Chapter 169: The Inception of Convolutional Neural Networks(1/2)
Chapter 169 The Prototype of Convolutional Neural Network
Author: September Sauce
Chapter 169 The Prototype of Convolutional Neural Network
Looking at the energetic Wang Jiancheng, Chen Nian felt more and more that this boy was really not simple.
If when he proposed a series of future weapon concepts before, I just thought he was chasing dreams, then after talking to him today, I am afraid I have to change my words.
It’s not chasing dreams, it’s making dreams.
It's very difficult, but not impossible.
"How are you going to achieve that? Drone combat is a complex system, especially when it comes to coordinated operations."
"You don't have any understanding of the data link system, and your knowledge system is not complete enough. Isn't it too difficult to implement the loyal wingman program?"
After hearing Chen Nian's words, Wang Jiancheng nodded and replied:
"Senior, I also know that it is difficult, so I didn't plan to make the project too complicated from the beginning."
"What I want to do is a remote control system that responds in a timely manner."
"Generally speaking, the control of the drone is to be integrated into the existing fighter avionics system, and through simple control, the drone can follow, fly with, attack, and cover."
"When the technology becomes more advanced in the future, we can use it as a foundation to implant various tactical templates to achieve one-click control."
"For example, when our satellite and data link systems evolve to be sensitive and powerful enough, we can even equip a J-22 with several drones, including electronic jammers, small early warning aircraft, and ground attack aircraft.
There’s a machine, and there’s also a shield machine.”
".Awesome."
Chen Nian sighed sincerely.
What is awesome about this set of ideas is not the idea of drone tactics he proposed, but this sentence:
system integration.
He probably thought what he was doing was simple, but in fact, he was really digging the foundation!
If this integrated control system can run smoothly and undergo open iterative development, maybe the Republic's first generation of "loyal wingmen" will really be born in his hands.
"Do you have an idea?"
Chen Nian continued to ask.
"Yes!"
Wang Jiancheng smiled, pushed away the almost eaten dinner plate, and answered:
"My plan is this. This system sounds simple, but in fact it is quite complex to implement."
"So, I don't have to do it all at once, so I just need to do a simple imitation and following part first."
"The implementation method is also very simple, directly through data synchronization, but the current problem is that there are differences in performance, speed, and flight envelope between the drone and the host. We need to use algorithms to smooth this gap, not
Very easy."
"Do you still know programming?"
Chen Nian asked in surprise.
"You know, my dad bought me a computer when I was very young. I bought it from an orchard at that time. I had no other use for it, so I taught myself programming."
"what language?"
"C#, python."
"Awesome."
In just a few minutes, Wang Jiancheng made Chen Nian sigh at his awesomeness several times.
However, when I think about it carefully, what is even more awesome is his father.
Wang Jiancheng's family conditions are not very good, he has an older sister at home, and the financial pressure is not small. Under such circumstances, he dares to make such a big gamble on educational resources, which shows that he is definitely not a short-sighted person.
This can be regarded as a kind of tiger father without dog son to a certain extent.
"Then what algorithm are you going to use to solve it?"
"Well, what I'm thinking about now is machine learning, so that the wingman has certain decision-making capabilities as a supplement to imitation. It mainly uses ensemble learning, specifically the random forest algorithm, but I feel that this algorithm is still a bit weak."
"We don't have any more good ideas at the moment. Let's take one step at a time. I think even so, entering the special class is enough."
That is definitely enough.
Even if it's not enough, I will ask Chen Guo to arrange it for you.
Chen Nian thought secretly in his heart.
However, the random forest algorithm does seem to be a little behind in this era.
In my own memory, probably by next year, that is, 2006, deep learning algorithms will appear.
Instead of letting him do random forest, why not go directly to deep learning?
This algorithm is not only superior in machine learning, it can even be said that the emergence of deep learning algorithms has triggered a frenzy of explosive growth in AI.
If Wang Jiancheng is really allowed to do it, the future will be immeasurable.
Even if it is just a start and a foundation for future generations, it is still a good attempt.
Thinking of this, Chen Nian said:
"I have an idea, you can try it."
"Now there should be some machine learning training models about layer-by-layer initialization appearing in the paper library. Based on this model, there is more potential that can be explored deeply."
"With your ability, it shouldn't be too difficult to use ensemble learning at this age and continue to explore in depth."
"You can first take a look at the most advanced experience in the world and try to apply it in your own practice."
"You can focus on the SVM method. If you have difficulties, you can ask a teacher majoring in statistics for help."
Having said this, Chen Nian paused for a moment.
He suddenly thought that at this time, it would be difficult for the other party to find more useful materials.
However, if you let yourself analyze this part of the content for him.
Wipe, 6.9 source points are consumed.
It's better to pull it down then.
So, he recalled carefully for a moment and continued:
"There are a few keywords that should be useful in your research process."
"Forget about kernel function expansion, linear regression function, and recursion. I will draw the general principles for you to see."
Chen Nian picked up his chopsticks, dipped them in the soup and began to write and draw on the table.
What he drew was the basic structure diagram of the convolutional neural network most commonly used for deep training of game AI at that time.
In fact, the entire principle of convolutional neural networks is not complicated. Even ordinary people with no programming foundation can understand the most basic parts in half a day.
The real difficulty is how to implement it with complex programs.
This is not what Chen Nian is good at. In his previous life, he only learned a little bit out of interest.
"Here, this is the filtering process diagram of the convolutional neural network. This is the C1 layer. Its function is to generate a feature map. Subsequently, the features are summed and weighted."
The soup gradually formed criss-crossing pictures on the table. Wang Jiancheng had been listening with all his heart from the very beginning, so even if Chen Nian only spoke once, he had already memorized the entire content in his heart.
When Chen Nian put down his chopsticks, the way he looked at Chen Nian had completely changed.
I originally thought that this was just an ordinary, enthusiastic senior with some minor achievements.
But now it seems.
What kind of monster is this!
He said he didn't understand programming, but his backhand came out with an algorithm that was obviously more powerful than random forest.
Although this algorithm is indeed not up to the level of immediate use, his ideas are definitely top-notch!
You also said you don’t know martial arts!?
Wang Jiancheng carefully wiped off the excess part of the picture with a tissue, not knowing what to say for a while.
The students passing by looked at this scene curiously, with funny expressions on their faces.
He probably felt that these two people were crazy about studying and pushed their reasoning to the dining table.
But what he didn't know was that the picture on the dining table might become a popular story many years later.
After bidding farewell to Wang Jiancheng, Chen Nian strolled back to his residence by himself.
In fact, revealing information about convolutional neural networks to him is not a too advanced technology leak, because as early as 2002, the basic principles of convolutional neural networks were born.
Until the outbreak in 2006, most of the work of scholars in this process was actually supplementing and improving the basic principles.
What I just did was to convey what I already had to Wang Jiancheng. As for how to actually apply it, he still needs to explore it.
To be continued...