当前位置：网站首页>Machine learning - matrix derivation basic formula + common formula

Machine learning - matrix derivation basic formula + common formula

2022-07-18 04:06:00 【Eschatological light】

stay depth When learning the back propagation of neural network , According to the last output loss Update the parameters of the network in reverse , The training process of neural network is actually the updating process of parameters , The parameters are updated according to the gradient descent method , The calculation of gradient requires partial derivatives .

d(A^T*x)/dx = d(x^T*A)/dx = A

Transpose of scalar equals scalar itself ：

a*b^T = c      c Is constant       c^T = c

  therefore ：a*b^T = (a*b^T)^T = a^T*b

 d(x^T*A*x)/dx = (A + A^T)*x

The matrix cookbook download ： link   Extraction code ：s8h3 

原网站

版权声明
本文为[Eschatological light]所创，转载请带上原文链接，感谢
https://yzsam.com/2022/198/202207141651558433.html

边栏推荐

猜你喜欢

随机推荐