Biomedical Engineering Reference
In-Depth Information
We can derive a similar formula to compute the transfer entropy. To do so, we use the
similarity between the transfer entropy and mutual information. Using Eq. (C.45),
the conditional entropy
H(
y
|
x
,
y
)
is rewritten as
H(
y
|
x
,
y
)
=
H(
y
,
x
|
y
)
−
H(
x
|
y
).
(8.97)
Substituting this equation into Eq. (
8.81
), we get
H
x
ₒ
y
=
H(
x
|
y
)
+
H(
y
|
y
)
−
H(
y
,
x
|
y
).
(8.98)
Comparing the equation above with Eq. (C.46), it can be seen that the transfer entropy
is equal to the mutual information between
y
(
t
)
and
x
(
t
)
, when
y
(
t
)
is given.
We define
ʣ
u
,v
|
w
as
ʣ
u
,v
|
w
=
ʣ
u
v
−
ʣ
u
w
ʣ
−
1
T
ww
ʣ
vw
.
(8.99)
Then, using Eq. (C.61), we can express
H(
y
|
y
)
, and
H(
x
|
y
)
, such that
2
log
ʣ
y
,
y
|
y
,
1
H(
y
|
y
)
=
(8.100)
2
log
ʣ
x
,
x
|
y
.
1
H(
x
|
y
)
=
(8.101)
On the basis of Eq. (C.59), we also derive
2
log
.
T
˜
1
ʣ
y
,
y
|
y
ʣ
y
ʣ
x
,
y
|
y
ʣ
x
,
x
|
y
x
,
y
|˜
H(
y
,
x
|
y
)
=
(8.102)
Thus, substituting the equations above into Eq. (
8.98
), we obtain
1
2
log
1
y
H
x
ₒ
y
=
.
(8.103)
−
ʣ
−
1
y
y
ʣ
y
,
x
|
y
ʣ
−
1
T
y
I
y
ʣ
,
y
|˜
x
˜
,
˜
x
|˜
,
˜
x
|˜
Accordingly, defining the eigenvalues of a matrix,
ʣ
−
1
y
y
ʣ
y
,
x
|
y
ʣ
−
1
T
y
y
ʣ
y
,
(8.104)
,
˜
x
|˜
,
y
|˜
x
˜
,
˜
x
|˜
as
ˇ
j
(
j
=
1
,...,
d
), the transfer entropy is given by
d
1
2
1
H
x
ₒ
y
=
−
ˇ
j
.
log
(8.105)
1
j
=
1