Biomedical Engineering Reference
In-Depth Information
μ
+
Namely,
x
2
obeys the Gaussian distribution with a mean equal to
A
c
and a
A
T
.
Bayesian inference quite often uses the precision, instead of the variance. Let
us define the precision matrix
ʣ
covariance matrix equal to
A
ʛ
that corresponds to the covariance natrix
ʣ
, i.e.,
ʛ
=
ʣ
−
1
. When the precision
ʛ
is used, the notation in Eq. (
C.2
) is used such that
|
μ
,
ʛ
−
1
x
∼
N(
x
).
(C.4)
Here, since we maintain the notational convenience
N(
random variable
|
mean
,
covariance matrix
),
ʛ
−
1
in this notation. Using the precision matrix,
we must use
ʛ
, the explicit form
of the Gaussian distribution is given by
2
exp
1
/
2
|
ʛ
|
1
2
(
T
p
(
x
)
=
−
x
−
μ
)
ʛ
(
x
−
μ
)
.
(C.5)
N
/
(
2
ˀ)
Let us compute the entropy when the probability distribution is Gaussian. The
definition of the entropy for a continuous random variable is
p
E
log
p
)
.
H
[
p
(
x
)
]=−
(
x
)
log
p
(
x
)
d
x
=−
(
x
(C.6)
Substituting the probability distribution in Eq. (
C.1
) into the equation above, we get
E
log
1
H
[
p
(
x
)
]=−
N
/
2
1
/
2
(
2
ˀ)
|
ʣ
|
2
E
1
T
ʣ
−
1
+
(
x
−
μ
)
(
x
−
μ
)
,
(C.7)
where constant terms are omitted. In the equation above, the first term on the right-
hand side is equal to
2
log
|
ʣ
|
. The second term is equal to
2
E
2
E
tr
T
1
1
T
ʣ
−
1
ʣ
−
1
(
x
−
μ
)
(
x
−
μ
)
=
(
x
−
μ
)(
x
−
μ
)
2
tr
ʣ
−
1
E
T
1
=
(
x
−
μ
)(
x
−
μ
)
2
tr
1
N
2
.
ʣ
−
1
=
ʣ
=
(C.8)
Therefore, omitting constant terms, the entropy is expressed as
1
2
log
H
[
p
(
x
)
]=
H(
x
)
=
|
ʣ
|
.
(C.9)