Shulou(Shulou.com)06/02 Report--

Several common problems of entering AI

As a practitioner of AI, the author is often asked about joining or switching to AI, among which there are some high-frequency repetitions. Today, a few are summarized for your reference.

Question1: I am too old, and I was not a computer-related major before. How can I change my career to be an AI?

This question is often asked. Many friends, after working for a few years, want to turn to AI technical positions. However, I feel that I am older, or have the pressure of life, so it is not realistic to take the full-time postgraduate examination.

Unfortunately, when most companies recruit AI engineers, candidates are required to have academic endorsements such as a master's degree in related fields. As a career changer, how to improve your professional background?

In view of this situation, the author's suggestion is: start from your original professional field.

For example, an automation engineer in the original manufacturing industry must be familiar with all kinds of industrial production equipment, understand all kinds of motors and sensors, and know the physical meaning of the output signals of these devices.

Then, when machine learning methods are used in the industrial field, automation engineers have a considerable advantage in feature engineering, which is not comparable to those who engage in pure machine learning.

At least at this stage, the actual landing of machine learning and deep learning is mainly based on data-it is more important to be able to convert practical problems into numerical values than to do the operation itself (algorithm) exquisitely.

Therefore, if people who change careers can make full use of their previous professional background instead of abandoning it, the previous major will become an additional item. Switching to an AI as a breakthrough may be easier than going to a degree.

In addition, many industries are trying to apply AI technology to traditional business by AI+,. Against this background, if you want to switch to an AI position, you might as well look for opportunities in your own industry.

Maybe your current unit is doing or preparing to do an AI+ project, so you can try to join in first. After all, it is generally easier to find a new job within the organization than to change jobs.

When many enterprises try AI transformation, they will hire consultants with AI background to make plans and solutions for them.

Outsiders have knowledge and skills in machine learning, but they are not familiar with the business field and need the assistance of people in their own industry-see if they can become their helpers at this time, and learn the application of AI in practical fields by working with professionals.

Assuming that you can have this kind of experience, or at least have a deep understanding of your own industry data, and think about and try out business applications, you will have practical experience in XX industry AI. You don't have to improve your background by reading or playing games.

Question2: I want to join the artificial intelligence industry, but I find that there are too many things to learn, and they are all so difficult. Do you want to study for two more years before you look for a job?

This problem is also quite typical.

Objectively speaking, a person to apply for an artificial intelligence-related graduate student, or their own peace of mind at home to study for two or three years, and then to apply for a job, is it really competitive ability will be improved?

Although it is true that the academic background is an additional item, the author does not recommend the practice of "wait two years before joining the job".

First of all, when a person has this idea, it is actually a manifestation of fear of difficulties. To put it bluntly, "wait two years" is an escape, not a plan.

If you don't do it now, you often don't really learn, but give up from then on.

In addition, even if some people do not give up and really immerse themselves in learning, do not forget that there is a window for the rapid rise of any industry.

At present, artificial intelligence is in such a window period. Because demand far exceeds supply, it provides opportunities for a large number of people with no academic background to enter the profession.

However, such a window is fleeting. It is likely to be as short as a year or two, and in the long run it will be closed in three or four years. At that time, if you want to join the AI, you don't just have to study by yourself, but you really must have a diploma.

If you want to enter the profession instead of grabbing the time point, take advantage of the tuyere period and say you want to wait there, the result of such a high probability is that you will miss the opportunity.

Question 3: this is a problem similar to question 2-I want to be a machine learning engineer, but I feel too difficult. Shall I first try a position with lower barriers, such as data tagging? If you do a good job in data tagging, can you also be "upgraded" to an algorithm engineer?

It should be said that the probability of gradually reaching position promotion against the difficulty within the industry is not zero, but in fact it is very unlikely.

For an individual, entering the profession with the position of "doing data" has already put a label on himself, and others will use it to classify TA.

The original threshold for posts with different difficulties is different, so people will naturally draw boundaries for them. If you want to move from a data job to an engineering position, you have to break through a ceiling.

In reality, very few people can really break through the ceiling. Not to mention this already very small possibility, but also subject to the general trend of the industry.

When the job gap in the industry is very large, it is relatively easy to get in.

If a person entered the profession from data tagging a few years ago, he could get in touch with a lot of algorithmic experts and machine learning engineers, studying hard through the process of doing a project, while waiting for opportunities-- such as machine learning engineers. In a period of time, there will be a great need for talent-as soon as you encounter a vacancy in the job of engineering, rush to it.

The boss knows that this is a reliable person, willing to learn, have the ability to learn, and have done data work, may indeed give TA a chance to advance. But over time, as the industry gap becomes smaller and smaller, the possibility of such job jumps doubles.

Therefore, the author suggests: if you want to be a machine learning engineer, you should study hard with this goal, don't hesitate, don't wait, don't run away, and learn quickly from now on!

Question 4: many students will say: "I want to work in the field of XXX (there are a variety of fields here, such as stock forecasting, treatment of difficult and complicated diseases with AI, computer vision, speech recognition, reinforcement learning, etc.). What knowledge do I need to learn? how to prepare for the written interview?"

First of all, we should distinguish between research and engineering practice.

If you want to do research-as an algorithm scientist in a university, research institute, or research institute of a large company, you can focus on a specific technology, such as reinforcement learning.

Suppose you just want to study reinforcement learning, then you can go to universities, research institutes, and some companies that are at the forefront of this field, such as Facebook,DeepMind, etc., to apply for jobs and do algorithm research. Of course, you usually have to have at least a doctorate at this stage.

But if what you want to do is engineering, and the job you are looking for is an engineering technical position, you don't use a certain technology to make a distinction.

In industry, the distinction of domain is product-oriented, and the target is a variety of application directions, such as face recognition, speech recognition and so on.

Under the direction of application, the specific technology to be used depends on the requirements at that time. It's not that you want to do computer vision, you just know how to use cnn. In order to solve practical problems, we often combine the achievements of various academia and improve them according to specific limitations and requirements.

In this process, the determinants of whether a certain technology will be used are very diverse, depending on whether the technology can support the solution of needs; whether objective equipment, personnel, and time allow the use of this technology; whether your boss has the motivation to try this, and so on.

There are many tools to be used in the process of doing engineering, and no matter doing engineering or doing research, we must have the most basic knowledge!

Classical machine learning models, common deep learning networks, and the whole process of model training and inference (prediction) must be mastered.

If you are particularly interested in an application, such as predicting stocks, you can try it yourself. After all, stock data are everywhere.

Although many talents and institutions have long tried to use machine learning to solve the problem of investing in stocks, until now, machines have been completely defeated by humans in terms of long-term stock forecasts.

If you want to do Internet finance, or if you want to go to a financial company to do machine learning, there are corresponding jobs to look for, but there doesn't seem to be any kind of industrial job that allows you to sit there all day and only predict stock prices.

As for the landing of AI in medical care, it is not a technical problem, but a problem of institutional barriers and data access.

To do medical AI, you must first be able to connect with the hospital. Even if there are channels to do this, a large number of cases are handwritten, and the doctor's handwritten, is there any way to make it electronic and managed by computer?

The first step in obtaining data is a problem that AI+ Healthcare has not yet solved. There is not even the most basic data management and statistics, what data analysis is there to talk about, what AI is to talk about?

When it comes to written tests and interviews, all written tests and interviews must start with basic knowledge.

For example, during an interview, the interviewer will often choose a classic model (linear regression in the early years, logical regression in the previous two years, and may have reached SVM now) and ask you what its model function is. What is the objective function? What are the optimization algorithms? Then see if you can write down the function formula clearly and explain the process step by step.

We don't have to be too utilitarian. It's good to have specific career goals, but before you do that, you have to master the basics.

First learn the most classic models, such as Linear Regression,Logistic Regression,Naive Bayes,Decision Tree, SVM,HMM,CRF,Clustering,GMM,PCA, etc.

Settle down to lay a solid foundation, and when it comes to the interview, you will naturally be able to answer fluently.

Https://mp.weixin.qq.com/s/YkRy_5xCKX4V_1RxWGPyYQ

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.