Professional Documents
Culture Documents
2.4.6 Repn MNs Loglinear
2.4.6 Repn MNs Loglinear
Graphical Models
Log-Linear
Models
Daphne Koller
Log-Linear Representation
Each feature fj has a scope Dj Different features can have same scope
Daphne Koller
Daphne Koller
Features: word capitalized, word in atlas or name list, previous word is Mrs, next word is Times,
Daphne Koller
Ising Model
Daphne Koller
Metric MRFs
All Xi take values in label space V
Xi
Xj
Distance function : V V R
Reflexivity: (v,v) = 0 for all v Symmetry: (v1,v2) = (v2,v1) for all v1, v2 Triangle inequality: (v1,v2) (v1,v3) + (v3, v2) for all v1, v2, v3
Daphne Koller
Metric MRFs
All Xi take values in label space V
Xi
Xj
Distance function : V V R
lower probability
Daphne Koller
(vk,vl) =
vk=vl otherwise
0 1 1 1
1 0 1 1
1 1 0 1
1 1 1 0
(vk,vl)
vk-vl
(vk,vl)
vk-vl
Daphne Koller
(vk,vl) =
vk=vl otherwise
0 1 1 1
1 0 1 1
1 1 0 1
1 1 1 0
Daphne Koller
original
denoised
denoised
(vk,vl) = |vk-vl|
(vk,vl) = min(|vk-vl|,d)
vk-vl
Daphne Koller