What is Machine Learning (ML)?

How it works

Machine learning is the process by which computers learn to recognize patterns and make predictions. Artificial intelligence analyzes data and accumulates experience, similar to how people learn. From movie recommendations on Netflix to facial recognition on a smartphone, machine learning is behind it all.

Imagine that you are training a dog. You teach it different commands. For example: "sit", "lie down", "come to me". You praise your pet and give a treat when it performs the command correctly. Gradually, the dog learns, remembers, and begins to understand that a specific command means a specific action.

Machine learning works on a similar principle. Only instead of a dog, there's a computer program that learns from data. For example, you show it photos of different objects and it learns to distinguish them from each other. The more data you give it, the better it learns and the more accurate the result becomes.

Machine learning is closely related to artificial intelligence. AI is a broad concept that includes various technologies, such as computer vision or natural language processing, which allow computers to "think" and act like humans. Machine learning is just one of the tools for creating AI.

In a year, you will become an ML engineer: get mathematical training, master Python, learn to work with data and create your first machine learning models on the Machine Learning course from scratch to Junior.

Learn more

How it appeared

In 1959, American researcher and IBM employee Arthur Samuel created a program that learned to play checkers. It was able to learn from its mistakes and gradually became stronger. This was the first serious achievement in the field of machine learning.

In the late 1980s, the ChipTest chess machine was created at Carnegie Mellon University. It calculated up to 50,000 moves per second. ChipTest became the prototype of the Deep Blue supercomputer, which in 1996 played chess with the great grandmaster Garry Kasparov and won.

Deep Blue, unlike humans, could not "feel" game situations. But he could analyze many times more moves and calculate which move would be profitable.

Deep Blue marked the beginning of a new stage in the development of machine learning. In 2011Google created a special department devoted to artificial intelligence — Google Brain. Three years later, Amazon and Microsoft launched their machine learning platforms. And Facebook presented the DeepFace algorithm that recognized people's faces.

New technologies such as GANs — generative adversarial networks and transformers — have opened up new possibilities for creating artificial intelligence that generates text, images, and video.

💡 GAN is a machine learning technology that includes two components: a generator and a discriminator. The generator creates new data. For example, images, imitating real ones. And the discriminator tries to figure out which data is real and which is artificially created. The components compete and improve. The generator is better at creating realistic data, and the discriminator is better at discriminating.

It's like two artists competing with each other. One artist creates images. The other tries to determine which images are created by a person and which by AI. By competing, both artists become better.

How it learns

Supervised learningImagine teaching a dog to obey commands. First, you say "sit" and help her sit, then praise and give a treat. Gradually, the dog understands that the word "sit" means a specific action and carries out the command on its own. In supervised learning, the algorithm is similarly given a data set, where each record contains examples with correct answers—labels. The algorithm learns from this data so that it can then independently classify new examples;

💡 Labelsare like stickers. They indicate which data belongs to which category. Imagine you have a box of fruit and someone sticks labels "apple," "banana," and "orange" on each piece of fruit. These labels help the algorithm understand how to distinguish and classify new data in the future.

Unsupervised Learning. The algorithm is given a dataset, but without labels. The algorithm must itself discover patterns and divide the data into groups. These groups are called clusters;

💡 Data — a set of information that is used to train the model.

💡 Model — a mathematical description that describes the relationship between the data and the expected outcome. It is like a map or diagram that helps you understand how the data relates to the outcome.

Reinforcement learning.This is learning based on trial and error. The algorithm receives a reward for correct actions and penalties for incorrect ones. For example, a computer game where you control a character learns to complete levels. Machine learning learns by receiving a reward for reaching a goal and penalties for colliding with obstacles.

💡 Algorithm— a set of instructions that tell the model how to learn from data and make predictions.

These algorithms and methods are most often used for machine learning:

Linear Regression.Predicts values of a continuous variable. For example, predicting real estate prices based on area;

Imagine you want to predict a person's height based on their weight. You already have data on the height and weight of several people. Linear regression constructs a line that best describes the relationship between height and weight. It then uses this line to predict the heights of other people.

Decision Trees.Group data. The decision tree algorithm creates a series of questions to classify an object. For example, to separate email into spam and regular emails;
Neural Networks.Recognize images, process natural language, and predict constantly changing data. For example, the weather;
Clustering.Groups objects into clusters based on similarity. For example, to group online store customers by purchasing preferences.

Where it is used

Large global companies use machine learning. Amazon uses it to recommend products, Google uses it to improve search results, and Tesla uses it for autopilot in cars. Let's consider its application in different areas:

Healthcare

Performs early diagnosis of diseases. Machine learning algorithms analyze MRI, CT, and X-ray images to detect signs of disease in the early stages. This helps doctors prescribe treatment in a timely manner and increases the chances of recovery;
Development of new drugs.Machine learning accelerates the search for new drugs by analyzing huge amounts of data on chemical compounds and their interactions with the body.

Finance

Detects fraud.Algorithms analyze financial transactions and identify suspicious activities. This protects bank accounts from unauthorized access. For example, in 2023, Mastercard launched a special system based on machine learning to find fraudsters;
Assesses credit risks.Machine learning algorithms analyze data about borrowers. They evaluate their solvency and the risk of loan default.

Marketing

Personalizes advertising.Machine learning algorithms analyze purchases, interests, and user behavior on the Internet. Thus, services offer advertising that is tailored to the interests of visitors;
Automates marketing campaigns.Machine learning optimizes newsletters, customizes advertising, and creates content. For example, a neural network analyzes user behavior and guesses which topics and formats are interesting to the audience. This makes newsletters personalized.

Machine learning also helps with content creation with ideas, processing large amounts of data, searching for trends, and writing texts.

Automotive industry

Drives a car. Machine learning algorithms learn to drive a car. Autopilot analyzes the road situation, traffic lights, and the behavior of road users; Machine learning algorithms analyze the behavior of road users. Machine learning algorithms analyze data about the car's operation. This service predicts breakdowns and alerts the driver to the need for repairs.

Other areas

Helps in agronomy. Machine learning algorithms help farmers optimize watering, fertilization, and harvesting. This increases yields and reduces costs;
Provides security. Machine learning is used for facial recognition and access control.

Pros and Cons

Machine learning is not perfect, like any technology. But developers are already working on problems that hinder users. For example, at Yandex, artificial intelligence is trained using high-quality and verified information. For this purpose, AI trainers are hired to check the information for training.

Advantages of machine learning

Processing large volumes of data. With the help of machine learning, it is possible to "read" a large amount of information. A person cannot master such an array of data and find hidden patterns;
Automation of routine tasks. For example, machine learning automatically sorts emails in your inbox into spam and important messages;
Improving forecast accuracy. Machine learning predicts which products will be in demand and which roads will be congested during rush hour. An approximate forecast allows you to make the right decisions.

Disadvantages of machine learning

Problems with data quality. Machine learning only works with high-quality data. If the data is wrong, then the predictions will be wrong.
Ethics and privacy issues.It is important to use machine learning ethically and responsibly so as not to violate people's rights.

Algorithms can be biased if they are trained on data that contains stereotypes about events and people. For example, a tabloid article that denigrates a media personality is not the best information for training algorithms.

The Future of Machine Learning. Useful materials from the editors of Skillbox.by

There are many myths circulating around artificial intelligence. For example, that neural networks will take over the jobs of IT specialists and replace human labor. The internet is full of unsubstantiated assumptions, so we will share the facts about the present and future of machine learning:

Large-scale changes.Machine learning models are becoming more complex - they require large amounts of data for training. For example, the GPT-3 language model, which learned to write articles for newspapers, contains 175 billion parameters. For comparison, this is more than the number of words in Wikipedia. And now there is a more advanced GPT-4;

Therefore, companies are actively investing in powerful computers and technologies that "feed" models with huge amounts of information.

Accessibility.Many companies share their trained models that can be used in various projects. For example, anyone can connect ChatGPT to their social media account and the neural network will communicate for them;
Machine learning on all devices. Instead of transferring all the data to servers, ML developers have begun integrating models directly into our devices. For example, Apple devices will soon have artificial intelligence built into them, called Apple Intelligence. This AI will be able to reply to messages, generate emoji, and summarize incoming emails;
New Features. Developers are now looking for ways to integrate ML into robots and create real androids. For example, the company Figure has created an android robot with a built-in neural network. It can communicate and execute commands. For example, bringing objects to a person.

For those who want to develop artificial intelligence and learn machine learning, we have compiled a list of useful materials.

References:

«Elements of Statistical Learning: Data Mining, Inference, and Prediction. Second Edition» Jerome Harold Friedman, Robert Tibshirani, and Trevor Hastie - for beginners in this topic;
"Pattern Recognition and Machine Learning" Christopher Bishop — for those who like to understand the essence.

Blogs:

Colah's blog — a blog for those who have already read a couple of books on the topic;
learnopencv — a developer's personal blog about the technical part of the MO with guides.

Games:

Kaggle and DrivenData — small competitive games for practicing coding and working with machine learning.

Master the profession of "Machine Learning Engineer" with Skillbox

A Machine Learning specialist, Data Scientist, or ML engineer analyzes large volumes of data, develops machine learning models, neural networks, and creates large GPT-like language models. There is a demand for such specialists everywhere: in business, medicine, industry, and other sectors.

Get access

What is machine learning and where is it useful?

Contents: