An alternating direction method for dual MAP LP relaxation

Ofer Meshi; Amir Globerson

doi:10.1007/978-3-642-23783-6_30

An alternating direction method for dual MAP LP relaxation

Ofer Meshi^*, Amir Globerson

^*Corresponding author for this work

Hebrew University of Jerusalem

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

32 Scopus citations

Abstract

Maximum a-posteriori (MAP) estimation is an important task in many applications of probabilistic graphical models. Although finding an exact solution is generally intractable, approximations based on linear programming (LP) relaxation often provide good approximate solutions. In this paper we present an algorithm for solving the LP relaxation optimization problem. In order to overcome the lack of strict convexity, we apply an augmented Lagrangian method to the dual LP. The algorithm, based on the alternating direction method of multipliers (ADMM), is guaranteed to converge to the global optimum of the LP relaxation objective. Our experimental results show that this algorithm is competitive with other state-of-the-art algorithms for approximate MAP estimation.

Original language	English
Title of host publication	Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2011, Proceedings
Pages	470-483
Number of pages	14
Edition	PART 2
DOIs	https://doi.org/10.1007/978-3-642-23783-6_30
State	Published - 2011
Externally published	Yes
Event	European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2011 - Athens, Greece Duration: 5 Sep 2011 → 9 Sep 2011

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Number	PART 2
Volume	6912 LNAI
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2011
Country/Territory	Greece
City	Athens
Period	5/09/11 → 9/09/11

Keywords

Approximate Inference
Augmented Lagrangian Methods
Graphical Models
LP Relaxation
Maximum a-posteriori

Access to Document

10.1007/978-3-642-23783-6_30

Cite this

Meshi, O., & Globerson, A. (2011). An alternating direction method for dual MAP LP relaxation. In Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2011, Proceedings (PART 2 ed., pp. 470-483). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 6912 LNAI, No. PART 2). https://doi.org/10.1007/978-3-642-23783-6_30

Meshi, Ofer ; Globerson, Amir. / An alternating direction method for dual MAP LP relaxation. Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2011, Proceedings. PART 2. ed. 2011. pp. 470-483 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); PART 2).

@inproceedings{a9666a29efb847d48e540f2a51f7e137,

title = "An alternating direction method for dual MAP LP relaxation",

abstract = "Maximum a-posteriori (MAP) estimation is an important task in many applications of probabilistic graphical models. Although finding an exact solution is generally intractable, approximations based on linear programming (LP) relaxation often provide good approximate solutions. In this paper we present an algorithm for solving the LP relaxation optimization problem. In order to overcome the lack of strict convexity, we apply an augmented Lagrangian method to the dual LP. The algorithm, based on the alternating direction method of multipliers (ADMM), is guaranteed to converge to the global optimum of the LP relaxation objective. Our experimental results show that this algorithm is competitive with other state-of-the-art algorithms for approximate MAP estimation.",

keywords = "Approximate Inference, Augmented Lagrangian Methods, Graphical Models, LP Relaxation, Maximum a-posteriori",

author = "Ofer Meshi and Amir Globerson",

year = "2011",

doi = "10.1007/978-3-642-23783-6_30",

language = "אנגלית",

isbn = "9783642237829",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

number = "PART 2",

pages = "470--483",

booktitle = "Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2011, Proceedings",

edition = "PART 2",

note = "European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2011 ; Conference date: 05-09-2011 Through 09-09-2011",

}

Meshi, O & Globerson, A 2011, An alternating direction method for dual MAP LP relaxation. in Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2011, Proceedings. PART 2 edn, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), no. PART 2, vol. 6912 LNAI, pp. 470-483, European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2011, Athens, Greece, 5/09/11. https://doi.org/10.1007/978-3-642-23783-6_30

An alternating direction method for dual MAP LP relaxation. / Meshi, Ofer; Globerson, Amir.
Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2011, Proceedings. PART 2. ed. 2011. p. 470-483 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 6912 LNAI, No. PART 2).

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

TY - GEN

T1 - An alternating direction method for dual MAP LP relaxation

AU - Meshi, Ofer

AU - Globerson, Amir

PY - 2011

Y1 - 2011

N2 - Maximum a-posteriori (MAP) estimation is an important task in many applications of probabilistic graphical models. Although finding an exact solution is generally intractable, approximations based on linear programming (LP) relaxation often provide good approximate solutions. In this paper we present an algorithm for solving the LP relaxation optimization problem. In order to overcome the lack of strict convexity, we apply an augmented Lagrangian method to the dual LP. The algorithm, based on the alternating direction method of multipliers (ADMM), is guaranteed to converge to the global optimum of the LP relaxation objective. Our experimental results show that this algorithm is competitive with other state-of-the-art algorithms for approximate MAP estimation.

AB - Maximum a-posteriori (MAP) estimation is an important task in many applications of probabilistic graphical models. Although finding an exact solution is generally intractable, approximations based on linear programming (LP) relaxation often provide good approximate solutions. In this paper we present an algorithm for solving the LP relaxation optimization problem. In order to overcome the lack of strict convexity, we apply an augmented Lagrangian method to the dual LP. The algorithm, based on the alternating direction method of multipliers (ADMM), is guaranteed to converge to the global optimum of the LP relaxation objective. Our experimental results show that this algorithm is competitive with other state-of-the-art algorithms for approximate MAP estimation.

KW - Approximate Inference

KW - Augmented Lagrangian Methods

KW - Graphical Models

KW - LP Relaxation

KW - Maximum a-posteriori

UR - http://www.scopus.com/inward/record.url?scp=80052407878&partnerID=8YFLogxK

U2 - 10.1007/978-3-642-23783-6_30

DO - 10.1007/978-3-642-23783-6_30

M3 - ???researchoutput.researchoutputtypes.contributiontobookanthology.conference???

AN - SCOPUS:80052407878

SN - 9783642237829

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 470

EP - 483

BT - Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2011, Proceedings

T2 - European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2011

Y2 - 5 September 2011 through 9 September 2011

ER -

Meshi O, Globerson A. An alternating direction method for dual MAP LP relaxation. In Machine Learning and Knowledge Discovery in Databases - European Conference, ECML PKDD 2011, Proceedings. PART 2 ed. 2011. p. 470-483. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); PART 2). doi: 10.1007/978-3-642-23783-6_30

An alternating direction method for dual MAP LP relaxation

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this