Machine learning algorithms

In the world of artificial intelligence, more specifically automatic learning or better known as machine learning, there are a large number of statistical methods and algorithms.

Knowing all of them and knowing when and how to use them is the job of machine learning engineers or data scientists.

This article gives an introduction to different types of machine learning algorithms, briefly explaining what each one consists of.

Stay and find out!

Within machine learning we find 4 main types of models depending on the way the system is trained.

In supervised learning the system is trained using labeled data. For example, if we are training an image classification system we will have to label each one saying what type of image it is.

The model learns to predict an output data from different input characteristics also known as features . In this type of learning we can distinguish between classifiers and regressors.

The goal of unsupervised learning is to discover hidden structures and patterns within a data set. Unlike supervised learning, these methods are not applied to regression or classification problems since there is no data on the output results.

Some of these algorithms are grouping algorithms or better known as clustering. An example is K-means clustering.

Semi-supervised algorithms make use of the combination of unsupervised and supervised learning to improve prediction power.

In different problems it has been proven that, if we use a large set of unlabeled data together with a small amount of labeled data, the results improve considerably.

Reinforced learning or better known in English as reinforcement learning It is a learning system where an agent learns in a simulated environment through trial and error by applying a reward system.

When the agent does a correct action, he or she is given a reward. On the other hand, when she makes an incorrect one, she is penalized.

The objective of reinforcement learning is to maximize the reward function so that the system learns to execute actions that will give it a greater number of rewards.

Below we show you the types of machine learning most used within the artificial intelligence sector.

Classification algorithms aim to classify an instance into a specific class. For example, in banking transactions, there are anti-fraud systems whose function is to classify a transaction as fraudulent or secure.

As we see, the classification is discrete data and can be binary or multiclass. An example of multiclass classification would be a model capable of classifying images of fruits into their corresponding label: bananas, kiwis, apples, pears or peaches.

Regression aims to map a continuous function, with continuous output results. An example would be a machine learning model that would like to predict the exact temperature next week based on data from the current week.

The output results of this model could be either exact numbers or decimal numbers. Therefore, this algorithm would be regression.

Deep learning algorithms are a special type of machine learning that uses deep neural network architectures. These neural networks are made up of neurons, connections between them and different layers.

These types of algorithms are those used for the most complex artificial intelligence problems such as autonomous driving or text generation.

Clustering algorithms are unsupervised methods that aim to group data into various groups based on their characteristics. For example, we could generate clusters from the characteristics of a group of flowers.

The most similar ones, which belong to the same species or very similar species, will be found within the same group.

These algorithms use different mathematical distance metrics to infer how far or close each point is.

It is a type of supervised algorithm that uses decisions or rules to divide the space of predictors by grouping instances with similar feature values.

In this algorithm, the values ​​or thresholds that determine whether an instance belongs to one group or another are optimized. These types of methods can be used for both classification and regression problems.

Lazy learning algorithms are those that do not have a training phase. Inference from the test data is done directly on the training data set.

The most popular example of algorithms based on lazy learning is K-nearest neighbors.

There are multiple algorithms within the field of artificial intelligence. To know which one to use, it is important to understand the advantages and disadvantages of each one, in addition to perfectly understanding the problem to be solved.

However, sometimes we will have to try different models to see which one works best with our data set.

Another factor that will help us decide on one method or another is the experience we have in the world of machine learning. Over the years, creating multiple models for different objectives, we will acquire the necessary experience to know a priori which may be the most appropriate algorithm.