L ( P , w ) ≡ − H ( P ) + w 0 ( 1 − ∑ y P ( y ∣ x ) ) + ∑ i = 1 n w i ( E p ‾ ( f i ) − E P ( f i ) ) = ∑ x , y P ~ ( x ) P ( y ∣ x ) log P ( y ∣ x ) + w 0 ( 1 − ∑ y P ( y ∣ x ) ) + ∑ i = 1 n w i ( ∑ x , y P ~ ( x , y ) f i ( x , y ) − ∑ x , y P ~ ( x ) P ( y ∣ x ) f i ( x , y ) ) \begin{aligned} L(P, w) & \equiv-H(P)+w_{0}\left(1-\sum_{y} P(y | x)\right)+\sum_{i=1}^{n} w_{i}\left(E_{\overline{p}}\left(f_{i}\right)-E_{P}\left(f_{i}\right)\right) \\=& {\color{red}\sum_{x, y} \tilde{P}(x) P(y | x) \log P(y | x) } +{\color{blue} w_{0}\left(1-\sum_{y} P(y | x)\right)} \\ &+\sum_{i=1}^{n} w_{i}\left(\sum_{x, y} \tilde{P}(x, y) f_{i}(x, y)-\sum_{x, y} \tilde{P}(x) P(y | x) f_{i}(x, y)\right) \end{aligned} L(P,w)=≡−H(P)+w0(1−y∑P(y∣x))+i=1∑nwi(Ep(fi)−EP(fi))x,y∑P~(x)P(y∣x)logP(y∣x)+w0(1−y∑P(y∣x))+i=1∑nwi(x,y∑P~(x,y)fi(x,y)−x,y∑P~(x)P(y∣x)fi(x,y))
L ( P , w ) L(P, w) L(P,w)对P(y|x)求导 假如是对 P ( y 1 ∣ x 1 ) P(y_1|x_1) P(y1∣x1)求导,如下
∂ L ( P , W ) ∂ P ( y 1 ∣ x 1 ) = P ~ ( x 1 ) ( l o g P ( y 1 ∣ x 1 ) + 1 ) − w 0 + ∑ i = 1 n w i ∑ x , y P ~ ( x ) P ( y ∣ x ) f i ( x , y ) \frac{\partial L(P,W)}{\partial P(y_1|x_1)}={\color{red}\tilde {P}(x_1)(logP(y_1|x_1)+1)}{\color{blue}-w_0}+\sum_{i=1}^nw_i\sum_{x,y}\tilde P(x)P(y|x)f_i(x,y) ∂P(y1∣x1)∂L(P,W)=P~(x1)(logP(y1∣x1)+1)−w0+∑i=1nwi∑x,yP~(x)P(y∣x)fi(x,y)
其中红色部分是因为只有 x = x 1 , y = y 1 x=x_1,y=y_1 x=x1,y=y1那一项才含有 P ( y 1 ∣ x 1 ) P(y_1|x_1) P(y1∣x1)
应用
∑
x
P
~
(
x
)
=
1
\sum_x\tilde P(x)=1
∑xP~(x)=1
=
P
~
(
x
1
)
(
l
o
g
P
(
y
1
∣
x
1
)
+
1
)
+
∑
x
P
~
(
x
)
w
0
+
∑
x
,
y
P
~
(
x
)
∑
i
=
1
n
w
i
P
(
y
∣
x
)
f
i
(
x
,
y
)
=\tilde{P}(x_1)(logP(y_1|x_1)+1) +\sum_x\tilde P(x)w_0+\sum_{x,y}\tilde P(x)\sum_{i=1}^nw_iP(y|x)f_i(x,y)
=P~(x1)(logP(y1∣x1)+1)+∑xP~(x)w0+∑x,yP~(x)∑i=1nwiP(y∣x)fi(x,y)
问题是最后怎么推出
∑
x
,
y
P
~
(
x
)
(
log
P
(
y
∣
x
)
+
1
−
w
0
−
∑
i
=
1
n
w
i
f
i
(
x
,
y
)
)
\sum_{x, y} \tilde{P}(x)\left(\log P(y | x)+1-w_{0}-\sum_{i=1}^{n} w_{i} f_{i}(x, y)\right)
x,y∑P~(x)(logP(y∣x)+1−w0−i=1∑nwifi(x,y))
的???