It's never too late to be an AI engineer

Since the beginning of the year, a number of major international developers of the Conference, whether it is Microsoft Build, Facebook F8 or later Google I/O, the "AI first" banner to the sky.

If this wave of AI is just sloganeering a few slogans, empty to mention a few strategies, there are several hot start-up companies, it is certainly not a climate. But under the wind and waves, what we see is that Google's line of major businesses have switched to deep learning, outdated mobile era of Microsoft has pulled up a nearly million people AI team. But the situation of the domestic first-tier manufacturers, I am afraid is similar.

This is a sign that is worth focusing on for programmers, especially for prospective programmers who will be on the verge of technology. Looking back at the opportunities brought by the mobile Internet, it is easy to debt settled and master how deep learning can bring advantages to the first-line job. However, unlike mobile development, demanding mathematical thresholds and costly combat training, the training cycle of AI talent has soared to more than 5 years ... It seems that there is no Master's and Doctor's knowledge reserve, we must say goodbye to AI-related technical work. Is the truth true?

This article integrates the AI Engineer Career Guide series, which starts with programmers, and the authors analyze how qualified engineers from various technical positions in the AI field are refined from a practical point of view. You will learn what kind of AI skill tree can meet the needs of their first-line business, the data science, machine learning algorithms, heterogeneous parallel computing and speech recognition, recommendation systems and dialogue systems and other fields of skill in the development of how to expand, especially among them the school path and the actual practice of how to choose.

Click the blue text below to read the article how to become a machine learning algorithm engineer (Zhang, the head of recommended algorithms)

Becoming a qualified development engineer is not a simple matter, and requires a range of competencies, from development to commissioning to optimization, each of which requires sufficient effort and experience to master. To become a qualified machine Learning algorithm engineer (hereinafter referred to as algorithmic engineer) is even more difficult, because in mastering the general skills of engineers, but also need to master a small machine learning algorithm knowledge network. This article will be a qualified algorithm engineer for the skills required to split, together to see what needs to master what skills to be considered a qualified algorithm engineer. Recommended system engineer (Chen Kaijiang, CTO of Technology)

Now, despite the "big Data", "AI", these words 360 degrees a day no corner bombing us, let us easily impetuous abnormal anxiety, but we have to admit that this is as a recommended system engineer a good time. Recommended system engineer and normal code, compared to the need to throw the PM to the pixel-level implementation, so heap code Chengshan; compared with the machine learning researcher, there is no need to indulge in mathematical deduction, suppress a beautiful self-consistent model, unified academic controversy; compared with data analysts, there is no need to draw beautiful charts, Make a cool PPT to report to the CEO, to the pinnacle of life. What is the positioning of the recommended system engineer? Why do you need the skills mentioned earlier? The author will combine their own experience to answer each. Dialogue Systems Engineer (Wu Jinlong, partner of Einstein Interactive Technology)

The dialogue System (dialogue robot) is essentially a machine-learning and artificial intelligence technology that allows machines to understand human language. It contains the integration of many disciplines and methods, is a technical focus in the field of artificial intelligence training camp. With the development of speech recognition, NLP and other technologies, the stage of dialogue robot will be bigger and larger with the advent of the Internet of everything. Data scientist (Forest Club, American DuPont Business data scientist)

Before you answer this question, I want you to think about another question: why you should be a data scientist. Of course, if you're looking for a $100,000 annual salary, I really hope that you will be able to relate this career to your own sense of worth. Because it's hard to be a data scientist, but if you look at it as a way to realize your personal value, pursuing your goals can lead to a lasting sense of accomplishment, and you'll be happy and motivated in the process. Data scientists this position is still relatively new, so from the team building and career trajectory are still developing, with good prospects. I hope you can become a data scientist who keeps thinking and learning for life. Heterogeneous Parallel Computing Engineer (Liu Wenzhi, head of High Performance computing division, Shang Tang Technology)

With the hot of deep learning (artificial intelligence), heterogeneous parallel computing has been paid more and more attention by the industry. From the beginning of deep learning must talk about GPU, to talk about deep learning must talk about computational power. The computational force is not only related to the specific hardware, but also to the level (i.e. heterogeneous parallel computing power) possessed by the person who can exert the hardware capability. A simple analogy is: the two-chip computing power is 10T and 20T, someone's heterogeneous parallel computing capacity of 0.8, he got the computing power of 10T chip, and the heterogeneous parallel computing capacity of 0.4 of the people to get the power of the calculation of the 20T chip, and in fact the final result of two people may not be small. The ability of a person with a strong heterogeneous parallel computing ability can be better played, and the goal of this article is to tell the reader what knowledge to learn from becoming a heterogeneous parallel computing power engineer. Speech Recognition Engineer (Chen Xiaoliang, founder of Sound and Wisdom technology)

At present, the accuracy and speed of speech recognition depends on the actual application environment, in quiet environment, standard accent, common vocabulary speech recognition rate has more than 95%, fully reached the available state, this is the current speech recognition is more fiery reason. Academia discusses a lot of speech recognition technology trends, there are two ideas is very noteworthy, one is the end-to-end speech recognition system, the other is G.E. Hinton recently proposed capsule theory, Hinton capsule theory academic controversy is still relatively large, whether in the field of speech recognition can be reflected in the advantage is worth exploring. This article takes the science popularization as the main, will connect the knowledge vertically and horizontally, and can combine the practice to understand the article, for the comprehensive understanding speech recognition is very helpful. Technical breakthroughs: A professional path for deep learning (Melody, chief executive officer, Christie)

Deep learning is essentially a deep artificial neural network, it is not an isolated technology, but a combination of mathematics, statistical machine learning, computer science and artificial neural networks, and many other fields. The understanding of deep learning can not be separated from the most basic mathematical analysis (advanced mathematics), linear algebra, probability theory and convex optimization in undergraduate mathematics, and the mastery of deep learning technology is inseparable from the hands-on practice of programming as the core. No solid mathematical and computer Foundation to do support, deep learning technology breakthrough can only be in the castle. Therefore, it is necessary to understand the significance of these basic knowledge in deep learning for beginners who want to be successful in deep learning technology. In addition, our professional path will also introduce deep learning from the theoretical dimension of structure and optimization, and analyze the advanced path based on the practice of deep learning framework. This article will also share practical experience in deep learning and experience in acquiring cutting-edge information in depth learning. Actual combat path: Programmer's Advanced Method of machine learning (Lulang software co-founder)

If we are not able to sniff the opportunity of machine learning in school, we choose other fields to study and work ... Now is going to halfway decent, diverted machine learning, what should be done to be as good as these people. Or, at least, is good enough. The author's experience of painful transformation, say for everyone to reference.