Deep Neural Networks - Building blocks of deep neural networks

王彩旗 edwardwangcq.com

于 2021-07-22 12:50:44 发布

阅读量119

点赞数

分类专栏：人工智能 # Neural Networks and Deep Learning

版权声明：本文为博主原创文章，遵循 CC 4.0 BY-SA 版权协议，转载请附上原文出处链接和本声明。

本文链接：https://blog.csdn.net/edward_wang1/article/details/118996581

版权

人工智能同时被 2 个专栏收录

142 篇文章 0 订阅

订阅专栏

Neural Networks and Deep Learning

32 篇文章 0 订阅

订阅专栏

Take the deep NN of figure-1 as the example:

Figure-1

Following figure-2 shows the building blocks of this deep NN while generalized the figure for NN with L layers:

Figure-2

Forward and backward propagation for single example x:

Forward propagation:

Input $a^{[l-1]}$ , output $a^{[l]}$ , cached $z^{[l]}$ :

$z^{[l]}=W^{[l]}a^{[l-1]}+b^{[l]}$

$a^{[l]}=g^{[l]}(z^{[l]})$

The initial input is vector x

Backward propagation:

Input $da^{[l]}$ , output $da^{[l-1]}, dW^{[l]}, db^{[l]}$ :

$dz^{[l]}=da^{[l]}*g^{[l]'}(z^{[l]})$

$dW^{[l]}=dz^{[l]}a^{[l-1]^{T}}$

$db^{[l]}=dz^{[l]}$

$da^{[l-1]}=W^{[l]^{T}}dz^{[l]}$

The initial input is $da^{[L]}=-\frac{y}{a}+\frac{(1-y)}{(1-a)}=-\frac{y}{\hat{y}}+\frac{(1-y)}{(1-\hat{y})}$

A side notes here is how to calculate $dz^{[l]}$ from previous layer $dz^{[l+1]}$ :

$dz^{[l]}=W^{[l+1]^{T}}dz^{[l+1]}*g^{[l]'}(z^{[l]})$

Forward and backup propagation for m examples X:

Forward propagation:

Input $A^{[l-1]}$ , output $A^{[l]}$ , cached $Z^{[l]}$

$Z^{[l]}=W^{[l]}A^{[l-1]}+b^{[l]}$

$A^{[l]}=g^{[l]}(Z^{[l]})$

The initial input is X like

Backward propagation:

Input $dA^{[l]}$ , output $dA^{[l-1]}, dW^{[l]}, db^{[l]}$

$dZ^{[l]}=dA^{[l]}*g^{[l]'}(Z^{[l]})$

$dW^{[l]}=\frac{1}{m}dZ^{[l]}A^{[l-1]^{T}}$

$db^{[l]}=\frac{1}{m}np.sum(dZ^{[l]}, axis=1, keepdims=True)$

$dA^{[l-1]}=W^{[l]^{T}}dZ^{[l]}$

The initial input is:

$dA^{[L]}=\begin{bmatrix} -\frac{y^{(1)}}{a^{(1)}}+\frac{(1-y^{(1)})}{(1-a^{(1)})} & -\frac{y^{(2)}}{a^{(2)}}+\frac{(1-y^{(2)})}{(1-a^{(2)})} & ... & -\frac{y^{(m)}}{a^{(m)}}+\frac{(1-y^{(m)})}{(1-a^{(m)})} \end{bmatrix}$

<end>

评论

被折叠的条评论为什么被折叠?

到【灌水乐园】发言

查看更多评论

添加红包

成就一亿技术人!

hope_wisdom

发出的红包

实付元

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。