When he left the Mississauga campus, Lin Zhiling still felt a little emotional. Pen @ fun @ pavilion wWw. ｂｉｑUｇE。 ｉｎｆｏ

"Why can't these people bring out their best value in such a good academic atmosphere in Toronto?"

Gu Cheng didn't feel any emotion about this, obviously he had no heart for a long time: "Science is a very rigorous thing. If you have cancer, science has pronounced a death sentence on you, and now there is a drug that has a 10% probability of being confused after taking it, but the mechanism of effect is unknown, do you take it or not? Do you do or not do the business of making money with this drug?

If you want to win the Turing Award, you can't do this kind of business, and you can't do this kind of research and development. If it's just money, it's practical, it doesn't matter. I've never wanted to embrace science in my life, I've just used it at best. But if science is not good at witchcraft on a particular problem, I'll use witchcraft. ”

Stephen Cook was a great man, but he didn't get along with Gu Chengming's circle.

Therefore, Gu Cheng can only respect the other party's character, but the Tao is not the same, and then try to poach those scholars who have not yet been shaped and are not so full of "discipline" from his subordinates.

As long as he is willing to exchange his name for money, Gu Cheng has hope to dig it up.

He spent two days in Toronto to give Jeff Hinton more time to do his people's ideological work.

Everything was going well, and things were completely progressing in the direction Gu Cheng expected. In the face of huge research funds and salaries, and even possible basic dividends in the future, and the opportunity to become famous in other media channels, I finally overcame the desire to win academic awards and impact factors.

In addition to Jeff Hinton, a school of research on "deep learning algorithms", several other professors under Stephen Cook have also loosened slightly, and Gu Cheng has poached at least two associate professors and a bunch of doctoral students.

What Gu Cheng wants is this effect. "If you follow Gu Chengming, even if you can't get on the core journal, you can still be famous in other serious media channels and be remembered by the world in another way."

The first step is always difficult, and once this idea takes root in the minds of the nimble basic scientific research talents, Gu Cheng's snowball will get bigger and bigger.

In order to accomplish all this, Gu Cheng also initially wrote a check for a full 30 million US dollars to inject funds into his newly registered shell research institute and foundation.

This money is almost equivalent to the gross profit of the "legend" for a quarter, and at present, this institution can only be set up in the United States - Gu Cheng tentatively asked Jeff Hinton, "If you let everyone go to work in China and increase the money, how many people are willing to leave", but the result is that the number of people willing to follow him will be reduced by at least two-thirds.

The impression that China has on white people today is still too opaque, and if it were another five or ten years, the situation would definitely be reversed.

Therefore, Gu Cheng decided to set up a research institute in the United States first, as well as a subsidiary of YY Network Technology, and wait for him to become bigger in two or three years, and then move the core R&D personnel to Huaxia in the long run.

……

"Deep learning algorithm" is the originator of artificial intelligence in the future, although it is not the only way to achieve it, but it has opened up a way for human beings to "let the machine slowly correct itself according to the data identification".

HISTORICALLY, THE INITIAL APPLICATION SCENARIOS OF THIS TECHNOLOGY WERE GOOGLE'S IMAGE SEARCH ENGINE AND FACEBOOK'S FACIAL RECOGNITION TECHNOLOGY.

Before that, Google search could only search for text information, but not for pictures.

Don't look at Baidu and Google, they have opened up functions such as "Baidu Image" very early, but the original "Baidu Image" did not determine the search results based on the content of the image, but relied on the text title of the web link to which the image belonged.

So 09 years ago, the "Baidu picture" function was not really "read" the content of the picture. It's just that laymen only look at the efficacy, so they didn't notice any abnormalities in this fairly smooth technical transition, and thought that the later "Baidu Pictures" picture recognition rate "naturally" improved.

Only after the new generation of artificial intelligence represented by deep learning is truly applied, humans have learned to let machines directly read the map itself.

However, although this application scenario is very ambitious, it has nothing to do with Gu Cheng's business. That's something that needs to be worried about by Baidu Boss Li and Ahri Mafeng. Boss Li may be able to do a picture search in the future, and Ma Feng can do Taobao to find the same model. Gu Cheng will at most mention it at the right time in exchange for his own benefits, but he will not end up in person.

THE SECOND APPLICATION SCENARIO OF DEEP LEARNING ARTIFICIAL INTELLIGENCE, THAT IS, FACEBOOK'S FACE RECOGNITION IN PARALLEL TIME AND SPACE, IS VERY COMPATIBLE WITH GU CHENG'S CURRENT BUSINESS.

Today, Zuckerberg is not an ambitious guy, he only wants to make a name for himself among Harvard girls, be admired by countless people, and blend into the top big clubs. So nine times out of ten, he will be persuaded by Gu Cheng to his command. GU CHENG DOESN'T PLAN TO START ANOTHER FACEBOOK, BUT PLANS TO DIRECTLY OPERATE "YY NETWORK" OVERSEAS.

Considering the factors of the wall, as well as the difference in content inside and outside the wall, when the time comes, the domestic part will turn the English "YY network" upside down and change it to "Renren network".

The name doesn't matter, anyway, both will end up being YY-based circle of friends space products.

On the domestic Tengyun side, Ma Teng is now doing QQ space, and Tengyun's capital chain is relatively tight. After QQ Space went astray, it was too late for Gu Cheng to announce his development plan and teach him.

According to this plan, Gu Cheng estimates that after he returns to China, he will comprehensively promote the development plan of "YY.com" and "Renren.com", and launch the website around the Golden Week in the fourth quarter and November, which will basically be stuck at several key time points.

IN ADDITION TO WHAT FACEBOOK AND GOOGLE HAVE DONE IN HISTORY, "DEEP LEARNING" NATURALLY HAS ITS OWN UNIQUE USE IN GU CHENG'S HANDS, THAT IS, "USER PREFERENCE ANALYSIS".

This started much later than the first two applications in parallel time and space, but Gu Cheng knew that it was not that this matter was technically much more difficult than the first two, but because the giants who first came into contact with deep learning artificial intelligence in parallel time and space were not involved in the entertainment/content industry.

In other words, if the first batch of people to come into contact with deep learning artificial intelligence are replaced by Amazon, "user preference analysis and push" will definitely become the first priority.

Gu Cheng's business has a very high degree of overlap with Amazon, and he is a person who knows what it is, so of course he will not let go of the layout in this field.

It's just that this piece of work is relatively large, on the one hand, it is necessary to stack algorithms, and on the other hand, it is also necessary to label and classify and identify a large number of entertainment works on the market, and refine the subdivision data representation step by step for many years in the future.

According to the most optimistic estimates, "user preference analysis and push" will have to lie in the laboratory for at least two or three years before the trial operation can be discussed.

Fortunately, Gu Cheng has a lot of money and can afford this kind of long-term investment.

……

After three or four days in Toronto and digging up enough people, Gu Cheng was ready to drive back to Boston to settle the bet with Zuckerberg.

However, counting the days, the one-week appointment with Zuckerberg had not expired, so Gu Cheng had to go to New York for a vacation of two or three days.

With his busy status, even in New York, he will not be very idle, at least he has to call to control the business.

No, when he was still in Toronto, he sent Liu Qian, the company's prospective CFO, from New York to San Francisco, and asked her to invest in an office plot in Silicon Valley, where she registered a YY subsidiary, and at the same time kept an eye out for a start-up that was less than two years old, and tried to acquire it.

The company that Gu Cheng was eyeing was AGEIA, which later came up with the world's three major physical computing engines, PhysX, in 05. Historically, the company was acquired by NVIDIA in '08, which later led to Nvidia's complete GPU career.

If Gu Cheng wants to engage in deep learning artificial intelligence and convolutional neural networks, it is very necessary to dig up such a company.

As Gu Cheng and Professors Stephen Cook talked about a few days ago, the biggest difference between any "neural network" and a traditional computer network is that "there is no center, and each neuron node is completely equal and completely distributed in the cloud".

Therefore, when performing the corresponding operations of "neural networks", the CPU efficiency of traditional human computers is actually not very high, whether it is Intel or AMD.

BECAUSE PEOPLE WHO KNOW A LITTLE BIT OF COMPUTER COMMON SENSE KNOW THAT CPU IS A "TIME AND TIME OCCUPATION" COMPUTING HARDWARE, IN LAYMAN'S TERMS, A 4G CPU WITH A MAIN FREQUENCY, ONLY ONE SECOND CAN BE CALCULATED 4 BILLION TIMES, BUT EACH MOMENT CAN STILL ONLY BE CALCULATED ONCE, WINDOWS'S "MULTITASKING SYSTEM", IN ESSENCE, IS JUST "THE TIME OCCUPATION OF THE CPU IS SUBDIVIDED, AND EACH BACKGROUND PROGRAM OCCUPIES SO MANY SUBTLETIES IN EVERY SECOND".

For example, when a person plays a CS game, a QQ is opened in the background. Running QQ requires CPU processing resources of "100 million operations per second", so the essence is that "the CPU with a 4G main frequency processes QQ for 25 milliseconds per second", rather than "processing CS and QQ at the same time" in the physical sense.

This mode of single-core operation is destined to be unsuitable for more and more "parallel operations" of convolutional neural networks in the future. So when the dawn of convolutional neural networks appeared in 06, Intel companies in parallel time and space were not unaware of this and struggled. It's just that Intel's initial struggle was to "develop multi-core CPUs".

This is what became known as "Intel Core Dual-Core/Quad-Core".

Unfortunately, history has finally proved that no matter how many cores the CPU has, it is a drop in the bucket to meet the vast parallel operation.

To fully satisfy the craving appetite of neural networks, it is necessary to rely on the GPU that was originally used for physical computing on graphics cards.

This is why after the outbreak of artificial intelligence in the software field, the market value of hardware computing companies fluctuated violently. Nvidia, which is a graphics card GPU, has jumped more than ten times in two years, and it seems to be competing with Intel.

To explain the difference in a phrase that even liberal arts students can understand: why all graphics cards don't have the concept of "dual-core/quad-core"? When a graphics card processes a computer image, each pixel is processed separately at the same time. There is no bottleneck of "time and time usage", resulting in GPUs obviously not requiring multiple cores like CPUs.

(Note: Some of the graphics cards in the GTX-Titan series are called "dual-core", which is actually a misadvertisement by merchants, and the essence of those graphics cards is "two graphics cards" rather than "dual-core".) ）

Gu Cheng's plan is that when Intel is still ready to struggle with multi-core, he will directly see through the big hole in one step and jump directly to the road of GPU domination.

Today's AGEIA company has been established for less than two years, and there is no marketization of scientific and technological achievements. Liu Qian set off with Gu Cheng's checkbook, and there was no disadvantage.