The reporter | Zhou Xiang



How hot is AI in China? The following set of numbers may illustrate this.

Megvii closed a $100 million Series C funding round in December.

In May, Yitu raised Rmb380m in a Series C financing round.

Two months ago sensetime, less than three years old, raised $410m in series B funding at a valuation of more than $1.5bn. The move not only set a record for the highest single round of financing in the global AI field, but also made Sensetime the AI unicorn company with the highest financing amount in the world.

And these are just computer vision startups.

Futility or not, the amount of money sensetime has raised, buoyed by The State Council’s New Ai Development Plan, has clearly touched a nerve.

But how does sensetime view the fundraising internally? What should I do with the money I raised? AI Tech Camp interviewed Sensetime co-founder Yang Fan, hoping to get a full scan of the “Internet celebrity” company.



                                             

                                                   
Yang Fan, co-founder and vice president of Sensetime

Why is shang Tang “so expensive”?

Compared with its competitors, sensetime has raised so much money that some people even questioned the authenticity of the figure online.

However, Yang Fan does not care about this “suspicion”, after all, Ding Hui, race lead, Huaxing, cornerstone and other 20 large investment institutions have come forward to endorse.

In his opinion, there are two main reasons why shang Tang is “more expensive” than Kuang Shi and Yitu.

First of all, as a technology-oriented company, sensetime’s core team has very strong original ability. In addition to publishing several papers on CVPR, ECCV and ICCV, the three top conferences on computer vision every year, Shangshang-Time has also made remarkable achievements in mainstream academic competitions such as LFW and ImageNet. In addition, Sensetime has more than 350 patents, 90% of which are invention patents.

Secondly, Sensetime has a strong technical liquidity ability. At present, Sensetime has more than 400 corporate customers, including such top customers as UnionPay and Xiaomi.

In addition, this may also have something to do with the field sensetime is in. After all, the most commercialized area of AI right now is computer vision, and that’s where Sensetime’s money is. Yang fan explained to AI Tech Camp why “vision” is so important.

Who has been the most profitable Internet player of the past 30 years? Search engines. Search engines are all about presenting all the information in the world to users. Text information is refined twice, the cost of information refining is relatively high, and there will be information loss or information doping in the process of refining. Image is the most informative way for human beings to interact with the outside world, and vision accounts for more than 90% of the information interaction between human beings and the outside world.
When you have the ability to structurally extract and analyze the transmission of visual information, the value it brings is far beyond imagination.

What should I do with the money I raised?

Around computer vision, Sensetime’s biggest businesses are still security, finance, and smartphone-based solutions such as smart photo albums and smart beauty. But Sensetime’s ambitions clearly go beyond that.

Yang Fan said that after obtaining financing, in addition to expanding investment in the underlying technology platform, it will also expand into new areas:

First, the underlying platform requires continuous investment of human resources. There is still a huge gap between what machines can do and what humans can do in terms of understanding video content.



Second, when we provide solutions, we need to cooperate with industry partners, so we need to know more about the upstream and downstream of the industry, which involves the iterative evolution of technology, so that our solutions can be more targeted.



Third, we will advance the layout of some industries, including unmanned driving, medical imaging and so on.

According to introducing,
Sensetime has been promoting driverless driving since last year. Now it has formed an in-depth cooperation with one of the world’s top five automobile companies and will invest more resources in the future.

What are the technical challenges?

Of course, there is money in computer vision, but there are plenty of technical hurdles. After all, AI today is mostly in the “perceptual” stage, which is a long way from being “cognitive.”

The same goes for shang Tang.

Sensetime used to focus on perception, but now it’s branching out deeper:

One direction is image intelligent enhancement. If you really want to do an image recognition solution, you have to think about how to do it from the very beginning. For example, if you do security, you have to think about how to solve the problem.



On the other hand, when we really deepen the industry solutions, you will find that the technical problems are no longer just technical points. For example, over a thousand videos are put together and I want to do a quick content search, which is actually a mixture of pure visual technology research problem and big systems engineering problem.

In addition, underlying issues such as the fundamental mechanism behind deep learning are also a research direction of Sensetime.

However, the AI Renaissance sparked by deep learning, big data and powerful computing won’t solve all the problems, and while the industry as a whole is making progress, it still faces many challenges. Yang fan said the current challenges fall into two main categories:

The first challenge: The better you get at deep learning, the deeper the network, which means a lot of computing costs. How to optimize the structure of the network and its computational overhead while doing better identification and deeper network is a common problem faced by the whole artificial intelligence industry.



The second challenge: Today’s deep learning is based on a large amount of data. How to reduce the dependence on data needs to be done.

In addition, although China may be leading the world in the research and development and implementation of AI application technologies, there is still a huge gap between China and the United States in basic scientific research, which is also a challenge for China as a whole.
Yao Chi-chi, the first Asian winner of the Turing Prize, believes that the weakness of Chinese AI lies in systems and theory. If computer technology is to develop comprehensively, both engineering and theory must be considered.

Yang Fan also holds the same view, “China is still too focused on applied research, while basic research is to solve the most essential problem, why can deep neural network work? Why can machines learn better than humans? He’s going to think about it, I think

At that level, there is still too little done in China.”

How do you view your competitors?

Of course, sensetime faces plenty of competitors in addition to its technological challenges. In addition to startup algorithms like Megvii and Yitu, some big companies are also pushing into computer vision.

For example, Hikvision is currently the leader in the field of security, providing a complete set of hardware + software solutions, and Sensetime’s largest revenue comes from security. However, Yang fan believes that What Sensetime provides is the ability of platformization, and the benefits of this model are good openness and traceability with partners, while companies in different industries definitely need to do some work like this, and ultimately the results will speak.

However, BAT and other technology giants are positioning themselves as platform-based companies and providing AI capabilities including computer vision through their cloud platforms. How can Sensetime compete with these big companies in a differentiated way?

Yang Fan doesn’t seem worried.

In fact, we are quite different from them. From the United States to China, all high-tech companies are embracing AI and taking AI as their top strategy. I think this shows the importance of this issue without a doubt. However, any commercial company, especially one with the size of BAT, must serve its own business strategy and planning for the purpose of AI technology evolution and the AI technology solutions it provides.



What does shang Tang offer? Sensetime’s AI strategy is to provide solutions for enterprises in different industries to help them solve existing problems. In this regard, I think each BAT company has its own most important business and development direction, and their AI will give priority to solving these things. So at this point, I don’t think there’s a conflict.

But just as today’s Internet is almost monopolized by a few big companies, how many players will there be in computer vision? Will it end up winner take all, too?

Yang Fan told AI Tech Base,

Any market is ultimately the same, it is a long tail state, even the Internet, in fact, it is still a long tail state, the head (companies) must be very few, and then some long tail in some segments of the industry to survive, this is a high probability event.

In addition, AI is different from the Internet. AI itself is not an industry, nor is computer vision itself an industry. It is actually a combination of various industries, or an ecosystem.

How to compete with BAT for talent?

However, even if there is no conflict in some businesses, competition for talent is a problem sensetang has to face.

According to Yang, sensetime’s r&d staff now accounts for more than 70 percent of the company’s total staff, with more than 300 people, including 18 professors and more than 120 doctoral students from prestigious universities around the world.

However, due to the sudden popularity of AI in recent years, the training speed of related talents has not kept pace with the demand of the industry, and the salary of these AI talents has also increased. According to a report released by Tencent, the annual salary of 300,000 to 600,000 yuan for AI-related technical positions is basically the mainstream income level.

With AI talent in short supply and salaries soaring, how can Sensetime compete with giants like BAT for talent?

Offering a very competitive salary is only one thing, Yang said. For really top talent, they value the following two things:

First, whether the enterprise can give him the space and opportunity to show his ability. How to understand? You have hired a person or a team with a high salary, but the team wants to do one thing or put one thing into practice. The cycle is very long and the process is very complicated. For truly outstanding talents, this situation is very uncomfortable.



Second, talented people value concentration, which means they attract each other. He thinks I need to work with people of my own calibre so that I can be more valuable.

Moreover, Sensetime does not exclude talents who switch to AI. Yang Fan believes that from the perspective of product implementation, such talents are very valuable.

Is there a bubble?

Back to Sensetime’s $410 million raise, which was more than many expected at the time. But with AI startups so hot, it’s probably only a matter of time before the $410 million record is broken.

As of June 31, 2017, The total financing amount of Chinese AI companies reached 63.5 billion yuan, ranking second, accounting for 33.18% of the total financing amount of global AI companies, according to the 2017 Sino-US AI Venture Capital Status and Trend Research Report released by Tencent Research Institute. At the same time, data show that The investment rate of Chinese AI companies is 69 percent, exceeding the 51 percent of American companies, showing a trend of lagging behind.

So is this a sign of a bubble?

In Yang fan’s view, there is not only a bubble in AI entrepreneurship, but also a big bubble,
The main culprit of this bubble is too many unprofessional people.

So the question for us today is, you know, people want to have a little bit of a sense of what this is, and they want to do something about it, and I think this is actually the biggest bubble.I think at any time, anywhere, in today’s highly divided society, it must be the most professional people to do the most professional things.

However, every coin has two sides. The dot-com bubble from 1995 to 2001, for example, has left many valuable legacies for the mobile Internet boom. This round of AI bubble is obviously not all bad and all good.

Yang fan believes that in the next decade, AI will form a very big trend, and will change every corner of life more profoundly than most people think,
The advantage of bubbles is that more people feel the coming of such a big wave and are ready to accept it.