The foundation of deep learning--the beginning of neural network

The foundation of deep learning--the beginning of neural network

Original address fundamentals of Deep learning–starting with Artificial neural network preface

Deep learning and neural networks are now driving advances in computer science, both of which have a strong ability to learn from data and the environment, which also leads them to be the preferred tool in the field of machine learning.
Deep learning and neural networks are the core of automatic driving, image recognition and referral systems. Moreover, they can be used as common algorithms for many types of data.
Some people may think that neural networks are very complex. Therefore, many people who have not used or even used it as a black box, what is black box, which is not concerned about the implementation of in-depth learning details. But if you want to use them skillfully, you must understand them.
In this article, I will try to explain the concept of neural network in a simple and simple sense. Understanding this article only requires your mastery of simple biological knowledge and the patience to watch. When you finish reading this article, you will become a data analyst who can confidently use the neural network.
main content: What is the neural network of how a single neuron works why often use the basic structure of multilayer network neural network feedback (important. What is a neural network

Neural networks also known as artificial neural (ANN) networks, as the name suggests, are artificial models of the human nervous system. Let's learn the human nervous system again.

The main components of the human nervous system are: dendritic: A functional neuron that receives stimulation and passes the impulses into the cell body: the structure of the nervous system and the functional unit synapses: the function that will stimulate the efferent cell body.

In simple terms, each neuron receives stimulation from a plurality of neurons through the dendrites. It then carries out the necessary processing of the received stimulus and transmits the pulse current to the synaptic terminal according to a specific rule.

The working mode of Artificial Neural network (ANN) is very similar. The basic working framework of Neural network is as follows

The diagram shows how a single neuron works in a typical neural network, which is explained in detail below.
Like the human nervous system, data input is the same as the dendrites that receive stimuli and then the neuron checks and processes the input. Finally, the data is transferred to another neuron (or next layer) that is connected to it. The neural network can be divided into three layers: input layer: The neurons in the input layer are used to enter training data or test data hiding layers: this is a layer hidden between the input and output layers, which enables the neural network to learn the intrinsic complex contact output layer in the data: The output layer is the result output of the first two layers processing. For example: In a five-class system, the output layer has five neurons.

Then we will learn the function of individual neurons by example. how a single neuron works

In this section, we will look at the work of individual neurons through a simple example. By this example, there is a subconscious understanding of how neurons can be outputted through input calculations. The typical neurons are as follows:

The components of a single neuron are:
1. X1 x_1, X2 x_2,..., XN x_n: input to the neuron data, can be the initial input data can be the hidden layer output of the intermediate data
2. X0 x_0: Deviation value, introduced as a constant to adjust the result.
3. W0 w_0, W1 w_1, W2 w_2,..., WN w_n: Specify the weights for each input. Even the deviation value has the weight
4. A: The output of neurons, calculated as

Here F f is called an activation function, which makes the neural network very flexible and gives the neural network the ability to estimate the complex non-linear relationships in the data. It can be a Gaussian function, a logical function, and a hyperbolic function can also be a simple linear one.

We will then use the neural network to implement or,and,not three basic logical operators to deepen our understanding of neural networks. You can also think of this problem as a classification problem.

For the moment, we'll abstract our questions into the following activation functions:

Example 1:and

We can think of and as the following neurons.

For this neuron output is
A=f (−1.5+x1+x2) a=f ( -1.5+x_1+x_2)

With the activation function, we can draw the truth table

Through the truth table we can see that the function of the AND operator has been implemented, and the result of column A is ' x1andx2 x_1 and X_2 '. Here our deviation value is set to-1.5, but this value is not fixed, as long as the value of the activation function is positive when the X1 x_1, X2 x_2 is positive, so X0 x_0 can take any value example between ( -1,2) 2:or

Similarly, we can use the following neurons to implement or:

The output of the neuron is
A=f (−0.5+x1+x2) a=f ( -0.5+x_1+x_2)

Truth table for

Same as a column ' X