Chapter 487 Physicists Who Can’t Wait

While HiSilicon and Huaxin were busy with design and chip breakthroughs, Xu Chuan was also studying these mathematical problems related to chip design in the villa.

NPU neural network architecture is an emerging computer field compared to CPU architecture and GPU architecture.

It was first proposed by psychologists and neurobiologists, and applied to computers by simulating the structure and operation of human or other living brain nerve cells.

Compared with the original flat computing architecture, the neural network architecture is more similar to 3D properties, and can often provide a relatively simpler method when solving complex problems.

Therefore, the neural network architecture has received more and more attention from the computer industry in recent years.

HiSilicon and Huaxin chose this breakthrough, on the one hand, they wanted to overtake on the curve, and on the other hand, they were helpless.

There is no way. Compared with those Western countries and companies, China started too late in the basic research of CPU architecture and GPU architecture, and is far inferior to others' strong experience and strength.

Of course, there are also the most critical patents.

Relying on the moat of patents, Western companies such as Qualcomm, Apple, and Google have almost completely monopolized this field.

If you want to bypass it, you can only find another way.

This is also the reason why HiSilicon and Huaxin chose to start with the NPU neural network architecture.

Although this architecture is still developed from the ARM architecture studied by ARM, it can lay a solid foundation for future independent research and development.

Flipping through the information on the computer, Xu Chuan studied and analyzed it carefully.

Generally speaking, the neural network architecture mainly uses various mathematical models to describe and simulate the biological nervous system at different levels from different angles.

Representative network models include BP network, RBF network, Hopfield network, self-organizing feature mapping network, etc.

Compared with the traditional CPU architecture and GPU architecture, he started with this architecture faster.

After all, mathematical models and modeling, as well as the mathematical formulas behind them, are one of his research areas.

For example, the RBF neural network architecture (radial basis function network architecture) is a type of forward network constructed based on function approximation theory.

The learning of this type of network is equivalent to finding the best fitting plane for training data in multidimensional space, and each hidden layer neuron activation function of its radial basis function network constitutes a basis function of the fitting plane.

Using mathematical theory to explain, the principle is actually very simple.

That is, find a function g in a selected class of functions so that it is an approximate representation of a known function in a certain sense, and find the error caused by the approximate representation with g. "

This is one of the basic functions of high school mathematics.

For Xu Chuan, studying these things was initially just to solve the problems of HiSilicon and Huaxin, but as time went on, he became more and more interested in this neural network.

This structure, which starts from mathematical mechanisms and uses models to complete descriptions and simulations at different levels, has many applications in other fields besides chips.

For example, artificial intelligence, automatic control, robots, statistics, etc.

These are all areas that Xu Chuan is very interested in.

Needless to say, artificial intelligence is a trend of future development.

Automatic control and robots are one of the core key points of social productivity progress.

As for statistics, it seems that it is an applied mathematics that mainly works for other fields.

But in fact, it can also be applied in the field of mathematics.

Of course, he pays more attention to the statistics and analysis of data in the field of high-energy physics or materials science through neural networks.

The data volume of the latter two is quite large, requiring more sound and simple mathematical statistics and analysis methods.

And after studying these days, Xu Chuan's keen sense It seems that this neural network is more suitable for quantum chips and quantum computers than ordinary silicon-based chips.

If the traditional silicon-based chip computer is regarded as 2D or 2.5D, then the quantum computer is 3D, which is naturally beyond one level in terms of dimension.

From the biological point of view, the neural network computing method can actually be seen as a 3D three-dimensional structure.

Of course, this statement may be a little inaccurate.

But he does think that this may be more suitable for quantum computers.

It's just that there is no complete quantum computer that can be used to verify this idea for him.

It will not take just a few days to solve the problems of HiSilicon and Huaxin. While studying and researching, Xu Chuan also devoted part of his energy to other work.

The days passed one by one, and in the blink of an eye, the Lantern Festival arrived.

In the villa, Tang Sijia knocked on the door of the study.

After a while, there was a movement inside. Xu Chuan came out, took a curious look at the life assistant in a white coat, and asked, "What's the matter? "

It was the first time he saw this life assistant wearing a doctor's uniform, which looked somewhat similar to the white coats in the laboratory.

Tang Sijia carried two bags and said concisely: "Today is your quarterly physical examination date. I'll take samples and do a physical examination."

"Oh, okay, wait a moment, I'll save the information on the computer." Xu Chuan suddenly understood and nodded.

Since he was assigned a life assistant last year, physical examinations have become a regular thing.

At first, he would have a checkup every month, but later he felt that it was too troublesome to come every month, so he wanted to change it to once a year.

After all, ordinary people usually only have a physical examination once a year, but Tang Sijia argued with him, and finally decided to do a simple physical examination once a quarter and a detailed physical examination once a year.

Calculating the time, today is almost the first quarter.

After drawing two blood samples for physical examinations, Tang Sijia first saved the samples and then handed over a tool.

"This is for collecting urine. Do you need my help?"

Xu Chuan coughed lightly and took the tool: "I can do it myself."

How can others help with this kind of thing?

After handling the physical examination, Tang Sijia checked and recorded physical information such as blood pressure, weight, height, vision, hearing, etc. After packing up the tools, he handed another bag he brought to Xu Chuan.

Xu Chuan took it and opened it to take a look. It looked like a lunch box inside?

"What is this?"

Tang Sijia smiled and said, "Tangyuan, isn't today the Lantern Festival? I guessed you probably wouldn't remember, so I asked Aunt Lin to make some for reunion."

Xu Chuan took out the lunch box and opened it. Inside, there were really white and moist dumplings, steaming.

After closing the lid, he smiled and said, "Thank you."

Eating dumplings on the Lantern Festival is a custom in the south, but he hasn't eaten it for a long time.

After all, at this time of the Lantern Festival, he has basically left his hometown.

If he lives alone, he won't remember to order dumplings on this day.

With the dumplings, Xu Chuan turned and entered the study.

On the desk, the phone was ringing. It was the president of Nanjing University calling.

"Academician Xu, have a good Lantern Festival." The phone was connected, and Tan Shaoyuan's cheerful voice came from the other end.

Xu Chuan smiled and responded: "Let's have fun together, and I wish President Tan good health."

After exchanging a few pleasantries, Tan Shaoyuan smiled and asked: "I wonder when Academician Xu is going to return to Jinling."

Xu Chuan: "I'm already back in Jinling, is there anything?"

Tan Shaoyuan: "That's right, I called you this time mainly because of the unified framework theory of strongly correlated electron systems that you proved perfect at the end of last year. The school has received many inquiries from physicists at home and abroad, all asking when you will give a lecture to explain it."

"So I want to ask you when it's convenient, so that the school can prepare in advance."

Hearing this, Xu Chuan thought for a while and said: "Then let's do it in ten days. I'm busy with other things for the time being."

These days he has become familiar with a lot of things related to neural network architecture, and he plans to use these ten days to solve the problems of HiSilicon and Huaxin.

Although the time is short and it may not be possible to make the best solution, from the previous communication, they are very anxious about this matter.

In this case, let's solve the problem first. As for optimization, we can do it later when we have time.

This kind of thing can't make a big fat man in one bite.

Tan Shaoyuan: "No problem, just follow your time, the school can do it anytime. Then let's make it like this, the school will send out the notice and announcement first, and be prepared."

Xu Chuan nodded and smiled: "Okay, it's just a bother to the school."

Tan Shaoyuan: "Well, how can this be called a bother? You know, Shuimu Peking University is hard to find. Then I won't bother you any more today, Academician Xu, we'll talk again when we have time."

"Okay."

After hanging up the phone, Xu Chuan looked at the black screen of the phone and smiled and shook his head.

Speaking of the unified framework of strongly correlated electron systems, it has been almost a month since the beginning of the year. It is estimated that many people in the physics community can't wait. If they can't find him, they can only look for Nanjing University.

However, after so much time, it is time to give a response to the outside world.

PS: Second update for monthly tickets