Classification active learning based on mutual information

Jamshid Sourati, Murat Akcakaya, Jennifer G. Dy, Todd K. Leen, Deniz Erdogmus

Research output: Contribution to journalArticle

12 Citations (Scopus)

Abstract

Selecting a subset of samples to label from a large pool of unlabeled data points, such that a sufficiently accurate classifier is obtained using a reasonably small training set is a challenging, yet critical problem. Challenging, since solving this problem includes cumbersome combinatorial computations, and critical, due to the fact that labeling is an expensive and time-consuming task, hence we always aim to minimize the number of required labels. While information theoretical objectives, such as mutual information (MI) between the labels, have been successfully used in sequential querying, it is not straightforward to generalize these objectives to batch mode. This is because evaluation and optimization of functions which are trivial in individual querying settings become intractable for many objectives when we are to select multiple queries. In this paper, we develop a framework, where we propose efficient ways of evaluating and maximizing the MI between labels as an objective for batch mode active learning. Our proposed framework efficiently reduces the computational complexity from an order proportional to the batch size, when no approximation is applied, to the linear cost. The performance of this framework is evaluated using data sets from several fields showing that the proposed framework leads to efficient active learning for most of the data sets.

Original languageEnglish (US)
Article number51
JournalEntropy
Volume18
Issue number2
DOIs
StatePublished - 2016
Externally publishedYes

Fingerprint

learning
problem solving
classifiers
marking
set theory
education
costs
optimization
evaluation
approximation

Keywords

  • Active learning
  • Classification
  • Mutual information
  • Submodular maximization

ASJC Scopus subject areas

  • Physics and Astronomy(all)

Cite this

Sourati, J., Akcakaya, M., Dy, J. G., Leen, T. K., & Erdogmus, D. (2016). Classification active learning based on mutual information. Entropy, 18(2), [51]. https://doi.org/10.3390/e18020051

Classification active learning based on mutual information. / Sourati, Jamshid; Akcakaya, Murat; Dy, Jennifer G.; Leen, Todd K.; Erdogmus, Deniz.

In: Entropy, Vol. 18, No. 2, 51, 2016.

Research output: Contribution to journalArticle

Sourati, J, Akcakaya, M, Dy, JG, Leen, TK & Erdogmus, D 2016, 'Classification active learning based on mutual information', Entropy, vol. 18, no. 2, 51. https://doi.org/10.3390/e18020051
Sourati J, Akcakaya M, Dy JG, Leen TK, Erdogmus D. Classification active learning based on mutual information. Entropy. 2016;18(2). 51. https://doi.org/10.3390/e18020051
Sourati, Jamshid ; Akcakaya, Murat ; Dy, Jennifer G. ; Leen, Todd K. ; Erdogmus, Deniz. / Classification active learning based on mutual information. In: Entropy. 2016 ; Vol. 18, No. 2.
@article{38965f3385014b0da50cdc0070e9d640,
title = "Classification active learning based on mutual information",
abstract = "Selecting a subset of samples to label from a large pool of unlabeled data points, such that a sufficiently accurate classifier is obtained using a reasonably small training set is a challenging, yet critical problem. Challenging, since solving this problem includes cumbersome combinatorial computations, and critical, due to the fact that labeling is an expensive and time-consuming task, hence we always aim to minimize the number of required labels. While information theoretical objectives, such as mutual information (MI) between the labels, have been successfully used in sequential querying, it is not straightforward to generalize these objectives to batch mode. This is because evaluation and optimization of functions which are trivial in individual querying settings become intractable for many objectives when we are to select multiple queries. In this paper, we develop a framework, where we propose efficient ways of evaluating and maximizing the MI between labels as an objective for batch mode active learning. Our proposed framework efficiently reduces the computational complexity from an order proportional to the batch size, when no approximation is applied, to the linear cost. The performance of this framework is evaluated using data sets from several fields showing that the proposed framework leads to efficient active learning for most of the data sets.",
keywords = "Active learning, Classification, Mutual information, Submodular maximization",
author = "Jamshid Sourati and Murat Akcakaya and Dy, {Jennifer G.} and Leen, {Todd K.} and Deniz Erdogmus",
year = "2016",
doi = "10.3390/e18020051",
language = "English (US)",
volume = "18",
journal = "Entropy",
issn = "1099-4300",
publisher = "Multidisciplinary Digital Publishing Institute (MDPI)",
number = "2",

}

TY - JOUR

T1 - Classification active learning based on mutual information

AU - Sourati, Jamshid

AU - Akcakaya, Murat

AU - Dy, Jennifer G.

AU - Leen, Todd K.

AU - Erdogmus, Deniz

PY - 2016

Y1 - 2016

N2 - Selecting a subset of samples to label from a large pool of unlabeled data points, such that a sufficiently accurate classifier is obtained using a reasonably small training set is a challenging, yet critical problem. Challenging, since solving this problem includes cumbersome combinatorial computations, and critical, due to the fact that labeling is an expensive and time-consuming task, hence we always aim to minimize the number of required labels. While information theoretical objectives, such as mutual information (MI) between the labels, have been successfully used in sequential querying, it is not straightforward to generalize these objectives to batch mode. This is because evaluation and optimization of functions which are trivial in individual querying settings become intractable for many objectives when we are to select multiple queries. In this paper, we develop a framework, where we propose efficient ways of evaluating and maximizing the MI between labels as an objective for batch mode active learning. Our proposed framework efficiently reduces the computational complexity from an order proportional to the batch size, when no approximation is applied, to the linear cost. The performance of this framework is evaluated using data sets from several fields showing that the proposed framework leads to efficient active learning for most of the data sets.

AB - Selecting a subset of samples to label from a large pool of unlabeled data points, such that a sufficiently accurate classifier is obtained using a reasonably small training set is a challenging, yet critical problem. Challenging, since solving this problem includes cumbersome combinatorial computations, and critical, due to the fact that labeling is an expensive and time-consuming task, hence we always aim to minimize the number of required labels. While information theoretical objectives, such as mutual information (MI) between the labels, have been successfully used in sequential querying, it is not straightforward to generalize these objectives to batch mode. This is because evaluation and optimization of functions which are trivial in individual querying settings become intractable for many objectives when we are to select multiple queries. In this paper, we develop a framework, where we propose efficient ways of evaluating and maximizing the MI between labels as an objective for batch mode active learning. Our proposed framework efficiently reduces the computational complexity from an order proportional to the batch size, when no approximation is applied, to the linear cost. The performance of this framework is evaluated using data sets from several fields showing that the proposed framework leads to efficient active learning for most of the data sets.

KW - Active learning

KW - Classification

KW - Mutual information

KW - Submodular maximization

UR - http://www.scopus.com/inward/record.url?scp=84960455436&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84960455436&partnerID=8YFLogxK

U2 - 10.3390/e18020051

DO - 10.3390/e18020051

M3 - Article

AN - SCOPUS:84960455436

VL - 18

JO - Entropy

JF - Entropy

SN - 1099-4300

IS - 2

M1 - 51

ER -