Publication Type

Conference Proceeding Article

Publication Date

2-2017

Abstract

This paper investigates the problem of online activelearning for training classification models from sequentiallyarriving data. This is more challenging than conventional onlinelearning tasks since the learner not only needs to figure outhow to effectively update the classifier but also needs to decidewhen is the best time to query the label of an incoming instancegiven limited label budget. The existing online active learningapproaches are often based on first-order online learning methodswhich generally fall short in slow convergence rate and suboptimalexploitation of available information when queryingthe labeled data. To overcome the limitations, in this paper,we present a new framework of Second-order Online ActiveLearning (SOAL), which fully exploits both first-order andsecond-order information to achieve high learning accuracy withlow labeling cost. We conduct both theoretical analysis andempirical studies for evaluating the proposed SOAL algorithmextensively. The encouraging results show clear advantages of theproposed algorithm over a family of state-of-the-art online activelearning algorithms

Keywords

online learning, active learning, machine learning

Discipline

Databases and Information Systems | Online and Distance Education

Research Areas

Data Management and Analytics

Publication

ICDM2016: Proceedings of IEEE International Conference on Data Mining: Barcelona, Spain, 2016 December 12-15

First Page

931

Last Page

936

ISBN

9781509054749

Identifier

10.1109/ICDM.2016.0115

Publisher

IEEE

City or Country

USA

Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.

Additional URL

1

Share

COinS