Artificial intelligence as a current popular hot industry, releasing a large number of jobs, and because of the lack of artificial intelligence core of talent, making payroll-related jobs is increasingly rising, especially AI algorithm engineer Kong, in Beijing, Shanghai, Guangzhou, Shenzhen first-tier cities, such as Hangzhou, the average monthly salary of up to 23K or more. Talent gap is large, high salary, comes with AI algorithms BUFF's post inspired many computer engineers, software engineering, automation, and other related professional students' morale. " So what exactly have the knowledge and skills to become a qualified engineer AI algorithms it?
Working on the necessary basics of AI algorithms
/ Mathematical basis /
Artificial intelligence to learn the most basic of the high number of line generation must master the theory of probability, at least have to be Gaussian function matrix derivation, gradient descent understand how it is, otherwise the basic principles of the model can not understand, model tone participate in the training does not arise.
/ Programming Fundamentals /
For doing algorithm engine development or application development engineers, the ability to code implementation will directly determine the quality and efficiency of work output. Therefore, you want to do AI engineers need proficiency in at least one programming language (preferably python or C ++), and master supporting tools commonly used libraries.
/ Algorithm capability /
AI algorithm engineers need to know the advantages and disadvantages of the model, application scenarios, model selection, parameter adjusting superior technology, which requires you to have a system of machine learning and deep learning of theoretical knowledge, depth of learning. Students can write the code, blog, through the brush LeetCode, ACM exam to improve the ability of the algorithm. Knowing large number of algorithms and algorithmic thinking, reading other people when the code will be able to quickly distinguish whether the code is of high quality, inadequate where, how to optimize, and test algorithm main job is to test your thinking ability.
/ Engineering capabilities /
Engineering capability ensures fast algorithm engineer's idea of landing. The best students master a scripting language, usually require hand-coding the interview, problems and optimization methods for the realization of the code may appear in a minimum level of countermeasures. This requires students to have some programming experience and R & D experience. Improve in this area can be find some more projects, student groups or joint fight Kaggle, Tianchi game as a training approach, achieved through continuous adjustment algorithms and parameters to improve engineering capabilities.
/ Operational capacity /
AI refers to the operational capacity of the company's engineers need specific business content, business processes have full knowledge, which can clearly grasp the business aspects of the processed data from, how mainly on the ability of students in a particular vertical applications in the field of AI. This requires the students in the learning process of AI to choose a good school to go deep vertical applications, but also to the enterprise-level projects as soon as possible to train yourself - because of the complex structure of the data generated by the specific application scenarios, the dynamic, far from open source projects publicly available data sets can be compared.
Thus, AI is a challenging subject, complex knowledge, expertise abstruse, algorithm engineers need to be versatile swordsmen. And because most of the current "211", "985" college had not been opened in artificial intelligence, even opened the relevant professional, teacher supply and training system is not mature enough. So whether undergraduate and graduate, in addition to master basic math, programming knowledge, professional skills of their way to learn AI mostly self-taught, in this process, both also show their different strengths and weaknesses.
Many undergraduates have been a view that is misleading: Algorithms Engineer Gang recruit only degree or above. In actual fact, in addition to BAT, the core algorithm engineer Kong Huawei, Xinjiang and other large manufacturers often required degree or above, at least half of small and medium sized, entrepreneurial technology companies in Internet recruiting engineers algorithm Kong requires only undergraduate degree. Moreover, if the undergraduates on machine learning prior to graduation, the depth of learning to master the basics of solid and did some artificial intelligence-related projects, and campus recruiting by interpolation to find an internship or job algorithm Kong, graduate students and more than a 2 - - 3 years of work experience.
But the reality is: the majority of undergraduate students at the school had no contact with machine learning, deep learning courses, they generally look watermelon by the book, Li Hang, "statistical learning methods," Andrew Ng ML introductory courses in artificial intelligence and other information related to the basic theory, but encountered a problem I do not know who to ask, specialized teachers is not artificial intelligence research directions, leading to frequent encounters pit and can not be filled in a timely manner; read a few lines of code or error code once it may take several weeks; now some AI knowledge and technical reserves can not be achieved hit kaggle, level Tianchi the games, but no opportunity to do artificial intelligence-related projects, algorithm engineering capabilities and the ability to question, to find internship class of algorithms is very difficult. Therefore, these manufacturers want to find high-paying undergraduate AI algorithms Kong, it is necessary to find a suitable learning path, speed up learning artificial intelligence, must not wait until after graduate school, then let's shortcomings head-on with qualifications .
And with respect to the graduate students, undergraduates will be more than three years of study in school time, access to learning resources are relatively more abundant in the study of artificial intelligence knowledge have problems, it can be more convenient with the high level of the students, teacher exchanges and discussions. If the selected specialty is artificial intelligence research, there are some opportunities in lab-related projects, and its undergraduate and more than education itself an advantage.
However, most students in order to complete laboratory work no energy or machine learning system to learn the depth of learning, and for Tensorflow frame, master DNN, CNN, RNN and other classic neural network model mostly remain at the theoretical level, maybe a little better can simply reproduce the code, but still can not meet the employment needs of enterprises; in addition, some students are learning artificial intelligence to send essays, much larger than the theoretical practice, laboratory code can not be representative of the code works, but more companies willing actual project experience recruiting graduating students; Another point, if the tutor more stringent requirements for graduate students, graduate students in school is difficult to find time to practice algorithms Kong companies, less real-world business experience.
Obviously, both undergraduate and graduate, in the face of complex and abstruse artificial intelligence, knowledge and skills, you may encounter many difficulties in order to tie him to live the pace of learning.