Studying active learning in the cost-sensitive framework

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

3 Scopus citations

Abstract

Active learning is a learning paradigm that actively acquires extra information with an "effort" for a certain "gain" when building learning models. This paper unifies the effort and gain by studying active learning in the cost-sensitive framework. The major advantage of studying active cost-sensitive learning aims at the business goal of minimizing the total cost directly, thus the potential applications of the proposed methods are significant. We first study a simple random active learner "buying" additional examples at random in order to reduce the total cost of example acquisition and future misclassifications. Then we propose a novel pool-based cost-sensitive active learner "buying" labels of unlabeled examples in a pool. We evaluate our new cost-sensitive active learning algorithms and compare them to previous active cost-sensitive learning methods. Experiment results show that our pool-based cost-sensitive active learner requires a fewer number of examples yet it produces a smaller total cost compared to the previous methods.

Original languageEnglish
Title of host publicationProceedings of the 45th Annual Hawaii International Conference on System Sciences, HICSS-45
PublisherIEEE Computer Society
Pages1097-1106
Number of pages10
ISBN (Print)9780769545257
DOIs
StatePublished - 2012
Event2012 45th Hawaii International Conference on System Sciences, HICSS 2012 - Maui, HI, United States
Duration: Jan 4 2012Jan 7 2012

Publication series

NameProceedings of the Annual Hawaii International Conference on System Sciences
ISSN (Print)1530-1605

Conference

Conference2012 45th Hawaii International Conference on System Sciences, HICSS 2012
CountryUnited States
CityMaui, HI
Period01/4/1201/7/12

Fingerprint Dive into the research topics of 'Studying active learning in the cost-sensitive framework'. Together they form a unique fingerprint.

Cite this