Andrew Ng--------Gradient Desecent

Gradient Descent:


反向传播求导过程:


根据链式法则(chain rule)

dj/dv = 3

dj/da = dj/dv  *  dv/da = 3 * 1 = 3

dj/db = dj/dv  *  dv/du  *  du/db = 3*c= 6


当你编程实现反向传播时,在代码里:dj/db写成d b   即可!





猜你喜欢

转载自blog.csdn.net/qq_26577455/article/details/79951327