Introduction of Neural Network 神经网络总论
History of NN
Maybe I will write this part later.
Perceptron
Skip.
Neuron
If you don’t know basic neuron model, please search it.
You can add a neural
z
0
(
l
)
/
x
0
z_0^{(l)} / x_0
z0(l)/x0 to every layer or use bias
b
b
b. Both of them are correct. I will use bias item here.
图片来自Andrew Ng在Coursera的课程
以下图片均来自浙大课程截图
Attention! The teacher reversed
a
a
a and
z
z
z by mistake.
{
z
1
=
ω
11
x
1
+
ω
21
x
2
+
b
1
z
2
=
ω
12
x
1
+
ω
22
x
2
+
b
2
a
1
=
ϕ
(
z
1
)
a
2
=
ϕ
(
z
2
)
y
=
w
1
a
1
+
w
2
a
2
+
b
\left\{\begin{array}{l} z_1 = \omega_{11}x_1 + \omega_{21}x_2 + b_1 \\ z_2 = \omega_{12}x_1 + \omega_{22}x_2 + b_2 \\ a_1 = \phi(z_1) \\ a_2 = \phi(z_2) \\ y = w_1a_1 + w_2a_2 + b \end{array}\right.
⎩⎪⎪⎪⎪⎨⎪⎪⎪⎪⎧z1=ω11x1+ω21x2+b1z2=ω12x1+ω22x2+b2a1=ϕ(z1)a2=ϕ(z2)y=w1a1+w2a2+b
ϕ
\phi
ϕ is activation function. We use step function for now.
ϕ
(
x
)
=
{
0
(
x
<
0
)
1
(
x
≥
0
)
\phi(x) = \left\{\begin{array}{l}0 (x <0) \\ 1 (x\geq0)\end{array}\right.
ϕ(x)={0(x<0)1(x≥0)
Data scientists have proved that all kinds of decision hyperplanes can be simulated by three layer neural network.
扫一扫
267
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
