Not in first-tier cities, a group of AI companies founded by "mud-legged people" have made money, is this possible?
Looking at the industry, star companies in the front line are still difficult to make profits, and there is no mature experience in the circle to follow. Such an AI company can not help but make people curious.
Recently, a profitable AI company has entered the public eye -
Sixth Mirror, an AI company that originated in Xi'an.
In 2014, three students of Northwestern Polytechnical University decided to start a business, and the sixth mirror was born, and the team has reached more than 100 people and is valued at 700 million.
As an AI startup, although its external appearance is limited, in the industry, it has attracted attention for achieving profitability early.
What kind of commercialization path did they take? What is the significance of such an "atypical" sample for the industry?
Don't rush to answer, let's first peel back the cocoon, connect the dots into a line, and see what kind of company this is.
The post-90s generation piled up into the factory
Can you imagine?
A group of post-90s people concentrate on taking root in the front line of the factory, and their daily work is to tackle the research of production lines, equipment research and development and design.
In a rolling mill in Hebei, such a thing happened.
Different from the traditional positions, these post-90s are debugging a high-precision measurement and control system, with the goal of using multiple sensors and multiple algorithms to perform 2D+3D detection of hot rails.
In the past, the traditional production line of rolling steel relied on manual handheld equipment for sampling, and the inspection process required waiting for high-temperature steel to cool, which took 10-20 minutes. This leads to problems such as low measurement accuracy, missed inspection and detection lag, which can easily lead to batch quality accidents and even production safety accidents.
This long product inspection system can automatically realize real-time defect detection and contour measurement in environments up to 1400 °C and rail operating speeds of up to 6 m/s.
Compared with the traditional method, their measurement accuracy reaches ±0.1mm, and greatly reduces the probability of missed inspection accidents, improves efficiency, and saves about 5 million costs per year on a single production line, which is said to be the first in China.
This batch of post-90s has also entered the photovoltaic power plant.
At the power farm, they control the drone equipped with thermal imaging cameras and visible cameras, identify photovoltaic panel defects through CV + AI technology in the air, screen the upper heat spots, and prevent the components in the shadows from becoming loads, consuming energy, and excessive heating and detracting battery life.
These young people even went to coal mines and became safety officers.
More precisely, they taught the "camera + system" to identify illegal operations and behaviors, through computer vision technology, to identify people sitting on belts, not wearing hard hats, illegal chasing monkey cars (a ropeway down the mine, similar to a ski resort single-person standing lift) and other unsafe behaviors.
There are many similar scene cases, and their figures also cover many industrial fields such as coal, steel, and energy.
In fact, the rare scene of this group of post-90s people getting together in the factory stems from an accidental factory investigation.
At that time, they found that although a heavy industry production line was highly automated, it was not completely autonomous and controllable.
The core equipment and industrial control software systems almost all come from foreign manufacturers, which has caused a lot of trouble for the digital transformation of factories, the industrial Internet and the Internet of Things.
Coupled with various external factors such as large industry span and large information gap, domestic emerging AI companies rarely intersect with traditional industries, and many needs are difficult to be concerned. The post-90s of the sixth mirror are one of the few teams to see the demand.
The team only circled the production line and sorted out more than 30 needs.
Now, after two years of exploration, they have covered steel, tobacco, coal mines, roads and bridges and other scenarios in the industrial AI sector, covering hardware equipment, algorithms, solutions and other products.
In the future, companies say that industries such as sewage treatment, petrochemicals, and rubber are all within their consideration.
In recent years, the value of industrial AI has gradually been recognized by the industry, and various giant players are also laying out. Because of the attention and continuous creation of value in advance, some time ago, several traditional enterprises signed strategic cooperation with the sixth mirror.
Among them, there are veterans of the foundry industry, emerging casting pipes, there are also local leaders such as Hegang Digital and Hebei Construction Investment, and the 55th Research Institute of China Electronics Technology Group Corporation...
Coal coke chemical head enterprise, Xuyang Group also officially announced the sixth mirror of strategic investment.
"Touched a nose of ash" at the beginning
The road to entrepreneurship has never been smooth, in fact, this group of young people has not been easy to walk at the beginning, especially when they first landed in the scene, they "touched a nose of ash".
Urban AI is their preferred business, that is, to provide "people-vehicle-things" recognition algorithms and applications for pan-park scenarios, including parks, schools, communities, airports, etc.
In 2017, they entered the community with face recognition technology that had just broken the world record (LFW/FDDB) and used the leading face recognition algorithm to do smart community applications, but they did not expect the problem to erupt and catch the team off guard.
Simply put, the algorithm model does not match the scene. This is also a common problem encountered by many enterprises when landing.
As we all know, the datasets used by laboratories to train algorithms are basically open source or clear media, film and video public images.
Although this kind of algorithm has considerable accuracy in the ideal environment, it is applied in real life, encountering light environments such as backlight/night/overexposure, or dust dry spots caused by bad weather, and even the most common "brush face" may encounter dynamic blur of images, unsatisfactory clarity, etc., resulting in a great reduction in effect.
In addition, there are recognition speed, use costs, these problems that have not been considered in the laboratory, all have to be solved in the real scene.
The deployment of the entire solution was originally planned to be completed in one to two weeks, but it was postponed to three months. During this period, the team and the scene manager worked together to solve the pain points one by one before completing the project.
It is also because of this experience that they realize that the scene and the manager of the scene are the best teachers, which has become their consistent style since then.
After a brief period of pain, they ushered in the first wave of victories -
At the end of the year (2017) when the all-round urban AI was landed, the company achieved breakeven.
Nowadays, the Sixth Mirror City AI has built a complete set of product systems for algorithm tools- algorithms- hardware-software-application platforms-solutions, and the solution product line covers public rental housing, schools, parks, communities, police and other scenarios.
In terms of hardware equipment alone, it has formed a product matrix including face recognition all-in-one machines, intelligent surveillance cameras, face brush payment terminals, and intelligent visitor machines.
Even so, the market challenges the team faced were still not small.
On the one hand, urban AI is far ahead in the industrial share of the mainland artificial intelligence market, accounting for up to 49%, and unlike the systematic and integrated business settings of smart cities, urban AI is a single-point breakthrough -
Providing customized services for scenarios is a feature that drives more technology companies to get a piece of the pie.
△ iResearch "Chinese Intelligent Industry Research Report (2020)"
On the other hand, AI is gradually evolving into a basic capability, and the technical threshold is constantly decreasing.
In the face of these situations, there are undoubtedly two ways to solve it, in addition to continuing to dig up the surplus value of urban AI, it is to find another way - to find the company's second growth curve. So, the sixth mirror aims at industrial AI.
Self-proclaimed AI "mud legs"
Seven years ago, Liu Chuang, a junior in the Department of Automation at Northwestern Polytechnical University, and two other students decided to start a business because of a pattern recognition class.
In the early years, due to insufficient funds, the company once relied on outsourcing projects such as website development to support the AI research and development team. Team members are also accustomed to describing themselves in a slightly unpretentious term:
Mud legs.
This statement is simple and appropriate, and it is not unrelated to their dead scenes and deep digging into industrial needs. Looking back now, the team has come up with its own methodology.
In a word, ai algorithm production platform + high standards landed.
As a technology company, the strength of the algorithm lies not in a single product, but in the ability to produce the algorithm at scale and in batches.
For the sixth mirror, it is the AI algorithm production platform.
The platform includes a deep learning training framework, a distributed end-edge cloud AI collaboration framework, a distributed decision-making and reinforcement learning toolchain, and a tool library of AI basic algorithms.
According to reports, in order to cope with the diversified needs of urban AI and industrial AI, the platform is oriented to customized scenarios, which can achieve rapid response, from data cleaning and labeling, model training to SDK package delivery, the whole process without excessive human intervention.
The AI wave, seen as the intelligent wave, is a revolution that replaces simple repetitive mental work. And this brain power comes from algorithms.
But light has a brain, but also needs to have a "torso" coordination. This torso, in the sixth mirror' view, is the landing methodology.
Specifically, it is to find the highest demand standards and develop modular solutions. Later, when sinking into a similar scene, the module is split.
Visualize understanding, the business from 0 to 1 time to do addition, and then broaden to 100 time to do subtraction.
For example, for the wisdom service of colleges and universities, first find a double first-class 985 university, because the requirements of scene managers must be the highest.
In this way, with the "brain" and "torso", we can better direct the rapid landing of the two major scenes.
In addition, in the face of the most common industrial problem in the industry, many AI companies folded the industrial small sample problem, the sixth mirror derived its own landing ideas.
Generally speaking, to ensure a high yield rate in actual production, it is necessary for AI models to have higher accuracy. With traditional deep learning features, this requires large-scale data to train.
This is difficult to achieve in industrial production.
Because often the more fatal a defective sample, the lower the probability of occurrence, the less sample data available for training, coupled with the long tail effect, more unknown scenarios can not be taken into account.
The solutions that exist in the market are nothing more than the optimization of algorithms and data, such as transfer learning and data gain.
On this basis, the sixth mirror also chooses to hit the "point" (scene) breakthrough to connect the upstream and downstream.
The technician explained that since the problem of small industrial samples is widespread, it is necessary to optimize the technology on a point (scene) basis and work with researchers of upstream sensors and downstream industrial software to solve the problem.
At present, the team has cooperated with a number of enterprises and universities. For example, the joint laboratory of industrial software and intelligent manufacturing was jointly built with the School of Mechanical and Electrical Engineering of Xidian University, and the "AI+Steel" Machine Vision Innovation Center was jointly built with Hegang Digital Technology Co., Ltd. and the School of Automation of Northwestern Polytechnical University.
The reaction of all parties has hinted that as long as AI companies are seen by the outside world as long as they have the strength and attitude of landing, they do not have to worry about the accumulation of customer value.
Similar cases are very common in the AI industry, the landing path is different, and even as the business gradually deepens to the industry, but everything is one, the road is simple, and the ultimate attribution is only one - the object of AI service.
Second-generation AI companies came to center stage
The sixth mirror is not a "typical" AI company, not only profitable.
This is a team that is not too focused on cutting-edge algorithms, papers, and patents, and accuracy is no longer the only criterion: the technology that the scene uses well and the scene manager uses well is the technology that the scene manager uses well.
This team does not appear much, and the technology is not top-notch、...... This also indirectly shows that AI has long been not as high as it was at the beginning, and more enterprises and technicians are involved to create industrial value.
More importantly, they are no longer looking for nails with hammers at the beginning, but quickly landing according to specific scenes.
From the obscure "mud legs" to walking to the front of the stage and entering the public vision, the sixth mirror is not only the epitome of the tide of AI empowerment, but also a typical example of starting from demand and creating value.
With the development of the industry, such second-generation AI companies have gradually crossed the original accumulation and come to the center of the historical stage.
From the perspective of traditional industry, the lack of young talents has always been a long-term accumulation of industrial diseases, and it has become more and more intense with the intensification of aging.
There was even a workshop worker who half-joked with the research team:
I don't know when I'll retire, but there are now some positions that no one has taken over.
Factories can't wait for new blood, how can industry talk about entering the core vision of science and technology? Fortunately, this need is receiving increasing attention.
A variety of cutting-edge technology companies and technicians are looking at traditional industries.
Accordingly, a series of cutting-edge products such as industrial-grade AR glasses of the same model as "out of control players", operating systems on coal mines, and safety helmets developed for workers to monitor brain waves have been developed.
At the same time, AI emerging companies represented by the sixth mirror are using business to link factories with young people, so that more young technicians can devote themselves to industrial data transformation.
Responding to the changes in the global market with domestic substitution is another value embodiment.
These cases have shown that with the wave of technology moving forward, the acceleration of industrial digitalization and intelligentization is the direction of the road.
The sixth mirror is just a microcosm, but it also represents a definite trend.