This article is the first in a series dedicated to Deep Learning: After a general introduction to the functioning and applications of neural networks, you will discover in the following articles the main types of networks and their architectures, as well as methods and various examples of applications of Deep Learning today. Let’s start without further ado our Introduction to Deep Learning .
Deep Learning: Définition et applications
These last few years, a new lexicon linked to artificial intelligence emerging in our society has flooded scientific articles, and it is sometimes difficult to understand what it is. When we talk about artificial intelligence, we often refer to associated technologies such as Machine learning or Deep Learning. These two terms are widely used with more and more applications, but are not always well defined.
We will start by going back to these three essential definitions:
Artificial intelligence: is a field of research bringing together all the techniques and methods that tend to understand and reproduce the functioning of a human brain
Machine Learning: is a set of techniques giving machines the ability to automatically learn a collection of rules from data. Contrary to programming, which is the execution of predetermined rules.
Le Deep Learning ou apprentissage profond : c’est une technique de machine learning reposant sur le modèle des réseaux neurones: des dizaines voire des centaines de couches de neurones sont empilées pour apporter une plus grande complexité à l’établissement des règles.
Deep Learning: is a machine learning technique based on the model of neural networks: dozens or even hundreds of layers of neurons are stacked to bring forth greater complexity to rule making.
What is Machine Learning: Supervised and unsupervised learning, don’t panic, we’ll explain it all !
Machine Learning is a set of techniques which enables machines to learn, unlike programming which consists of executing predetermined rules.
The two main learning types in Machine Learning are supervised and unsupervised learning.
In supervised learning, the algorithm is guided with prior knowledge of what the output values should be. Consequently, the model adjusts its parameters in order to reduce the gap between obtained results and expected results. The margin of error is thus reduced over the course of model training, so as to be able to apply it to new cases.
However, unsupervised learning does not use labeled data. It is therefore impossible for the algorithm to calculate without any doubts a pass score. Its goal is thus to deduce the groupings present in our data.
Let’s use the example of a set of data of flowers, we try to group them into classes. Here we do not know the species of the plant, but we want to try to put them in a group together:
Example: if the shapes of the flowers are similar then they are related to the same corresponding plant.
There are two main areas of domains in unsupervised learning in order to find clusters:
- Partitioning methods: k-means algorithms.
- Hierarchical grouping methods: ascending hierarchical classification (CAH)
What about Deep Learning?
Deep Learning is one of the main technologies of machine learning. With Deep Learning, we deal with algorithms capable of mimicking the actions of the human brain thanks to artificial neural networks. Networks are made up of tens or even hundreds of “layers” of neurons, each receiving and interpreting information from the previous layer.
Each artificial neuron represented in the previous image by a circle, can be seen as a linear model. By interconnecting neurons as a layer, we turn our neural network into a very complex non-linear model.
To illustrate the concept, let’s take a dog-cat classification problem from an image. During training, the algorithm will adjust the weights of the neurons in order to reduce the difference between the obtained results and expected results. The model will be able to learn to detect triangles within an image since cats’ ears are more triangular ears than dogs’.
Why do we use Deep Learning?
Deep Learning models tend to work well with a large amount of data, whereas more traditional machine learning models stop improving after a saturation point.
Over the years, with the emergence of big data and increasingly powerful computing components, power and data intensive Deep Learning algorithms have overtaken most of the other methods. They seem to be ready to solve many problems: recognising faces, defeating go or poker players, allowing the driving of autonomous cars or even finding cancer cells.
Implication of artificial intelligence
Almost all industries are affected by AI. Machine learning and Deep Learning play an important role in this.
Whether you are a healthcare professional or a lawyer, it is possible for a highly autonomous model to assist you or even replace you one day.
In the health professions, there are already applications for automatically diagnosing a patient.
The automotive industry has also been shaken up with the arrival of assisted driving.
It is also using Deep Learning that Google’s Alpha Go model succeeded in beating the best Go champions in 2016. The search engine of the American giant is itself more often based on learning by Deep Learning rather than on written rules.
Today, Deep Learning is even capable of “creating” a painter’s painting on its own. This is called Style Transfer. If this topic interests you, an article entirely devoted to this topic will soon be on our blog!
Next, we will introduce you to neural networks with a new approach that we hope you’ll enjoy !
Article 1 : Introduction au deep learning
Article 2 : Réseaux de neurones: Biologique VS Artificiels
Article 3 : Convolutional neural network