MRF optimization via dual decomposition

In dual decomposition a problem is broken into smaller subproblems and a solution to the relaxed problem is found. This method can be employed for MRF optimization.[1] Dual decomposition is applied to markov logic programs as an inference technique.[2]

Background edit

Discrete MRF Optimization (inference) is very important in Machine Learning and Computer vision, which is realized on CUDA graphical processing units.[3] Consider a graph  with nodes  and Edges  . The goal is to assign a label  to each  so that the MRF Energy is minimized:

(1)  

Major MRF Optimization methods are based on Graph cuts or Message passing. They rely on the following integer linear programming formulation

(2)  

In many applications, the MRF-variables are {0,1}-variables that satisfy:   label   is assigned to  , while   , labels   are assigned to  .

Dual Decomposition edit

The main idea behind decomposition is surprisingly simple:

  1. decompose your original complex problem into smaller solvable subproblems,
  2. extract a solution by cleverly combining the solutions from these subproblems.

A sample problem to decompose:

 where  

In this problem, separately minimizing every single  over   is easy; but minimizing their sum is a complex problem. So the problem needs to get decomposed using auxiliary variables  and the problem will be as follows:

 where  

Now we can relax the constraints by multipliers   which gives us the following Lagrangian dual function:

 

Now we eliminate   from the dual function by minimizing over   and dual function becomes:

 

We can set up a Lagrangian dual problem:

(3)   The Master problem

(4)  where  The Slave problems

MRF optimization via Dual Decomposition edit

The original MRF optimization problem is NP-hard and we need to transform it into something easier.

 is a set of sub-trees of graph  where its trees cover all nodes and edges of the main graph. And MRFs defined for every tree   in  will be smaller. The vector of MRF parameters is  and the vector of MRF variables is  , these two are just smaller in comparison with original MRF vectors  . For all vectors  we'll have the following:

(5)  

Where  and  denote all trees of  than contain node  and edge  respectively. We simply can write:

(6)  

And our constraints will be:

(7)  

Our original MRF problem will become:

(8)  where  and  

And we'll have the dual problem we were seeking:

(9)  The Master problem

where each function  is defined as:

(10)  where  The Slave problems

Theoretical Properties edit

Theorem 1. Lagrangian relaxation (9) is equivalent to the LP relaxation of (2).

 

Theorem 2. If the sequence of multipliers  satisfies  then the algorithm converges to the optimal solution of (9).

Theorem 3. The distance of the current solution  to the optimal solution  , which decreases at every iteration.

Theorem 4. Any solution obtained by the method satisfies the WTA (weak tree agreement) condition.

Theorem 5. For binary MRFs with sub-modular energies, the method computes a globally optimal solution.

References edit

  1. ^ "MRF Optimization via Dual Decomposition" (PDF). {{cite journal}}: Cite journal requires |journal= (help)
  2. ^ Feng Niu and Ce Zhang and Christopher Re and Jude Shavlik (2012). Scaling Inference for Markov Logic via Dual Decomposition. 2012 IEEE 12th International Conference on Data Mining. IEEE. CiteSeerX 10.1.1.244.8755. doi:10.1109/icdm.2012.96.
  3. ^ Shervin Rahimzadeh Arashloo and Josef Kittler (2013). Efficient processing of MRFs for unconstrained-pose face recognition. 2013 IEEE Sixth International Conference on Biometrics: Theory, Applications and Systems (BTAS). IEEE. doi:10.1109/btas.2013.6712721.