Machine Learning - An Introduction

Lately, it seems that every cutting-edge application of computer science has the same buzzword to hipe the public and try to get all the attention “Machine Learning”. But what is it exactly? Why is it often associated with the words  “Artificial Intelligence”? How does it work? And what is the difference with deep learning? 

What is it?

Machine learning refers to the use of algorithms that use data and experience to improve performance without the need to explicitly program a behavior. The same algorithm can be trained to analyze text, recognize patterns on a set of data, predict results using a set of data as an example, and much more. It can “learn” like a person to recognize patterns on the data provided.

It is used for a lot of applications like recognizing speech, images, searching for patterns on data sets, predicting results, e.g. results on an NBA match using the statistics of the players, predicting future behavior using past data, etc.  

Why is it often associated with “Artificial Intelligence”?

Artificial Intelligence refers to all the technology that gives a machine the ability to behave like a human. The main goal is to develop self-reliant machines that can act and think like humans. As you may have guessed, this is a broad term that encapsulates Machine Learning. People often use it to describe any application that improves and learns on its own but in our case, we are talking about a specific subset of those applications 

How does it work?

It depends on the type of processing we are trying to do. You wouldn’t use a hammer to drill a hole. Similarly, you wouldn’t use the same type of algorithm on different applications. The 3 most important types of Machine Learning methods are:

Supervised learning 

In supervised learning, the data is already labeled, which means we know the target variable. Using this method, the system can predict future outcomes using past data as based for the assumptions. To train the model we need a set of already complete data to train the model before we start to make predictions.

One of the most famous examples of this is the Titanic problem, using the information of sex, age, and a class of ticket of some passengers and the information of whether they lived or not, we can train the algorithm to predict based on age, sex and ticket if a passenger would have made it. 

Unsupervised learning

Unsupervised learning is a method used when we don’t have the data labeled and can see a clear pattern in the dataset. We let the algorithm search for patterns and divide the data according to patterns that it recognizes, it could be any pattern and it may not necessarily separate the data by the same pattern every time.

For example, let's say that we have a group of geometrical forms, triangles, circles and squares of different colors, blue, red, and green, and  3 sizes, the algorithm may separate the shapes by any of the 3 different characteristics, or a combination of 2, depends on what it thinks that the most defining characteristic of the set is.

Reinforcement Learning

This method is used when we need to complete a task in an uncertain environment. The algorithm receives corrections or rewards from the environment and reacts to it. The reward is a measure of how successful the method was in completing the task. 

An interesting example would be an algorithm that plays a computer game, let’s say Mario Bros., we can define the way the algorithm can communicate with the environment, using the game controls, and how the environment rewards the system for the performance, the score on the screen. Every time the system plays the game it learns how to get farther in the game, and in doing so is rewarded by a high score. If we leave the algorithm to play the game, eventually it will win the stage, after learning the needed inputs to get the highest score.

What is the difference between ML and Deep learning

Deep learning is a subset of Machine learning that utilizes complex algorithms and neural networks to train the model inspired by the works of the human brain. The main difference between the 2 is the vast amount of data processed by a Deep learning algorithm and the lack of a need for structured data, that's one of the reasons that DL is used for processing lots of images for a variety of purposes. 

Conclusion

Machine learning is a powerful tool for analyzing, understanding, and identifying patterns in data. At the same time, it provides an accurate insight into how the customer behaves, as well as business patterns, while also helping develop new products. For all of these reasons, this is a tech trend that we need to keep in mind and continue to research, in an effort to make more precise predictions of business needs.