Belief Propagation
description
Transcript of Belief Propagation
![Page 1: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/1.jpg)
Belief Propagation
Kai Ju LiuMarch 9, 2006
![Page 2: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/2.jpg)
Statistical Problems
• Medicine• Finance• Internet• Computer vision
![Page 3: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/3.jpg)
Inference Problems
• Given data B, infer A: p(A|B)• Computer vision
– Given image, find objects– Given two images, resolve 3D object– Given multiple images, track object
![Page 4: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/4.jpg)
Conditional Probability
• Given event B, what is probability of A?
• Independence: p(A|B)=p(A)
Bp
BApBAp ,|
A B
![Page 5: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/5.jpg)
Bayes’ Rule
Bp
ApABpBp
BApBAp |,|
![Page 6: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/6.jpg)
e.g.
Cold Weekday
Party
Hangover
Marginal Probability: WCP xxx
WCPHH xxxxpxp,,
,,,
8-sum
WCWCPPH
WCPH
xpxpxxxpxxpxxxxp
,|| ,,, Joint Probability:
![Page 7: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/7.jpg)
e.g. (cont.)
Cold Weekday
Party
Hangover
Marginal Probability:
WCP xxxWCPHH xxxxpxp
,,
,,,
8-sum
sum-2|
sum-4,|,
P
WC
xPPHH
WCxx
WCPP
xpxxpxp
xpxpxxxpxpLocalize probabilities:
![Page 8: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/8.jpg)
Approach
• Define variables and connections
• Calculate marginal probabilities efficiently
• Find most likely configuration
PHWCPWCHPWC xxpxxxpxpxpxxxx |,,|,,;,,,
PWC xxx
HPWCH xxxxpxp,,
,,,
Hx
xpH
maxarg
![Page 9: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/9.jpg)
Pairwise Markov Random Field
1 2 3
4
5
• Basic structure: vertices, edges
![Page 10: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/10.jpg)
Pairwise Markov Random Field
• Basic structure: vertices, edges• Vertex i has set of possible states Xi
1X 2X 3X
4X
5X
and observed value yi
1y 2y 3y4y
5y
• Compatibility between states and observed values, iii yx ,
1 2 3
4
5
• Compatibility between neighboring vertices i and j, jiij xx ,
12 23
34
35
45
![Page 11: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/11.jpg)
Pairwise MRF: Probabilities
• Joint probability:
1X 2X 3X
4X
5X
1y 2y 3y4y
5y
1 2 3
4
5
12 23
34
35
45
ij
jiiji
iii xxyxZ
xxp ,,1,,5
151
• Marginal probability:
ijjXx
iijj
xxpxp,51,
51 ,,
– Advantage: allows average over ambiguous states– Disadvantage: complexity exponential in number of vertices
![Page 12: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/12.jpg)
Belief Propagation
1 2 3
4
5
![Page 13: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/13.jpg)
Belief Propagation
1b 2b 3b
4b
5b
• Beliefs replace probabilities:
iNj
ijiiiii
ii xmyxz
xb ,1
• Messages propagate information:
jj Xx ijNk
jkjijjijjjiji xmxxyxxm\
,,
212 xm
121 xm
323 xm
232 xm
434 xm
343 xm
535 xm
353 xm
![Page 14: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/14.jpg)
Belief Propagation Example
1 3
4
5
2b
23221222222
222222
22 ,1,1 xmxmyxz
xmyxz
xbNj
j
212 xm
232 xm
33
3533432332333232 ,,Xx
xmxmxxyxxm
343 xm
353 xm
11
2112111212 ,,Xx
xxyxxm
5544
35535553533443444343 ,,;,,XxXx
xxyxxmxxyxxm
![Page 15: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/15.jpg)
BP: Questions
• When can we calculate beliefs exactly?• When do beliefs equal probabilities?• When is belief propagation efficient?
Answer: Singly-Connected Graphs (SCG’s)• Graphs without loops• Messages terminate at leaf vertices• Beliefs equal probabilities• Complexity in previous example reduced from 13S5 to
24S2
![Page 16: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/16.jpg)
BP on Loopy Graphs
• Messages do not terminate• Possible approximate solutions
– Standard belief propagation– Generalized belief propagation
BP-TwoGraphs: Goals• Utilize advantages of SCG’s• Be accurate and efficient on loopy graphs
![Page 17: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/17.jpg)
BP-TwoGraphs: SCG’s
• Calculate beliefs on each set of SCG’s:–
• Select maximum beliefs from both sets– i
Hii
Giii xbxbxb ,max
iHii
Gi xbxb and
n
n
HHGG,,,,
1
1
• Consider loopy graph with n vertices• Select two sets of SCG’s that approximate the graph
–
![Page 18: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/18.jpg)
BP-TwoGraphs: Vision SCG’s
• Rectangular grid of pixel vertices• Hi: horizontal graphs
• Gi: vertical graphs
![Page 19: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/19.jpg)
Image Segmentation
add noise segment
![Page 20: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/20.jpg)
Image Segmentation: Results
![Page 21: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/21.jpg)
Real Image Segmentation
![Page 22: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/22.jpg)
Real Image Segmentation: Training
![Page 23: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/23.jpg)
Real Image Segmentation: Results
![Page 24: Belief Propagation](https://reader034.fdocuments.in/reader034/viewer/2022051402/56815fab550346895dcea53e/html5/thumbnails/24.jpg)
Stereo Vision