Active learning for pipeline models

Dan Roth, Kevin Small

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

Abstract

For many machine learning solutions to complex applications, there are significant performance advantages to decomposing the overall task into several simpler sequential stages, commonly referred to as a pipeline model. Typically, such scenarios are also characterized by high sample complexity, motivating the study of active learning for these situations. While most active learning research examines single predictions;, we extend such work to applications which utilize pipelined predictions. Specifically, we present an adaptive strategy for combining local active learning strategies into one that minimizes the annotation requirements for the overall task. Empirical results for a three-stage entity and relation extraction system demonstrate a significant reduction in supervised data requirements when using the proposed method.

Original languageEnglish (US)
Title of host publicationAAAI-08/IAAI-08 Proceedings - 23rd AAAI Conference on Artificial Intelligence and the 20th Innovative Applications of Artificial Intelligence Conference
Pages683-688
Number of pages6
StatePublished - Dec 24 2008
Event23rd AAAI Conference on Artificial Intelligence and the 20th Innovative Applications of Artificial Intelligence Conference, AAAI-08/IAAI-08 - Chicago, IL, United States
Duration: Jul 13 2008Jul 17 2008

Publication series

NameProceedings of the National Conference on Artificial Intelligence
Volume2

Other

Other23rd AAAI Conference on Artificial Intelligence and the 20th Innovative Applications of Artificial Intelligence Conference, AAAI-08/IAAI-08
CountryUnited States
CityChicago, IL
Period7/13/087/17/08

Fingerprint

Pipelines
Learning systems
Problem-Based Learning

ASJC Scopus subject areas

  • Software
  • Artificial Intelligence

Cite this

Roth, D., & Small, K. (2008). Active learning for pipeline models. In AAAI-08/IAAI-08 Proceedings - 23rd AAAI Conference on Artificial Intelligence and the 20th Innovative Applications of Artificial Intelligence Conference (pp. 683-688). (Proceedings of the National Conference on Artificial Intelligence; Vol. 2).

Active learning for pipeline models. / Roth, Dan; Small, Kevin.

AAAI-08/IAAI-08 Proceedings - 23rd AAAI Conference on Artificial Intelligence and the 20th Innovative Applications of Artificial Intelligence Conference. 2008. p. 683-688 (Proceedings of the National Conference on Artificial Intelligence; Vol. 2).

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

Roth, D & Small, K 2008, Active learning for pipeline models. in AAAI-08/IAAI-08 Proceedings - 23rd AAAI Conference on Artificial Intelligence and the 20th Innovative Applications of Artificial Intelligence Conference. Proceedings of the National Conference on Artificial Intelligence, vol. 2, pp. 683-688, 23rd AAAI Conference on Artificial Intelligence and the 20th Innovative Applications of Artificial Intelligence Conference, AAAI-08/IAAI-08, Chicago, IL, United States, 7/13/08.
Roth D, Small K. Active learning for pipeline models. In AAAI-08/IAAI-08 Proceedings - 23rd AAAI Conference on Artificial Intelligence and the 20th Innovative Applications of Artificial Intelligence Conference. 2008. p. 683-688. (Proceedings of the National Conference on Artificial Intelligence).
Roth, Dan ; Small, Kevin. / Active learning for pipeline models. AAAI-08/IAAI-08 Proceedings - 23rd AAAI Conference on Artificial Intelligence and the 20th Innovative Applications of Artificial Intelligence Conference. 2008. pp. 683-688 (Proceedings of the National Conference on Artificial Intelligence).
@inproceedings{abf5d9c5f4ea4bc1b844351afa166ec4,
title = "Active learning for pipeline models",
abstract = "For many machine learning solutions to complex applications, there are significant performance advantages to decomposing the overall task into several simpler sequential stages, commonly referred to as a pipeline model. Typically, such scenarios are also characterized by high sample complexity, motivating the study of active learning for these situations. While most active learning research examines single predictions;, we extend such work to applications which utilize pipelined predictions. Specifically, we present an adaptive strategy for combining local active learning strategies into one that minimizes the annotation requirements for the overall task. Empirical results for a three-stage entity and relation extraction system demonstrate a significant reduction in supervised data requirements when using the proposed method.",
author = "Dan Roth and Kevin Small",
year = "2008",
month = "12",
day = "24",
language = "English (US)",
isbn = "9781577353683",
series = "Proceedings of the National Conference on Artificial Intelligence",
pages = "683--688",
booktitle = "AAAI-08/IAAI-08 Proceedings - 23rd AAAI Conference on Artificial Intelligence and the 20th Innovative Applications of Artificial Intelligence Conference",

}

TY - GEN

T1 - Active learning for pipeline models

AU - Roth, Dan

AU - Small, Kevin

PY - 2008/12/24

Y1 - 2008/12/24

N2 - For many machine learning solutions to complex applications, there are significant performance advantages to decomposing the overall task into several simpler sequential stages, commonly referred to as a pipeline model. Typically, such scenarios are also characterized by high sample complexity, motivating the study of active learning for these situations. While most active learning research examines single predictions;, we extend such work to applications which utilize pipelined predictions. Specifically, we present an adaptive strategy for combining local active learning strategies into one that minimizes the annotation requirements for the overall task. Empirical results for a three-stage entity and relation extraction system demonstrate a significant reduction in supervised data requirements when using the proposed method.

AB - For many machine learning solutions to complex applications, there are significant performance advantages to decomposing the overall task into several simpler sequential stages, commonly referred to as a pipeline model. Typically, such scenarios are also characterized by high sample complexity, motivating the study of active learning for these situations. While most active learning research examines single predictions;, we extend such work to applications which utilize pipelined predictions. Specifically, we present an adaptive strategy for combining local active learning strategies into one that minimizes the annotation requirements for the overall task. Empirical results for a three-stage entity and relation extraction system demonstrate a significant reduction in supervised data requirements when using the proposed method.

UR - http://www.scopus.com/inward/record.url?scp=57749169475&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=57749169475&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:57749169475

SN - 9781577353683

T3 - Proceedings of the National Conference on Artificial Intelligence

SP - 683

EP - 688

BT - AAAI-08/IAAI-08 Proceedings - 23rd AAAI Conference on Artificial Intelligence and the 20th Innovative Applications of Artificial Intelligence Conference

ER -