Have you ever wondered how the human brain processes information? How do we see, hear, speak, think? Behind these complex functions lies an amazing network of 86 billion neurons. 人工神经网络，就是一种受生物学启发的编程范式，能够让计算机从可观测数据中学习，让它自己找出解决问题的方法。It is a network of interconnected artificial neurons. In this article, I will start with artificial neurons and take you into this intelligent world.
Perceptron
In 1943, American psychologist McCulloch and mathematical logician Pitts first proposed a binary neuron model, known as the McCulloch-Pitts model (MP model). They regard neurons as binary switching elements, and combine them in different ways to complete various logic operations, such as AND, OR, NOT, etc. Influenced by this research, in 1957, American scholar Rosenblatt proposed the perceptron. As the first model to lead neural network to engineering application, it has had a great impact on neural network research.
感知机接收若干个二进制输入，并生成一个二进制输出。权重为实数，表示输入对于输出的重要性。输出是0还是1,则由加权和小于或大于某个阈值来决定。
A perceptron can be thought of as a machine that makes decisions based on weights. Suppose you are considering whether to bookmark this article, you can make a decision by setting weights for the following 3 factors.
| factor | Weights |
|---|---|
| Is the content of the article valuable? | 0.8 |
| Is the style of the article interesting | 0.6 |
| Is the length of the article appropriate? | 0.4 |
If we set the threshold to 1.2, then when the input is (1,1,1), that is, when the content of the article is valuable, the style is interesting, and the length is appropriate, the output is 1, indicating a favorite; when the input is (0,1,0 ), that is, when the content of the article is worthless, the style is interesting, and the length is inappropriate, the output is 0, indicating that it is not favorited. By adjusting weights and thresholds, different decision models can be obtained.
sigmoid neuron
Small changes to the weights or biases on a single perceptron can sometimes cause the output to flip completely, such as 0 to 1. This is bad for learning, since we want to be able to fine-tune the parameters to improve the performance of the network. A sigmoid neuron can be introduced to solve this problem. A sigmoid neuron is similar to a perceptron, but small changes to the weights and biases cause only small changes in the output.
sigmoid 神经元的输入可以取0到1的任意值，而不限于0或1。
Summarize
This paper introduces the basic building block of artificial neural network—artificial neuron. We started with the earliest perceptron and learned about its structure, principle and application. We then introduce sigmoid neurons, addressing some of the limitations of perceptrons and laying the groundwork for subsequent learning.
