Future Super Intelligent System

Everyone who had already started to get tired subconsciously lifted their spirits. In normal companies, there are two people who come to participate in this kind of proposal, a project leader and a technician.

Therefore, whether it is a project manager who wanders in the market all year round or a technician wearing a plaid shirt, they are naturally full of interest in what Liu Fan will say next.

"The reason why our face recognition technology can realize the application of the scene you saw just now is based on the logic theory of dynamic features we proposed."

Liu Fan started to explain, and some people started to record, some started to video, and some people started to take notes.

"The core of our current mainstream face recognition technology is features. Whether it is overall geometric features or local features, or whether it is traditional algorithms or neural network simulations, the root of everything is to capture features and process them. But no matter the follow-up How dynamic or how high-dimensional is the processing method, the features we capture are themselves static.

It is also impossible to avoid being easily affected by the environment. So we have long considered whether we can capture dynamic features instead of static features. Later, we realized that simply capturing dynamic features was not enough. If big data can be used to mine dynamic feature logic, then no matter how the environment affects, as long as we can build a large enough dynamic feature logic model, even if we rely on very vague Image data may also enable face recognition.

I simply give an example. "

Liu Fan said, and clicked the PPT to the next page, "First of all, everyone here must know a basic principle, that is, one leaf, one bodhi. There are no two identical leaves in this world, and there are no two leaves with the same growth process.

You can watch the PPT. This is the first time we discovered the dynamic feature logic. After continuously mining these 30 million dynamic faces, the computer came up with an unexpected answer. There is a functional relationship between the rotation and the muscle changes at the position of the red point below the cheek, and this function is the same between different people, the only difference is in the coefficient.

The specific function involves commercial privacy, so I won’t show it here. I just make a simple analogy. The functional relationship between A’s eyeballs and cheeks is F(X)=y, and the functional relationship between B’s is F(X)=1.1y. The functional relationship is F(X)=1.2y, and so on, so when the video captures the variable X of a certain person's eyeball rotation, the corresponding function result is unique, and we currently have a dynamic function like this on the face 26 were found, and we're still working on it.

I believe everyone can understand that the advantages of dynamic functions are very obvious. For example, if a criminal covers his face, but as long as his eyes can be seen, as long as his eyes provide a variable X, then several combinations with eyes can be performed. Calculation of related functions, maybe he also covers his eyes, but the muscle changes in a certain part of the face also have matching corresponding functions, then..."

"You let the algorithm deduce the algorithm!?" Liu Fan was halfway through speaking, when someone interrupted him suddenly.

And when the man in the black plaid shirt said this, the audience immediately became restless.

They just listened to Liu Fan's talk, but they haven't had time to realize what Liu Fan's algorithm would mean to the artificial intelligence industry if it really existed.

The so-called deep learning algorithm is the learning of ability. To give a slightly one-sided example, let the computer keep learning multiplication, and the multiplication calculation speed of the computer will gradually become faster and faster.

To use a more realistic example, many companies are now starting to develop robots. When you see robots interacting with people, you will feel that the era of artificial intelligence has really come. But in fact, in the process of a robot talking to a human, we simplify the process of deep learning. In fact, what to say is an optimized answer under a certain program setting.

For example, a girl told the robot that I was angry. The current thinking mode of the intelligent robot at this time is like this. She said she was angry. Based on the principle that women are unreasonable when they are angry, the optimal solution is obtained at this time. , Just apologize and send her a red envelope.

But how do real people think about things? Is she really angry? Why is she angry? So what should I do in such a situation? If it's not a big deal, then please apologize, but if it's a matter of principle, you must not spoil her, because it may ruin her and yourself.

This is just an example. Boys will have many more thoughts than this in real relationships.

In fact, such a comparison makes it easy for everyone to understand the difference between humans and artificial intelligence. One is to follow the set program to analyze and get the optimal solution, and the other is to react differently based on life experience and emotions. . And this instant response is theoretically completely chaotic and unpredictable.

We can find that people go through several processes when dealing with one thing: perception, analysis, and decision-making.

What artificial intelligence is doing now is information input, information analysis, and output of optimal results.

Therefore, in terms of information processing, artificial intelligence looks similar to human beings, but once emotions and disordered things are involved, the difference between artificial intelligence and human beings will come out.

People will die for those they love, and today's artificial intelligence will never.

At this time, look back at the algorithm implemented by Liu Fan. If the algorithm can derive various functions based on the continuously input data, the behavior pattern at this time can be simulated to a certain extent: people start from a completely ignorant baby. Growing up, I learned more and more about the world.

In fact, for artificial intelligence, the so-called algorithms, functions, and rules may be the survival rules of human beings. Isn't a person's whole life a process of constantly understanding life?

If we refine the process of human growth, is it possible that we are constantly discovering countless rules? It's just that the brain handles these problems, and we don't realize it ourselves.

So when the algorithm can discover the rules by itself, it is equivalent to opening another door for artificial intelligence. Although this direction is not necessarily completely correct, theoretically speaking, it is likely to be closer to the door of real artificial intelligence.

But this thing is easy to talk about, but it is too difficult to realize it. Just like Zhang Kaixiang has a strong mathematical foundation and corresponding conjectures, but he has no way to start. If Liu Fan didn't have the abnormal plug-in of the system, he might not be able to do it in his life if he wanted the algorithm to independently mine functions.

Of course, in fact, Liu Fan’s current algorithm is still far from the ideal state. On the one hand, there is still a lot of room for the accuracy of function derivation and mining capabilities, because it will be limited by the amount of data and data types. . For example, before this face recognition, Liu Fan's data association inverse calculation algorithm has not made any breakthroughs.

Although in the past, it is indeed possible to realize the use of unordered data as he imagined before, but the previous functions did not break through the traditional mathematical framework underlying the algorithm, which leaves the question of whether these functions are derived by the algorithm itself remains to be seen. Discussed.

However, this experiment on face recognition allowed Liu Fan to see a breakthrough, and the self-derivation of the algorithm seems to be really feasible.

Another problem is that Liu Fan's algorithm can only analyze, not make decisions.

He tried many methods, but could not achieve decision-making ability.

He gradually realized that in order for the algorithm to have decision-making ability, it is necessary to break the existing underlying algorithm principles.

But this does not affect the shock caused by Liu Fan's current algorithm to the technicians present today. Liu Fan and Zhang Kaixiang will have such conjectures, and others may have such conjectures, but no one can overcome the technical difficulties That's all, so when everyone realized what Liu Fan's presentation would mean if it was true, it was difficult for everyone to control their emotions.

And everyone's shock was exactly what Liu Fan expected. From the moment the GSCT was headed, Wood Dragon Technology was an army standing on the battlefield ready to attack the city, and there was no need to hold back...