神经网络模型 更多见：iii.run

Neural Networks Model

A single neuron model: logistic unit

$$\begin{bmatrix}x_0 \newline x_1 \newline x_2 \newline \end{bmatrix}\rightarrow\begin{bmatrix}\ \ \ \newline \end{bmatrix}\rightarrow h_\theta(x)$$

• Takes 3+1 inputs(the extra input called bias is just like $θ_0$ in logistic regression, not shown in picture).
• Both input and output could be represented as vectors, in which each unit has its own parameters $θ$
• All the units in the same layer take the same input $x$, as the pic shows.
• Each unit has only one output: $sigmoid(θ^Tx)$. Of course there're other choices for sigmoid function.

Neural Networks

$$\begin{bmatrix}x_0 \newline x_1 \newline x_2 \newline x_3\end{bmatrix}\rightarrow\begin{bmatrix}a_1^{(2)} \newline a_2^{(2)} \newline a_3^{(2)} \newline \end{bmatrix}\rightarrow h_\theta(x)$$

Calculation from one layer to the next

In the picture above, we have the networks from layer j to layer j+1, in which layer j has 3(+1) units while layer j+1 has 3 layers. Let $s_j=3$, $s_j+1=3$

• $α^{(j)}$ : Output of the $j_{th}$ layer. $s_j+1$ dimension vector.
• $θi^{(j)}$ : Parameters in the $i{th}$ unit of $(j+1)_{th}$ layer. $s_j+1$ dimension vector.
• ${\theta^{(j)}} = \begin{bmatrix} \theta_1^{(j)} & \theta_2^{(j)} & \cdots & \theta_{s_(j + 1)}^{(j)} \end{bmatrix}^T$ : All the network parameters from $j_{th}$ layer to ${(j+1)}_{th}$ layer.
• We have: $\alpha^{(j+1)} = sigmoid(\mathbf{\theta^{(j)}}\alpha{(j)})$ add $\alpha_0^{(j+1)}$

Multiclass Classification

分类器

Each $y^{(i)}$ represents a different image corresponding to either a car, pedestrian, truck, or motorcycle. The inner layers, each provide us with some new information which leads to our final hypothesis function. The setup looks like:

Summary

Example: layer 1 has 2 input nodes and layer 2 has 4 activation nodes. Dimension of $\Theta^{(1)}$ is going to be 4×3 where $s_j = 2$ and $s_{j+1} = 4$, so $s_{j+1} \times (s_j + 1) = 4 \times 3$$

本小节笔记，有大段参考自 shawnau.github.io