregate overall reward without knowing its own individual reward and without exchanging information with other classifiers. We prove that the proposed algorithms achieve logarithmic learning regret uniformly over time and hence, they are order optimal. Therefore, the long-term time average performance loss tends to zero. We also design learning algorithms whose regret is linear in the number of classification functions. This is much smaller than the regret results which can be obtained using existing bandit algorithms that scale linearly in the number of classifier chains and exponentially in the number of classification functions.

Original languageEnglish (US)
Title of host publication2013 51st Annual Allerton Conference on Communication, Control, and Computing, Allerton 2013
PublisherIEEE Computer Society
Pages512-519
Number of pages8
ISBN (Print)9781479934096
DOIs
StatePublished - 2013
Externally publishedYes
Event51st Annual Allerton Conference on Communication, Control, and Computing, Allerton 2013 - Monticello, IL, United States
Duration: Oct 2 2013Oct 4 2013

Other

Other51st Annual Allerton Conference on Communication, Control, and Computing, Allerton 2013
CountryUnited States
CityMonticello, IL
Period10/2/1310/4/13

Fingerprint

Data mining
Classifiers
Feedback
Big data
Parallel algorithms
Learning algorithms
Processing
Costs

ASJC Scopus subject areas

  • Computer Networks and Communications
  • Control and Systems Engineering

Cite this

Xu, J., Tekin, C., & Van Der Schaar, M. (2013). Learning optimal classifier chains for real-time big data mining. In 2013 51st Annual Allerton Conference on Communication, Control, and Computing, Allerton 2013 (pp. 512-519). [6736568] IEEE Computer Society. https://doi.org/10.1109/Allerton.2013.6736568

Learning optimal classifier chains for real-time big data mining. / Xu, Jie; Tekin, Cem; Van Der Schaar, Mihaela.

2013 51st Annual Allerton Conference on Communication, Control, and Computing, Allerton 2013. IEEE Computer Society, 2013. p. 512-519 6736568.

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

Xu, J, Tekin, C & Van Der Schaar, M 2013, Learning optimal classifier chains for real-time big data mining. in 2013 51st Annual Allerton Conference on Communication, Control, and Computing, Allerton 2013., 6736568, IEEE Computer Society, pp. 512-519, 51st Annual Allerton Conference on Communication, Control, and Computing, Allerton 2013, Monticello, IL, United States, 10/2/13. https://doi.org/10.1109/Allerton.2013.6736568
Xu J, Tekin C, Van Der Schaar M. Learning optimal classifier chains for real-time big data mining. In 2013 51st Annual Allerton Conference on Communication, Control, and Computing, Allerton 2013. IEEE Computer Society. 2013. p. 512-519. 6736568 https://doi.org/10.1109/Allerton.2013.6736568
Xu, Jie ; Tekin, Cem ; Van Der Schaar, Mihaela. / Learning optimal classifier chains for real-time big data mining. 2013 51st Annual Allerton Conference on Communication, Control, and Computing, Allerton 2013. IEEE Computer Society, 2013. pp. 512-519
@inproceedings{41b6a3b844f24fd2bac205bb06533d08,
title = "Learning optimal classifier chains for real-time big data mining",
abstract = "A plethora of emerging Big Data applications require processing and analyzing streams of data to extract valuable information in real-time. For this, chains of classifiers which can detect various concepts need to be constructed in real-time. In this paper, we propose online distributed algorithms which can learn how to construct the optimal classifier chain in order to maximize the stream mining performance (i.e. mining accuracy minus cost) based on the dynamically-changing data characteristics. The proposed solution does not require the distributed local classifiers to exchange any information when learning at runtime. Moreover, our algorithm requires only limited feedback of the mining performance to enable the learning of the optimal classifier chain. We model the learning problem of the optimal classifier chain at run-time as a multi-player multi-armed bandit problem with limited feedback. To our best knowledge, this paper is the first that applies bandit techniques to stream mining problems. However, existing bandit algorithms are inefficient in the considered scenario due to the fact that each component classifier learns its optimal classification functions using only the aggregate overall reward without knowing its own individual reward and without exchanging information with other classifiers. We prove that the proposed algorithms achieve logarithmic learning regret uniformly over time and hence, they are order optimal. Therefore, the long-term time average performance loss tends to zero. We also design learning algorithms whose regret is linear in the number of classification functions. This is much smaller than the regret results which can be obtained using existing bandit algorithms that scale linearly in the number of classifier chains and exponentially in the number of classification functions.",
author = "Jie Xu and Cem Tekin and {Van Der Schaar}, Mihaela",
year = "2013",
doi = "10.1109/Allerton.2013.6736568",
language = "English (US)",
isbn = "9781479934096",
pages = "512--519",
booktitle = "2013 51st Annual Allerton Conference on Communication, Control, and Computing, Allerton 2013",
publisher = "IEEE Computer Society",

}

TY - GEN

T1 - Learning optimal classifier chains for real-time big data mining

AU - Xu, Jie

AU - Tekin, Cem

AU - Van Der Schaar, Mihaela

PY - 2013

Y1 - 2013

N2 - A plethora of emerging Big Data applications require processing and analyzing streams of data to extract valuable information in real-time. For this, chains of classifiers which can detect various concepts need to be constructed in real-time. In this paper, we propose online distributed algorithms which can learn how to construct the optimal classifier chain in order to maximize the stream mining performance (i.e. mining accuracy minus cost) based on the dynamically-changing data characteristics. The proposed solution does not require the distributed local classifiers to exchange any information when learning at runtime. Moreover, our algorithm requires only limited feedback of the mining performance to enable the learning of the optimal classifier chain. We model the learning problem of the optimal classifier chain at run-time as a multi-player multi-armed bandit problem with limited feedback. To our best knowledge, this paper is the first that applies bandit techniques to stream mining problems. However, existing bandit algorithms are inefficient in the considered scenario due to the fact that each component classifier learns its optimal classification functions using only the aggregate overall reward without knowing its own individual reward and without exchanging information with other classifiers. We prove that the proposed algorithms achieve logarithmic learning regret uniformly over time and hence, they are order optimal. Therefore, the long-term time average performance loss tends to zero. We also design learning algorithms whose regret is linear in the number of classification functions. This is much smaller than the regret results which can be obtained using existing bandit algorithms that scale linearly in the number of classifier chains and exponentially in the number of classification functions.

AB - A plethora of emerging Big Data applications require processing and analyzing streams of data to extract valuable information in real-time. For this, chains of classifiers which can detect various concepts need to be constructed in real-time. In this paper, we propose online distributed algorithms which can learn how to construct the optimal classifier chain in order to maximize the stream mining performance (i.e. mining accuracy minus cost) based on the dynamically-changing data characteristics. The proposed solution does not require the distributed local classifiers to exchange any information when learning at runtime. Moreover, our algorithm requires only limited feedback of the mining performance to enable the learning of the optimal classifier chain. We model the learning problem of the optimal classifier chain at run-time as a multi-player multi-armed bandit problem with limited feedback. To our best knowledge, this paper is the first that applies bandit techniques to stream mining problems. However, existing bandit algorithms are inefficient in the considered scenario due to the fact that each component classifier learns its optimal classification functions using only the aggregate overall reward without knowing its own individual reward and without exchanging information with other classifiers. We prove that the proposed algorithms achieve logarithmic learning regret uniformly over time and hence, they are order optimal. Therefore, the long-term time average performance loss tends to zero. We also design learning algorithms whose regret is linear in the number of classification functions. This is much smaller than the regret results which can be obtained using existing bandit algorithms that scale linearly in the number of classifier chains and exponentially in the number of classification functions.

UR - http://www.scopus.com/inward/record.url?scp=84897723649&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84897723649&partnerID=8YFLogxK

U2 - 10.1109/Allerton.2013.6736568

DO - 10.1109/Allerton.2013.6736568

M3 - Conference contribution

SN - 9781479934096

SP - 512

EP - 519

BT - 2013 51st Annual Allerton Conference on Communication, Control, and Computing, Allerton 2013

PB - IEEE Computer Society

ER -