机器学习-白板推导 P2_6
Jensen’s Inequality 杰森不等式
假设
f
(
x
)
f(x)
f(x)是convex function
则
E
[
f
(
x
)
]
≥
f
(
E
(
x
)
)
E[f(x)]\geq f(E(x))
E[f(x)]≥f(E(x))
因为
f
(
x
)
f(x)
f(x)是凸函数
所以
∀
x
,
f
(
x
)
≥
l
(
x
)
\forall x, f(x) \geq l(x)
∀x,f(x)≥l(x)
对两边同时求期望：
E
[
f
(
x
)
]
≥
E
[
l
(
x
)
]
=
E
[
a
x
+
b
]
=
E
[
a
x
]
+
E
[
b
]
=
a
E
[
x
]
+
b
=
f
(
E
[
x
]
)
\begin{aligned} E[f(x)] &\geq E[l(x)]\\ &=E[ax+b] \\ &=E[ax]+E[b]\\ &=aE[x]+b \\ &=f(E[x]) \end{aligned}
E[f(x)]≥E[l(x)]=E[ax+b]=E[ax]+E[b]=aE[x]+b=f(E[x])
常见用法
6462
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
