Day 5 neural Networks neural network

Neuron Model

　　

Neurons can be thought of as a computational unit that receives certain information from the input nerves, makes some calculations, and then transmits the results to other nodes or other neurons in the brain through axons.

The neuron is modeled as a logical unit, as follows:

　　

In, the input unit is X1 X2 X3, sometimes can also be added x0 as offset units, the value of x0 is 1, whether to add bias units depends on whether it is advantageous to the example.

The Orange small Circle in the middle represents a single neuron, and the neural network is actually a collection of different neurons combined together.

The output is the result of the calculation h (x).

Neural network

　　

　　

The input unit is X1 X2 X3, and the bias unit x0 can be added.

There are three neurons A1 (2) A2 (2) A3 (2) in the middle layer, and a A0 (2) can be added if the bias unit is added.

The last layer of the node is the output of the middle three nodes, assuming the result of the function.

Among them, the first layer is called the input layer, the middle layer is called the hidden layer, the last layer is called the output layer.

　　

　　

For example, a superscript (2) Subscript 1 represents the first excitation of the 2nd layer, that is, the first excitation of the hidden layer. The so-called excitation (activation) refers to a specific neuron after reading the information, need to use the parameter matrix, after a series of calculations and then pass the value to the next layer, wherein the calculation process is S-excitation function or called the logical excitation function.

Forward propagation forward propagation

　　

We start with the excitation of the input layer, then forward to the hidden layer and calculate the excitation of the hidden layer, and then proceed forward propagation to calculate the excitation of the output layer.

Neural Network and Logistic regression

　　

If we cover the input layer of the neural network, we will find that this is really a logical regression, in the logistic regression we predict H (x) with the given eigenvalues, and in the neural network we use the calculated values of the hidden layer to predict H (x).

Other neural networks

　　

　　

Day 5 neural Networks neural network