F IXING M AX -P RODUCT : A U NIFIED L OOK AT M ESSAGE P ASSING A LGORITHMS Nicholas Ruozzi and Sekhar Tatikonda Yale University.

Slides:



Advertisements
Similar presentations
The Primal-Dual Method: Steiner Forest TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: AA A A AA A A A AA A A.
Advertisements

Algorithms for MAP estimation in Markov Random Fields Vladimir Kolmogorov University College London Tutorial at GDR (Optimisation Discrète, Graph Cuts.
1 LP, extended maxflow, TRW OR: How to understand Vladimirs most recent work Ramin Zabih Cornell University.
Linear Time Methods for Propagating Beliefs Min Convolution, Distance Transforms and Box Sums Daniel Huttenlocher Computer Science Department December,
Variational Methods for Graphical Models Micheal I. Jordan Zoubin Ghahramani Tommi S. Jaakkola Lawrence K. Saul Presented by: Afsaneh Shirazi.
1 LP Duality Lecture 13: Feb Min-Max Theorems In bipartite graph, Maximum matching = Minimum Vertex Cover In every graph, Maximum Flow = Minimum.
Geometry and Theory of LP Standard (Inequality) Primal Problem: Dual Problem:
Approximation Algorithms Chapter 14: Rounding Applied to Set Cover.
Join-graph based cost-shifting Alexander Ihler, Natalia Flerova, Rina Dechter and Lars Otten University of California Irvine Introduction Mini-Bucket Elimination.
Introduction to Algorithms
ROTAMER OPTIMIZATION FOR PROTEIN DESIGN THROUGH MAP ESTIMATION AND PROBLEM-SIZE REDUCTION Hong, Lippow, Tidor, Lozano-Perez. JCC Presented by Kyle.
ICCV 2007 tutorial Part III Message-passing algorithms for energy minimization Vladimir Kolmogorov University College London.
Basic Feasible Solutions: Recap MS&E 211. WILL FOLLOW A CELEBRATED INTELLECTUAL TEACHING TRADITION.
Junction Trees: Motivation Standard algorithms (e.g., variable elimination) are inefficient if the undirected graph underlying the Bayes Net contains cycles.
Visual Recognition Tutorial
Distributed Message Passing for Large Scale Graphical Models Alexander Schwing Tamir Hazan Marc Pollefeys Raquel Urtasun CVPR2011.
Tirgul 12 Algorithm for Single-Source-Shortest-Paths (s-s-s-p) Problem Application of s-s-s-p for Solving a System of Difference Constraints.
UMass Lowell Computer Science Analysis of Algorithms Prof. Karen Daniels Fall, 2006 Lecture 9 Wednesday, 11/15/06 Linear Programming.
Convergent and Correct Message Passing Algorithms Nicholas Ruozzi and Sekhar Tatikonda Yale University TexPoint fonts used in EMF. Read the TexPoint manual.
Global Approximate Inference Eran Segal Weizmann Institute.
Approximation Algorithms
Message Passing Algorithms for Optimization
EE 685 presentation Optimization Flow Control, I: Basic Algorithm and Convergence By Steven Low and David Lapsley Asynchronous Distributed Algorithm Proof.
Integer Programming Difference from linear programming –Variables x i must take on integral values, not real values Lots of interesting problems can be.
Tirgul 13. Unweighted Graphs Wishful Thinking – you decide to go to work on your sun-tan in ‘ Hatzuk ’ beach in Tel-Aviv. Therefore, you take your swimming.
Constrained Optimization Rong Jin. Outline  Equality constraints  Inequality constraints  Linear Programming  Quadratic Programming.
1 Linear Programming Supplements (Optional). 2 Standard Form LP (a.k.a. First Primal Form) Strictly ≤ All x j 's are non-negative.
MAP Estimation Algorithms in M. Pawan Kumar, University of Oxford Pushmeet Kohli, Microsoft Research Computer Vision - Part I.
Daniel Kroening and Ofer Strichman Decision Procedures An Algorithmic Point of View Deciding ILPs with Branch & Bound ILP References: ‘Integer Programming’
Decision Procedures An Algorithmic Point of View
V. V. Vazirani. Approximation Algorithms Chapters 3 & 22
CS774. Markov Random Field : Theory and Application Lecture 08 Kyomin Jung KAIST Sep
Topics in Algorithms 2005 Constructing Well-Connected Networks via Linear Programming and Primal Dual Algorithms Ramesh Hariharan.
1 Minimum Cost Flows Goal: Minimize costs to meet all demands in a network subject to capacities (combines elements of both shortest path and max flow.
Probabilistic Graphical Models
EE 685 presentation Utility-Optimal Random-Access Control By Jang-Won Lee, Mung Chiang and A. Robert Calderbank.
An Efficient Message-Passing Algorithm for the M-Best MAP Problem Dhruv Batra (Currently) Research Assistant Professor TTI-Chicago (Spring 2013) Assistant.
Algorithms for MAP estimation in Markov Random Fields Vladimir Kolmogorov University College London.
Discrete Optimization Lecture 3 – Part 1 M. Pawan Kumar Slides available online
Probabilistic Inference Lecture 5 M. Pawan Kumar Slides available online
Dynamic Tree Block Coordinate Ascent Daniel Tarlow 1, Dhruv Batra 2 Pushmeet Kohli 3, Vladimir Kolmogorov 4 1: University of Toronto3: Microsoft Research.
CSE 589 Part VI. Reading Skiena, Sections 5.5 and 6.8 CLR, chapter 37.
Update any set S of nodes simultaneously with step-size We show fixed point update is monotone for · 1/|S| Covering Trees and Lower-bounds on Quadratic.
EE 685 presentation Optimization Flow Control, I: Basic Algorithm and Convergence By Steven Low and David Lapsley.
Linear Program Set Cover. Given a universe U of n elements, a collection of subsets of U, S = {S 1,…, S k }, and a cost function c: S → Q +. Find a minimum.
Approximation Algorithms Department of Mathematics and Computer Science Drexel University.
Implicit Hitting Set Problems Richard M. Karp Erick Moreno Centeno DIMACS 20 th Anniversary.
Log Truck Scheduling Problem
Lecture.6. Table of Contents Lp –rounding Dual Fitting LP-Duality.
Join-graph based cost-shifting Alexander Ihler, Natalia Flerova, Rina Dechter and Lars Otten University of California Irvine Introduction Mini-Bucket Elimination.
Introduction to Optimization
Tightening LP Relaxations for MAP using Message-Passing David Sontag Joint work with Talya Meltzer, Amir Globerson, Tommi Jaakkola, and Yair Weiss.
The minimum cost flow problem. Solving the minimum cost flow problem.
Distributed cooperation and coordination using the Max-Sum algorithm
Approximation Algorithms Duality My T. UF.
Approximation Algorithms based on linear programming.
Tuesday, March 19 The Network Simplex Method for Solving the Minimum Cost Flow Problem Handouts: Lecture Notes Warning: there is a lot to the network.
Chapter 8 PD-Method and Local Ratio (5) Equivalence This ppt is editored from a ppt of Reuven Bar-Yehuda. Reuven Bar-Yehuda.
Introduction of BP & TRW-S
The minimum cost flow problem
Chap 9. General LP problems: Duality and Infeasibility
Introduction to linear programming (LP): Minimization
Analysis of Algorithms
MURI Kickoff Meeting Randolph L. Moses November, 2008
Chapter 5. The Duality Theorem
Expectation-Maximization & Belief Propagation
Flow Feasibility Problems
(Convex) Cones Def: closed under nonnegative linear combinations, i.e.
1.2 Guidelines for strong formulations
Presentation transcript:

F IXING M AX -P RODUCT : A U NIFIED L OOK AT M ESSAGE P ASSING A LGORITHMS Nicholas Ruozzi and Sekhar Tatikonda Yale University

P REVIOUS W ORK Recent work related to max-product has focused on convergent and correct message passing schemes: TRMP [Wainwright et al. 2005] MPLP [Globerson and Jaakkola 2007] Max-Sum Diffusion [Werner 2007] “Splitting” Max-product [Ruozzi and Tatikonda 2010]

Typical approach: focus on a dual formulation of the MAP LP Message passing scheme is derived as a coordinate ascent scheme on a concave dual: MPLP TRMP Max-Sum Diffusion P REVIOUS W ORK MAP MAP LP Concave Dual

Many of these algorithms can be seen as maximizing a specific lower bound [Sontag and Jaakkola 2009] The maximization is performed over reparameterizations of the objective function that satisfy specific constraints Different constraints correspond to different dual formulations P REVIOUS W ORK

T HIS W ORK Focus on the primal problem: Choose a reparameterization of the objective function Reparameterization in terms of messages Construct concave lower bounds from this reparameterization by exploiting concavity of min Perform coordinate ascent on these lower bounds MAP Reparamet- erization Concave Lower Bound

T HIS W ORK Many of the common message passing schemes can be captured by the “splitting” family of reparameterizations Many possible lower bounds of interest Produces an unconstrained concave optimization problem MAP Reparamet- erization Concave Lower Bound

O UTLINE Background Min-sum Reparameterizations “Splitting” Reparameterization Lower bounds Message Updates

M IN -S UM Minimize an objective function that factorizes as a sum of potentials (assume f is bounded from below) (some multiset whose elements are subsets of the variables)

C ORRESPONDING G RAPH 2 1 3

R EPARAMETERIZATIONS We can rewrite the objective function as This does not change the objective function as long as the messages are finite valued at each x The objective function is reparameterized in terms of the messages No dependence on messages passed from i to ®

B ELIEFS Typically, we express the reparameterization in terms of beliefs (meant to represent min-marginals): With this definition, we have:

M IN -S UM The min-sum algorithm updates ensure that, after updating m ®i, In other words, Can estimate an assignment from a collection of messages by choosing Upon convergence,

C ORRECTNESS G UARANTEES The min-sum algorithm does not guarantee the correctness of this estimate upon convergence Assignments that minimize b i need not minimize f: Notable exceptions: trees, single cycles, singly connected graphs

L OWER B OUNDS Can derive lower bounds that are concave in the messages from reparameterizations: Lower bound is a concave function of the messages (and beliefs) We want to find the choice of messages that maximizes the lower bound This lower bound may not be tight

O UTLINE Background Min-sum Reparameterizations “Splitting” Reparameterization Lower bounds Message Updates

“G OOD ” R EPARAMETERIZATIONS Many possible reparameterizations How do we choose reparameterizations that produce “nice” lower bounds? Want estimates corresponding to the optimal choice of messages to minimize the objective function Want the bound to be concave in the messages Want the coordinate ascent scheme to remain local

“S PLITTING ” REPARAMETERIZATION where c i, c ®  0 and the beliefs are defined as:

“S PLITTING ” REPARAMETERIZATION TRMP: is a collection of spanning trees in the factor graph and ¹ is a probability distribution on spanning trees Choose c ® = ¹ ® Can extend this to a collection of singly connected subgraphs [Ruozzi and Tatikonda 2010]

“S PLITTING ” REPARAMETERIZATION Min-sum, TRMP, MPLP, and Max-Sum Diffusion can all be characterized as splitting reparameterizations One possible lower bound: We could choose c such that f can be written as a nonnegative combination of the beliefs

“S PLITTING ” R EPARAMETERIZATION TRMP lower bound: Max-Sum Diffusion lower bound:

O UTLINE Background Min-sum Reparameterizations “Splitting” Reparameterization Lower bounds Message Updates

F ROM L OWER BOUNDS TO M ESSAGE U PDATES We can construct the message updates by ensuring that we perform coordinate ascent on our lower bounds Can perform block updates over trees [Meltzer et al. 2009] [Kolmogorov 2009] [Sontag and Jaakkola 2009] Key observation: Equality iff there is an x that simultaneously minimizes both functions

M AX -S UM D IFFUSION Want Solving for m ®i gives Do this for all ® 2  i

S PLITTING U PDATE Suppose all coefficients are positive and c i > 0 We want Solving for m ®i gives Do this for all ® 2  i

C ONCLUSION MPLP, TRMP, and Max-Sum Diffusion are all instances of the splitting reparameterization for specific choices of the constants and lower bound Different lower bounds produce different unconstrained concave optimization problems Choice of lower bound corresponds to choosing different dual formulations Maximization is performed with respect to the messages, not the beliefs Many more reparameterizations and lower bounds are possible Is there a reparameterization in which the lower bounds are strictly concave?

Q UESTIONS ?