Introduction to neural networks (serialization)

 

. AI technology in game programming
. (Serialization)

Introduce Neural Networks in common languages

(Neural Networks in plain English)

Because we don't have a good understanding of the brain, we often try to use the latest technology as a model to explain it. In my childhood, we all believed that the brain was a telephone switch. (What else can it be ?) At that time, I also saw Xie Linton, a famous British neuroscientist, cool up the brain's work as an electronic reporter. Earlier in the year, froyd often compared his brain to a hydro-electric generator, while lavenitz compared it to a mill. I also heard people say that the ancient Greek think of the brain function as a slingshot. Obviously, it is only a digital electronic computer to compare the brain. John R. Searle [NOTE 1]

 

Introduction to Neural Networks
Networks)
For a long time, artificial neural networks were completely mysterious to me. Of course, I have read about them in the literature. I can also describe their structure and working mechanism, but I have never been able to "Aha !" Just as a concept in your mind that is difficult to understand and suddenly understands. I seem to have been hitting a ghost on my head, or a movie named animal.
The one in house (named "Animal House" in Chinese) is screaming bitterly. "Thank you, sir. Please give me another one !" The poor guy. I cannot translate the concept of mathematics into practical applications. Sometimes I even want to catch the authors of all the neural networks I have read, bind them to a tree, and yell at them loudly: "Don't give me any more mathematics, give me something practical! ". But it is never possible. I had to fill this gap myself... so I did the only thing I could do under that condition. I started to work. <Smile>

A few weeks later, on a beautiful day, when I was on vacation at the shore of Scotland, my mind was suddenly shocked when I stared at the narrow bay through a mist. I suddenly realized how artificial neural networks work. I got "Aha !" Now! But now I have only one tent and a sleeping bag, and half a box of cornflakes. Without a computer, I can quickly write some code to prove my intuition. Arghhhhh! Then I thought that I should buy a laptop. Anyway, a few days later, when I got home, I immediately put my fingers on the keyboard. A few hours later, my first artificial neural network program was finally compiled and run, and it worked quite well! Naturally, code writing is messy and needs to be organized, but it does work and, more importantly, I still know why it works! I can tell you that I was a very proud person that day.

I hope this book will be delivered to you !" Feeling. When we finish learning the genetic algorithm, you may already have a taste of it, but if you want it to be wonderful, you have to finish learning the neural network part.

Biological neural network-brain (a biological neural network-the brain)
Your brain is gray and frozen. It does not use a single processing unit to work like a CPU in a computer. If you have a corpse that is fresh and stored in formalin, you can see the familiar wrinkles in the brain tissue after carefully sawing its skull with a saw. The outer layer of the brain is like a big walnut, all of which are wrinkled [0 left]. This layer of tissue is called cortex ). If you carefully use your fingers to remove the entire brain from the middle of your head, take a surgeon's surgical knife and cut the brain into slices, then you will see two layers of the brain [figure 0 right]:
The gray outer layer (this is the source of the word "gray", but the fresh brain that has not been fixed with formalin is actually pink .) And the White inner layer. The gray layer is only several millimeters thick, compress billions of tiny cells called neurons. The white layer occupies most of the space in the cortex under the gray matter of the cortex and is composed of countless connections between neural cells (but there is no neural cell itself, just like the back of the printed circuit board, only the link of the component, but not the component itself ). The cortex is wrinkled like a walnut, which can insert a large surface area into a small space. This can accommodate more nerve cells than smooth cortex. The human brain contains about 10g (or 10 billion) such tiny processing units; an ant brain also has about 250,000.

Table 1 below shows the number of nerve cells in humans and several animals.

Table L Number of nerve cells in humans and several animals Animation category Number of nerve cells (orders of magnitude) Youniu 10,000 (= 104) Beebot 100,000 (= 105) Fengque 10,000,000 (= 107) Old Rat 100,000,000 (= 108) Persons 10,000,000,000 (= 1010) Elephant 100,000,000,000 (= 1011)

 

Figure 0-1 walnut in the cerebral hemisphere Figure 0-2 cerebral cortex consists of gray and white matter Figure 0 shape and slice shape of the brain

 

Figure 1 Neural Cell Structure

In the first nine months of human life, these cells were created at an astonishing rate of 25,000 per minute. Nerve cells are very different from any other type of cell in humans. Each nerve cell has a wire-like things called axon, it is sometimes stretched several centimeters in length to send signals to other nerve cells. The structure of neural cells is shown in Figure 1. It consists of a cell body (SOMA), some DCs (dendrite)
, And can be a long axon. The neural cell body is a star shape with a nucleus in it ). The DC grows from the cell body to each direction, and can have branches, which are used to receive signals. There are also many branches of the axon. By connecting the branch ends (terminal) with the DCS of other nerve cells, the axon forms the so-called synapse (not shown in the figure ), A nerve cell sends signals produced by the delivery of these signals to other nerve cells through the axon and syn.

Each nerve cell is connected to about 10,000 other nerve cells through its DCS. This makes the total number of connections between all nerve cells in your mind may be 1 to 000,000,000,000,000. This is more connected than 100 m modern telephone switches. So it's not surprising why we sometimes have headaches!

Interesting fact It has been estimated that, if a person's brain connects all the nerve cells of the axon and the DC, and pulls them into a straight line, the Earth can be connected to the moon, then return from the moon to the Earth. If we connect the neurons in the brains of all people on the earth to the DCS, we can stretch out to the nearest galaxy!

Neural Cells exchange signals through electro-chemical processes. The input signal comes from other nerve cells. These nerve cells, the terminal, and the neighborhood of the nerve cells form a synapse, and the signal enters the local cell from the SYN on the DC. How signals are actually transmitted in the brain is a very complicated process, but as far as we are concerned, it is important to look at it like modern computers, use a series of 0 and 1 operations. That is to say, there are only two States of brain nerve cells: Fire and not excited ). The intensity of the transmitted signal remains unchanged, but only the frequency is changed. Neural cells use a method we do not know to add all the signals from the DC. If the sum of all signals exceeds a threshold value, it will trigger the nerve cells into the fire state, and then an electrical signal will be sent out to other nerve cells through the axon. If the total signal does not reach the threshold, the nerve cells will not be excited. This explanation is a bit too simple, but it can satisfy our purpose.

It is precisely because of the massive number of connections that the brain has incredible capabilities. Although each nerve cell only works at a frequency of about Hz, because each nerve cell is working in parallel in an independent processing unit, the human brain has the following obvious characteristics:

It can achieve unsupervised learning. One of the incredible facts about our brains is that they can learn on their own without the supervision and guidance of mentors. If a nerve cell is subject to high-frequency stimulation for a period of time, the connection strength between it and the nerve cell that inputs the signal changes according to a certain process, this makes the nerve cell more excited when it is motivated next time. This mechanism was written by donard Hebb more than 50 years ago in organination.
Of behavior. He wrote:

"When one axon of nerve cell a repeatedly or persistently inspires another nerve cell B, then one or both of the nerve cells will have a growth process or metabolic change, which will increase the efficacy of a cell that inspires one of the B cells"

 

On the contrary, if a nerve cell is not motivated for a period of time, its connection effectiveness will gradually decline. This phenomenon is called plasticity ).

Tolerance ). Even if a large part of the brain is damaged, it can still perform complex work. A famous test is to train a mouse to walk in a maze. Scientists then cut off part of their brains from the ground up. They found that even if a large part of the mouse's brain was cut off, they could still find walking paths in the maze. This proves that in the brain, knowledge is not stored in a local place. Other experiments show that if a small part of the brain is damaged, neural cells can regrow the damaged connection. [Note: I think this can also happen in Humans: after a period of rehabilitation training, patients with cerebral infarction that causes large-scale brain necrosis due to cardiovascular disease or other causes can also recover to health, especially, memory is not impaired.]

Information processing efficiency is extremely high. The transmission of electrical-chemical signals between neural cells is very slow compared with the data transmission of CPU in a digital computer. However, due to the parallel operation of neural cells, this allows the brain to process large amounts of data at the same time. For example, when the brain visual cortex processes an image signal input through our retina, it takes about Ms. Considering that the average working frequency of your nerve cells is only 100Hz, Ms means that only 10 computing steps can be completed! Think about how much data we see in our eyes, and you can see that this is an incredible great project.

Good at inductive promotion. Unlike a digital computer, a brain is very good at pattern recognition and can be generalized Based on familiar information (generlize ). For example, we can read the texts in others' manuscripts, even if we have never seen anything written by others before.

It is conscious. Consciousness is a broad and heated topic for neuroscientists and AI researchers. A large number of documents have been published on this topic, but there is no substantive unified view on what consciousness is. We cannot even agree that only humans are conscious, or that humans, including their close relatives in the animal kingdom, are conscious. Is an orangutan conscious? Is your cat conscious? Is the fish you eat conscious last week's dinner?

 

Therefore, an artificial neural network (
Artificial Neural Network (ANN) is to simulate a large amount of parallelism under the constraints of the existing scale of contemporary digital computers, so that it can display many similar features as the biological brain. Let's take a look at their performances!