Deep structured prediction with nonlinear output transformations

Colin Graber, Ofer Meshi, Alexander Gerhard Schwing

Research output: Contribution to journalConference article

Abstract

Deep structured models are widely used for tasks like semantic segmentation, where explicit correlations between variables provide important prior information which generally helps to reduce the data needs of deep nets. However, current deep structured models are restricted by oftentimes very local neighborhood structure, which cannot be increased for computational complexity reasons, and by the fact that the output configuration, or a representation thereof, cannot be transformed further. Very recent approaches which address those issues include graphical model inference inside deep nets so as to permit subsequent non-linear output space transformations. However, optimization of those formulations is challenging and not well understood. Here, we develop a novel model which generalizes existing approaches, such as structured prediction energy networks, and discuss a formulation which maintains applicability of existing inference techniques.

Original languageEnglish (US)
Pages (from-to)6320-6331
Number of pages12
JournalAdvances in Neural Information Processing Systems
Volume2018-December
StatePublished - Jan 1 2018
Event32nd Conference on Neural Information Processing Systems, NeurIPS 2018 - Montreal, Canada
Duration: Dec 2 2018Dec 8 2018

Fingerprint

Computational complexity
Semantics

ASJC Scopus subject areas

  • Computer Networks and Communications
  • Information Systems
  • Signal Processing

Cite this

Deep structured prediction with nonlinear output transformations. / Graber, Colin; Meshi, Ofer; Schwing, Alexander Gerhard.

In: Advances in Neural Information Processing Systems, Vol. 2018-December, 01.01.2018, p. 6320-6331.

Research output: Contribution to journalConference article

@article{e5b679f3761c402095d146cd8e02c1b9,
title = "Deep structured prediction with nonlinear output transformations",
abstract = "Deep structured models are widely used for tasks like semantic segmentation, where explicit correlations between variables provide important prior information which generally helps to reduce the data needs of deep nets. However, current deep structured models are restricted by oftentimes very local neighborhood structure, which cannot be increased for computational complexity reasons, and by the fact that the output configuration, or a representation thereof, cannot be transformed further. Very recent approaches which address those issues include graphical model inference inside deep nets so as to permit subsequent non-linear output space transformations. However, optimization of those formulations is challenging and not well understood. Here, we develop a novel model which generalizes existing approaches, such as structured prediction energy networks, and discuss a formulation which maintains applicability of existing inference techniques.",
author = "Colin Graber and Ofer Meshi and Schwing, {Alexander Gerhard}",
year = "2018",
month = "1",
day = "1",
language = "English (US)",
volume = "2018-December",
pages = "6320--6331",
journal = "Advances in Neural Information Processing Systems",
issn = "1049-5258",

}

TY - JOUR

T1 - Deep structured prediction with nonlinear output transformations

AU - Graber, Colin

AU - Meshi, Ofer

AU - Schwing, Alexander Gerhard

PY - 2018/1/1

Y1 - 2018/1/1

N2 - Deep structured models are widely used for tasks like semantic segmentation, where explicit correlations between variables provide important prior information which generally helps to reduce the data needs of deep nets. However, current deep structured models are restricted by oftentimes very local neighborhood structure, which cannot be increased for computational complexity reasons, and by the fact that the output configuration, or a representation thereof, cannot be transformed further. Very recent approaches which address those issues include graphical model inference inside deep nets so as to permit subsequent non-linear output space transformations. However, optimization of those formulations is challenging and not well understood. Here, we develop a novel model which generalizes existing approaches, such as structured prediction energy networks, and discuss a formulation which maintains applicability of existing inference techniques.

AB - Deep structured models are widely used for tasks like semantic segmentation, where explicit correlations between variables provide important prior information which generally helps to reduce the data needs of deep nets. However, current deep structured models are restricted by oftentimes very local neighborhood structure, which cannot be increased for computational complexity reasons, and by the fact that the output configuration, or a representation thereof, cannot be transformed further. Very recent approaches which address those issues include graphical model inference inside deep nets so as to permit subsequent non-linear output space transformations. However, optimization of those formulations is challenging and not well understood. Here, we develop a novel model which generalizes existing approaches, such as structured prediction energy networks, and discuss a formulation which maintains applicability of existing inference techniques.

UR - http://www.scopus.com/inward/record.url?scp=85064831684&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85064831684&partnerID=8YFLogxK

M3 - Conference article

AN - SCOPUS:85064831684

VL - 2018-December

SP - 6320

EP - 6331

JO - Advances in Neural Information Processing Systems

JF - Advances in Neural Information Processing Systems

SN - 1049-5258

ER -