Machine learning
Machine learning is the use of mathematical algorithms by which a computer is able to to be trained without direct instructions and interventions.
A lot of scientific articles have been written about machine learning and active work is underway to study and develop it, it is considered one of the forms of artificial intelligence. In machine learning using mathematical data reveal patterns, with the help of which self-learning takes place.
In this article, we will try to understand with you in simple language what it really is machine learning, let's analyze the question without using scientific terminology, but let's take examples of real tasks, and we will analyze their solution.
The purpose of machine learning
The main purpose of machine learning is to be able to predict a number of events. Based on mathematical functions, an algorithm is determined by which a conclusion is made on a particular event.
As an example, let's take the purchase of a car. We can independently study the market, view ads and draw conclusions about the cost. For example, a new car costs $20,000, a car, which year - costs $19`000, which two years - $18`000 and so on.
Based on the information received, we derive a formula that shows that the initial price of the car $20,000, and decreases by $1,000 each year.
Such data analysis in machine learning is called regression – predicting an event from known data. But, as a rule, there are many more non-obvious factors in life that need to be taken into account in order to make a correct conclusion and avoid mistakes.
To solve more complex problems, robots were created that use mathematical functions and algorithms to find patterns in the data obtained and learn to predict the answer. Based on the work done data were obtained that showed that there are such patterns that a person does not even guess about.
Thus, machine learning was created.
Components of machine learning
There are three components by which the result is predicted. The more diverse the data obtained, this will make it easier for the machine to find a number of patterns and get a more accurate result.
Let's analyze each component separately.
Data collection
To make a conclusion, first of all we need to collect data. Examples are needed to identify spam spam emails, in order to predict the stock price, it will be necessary to reflect the price history, in order to analyze a person's interests, you need to get data about his actions, for example, in social networks. The more data, the more accurate the result. Thousands of examples are the minimum with which it is possible make an approximate conclusion.
Data is collected in all possible ways. In some cases, they are made manually, so it takes longer, at the same time, the number of errors is reduced. Or automatically, when all kinds of data are loaded that it turned out to be found with the belief that the correct result will be obtained.
For example, Google analyzes the data of its users in order to receive them for free , according to which the analysis takes place in the future.Obtaining reliable and high-quality data is the number one task for large corporations and companies, which, in turn, only under rare pretext disclose the information received.
Signs
Signs are understood as the properties and characteristics of a particular phenomenon, object or person. These can be signs such as the gender of the user, the stock price or the frequency counter of the word in the text. All of the above is a sign and based on the received signs, an analysis is performed.
The machine needs to know what needs to be analyzed. The ideal machine learning option is when the data they are in the tablets, there is their name and all the necessary information. Otherwise, when there are too many signs, the model starts working less efficiently and slower. In this regard, the process of selecting the right signs often takes longer than the rest of the training. But there are reverse situations when connected the human factor. A person, based on his beliefs and attitudes, begins to select himself, according to his according to the "correct" signs, thereby introduces subjectivity into the model, and in this case the machine begins to lie.
Algorithm
It is often possible to solve the same problem by different methods. Accuracy and speed depend on the choice of the method works and size of the finished model. At the same time, there is one caveat: if the data was obtained incorrectly, then even the best algorithm will not be able to cope with the task. It is important not to focus on percentages, but to collect more the amount of data.
Intelligence and learning
It is important to see and understand the difference between artificial intelligence and machine learning.
Artificial intelligence is a name that covers the entire field of science. In which they already exist sections such as neural networks, machine learning, etc.
Machine learning, in turn, is a branch of artificial intelligence. It is an important element, but he's not the only one.It is impossible to compare artificial intelligence and machine learning, these are things that are at different levels of science. But we can compare what has been achieved in this area to date, and see what the machine can do.
| The machine can | The machine cannot |
|---|---|
|
|
The world of machine learning.
It is possible to classify algorithms in a dozen different ways. It does not happen that one or another problem is solved only one method.
Currently, there are only four main directions in machine learning:
| Classical training | Simple data, clear signs |
| Reinforcement Learning | There is no data, but there is an environment with which you can interact |
| Ensemble methods | There is data, but their quality is very low |
| Neural networks and deep learning | Complex data, unclear signs |
