Fixing max-product: Convergent message passing algorithms for MAP LP-relaxations

Amir Globerson; Tommi Jaakkola

Fixing max-product: Convergent message passing algorithms for MAP LP-relaxations

^*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

We present a novel message passing algorithm for approximating the MAP problem in graphical models. The algorithm is similar in structure to max-product but unlike max-product it always converges, and can be proven to find the exactMAP solution in various settings. The algorithm is derived via block coordinate descent in a dual of the LP relaxation ofMAP, but does not require any tunable parameters such as step size or tree weights. We also describe a generalization of the method to cluster based potentials. The new method is tested on synthetic and real-world problems, and compares favorably with previous approaches.

Original language	English
Title of host publication	Advances in Neural Information Processing Systems 20 - Proceedings of the 2007 Conference
Publisher	Curran Associates Inc.
ISBN (Print)	160560352X, 9781605603520
State	Published - 2008
Externally published	Yes
Event	21st Annual Conference on Neural Information Processing Systems, NIPS 2007 - Vancouver, BC, Canada Duration: 3 Dec 2007 → 6 Dec 2007

Publication series

Name	Advances in Neural Information Processing Systems 20 - Proceedings of the 2007 Conference

Conference

Conference	21st Annual Conference on Neural Information Processing Systems, NIPS 2007
Country/Territory	Canada
City	Vancouver, BC
Period	3/12/07 → 6/12/07

Cite this

@inproceedings{523e1c5a4db9484f9918e7587e7475a6,

title = "Fixing max-product: Convergent message passing algorithms for MAP LP-relaxations",

abstract = "We present a novel message passing algorithm for approximating the MAP problem in graphical models. The algorithm is similar in structure to max-product but unlike max-product it always converges, and can be proven to find the exactMAP solution in various settings. The algorithm is derived via block coordinate descent in a dual of the LP relaxation ofMAP, but does not require any tunable parameters such as step size or tree weights. We also describe a generalization of the method to cluster based potentials. The new method is tested on synthetic and real-world problems, and compares favorably with previous approaches.",

author = "Amir Globerson and Tommi Jaakkola",

year = "2008",

language = "אנגלית",

isbn = "160560352X",

series = "Advances in Neural Information Processing Systems 20 - Proceedings of the 2007 Conference",

publisher = "Curran Associates Inc.",

booktitle = "Advances in Neural Information Processing Systems 20 - Proceedings of the 2007 Conference",

note = "null ; Conference date: 03-12-2007 Through 06-12-2007",

}

Globerson, A & Jaakkola, T 2008, Fixing max-product: Convergent message passing algorithms for MAP LP-relaxations. in Advances in Neural Information Processing Systems 20 - Proceedings of the 2007 Conference. Advances in Neural Information Processing Systems 20 - Proceedings of the 2007 Conference, Curran Associates Inc., 21st Annual Conference on Neural Information Processing Systems, NIPS 2007, Vancouver, BC, Canada, 3/12/07.

Fixing max-product : Convergent message passing algorithms for MAP LP-relaxations. / Globerson, Amir; Jaakkola, Tommi.

Advances in Neural Information Processing Systems 20 - Proceedings of the 2007 Conference. Curran Associates Inc., 2008. (Advances in Neural Information Processing Systems 20 - Proceedings of the 2007 Conference).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Fixing max-product

AU - Globerson, Amir

AU - Jaakkola, Tommi

PY - 2008

Y1 - 2008

N2 - We present a novel message passing algorithm for approximating the MAP problem in graphical models. The algorithm is similar in structure to max-product but unlike max-product it always converges, and can be proven to find the exactMAP solution in various settings. The algorithm is derived via block coordinate descent in a dual of the LP relaxation ofMAP, but does not require any tunable parameters such as step size or tree weights. We also describe a generalization of the method to cluster based potentials. The new method is tested on synthetic and real-world problems, and compares favorably with previous approaches.

AB - We present a novel message passing algorithm for approximating the MAP problem in graphical models. The algorithm is similar in structure to max-product but unlike max-product it always converges, and can be proven to find the exactMAP solution in various settings. The algorithm is derived via block coordinate descent in a dual of the LP relaxation ofMAP, but does not require any tunable parameters such as step size or tree weights. We also describe a generalization of the method to cluster based potentials. The new method is tested on synthetic and real-world problems, and compares favorably with previous approaches.

UR - http://www.scopus.com/inward/record.url?scp=85138475821&partnerID=8YFLogxK

M3 - פרסום בספר כנס

AN - SCOPUS:85138475821

SN - 160560352X

SN - 9781605603520

T3 - Advances in Neural Information Processing Systems 20 - Proceedings of the 2007 Conference

BT - Advances in Neural Information Processing Systems 20 - Proceedings of the 2007 Conference

PB - Curran Associates Inc.

Y2 - 3 December 2007 through 6 December 2007

ER -

Fixing max-product: Convergent message passing algorithms for MAP LP-relaxations

Abstract

Publication series

Conference

Other files and links

Fingerprint

Cite this