互信息
定义
两个离散随机变量
( X , Y ) ∼ p ( x , y ) (X,Y)\sim p(x,y) (X,Y)∼p(x,y),一对离散随机变量,互信息: I ( X ; Y ) = ∑ ( x , y ) ∈ X × Y p ( x , y ) log p ( x , y ) p ( x ) p ( y ) = H ( X ) − H ( X ∣ Y ) = H ( Y ) − H ( Y ∣ X ) = H ( X ) + H ( Y ) − H ( X , Y ) \begin{aligned} I(X ; Y) & =\sum_{(x, y) \in \mathcal{X} \times \mathcal{Y}} p(x, y) \log \frac{p(x, y)}{p(x) p(y)} \\\\ & =H(X)-H(X \mid Y) \\\\ & =H(Y)-H(Y \mid X) \\\\ & =H(X)+H(Y)-H(X, Y) \end{aligned} I(X;Y)=(x,y)∈X×Y∑p(x,y)logp(x)p(y)p(x,y)=H(X)−H(X∣Y)=H(Y)−H(Y∣X)=H(X)+H(Y)−H(X,Y)
两个连续随机变量
( X , Y ) ∼ f ( x , y ) (X,Y)\sim f(x,y) (X,Y)∼f(x,y),一对连续随机变量,互信息: I ( X ; Y ) = ∫ f ( x , y ) log f ( x , y ) f ( x ) f ( y ) d x d y = h ( X ) − h ( X ∣ Y ) = h ( Y ) − h ( Y ∣ X ) = h ( X ) + h ( Y ) − h ( X , Y ) \begin{aligned} I(X ; Y) & =\int f(x, y) \log \frac{f(x, y)}{f(x) f(y)} d x d y \\\\ & =h(X)-h(X \mid Y) \\\\ & =h(Y)-h(Y \mid X) \\\\ & =h(X)+h(Y)-h(X, Y) \end{aligned} I(X;Y)=∫f(x,y)logf(x)f(y)f(x,y)dxdy=h(X)−h(X∣Y)=h(Y)−h(Y∣X)=h(X)+h(Y)−h(X,Y)
一个离散,一个连续
X ∼ p ( x ) X\sim p(x) X∼p(x), Y ∣ { X = x } ∼ f ( y ∣ x ) Y|\{X=x\}\sim f(y|x) Y∣{X=x}∼f(y∣x),互信息: I ( X ; Y ) = h ( Y ) − h ( Y ∣ X ) = H ( X ) − H ( X ∣ Y ) I(X ; Y) = h(Y)-h(Y \mid X) = H(X)-H(X \mid Y) I(X;Y)=h(Y)−h(Y∣X)=H(X)−H(X∣Y)
条件互信息
定义
连续变量
( X , Y ) ∣ { Z = z } ∼ F ( x , y ∣ z ) (X,Y)|\{Z=z\}\sim F(x,y|z) (X,Y)∣{Z=z}∼F(x,y∣z), Z ∼ F ( z ) Z\sim F(z) Z∼F(z),互信息: I ( X ; Y ∣ Z ) = ∫ I ( X ; Y ∣ Z = z ) d F ( z ) I(X ; Y \mid Z)=\int I(X ; Y \mid Z=z) d F(z) I(X;Y∣Z)=∫I(X;Y∣Z=z)dF(z)
离散变量
( X , Y , Z ) ∼ p ( x , y , z ) (X,Y,Z)\sim p(x,y,z) (X,Y,Z)∼p(x,y,z),互信息: I ( X ; Y ∣ Z ) = H ( X ∣ Z ) − H ( X ∣ Y , Z ) = H ( Y ∣ Z ) − H ( Y ∣ X , Z ) = H ( X ∣ Z ) + H ( Y ∣ Z ) − H ( X , Y ∣ Z ) \begin{aligned} I(X ; Y \mid Z) & =H(X \mid Z)-H(X \mid Y, Z) \\\\ & =H(Y \mid Z)-H(Y \mid X, Z) \\\\ & =H(X \mid Z)+H(Y \mid Z)-H(X, Y \mid Z) \end{aligned} I(X;Y∣Z)=H(X∣Z)−H(X∣Y,Z)=H(Y∣Z)−H(Y∣X,Z)=H(X∣Z)+H(Y∣Z)−H(X,Y∣Z) I ( X ; Y ∣ Z ) = E p ( x , y , z ) log p ( X , Y ∣ Z ) p ( X ∣ Z ) p ( Y ∣ Z ) I(X ; Y \mid Z)=E_{p(x, y, z)} \log \frac{p(X, Y \mid Z)}{p(X \mid Z) p(Y \mid Z)} I(X;Y∣Z)=Ep(x,y,z)logp(X∣Z)p(Y∣Z)p(X,Y∣Z)
例子
一个例子,及其互信息公式的推导:
X
∼
P
X
(
x
)
X\sim P_X(x)
X∼PX(x)与
S
R
S_R
SR没关系,是相互独立的,
Y
Y
Y则是同时受到
X
,
S
R
X,S_R
X,SR的影响。
互信息:
I
(
P
X
;
P
Y
∣
X
)
=
∑
x
∑
y
P
X
(
x
)
P
Y
∣
X
(
y
∣
x
)
log
P
X
Y
(
x
,
y
)
P
X
(
x
)
P
Y
(
y
)
\mathcal{I}\left(P_X ; P_{Y \mid X}\right)=\sum_x \sum_y P_X(x) P_{Y \mid X}(y \mid x) \log \frac{P_{X Y}(x,y)}{P_X(x) P_Y(y) }
I(PX;PY∣X)=x∑y∑PX(x)PY∣X(y∣x)logPX(x)PY(y)PXY(x,y)
I
(
P
X
;
P
Y
∣
X
)
=
∑
x
∑
y
P
X
(
x
)
P
Y
∣
X
(
y
∣
x
)
log
P
X
∣
Y
(
x
∣
y
)
P
X
(
x
)
\mathcal{I}\left(P_X ; P_{Y \mid X}\right)=\sum_x \sum_y P_X(x) P_{Y \mid X}(y \mid x) \log \frac{P_{X \mid Y}(x \mid y)}{P_X(x)}
I(PX;PY∣X)=x∑y∑PX(x)PY∣X(y∣x)logPX(x)PX∣Y(x∣y)
条件互信息: I ( P X ; P Y ∣ X S R ∣ P S R ) = ∑ s R P S R ( s R ) ∑ x ∑ y P X ( x ) P Y ∣ X S R ( y ∣ x s R ) log P X Y ∣ S R ( x , y ∣ s R ) P X ∣ S R ( x ∣ s R ) P Y ∣ S R ( y ∣ s R ) \mathcal{I}\left(P_X ; P_{Y \mid X S_R} \mid P_{S_R}\right)=\sum_{s_R} P_{S_R}\left(s_R\right) \sum_x \sum_y P_X(x) P_{Y \mid X S_R}\left(y \mid x s_R\right) \log \frac{P_{X Y | S_R}\left(x,y|s_R\right)}{P_{X|S_R}(x|s_R)P_{Y|S_R}(y|s_R)} I(PX;PY∣XSR∣PSR)=sR∑PSR(sR)x∑y∑PX(x)PY∣XSR(y∣xsR)logPX∣SR(x∣sR)PY∣SR(y∣sR)PXY∣SR(x,y∣sR) I ( P X ; P Y ∣ X S R ∣ P S R ) = ∑ s R P S R ( s R ) ∑ x ∑ y P X ( x ) P Y ∣ X S R ( y ∣ x s R ) log P X Y S R ( x , y , s R ) P X ∣ S R ( x ∣ s R ) P Y S R ( y , s R ) \mathcal{I}\left(P_X ; P_{Y \mid X S_R} \mid P_{S_R}\right)=\sum_{s_R} P_{S_R}\left(s_R\right) \sum_x \sum_y P_X(x) P_{Y \mid X S_R}\left(y \mid x s_R\right) \log \frac{P_{X Y S_R}\left(x,y,s_R\right)}{P_{X|S_R}(x|s_R)P_{Y S_R}(y,s_R)} I(PX;PY∣XSR∣PSR)=sR∑PSR(sR)x∑y∑PX(x)PY∣XSR(y∣xsR)logPX∣SR(x∣sR)PYSR(y,sR)PXYSR(x,y,sR) I ( P X ; P Y ∣ X S R ∣ P S R ) = ∑ s R P S R ( s R ) ∑ x ∑ y P X ( x ) P Y ∣ X S R ( y ∣ x s R ) log P X ∣ Y S R ( x ∣ y s R ) P X ( x ) \mathcal{I}\left(P_X ; P_{Y \mid X S_R} \mid P_{S_R}\right)=\sum_{s_R} P_{S_R}\left(s_R\right) \sum_x \sum_y P_X(x) P_{Y \mid X S_R}\left(y \mid x s_R\right) \log \frac{P_{X \mid Y S_R}\left(x \mid y s_R\right)}{P_X(x)} I(PX;PY∣XSR∣PSR)=sR∑PSR(sR)x∑y∑PX(x)PY∣XSR(y∣xsR)logPX(x)PX∣YSR(x∣ysR)