Wang Haoquan: It is not a scientist who changed the world. It is a scientific business family 丨 scientist entrepreneurship series 06

Monkeys have been difficult to evolve, and people become entrepreneurs. Scientists become uncomfortable.

2012 "Entrepreneur Magazine" has made a special planning, discussing "When scientists become entrepreneurs." This role will encounter a wide variety of problems. For example, when scientists / entrepreneurs encounter the system, encounter the market and encounter capital. We even summed up the "seven sins" of scientists. The reporter of "Entrepreneur Magazine" found that many scientists are "pseudo-entrepreneurship", many entrepreneurs "personality division", but in front of entrepreneurship, equal equality, scientists especially self-revolution.

In 2017, "Dr. Shui Dynasty" is sprinkled and wrote an 87,000 word "technology stagnation." Only for discussion: human science and technology levels come to the bottleneck, will face the stagnation of technology development in the future. He smashed the technology explosion and exploded by fireworks: the moment of fireworks explosion was beautiful and beautiful, but when people saw it, the explosion was close to the end.

I didn’t pay attention to Wang Yuquan to see his views on technology stagnation. But from interviews you can feel that he is a technology musician. Shortly after the interview, Wang Hao has conducted a lecture called "Technical and Cultural Symbolic Evolution".

This article is the sixth article, and the trend is related to the number of scientists. In response to technological innovation, technology companies, Wang Yizhen shared his observation and thinking from the two perspectives of supply.

From the perspective of demand, Wang Yizhen believes that the technology revolution roll forward. Behind every scientific revolution, breakthrough new technologies will lead the whole society to form a new growth paradigm and production paradigm. The steam technology revolution, the core of the power technology revolution is the problem of product scale. The third scientific and technological revolution is solved by the problem of service scale. At present, the global industry is in the early stage of service scale.

From the perspective of supply, technology has become more complicated. The realization of science and technology innovation is inseparable from the prosperity of industrial ecology. Industrial prosperity is the core industrial links have corresponding leading companies. Collaboration between them constitutes a prosperous industry. This or one of my country encourages the new reasons for developing specialty. "I have to buy" to realize a domestic substitute for a part, and it is difficult to establish a sound industrial ecology.

After talking about "service size", he even mentioned "creative scale". "Creative scale" can be achieved after "service size". In the context of Wang Yuquan, "service scale" corresponds to the information revolution. "Creative scale" belongs to "Creating Revolution", and mankind or will be able to get rid of evolution.

"Technology is changing the world of the times, everyone must bend into the office." Wang Yuquan said.

01 side of the coin: service scale

Before "serving sizes" is the "scale commercialization." "Scale product" is the trend of the two industrial revolutions appeared before the PC.

"Scale commercialization," can be traced back spinning jenny, rather than the Watt steam engine. Combined with the spinning jenny and water, rather than steam, allowing large-scale production of textiles is possible. Europe was the hydraulic power plant industrialized areas commonly used.

The story is this steam engine. Lift the steam engine, Watt had to bring. Watt improved the lives of more than 20 steam engine. In fact, Watts Industrial Revolution is not the most typical figure. Because of that era in the life of Watt, the steam engine has not been large-scale applications. Watt and before, there are a few people’s names should be remembered: Galileo’s student Torricelli, Boyle scientist, scientists Papin, the British army engineer Savery, Newcomen hardware merchant and Corey. Torricelli, Boyle pressure were found (1643), over the pressure of the gas is inversely proportional to the volume (Boyle 1662). Papin who served as Boyle’s assistant, he used the liquid boiling point with elevated pressure and elevated principle, in 1679 invented the world’s first pressure cooker, and published works of the steam engine in 1690.

Savery put pressure pumping principle with mine together, and learn the principles of Papin’s invention of the steam engine called "Friends of the miners of" pump. Hardware businessman Newcomen and Savery Corey is the fellow who helped make steam Savery pump model. They create the world’s first all-weather continuous operation, running costs are relatively low steam drainage machine – Newcomen engine, and patented in 1712.

1763 winter, the University of Glasgow purchased a Newcomen steam engine model for physics practice. Experimenter James Watt is responsible for maintaining it. The story behind it is written in the history books. Watt developed a prototype in 1768, patented the following year. After Watt improved the steam engine, brought a series of technological revolution, to promote the production of manual labor leap to power the machine.

The core of the industrial revolution in automation. Automated production of huge significance. During the interview, Wang Yuquan special emphasis on the fact that an easily overlooked:Those who built the first mill in the UK to become the richest person in Britain."Before the richest man in all kings, generals, are playing down. But the richest man is earned out. This is the first time the world has changed the rules of the game and you can become the richest man by money." "If you can not murder and arson will be able to make a profit two percent or three hundred, why do you want to murder and arson? "Wang Yuquan said.

After the discovery of this new way to get rich, we embarked on the entrepreneurial path. This road and the road is not a zero sum game. Entrepreneurs lead the way is not looting, but as much as possible large-scale production, will provide products to as many people to use.Since the industrial revolution, in fact, another identity is the world’s richest technology entrepreneurs.John D. Rockefeller bought out of oil refining technology, by investing in the construction of large-scale technology into practice. Now the world’s richest elon Musk (Elon Musk) also technology entrepreneurs. Elon Musk has optimized thinking typical of engineers. Musk has said: There is no rocket scientist, rocket engineer only. Because scientists made no rocket, engineers can build rockets. Wang Yuquan was approved this sentence. He also believes that promoting social progress are engineers, entrepreneurs, rather than scientists. "Section and technology are two different things children, Chinese scientists over-emphasized, because the technology is put together,‘Branch’ light stick ‘technology’, the real push society is ‘technology’.Because there are a lot of ‘technology’ is no ‘families’, but no ‘families’, technology is possible progress, but there is no ‘technical’, ‘branch’ on what did not. "

Large-scale automated production make the product possible. Although the product can scale to achieve the scale of production, but there is no way to meet individual needs. Before PC, then technology entrepreneurs is how to solve?The service to the product solidify in.Such as how to meet the needs of users iced watermelon? Sold to users is achieved by a refrigerator. The user’s own holding melon in the refrigerator, you can get iced watermelon. That user through the products and services it provides to its own. After the PC, the digital revolution began, and then after the Internet, mobile Internet era to today’s AI. Wang Yuquan judgment, has now entered the digital revolution to the end. Marked by artificial intelligence so that "the scale of service" as possible.

Specifically, the "service scale" on the premise that meet the individual needs of users, large-scale production of artificial intelligence can be replicated personalized service. "Service scale" What are the characteristics? Still iced watermelon, for example, users no longer want to refrigerators (product) but iced watermelon (service). The user is not concerned with the product (refrigerator) but to serve (iced watermelon). The needs of users (iced watermelon) are directly met. If the "service scale" to see Microsoft, Apple, you can see they are not the same. Microsoft to embrace subscription system, the original can be a one-time buyout of Office productivity software products become Office 365. Office 365 is equivalent to one kind of subscription service, or need to pay a monthly subscription fee per year. Apple is also true. Apple has a large number of mobile phone subscriptions APP is made. In early November this year, well-known application notes Notability sudden change from a subscription-based buyout system, and being on the hot search. In addition, Apple phone AppleCare + (US version), music, news, videos, etc. are subscription-based. The future, mobile phones will be priced less than the total value of the various services the user subscribed.

The service scale, the subscription system is being promoted, and the trend is visible. In the future, even the car can also use a subscription system. Essentially, users buy cars and buy transportation. One-time bought-in delivery cannot be personalized. From the perspective of the service, this approach has a shortcoming. If there are 3 vehicles in 10,000 vehicles, it is bad, it happens to be bought by Zhang San. So, what is the consequence of the car? This is unfair. After changing into the subscription, Zhang San pays for money, encounters the bad car, and the car enterprise can be replaced in time. In this case, Zhang San’s car will never be broken. Under the subscription, the vehicle enterprise needs to provide long-term services to our customers, while IOT equipment such as sensors will continue to monitor the vehicle to ensure uninterrupted providing of quality services.

"The rules of the whole time have changed." Wang Wei commented on the transformation from the size of the product to the scale of service. In the interview, he passed"The synergy of science and technology and culture makes people today"To further interpret the logic behind this shift. he thinks,Society is choosing technology, technology is also choosing society.What technologies need for society, it will make it vital. Conversely, technology is also choosing people. Because the society is not supported, it will not develop. For example, the Industrial Revolution makes the Britain strong, so that it exceeds China that is still in feudal society. The power of the United Kingdom is the result of the two-way selection. This is the same as rice and human society. This two-way selection is now continuing and continues in acceleration. The requirements and techniques of service scale and techniques are mutually selection, mutual promotion, and the era of human society into service scale.

Now, the service is in the early stage. But not all industries are suitable for service size. In the interview, Wang Yuquan talked about two industries. The catering industry currently has a service-scale seedlings. The training industry is difficult to achieve.

Catering, especially high-end catering, is difficult to make universal, and the Ford is so big. What is the reason? Unmatched. The higher the higher, the more difficult it is. Because the chef is top, the top is not significant. The original millet ecological chain vice president Tang Mu created a company called a smart. Such as shadow intelligence has created a kitchen robot. It is a fried steak’s arms robot. Wang Yuquan said that the service is to pass is an experience. The core of service is the copy of excellent experience. There is a classic misunderstanding for service size, "personalization is equal to good service", actually personalization is not equal to good service.

The robot is fried, and the product is solidified to the service to solve the scale, but the new problem comes, and how to let the dish for chain catering brands. Three years without changing dishes means that customers lose their freshness, but the cost of replacing menus is high. The dish will change the menu, change the chef, etc. But a chain restaurant using the cooking robot can be implemented by updating the software. While the ingredient is completed, the new dishes are instantaneously completed in all its restaurants. This means that the catering industry has the possibility of giant and Macon.

"The production logic of the gigabaters has changed. Previously, the logic of the giant Macro business was in a sufficiently large field; now, the logic of the giant Macro business is a big enough The field can provide a good high-end service. "Wang Yuquan said.

(Such as the smart kitchen robot)

The training industry is not suitable. IT and consulting combinations are not easy, because IT companies and consulting companies are different: consulting companies need a lot of manpower, customize different projects for different enterprises, unable to achieve rapid expansion; and IT companies or said a software company has Scale benefits and marginal benefits have the potential for rapid expansion.

Wang Yuquan team shared the most mysterious big data company of Silicon Valley. Even now, there are two main products in Palantir: Gotham (a large number of customized) facing business customers, and Foundry for business customers (more standardized, modular). At present, Palantir is still exploring the path of scale growth, stepping on the stairs between customization and scale. Wang Yuquan believes that artificial intelligence undoubtedly broaden the area of ??service scale and application range, but not all industries have a chance, if there is no chance in one field, you need to avoid it.

For individual companies, US groups calculate half of service-scale companies. Wang Yuquan believes that the current US group has achieved a certain degree of scale, but the weakness is delivering goods. Unmanned distribution is the direction of the current US group being explored. According to incomplete statistics, as of June this year, the US group drone has completed more than 220,000 flight tests, and the long-term length of 600,000 minutes, and the relevant orders for real users have completed 2500 orders. Wang Xing’s public speech in 2018 mentioned: In 2019, it will fully operate unmanned distribution.

2020 US group investment indoor distribution robot enterprise Purdue Technology. In addition to the partners of Purdu Technology, US Mission unmanned distribution platforms, there are autox, SEG Wei, Experimental technology, etc. The figure below shows the list of partners disclosed by the US group unmanned distribution platform.

(US group unmanned distribution platform disclosure partner)

In response to the exploration of unmanned distribution of the US group, Wang Haoquan believes that people have their own use in the age of service. Technical alternatives have two alternatives: one is an occupational alternative, an alternative to machinery, is also the most common practice. The other is occupational enhancement. Simple point, professional enhancement is that 1 person plus 5 machines can do 5 people. For example, Tesla electric truck. Tesla’s logic is not alternative, but the truck can be high-efficient and a driver can open three cars. Under the human machine, the complicated road conditions of automatic driving cannot be handled by people.

Drip is a "hard brother" of the US group."The tactics have become, the strategy is defeated"It is a review of Wang Yizhen to the drip. The new T3 travel made Wang Yizhen very emotion: "When did you see a industry, a rivers and lakes, and some people challenge you." This is becauseWhen automatic driving + electric vehicles come out, the travel industry has emerged.Only in the context of continuous curve, there is an endless newcomer challenge because the next corner has a chance to overtake. In the age of service, the burden of the drip lies in the ownership of the car belongs to the individual owner. Diards don’t have the right when accurate statistics and precise dispatch vehicles. But change has occurred. Last month, the US car rental company Hertz has decided to initially order 100,000 Tesla electric cars.

(Cloud Story Technology Hotel service robot "Xiaituz")

Yunu Technology is "excellent students" of Wang Yizheng recognized by Wang Yizheng. It is also an investment project of sea silver capital. In Wang Yuxi’s mouth, "Xiaolitz" is a cute word. In the hotel, especially late night delivery, "words" it is ignorant, not quietly, which can make guests feel safe. The robot of cloud lavage technology can independently complete the delivery of the delivery from the directive to complete the delivery. It can alleviate the "separation" of the front desk service personnel to a certain extent.

As a investor of cloud lavage technology, Wang Yuxi has enhanced valuation to the cloud of the cloud, and the venture capital negotiations: First, you admit that it does not admit that the future is a service robot? There is always a service robot to go to the community and the residents. Second, in addition to the cloud robot, you can’t find the second productive robot. The founder of Yunu Technology has joked: She is the largest labor outside the head. Because more than 10,000 machines are outsourced into the hotel.

However, Wang Yuquan, but I think she is the most "black heart" capitalist. "There is no salary to the robot. I am all taken."

The other side of 02 coins:Future layout technology industry ecology

The other side of the coin is the scientific industry ecology, on the side of the supply, echoes the changes in the demand side mentioned earlier. One of the changes is that the service scale makes some artificial opportunities for corner overtaking.

The chip industry has this opportunity. In 2018, the Science and Technology Daily has launched the "core technology that needs to be overcome". This column has been published in 35 reports. The first report reported behind the lagging status of the top lightning machine required for high-end chip manufacturing. my country is dispatched at the integrated circuit supply and demand gap. A large number of chips rely on imports. According to the statistics of the General Administration of Customs, in 2020, my country’s integrated circuit imports were 35.5 million US dollars, and exports were $ 116.6 billion. Its difference increased by $ 29.55 billion than $ 203.9 billion in 2019. If the exchange rate of $ 6.3787 yuan is converted to $ 188.43 billion, it is more than 2020 GDP 127.66 billion yuan than Tongliao.

China Chip businesses are in medium and low-end sectors, logic, storage and other high-end chips depend on imports in the global integrated circuit industry. High-end chip manufacturing mainly relying on the manufacturing equipment of the photolithography, the current cutting-edge momentary technology is monopolized by the Netherlands ASML.

The reason why the radiator in the chip field is over-vehicle, because with the rise of the new generation of artificial intelligence, the chip industry has a continuous curve. Just as mentioned in the example of the drip, only the chance of overtaking in a continuous curve. At present, the pattern of general computing chips represented by the CPU has been set, and China is more difficult to make a difference under the established pattern, but there is a chance to make breakthroughs in the special calculation chip field. A group of AI chip companies represented by Cambrian were currently rising. In fact, the opportunity of corner overtaking also appears in the fields of electric vehicles, power batteries. The photovoltaic industry is also a comprehensive leading industry through industrial ecological layout.

"From the perspective of development, when there is a major new technological change in the subverting industry, there is a chance to break the original technical monopoly, and there is also a chance to restruct an industrial ecology. This is precisely the later people with excellent anti-super chance."

(Professor Chen Yunyu, Professor Chen Yun, first released AI chip architecture in the world.

Wang Yuquan mentioned that the two years of China and the United States have also let the industry recognize the core of scientific and technological innovation is the core of the big country, the pursuit of a single indicator is leading, but the industry is not just a single business, but the entire industrial ecology. The strength comparison. As mentioned earlier is the only company in the world that produces high-end lightning machines, and is also a company that can support the company’s US semiconductor industry ecology based on SEMATECH (US Semiconductor Manufacturing Technology Strategic Alliance). SEMATECH is a non-profitred R & D institution in 1987 that the US government has established a non-profit of the 14 largest electronic enterprises in native. These companies are taking into a collaboration together with the best research and development power, and joint attack on the core subjects in the electronics industry. $ 200 million in research funds per year, in which the US government and these 14 companies have taken half. The establishment of Sematech has promoted the industry’s internal mutual research and development and achievement sharing, and gradually established a complete industrial ecology, and the Japanese semiconductor industry has solved the "card neck" problem.

Then, how to build a scientific and technological industry ecology in the future of service scale

There are two main body of the industrial ecology, one is the enterprise, and the other is the government.

At the logic level of the enterprise, enterprises should abandon the traditional practice: each battle, trying to create a complete industrial chain inside the company, and fight the competitors and occupy the market share as much as possible. Wang Haoquan believes that industrial ecology is collaborative, it is common prosperity. If a tree covers the sun, it will be trimmed. Now, whether it is a big company or a "special new" company, we should consider what value can provide for the industrial ecology, what contributions can be made for society, but also guarantees good users’ privacy and interests.

At the government level, Wang Haoquan believes that the government should do, it is to cultivate a tropical rainforest, rather than transplanting a big tree.

So how should the Government create an industrial ecology?

The first is to find local features, to create industrial ecology (Qingdao longevity & Technology) has its own advantages.

In 2020, Wang Yuquan team and Qingdao cooperation from the image of the local infirmary of the Holy Land, around the rich Wellness infrastructure, a pleasant climate advantage, for "longevity" and "healthy" demand, to design a "longevity technology" eco-industrial development direction . By combing the key aspects of industrial clusters and industrial collaboration, the team found Wang Yuquan time-consuming and inconvenient problem of traditional health screening, judged by that this is a short board "longevity" industry. On the other hand large medical data companies lack a lot of data can be trained, so they introduce artificial intelligence medical companies Airdoc to Qingdao, get through the demand across the industry. At present, the first batch of 150,000 people of Qingdao have been quicker to enjoy health screening services. This is the docking requirements, new ways to complement each other, and a change in the past to use the money to "buy" investment approach business.

(Qingdao, images from the network)

From the Qingdao example, Wang Yuquan summarizes the key points of the layout of industrial ecology two places in the city: the first, for the future; second, not greedy.

Facing the future, it comes in front of "continuous curve", "turn to overtake," industrial cities should be future-oriented industrial layout. Wang Yuquan think cultivate ecology, should start from scratch, start from early intervention and find their own ecological position, do not think picking the fruit, not in order after ecological and other mature, cut directly into the ecology taking over someone else’s. This road has been demonstrated in the chip industry is not going to work.

"It is a technology any ecological product. Any of the N links have an ecological composition, this corresponds plate consisting of N ‘barrel’ Eco early, the plates are short, only the width of the other. With ecological maturity, these plates grow up together. complete means that a number of enterprises from ecological shortcomings become long board. "he reminded not to ignore the ecology of narrow boards, such as the chip industry this board though narrow lithography machine but it will be deprived of the chip’s "bucket" on the leak. "In the past we always stare at other people the widest plate (Qualcomm), invisible narrow board (lithography). The problem is that even if you put the widest board clutched in his hand, still did not control this ecosystem, because the narrow strip in the hands of people. people are not high-pass, but kill you with ASML. "

Greedy, Wang Yuquan mouth is in place plans for industrial common mistake. For example, after the AI ??fire, many cities have rushed to engage in AI. "We believe that artificial intelligence will not form a sufficiently large industry ecosystem in China more than five. If the domestic front row less than 5, do not play. The top five is a high probability the north of Guangzhou-Shenzhen Hang." Wang Yuquan said suitable What, and local resource endowments and capabilities closely related genes.

Second, setting up enterprise communication platform (Chongqing & electronics industry).

In 2008, the electronics industry in the Midwest is still a blank. From the perspective of industrial Chongqing foster integration starting around parts, raw materials localization, to build a collaborative platform between enterprises. Chongqing and HP docking first production needs, and with Foxconn confinements, final orders to Foxconn for HP’s 20 million units per year, the promotion of their joint investment to build production lines in August 2009. Then seven computer manufacturers and more than 1,000 settled in the surrounding parts factory support, forming a collection of machine parts production and integration, processing, research and development and settlement of the whole industry chain integration. Through research and discussion, Chongqing learned from the enterprise logistics and transport constraints on industrial development, the railway $ 100 per container / km price negotiations fell through the year 2015 to $ 0.5 / km and meet a few local million units of products to North America, Asia and Europe distribution needs.

Wang Yuquan said, Chongqing Municipality on local entrepreneurs to build communication platform efforts, regularly organized industry chain entrepreneur forum, docking and other industrial cooperation activities downstream, the use of these communication platform in time to help each depth docking enterprise needs, understand each other technical strength, shapeAs "the ability to circle" based on the local industry chain,When you need to quickly understand the willingness to cooperate with each other to achieve cooperation.

Third, improve the supporting collaborative system (Suzhou & Biomedical).

(Cold Spring Harbor Laboratory, images from the network)

"Biopharmaceutical war has been more than half, Suzhou has win the game." Wang Yuquan said.

In fact, Suzhou foundation in the biopharmaceutical field is not thick. Turning point occurred in 2009. Suzhou Industrial Park in cooperation with famous American Cold Spring Harbor Laboratory, local spending $ 1 million a year by the establishment of the International Conference on Cold Spring Harbor Asia platform, has in Suzhou and the surrounding hosted 130 international top level conference sessions, attracting the world from 60 20,000 passengers a number of countries participating scientists, including 22 Nobel Prize winners.

Office will greatly promote the rapid gather high-level personnel and attract new NUS Su Research Institute, Institute for Advanced Study at Oxford University (Suzhou) and other 18 international innovation cooperation platform floor, there are 12,000 foreign-related personnel to engage in Suzhou research and other related work. This not only allows Suzhou Asia-Pacific region has become the largest bio-pharmaceutical industry information distribution center, also required to meet the needs of biopharmaceutical companies and a large number of high-level personnel to quickly docking latest R & D results, local and therefore the emergence of a more than 1,600 biomedical companies. Ecological bio-pharmaceutical industry continues to prosper.

"What the industry needs to build pondering? Be wondering how this ecosystem can flourish. Rockery bonsai ecology is not dead, but a living rainforest, you need to think about how to make the metabolism, how to make the exchange of information, exchange of energy. These things are important. "(This article appeared in" John Doe "as a pseudonym)


[1] Wang Yuquan. Eco-technology innovation to build for the future

[2] 2020 Shu 18th century "black technology": the Central Commission for Discipline Inspection of the steam engine Birth Control Yuan national website [2020-11].

[3] Wang Yuquan: build industrial ecology, we need to cultivate a forest rather than a tree transplanting global outlet [2021-8].

[4] Silicon Valley’s most mysterious Palantir, in the end is a consulting firm or IT company Global outlet. [2021-8]

Weima car smart driving again, OTA upgrade brings a new experience

Committed to creating a Weima car who can buy a smart car to buy a smart car, a L4-level unmanned production model Weimar W6, to solve the parking problem for the majority of users, so that smart cars better Serving users, truly doing technology among Ware.

Weima Apollo works with one of the world’s largest automatic driving open platforms that reflects the long-looking to artificially intelligent in new energy vehicles. In 19 years, Weima’s automatic driving assistance system, which is a real L2-level automatic driving assist system, so that more users and families can enjoy smart driving. Weima W6 is based on the user’s pain points to realize the L4 unmanned driver of parking scenes, equipped with AVP unmanned self-service parking system, which increases the intelligence of new energy vehicles. And the power of the domain OTA upgrade also makes the car recently have a new feeling after the first OTA upgrade, increasing the KTV, intelligent voice assistant’s upgrade, and the control increase function of Huawei negative screen. Subsequent, Weima W6 will be based on open HAVP (autonomous learning parking), through OTA upgrades online PAVP (high-precision). HAVP and PAVP are suitable for fixed parking space scenes and non-fixed parking spaces, and strong combination provides users with the most practical driver’s technology experience.

In the second half of the year, Weima car also released hardware and software more intelligent advanced Weima M7. It is reported that Weima M7 will also support unmanned domain, will automatically drive more upstairs. According to industry related companies, 2020 to 2025 will become a critical period of my country to high-end automatic driving, by 2030, it is expected to realize the height of the suburbs and urban roads, and it is expected to be fully automatic after 2035. Our life will also bring more changes because of technology.

Hippocampus | Mask: Humanity is very likely to live in a matrix game of higher civilized simulation

Mask’s "The Joe Rogan Experience" program hosted by comedian Joe Rogan, smoking and drinking, playing big treasure sword, can be said to be eyeball.

In fact, in the two and a half-hour podcast, Mask comprehensively answered his own values, especially the explanation of him firmly, "We live in simulation", think that human civilization is likely Like the game, it is part of many simulated civilizations.
Humans may live in oneHuge and advanced computer games

Mask’s "Matrix-Style Simulation" Theory is based on the fact that the universe has existed 13.8 billion years.
Because this universe has nearly 14 billion years of history, the history of human beings is less than 10,000 years, so this time is enough for other civilization. He believes that the older civilization is likely to be our Creator and compare real life as a progress in the past decades.
"From the statistical perspective, there is a possibility that there is a civilization in such a long time, and they have found a very credible simulation method. Once the situation exists, then they build their own virtual multiple space is just a time. The problem. "
In fact, many people have mentioned that many people think that this is true, can create this simulation experiment, and they like to create "toys", and even create the universe is possible, this It is the low dimension of human creature, like high dimensions, how they created, we can’t imagine, because human thinking is always on a certain basis, we are hard to break through.
Mask also said that the analogical arguments are very sufficient, and also remind us that we don’t try to speed up the evolution of civilization, otherwise it will make the boundaries ambiguous, so that civilization is ending.
"One of these two things will happen. Because we exist, we are likely to be in the simulation."
He said that if it is this situation, then "Basic Reality" "" Basic Reality "that is used to simulate our real life may be very boring.
This is not Mask’s first sharing this idea, and he said on the Recode’s Annual Code Conference in 2016: He said:
"In view of us, we are obviously unable to distinguish between games, and these games can be played on any set of top boxes or PCs and anything else, and there may be billions of such computers or devices, then we are in the foundation reality. The probability is only one hundredty-one. "
"40 years ago, we have" pong ", which is two rectangles and a point. This is the beginning of the game. For 40 years, we have 3D simulation, and millions of online games. And technology is still developing, We will have VR and AR world soon. "
Although I can imagine that all of us may actually live in a huge and advanced computer game, physicists are indeed attracted to this idea, and in theory, it can be at least a possibility.
In fact, in addition to Mask, many science and technology leaders are obsessed with simulation theory and invest hundreds of millions of dollars. The Silicon Valley, which is a pile of companies such as Apple, Google and Facebook, and is the forefront of studying in this regard.
In 2016, the incubator Y Combinator president Sam Altman’s "New Yorkers" said that the entire Silicon Valley, including him, is very obsessed with the concept of computer simulation. He said: "Many people in Silicon Valley are very obsessed with this simulation assumption. They think that the reality we experience is the computer generated. The billionaires in the two science and technology community have been secretly recruiting scientists, hoping to bring us from simulation Liberate. "
In an interview, Mask also reaffirmed his serious concern about artificial intelligence, this topic he publicly discussed many times.But for the risk of artificial intelligence, he feels that people’s attitude is still not paying enough.
He said that one of the performances of insufficient attention is that people ignore the integration of humans and technology, but this integration has been carried out in an amazing speed.
"You are already a semi-mechanic." He said, "iPhone is actually your own extension, but now you can control the extension items you control, such as mobile phones and computers, communication between the mobile phones, and the data rate is slow. "
Mask latest five major attention:The network is the projection of the ID

In this podcast, Rogan conversed with Mask’s two and a half hours, the topic mainly focused on five aspects:
1. Main dangers of artificial intelligence are human beings become a weapon
Mask has long been a warning for the dangers of artificial intelligence. In March of this year, he said in south of Southwest China that artificial intelligence is far more dangerous than nuclear weapons, and the government should take action to regulate the development of artificial intelligence.
Mask said that the main dangers are not artificial intelligence to human attack. "There is a tricky problem here, that is, it is very attractive to use artificial intelligence as weapons. The danger is in human beings use it." On the other part of the podcasts, China Mask also added, "I tried to convince people to slow down." Slow down the speed of artificial intelligence, but this is futile. I have worked hard for many years, no one listened. "
Luogen also said, "This seems like a movie, the robot will take over everything, and you are frightened."
"But no one listens" Mask said.

2. Soon the significant progress of neural connection technology can be announced
If you can’t beat Ai, add AI.
This is the basic argument of Mask.
He believes that the future of artificial intelligence is to find ways to fuse human and machinery. In some respects, we have done it now: smartphones can be considered to have its own extension.
However, this extension and artificial intelligence relationship have bandwidth problems.
"You can’t communicate with your fingers, because it is too slow." Mask said.
Our goal is to greatly improve our biological self and digital self-communication channels, which can be achieved by neural-link technology, which helps control the long-term evolution of human and artificial intelligence.
"From the long-term perspective, this is like the purpose of neural connection, which is created a high-band width brain interface so that we can symbio with artificial intelligence."
Masque has found NEURALINK, as for the latest progress of this company, he revealed: "We will have some interesting things to announce, at least a number of levels than anything, may be considered more than anyone Be good. "
He depicted a long-term vision for this technology: adding artificial cognition to the brain "A.I. EXTENSION OF YourSelf" – The cerebral skin and brain edge system forms a synergistic relationship.
3. The society is playing with the earth "crazy game"
Mask said that in the process of more sustainable energy transitions, electric vehicles should be important as soon as possible.
"We really play a crazy game about the atmosphere and ocean. We collected a lot of carbon from the depths of the ground, then release these carbon in the atmosphere, this is crazy. We should not do this, this is very To speed up the transformation of sustainable energy. It is obvious that in the long run, we will exhaust oil. The oil we exploit and combustion is only so much. We must have a sustainable energy transportation and energy infrastructure. .
We extract hundreds of tons of carbon from the ground and discharge it into the atmosphere and ocean. This is a crazy experiment. This is the most stupid experiment in human history. Why do we do this? This is too crazy. "
4. Network landscape is the projection of ID
Mask said that the most successful online platform is the platform that resonates with our brain edge system – part of the brain is responsible for emotion, stimulation and memory, and these systems, such as social media, the intelligence of the society The share of the share is getting bigger.
"Imagine all of these things, including that original motivation, all we like, annoying, scared things, are on the Internet, they are projected by our brain edge system."
5. Once artificial intelligence becomes dangerous, supervise it is too late.
Mask said that before the government truly began to supervise an industry, it needs to develop and implement many years. Taking the safety belt regulations in the automotive industry as an example, this provision actually takes 10 years.
"This time framework is independent of artificial intelligence, starting from dangerous moments, you are impossible (supervise it) 10 years. Too late."
However, when artificial intelligence reaches the so-called singular point, what happens, no one can say it.
"It’s hard to predict, just like black holes, what happens outside your line of sight. It may be terrible, it may be very great. It is still unclear. One thing is affirmative: We can’t control it."
Mask also admitted that he still hopes to make the best contribution to humans. "I would rather keep a mistake but optimistic attitude."

finallyLet’s talk about smoking and drinking the big treasure sword.
During the interview, Masque took a whiskey, but he also took a mariha, but he didn’t like to do this, because this would make him proud efficiency decreased.
"This is like pumping a cup of coffee," Mask said, "I like to do things well, also like useful things."
The program is recorded in California, according to local laws, smoking cannabis
Whether you love him or hate him, you can’t noise Mask has such an interesting point.

Apache TVM or will change the training method of neural network

In recent years, artificial intelligence procedures have been driving changes in computer chip design, while new computers also make new neural networks in artificial intelligence.There is a feedback loop in progress, very powerful.

Its core is to convert a neural network program to software technology running on a novel hardware, and recently being able to get powerful open source projects.

Apache TVM is a compiler that is different from other compilers.It is not a typical chip command that converts the program to a CPU or GPU, but is a Tensorflow (end-to-end open source machine learning platform) or Pytorch form, study the calculation operation in the neural network, such as convolution andOther conversions and identify how these operations are best mapped to hardware based on the dependence between the operation.

The core of this operation is an Octoml, which is a two-year start-up, providing ApachetVM service and helping AI operations in mlops.

In the latest developments of hardware and research feedback cycles, TVM optimization processes may have shaped all aspects of manual intelligence development.

Today, TVM is specifically used for reasoning, which can be seen as part of artificial intelligence, which is fully developed neural network for prediction according to new data. But in the future, TVM will extend to training, which is a process of developing neural networks.

Apache TVM "Refining"

"Training and architecture search in our road map." Octoml joint founder and CEO Luis Ceze said, here, he refers to the best network design by letting the neural network to achieve automatic design The process of neural network architecture.

Will the neural network developers use TVM to affect their training methods?

"If they have not yet, I suspect that they will affect their training methods." Ceze said. "Some people come to us to train, we can train models", while tracking the performance of the model after training.

The TVM and Octoml services continue to expand, because the technology is a broader platform than the compiler.

"Users can extend TVM and OCTOML into a flexible, ML-based automation layer is used to accelerate, run on various hardware running in the machine learning model." CEZE said: "Every one in these hardware, no matter which one, Have their own way of writing and executing code, writing code and figure out how to best utilize today’s hardware, can be handled by ML developers and hardware suppliers. "

This also means that compilers and services have replaced this manual tuning – "Today" in the reasoning level, the model is ready to deploy, "Tomorrow", may be practical to develop or train.

The key to TVM attraction is higher performance in throughput and delay, as well as computer power efficiency. This becomes more and more important for increasingly challenging neural networks that have become increasingly challenging.

"There are some models that use crazy computations." CEZE observed, especially natural language processing models, such as Openai’s GPT-3, they are extending to 10 trillion neurological weights or parameters.

With the expansion of these models, they will also bring "extreme costs". He said, "Not only training time, but also in service time, all modern machine learning models is true."

Therefore, if it is not "an order of magnitude" optimization model, the most complex model is not truly feasible in production, and they still have to study curiosity.

However, performing optimization using TVM involves its own complexity. "It is necessary to get results in the way they need, this is a heavy work." Ceze said.

Octoml simplifies things by making TVM more like a button event. "From the end user’s point of view, they upload models, compare models, and optimize values ??on a large number of hardware targets." Ceze said: "The key is automatic – no low-level engineers write code sweat and tears." So Octoml The development work is to ensure that the model can be optimized for increasingly hardware.

"The key here is to make full use of each piece of hardware. This means that the machine code is dedicated to a particular machine learning model on a particular machine learning model." Things typically convolutive neural networks may be optimized to accommodate specific hardware blocks of a particular hardware accelerator.

The result is proven. In the baseline test of the MLPERF test kit for neural network reasoning, Octoml’s reasoning performance of the ancient RESNET (residual network) image recognition algorithm can get the highest score.

Reflect the value of Octoml in cooperation

Since December last year, Octoml services have been in the forecast and preemptive experience.

In order to advance its platform strategy, Octomal announced earlier this month, it has obtained $ 85 million C-round financing from hedge fund Tiger Global Management and existing investors Addor, Madrona Venture Group and Amplify Partners. This round of financing makes Octoml’s total financing rate of $ 132 million.

This fund is Octoml efforts to spread ApachetVM to a growing AI hardware. Also in this month, Octoml announced the establishment of partnerships with British company ARM Ltd., which is being acquired by AI chip giants NVIDIA. Prior to this, the company announced its partnership with Advanced Micro Devices and QualComm.

In addition, the ARM partnership will extend the use of Octoml services to ARM CPU core licensed, the latter leads mobile phones, networks, and Internet of Things.

In addition to the design of the neural network, the feedback cycle may cause other changes. It may broaden the way ML’s business deployment method, after all, this is the meaning of mlops.

CEZE predicts that the technology can greatly improve the portability of ML services as optimized by TVM.

Since clouds provide various trade-offs for various hardware products, dynamic optimization can eventually be used for different hardware objectives, it means to be more flexible from one goal to another.

"From essentially, it is useful to extrude more performance from any hardware target in the cloud because it provides more target flexibility." CEZE describes it: "Can automatically optimize portability And portability provides a choice.

This includes any available hardware in the cloud configuration, but also includes hardware such as delay, throughput, and cost that happens to happen to the same SLA. "

Next generation tumor immunization technology: Single-cell analysis and artificial intelligence from immunohistogenesis


The huge progress of immunotherapy has changed the model of cancer treatment, however, in view of only a few patients have responding to the blocked the immunogenic examination and other immunotherapy strategies, more new technologies need to be used to decipher the micro-micro-microenvironment of tumor cells and tumors (TME The complex interaction between the ingredients.

Tumor immunougueology refers to a comprehensive study of TME using multi-synthetic data using immunologically, immunological proteomics, immunogenic information, etc. reflects tumor immunization state, depending on the rapid development of next-generation sequencing technology. The high-throughput genome and transcriptum data can be used to calculate the abundance of immunocytes and predict tumor antigens, however, due to batch sequencing represents the average characteristics of heterogeneous cell populations, different cell subtypes cannot be distinguished. Single-cell-based technology can better analyze TME through accurate immune cell subpopulations and spatial structures. In addition, the depth learning model based on radiographics and digital pathology has a large extent to the study of tumor immunity, which is well shown in predicting immunotherapy response. The progress and breakthroughs of these new technologies have far-reaching significance for cancer treatment.

Tumor immunization micro environment

In the past few years, the research progress of tumor immunity has made us fundamentally changed the understanding of tumors. The definition of tumors also evolved from a simple tumor cell to a complex organ pattern structure, consisting of tumor cells, immunocytes, fibroblasts, vascular endothelial cells, and other matrix cells. Various cells and components near tumors, such as structural, such as immunofuncing cells, blood vessels, extracellular matrices, etc., also known as tumor immunochemical, and have become one of the topics of tumors. TME has been proven to play a decisive role in cancer, tumor progression, metastasis and recurrence.

TME includes extremely diverse immune cells, including T lymphocytes, B lymphocytes, natural killing (NK) cells, macrophages, dendritic cells (DC), granulocytes and mycinoid inhibitory cells (MDSCs) Wait. Typically, T cells, B cells, NK cells and macrophages help inhibit tumor growth, while MDSC and regulatory T cells (Treg) tend to inhibit anti-tumor immunity. However, existing studies have confirmed that in view of the complex interaction with tumor cells, the specific role of immune cells may have changed, and even completely opposite.

In summary, various immune cell types, even different functional conditions of specific immune cell types may have an opposite effect against tumor immunity. Therefore, this requires the most advanced bioinformatics techniques to minimize the immunological characteristics of tumors to the greatest extent, and provide more information to enhance our understanding of tumor immunity.

Immune genomics in NGS era

Over the past twenty years, NGS, including all genome sequencing (WGS), full exon group sequencing (WES), and RNA sequencing (RNA SEQ) has been successfully developed and applied to acquire human genome information. NGS produces high-throughput genome and transcriptional data to lay the foundation for the study of multi-step immune responses.

Immune cells in TME

TME consists of a variety of immunocytes, for quantification of tumor immunocytocyte components, conventional methods, such as flow cytometry and immunohistochemistry (IHC), due to its high cost and low tissue availability, is not applicable to large scale analysis. With the rapid development of NGS, we can estimate the abundance of dozens of immune cell types through NGS data, which are also proven to be reliable. These analyzes are mainly DNA and RNA sequencing, especially the latter. Regarding RNA sequence data, the principle of calculation methods is mainly divided into gene set enrichment analysis (GSEA) and inverse volume.

Typically, the representative algorithm based on GSEA includes Estimate, XCell, and MCP counts. A common feature of a GSEA-based approach is to establish specific genomic sets for each of the immune cell subsets of interest. The inverse volume of the cellular component is the reverse process of cell subtype convolution in a body tissue based on gene expression characteristics. Tools based on reverse volume include Decornaseq, Pert, Cibersort, Timer, EPIC, QuantiseQ and Deconf.

Identification of tumor antigens

Body cell DNA mutation, including mononucleotide variation (SNV) and insertion and deletion (INDEL), is the main source of abnormal antigen. Currently, the Genome Analysis Toolkit (GATK) is the industry standard of SNV and Indel by analyzing WES, WGS and RNA sequence data. It is also expanded to encompass copy number variation (CNVS) and structural variation (SVS).

Furthermore, the abnormal peptide needs to be combined with HLA to assist the T cell receptor (TCR) to initiate an immune response. The prediction of HLA is essential to identify tumor antigens. HLA Miner and SEQ2HLA are two early tools for HLA typing from NGS data, Four, six and eight-bit resolutions in PHLAT, HLAREPORTER, SNP2HLA, HLA-HD, Optype, and HLA-VBSEQ in different cancers. The performance is quite good. In these tools, Polysolver is one of the recognized standard tools currently using low coverage WES data.

In addition to identifying an abnormal peptide and HLA type, antigen MHC binding pro and is the next focus of tumor antigen prediction. Many peptide-MHC-I (PMHC-I) affinity prediction tools are based on artificial neural network (ANN) training methods and position specific score matrices (PSSM), as currently widely used tools NetMHC and NetMHCPAN. Due to the diversity of MHCII binding peptide and "openness" of the binding zone, predicting PMHC II affinity is more challenging, the number of PMHC II affinity prediction methods is far less than PMHC-I.

Single cell age immunohistology

Although the study of tumor immunity is used to study NGS technology to greatly promote the development of tumors, batch sequencing may cause signals to be diluted below the detection limit and mask the reaction of individual cells. This may hide many important biological phenomena. Until recently, the technical breakthrough of single cell-related methods thoroughly changed our understanding of tumor immunity, and transitioned from regional levels from regional levels to a single cell level.

Multi-color flow cytometry

Multi-parameter analysis is functional and physically distinguished by different immune cell subsets. In addition, with the advancement of technology, the instrument design of more parameters can be measured, such as 30 parameters and 50 parameter flow cytometry. However, due to the lower parameter accuracy, or higher precision measuring parameters, especially due to overlap between fluorescent dye emission spectroscopy, these disadvantages limit the application of multi-color flow cytometry to a certain extent And further development.

Matrix flow cytometry

The mass spectrometry is a new innovation in the field, also known as flight time (CYTOF), combined with streaming cytometry with mass spectrometry. Compared to conventional flow cytometry, mass spectrometry is not a fluorom antibody with metal isotope, and then quantifies the signal using the flight time detector, the detector detects at least 40 parameters, and avoids spectral overlap. Cytof has been confirmed to be a precise tumor tissue high-dimensional analysis method for explorating immunoassays and biomarkers found.

Although in theory, mass spectrometry flow cytometry allows us to detect up to 100 parameters per cell, but the processing speed and flux are limited by ion flight. After atomization and ionization, the cells are completely destroyed during the pretreatment, resulting in the end of subsequent cell classification. In addition, Cytof may not be appropriate for the measurement of certain low expression molecular features.

Spectral flow cytometry

Spectrum flow cytometry is another latest technological advancement that promotes traditional flow cytometry. Unlike the mass spectrometer, spectral flow cytometry is still labeled antibodies with fluorescent dye, but a new detector with a dispersion optical and measuring full emission spectrum replaces conventional optical and detectors. Based on the same principle, traditional flow cytometry and spectral flow cytometry maintain considerable compatibility, particularly in terms of commercial antibodies, but can better eliminate confusion factors, such as spectral overlap to improve efficiency . With the development of compensation technology, spectral flow cytometry may replace multi-color flow cytometry.

Single cell RNA sequencing

Based on streaming cytometry combines a specific label with the corresponding cell subgroup and identifies the label, indicating that the target must be determined before the sample collection, and the initial goal limits the information obtained from these technologies, only these Technologies find "known unknown".

The emergence of single cell sequencing technology pushes a single cell sequence to a new height. Single cells can be sequenced using a standard NGS protocol for predetermined targets such as streaming cytometry.

At present, SCRNA-SEQ applications are more mature than other methods, and the re-tumor immunotherapeutic field provides us with a lot of very valuable discovery and revelation. However, the technical noise produced by trace substance is still the most significant challenge. How to separate a single cell and maintain its bioactivity, how to solve the enlarged huge technical noise and improve sensitivity, how to get the highest number of measurable genes at the lowest price, how to analyze data more effectively, these are greatly improved single cells The threshold for sequencing limits its wide application.

Immunohistology and artificial intelligence

Artificial intelligence in tumor immunization studies mainly involve the following aspects: (1) Alleviation of immunological infiltration on artificial identification of pathological sections; (2) Provide an alternative technology to identify immune cells that are difficult to identify in naked eye The subpopulation and spatial structure; (3) Provides a non-invasive method to predict the TME characteristics of a particular patient and the reaction of immunotherapy.

Tumor antigen prediction based on deep learning method

The first step in deciphering the tumor antigen is to predict the abnormal peptide. In addition to identifying a variety of algorithms of SNV, the recently designed CN learning tool is also designed to detect CNV and exhibit good performance. About HLA profiles, Bulik et al. Generated a large integrated data set, including various types of cancer tissue HLA types and HLA peptides, which announced data that can be used to train intact mass spectrum depth learning model EDGEs, which is already non-small Verification is obtained in patients with cell lung cancer (NSCLC). In addition, two very promising depth learning methods Maria and MixMHC2PRED have recently been developed, which greatly increases MHC-II prediction accuracy.

Application of radiology in tumor immunity

With the development of artificial intelligence in medical imaging, the image is not just a picture, but a large-scale digital data, the process of analyzing the imaging data using AI technology is the radiographic. Radiological techniques applied to tumor immunization are mainly used to identify biomarkers reflecting immunoff and predict the therapeutic response of ICB treated patients.

Calculation of tumor immunity

AI, or so-called digital pathology, by calculating analysis, providing new insights for exploring the interaction between immune cells and tumor cells and the key behavior of cancer biology.

Similar to radiology, digital patients combine deep learning from the image to excavate the invisible information, so that we can understand TME in cell or molecular level. Digital pathology may be a promising method for studying the relationship between TME structure and cancer biology and treatment.

Application of immunohistology in tumor immunotherapy

Identifying ICB biomarkers for patient layers

As a target of ICB, the PD-L1 expression level detected by IHC is the first predictive biomarker, but some clinical trials have shown that ICB has only slight effect on some PD-L1 high expression patients, and ICB also Will respond to PD-L1 low expression patients. Therefore, other biomarkers are urgently needed to fill this gap.

In 2014, the researchers were linked to the clinical survival rate of tumor mutation (TMB) with patients who were treated with CTLA-4 inhibitors by WES. Subsequently, other retrospective studies also prove that high TMB is related to lasting clinical benefits. Regarding the method used to evaluate TMB, due to the high cost and complexity of WES, the FDA approves two alternative NGS platforms, which is FoundationOndongOne CDX (F1CDX) and MSKCC operable cancer target integrated mutant (MSK-IMPACT), and passed Prospective study of multiple cancers has been verified.

On the other hand, immune cell infiltration, especially TIL, a key role in immune response. In order to find more desirable treatment and prognosis biomarkers, single-cell sequencing is used to identify more immune cells. TCF7 + memory T cells have been found to be related to clinical improvement in patients with melanoma after anti-PD1 treatment, while stem cell-like TCF1 + PD1 + T cells have been confirmed, which contributes to tumor control in ICB treatment. Thorbing T cell subpopulations and functional conditions associated with treatment and prognosis are determined by single-cell sequencing techniques.

Prediction of new antigen in ACT treatment

Overpounding Cell Therapy (ACT) is re-transferred to the patient by transgene or amplified autologous or allogeneic T cells to enhance anti-tumor immunity. Immunoatology is mainly used to identify the ideal tumor antigen in ACT treatment.

At present, the new antigen-specific TCR-T cell has not yet entered the clinical application. However, it is gratifying that some case reports show the neat colorectal cancer, breast cancer and cholangiocarcinoma patients with immunohistology, T cell recognition tumor new antigen Effectiveness. TRAN et al. WGS on the sample of metastatic cholangiocarcinoma patients, determined 26 species of cell mutation. The series microchrome composed of mutant genes was transcribed and transfected into autologous APC, and then the new antigen was presented with TIL of APC and the patient’s source of TIL, and finally identified antigen-specific CD4 + VB22 + T cell clones and induced epithelial cancer.

The conventional autologous APC and T cell co-cultured new antigen selection is limited due to its low flux, high cost and time consuming characteristics. In order to eliminate these obstacles, more high-throughput immunogenic new antigen detection techniques were developed. Li et al. Established a Trogocytosis-based platform where the TCR and PMHC were combined, the surface marker protein was transferred from the APC to T cells. Therefore, the ideal new antigen can be identified by analyzing labeling positive cells. In the future, these emerging immunologies will achieve high-throughput antigen selection.

Choose a new antigen for individual tumor vaccines

The immunohistological method has been widely used in vaccine development in clinical studies. In general, a new antigen for generating a personalized vaccine is identified by analyzing a tumor and normal tissue WES and RNA sequences, and is identified by an algorithm (such as NetMHCPAN).

Similar to ACT, key parameters for tumor vaccine development are ideal for new antigen identification. In order to improve the accuracy of the new antigen prediction and the new epitope selection of the immunogenicity, immunohistology technology has made unremitting efforts in these respects. In a recent study, Wells et al. Compiled all new antigen predictions and selection methods and provided a new candidate measurement pipe, including 14 immunogenic characteristics of MHC presence and T cell recognition. This study laid a solid foundation for improving the efficacy of tumor vaccine and the adverse cell therapy.


In recent years, with the huge leap of emerging technologies in the field of immunohma, we can now analyze tumor immunity before.

In the bulk sequencing era, we can better explore the individual infiltration model of tumor immunocytes, and the prediction of abnormal peptides, HLA-based and tumor antigen MHC binding pro-and predictions, using immunogenics, predicting tumor antigens have been in front of it. It proves its reliable effect in clinical research.

In addition, with the development of single-cell immune related technologies, from multi-color flow cytometry to CYTOF, single-cell tumor immunization maps help us to classify the immune cell subtracitions to decide the TME component. The emergence of artificial intelligence also provides a new direction for the development of immunohistology.

With the booming of immunohistology, sustainable development needs several issues. First, although many methods of quality control and improvement algorithm have been implemented, the effectiveness of these technologies remains to be improved. Especially in tumor antigen prediction, single cell sequencing and space resolving transcriptional groups, technical noise and mixing factors have hindered subsequent analysis. Second, we look forward to the more cost-effective, easier to obtain and more automated techniques, thoroughly change the development of the discipline. Third, we also expect researchers to make full use of prior art to explore tumor immunity and promote clinical transformation.

Although there is still a lot of work to do, the immunohistology is likely to dominate the future in the future, and its clinical value will undoubtedly promote the development of the subject in the field of immunohistology, single cells and artificial intelligence.


1.Technological Advances in Cancer Immunity: from Immunogenomics to Single-Cell Analysis and Artificial Intelligence. SignalTransduct Target Her. 2021; 6: 312.

[Heart News] Artificial intelligence ECG and clinical risk factors forecast atrial fibrillation

Artificial Intelligence (AI) Analysis 12 Level ECG (ECGS) helps effectively evaluate the risk of trembling (AF). However, except for clinical risk factors, it is not clear whether artificial intelligence provides meaning and further improvement of accurate prediction of AF.

The study was conducted by 12 lead electrocardiograms for long-term primary health care patients (MGH) in Massachusetts General Hospital (MGH), which was inferred by convolutional neural network ("ECG-AI"). Subsequently, three Cox proportional risk models were constructed, each model included: a) ECG-AI 5-year room fuse, b) CHARGE-AF clinical risk score, and C) ECG-AI and Charge-AF rating ("CHAI"). The researchers evaluate model performance by calculating judgment (under the area of ??ROC curve, AUROC) and calibration in an internal test set and two external test sets (Brigham and Women’s Hospital and British Bioconbound). The model is re-calibrated in the British Biological Bank, and the risk of 2 years of room tribin is estimated. The most significant electrocardiogram characteristics of ECG-AI risk predictions were studied by significant mapping, and the correlation between ECG-AI and Charge-AF linear prediction factors were evaluated.

The training set includes 4,577 (age 55 ± 17 years old, 53% women, 2171 AF incidents), including 83,162 people (age 59 ± 13 years old, 56% women, 2424 AF events). CHARGE-AF (MGH 0.802, 95% Ci 0.767-0.752, 95% Ci 0.741-0.752, 95% Ci 0.741-0.763; UK Biobank 0.732, 95% Ci 0.704-0.759) and ECG-Ai (MGH 0.823, 95% Ci 0.790- 0.856; BWH 0.747, 95% Ci 0.736-0.759; UK BIOBANK 0.705, 95% CI 0.673-0.737) Auroc comparison. The CH-AI group Auroc has the highest: MGH 0.838, 95% Ci 0.807-0.869; BWH 0.777, 95% Ci 0.766-0.788; UK BIOBANK 0.746, 95% CI 0.716-0.776). ECG-AI (MGH 0.0212; BWH 0.0129; UK Biobank 0.0035) and CH-Ai (MGH 0.012; BWH 0.0108; UK BIOBANK 0.0001) Calibration error. In significant analysis, the effect of electrocardiogram P wave on artificial intelligence model predictions is the largest. ECG-AI and Charge-AF linear predictors (Pearson R MGH 0.61, BWH 0.66, UK BIOBANK 0.41).

Table 1 Baseline Information

This study developed a depth learning model ECG-AI, using 12 lead electrocardiogram data precisely predicts the time of the atrial fibrillation. ECG-Ai uses approximately 100,000 electrocardiogram training from more than 40,000 patients from the primary health queue. CH-AI is a model of ECG-AI and Charge-AF, compared with Charge-Af, in patients with more than 80000 clinical features, a large number of clinical features of more than 80000 cases, in multiple prognostic model indicators The effect is better. ECG-AI evaluates the risk of CHARGE-AF rating containing 11 components. In addition, the study found that ECG-AI and CHARGE-AF are highly correlated, indicating that ECG-AI can largely reflect the ECG performance of the determined atrial fibrillation clinical hazardous factor. The results show that the clinical risk signal based on deep learning provides similar predictive utility, ECG-AI and clinical risk factors complement each other, enhance risk prediction skills.

Table 2 Test Concentrated atrial fuse

The Attia team has developed a depth learning model that is 80% for the model of atrial fibrillation state classification of diagnosis room fibrillation and sinus rhythm. Raghunath et al. Subsequently developed a neural network that utilizes 12 lead-optic maps to predict atrial fibrillation event. In one year, the atrial fibrillation recognition capacity is good, in subset analysis, compared with Charge-Af, performance is better. The study made substantial supplements in the aforementioned work, introduced a deep learning model, explicitly incorporating the survival time, and conducted strict epidemiological surveys, including quantitative errors, calibration, re-classification and extensive external validation. However, due to the data is not applicable, the study cannot be directly compared to the aforementioned model.

The results show that when using a strict epidemiological index to assess, the depth learning model uses an electrocardiogram to assess the risk of atrial fibrillation is reliable and effective. Specifically, the study evaluated ECG-AI of the test set consisting of independent individuals from: a) the same institution with the training set, b) independent institutions in the same medical network, c) from atrial fibrillation risk Lower different region forward-looking research queues. With the previous research results, ECG-Ai is best in the closest population closest to the training set, and the difference in different samples is small, highlighting the importance of extensive external verification in assessing clinical utility. In the end, researchers suspect that the differences in identification ability may differ from different characteristics (eg, age, baseline atrial risks), and thus the specific electrocardiogram characteristics are different from the relevance between the risks of the long-term atrial fibrillation.

This study is of great significance to the relationship between depth study of ECG hazardous signals and traditional atrial fibrillation clinical risk factors. First, the clinical hazardous factors can be performed on an electrocardiogram in a perceived manner in the deep learning model. Specifically, the ECG-AI probability and the Charge-AF score have moderate correlations in each test. In addition, the characteristics map and median waveform analysis have found that ECG-Ai probability is largely influenced by atrial decoder and a replenishment period, and the atrial structure and function may be affected by chronic diseases such as age and hypertension.

Second, the depth learning model seems to extract a risk factor complementary to clinical hazardous factors. CH-AI is a combination of CHARGE-AF and ECG-AI, which always shows better AF identification, calibration, and re-classification. This shows that the predicted effect of the model combined with the clinical risk information and the artificial intelligent veneer risk is significantly improved compared to any of the methods alone. Similarly, atrial fibrillation is significantly higher based on the risk of ECG-AI and Charge-Af, based on the use of two models alone. In addition, the predictive differences in CH-AI and Charge-AF are more significant over time, which may be related to the time accumulation effect of clinical risk factors, while the authentication of ECG-Ai remains relatively unchanged, prompting an electrocardiogram-based atrial fibrillation Risk predictions can contribute the most in short-term predictions.

The design background is explained below. First, the researchers conduct ECG-Ai training for individuals who conduct at least one ECG clinical examination. At the same time, the patient baseline information includes each component of the charge-AF rating. These two requirements have introduced potential selection bias. However, the study included EHR samples accepted longitudinal primary care may reduce bias, and the research model continues to distinguish AF in a completely independent forward-looking queue study. Second, the research training set represents an individual of a single mechanism. Training for larger samples in multiple institutions helps to generate more accurate and more universal models. Third, due to the limited follow-up of British bio-bank, the evaluation time window is 2 years, while MGH and BWH are five years.

Fourth, ECG-AI is a black box model. However, compared with the previous AF prediction model, the study uses a significant mapping and median waveform analysis, and the biological reasonable electrocardiogram changes (for example, P-wave duration prolong) have the greatest impact on the AF risk assessment. Fifth, 5 years of predictive window may represent the risk of atrial fibrillation, but it cannot be intervened immediately. However, the study also has a consistent difference in a shorter time window. It is not yet excluded that ECG-AI recognizes the previously have unexposed atrial vibration. Sixth, ECG-AI and CH-AI need to recalibrate in Biobio Bioconbanks. It is necessary to recalibrate when replacing the prognostic model. It has been very accurately estimated by simple recalibration of the incidence of the atrial fibrillation of BBBs. Tip The initial calibration error can be completely attributed to the incidence of baseline atrial fibrillation in the British branch, approximately one-third of MGH and BWH. In addition, CH-Ai has a better calibration effect than Charge-Af, although Charge-AF has a similar recalibration in Bio Biological Bank.

Figure 1 Research overview

Figure 2 identification of atrial fibrillation

Figure 3 Correction of atrial fibrillation

Figure 4 In accordance with predicted risk stratified atrial fuse

Figure 5 ECG-AI depth learning model two visual behavior forms

In summary, in three independent test sets of more than 80,000 people, ECG-AI (a depth learning model using 12 lead electrocardiogram clearly predicts the time of AF occurrence time) is compared to Charge-AF clinical risk score containing 11 components. Provide a 5-year AF risk forecast similar to the effect. The CH-A model combines Charge-Af and ECG-AI to identify, calibrate, and re-categorize more accurately. Based on the depth learning ECG’s atrial risoting assessment has the potential for wide application, it can provide accurate and promotion of atrial fuser risk assessment within a few years of ECG inspection.


Khurshid Shaan, Friedman Samuel, Reeder ChristoPher, et al. Electrocardiogram-based deep Learning and Clinical Risk Fabrtors to Predict Atries, 2021, In Press.

Weifang Qingzhou: Smart agriculture is injecting new vitality for rural villages

"Xiaofeng, how much is it now?" "The real-time temperature in the shed is 25 ° C", "" On the 2nd front airproof "" No. 2 ventilation mouth has been opened ". On November 15th, in the Poverty Alleviation Project in the Nanzhuwangwang Ecological Park in Qingzhou, Qingzhou City, Yunshi Town, the planting households were using mobile phones to carry out the wind. A variety of operations in the wind. "This artificial intelligence can also forecast the weather in advance. When you encounter the rain, you will close all the wind in advance, eliminate the possibility of rain into the shed. When the temperature in the shed exceeds 30 ° C, you can automatically open all the wind Avoid the temperature too high. "The Demide slid on the phone screen and introduced it. Automatic wind turbine, automatic temperature control, intelligent water fertilizer integrated equipment, intelligent fill light, etc., thereby achieving remote management and intelligent operation, etc. Even if there is no sun, rain and snow weather does not affect the growth of vegetables in the shed. In the greenhouse computer control room, you can see the environmental data in the shed on the screen, including light, temperature, humidity, carbon dioxide concentration. "This is our southern Xiao Wang Zhijia agricultural data platform, which is our own research and development, a total of the Internet of Things system, automated central control system, price analysis, planting warning system, ozone water straw returning system and fire eye gold eye disease intelligent identification system six Match. "Sun Guo, secretary of the Party Branch of Nan Xiaowang Village, told reporters.

Yuancai "new track" broke out! 8 "virtual people" top faucet, relieved!

The Yuan Universe is the next generation of Internet revolutionary interactions. The Yuan Universe is parallel to the real world, providing an open virtual world of games, shopping, social immersion experiences. In the stage of moving Internet traffic, the number of users is weak, and the Yuan Universe is the current focus of the technology company.

Today, the speculation of the Yuan universe, and pushing up new, the virtual man became the top theme today. It is estimated that many stockings are foggy, what is "virtual person"?

The virtual person is actually a virtual person with digital shape. If it is difficult to understand, it can also be seen as a similar Taobao, the smart customer service in the Jingdong background, but only in the video manner. This year, Double Eleven Shopping Festival, quickly realize the virtual anchor "Xiaofang" live band; Tencent cloud launched a new intelligent product matrix based on new generation multi-modal interaction technology, Tsinghua University Virtual Student, Tsinghua University, Tsinghua University Ice is also a beautiful virtual person.

From the industry market space, the data shows that the size of China’s virtual idol core market is 3.46 billion yuan, which is expected to reach 6220 million yuan in 2021; 2020 virtual idol drive the surrounding market size of 64.56 billion yuan, expected to be 1074.9 in 2021 100 million yuan.
The city’s universe is still far away, some is somewhat deficient, but virtual human development has matured, and it is also important to become an important tissue part of the Yuancosian.
Regardless of the Yuanyuan or the virtual person, it is currently the concept of the concept. The hype is the future imagination, and the performance is still far away. The speculation of the Yuan Universe has spread from the game that has just started, which is a multi-segment area, which is a common phenomenon of super subject. New energy vehicles are also spreading from the initial lithium ore, lithium battery, upstream materials to components and chips.

8 "virtual people" top faucet

1, blue cursor

Beijing Blue Cursor Data Technology Co., Ltd. is a public relations consulting service and advertising service. The core business is to provide brand management services and advertising design, production, proxy, release business, and to undertake exhibitions. In 2019, the company’s founding strength continued to obtain the industry height recognition, and won the golden offer the audience award, the IN2 Sabre Awards Gold Award, Times Golden Image Award Best Visual Design, etc. More than 100 innovation top awards.

2, Sichuan Media

The main business of Sichuan News Net Media (Group) Co., Ltd. is a new media integrated marketing, mobile information service, interactive TV business. The company’s main products and services are publicity and promotion services, online public opinion services, advertising agency operations, technical services, mobile news information dissemination, mobile phone report value-added services. The company’s "Spicy Community" has also been rated as a "China Internet Site Brand Column (Channel) recommended by the State Council Information Office and China Internet Association.

3, beautiful culture

Msusheng Cultural Creative Co., Ltd. is closely related to the cultural industry and is committed to the cultural industry related business. Since the company’s listing, based on the original animation derivatives, the company focuses on the upstream business, improve the industrial chain, in the cultural industry chain, including anime, games, film and television, derivatives, etc., the strategic layout, and implement the transformation and upgrading, preliminary The construction of the cultural ecotrial circle of "Self-owned IP + Content Production + Content Distribution and Operation + New Media Operation + Derived Development Design + Online Retail Channel". The company’s products mainly include services such as IP derivatives, animation, games, film and television products and operations such as light game service platform business.

4, Mango Super Media

The main business of Mango Super Media Co., Ltd. is mainly composed of three parts of the new media platform operation, new media interactive content production and media retail.

5, Jiecheng Shares

Beijing Jiecheng Century Technology Co., Ltd. is a professional engage in new media copyright operations, film and television content production and distribution, audio and video technology services and digital education cloud platform construction. The company’s main products include media asset management system solutions, high standard cleaning, non-editing network solutions, all-stage multi-heterogeneous network solutions and all-Taiwan unified monitoring and monitoring solutions.

6, Fangzhi Technology

Shenzhen Fangzhi Technology Co., Ltd. is mainly engaged in research and development and sales of primary and secondary school synchronous education software and providing online online services. With the experience of information technology accumulated for many years, the deep understanding of teaching, management, discipline in-depth integration and education teaching needs, the use of mobile interconnection, cloud computing, big data, artificial intelligence, etc., established advanced resource management operation platform, effective Teaching methods and information technology organically combined.

7, Zhen Electronics

The main business of Shen Si Electronic Technology Co., Ltd. is the identity certification business, industry deep-cultivation business (financial industry, security industry, medical industry), artificial intelligence business (Shen Siyun, single AI products, and AI cloud service solution). The company’s main products are ID card reading equipment, mobile exhibition industry, silver medicine self-service, computer vision, convenient payment. The company has won the second prize of "Chuangke China" SME Competition, the company’s wisdom catering enters the 2019 enterprise canteen information service service national TOP10, company with "Smart Business Service Robot Project" finalists in China New generation of artificial intelligence industry innovation Key task potential units.

8, Wanxing Technology

Wanxing Technology Group Co., Ltd. continues to focus on video creative software business, actively promoting the technological innovation and quality of drawing creativity, document ideas and utilities, quickly responding to market demand, strengthening mobile service layouts, and constantly exploring new products. The company’s main products include digital creative products (video creative software, drawing creative software, document creative software, utility software), and the company continues to obtain "national planning layout" certification in the "National Planning Layout" certification.

Shanghai: Strengthening Digital Infrastructure Planning and Layout Construction A New Generation Communication Network, Data Center, Artificial Intelligence Platform and other major infrastructure

Shanghai Municipal People’s Congress Standing Committee: "Shanghai Data Ordinance" has been passed on November 25, 2021, was appointed on November 25, 2021, from Shanghai’s 15th People’s Congress.From now on.Strengthen the digital infrastructure planning and layout, improve the service capabilities of e-government clouds, e-government external networks, build a new generation of communication networks, data centers, artificial intelligence platforms, etc., establish digital numbers, storage, calculation, and safetyInfrastructure system.Support Pudong New Area to strengthen data transactions related to digital trust system construction, innovation integration big data, block chain, zero trust and other technologies, build digital trust infrastructure to ensure trusted data trading services.

Around the hardware bottleneck, double the chip power, and the software level is deeply excavated by the chip performance?

If you are inventory in the past two years of industry hot words and social hot words list, the "chip" must be famous.

With the broad practice of AI technology in all walks of life, the application layer is getting higher and higher. Correspondingly, the requirements for deep learning on chip power increases. Information age, everywherechipHowever, the chip is a scarce resource.

There are two ways to work in the industry, one is a custom chip, one is a modification of the model. Reduce the requirements of force by using a small model or compression model. There are highly and disadvantages in two ways, and the custom chip performance is strong but cost, cycle, and the risk is very low; the cost of small models or compression models is low, but the cycle is short, but it will lead to decline in accuracy, it is difficult to high-precision and high performance. A better balance between it.

Excessive redundant calculations in the existing AI calculation and the ability to operate the capacity of the work engine constrain the mining of chip performance. In the case of the imbalance of the chip resources supply and demand, the mainstream practice is a challenge to attack productivity.

There is also a technical team. A family called CoCoPie AI company announced that you can dig existing chip points from the software level by compression and compiling collaborative design technologies, which is expected to make existing chip performance. So we found the person in charge of Cocopie Li Xiaofeng. According to him, CoCopie has already built Coco-Gen and Coco-Tune and other products. These products are able to process artificial intelligence applications in real time without additional additional artificial intelligence hardware.

He told infoq: "Cocopie unique AI software technology stack, solves the development and popular bottleneck problem of end side, which is still unique in the industry. Test data and customer feedback indicate that the comparative advantage with other programs is very obvious. There is a big chance to win in the tide of the end of the equipment wisdom. "

Bypassing the hardware bottleneck, double the chip power, is the software level increase whether the chip performance is feasible? In order to further understand the techniques used by Cocopie, the answer to this problem is obtained, and the InfoQ has interviewed Li Xiaofeng.

Q: What is the specific technical implementation and academic papers support by optimizing compression and compiling collaborative design, solving performance issues, specific technical implementation and academic papers?

Li Xiaofeng: Cocopie’s technical core is the three professors in the founding team. They are people who are very high and unusually diligent, and they are leaders in their respective fields. Among them, Wang Yanzhi professor has focused on the AI ??model algorithm, Ren Bin’s focus on the AI ??model compiled, and Gu Xi Peng ‘s professor’s system engine. These research areas are technically a well complement, which constitutes the iron triangle of AI calculation optimization technology, mutual indispensable, jointly builds the company’s core competitiveness, and is also a kind of sky.

First introduce the basic techniques of the AI ??model optimization. An AI task runs on the device, actually the process of maping the AI ??model to a chip instruction sequence. Compression and compilation are two key steps to perform. The compression optimization of the model itself will be reduced by the compression optimization of the model itself by weight-bearing pruning, weight quantization, reducing the complexity of the model itself. The compiled model is optimized for the compressed model. In this way, the AI ??task is more efficient, and the chip capacity can be made to the other hand.

But compress and compile these two steps, currently do not do very well in the industry. Existing technologies can only compress, or only compile, or although both have, they are designed to isolate each other, there is no good collaborative design, so it is difficult to achieve both effects of reasoning accuracy and ensuring operational efficiency. .

The core of CoCopie technology is to compress and compile two steps of "collaborative design", that is, consider compression and hardware preferences when designing compression, and select the compression model when designing the compiler. Compilation Optimization Method. Corresponding to compression and compiling two steps, we design two components for the Cocopie frame: Coco-Gen and Coco-Tune. Coco-Gen generates an efficient execution code by generating phase associations based on mode-based neural network twigs, and Coco-Tune can significantly shorten the process of compression and training of DNN models.

Cocopie technology is common and can be widely used in various CPUs.GPU, DSP and AI special chips such as NPU, APU, TPU, etc.

Cocopie published a large number of top international conference papers in the relevant field, from the upper AI application optimization technology, AI model design technology, to compiler optimization technology, underlying hardware related optimization technology. In particular, Cocopie’s technical introduction article was published in the Communications of ACM in June this year, which is the flagship publication of the US computer society, which is released with this year’s Tuling Award, which means that the academic community is highly recognized to COCOPIE.

Q: Can the current core product COCO-GEN and COCO-TUNE can be used alone?

Li Xiaofeng: These two products provide key technologies for our AI model optimization, Coco-Gen generates efficient execution code by generating phase-based code-based code generation, Coco-Tune can be significantly shortened by generating phases associated with mode-based neural network twigs with mode-based code; COCO-Tune can significantly shorten DNN model compression and training process.

Coco-Gen and Coco-Tune can be used separately. They constitute the core of the CoCopie toolchain, so it is prioritized. As a bridge that connects the upper layer AI task and the lower hardware, the Cocopie’s product system will continue to add new members.

Q: Solve the chip shortage problem from the software level, is there a similar software technology in the industry?

Li Xiaofeng: The current end Ai technology stack, only Cocopie optimization technology can meet or exceed the performance of the AI ??special chip on the mainstream chip, which is through a large measure of verification. Currently known techniques, or side heavy compression, either focused on compiling, and did not see the techniques of the two collaborative design, which is the patented technology of CoCoPie.

Because although the current mainstream chip has a good potential, we must use this potential, you must convert the AI ??task to a suitable vector calculation by compressing and compiling synergies. Quantity. This is the key to CoCopie’s technical critical.

Q: The ruler is short, the inch has a long, what is the advantages and limitations of this technology?

Li Xiaofeng: The advantage of COCOPIE is that on the one hand, it is possible to make a lot of the AI ??tasks that cannot be operated properly on the end side. On the other hand, it is dedicated to the end side. AI chipThe AI ??task can be run, and now you can run through the mainstream chip.

The implementation of the AI ??task will always be restricted by the chip’s strength. Cocopie technology has always its own limitations, and the liberated AI is not unlimited. In addition, Cocopie technology is currently focused on the AI ??inference tasks, as for the acceleration of special AI training tasks is not our focus.

Q: Cocopie technology can make the chip to increase 3-4 times, allowing the chip effectiveness to increase by 5-10 times, what is the standard? Can you achieve this level for different chips?

Li Xiaofeng: These data are actually measured, and the peer review has passed the customer’s identification. In other words, there is a technical support in technology, and there is a product in practice.

For example, comparison with universal chips and Google TPU-V2: Use Cocopie, VGG-16 Neural Networks in Mobile Equipment Samsung Galaxy S10 is nearly 18 times higher than the TPU-V2, and RESNet-50 has achieved 4.7 times. Efficiency is improved.

On the same SAMSUNG GALAXY S10 platform, both the C3D and S3Ds of behavior recognition have increased by 17 times and 22 times higher than Pytorch Mobile, respectively. Running Mobilenetv3, Cocopie’s speed increases by nearly 3 times and 4 times respectively, respectively, respectively, respectively, respectively.

In addition, the results of the QUALCOMM TREPN POWER POWER PROFILER have been shown that Cocopie is shortened than 9 times higher than the TVM, but the power is only less than 10%. In the work of DNN reasoning acceleration based on AQFP superconductance, our research is also the highest energy efficiency in all hardware devices through a low temperature test.

Q: The effectiveness of the performance will not be empty, what are the requirements for the operation of this software to the hardware environment?

Li Xiaofeng: Yes. Cocopie technology is not high, the mainstream chip can be satisfied, in particular, the chip requires vector computing power, such as the Neon instruction set, Intel’s SSE, AVX instruction set, RISC-V vector extension, etc. They are all current CPUs, GPU and APU / NPUs are not more useful. Of course, if there is no vector computing power, Cocopie’s technology can still work, but will be affected.

Q: What is the main challenge encountered during the technology practice?

Li Xiaofeng: The main challenge encountered in practice in practice is that our current product system is not very perfect, and customers’ needs are also a variety of different, and the specific service methods are very different. Therefore, we have not yet carried out large-scale business. Promotion, mainly for some key areas, key customers, such as representative mainstream chip providers, equipment providers, software service providers, etc., selectively provide services in accordance with our product development strategy. We will carry out this process, and we will work together with a variety of customer needs and continuously explore the best product service system.

Q: Is this technology currently have a practical floor case?

Li Xiaofeng: There are still more than a dozen people in collaborative customers. These customers have multiple fields, such as Tencent, Drop, a famous chip platform provider, a famous mobile phone manufacturer, and the Ministry of Communications, the world-renowned service provider Cognizant. Wait.

Q: The mainstream processor is a more popular solicity of real-time manual intelligence. Do you agree with this view?

Li Xiaofeng: Yes. For end-side devices, the mainstream processor is a more favorable solution to real-time manual intelligence.

1. From the functional say, the end side device resource is limited, the application scenario is killed, while the dedicated AI processor function is relatively curing, and the abnormal flexible functional demand for the end side has a large challenge. Mainstream processors can handle AI issues through software technology, of course, there is no need to have another branch.

2. From the technical statement, the practice of solving the special chip of the AI ??problem is actually increasing the processing capabilities of vector computing, improves memory access efficiency, and some are tensile calculation units. As mentioned earlier, the current mainstream chip is actually a vector computing unit. These vector computing units may be weak than the dedicated tensile processing chip, but the current AI task is generally sufficient, provided that there must be excellent model compression and compilation tools, able to make AI tasks through a delicate design Convert to a suitable vector calculation and control the overall calculation.

3, from the cost In addition to purchasing the cost of the AI ??chip itself, there are some recessive costs. Many chips will cause redesign of PCB, heat dissipation, and the like, and the package is also an extra cost. Many devices such as smart headphones, micro medical devices, etc. are sensitive to these factors.

4. It is important to emphasize that Cocopie’s technology does not exclude special AI chips. As a full stack of AI, Cocopie also supports the AI ??processor to allow AI processors to play a greater performance. Therefore, we also have an important role in a specific application area.

Q: Is Cocopie a product transition time?

Li Xiaofeng: Only, because of the generalization of the terminal AI, CoCopie is the leader of the advanced technology in this field, and we believe that its future development space is very large. We have a complete set of product development strategies in our inside, and future product forms will be different, but core technology is inherent.

In addition, it is assumed that there is a popularity of the ai-specific chip to a day, nor does it adversely affect the living space of COCOPIE. First of all, the AI ??special chip will never be more popular than the general-purpose chip. For universal chips that can be done well, it may still be more effective in universal chips; secondly, the AI ??chip has developed even if it is developed. Enhance the optimization technology. Our technology will make the AI ??chip’s ability to further improve. In fact, universal chips are the same, such as CPU or GPU, no matter how cheap, high performance, still require high performance compiler support, such as LLVM or NVCC, etc.

In a long time, the development of AI technology stacks will only get more and more demand for Cocopie software technology, just like mobile SOC chips, the function is not more and more simple, but more powerful, 8 Nuclear phones are common, and the requirements for software technology are getting higher and higher. In fact, AI calculates the requirements for integration to exceed the development speed of the AI ??hardware capabilities. According to the research report of the US MIT University, AI calculations have been developed in recent years, the development is 700 times every two years. This development speed is only impossible to meet the requirements of hardware capabilities, and must be better in software technology.

Q: How do you think about today’s popular big model technology?

Li Xiaofeng: The big model often achieves higher AI capabilities in the case of training data. This is the way to explore unknown world exploration. This is like a high-energy physics community, in order to make new discovery, it continuously constructs a higher energy of particle crash. However, this thing is also needed to see from two aspects. If you pursue a larger model, the amount of training data, training time, integration support, energy consumption, etc. are increasing, and the marginal benefits will be getting smaller and smaller. This trend is obviously unsustainable. It may only be practiced in the work of individual significant challenges in the future.

Here is a number on the big model gain, and the RESNET is a famous computer visual model released in 2015. The improved version of this model is called ResNext, which was advent in 2017. Compared with RESNET, the computing resources required by ResNext (measured with total floating point calculations), but accuracy is only 0.5%.

Give a number on carbon emissions, according to the Report of Forbes Magazine, since the depth study has developed in 2012, the calculation resources needed to generate first-class artificial intelligence models, an average of 3.4 months This means that the energy required to train the AI ??model has increased by 300,000 times from 2012 to 2018.

If you do a contrast, the ability to learn in deep learning is still a big gap even if you compare with a baby, let alone compare with the brain of adults. And our adult brain operation, only 20 watts of energy, which can only supply a bulb.

We obviously impossible only by expanding the model to improve the machine’s intelligence, and the academic circles are constantly exploring new methods.

Q: Can you talk about the basic software industry that is currently being floating? Simply talk about your judgment on the chip industry?

Li Xiaofeng: The importance of basic software is getting bigger and bigger. There are two reasons, one is that the technology has developed very quickly in recent years, and the basic software has actual needs, and it is necessary for the future; the other is that a large number of high-quality engineers have been cultivated.

The chip industry will continue to flourish. The development trend of science and technology is to continuously penetrate the digital world into all aspects of the physical world, while the simultaneization is the constant implantation of the chip in various equipment. The core means of intelligent in the upper wave of equipment is to implant the chip on the device, and the core means of running applications, and this wave of intelligent core means is to run deep neural networks on the equipment, which is a great trend of vastness. This is also the fundamental opportunity of CoCopie.

Li Xiaofeng told InfoQ, Cocopie’s technology leading advantages have been at least a few years, enough to compete in a computer field. Cocopie technology is not to solve the problem of chip shortage, but to achieve the generalization of the AI ??task, it is just a matter of happening, which is a by-product of CoCopie technology capabilities.

In Li Xiaofeng, in the face of complex and diverse scenes and terminals, the existing technical level cannot fully exert the ability of the mainstream chip, so there is a Cocopie’s development space. It can be confirmed that the development of Cocopie provides a new idea to put the chip capability "things".