Bayesian Inference M = { set of events derived from a model} Bayes theorem: P( M|D) = P(D) P(D|M) P(M) D = { corpus of known data} Baldomero Oliva Miguel U
n
i
v
e
r
s
i
t
a
t
P
o
m
p
e
u
F
a
b
r
a
Hidden Markov Models Baldomero Oliva Miguel U
n
i
v
e
r
s
i
t
a
t
P
o
m
p
e
u
F
a
b
r
a A = a1a2a3a4a5
B = b1b2b3b4b5
C = c1c2c3c4c5
D = d1d2d3d4d5
E = e1e2e3e4e5 P1P2P3P4P5 P1(c1)P2(c2)P3(c3)P4(c4)P5(c5) = Ptotal
Bayesian Inference M = { set of events derived from a model} Bayes theorem: P( M|D) = P(D) P(D|M) P(M) D = { corpus of known data} Baldomero Oliva Miguel U
n
i
v
e
r
s
i
t
a
t
P
o
m
p
e
u
F
a
b
r
a
Baldomero Oliva Miguel U
n
i
v
e
r
s
i
t
a
t
P
o
m
p
e
u
F
a
b
r
a likelihood posterior Maximum a posteriori MAP
Maximum likelihood ML Minimizar F= - log(P(D|M)) -log(P(M))
Minimizar F= - log(P(D|M)) Bajo condiciones de contorno Yi(M) =0
Baldomero Oliva Miguel U
n
i
v
e
r
s
i
t
a
t
P
o
m
p
e
u
F
a
b
r
a
e HMM A A G G C C G G T T pe(A) pe(G) ..... Baldomero Oliva Miguel U
n
i
v
e
r
s
i
t
a
t
P
o
m
p
e
u
F
a
b
r
a
..... A G C G T e pe(A) pe(G) ..... i A A C G A A C G pi(A) pi(A) Tei Baldomero Oliva Miguel U
n
i
v
e
r
s
i
t
a
t
P
o
m
p
e
u
F
a
b
r
a HMM
A G C G T e pe(A) pe(G) ..... d i A A C G - - pi(A) pi(A) ..... Tei Tid Baldomero Oliva Miguel U
n
i
v
e
r
s
i
t
a
t
P
o
m
p
e
u
F
a
b
r
a HMM
e d i { Pe } { Pi } { Tab } Baldomero Oliva Miguel U
n
i
v
e
r
s
i
t
a
t
P
o
m
p
e
u
F
a
b
r
a
e d i start e d i e d i end Architecture HMM Baldomero Oliva Miguel U
n
i
v
e
r
s
i
t
a
t
P
o
m
p
e
u
F
a
b
r
a P(Sequence, path|model) = Product
Baldomero Oliva Miguel U
n
i
v
e
r
s
i
t
a
t
P
o
m
p
e
u
F
a
b
r
a A posteriori { Pe; Pi } state i DATA
sequence Q
path p
e d i start e d i e d i end Baldomero Oliva Miguel U
n
i
v
e
r
s
i
t
a
t
P
o
m
p
e
u
F
a
b
r
a Boundary conditions
Baldomero Oliva Miguel U
n
i
v
e
r
s
i
t
a
t
P
o
m
p
e
u
F
a
b
r
a
Baldomero Oliva Miguel U
n
i
v
e
r
s
i
t
a
t
P
o
m
p
e
u
F
a
b
r
a e d i ..T..
..T..
..S..
..T..
..Y.. eThr X=3/5
ePro X=0/5
Dirichlet
Baldomero Oliva Miguel U
n
i
v
e
r
s
i
t
a
t
P
o
m
p
e
u
F
a
b
r
a
Baldomero Oliva Miguel U
n
i
v
e
r
s
i
t
a
t
P
o
m
p
e
u
F
a
b
r
a INFORMATION
Comments