Train and test tightness of lp relaxations in structured prediction

Ofer Meshi, Mehrdad Mahdavi, Adrian Weiler, David Sontag

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

Abstract

Structured prediction is used in areas such as computer vision and natural language processing to predict structured outputs such as segmentations or parse trees. In these settings, prediction is performed by MAP inference or, equivalently, by solving an integer linear program. Because of the complex scoring functions required to obtain accurate predictions, both learning and inference typically require the use of approximate solvers. We propose a theoretical explanation to the striking observation that approximations based on linear programming (LP) relaxations are often tight on real-world instances. In particular, we show- that learning with LP relaxed inference encourages integrality of training instances, and that tightness generalizes from train to test data.

Original languageEnglish (US)
Title of host publication33rd International Conference on Machine Learning, ICML 2016
PublisherInternational Machine Learning Society (IMLS)
Pages2652-2664
Number of pages13
Volume4
ISBN (Electronic)9781510829008
StatePublished - 2016
Event33rd International Conference on Machine Learning, ICML 2016 - New York City, United States
Duration: Jun 19 2016Jun 24 2016

Other

Other33rd International Conference on Machine Learning, ICML 2016
CountryUnited States
CityNew York City
Period6/19/166/24/16

Fingerprint

Linear programming
Computer vision
Processing

ASJC Scopus subject areas

  • Artificial Intelligence
  • Software
  • Computer Networks and Communications

Cite this

Meshi, O., Mahdavi, M., Weiler, A., & Sontag, D. (2016). Train and test tightness of lp relaxations in structured prediction. In 33rd International Conference on Machine Learning, ICML 2016 (Vol. 4, pp. 2652-2664). International Machine Learning Society (IMLS).

Train and test tightness of lp relaxations in structured prediction. / Meshi, Ofer; Mahdavi, Mehrdad; Weiler, Adrian; Sontag, David.

33rd International Conference on Machine Learning, ICML 2016. Vol. 4 International Machine Learning Society (IMLS), 2016. p. 2652-2664.

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

Meshi, O, Mahdavi, M, Weiler, A & Sontag, D 2016, Train and test tightness of lp relaxations in structured prediction. in 33rd International Conference on Machine Learning, ICML 2016. vol. 4, International Machine Learning Society (IMLS), pp. 2652-2664, 33rd International Conference on Machine Learning, ICML 2016, New York City, United States, 6/19/16.
Meshi O, Mahdavi M, Weiler A, Sontag D. Train and test tightness of lp relaxations in structured prediction. In 33rd International Conference on Machine Learning, ICML 2016. Vol. 4. International Machine Learning Society (IMLS). 2016. p. 2652-2664
Meshi, Ofer ; Mahdavi, Mehrdad ; Weiler, Adrian ; Sontag, David. / Train and test tightness of lp relaxations in structured prediction. 33rd International Conference on Machine Learning, ICML 2016. Vol. 4 International Machine Learning Society (IMLS), 2016. pp. 2652-2664
@inproceedings{07982d2516744279874749ca7c5a17c6,
title = "Train and test tightness of lp relaxations in structured prediction",
abstract = "Structured prediction is used in areas such as computer vision and natural language processing to predict structured outputs such as segmentations or parse trees. In these settings, prediction is performed by MAP inference or, equivalently, by solving an integer linear program. Because of the complex scoring functions required to obtain accurate predictions, both learning and inference typically require the use of approximate solvers. We propose a theoretical explanation to the striking observation that approximations based on linear programming (LP) relaxations are often tight on real-world instances. In particular, we show- that learning with LP relaxed inference encourages integrality of training instances, and that tightness generalizes from train to test data.",
author = "Ofer Meshi and Mehrdad Mahdavi and Adrian Weiler and David Sontag",
year = "2016",
language = "English (US)",
volume = "4",
pages = "2652--2664",
booktitle = "33rd International Conference on Machine Learning, ICML 2016",
publisher = "International Machine Learning Society (IMLS)",

}

TY - GEN

T1 - Train and test tightness of lp relaxations in structured prediction

AU - Meshi, Ofer

AU - Mahdavi, Mehrdad

AU - Weiler, Adrian

AU - Sontag, David

PY - 2016

Y1 - 2016

N2 - Structured prediction is used in areas such as computer vision and natural language processing to predict structured outputs such as segmentations or parse trees. In these settings, prediction is performed by MAP inference or, equivalently, by solving an integer linear program. Because of the complex scoring functions required to obtain accurate predictions, both learning and inference typically require the use of approximate solvers. We propose a theoretical explanation to the striking observation that approximations based on linear programming (LP) relaxations are often tight on real-world instances. In particular, we show- that learning with LP relaxed inference encourages integrality of training instances, and that tightness generalizes from train to test data.

AB - Structured prediction is used in areas such as computer vision and natural language processing to predict structured outputs such as segmentations or parse trees. In these settings, prediction is performed by MAP inference or, equivalently, by solving an integer linear program. Because of the complex scoring functions required to obtain accurate predictions, both learning and inference typically require the use of approximate solvers. We propose a theoretical explanation to the striking observation that approximations based on linear programming (LP) relaxations are often tight on real-world instances. In particular, we show- that learning with LP relaxed inference encourages integrality of training instances, and that tightness generalizes from train to test data.

UR - http://www.scopus.com/inward/record.url?scp=84998953774&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84998953774&partnerID=8YFLogxK

M3 - Conference contribution

AN - SCOPUS:84998953774

VL - 4

SP - 2652

EP - 2664

BT - 33rd International Conference on Machine Learning, ICML 2016

PB - International Machine Learning Society (IMLS)

ER -