Sum Product Algorithm

# Algorithm sum product updates variable to local

This preview shows page 1. Sign up to view the full content.

This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: ates Variable to local function: µx →f (x ) = µh → x h∈n(x )\f Instructor: Arindam Banerjee The Sum-Product Algorithm Sum Product Updates Variable to local function: µx →f (x ) = µh → x h∈n(x )\f Local function to variable: µf →x (x ) = f (x ) ∼x Instructor: Arindam Banerjee µy →f (y ) y ∈n(f )\{x } The Sum-Product Algorithm Example: Step 1 µfA →x1 (x1 ) = fA (x1 ) µfB →x2 (x2 ) = fB (x2 ) µx4 →fD (x4 ) = 1 µx5 →fE (x5 ) = 1 Instructor: Arindam Banerjee The Sum-Product Algorithm Example: Step 2 µx1 →fC (x1 ) = µfA →x1 (x1 ) µx2 →fC (x2 ) = µfB →x2 (x2 ) µfD →x3 (x3 ) = fD (x3 , x4 )µx4 →fD (x4 ) ∼x3 µfE →x3 (x3 ) = fD (x3 , x5 )µx5 →fE (x5 ) ∼x3 Instructor: Arindam Banerjee The Sum-Product Algorithm Example: Step 3 µfC →x3 (x3 ) = fC (x1 , x2 , x3 )µx1 →fC (x1 )µx2 →fC (x2 ) ∼x3 µx3 →fC (x3 ) = µfD →x3 (x3 )µfE →x3 (x3 ) Instructor: Arindam Banerjee The Sum-Product Algorithm Example: Step 4 µfC →x1 (x1 ) = fC (x1 , x2 , x3 )µx2 →fC (x2 )µx3 →fC (x3 ) ∼x1 µfC →x2 (x2 ) = fC (x1 , x2 , x3 )µx1 →fC (x1 )µx3 →fC (x3 ) ∼x2 µx3 →fD (x3 ) = µfC →x3 (x3 )µfE →x3 (x3 ) µx3 →fE (x3 ) = µfC →x3 (x3 )µfD →x3 (x3 ) Instructor: Arindam Banerjee The Sum-Product Algorithm Example: Step 5 µx1 →fA (x1 ) = µfC →x1 (x1 ) µx2 →fB (x2 ) = µfC →x2 (x2 ) µfD →x4 (x4 ) = fD (x3 , x4 )µx3 →fD (x4 ) ∼x4 µfE →x5 (x5 ) = fD (x3 , x5 )µx3 →fE (x5 ) ∼x5 Instructor: Arindam Banerjee The Sum-Product Algorithm Example: Termination Marginal function is the product of all incoming messages g1 (x1 ) = µfA →x1 (x1 )µfC →x1 (x1 ) g2 (x2 ) = µfB →x2 (x2 )µfC →x2 (x2 ) g3 (x3 ) = µfC →x3 (x3 )µfD →x3 (x3 )µfE →x3 (x3 ) g2 (x2 ) = µfD →x4 (x4 ) g5 (x5 ) = µfE →x5 (x5 ) Instructor: Arindam Banerjee The Sum-Product Algorithm Belief Propagation in Bayes Nets Instructor: Arindam Banerjee The Sum-Product Algorithm Belief Propagation in Bayes Nets fA (x1 ) = p (x1 ) fB (x2 ) = p (x2 ) Instructor: Arindam Banerjee fC (x1 , x2 , x3 ) = p (x3 |x1 , x2 ) The Sum-Product Algorithm Belief Propagation in Bayes Nets fA (x1 ) = p (x1 ) fB (x2 ) = p (x2 ) fD (x3 , x4 ) = p (x4 |x3 ) Instructor: Arindam Banerjee fC (x1 , x2 , x3 ) = p (x3 |x1 , x2 ) fE (x3 , x5 ) = p (x5 |x3 ) The Sum-Product Algorithm...
View Full Document

## This document was uploaded on 03/18/2014 for the course CS&E 5512 at Minnesota.

Ask a homework question - tutors are online