If you cheat, money will follow

After listening to Academician Wen's introduction, Chang Le followed Liu Chaoyang to another building, the Computer Research Institute.

Along the way, I met many young and new faces, all of whom were Chinese.

Liu Chaoyang said: “In the past few years, many young people have joined the institute, some engaged in basic research and some conducting applied research.”

"There have been many small achievements in applied research in recent years, but basic research has not progressed quickly."

Chang Le said: "This is understandable. If basic research could spring up like bamboo shoots after a rain like applied research, humans would have flown out of the solar system long ago."

Liu Chaoyang sighed: "Yes, I will never see it again in this life."

Chang Le shook his head: "Don't be discouraged. Revolutionary results may suddenly break out one day and scare you."

Liu Chaoyang laughed: "I hope so."

"Boss, what we are meeting now is an interesting team. They are all young people in their 20s. Their ideas are very interesting and they can be regarded as keeping up with the times."

"Oh?"

The two walked into the Computer Research Institute and came to a large research room called the Deep Learning Research Laboratory.

In the research room, I saw eight young people and a monster computer equipped with dense graphics cards.

The style is similar to Tuba’s DIY products.

"Li Fei, all of you come here." Liu Chaoyang called the young man over and introduced him.

It turns out that these eight people are all master students in the Department of Computer Science and Technology of Tsinghua University and just graduated last year.

After they graduated, they took the initiative to apply for jobs at Pingtou Ge Research Institute.

Liu Chaoyang was very interested in their research project and signed them.

The leader of these eight people is called Li Fei, who is one word different from the great goddess from Stanford.

The latter hosted the 2012 Imagenet competition, which caused a sensation in the industry.

Li Fei saw his future career path from the Imagenet competition, and gathered seven other like-minded students to form a team to focus on deep learning.

Although the team only had eight people, it was five more than the Lake Tahoe Hinton team.

The latter's leather bag company was established in 2012 with no products and only a few public papers.

One month after its establishment, it was purchased by Google at a bidding price of US$4400 million, which opened the curtain of deep learning.

B degree also participated in the bidding, which stopped at 4000 million.

It’s not that you can’t bid if you have a B degree. Instead, Hinton ended the bidding and decided to join Google.

Li Fei's team's approach is very mainstream.

It is to further improve and improve on the basis of the convolutional neural network of the Lake Tahoe Hinton team, constantly feed image data to the algorithm, find errors and then optimize, and so on.

"Boss, Academician Liu, the current accuracy of the algorithm is not satisfactory, only 90%." Li Fei said.

The algorithm accuracy of the winning team in the 2012 Imagenet competition was 97.3%.

"Is the low accuracy related to computing power?" Chang Le asked.

"It's not all about computing power, the algorithm also has many flaws, for example..." Li Fei shook his head and tapped the keyboard a few times:

A big meat bag is displayed on the screen, but the answer given by the algorithm is not serious: breast.

The algorithm seems very confident and pulls out a series of pictures of breasts of different sizes, with fluctuations and sizes, to prove that the two are the same thing.

But the human brain can easily tell that the two are not the same thing.

Although there are similarities in some functions, they can both bite or lick.

"Where did these messy pictures come from?" Chang Le was curious.

"In the Goosechang database, they have opened up access to social apps to us, so we often see all kinds of weird photos." Li Fei said.

"I contacted Goose Factory. Goose Factory is very interested in our project and is ready to send a team to participate. In addition, they think that the computing power of Li Fei's team is too poor, and they are ready to invest in building CPUs and GPUs to increase computing power." Liu Chaoyang introduced .

"They thought it was simple. Stacking computing power cannot solve the problem. Google Cat stacked tens of thousands of CPUs, and the accuracy rate was only 75%, while the Alex team only used two GTX580 graphics cards, and the accuracy rate was more than 97%." Li Fei He shook his head and said:

"It's not time to pile up computing power yet. Of course, high computing power will help algorithm optimization."

Compared with CPU, GPU has stronger parallel computing capabilities. The leather-clad swordsman of "Silicon Valley Knife King", Lao Huang, can be proud.

However, deep learning is to simulate the human brain.

Through appropriate algorithms, continuously feed data to improve recognition accuracy.

For example, if you give a picture of "Knowing King" with half a head of hairstyle, a 1-year-old child may think it is poop.

Show it to adults, and you will know at a glance that you "understand the king".

It's the same as mistaking buns for breasts - seeing less.

"Yes. Algorithms are important, and computing power is also important. They can be done at the same time." Chang Le nodded and looked at Liu Chaoyang:

"Academician Liu, I remember there is a team called OPENAI. Have you contacted it?"

"I contacted them, but they were rejected. They are not very interested in crossing the ocean." Liu Chaoyang said regretfully.

Currently, the global AI resort is North America.

"Forget it if you refuse." Chang Le said to Li Fei: "Algorithm problems can be slowly optimized, and personnel and hardware facilities must be strengthened."

"Okay, boss." Li Fei nodded.

"Academician Liu, the budget of Li Fei's team can be further increased. The field of deep learning is a long march, and it is unknown who will make it to the end. Temporarily leading does not mean that we will lead in the end. It may be more advantageous to be behind temporarily, and there are many wrong routes that we do not need to follow. Trial and error." Chang Le said to Liu Chaoyang.

"Yes, I very much agree with this view." Academician Liu said.

After leaving the Computer Research Institute, I came to the Materials Research Institute.

"What is Shu Zhijie doing now?" Chang Le asked.

"The cost of solid-state batteries is still being reduced." Liu Chaoyang said.

"I know that he has some thoughts about the company not promoting this project." Chang Le chuckled.

Liu Chaoyang smiled and nodded in acquiescence.

"I can understand his mood. But the company's layout in the commercial field has more considerations." Chang Le explained:

"Although the domestic new energy industry is vigorous and in the ascendant, it has not really taken off in North America and Europe. Japan is even opening up another path independently and wants to go it alone. The global pattern has not yet been truly formed."

"If we forcefully promote solid-state batteries with performance beyond the times at this time, the likely result will be that we play stand-alone, and others will directly go offline and change tracks. After all, electric energy is not the only route, and they have more choices."

"That makes sense." Liu Chaoyang nodded.

"So, you go and tell him. The company will definitely push solid-state batteries at the appropriate time... The premise is that Europe and the United States will really make up their minds and invest huge costs and can't turn back."

"I think he must understand this truth. Now we can suggest that he expand in depth based on the existing research and development results, such as graphene batteries, solar cells, etc. If these technologies can be realized, the country will be invincible .”

"Boss, please rest assured. This guy Zhijie will definitely understand the company's layout."

"That's good. He is a true direct descendant of the institute, and I will definitely not treat him badly." Chang Le thought:

"However, we have to remind him. Don't follow the example of Japan and eat alone. Don't run out of patents. Give others hope."

"Ha ha……"