A short introduction to machine learning for beginners
Machine learning is a branch of artificial intelligence based on the idea that systems can learn.
Lady Marcela Sánchez Moreno
Machine learning is a branch of artificial intelligence, a product of computer science and neuroscience, which is responsible for developing and creating systems or techniques that enable a computer or machine to have the ability to learn.
The machine or computer, together with an algorithm designed for learning, learns without being explicitly programmed, as explained by Arthur L. Samuel in 1959, who was a pioneer in demonstrating the concept of artificial intelligence in its early days.
To be a little clearer with this definition, through algorithms that are a sequence of instructions that lead to the solution of a problem, the machine or computer can make relevant discoveries about a group of millions of data that reviews by identifying and analyzing complicated patterns and with them can predict behavior after this action, as acceptable output values within a range that is given by the algorithm (general and specific tasks) being able to learn as a human would, leaving aside the need to depend on precisely a human to do by code or instructions by the machine.
This implies that in the future the machine will also be able to improve autonomously without human intervention in its learning, this could be seen as if the machine develops “intelligence” in a certain way.
Going a little deeper, what are the algorithms of machine learning?
There are three types of algorithms that help in machine learning: supervised, unsupervised and reinforcement.
The main ingredient for machine learning to work is data. -We will talk about the functionality of machine learning in the real world later on.
How does machine learning work?
In the past, the only way for a machine to do anything was to send it a series of instructions specific to each action. Today the algorithms used for the development of machine learning, these are mostly automatic, because when collecting the data, the machine makes its own calculations and the more data, the results can be a little more accurate and better, all new data becomes an algorithm and this intervenes in the complexity and effectiveness of the machine and how it works.
An interesting thing about how the machine learns is that there is a kind of data (structured and unstructured) that you have to understand before you can understand it.
Structured data: This data is used by companies, usually in databases, and is organized in a systematic way, so it can be easily processed. For example an excel or access file, where you can find the information of the clients that manage in a company, with all their data, something like a virtual file.
Unstructured data: These are binary data that, since they are not organized, can be difficult to identify at first and without structure, until they are processed, they have no use and could not be stored in an original way.
For example, the e-mails, when these have been filtered, can be located easily and can have a classification to be able to obtain information about them, in this case, this information in mass would be processed word by word, for its comparison and classification according to the patterns that are looking for.
At this point, it is necessary to be extremely careful with how the data is extracted and processed, which, together with machine learning, organizes the data for its proper handling, especially the unstructured data.
Knowing this, let’s talk in a little more detail about the algorithms used for machine learning.
1. Supervised learning:
In this type of learning, the machine is trained where it is provided with a quantity of data that is defined with detailed labels. Once it is provided with a sufficient amount of data, new data could be introduced without the need to label anything, since the machine will be able to identify the different patterns that in its training register, this is known as classification.
Another method consists of predicting a continuous value, using different parameters that when combined in the introduction of new data, lead to the prediction of a certain result, this is known as regression.
In this type of learning, different examples are used from which new cases of data analysis can be generated.
2. Unsupervised learning:
In this type of learning, no labels are used as in supervised learning. Here the purpose is to abstract patterns of information directly and for the machine to be able to understand them, this is known as data clutter, this training is very similar to the way humans process information.
With this method, loss of information is avoided and the data is visualized for better understanding.
3. Learning by reinforcement:
In this type of learning, the machine learns through experience, reinforcement. It is based on trial and error and the method of reward that reinforces optimizes and makes effective the way the machine performs its tasks and how it behaves.
It is one of the most interesting ways of learning machine learning because large amounts of information are not introduced.
In this case, the machine knows what the result will be from the beginning, but not knowing what decision to take to get those results, the algorithm evolves and associates the patterns with which it achieves success, repeats them, perfects them to make them effective.
Like the previous method, it is very similar to how humans learn, over time the machine will improve and make mistakes until it finds the best way to solve the situation it faces.
explaining a little about how machine learning works and its different types, we can specify some of the algorithms that help in this task.
There are 8 different machine learning algorithms, in this case, I talk about the 4 algorithms he considers most important and relevant:
1. Linear Regression:
Linear regression is a supervised learning algorithm used in Machine Learning, statistics, prediction and data forecasting. This program must evaluate and understand the relationship between the variables, what basically is done to exemplify this is to draw a line that shows us how the variables mentioned above are related (x and y, perhaps on a Cartesian plane), the analysis of this regression focuses on how changeable the variables can be.
2. Logistic Regression:
Logistic Regression is a supervised algorithm used to classify two classes or variables. With it you can calculate the possibility of something happening, its implementation is easy and simple, it is the basis of any binary classification problem.
There are types of logistic regression such as binary logistic regression, this has two possible results, multinomial logistic regression, which predicts between three or more categories, and ordinal logistic regression, which has the ability to classify between several multivariate elements.
The results of logistic regression are discrete and to say the least, like yes or no.
3. Decision Trees:
A decision tree is a supervised learning method that combines regression and classification. It is one of the most used methods in machine learning, it is simple and powerful at the same time. It works like when we make any decision in our daily life.
These graphical representations have a node called “root” and then they are detached from the rest of the attributes in two branches that can have a true or false possibility, depending on the decision that the machine takes in its learning process, everything forks two by two until arriving at the final nodes, that are equivalent to the answer, yes or no, true or false.
During all this process the algorithm usually measures somehow the predictions it made, evaluates them and compares them to obtain, as previously mentioned, the best decision.
4. Support Vector Machines:
Vector Support Machines (SVM) is a supervised learning algorithm that creates a line between different categories of data.
This vector is calculated, using as reference the dividing line to see which group of data is closer or further away from it. It is discriminatory because it defines the limits of the decision to be made, in this process is also made a mapping of the data for its organization and classification (binary or regressive).
The training of the vector machine consists of two phases: the first one transforms the input data into a dimensional feature space, it is called kernel trick.
The support vector machines belong to a class of Machine Learning algorithms called kernel methods and are also known as kernel machines, then classifies these characteristics into two classes, this numbering depends on the number of support vectors and finally builds the support vector space.
Solve a quadratic optimization problem that fits an optimal hyperplane to classify the transformed characteristics into two classes. The number of transformed characteristics is determined by the number of support vectors.
Only the support vectors selected from the training data are required to build the decision surface. Once trained, all other training data are irrelevant.
Machine learning is a tool that gives us the possibility of developing a better and innovative future, giving you the possibility of having technology with “intelligence” that can be useful in your daily life, it is also a discipline that has become indispensable for programmers and software engineers who want to create the future through technology.
Although many people thought that it would never be possible for a machine to think similarly to the human, this would not be possible with technological advances, we are much closer to this utopia, because now we see the ability of machines to interact with the world to an accelerated pace, as well as your learning will, with the rapid growth of the data we produce and the passage of new technologies.
Lady Marcela Sánchez Moreno
Software Engineer Student at Holberton School Colombia 💻 Enthusiastic weaver of macramé and crochet 🧶 Colombian Red Cross volunteer ❤