Bayesian estimation
n
Standard
Bayesian
approach:
n
Maintain a density p(
q
) over CPD parameters
n
Given data we update p
n
Happens to collapse down to a single network
n
For queries about the next instance
Y
X
E(
q
y0|x0
)
E(
q
y0|x1
)
E(
q
x0
)
Y
X
p(
q
y0|x0
)
p(
q
y0|x1
)
p(
q
x0
)