How Can A Programmer Become An Artificial Intelligence Specialist

According to a McKinsey report , there is a steady shortage of machine learning specialists in the United States: demand is growing at 12% per year, and supply at only 7%, with the result that in the near future there will be 250,000 more open potential candidates than potential applicants. In Russia, according to HeadHunter estimates , the number of positions for machine learning and artificial intelligence specialists in 2017 increased by almost 11 times.

AIMAGNUS.COM

CTO at LoyaltyLab and Binary District course speaker Alexander Kondrashkin compiled a step-by-step AI study guide. Alexander states: “Around the machine learning a halo of wild complexity has formed. This is so if you want to make discoveries, develop new algorithms and enter the history of science, but if you just apply well-known solutions in practice, the entry threshold is not at all large. ” Almost all programmers have the necessary basis for building a career in artificial intelligence.

What you need to start

Abstract thinking

Machine learning revolves around finding patterns in data. Data Scientist devotes much more time to generating hypotheses, preparing and conducting endless experiments on data arrays, than designing service architectures and debugging them. In the mind of a specialist, Yandex.Maps roads and intersections turn into graphs, and statistics on cash withdrawals at ATMs turn into time series in the analytical system. One cannot do without the skill of representing ordinary things in an abstract form.

General literacy in maths

A career in machine learning requires general literacy in mathematical disciplines. Probability theory, linear algebra, mathematical analysis - subjects that teach in the first year of any technical specialty - another cornerstone of Data Science.

Knowledge of Python and the basics of backend development

The third basic skill is programming. Most Machine Learning work is done in Python, but knowledge of any language will speed learning.

At the stage of working with prepared models, backend development skills will come in handy. Even with superficial knowledge in this area, a neural network is much easier to turn into a user-friendly microservice.

If you already have this base, you can safely go to learn cars.

Where to begin

The path of the future Machine Learning specialist partially repeats the history of the industry and begins with the classic learning algorithms with and without teacher, created in the last century.

For starters, the Bayesian classifier, linear regression, and decision trees are simple, intuitive methods for automatically sorting objects. Unlike capricious neural networks, it is easier to get positive feedback from them, to see: “Wow, machine learning really works!” - and get motivation to understand the topic further.

After two or three months of studying the basics of syntax and classical algorithms, it's time to move on to neural networks of simple architecture - single-layer perceptrons. The next logical step is to make the multilayer neural network work, and after that you should pay attention to reinforced learning.

How, where and what to learn

To master the basic methods, a reliable laptop with Internet access is enough from the equipment. Beginners rarely need really fast hardware, and in extreme cases, there are services from Google, Amazon, Microsoft and other cloud providers that lease out power.

Courses and trainings

At educational sites on the network, many comprehensive training programs in English and Russian are published.

One of the most popular English language courses is Stanford University's Machine Learning . Not so well-known, but also useful are the free programs Intro to Machine Learning and Become a Machine Learning Engineer , which are offered at Udacity.

Those who want to understand the nuances of using Python for machine learning should pay for the " Data Science, Deep Learning and Machine Learning with Python " or the " Python for Data Science and Machine Learning Bootcamp " at Udemy. And for all those who want to establish feedback with teachers and gain knowledge in person, an intensive practical AI course is opened in the Binary District .

Practical tasks

You can practice on your own invented tasks, but such experiments are ineffective for study. It will take a huge amount of time and effort to collect data through the API, clear it and prepare it, which can be spent on mastering knowledge. It is much more convenient when everything is thought up and made up to you. The Kaggle educational platform publishes machine learning tasks with a detailed description, predefined conditions and prepared datasets.

For example, the classic tasks of predicting the survival of Titanic passengers or real estate prices in Ames, Iowa will help to thoroughly understand classification and regression, and recognizing handwritten numbers will allow you to immerse yourself in working with pictures through a classic problem.

What to do next

Starting almost from zero level, in half a year it is quite possible to get an internship, and soon to get a job. To master neural networks of complex architectures and ensembles on work tasks with the support of colleagues will not be as difficult as alone.

However, this is not the only way to test acquired skills in battle. If your company has collected a lot of data, it is worth developing a machine learning culture inside, extracting additional benefits from the databases and, quite possibly, money.

Practice, read, experiment and share your experience - there is no ceiling in this area. Further professional growth in Data Science is the result of daily work, a consequence of the complexity of tasks, communication with colleagues, participation in conferences and immersion in business metrics. Any model can be made a little better, but over time there will come an understanding of when an additional percentage of accuracy is not worth the effort and will not benefit the company.

Of course, AI will not solve all the problems of mankind, and will not take away the work of programmers. But now machine learning is opening up great opportunities, technological and professional prospects.