A Simple Understanding of What Neural Networks Are

The term 'deep learning' refers to training neural networks. 'Deep' means very large neural networks. So what exactly are neural networks?63;After reading this article, you will have a very intuitive understanding.

Let's start with an example of housing price prediction. Because now the housing prices are too damn high, early...8Before the New Year, my parents said they would help me buy a house in Beijing, but I thought I couldn't rely on my family, so I refused. Now I think, I'm a moron. If I had bought it then, I wouldn't have to write blogs now~~

It is said that the housing prices are all driven up by the Chinese people themselves, but in addition to speculation, there are still some real factors that affect housing prices. By analyzing these factors, we can predict housing prices. Suppose you have a dataset (the area and price of six houses). You want to find a method (i.e., build a function) to predict the price from the area. If you are familiar with linear regression (if you are not familiar with it, you can just consider it as a mathematical theory), then you can draw a straight line near the actual data according to this theory, as the red line in the figure above. The blue small circles around it represent the points corresponding to the area and price of the six houses, that is, to find the price corresponding to a certain area according to this line (this function). Then, except for some errors, it is basically accurate. Moreover, we know that the price will never be negative, unless you are a triad, occupying someone else's house and still making them pay you money. Therefore, when the area is zero, the price is also zero, so the line above needs to turn a corner. You can consider this function that predicts the price based on the area as a very simple neural network (which is almost the simplest neural network). As shown in the figure on the right, the yellow circle is a neuron, the size of the house is x, which enters the neuron as input, and then the neuron outputs the housing price y.

In the literature related to neural networks, you will often see this function (the red line above). This function is 0 for a certain period of time, and then suddenly takes off and soars. It is called the linear unit function (ReLU, the full name is rectified linear unit). Now you don't have to worry if you don't understand it, you will understand it gradually later!

The example of predicting house prices using area above is a tiny single-neuron neural network, and by stacking many such neurons together, we can form a larger neural network. You can think of a neuron as a building block, and you can obtain a larger neural network by stacking many such building blocks together. In the example of house prices, as shown in the figure above, the factors affecting house prices are not only area, but also, for example, the number of bedrooms. In addition to area, the number of bedrooms also determines whether the house is suitable for your family, for example, if there is}}3person or5person. Another factor is the postal code. The postal code here represents the geographical location, and if the location is prosperous, life will be more convenient. Then there is another factor, the quality level of schools, which is known as school district housing in China. These factors, we also call them features. According to the area and the number of bedrooms, we can judge whether it meets the family size. According to the postal code, we can estimate the convenience of living, including going to school. Finally, the school level can evaluate the quality of education. People consider these four factors when buying a house, that is, they determine the house price, and there is also the他妈的 speculation, let's ignore this factor for the time being. So in this example, x represents this4The input, y is the price to be predicted. The left figure above is our artificial analysis process, we analyze which inputs affect the matching degree of family size, living convenience, and educational quality, and then deduce the price in our hearts. One of the amazing things about neural networks is that we only need to provide input x (area, number of bedrooms, etc.) and the desired result y (house price), as well as the real data used for training (the area, number of bedrooms, etc. of the six houses above and the price), and the intermediate part of family size matching, living convenience, and educational quality will be automatically determined by the neural network according to the actual data, as shown in the right figure. When you input the second7After determining the characteristics of a house (area, postal code, etc.), this neural network will give you the final price based on the previously trained intermediate part. That is, the internal process of the neural network replaces the analytical process of humans. To give another example, when we teach children to recognize cats, we bring some white cats and tell them this is a cat, and then bring some black cats and tell them this is also a cat, and then bring some dogs and tell them this is not a cat, and finally bring some calico cats and ask them, and they will tell you this is a cat, but how do they know? What is the process of this intermediate judgment? We don't know. This is why it is said that neural networks are terrifying - their working principle is too similar to humans - we don't know how they operate internally, that is, we don't know how they think, just like we don't know how children specifically distinguish between cats and dogs. Musk said that artificial intelligence is dangerous and may destroy humanity, which is not without reason.

The neurons in the middle part mentioned above are called hidden neurons. You may have noticed that in the neural network shown in the right figure, each hidden neuron receives all the feature inputs. In the left figure, it is our human analysis process. We analyze that the area and the number of bedrooms are related to the size of the family, and in the right figure, we provide all the features to each hidden neuron to let the neural network analyze which features will affect the price of the house from which aspects. Maybe the analysis of the neural network is more efficient and comprehensive than that of humans. I saw a news report before that Facebook's neural network invented a way of communication by itself because it thought that communicating with human language was too inefficient~~

After reading this article, you should still be somewhat confused about neural networks... That's right, if just one article can make you understand neural networks, you are a genius.

That's all about the content of this article about a simple understanding of what a neural network is. I hope it will be helpful to everyone. Those who are interested can continue to refer to this site:

Python and Artificial Neural Networks: Introduction to Using Neural Networks to Recognize Handwritten Images

70 lines of Java code to implement deep neural network algorithm sharing

Welcome to leave a message if there is anything insufficient. Thank you for your friends' support to this site!

Declaration: The content of this article is from the Internet, and the copyright belongs to the original author. The content is contributed and uploaded by Internet users spontaneously. This website does not own the copyright, has not been manually edited, and does not assume any relevant legal liability. If you find any content suspected of copyright infringement, please send an email to: notice#w3Please report via email to codebox.com (replace # with @ when sending email) and provide relevant evidence. Once verified, this site will immediately delete the infringing content.