Machine learning projects permeate almost every area of our lives. Teaching bots to respond like a human or to predict music preferences is fun, but not easy. Together with Victoria Tyufyakova, a mentor on the Data Science course, we figure out what kind of base you can go into machine learning with and what languages are popular in the field, except for Python.
What does an ML specialist do?
A machine learning specialist (ML specialist) uses algorithms to build models that can learn independently on various data sets, from tabular data to pictures and texts. It reveals complex patterns in datasets so that the model can more accurately predict the result, and it also helps automate decision-making processes for difficult problems in practice. Machine learning is a subsection of artificial intelligence, which also includes deep learning and neural networks.
The main difference between an ML specialist and a data scientist is that the former, in addition to everything else, has stronger development skills and can bring the model into production, for example, embed it into an existing product or package it into a new one.
Where is it needed?
Since the activity of an ML specialist is aimed at facilitating decision-making, they are looking for him both for everyday business tasks and for scientific projects in many industries.
- Transport. With the help of models, you can train unmanned vehicles, identify obstacles, traffic jams, build optimal routes or recommend convenient landing sites .
- Retail. You can plan deliveries, personalize ads and discount offers, or measure the effectiveness of promotions.
- Healthcare. Based on the results of the tests, it is possible to diagnose pathologies, make more accurate diagnoses, and choose the optimal path of treatment.
- Finance. Machine learning methods help banks make faster decisions on issuing loans, predict changes in the volume of deposits, while natural language processing allows you to set up chatbots and virtual assistants.
- Industry. You can predict when equipment will fail, or find defects.
- Entertainment. Algorithms in the gaming field can create more realistic avatars, predict the next move to win or lose, and create 3D models for VR glasses. Movie feeds in online cinemas and news in social networks adapt to our preferences, also thanks to machine learning.
Machine learning is used by well-known global businesses such as Amazon, Google, Netflix, Apple, as well as large Russian companies and banks: Beeline, Tinkoff, Sportmaster, Lenta, as well as small startups.
What does he need to know?
It is quite difficult to come into machine learning from scratch – you need serious mathematical skills already at the start. To develop algorithms, you need to understand linear algebra, mathematical analysis, statistics, and probability theory. ML specialists use linear, logistic regressions, decision trees, Bayesian classifier, boosting – this is not a complete list of algorithms that are popular in machine learning. They require a good mathematical background for optimal use depending on the structure or size of the data.
Since an ML specialist is constantly working with data, he needs to know SQL, be able to write database queries, and work with data warehouses.
Most often, ML specialists use Python (or R ) and libraries: Pandas, NumPy, Sklearn, Keras. For recommender systems and neural networks, the TensorFlow framework is suitable, and for natural language processing, specialists use PyTorch.
Scala is an industrial-scale efficient programming language, but it is less common in job postings. It is faster than Python and R, which makes it stand out when working with large databases. Also, ML specialists often use the Apache Spark framework, which allows you to work with unstructured or semi-structured data sets.
Often among the skills, you can also find C or C ++ languages. The fact is that they process large amounts of data faster than Python, and for machine learning tasks, learning speed can be very important. They can also deploy models with MATLAB, which is a software package that is used for ML in a scientific environment.
In the field, a big demand for MLOps skills is the standardization and optimization of machine learning model lifecycle management. This helps reduce risks when working with machine learning and automating processes.
In addition to technical skills, vacancies also mention knowledge of the English language. It is needed for reading technical documentation and scientific literature since many current works on the topic are not translated into Russian. In addition, it will allow you to enter the international market.
Well
Machine learning and neural networks
A set of advanced courses for mastering machine and deep learning from classical models to neural networks. Additional 5% discount with promo code BLOG.
To learn more
Types of machine learning
In machine learning, it is not necessary to get a perfect error-free result. Rather, the model should correspond as closely as possible to the metrics that the business sets along with the problem statement.
Different types of machine learning are used for different tasks: supervised, unsupervised, reinforcement learning.
Learning with a teacher
The model has data about past interaction with the system. She understands what correct answers should look like. This type of training is used most often. Supervised learning solves two main problems:
– classifications. The model must determine the category of the analyzed objects. For example, you can separate pictures of cats from pictures of dogs, or detect fraudulent activities of a bank customer.
– regressions. In such problems, on the basis of a time series, it is possible to predict, for example, the dollar exchange rate in the short or even long term.
Learning without a teacher
The model does not provide the correct answers, so it builds relationships between the data and interprets them. Most popular tasks:
– downsizing. This allows you to extract meaningful information from the data through fewer features. So you can reduce the size of the data if there are too many of them so that they take up less space during storage, for example, reduce the size of pictures.
– clustering. Clustering groups together similar data to find patterns among group members. This allows you to find insiders to, for example, apply targeted advertising. Most often, it is used in recommender systems that offer to buy a thematic product for a client.
Reinforced training
The closest type of machine learning to artificial intelligence. The algorithm, guided by a given set of actions and rules, interacts with the environment. By trial and error, he builds the most optimal order of his actions. Such tasks are popular in navigation: the algorithm moves around the room and builds a trajectory so as not to crash into an obstacle.
Prospects and salaries of ML-specialists
Machine learning every year penetrates deeper into all areas of human activity, we use it every day, for example, when we are looking for information in search engines. As Nadezhda Zueva, co-founder of the MIPT-based Deep Learning School, notes, more and more services use machine learning technologies, which entails an increase in jobs.
Most actively, machine learning is being implemented in information technology, financial, transport sectors and marketing. The market for chatbots, which are used in customer support, is growing significantly, which, of course, was greatly affected by the pandemic.
According to Gartner, in 2022, global revenue in the field will increase by 21.3%. 48% of IT leaders are already implementing machine learning technologies or plan to do so during 2022.
As of December 2021, there were about 1,600 vacancies on hh.ru for the query ML. At the same time, the lowest salary among these vacancies is 40 thousand rubles for a junior. However, on average, salaries range from 150 to 250 thousand rubles. Seniors with three years of experience in the field are paid from 400 thousand rubles.
However, the number of specialists practically does not increase, and there is a shortage of qualified employees in the sphere.
How can a newbie get a job?
A beginner can count on a job if he already has practical experience. In Data Science, it is easy to gain this experience by simply solving cases. You can collect data yourself, participate in machine learning competitions, for example on Kaggle, in hackathons, where participants are given data with which they can solve real business problems.
Such a task can be put into a case and posted on your GitHub. In this way, you can put together a portfolio to present your skills at an interview. The more detailed the case, the more likely it is that you will be hired because it will be clear that you have tried to delve into the topic.
A junior who knows the basic principles, but has not yet gained practical experience, can be given tasks to test hypotheses, and upload data. It does not greatly affect the final product. For a middle position, you need to work in the field for about two years. He can develop models that influence decision-making in the company. A senior with experience, in addition to developing models, can also implement them. Often, in addition to technical tasks, he also has managerial responsibilities to lead the team.
Where to begin?
Books
If you are just interested in machine learning, you can start with books to understand how easy it is for you to understand the basic concepts:
” Introduction to Machine Learning with Python ” by Andreas Müller,
” Applied Machine Learning with Scikit-Learn, Keras and TensorFlow ” by Aurélien Geron.
An extensive library with English and Russian books on machine learning and artificial intelligence is also available in the Machine Learning telegram channel .
YouTube
TensorFlow channel with explanatory videos about machine learning and conference recordings.
Channel Jon Krohn, which also focuses on mathematics.
A five- hour video about machine learning in Python from scratch.
Open course on machine learning from ShAD.
Learn on your own
If you already have basic mathematical skills and are ready to learn languages and libraries on your own, you can focus on ready-made roadmaps in order to structure and not forget all the information you need to know.
Collection on GitHub with a list of required skills, courses, competencies.
Communities
To stay up to date with the latest news, competitions, technologies, ask for help or find a job, you need to update the community. Some of the largest and most popular are ods.ai and Kaggle, not only for data scientists or analysts, but in general for all Data Science specialists, including ML.
Courses
You can start with free ones, Coursera always helps out: short introductory programs are offered by both Russian universities ( NRU HSE , MIPT ), banks, and many foreign companies and institutions from Amazon to Stanford.
You can also take a course in the machine and deep learning . Its advantage is in practice and mentors who will help you understand any nuances. For a specialization as complex as machine learning, the support of experienced professionals is more important than ever, especially if you are just starting out in the profession.
Leave a Reply