What is the criterion of in-depth learning algorithm engineer in my mind

What is the criterion of in-depth learning algorithm engineer in my mind?

There are three AI platforms that only concentrate on the original output, seldom talk about and do not touch hot spots, but recently many friends have asked, so they have to write a unified article to answer this question that everyone is very concerned about, of course, this is only a personal point of view.

Author & Editor

At present, there are many things that can be done by using this tool, such as image, voice, NLP, entertainment, finance, medical treatment and so on. A lot of non-professional technicians want to change careers. In other words, there are several people who are really of this line of origin. The disciplines are only recently opened.

So the question arises. How can we join the industry? There are many answers, I will not say, the following only elaborate personal views, just personal views, do not like movable!

If I recruit, what is the minimum requirement?

If I am recruiting an in-depth learning algorithm engineer, whether it is campus recruitment or social recruitment, the basic requirements are as follows:

(1) Proficient in Python and c++ programming, at least familiar with Caffe and T ensorflow/Pytorch frameworks.

(2) Proficient in the use and design of various model architectures.

(3) To skillfully manipulate the collation and use of data, we must have a deep understanding of the position of data in the task of in-depth learning.

If you can't satisfy these three points, if you have a very resounding origin, there may be drama. Now is not 16, 17 years to see two online courses run two GitHub projects can enter the market, there are people on the market, really can not choose the top school seedlings to post re-cultivation, curriculum vitae background is very important, the reality is that.

It seems that there are only a few sentences, but there are many subtexts here.

(1) Writing programs must be skilled.

Can you just write python? Primary school students will be good, unless you reach the level of writing a framework like faster rcnn, you can widen the gap with the public. It's enough to be able to write Python algorithms, but companies don't hire people to write papers just for your training model. Can Python deploy the model to mobile phones? Can Python optimize the underlying algorithm? Be sure to master C/C++ in advance, or you'll learn it again at work. That kind of pain and pressure will torture you.

In the same way, open source frameworks can't just tensorflow, let alone talk about it.

(2) Model + data, thoroughly grasp the two carriages of in-depth learning.

The normal way for a company to work is that the boss says there is a project, you do it, the data is not available, about two months later on line, a weekly meeting. No one gives you data unless it's a large-scale annotation that can be outsourced. Data comprehension is not in place, 99% of your project is dead, what amount of data is inappropriate, the data distribution is unreasonable, the data quality is not high enough, the test data is not good, there are many moths.

Besides, there are many open source models, but have you seen any core models that have been optimized on the product line? Open source is mostly the model of academic research. In order to occupy pits and brush indicators, the most important purpose is to occupy pits to declare sovereignty, but it is not directly available online. And there are different requirements for different tasks, from input size to step length, depth width to training methods, there are many doors, in-depth learning is known as alchemist has some close-knit skills.

(3) Poor ability, do you have any background?

Even if it is a training institute, it will say that we have Dr. XX who returned home from XX. There are very few schools in 985 that are not good for me to do this. Many bosses prefer to recruit students from top schools who are temporarily unable to do so, rather than a person of ordinary origin who is not prominent in ability. Where is there no circle when the rivers and lakes are big?

If your resume background is not good, you can only show the level that others do not have, and have more cards.

2 What skills does a master have?

Above is the minimum requirement, then what skills do the master have?

The low hand is the same, the master is different, there is no uniform standard, the following list of common skills of the master bar.

(1) Good writing.

This is the most familiar to you, it is also the most media propaganda, after all, a large group of students. Write really well, not in the way that you wash your manuscripts with water.

(2) Code/framework is well written.

If you can write a DarkNet or Caffe, the salary is arbitrary. Of course, someone else has written it out and you can't fix it again.

(3) Wide knowledge and practical ability.

Whatever the project, you can handle it on your own. This is the person who can solve the problem and make profits for the company.

3 How to learn

This is a very sensitive question. I will not say how to learn, but what learning methods are wrong. It only represents my personal position.

(1) Watching video has little effect.

Learning is an anti-human thing. If you are comfortable with it and can learn while lying down, you should first doubt whether you have really learned something. We have written more than 500,000 technical articles, that is, no video. Why? Because I think videos are limited to knowing one thing, and only by hand can they be learned. To take an extreme example, one of my fans reported tens of thousands of courses to study, and as a result, his project Linux has been working for a long time without using it.

(2) Don't indulge in theory.

It's no problem to learn maths well, but I met a few students who were addicted to deducing formulas and many people saw a lot of things.

Please read the Chinese version for details.