During BIGO work, interviewing many candidates, there are graduates, but also worked for several years, of course, sometimes there are like me, students from other areas referred for depth of learning. After the interview, for some students, often controversial before the interviewer, and some feel pretty good, the other interviewer feel that it must not be. So, after the interview many times, I often think, what kind of students that enterprises need, how to identify which outstanding students, and how to identify those bombastic generation of it?
Personally I think that a good algorithm engineers, following Odor:
Codes are strong
I think this is the most important capability. Often we encounter some paper models of various well aware, from faster rcnn series to ssd, yolo, all-knowing, but let him write code on a computing iou dumbfounded classmates. In my opinion, these students just transfer package and to Man in the job market is the lowest level of existence. Some students might refute me, some of the code Dr ability in general, but mathematics is good, and stronger model ah. On the contrary, some real Bullog I know, none of the code is not powerful capabilities, it is conceivable that they are sophisticated model, there is no ability to do powerful code support, how is groundbreaking achieve concurrent paper out of? So, who will be on paper, show me the code.
Really understand the algorithm, not only derivation formula
When I interview people rarely go directly to deduce what kind of formula (of course large column how to recognize a good algorithm engineers , various linear model I will study, which is the basic linear algebra capacity), but I would go visit its image understanding. For example, I often see students have written resume will svm, so I would ask them, svm and a logistic regression, when the data categories of serious imbalance, which is better should you choose? This problem will not need to derive the formula to solve, but was able to really examine whether the interviewer really understand svm starting point, if the problem can not answer it, I think, will derive the formula to solve even svm, it is relatively poor candidate.
Strong willingness to learn, do not pick live
Algorithm Engineer was called algorithm engineers, because they often encounter unknown problems, learn and try to solve it, so people learn the will is not strong enough, can not be, no matter how high knowledge base, also encountered an unknown problem. Pick live, not to the actual work, from data to find data cleansing, this is the only way algorithm landing, no one can have the privilege to select only wanted to do, threw the dirty work to others .
The actual interview, interview 10 people, to have two individuals can achieve these requirements pretty good. So, before know almost see a discussion, say 2019 school recruit depth study whether job competition is fierce, I would say, in fact, feel intense, most of those who have looked at this line of hot money, just looked at the video on the Internet to interview, in this business for real students have the ability, is not fierce, are rushing to each company's. And, after so long, the team never established, companies also understand that, in fact, the algorithm team to recruit people focusing on quality, not quantity, recruit 10 bully, as a great God, true field algorithm.