September 5, Beijing, sunny.
This day was the first weekend when Nolan's new film "Credo" went online. Everyone walked out of the cinema and seemed unable to speak well. They all made friends:! The second brush needs to be the first one. The brain is too hot
"Too brain burning" - the same voice echoed in a venue in the northwest of Beijing. More than 80 students of Baidu Whampoa College are experiencing a test that may be more challenging than the undergraduate graduation defense.
In a conference room, there were several serious judges, many of whom were outstanding architects and scientists of Baidu. Around them were a dozen students of all ages from all industries. Everyone listened attentively to the students who were defending their graduation projects with PPT.
These topics are the specific problems and puzzles they brought from the industry. After three months of courses, discussions, and practice, the corresponding AI technology solutions were finally presented, drawing a good end to the weekly hard courses.
They may not be able to watch the latest movie and play the hottest topic at the first time, but in this special 2020, they will become a trendsetter standing on the "singularity of time" of industrial intelligence landing.
Textbooks cannot grow AI, and the talent dilemma of industrial intelligence
The word "time singularity" sounds cool, but the process of AI landing industry is not cool at all, and it is not full of cyberpunk style sci-fi color as the public guess. The students of the third phase of Whampoa College felt this very deeply.
Most of them come from traditional industries. Some have more than ten years of experience as IT engineers, and some have a general understanding of AI. However, no public class or popular science article can tell them what problems AI can help their enterprises and businesses solve, how to solve them, and to what extent.
To some extent, this also reflects the practical difficulties of AI landing:
First of all, there are few practical cases and insufficient communication in the industry. Not only are traditional SMEs eager to learn, but many CTOs or technical directors of large state-owned enterprises are also eager to get practical and detailed guidance in intelligent transformation. However, in practice, most enterprises can only purchase the entire system or general algorithms from integrators or platforms, and the similarities in capabilities will encounter "acclimatization" when landing in specific scenarios.
Secondly, the platform and tools are scattered, the technical system is complex, various open classes, dry goods articles and salon sharing are rough and fragmented, and developers have high learning costs, which leads to low efficiency of AI project promotion;
Compared with the glittering academic giants, the ability requirements of the industrial intelligence landing stage are engineering talents who can transform needs into problems and find the best cost-effective solution. As early as June, Nvidia engineer Chip Huyen made his own conclusion about the trend of layoffs in machine learning posts, that is, after having engineering related knowledge background, it would be more promising to contact machine learning than to contact machine learning directly. So, how can these engineering capable people start shaking hands with machine learning?
The answer may be a set of systematic, structured and customized courses, an educational mechanism that can ensure learning effects and practical output. Of course, it is also essential to create a good environment to achieve interaction between a large number of (potential) practitioners, and to analyze and discuss cutting-edge cases.
These are the things Baidu Whampoa College has done for three periods.
Three month time curse,
Technical pursuit of 81 industrial talents
12 courses, three hours of intensive input each time, three times of expert guidance (in fact, experts have provided as many as 12 times of guidance for students), offline homecoming, industry gatherings, etc., all of which were acquired by the third phase of Baidu Whampoa College students in three months.
The support resources mobilized by Baidu are also obvious to all. Both the passionate guidance of the chief architect and scientists, as well as the meticulous assistance of experts and teaching assistants, have been selflessly poured on these future "chief AI architects".
Why did you do it? Why are the resources of Whampoa College increasing?
A class teacher of the third course told me that the student group of Whampoa College began to change from the first to the third period, and the proportion of technicians in traditional enterprises began to increase significantly. However, these students and enterprises have different requirements for AI implementation. What can and cannot be done by AI, where are the limitations, and how can data be used? Professionals are needed to help sort out, popularize and guide actual combat.
Therefore, from Phase I to Phase III, Whampoa College has gradually found a clearer path.
Since many traditional enterprises do not know the boundaries of AI and what AI can do, they will gather Baidu's own scientists and "senior engineers" to guide them hand by hand. Some students shared that during the first expert guidance, the students thought that what they wanted to do was very simple, but actually the technical implementation was very complex. Baidu senior engineering instructors would sort out for them what technical methods could be used to solve his business problems, what was the general technical idea and feasibility, and who to turn to if they encountered problems. Such guidance is time-consuming and laborious. In fact, the number of internal teachers mobilized by the third course of Whampoa College is unprecedented.
Since there are few industrial exchanges, it is difficult for technicians to find peers, so we should build an AI architect exchange platform for them. With the help of homecoming, online interaction and other forms, students can finally exchange what they need with peers in a digital space, and cultivate "classmate" friendship to deepen exchanges between different industries and applications;
Since there are few practical cases and lack of practical experience, we will use the combination of "devil like" high-intensity theory+industrial projects to enable students to quickly apply AI to the problems faced by their own businesses, and call the full stack AI capabilities of flying oars to create practical cases of industrialization. For example, a commercial aircraft manufacturer introduced the semantic segmentation algorithm of flying oars into the process of raw material defect detection. At present, it has received good feedback in actual production, and the recognition accuracy rate is more than 95%.
Based on this, students of Whampoa College can also have the ability of "chief AI architect". They are compound talents who understand both application scenarios and AI technology, can apply AI technology and solve practical problems in the scenarios during the industrialization process.
And with their graduation, a large number of industrial AI seeds will sprout from all over the world.
Insight into the future, the counter time code given by Whampoa College
From this perspective, we can also think about what kind of AI we need in the context of the international environment in the year of the epidemic?
In fact, the answer is almost out, that is, autonomous and controllable AI, AI that catalyzes the digital intelligence transformation of the industry, AI that has a high degree of landing and a long life cycle
When these appeals are combined, it is not so simple. Whampoa College is also in the process of talent cultivation, delivering various accumulation of flying oars to the industry:
1. Full stack AI capability. As an operating system in the AI era, the Flying Propeller integrates the underlying tools, hardware and software and technical modules required in the AI era, becoming the hub connecting industry and AI, and meeting the application needs at different stages. In the words of a student, "a novice can be taken care of by a nanny, while a person with some experience can get a powerful weapon". In reality, this feature has also triggered more and more students and enterprises to start moving from overseas in-depth learning framework to flying oars.
2. Close to industrial demand. If it is just a little spillover of technical capabilities, it is certainly not enough to attract developers to start complex code relocation projects. As a deep learning framework more in line with China's national conditions, flying oars can bring unique value to developers, which may also be the key to ecological growth. It can be clearly seen from the distribution of students in Huangpu College that AI is rapidly moving from the Internet field to more and more traditional industry enterprises, and their application scenarios of AI are different and varied. Baidu AI's application case has dispelled many enterprises' concerns.
3. Benign development ecology. From the differences between Whampoa College and other talent training programs, we can also summarize a better way to open AI platform ecology: systematization, high quality, and sustainability. Instead of pursuing "quick success", we use daily and weekly high-frequency exchanges to establish a lasting and in-depth student relationship, so that they will no longer stay in training, but become "vocational colleges" to incubate high-end talent circles in the industry. The relationship between many students and enterprises and Baidu has gone beyond their original ideas at the beginning of enrollment, and has generated deeper and broader cooperation in reality.
For example, during the course, a student discovered many fields that can apply AI through internal communication in the enterprise, such as using machine vision to assist the brakes of aircraft, and intelligent quality inspection to ensure production links. With this opportunity, Baidu's technical team also carried out in-depth communication with the enterprise's engineering team, For some technical problems at the practical level, such as how to realize machine vision when the screws are tightened to 6 360 degrees. To solve this problem, Baidu's vision technology department, robot and autonomous driving laboratory, deep learning technology platform department, big data laboratory and other research and development departments have all made suggestions to coordinate input of AI solutions that can meet the needs of the industrial end.
On the one hand, this kind of communication comes from the students' sincere sense of identity and mutual trust in Baidu Flyer's technical system in Huangpu College, and also from Baidu's "simple and reliable" engineer culture.
Similar industrial problems are still emerging, which also makes industrial intelligence a time node that must move from concept to pragmatism, and from algorithm to engineering.
As a lecturer of Whampoa University said, the implementation of in-depth learning is not a technical problem, and it usually tests not the algorithm ability, but the selection of a complete set of project implementation schemes.
Today, AI does not need to prove itself again and again, but how to make all industries run on the AI highway? Is it to create new "wheels" for each industry and enterprise, and then go on teaching one by one? Of course, it is necessary to throw qualified AI architects into various fields, so that they can really explore, integrate and test in-depth learning technology and industry needs, and ignite enthusiasm of one party.
Then the question arises again. What kind of ability does such an AI architect need? What Whampoa College has done is to define the ability, build weapons, and support energy for the "chief AI architect", open the door between AI and industry, and create an ecological advantage of travel alienation for Baidu AI to B.
The concept of reverse time mentioned by Nolan in "Credo" refers to that the universe starts to collapse from expansion at a certain time singularity. Therefore, time reversed and mountains and rivers flowed backwards.
And industrial intelligence itself, like the opening of counter time, may be difficult for people to observe the changes brought by it at the beginning. However, the world will eventually be changed through the small displacement of each person and step. The value provided by Baidu Whampoa College may be to let everyone who is eager to devote themselves to AI and use AI have partners, identify the direction, look at the lighthouse and drive into the future.