Link to the University of Pittsburgh Homepage
Link to the University Library System Homepage Link to the Contact Us Form

Active Learning of Classification Models from Enriched Label-related Feedback

Xue, Yanbing (2020) Active Learning of Classification Models from Enriched Label-related Feedback. Doctoral Dissertation, University of Pittsburgh. (Unpublished)

This is the latest version of this item.

[img]
Preview
PDF
Download (4MB) | Preview

Abstract

Our ability to learn accurate classification models from data is often limited by the number of available labeled data instances. This limitation is of particular concern when data instances need to be manually labeled by human annotators and when the labeling process carries a significant cost. Recent years witnessed increased research interest in developing methods in different directions capable of learning models from a smaller number of examples. One such direction is active learning, which finds the most informative unlabeled instances to be labeled next. Another, more recent direction showing a great promise utilizes enriched label-related feedback. In this case, such feedback from the human annotator provides additional information reflecting the relations among possible labels. The cost of such feedback is often negligible compared with the cost of instance review. The enriched label-related feedback may come in different forms. In this work, we propose, develop and study classification models for binary, multi-class and multi-label classification problems that utilize the different forms of enriched label-related feedback. We show that this new feedback can help us improve the quality of classification models compared with the standard class-label feedback. For each of the studied feedback forms, we also develop new active learning strategies for selecting the most informative unlabeled instances that are compatible with the respective feedback form, effectively combining two approaches for reducing the number of required labeled instances. We demonstrate the effectiveness of our new framework on both simulated and real-world datasets.


Share

Citation/Export:
Social Networking:
Share |

Details

Item Type: University of Pittsburgh ETD
Status: Unpublished
Creators/Authors:
CreatorsEmailPitt UsernameORCID
Xue, Yanbingyax14@pitt.eduyax14
ETD Committee:
TitleMemberEmail AddressPitt UsernameORCID
Committee ChairHauskrecht, Milosmilos@cs.pitt.edumilos
Committee MemberLitman, Dianedlitman@pitt.edudlitman
Committee MemberKovashka, Adrianakovashka@cs.pitt.eduaik85
Committee MemberVisweswaran, Shyamshv3@pitt.edushv3
Date: 16 September 2020
Date Type: Publication
Defense Date: 28 June 2020
Approval Date: 16 September 2020
Submission Date: 5 August 2020
Access Restriction: No restriction; Release the ETD for access worldwide immediately.
Number of Pages: 142
Institution: University of Pittsburgh
Schools and Programs: Dietrich School of Arts and Sciences > Computer Science
Degree: PhD - Doctor of Philosophy
Thesis Type: Doctoral Dissertation
Refereed: Yes
Uncontrolled Keywords: active learning, classification, multi-class, multi-label, enriched label-related feedback, probabilistic score, Likert-scale feedback, ordered class set
Date Deposited: 16 Sep 2020 15:30
Last Modified: 16 Sep 2020 15:30
URI: http://d-scholarship.pitt.edu/id/eprint/39554

Available Versions of this Item

  • Active Learning of Classification Models from Enriched Label-related Feedback. (deposited 16 Sep 2020 15:30) [Currently Displayed]

Metrics

Monthly Views for the past 3 years

Plum Analytics


Actions (login required)

View Item View Item