Shulou(Shulou.com)06/02 Report--

Last week, deep learning was basically equivalent to New Year!

On March 27, the 2018 Turing Prize was officially announced: Yoshua Bengio, Geoffrey Hinton and Yann LeCun, the three giants of deep learning, won the Nobel Prize in this computer field.

Not only does the field of artificial intelligence have glory, but it also makes the profession of deep learning algorithm engineer attract much attention.

In industry, deep learning has become the most "money scene" direction in recent years. Google, Microsoft, Baidu, Tencent and other big tech companies are spending a lot of money on recruiting, and deep learning engineers are worth $243000 a year (from Glassdoor2018 data) to lead the industry.

Both academia and industry have offered olive branches to deep learning. But how do you quickly step into the threshold of deep learning engineers? I think everyone has the same confusion as us.

As a professional technology self-media, how can we not take the lead!

Coincides with the first deep learning engineer professional certification examination officially opened, so we quickly sent our best mathematics myself, to find out (self-humiliation), in the gongs and drums, firecrackers in the happy atmosphere, for everyone to bring the deep learning engineer professional qualification examination first-hand battle report.

First, let's talk about the overall feeling:

From the point of view of a liberal arts student + industry observer, the difficulty of in-depth learning of engineers at the primary level is not high. More emphasis is placed on the investigation of basic knowledge and comprehensive skills. As long as a small amount of time is taken to prepare for the examination at ordinary times, it is not a problem to get certificates.

All Chinese textbooks and examination questions, theory + practice combination of investigation methods, for Chinese developers and examinees are also quite friendly;

However, if you want to really be able to be independent in algorithm development and model application, and complete the AI projects needed by the enterprise community, you still need to challenge the intermediate and advanced levels. In other words, beginner is just an "appetizer" for beginners, recognition in the professional market is not clear enough, it is unrealistic to rely on it to reach the peak of engineer life.

Deep learning does not need a certificate escort? Engineer certification is not worth testing? How to test again? This immersive experience report will tell you the answer. Let's Go!

To be a decent technology media practitioner, start with plucking up the courage to go to the examination room

Here's the story. I may have been AI head down, suddenly a faint head, what do not know, and so come back to God on the computer more than a pass...

Okay, it's bullshit. Although I am a liberal arts student, I have dealt with AI for so long. It is not a problem to understand a thesis and analyze a model. Help students who want to show their skills in the field of deep learning to experience how difficult it is to take exams in advance.

So I confidently walked into the exam room on the last weekend of March with a (guilty) smile and good wishes, and proudly became a "mouse" for the first deep learning engineer certification exam in human history.

First explain the preparation work needed before the exam.

The Deep Learning Engineer certification was launched by Baidu Institute of Technology in partnership with Pearson VUE Test Center. After purchasing the test ticket, log on Pearson VUE official website. After successful registration, reserve the examination room and test time by yourself. So easy~

In order to create a sophisticated academic atmosphere and sense of ceremony for myself, I chose the examination point of Wudaokou in the center of the universe.

Formal examination:

As a liberal arts student, I have basically given up treatment. On the day of the exam, I had nothing, put on my spiritual pet "mobile phone," bring my beloved ID card, so low-skilled people also dare to come!

P.S. There were a lot of exam items that day, and the world of programmers was really bald.

In my "magic calculation," brave and resourceful arrangement, really squatted outside the examination room to the same door examination examinee senior. The registration clerk at the front desk also told me that he scored 70 on the test. I think the mood of learning bully should be good, so I secretly rub up to get scriptures.

He never expected that Brainiac was a graduate student in artificial intelligence at Beihang University. There was an entire class of people who came to take the exam with him! Sure enough, when I asked which questions felt more difficult, he calmly replied to me: I think they are OK! Some knowledge is a bit vague! There was confidence in his words that he knew everything…Alright.

Although I am not on the same level as professional players, I am still relieved to know that I don't have to write code on the spot. When we have so many years of "three long and one short choose a short, three short and one long choose a long, really won't choose C" test skills are abandoned? I'll let heaven know I don't give up!

Finally to the appointment of the examination time, I took the identity card and fans of self-confidence into the examination room!

Before entering, I imagined that the examination room should be like this:

At worst, it was like this:

The real test is like this:

Emmmm, it doesn't look very intelligent. Of course, the unique experience comes later.

The staff helped me open the system and left. After confirming the information and reading the exam reminder, I could start answering questions. Don't fantasize about on-site classmates and off-site spectators,"Big Brother" watching you!

From an outside perspective:

There are 50 multiple-choice questions, each with 2 points. There will be a 90-minute countdown in the upper right corner.

Some questions are still relatively handy, such as in which areas collaborative filtering is applied, which algorithms have smaller variances, etc., and we are still relatively confident.

However, when I encounter some problems involving calculus of tuning parameters, I will say frankly that they are almost like this in my eyes:

Please select one type of

A,

B.

C.

D.

In the face of this hell-level question, I can only release the big move of "all choose C".

After 50 minutes of desperate struggle, I finally completed the answer and blind, frankly chose "submit the test." Unexpectedly, the system gave points on the spot after the exam, leaving no buffer for the students. So I learned that I, naturally, had fallen on the street and got a 44-point "death report card."

Out of the examination room, behind me fell a floor, not as white as a lamp draft paper, but also my broken AI engineer dream o()o.

After signing my name at the front desk, I got my "failing subject" report card and left the quiet science park. In the strong wind of early spring in the north, I watched the bustling traffic at Wudaokou and couldn't help falling into the myth of philosophy_

Who am I? Where am I? Why do I have to do this?

So, what exactly does a deep learning engineer test?

Even though I had successfully thrown myself into the street, the mood of the other students at the scene was still quite high. There were more than 30 people in the entire class of Beihang artificial intelligence major in the same exam as me. It seemed that deep learning and Baidu, the two artificial intelligence gold-lettered signboards, were still relatively appealing.

After the experience, we can understand the training situation and professional needs of deep learning, which is our ultimate goal.

So, what skills do deep learning engineers examine? What do practitioners think of professional certification?

Combined with the examination experience, simply split the skill tree of deep learning engineers:

Personal feeling, elementary corresponding basic theory and knowledge is relatively wide, more suitable for school students and graduates one or two years of engineers and developers to apply. Because it was still at the peak of basic knowledge of mathematics, programming, concepts, etc.

At the engineering level, the main language covered by the exam is Python, which is also in line with the actual needs of mainstream deep learning frameworks. Among them, practical operation requirements are also put forward for paddlepaddle platform development capability.

At present, it seems that industry-oriented deep learning junior engineers, the portrait is probably like this: have certain basic knowledge of advanced mathematics, understand basic machine learning theory, can combine Chinese development framework for simple model development and actual combat.

So, can current students and entry-level developers meet this requirement?

Through a round of internal communication into the examinee, the more intuitive feeling is that even professional students with software engineering background are faced with an awkward skill training environment:

First of all, the talent training system of colleges and universities emphasizes theory, English and low actual combat, which is out of line with the actual industrial demand.

There is examinee feedback, although professional learning has been exposed to deep learning development and other projects, but limited by GPU resources, data scale and so on, almost rarely really participate in practice, naturally it is difficult to master programming ability skillfully.

In addition, due to the first-mover advantage of AI ecological layout in the English world, it is obvious that most of the teaching, development and programming of deep learning are implemented on English frameworks such as Tensorflow and PyTorch. This creates two problems: one is that there is insufficient support for Chinese enterprises and development needs, and the other is that the security of technology and talent pool are facing certain risks. Chinese platform and certification standards still need to continue to exert efforts, Baidu deep learning engineer certification system building, this can really play a certain role in boosting.

More importantly, the current training and certification of skills for artificial intelligence and deep learning have steep learning curves.

Many technology companies, including Google, Apple, Alibaba, Tencent, Baidu and Huawei, regularly hold technical training and salon sharing internally, such as Huawei University, Ali Internal Training and Lakeside College. But only for a handful of top programmers. The tools provided are also "from 1 to 100" boosters.

Someone once joked about Tensorflow's learning process: start learning, stick with it, and then you succeed!

For most ordinary people who are in the difficult entry and rise period, there is currently no complete socially recognized vocational skill development path. But predictably, as artificial learning neural networks become the infrastructure of society, we will inevitably need more and more deep learning engineers to escort them.

Throughout the whole industry, Baidu deep learning engineer standard and certification pyramid structure is relatively complete, providing learning resources from basic science popularization to AI studio development actual combat platform, and then to paddlepaddle professional development platform, as the forerunner of deep learning vocational education, it can be seen that it is more sincere.

From this point of view, vocational certification forces AI education in colleges and universities to check gaps, or can play a positive role. Of course, this is also closely related to Baidu's own comprehensive layout in the artificial intelligence industry, the open mind of AI technology and its academic leading ability.

Specific to the engineer examination, from the perspective of the whole industry, the gold content of a single certificate remains to be examined.

There are students who have passed the exam on the spot and throw out relevant questions: "I don't know what this certificate can do.""Will there be extra points in the interview?" "。

At present, the exact news is that Baidu has added deep learning certification to the talent recruitment system, and interviewers who get certificates will be given priority.

What Baidu needs to do next may not only be to make in-depth learning "more suitable for Chinese physique" and "Baidu recognition," but also to carry out a series of planning, so that professional standards can further sink to the industry, and connect the real talent demand with the certification system.

On the whole, with a 500 yuan as the entrance, step into the door of deep learning industry, students who have spare power can try, if expect to earn a high salary by a paper certificate, may also seek more sharpening.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.