一个样本:
FP:
P:x变量个数
n1:第一层隐藏层个数
n2:第二层隐藏层个数
z[1] = w[1] x + b[1]
n1X1=n1XP * PX1 + n1X1
a[1] = g(z[1])
n1X1
z[2] = w[2] a[1] + b[2]
n2X1 = n2Xn1 * n1X1 + n2X1
a[2] = g(z[2])
n2X1
L(a[2],y)
1X1
bp:
dz[2] = a[2] - y
n2X1
dw[2] = dz[2] a[1].t
n2Xn1 = n2X1 * 1Xn1
db[2] = dz[2]
n2X1
dz[1] = da[1] *g’(z[1]) =w[2].t dz[2] *f’(z[1])
n1X1 = n1X1 * n1X1 = n1Xn2 * n2X1
dw[1] = dz[1] x.t
n1XP = n1X1 * 1XP
db[1] = dz[1]
n1X1
m个样本:
fp:
Z[1] = [z[1](1) , z[2](2) , … , z[m](m)]
Z[1] = w[1] X + b[1]
n1 X m = n1 X P * PXm + n1 X (1 * m)
A[1] = g(Z[1])
n1 X m
Z[2] = w[2] A[1] + b[2]
n2 X m &#