I am fortunate to participate in the AI pre-research group of my department. It has been almost half a year from semi-involvement in the beginning (taking into account the project while pre-studying and learning) to full involvement in the later period (only in charge of pre-studying and learning). The following is a summary of the challenges that an ordinary programmer will face when he changes his career to AI.

First of all, I would like to introduce my background, which is not 985,211, which can be said to be very low. I majored in software engineering in 2017. I am not a student with outstanding academic performance, but with general professional knowledge, mainly engaged in JAVA development, and then changed my career to computer vision.

Then I started to talk about the challenges I faced in moving into AI.

  • Because of the boss’s high requirements for English ability, we could only read English papers. For me, I only passed CET-4 and hardly read English literature in college. It was really difficult at the beginning. But this really is the threshold of the entry level, because the AI industry in recent years, the rapid development of the level gap year in ordinary in the IT industry can at least three or four years, so must be new, English reading ability must possess, and not just chasing the latest, before the old paper still have to see, because in the new paper reference old knowledge are direct reference papers, So you will read a lot of English papers.

  • Based machine learning algorithm, the basis of deep learning algorithm is really very important, I have seen some online teaching video, simply Wu En of machine learning, CS231n of computer vision, etc., behind the implementation model, adjust the parameter of time will find their own poor background, almost can only follow paper tuning parameter, oneself besides randomness, No decision making adjustments can be made. So if you want to change careers, the algorithmic foundation is really, really important, don’t rush.

  • Mathematical basis That is a really gives me a headache, if we can only do some machine learning and deep learning related work, I feel the demand of the mathematics is not much more special, but if you want to continue in-depth, migration of reinforcement learning, or learning, tutorial there will be a lot of mathematical deduction, is really a headache. In a team, we can also urge learning, if a person self-study, facing several pages of math formula, math foundation is not very good.

  • Programming skills Programming skills are still very important for an algorithm engineer, because in addition to writing models and tuning parameters, you often need to do engineering, unless you are a very good algorithm engineer and don’t need to do engineering. For the direction of the algorithm programmer, must not lose your programming ability, free or can write more code, and then on the Internet often see a lot of non-IT professional algorithms, such as mathematics, strongly suggest more solid programming ability, because all big factories almost will have algorithm pen questions.

  • In fact, the actual combat experience of the algorithm industry is also very large requirements for actual combat experience, for example, RECENTLY I want to achieve a paper model, the theoretical knowledge of the paper looks very simple, so I am confident to masturbate the model, and then found that how to masturbate is not quite right. Finally compared to other people’s source code, in fact, I missed a lot of small details, such as some data pretreatment what, these in the paper will not tell you, you need is experience, the industry shared experience, but I did not, so as an algorithm engineer, theory is important, but can not be divorced from the actual combat.

  • Degree required This is an issue I recently learned, is really eat degree of the industry, I tentatively for many computer vision company deliver the resume, have not received any response, because others request almost all is a master of more than 3 years of above, I also have a master’s colleagues to try, although there will be an interview, but the interviewer will ask your bachelor’s degree, Generally speaking, the requirements for academic qualifications are quite strict. Then after my personal analysis, I think this is a requirement is not too much, because, in my opinion, direction and other IT direction algorithm, are better than one in two heads are better than the industry IT is no use, an excellent algorithm engineer more than one hundred middle algorithm engineer, so if I am a business owner, I would also be filtered on the record of formal schooling.

  • This is also the industry is more important things, is a variety of competition ranking, what contribution to the open source community, and how many papers published, but now papers are flying, quality papers are less and less. If you want to be successful in this industry, you have to pay attention to all kinds of competitions and strive for some rankings. The big factories in China all do so. And then if you can write some good papers, you can make it in the industry like a duck to water.

One last word

The current AI industry is indeed a serious bubble phenomenon, many people want to enter this track to run, the bubble will be broken, especially in the recent economic winter, change industries need to be cautious!!

From the perspective of wage oriented program (oop), the algorithm engineer treatment is very good indeed, but it seems to me that only for excellent algorithm engineer, the general development engineers could be reduced to do algorithm (i.e., the others to write a good algorithm engineering), for this kind of work, treatment is not so good, and often also face model optimization is not good, the performance is bad.

The above is only a personal opinion, although more explicit, may hurt confidence, but because of this, see the facts, and still run forward, do not regret, the more effort, the luckier.