In December 2013, Meng Fanqi will go to Australia, XN, to participate in the top conference on computer vision, ICCV-2013, International Conference on Computer Vision.

He has not submitted any articles and published them in this conference, but everyone knows that this person who has not submitted any articles is the absolute core of this conference.

At the same time last year, one of the three AI giants, Hinton from the University of Toronto in Canada, released Alex with his students. It extracts knowledge that is difficult for humans to understand and explain from tens of millions of pictures.

As the only team that uses neural networks as the core algorithm, they defeated traditional algorithms in a crushing manner and won the championship in a complex visual task such as Thousand Classification.

In the new era of AI, deep learning begins from this moment. Giant technology companies led by Google, Microsoft, Baidu and other companies are swarming in, making the originally relatively pure AI field more pragmatic in the industrial world. Of course, there are also

The stench of some capital.

In 2013, the same event was held routinely. While universities and institutions were still scrambling to replicate the results of last year's Hinton team, a name that no one had ever heard of, Meng Fanqi, used Dream to use an even more exaggerated name.

The performance improvement stands proudly on the list of all tasks in the competition.

And the top-5 error rate of 4.8% has exceeded the human standard of 5.1% provided by the competition four years ago.

This terrifying improvement is completely beyond everyone’s imagination and understanding.

It has been nearly a month since the results were announced, and the entire academic community still has no idea how this result was achieved.

The ICCV conference party that co-organizes the event has already sent invitation emails to the top contestants.

Fortunately, this mysterious "Meng" quickly confirmed his attendance. Otherwise, no one really knows how to contact him.

Not only is it difficult for people who are not familiar with this field to imagine, but even the Hinton team, which has been deeply involved in this field for nearly half a century and topped the list with the power of three people last year, has no idea how such a terrifying performance improvement is achieved.

.

Someone has made such a terrible breakthrough in the field that he is best at, but he has no ideas or clues at all. This is something that is not easy for this gray-haired old man to accept.

In addition, the biggest pressure comes from his employer, Google.

Hinton, who was 65 years old at the time, directly improved the competition's Top-5 accuracy rate for thousands of categories from about 75 to 85 for a single model last year. This was a huge shock to Google at the time.

Although it did not officially participate in the competition, Google, which has always been very confident in processing massive data, did some verification on the competition data in private. It turned out that it had rich experience in big data processing and massive computing resources, and was defeated by Hinton.

A small academic team like the trio made a big splash.

Technology is the core productivity.

Keenly, Google quickly realized the potential of deep neural networks, but at the same time, there were other well-known companies and institutions. The phone number and email address of Hinton's team were quickly crowded.

Google sent engineering director Alan and even Jeff Dean, who is well-known in Silicon Valley, to visit Canada in person to ask the Hinton team to join Google.

Jeff Dean joined Google very early and was also one of the three early giants of Google. Because he was actually a great guy at Google, there were many jokes about him that were circulated in the public, and they were called brother-in-law jokes.

Most of them are like, "The speed of light used to be only 120km/h, and then Jeff spent a weekend making a small optimization of physics."

Or, "Jeff only uses buttons 0 and 1 when programming"

"The compiler never warns Jeff, it's Jeff who warns the compiler"

He founded Google Brain in 2011, vowing to establish Google's world-leading position in AI. And in 2012, it was able to recognize some content in its own Youtube videos, and greatly enhanced the accuracy of speech recognition.
p>

Jeff firmly believes that no one knows the value behind this better than him, and he will give Hinton a price he cannot refuse."

In fact, this is indeed the case. When the bidding price exceeded 20 million US dollars, the Hintons began to lose the concept of numbers. This feeling was very unreal, nihilistic and ethereal.

After the mutual bidding between Google and Baidu reached 44 million US dollars, Hinton stopped the bidding process and chose to join Google.

In fact, the amount Jeff Dean is willing to pay is far more than that.

A year later, Jeff Dean and Hinton once again stood at the negotiating table, but this time, they belonged to the same camp.

Their thoughts were the same, "Google must get him, and we must give him a price that cannot be refused."

But when they actually entered the conference room of XN, the two people looked at each other, and their thoughts were wavered at the same time.

The young man from China in front of me seems to be too young. In this field where at least master's and doctoral students are involved, a junior in his twenties does not seem so trustworthy.

"You are a very cheerful person and a very smart person. Let's not beat around the bush." ​​Jeff said straight to the point, "Most of the time, I believe that youth means immense potential. But judging from your experience and education

Looking at the background, I also have to consider for the company that your achievements this time are just a matter of luck and chance."

"This is an astronomical contract. We are just talking about some possibilities. I hope you don't have other ideas." Hinton added.

Meng Fanqi had already anticipated this. Before the rise of deep learning and deep neural network performance, Hinton had already become the originator in this field. Google dared to directly invest 40 million US dollars out of basic trust.

I have no historical achievements, I am so young, and my major is not even related, so it is reasonable for Google to worry about this.

"How about this, we can sign a gambling contract." When trust is not enough, a gambling contract is the best solution. You can punish situations that you don't want to happen to avoid losses, and you can punish situations that you want to happen. Transfer more benefits.

"We can sign a price based on the quality and quantity of the research results, or share them based on the adoption of these results or even the additional revenue brought in the future. If my method is not adopted by Google, or if it does not bring benefits after adoption, I will even You don’t need to take a penny.”

Jeff and Hinton looked at each other in confusion. If nothing else, the young man in front of them was second to none in terms of his confidence and courage.

They are willing to give up the astronomical three-year contract of US$1,000,000 proposed by the two of them, and use a low base salary to gamble on rewards and shares based on results.

Is he so sure of his own wisdom? Can he always achieve such amazing results?

The two giants signed the contract with the young man in front of them with doubts.

Later, when interviewed by the media, Jeff often lamented that after signing such a contract, he was so numb that my heart really hurt.

"But Meng laid the core and basic algorithms in so many fields during his time at Google. Aren't you happy?"
Chapter completed!