In recent months, companies such as Microsoft, Google, Apple, Facebook, and other tech activists have stated that mobile devices or smartphones have come in second with the advent of artificial intelligence. Today, smart voice assistants and other similar services are increasingly being developed to replace older ways of doing things on smartphones and computers.

Following the popularity of artificial intelligence in the world of technology, the two names machine learning and deep learning are often heard. These are some of how AI can be taught how to do its job, and its application goes far beyond intelligent voice assistants. In the continuation of this article, we will discuss the differences between machine learning and deep learning.

Computers today see, hear and talk

With the help of modern computer learning, they can predict the weather and the stock market, identify users’ buying habits, and control the robots of a factory. Google, Amazon, Netflix, Facebook, LinkedIn, and companies that serve a large number of users all use machine learning. But at the heart of all this learning is what we call an algorithm.

An algorithm is not a complete computer program; Rather, if we want to explain it simply, it is called a limited set of steps to solve an algorithm problem.

For example, a search engine displays search results based on its own search algorithm based on the user input and connection to a database. Achieving such a result requires several steps and steps.

Machine learning began around 1956 with the work of scientists such as Arthur Samuel.

Samuel did not intend to write long and detailed programs for the computer and through it to defeat a human opponent in the game of checkers.

He sought an alternative solution and eventually created an algorithm that allowed the computer to play against it thousands of times. In this way, the computer was able to learn to play against other opponents, and in 1962, it defeated the Connecticut Checkers champion.

As shown in the example above, machine learning is based on trial and error. We are not able to write programs for self-driving cars that recognize the difference between a pedestrian and a tree or a vehicle, But we can write an algorithm for this problem that the computer can use to solve the problem and other data.

Such algorithms are used in other cases, such as predicting the direction of a storm, early detection of Alzheimer’s, detection of the highest and lowest financial receipts of football stars, and so on.

Machine learning typically runs on low-level tools and breaks things down into smaller sections. Each section is solved separately and the final answer is obtained by combining the answers of all sections.

Tom Mitchell, a well-known machine learning activist at Carnegie Mellon University, explains that computer programs are constantly learning from their experiences and that their performance in performing the tasks assigned to them is evolving. Machine learning algorithms allow applications to make predictions and improve over time by using trial and error to make such predictions.

Here are four basic types of machine learning:

Supervised machine learning

In this scenario, the computer program is fed with labeled data. For example, if we want to identify pictures of girls and boys with an image sorting algorithm, pictures of girls with girls ‘labels and photos of boys with boys’ labels are introduced to the computer. These images are used as a set of data for computer training, and the tags remain in place as long as the program can detect at an acceptable rate.

Learning semi-regulatory machine

In this type, a limited number of photos are tagged. The computer program uses an algorithm to provide its best guess for detecting unlabeled images, and then the data is returned to the program for practice; Then other categories of photos are given to the program with a small number of tags. This is an iterative process and continues until you reach the right rate of correct answers.

Learn the car carelessly

This method of machine learning does not involve any labels. Instead, the program randomly selects photos of girls and boys using one of the two algorithms described below. The first algorithm is called clustering and groups objects based on hair length, jaw size, eye location, etc. Another algorithm is called dependency, which allows the program to construct conditional rules based on the similarities it finds. In other words, the program finds a pattern among the photos and arranges them accordingly.

Learning machine booster

Playing chess is one of the best examples to illustrate this type of algorithm. The computer program knows the rules of the game and how to play and goes through the steps to finish the game. The only information provided to the program is the result of winning or losing the game. Then the program continues to play and records its successful moves to finally succeed.

Now that we are somewhat familiar with machine learning, we move on to the deep learning discussion.

Deep learning

Deep learning is actually machine learning at a deeper level. You may say sarcastically that it was also found in his name, But deep learning is actually inspired by the way the human brain works and requires advanced tools such as powerful graphics cards for complex computations and large amounts of macro data. The small amount of data in this algorithm results in poorer results and performance.

Unlike standard machine learning algorithms, which break down problems into smaller parts and then solve them, deep learning solves problems completely. The more data volume and time are available to deep learning algorithms, the better the result.

We are transitioning from a world of mobile devices to a world of artificial intelligence

In the example of recognizing boy and girl photos in machine learning, we saw that these algorithms examined the photos as a batch of data; But in deep learning, the program scans all the pixels of the photos to get all the shapes and edges that may be useful in gender recognition.

The program then prioritizes the registered forms to know their importance in the gender identification process and places them in separate categories.

At a simpler level, machine learning algorithms can detect shapes such as triangles or squares with the definitions we have given them. For example, a triangle has three vertices and a square has four vertices.

In deep learning, the program does not start with predefined information.

Instead, the program tries to identify the number of lines in the shapes or how the lines relate to each other, such as intersecting or perpendicular.

As a result, if the program encounters a shape like a circle, it will recognize that such a shape does not fall into the category of shapes such as squares or triangles.

The deep learning process requires powerful hardware to process the metadata generated by this algorithm. Such hardware is usually located in data centers that, by creating an artificial neural network, can provide the processing power required by various intelligent programs and their metadata.

Also, programs that use machine learning need more time to learn because they operate without human help and shortcuts.

Today, deep learning has come to the aid of human beings in a way that has made it possible to build and improve many services.

Self-driving cars, health care prevention services, and even better movie offerings have all come to fruition today or will soon be fully operational.

According to Nvidia’s Michael Copeland, with the help of deep learning, AI is likely to reach the level of science fiction films we’ve long imagined.

Is the Skynet phenomenon likely soon?

The answer to this question seems to be clearly no, and fans of the Terminator movie will have to wait. The best example of deep learning is the translator program.

This technology can listen to an English speaker and translate it instantly into other languages ​​in the form of text or electronic voice.

Such success, achieved over the years, is due to the slow learning process due to linguistic differences and complexities and the maturation of hardware capabilities over time.

Deep learning has formed the basis of chats such as Cortana, Alexa, Facebook, Instagram, etc.

In social networks, the deep learning algorithm is responsible for introducing new people or pages to users, and these algorithms allow companies to personalize their ads based on different users.

One thing that can ultimately be predicted for the future is the elimination of many of today’s common computer forms from human life. As Google CEO Sander Pichai puts it:

“Over time, the computer, regardless of its form, will become an intelligent assistant to assist you in your day-to-day operations.” “We are a world based on artificial intelligence.”