The terms Machine Learning and Deep Learning will be often put in the same basket, but what are they and what is their role? To understand these aspects, the first step is their positioning within the larger umbrella of AI (AGI).
To address the limitations, characteristics and differences of these fields, it is necessary to know what an algorithm is since it is the raw material of artificial intelligence, and therefore machine learning and deep learning.
What is an algorithm?
An algorithm is a set of instructions that solve a problem. These instructions have to be finite, ordered and logical, in other words, they cannot be an infinite number of instructions. They are sequences and must be unambiguous.
When designing an algorithm that meets these characteristics, it is possible to program it on a computer, which can follow instructions and solve a given problem. A good analogy is a cooking recipe, since it complies with a set of finite instructions, a sequence and logic, with the aim of preparing a dish.
In that case, what is “AI”?
Starting from the point in history in which computer theory was formalized, AI has been normally defined as an area where you can investigate the design of processes that can solve problems represented mathematically, and automate these processes through computers that affect the ability to solve problems and often exceed the ability of humans.
Comparing the capacity of these teams with human and surgical intelligence, the following question arises: Are the teams smart? From a psychological perspective, there are multiple intelligences and variables, frequencies of contexts and the sciences that study this concept. So, there is no exact definition of artificial intelligence.
In principle, we can say that artificial intelligence is a new field of research within computing that is dedicated to designing algorithms that solve problems through computer learning. Each algorithm has its own methods and characteristics, and due to the above, they have different performances to solve specific problems.
In the last 50 years of human history, a large number of algorithms have been designed, which can be called “intelligent”, and since there are so many, it is necessary to categorize them according to their properties and the types of problems for which they provide solutions. The most outstanding category in the recent years is machine learning
Okay…what is machine learning, then?
It is a subfield of artificial intelligence that has selected algorithms for data analysis, learns from them and performs different tasks such as classification, prediction and pattern groupings, among others. This process is carried out through a large data set, through which these algorithms can be trained.
Within machine learning, we have two large groups: supervised and unsupervised learning.
It has a set of labelled data to train the algorithm, and what the algorithm is looking for is evaluating an unknown pattern, and deciding the class to which it belongs.
For example, if we want to classify which fashion season a garment belongs to, what we need is to have a set of garments with their characteristics and labels according to the season they belong to: spring / summer season or autumn / winter season. These characteristics can be: color, material, texture, design, brand, type of ironing, among others.
With this information, we can train the supervised learning algorithm to be able to decide which season a garment belongs to without having known it before.
In the case of unsupervised algorithms, their task is to find relationships between patterns that have not been previously labelled. In the case of the previous example, if the clothes were not labelled by fashion season, the algorithm would have to label the fall / winter season clothes by itself and differentiate them from the other class.
It should be noted that many of the algorithms that make up this field were created years ago, but due to the low computational capacity and lack of data, their use was not possible.
The endgame: what is deep learning?
It arises from the search to simulate the learning process of humans using the knowledge we have of the functioning of neural networks in the brain. In this way, it was attempted that computers learn in the same way as humans.
The first artificial neural network was created in 1943 by neurologist Warren McCulloch, and Walter Harry Pitts, who were engaged in computational neuroscience. From this moment onwards, several relevant works were published, but it was not until the 1980s that they began to have great relevance in the world of research, and it was in the last ten years when they entered into business practice.
A neural network is made up of a set of neurons connected to each other, each neuron has the ability to learn and transmit that knowledge to other neurons. In the following image you can see what a network is like, where the dots represent the neurons and the lines connecting the nodes between them.
Due to the fact that, in recent years, the computational and storage power has increased, it has been possible to create artificial neural networks with hundreds of thousands of neurons, which has led to naming the field of research dedicated to the study of these networks “Deep learning”.
Currently, deep learning is within the field of machine learning because neural networks solve the same type of problems as algorithms in this field, however, the area is growing rapidly and generating multiple branches of research.
Neural networks are mostly used to solve problems of image classification, natural language analysis, such as speech recognition and text creation, for example, creating the news, poems or micro-stories.
Provide your comments below