神经网络中的常用激活函数和导数

1、 sigmoid 函数

y = 1 1 + e − x y = \frac{1}{1 + e^{-x}} y=1+ex1

导函数:

d y d x = − ( 1 + e − x ) − 2 ⋅ ( 1 + e − x ) ′ = − ( 1 + e − x ) − 2 ⋅ 1 ⋅ ( e − x ) ′ = − ( 1 + e − x ) − 2 ⋅ 1 ⋅ ( e − x ) ⋅ ( − x ) ′ = − ( 1 + e − x ) − 2 ⋅ 1 ⋅ ( e − x ) ⋅ ( − 1 ) = ( 1 + e − x ) − 2 ⋅ ( e − x ) = e − x ( 1 + e − x ) 2 \begin{aligned} \cfrac{{\rm d}y}{{\rm d}x} &= -(1 + e^{-x})^{-2} \cdot (1 + e^{-x})^\prime \\ &= -(1 + e^{-x})^{-2} \cdot 1 \cdot (e^{-x})^\prime \\ &= -(1 + e^{-x})^{-2} \cdot 1 \cdot (e^{-x}) \cdot (-x) ^\prime \\ &= -(1 + e^{-x})^{-2} \cdot 1 \cdot (e^{-x}) \cdot (-1) \\ &= (1 + e^{-x})^{-2} \cdot (e^{-x}) \\ &= \cfrac{e^{-x}}{(1 + e^{-x})^{2}} \end{aligned} dxdy=(1+ex)2(1+ex)=(1+ex)21(ex)=(1+ex)21(ex)(x)=(1+ex)21(ex)(1)=(1+ex)2(ex)=(1+ex)2ex

又因为 1 − y = 1 − 1 1 + e − x = 1 + e − x − 1 1 + e − x = e − x 1 + e − x 1 - y = 1 - \cfrac{1}{1 + e^{-x}} = \cfrac{1 + e^{-x} -1}{1 + e^{-x}} = \cfrac{e^{-x}}{1 + e^{-x}} 1y=11+ex1=1+ex1+ex1=1+exex

所以 d y d x = y [ 1 − y ] \cfrac{{\rm d}y}{{\rm d}x} = y[1 - y] dxdy=y[1y]

2、Tanh 函数
y = t a n h ( x ) = e x − e − x e x + e − x = 2 s i g m o i d ( 2 x ) − 1 y = {\rm tanh}(x)=\frac{e^x-e^{-x}}{e^x+e^{-x}} = 2{\rm sigmoid}(2x)-1 y=tanh(x)=ex+exexex=2sigmoid(2x)1

导函数:

d y d x = ( e x − e − x e x + e − x ) ′ = ( e x − e − x ) ′ ( e x + e − x ) − ( e x − e − x ) ( e x + e − x ) ′ ( e x + e − x ) 2 = ( e x + e − x ) ( e x + e − x ) − ( e x − e − x ) ( e x − e − x ) ( e x + e − x ) 2 = e 2 x + e − 2 x + 2 − ( e 2 x + e − 2 x − 2 ) ( e x + e − x ) 2 = 4 ( e x + e − x ) 2 \begin{aligned} \cfrac{{\rm d}y}{{\rm d}x} &= \left(\cfrac{e^x-e^{-x}}{e^x+e^{-x}}\right)^\prime \\ &= \cfrac{(e^x-e^{-x})^\prime(e^x+e^{-x})-(e^x-e^{-x})(e^x+e^{-x})^\prime}{(e^x+e^{-x})^2} \\ &= \cfrac{(e^x+e^{-x})(e^x+e^{-x})-(e^x-e^{-x})(e^x-e^{-x})}{(e^x+e^{-x})^2} \\ &= \cfrac{e^{2x}+e^{-2x}+2-(e^{2x}+e^{-2x}-2)}{(e^x+e^{-x})^2} \\ &= \cfrac{4}{(e^x+e^{-x})^2} \end{aligned} dxdy=(ex+exexex)=(ex+ex)2(exex)(ex+ex)(exex)(ex+ex)=(ex+ex)2(ex+ex)(ex+ex)(exex)(exex)=(ex+ex)2e2x+e2x+2(e2x+e2x2)=(ex+ex)24

又因为
1 − y = 1 − e x − e − x e x + e − x = e x + e − x − ( e x − e − x ) e x + e − x = 2 e − x e x + e − x 1 - y = 1 - \cfrac{e^x-e^{-x}}{e^x+e^{-x}} = \cfrac{e^x+e^{-x}-(e^x-e^{-x})}{e^x+e^{-x}} = \cfrac{2e^{-x}}{e^x+e^{-x}} 1y=1ex+exexex=ex+exex+ex(exex)=ex+ex2ex

1 + y = 1 + e x − e − x e x + e − x = e x + e − x + ( e x − e − x ) e x + e − x = 2 e x e x + e − x 1 + y = 1 + \cfrac{e^x-e^{-x}}{e^x+e^{-x}} = \cfrac{e^x+e^{-x}+(e^x-e^{-x})}{e^x+e^{-x}} = \cfrac{2e^{x}}{e^x+e^{-x}} 1+y=1+ex+exexex=ex+exex+ex+(exex)=ex+ex2ex

所以

( 1 − y ) ( 1 + y ) = 1 − y 2 = 4 ( e x + e − x ) 2 = d y d x (1 - y)(1+y) = 1 - y^2 = \cfrac{4}{(e^x+e^{-x})^2} = \cfrac{{\rm d}y}{{\rm d}x} (1y)(1+y)=1y2=(ex+ex)24=dxdy

即:
y = t a n h ′ ( x ) = 1 − t a n h 2 ( x ) y = {\rm tanh}^\prime(x)=1-{\rm tanh}^2(x) y=tanh(x)=1tanh2(x)

3、ReLu 函数

y = f ( x ) = max ⁡ ( 0 , x ) y = f(x) = \max(0,x) y=f(x)=max(0,x)
导函数
y ′ = f ′ ( x ) = { 1 , x > 0 0 , x ≤ 0 y^\prime = f^\prime(x) = \begin{cases} 1,& x > 0\\ 0,& x \le 0 \end{cases} y=f(x)={1,0,x>0x0

4、Leaky ReLu 函数

y = f ( x ) = { x , x > 0 α ⋅ x , x ≤ 0 y = f(x) = \begin{cases} x,& x > 0\\ \alpha \cdot x,& x \le 0 \end{cases} y=f(x)={x,αx,x>0x0
导函数
y ′ = f ′ ( x ) = { 1 , x > 0 α , x ≤ 0 y^\prime = f^\prime(x) = \begin{cases} 1,& x > 0\\ \alpha,& x \le 0 \end{cases} y=f(x)={1,α,x>0x0

5、Softplus 函数

y = f ( x ) = log ⁡ ( 1 + e x ) y = f(x) = \log(1+e^x) y=f(x)=log(1+ex)

导函数
y ′ = f ′ ( x ) = ( 1 + e x ) ′ 1 + e x = e x 1 + e x = 1 1 + e − x = s i g m o i d ( x ) y^\prime = f^\prime(x) = \cfrac{(1+e^x)^\prime}{1+e^x} = \cfrac{e^x}{1+e^x} = \cfrac{1}{1+e^{-x}} = {\rm sigmoid}(x) y=f(x)=1+ex(1+ex)=1+exex=1+ex1=sigmoid(x)

(本节完)

  • 7
    点赞
  • 22
    收藏
    觉得还不错? 一键收藏
  • 1
    评论

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论 1
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值