According to Wikipedia:
Machine learning (ML) is the study of computer algorithms that improve automatically through experience. It is seen as a subset of artificial intelligence. Machine learning algorithms build a model based on sample data, known as “training data”, in order to make predictions or decisions without being explicitly programmed to do so. Machine learning algorithms are used in a wide variety of applications, such as email filtering and computer vision, where it is difficult or infeasible to develop conventional algorithms to perform the needed tasks.
So what does the above mean?
When we talk about machine learning we talk about artificial intelligence, because artificial intelligence is the attempt to make a device or an application be as or more intelligent than a human.
Machine learning is a series of algorithms that make a device or application artificially intelligent. Many years ago Alan Touring said that when a computer made you believe that he was a human, that you were talking to a human or interacting with a human you could say that this computer was artificially intelligent.
Therefore, artificial intelligence could be summarized as an adjective to describe those devices to applications that simulate human intelligence through machine learning.
In other words, artificial intelligence is like a car and machine learning is the engine that makes it move.
As its name indicates, it implies a learning stage, in which by means of data collection certain patterns can be found that later by means of code are classified into categories, to better understand this, let’s see the following example:
If you ask several people to think of a dog, surely all of them will think of different kinds of dogs, so if later you show them several photos of dogs and ask them to point out which is the most consistent with what they thought, each one will point to a different one.
Suddenly some chose this:
Or this beautiful guy:
And so on…
Which means that we all have different concepts of similar things. Now imagine that an application is developed that recognizes dogs and only the model was trained with images from the first photo. So when you see an image of the other groups of dogs you will not recognize them as such even though they are dogs.
Why did I give the above example?
Well, as with people, the more experiences and kinds of dogs you know, the greater and the better your recognition will be, that is, you will have a better classification system of what a dog is and what it is not.
Machine learning is something similar to teaching a child who is and who is not a dog, it is a learning process that takes time in which through a series of experiences the child will learn to distinguish, for example, small breed dogs from Large breed dogs and differentiate these with, for example, a cat.
Having already a broader idea about what machine learning is, we are going to give a view of what is necessary to implement it.
What is needed to apply machine learning?
The following are the minimum knowledge required to get started with machine learning
Math: Important to describe the model that describes the behavior of the collected data. Yes, it sounds a bit confusing, but you know everything is described by an equation and the math is accurate for this purpose.
Probability and statistics: They are necessary for data processing and to find or infer important information about the data we are working on.
As for mathematics, this would be the minimum necessary to enter the world of machine learning since with this we can understand how algorithms work in order to know what is going wrong in our training model and be able to improve it to obtain the expected results.
Programming: Algorithms are the fundamental axis of all development and machine learning is not the exception, to understand the fundamentals about them and about data structures (they are the different ways in which the algorithm data is organized to perform a job).
Divisions of Machine Learning:
Supervised: This type of learning is based on discovering the relationships between some input and output variables, in other words, It is to tell the algorithm what the desired result is for a given value. for example:
Let’s explain more clearly, try to guess the following:
if I tell you the following sequence:
1 -> 2
2 -> 4
4 -> 8
5 -> 10
10 -> ?
What do you think is the result of the number 10?
If you thought of 20, congratulations! You have been able by observation to find that the relationship between the first number (input variable) and the second number (output variable) is to multiply the first by two. This means that if I gave you another value for example in 20 you would be able to apply the rule that you have learned to find the result that is 40. This means that through observation you have been able to generalize your knowledge. And this is the key to supervised learning.
Not supervised: By definition it is the paradigm of producing knowledge only from the data that is provided as input without the need to explain to the system what result we want to obtain, it may be difficult to understand how it can be learned without obtaining a previous example, let’s see the following example to better understand the unsupervised learning.
Imagine that you are Indiana Jones and you find yourself in a temple about which you have no previous information, in which you find a series of symbols that seem to be a language which you do not know
But you notice certain patterns in said series of symbols which allow classify them, this problem that you have just solved is called clustering, which is a very important issue within the field of unsupervised training, so without having to supervise your answers you have been able to generate a type of valuable knowledge with the symbols what did you process.
Advantages and disadvantages
Unsupervised learning has an advantage and that is that the data sets necessary to train are less expensive to obtain, since in the case of supervised training, if you need to train an algorithm to differentiate dogs from cats, you need the input images and also For someone to previously classify all the images and classify them one by one as a dog or cat, a task that turns out to be very strenuous since to have a good machine learning model you will need at least 100,000 images of dogs and cats.
What unsupervised learning does is find patterns of similarity in the input data, for example, the symbols of a language.
Going back to the first example of asking a group of people what a dog looks like, each one would have their definition, they all understand the concept of what a dog is, but nevertheless, when we look at the reality we find that this concept has many different variants, so what is a dog really like? This is very interesting because it is something that the brain manages to do automatically and something that unsupervised learning is learning to do, determining whether one thing is similar to another.
Finally, it is not necessary to explain how this learning occurs, to understand how the learning paradigm works, the only thing you need to know is how the machine learning system interacts with the input information and the output information that they produce. which is known as a black box, that is, for us it is not necessary to know what is happening within the model, we are only interested in the interaction of this system with its environment, that is, to know the -what it does- and not the -how it does it-
One of the most popular techniques for developing machine learning is the use of neural networks.
The interesting thing about these is that they can learn in a staggered way, that is, the information is divided by levels, where the first layers learn very specific concepts of an object, such as how a dog’s muzzle is, how its legs are and in subsequent layers use previously learned information to learn more abstract concepts such as, what is a dog? a cat and a rabbit, this means that as we add more layers the information we obtain is more abstract and precise, currently the trend is to add more and more layers which leads to much more complex algorithms known as Deep Learning. Since we are currently immersed in the information age and we are surrounded by a large amount of data, training models based on Deep Learning are necessary to process and classify all this data set.
Neural networks were inspired by the architecture of the biological visual cortex. Deep learning is a set of techniques for learning in neural networks that involves a large number of “hidden” layers to identify features. Hidden layers come between the input and output layers. Each layer is made up of artificial neurons.
In a feed-forward network, the neurons are organized into distinct layers: one input layer, any number of hidden processing layers, and one output layer, and the outputs from each layer go only to the next layer.
In a feed-forward network with shortcut connections, some connections can jump over one or more intermediate layers. In recurrent neural networks, neurons can influence themselves, either directly, or indirectly through the next layer.
So… That's it’s all, Thanks for reading.
References: