Get to know Neural Network (a) Neural Networks

Feature Extraction

Feature Extraction: My understanding is that the data of the object, as an entity described as the computer can understand, then it is the definition of such a person, name, gender, hobbies, place of birth, date of birth ... so you can give him enough labels, you can define who he is, this is the label features.

Neural Networks

Using a neural network has four steps

1. Entity extraction from feature vectors as input to the neural network, different entities may be extracted in different feature vectors
2. Defines the structure of the neural network, and defines how to get from input to output of the neural network, the neural network process is the front propagation algorithm
3. To adjust the values ​​of the parameters of the neural network by training data, which is the process of training the neural network
4. The use of trained neural network to predict unknown data

Forward propagation

Look at this chart shows the process forward propagation

1. Corresponding to the input layer of the neural network input
2. The connection structure of the neural network is the connection between different nodes
3. Determining the parameter of each node, it is to determine the various values ​​of W

Such inputs X1 and X2 can be obtained by the propagation front Y.

Back-propagation algorithm

Before getting a lot after the dissemination of results, we need to define a loss function to describe the gap between the current predicted value and the true answer. Then adjust the value backpropagation neural network parameters so that the gap can be reduced.

This is commonly used to determine a value above an algorithm W

Train the neural network

The above procedure realize the whole process of training the neural network. From this process it can be concluded that the process of training the neural network can be divided into the following three steps:

1. Output structure and the forward propagation neural network is defined.
2. Loss function defined backpropagation algorithm selection and optimization.
3. Generating a session (tf.Session) and the training data is run repeatedly back propagation optimization algorithm.

Whatever the structure of the neural network changes, these three steps are the same.