Nearly two months without updating blog, focus on the aspects of the delivery of conference papers and autumn move to find work. Here briefly summarize some suggestions and feelings of autumn recruit interview.
Delivery is mainly NLP algorithm engineer jobs, mainly participated in the interview company Tencent (Spring trick), ants gold dress (spring strokes), chasing a science and technology, ape counseling, job help, according to map technology, Jingdong, fight a lot, star of Science and Technology (campus recruitment interview, whether by interview and follow-up arrangements need to October), Millward science and technology, as well as Industrial and Commercial Bank (school preach interview, written interview to the official bank to October).
The company's cattle pen by the side by the customer network, and many sites have to share their predecessors, and I will not repeat them, just sum up my feelings.
1. have their own thing
Candidates algorithm engineers very important point is to have our own things. I put this on the first point, it is to emphasize its importance. I know many students, including myself, we spent a lot of energy early to see someone else's paper models and algorithms, to understand the causes of these models may work, or even look at some books, like the face by them as "standard answer" to remember, and "examination" almost.
But, you see more, but also all other people's things, own things? To put it simply, reading papers and reading a novel on the same simple, but think and form their own algorithm model, validation, trial and error, improve and perfect the whole process on its own is the real harvest, which is the ability to reflect. In my opinion, this is a sub-engineer the algorithm, that is, not only have the ability to learn, and more importantly is to analyze, to think, the ability to form a solution. And not know a lot of other people's models, the ability to rote.
This is important in a job interview, the interviewer will be interested in what you have done to solve or improve what, where how to do, to do so in truth, there is no other possible way, what are the advantages and disadvantages of each of these methods ...? of course, this may seem a bunch of questions, very simple, if it is to seriously consider their own algorithm repeatedly proven, these are the issues we have been considered verified. So these seemingly difficult problem, but in reality is their ability algorithm, reflect practical ability, logical, experiment fully answer would be a plus for themselves. (In addition to these analyzes are in place, it will take a lot of interview time, will leave a good impression to the interviewer; And if the test code or logic behind the problem, the underlying problem will not be very demanding requirements of the interviewer first impressions important.)
In addition, I want to say is, there are some books by side, the blog for the understanding of classical algorithm may be wrong, learning when to have their own thinking, so the algorithm idea for our own use, rather than as a standard answer to the back, it is very regrettable, even a waste of time, misleading the algorithm ideas. These answers back time might as well let folks thinking to do their work. (For a simple example, XGBoost, and the contrast GDBT, GBDT is the first to use the second derivative, XGBoost GDBT algorithm follows the idea of also using the second derivative, not in the face by talking about "GDBT use only a with the second-order derivative XGBoost. "I would have a direct look at the paper.)
2. What companies need to know
Not many companies are holding the same education as the University of you, train you, you discover the unknown potential approach to recruiting, unless this post a lack of people, otherwise many companies are valued Will you now have the ability to direct the work. This is fair and reasonable. If you want to get the company to pay the salaries, need to know what the company needs and what they can do, not even a graduate student in-depth research company main business content, before the interview also need to be prepared in advance. Though unable to very expert in, it's best to have a basic understanding of the architecture, commonly used algorithms, models, tools.
For example, when chasing a major business NLP technology at this stage of the dialogue system, consider the lines can deliver their products and so on, and generally will not use end model but pipe line, so each module can be pre-dialogue system to solve the problem the main algorithm understanding; Jingdong Mall is the main job of scoring product reviews, mainly used in sentiment analysis, text classification, before the interview can understand some of the other second interview the interviewer asked about regularization, parameter adjustment, spark, in fact, the interviewer they usually like to ask the business is often used, if the early understand their business can be targeted better prepared. According to the plans of the technology only participated in one side, companies pay more attention algorithm programming capability, is the ability Shredded code for the project applicant's experience in passing. Done in accordance with plans to ask the interviewer during my undergraduate project involves CV, I feel very strange, it does not embody the self-righteous think NLP capabilities, ask the interviewer to change the contents of the talk of NLP. Later I discovered that he did not understand, the original interviewer is doing a CV ...
Talk so much, it is to say, to stand in the interviewer's perspective. An interview is not his own show, is looking for buyers for themselves. Whether it is the first point "own thing", said no matter how good, is for the first 2:00 of service that reflect if he can do the job. What companies need to know what they have, if not pre-fill some of what is required to make adequate preparations. In exchange, the interviewer will generally prefer to pick a topic and you are doing business in more relevant topics to ask, not only because the business needs, but also because he understands. Most people tend to understand that they have to find out about something a little, the interviewer also have this mentality.
3. firmer foundation
Interview mainly on basic skills, including, (1) the depth of learning to understand the basic algorithm for NLP or depth, this generally extends from your subject presentation; (2) some companies will be involved in machine learning algorithm; ( 3) as well as C ++ (Tencent), python (chasing a); (4) Shredded Code (basically tear); (5) other topics (exam puzzle in accordance with plans, fight a lot of exam questions probability).
These are cumulative, for (1) the depth of learning, and (2) machine learning algorithms, are also commonly asked and important part, if you usually pay attention to the accumulation of the paper model to figure out which part of the basic no problem, there are individual the more difficult it could not answer honestly say he did not consider in depth, also in line with the interviewer would like to hear your views on this approach to discuss the issue, and will not be difficult.
I personally think that a blog is a good way to show their accumulation of continuous learning, and have a good way to own deep thinking on the issue. Write a blog to me is very beneficial. In many cases, since that will either correct understanding of the content, in re-think and write down the finishing process, found a lot of places do not understand themselves understand, before or incorrect, will gather information and then issue Root of the Matter Rethinking sort out. This process, they themselves would clarify ideas, more in-depth understanding; the pen to paper to write down the process should also think about how to organize words, the use of illustration let people do not understand can understand that he also exercise expression the ability; users answer questions during the harvest will be some of the friendship ~; another unexpected results that link to blog posted on the resume, the interviewer really willing to turn over the pages! He will see that your continued efforts, will be happy to talk to you on certain issues, listen to your understanding of this process, we could really learn a lot. And this may be "normal results" came in the introduction, the interview "a test set for life," the high risk of spread out. Of course blog is only one way, such as github project, etc., are showing the way individual long-term effort, but also can help people in a short time more fully comprehensive understanding of their own.
For (3) c ++ you can see some of the chapters "c ++ primer" is, of course, the point often mistake some programming, such as wild pointers, memory leaks, etc. usually need to accumulate it; python is usually encounter problems smoothly investigation, mainly to understand clearly mutable objects immutable, process threads, I heard that some companies also test a decorator; entry python python friend's blog to see Liao Xuefeng time undergraduate, these have said very clearly, easy to understand, you can also go over.
For (4) basically is "to prove safety offer", leetcode brush questions; I heard that some companies Examination Shredded k-means this machine learning algorithms, but also good accumulation.
For (5) to see through surfaces.
4. Understand the environment
HR may face problems involved: you apply for jobs? Why choose this direction? Xx can work in the city do? Why want xx city? Where is home? Why did you choose this company? There are senior senior sister apprentice in the company do to understand the company? NLP engineers know what you mainly do? Your three advantages and three disadvantages? How under pressure? expected salary? Whether to accept the swap?
Advice is, know more about early to understand the development prospects of the company's operations, work (overtime) and so on.
For his part involved: do not need to say "completely true", but need to be justified, the core is to reflect the love and for the industry, can do the job, rather than reflect on their own analysis of how place; not under pressure only said "very good", the best example of the ability to reflect their own child out;
Salary: according to their average salary level required for NLP algorithm engineer jobs industry, the company has a basic level of understanding, say they think a reasonable value, not cheap themselves. If the HR side through the final stage to talk about salary, this time we can take a look at this salary level is reasonable based on their understanding of the industry situation; if the salary is low, you can try based on industry standards and their ability to talk and HR salary, a more ideal to fight for their salaries; of course, if the salary is high, means that the company or ask your expectations too high and need to do better.
5. know what they want
(1) not to job transfer?
People are really powerful harvester offer, but for a wider range of students, this year's algorithm Kong and easy to find. After hitting the wall in front, surrounded by many friends of the job transfer.
This year the demand for large jobs:
1) Client -> Requirements: No, it does not matter, you can come to learn. But the need to have a solid foundation of Java (for Android) and basic programming skills (Shredded Code). Also concerning the client's face can be seen through learning and understanding. When a friend candidates for the position, will not respect the client, the interviewer directly say, what would you can talk about, and then describes some friends prepare their own content, can articulate.
2) test development -> above, still need to have Java-based, and some understanding of the test work. Still have not attitude can come to learn. But some companies are still relatively strict, the interviewer will see at a glance whether you have done the work, and will say that the content in the surface through practical work and will not use and so on.
In fact, do more than two jobs, wages in some companies is also very high, and even some of the company's clients pay even higher than some developed algorithms Kong Kong. So, not to say that wages must be algorithm> Development> Client & measuring open.
So, are you willing to risk to find work adhere algorithm Kong? Or job transfer? To State Grid? go to bank? Civil servants? Du Bo? This is a personal choice. Here is the prioritization of the problem vary, there is no right or wrong. Perhaps the only fault of their own inherent obsession, an industry think how certain how, in fact, businesses are conducting their own development, so not only look at now, it also depends on the future.
(2) find a job how to do?
To find a job linked to a string, hang to doubt myself; but I think, reading is not intended to imprison themselves in a certain fixed direction, so-called "gentleman not" on, think they still have a lot of good ability, even though unemployment also still by virtue of their ability to make a living, so do not be afraid. Lack of ability can learn it. As for the ability but the interviewer did not see, thinking about their own intelligence and physique is still there, not the company, is the company's losses, they have to lose?
In fact, many people are experiencing a lot of times after being rejected, not to give up, finally have a result. If worth the wait, worth it stick. I hope everyone can have a satisfying job, and doing well at work!