In recent years, artificial intelligence (AI) has been receiving a lot of media attention. The focus of much of this attention has been directed at a particular AI sub-discipline: deep learning.
But what exactly is deep learning?
Deep learning involves convolutional artificial neural networks. Does that sound like a foreign language to you? Don’t panic: let’s decipher it together!
First, you need to know what an artificial neural network is. Then we’ll look at convolution. Once these two concepts are defined, you’ll grasp deep learning, no problem! Ready?
The Inspiration Behind Artificial Neural Networks
In the early days of artificial intelligence, the goal was to replicate the human brain or, more accurately, to draw inspiration from it. Researchers focused first on the functioning of the biological brain, specifically concerning neurons. These neurons communicate with one another using synapses.
Algorithms were developed in an attempt to simulate this neural architecture. Of course, artificial intelligence doesn’t recreate neurons; it simply draws inspiration from them.
Let’s see what that looks like!
What are Artificial Neural Networks?
To get a clearer picture of how artificial neural networks work, let’s look at an example of how they are used in image recognition. We’ll focus on image sorting. Digital photos tend to pile up, and sorting through them can become a real chore. Like in the video with cats and dogs, we’re going to build an artificial intelligence program that will do the sorting. Its mission will be to tag the people pictured below with their names.
To train an artificial intelligence system to identify family members who appear in snapshots, gather a series of photos of each individual.
To analyze family photos, we'll use an artificial neural network.
An artificial neural network can be imagined as a box with:
An entrance: the data input (a photo).
An exit: the expected output (the first name of the person in the photo).
A neural network (between the two): a series of layers of artificial neurons.
So, should we use supervised or unsupervised for this?
You can use deep learning with both supervised and unsupervised learning. In this instance, we are dealing with a typical supervised learning problem. We want to train an algorithm on annotated examples.
What is going on in this neural network? How does the learning take place?
Neurons make up the network, grouped into three different layers: input, hidden, and output.
The Input Layer
Thus, if an image has a million pixels, the first layer will be made up of a million neurons.
The Hidden Layers
The results of these mathematical computations—known as activation functions—determine whether a neuron will be activated or not, as each neuron takes input data from the preceding layer and returns a 1 (neuron activation) or a 0 (non-activation).
The Output Layer
In our example, this will be the name of the person in the photo.
We can represent the three types of layers schematically as follows:
OK, the neural network receives images as input and delivers a response as output. But how does it learn?
For the algorithm to learn, it must train itself by passing each photo through its different neuron layers until it obtains a result. In the beginning, the responses will be somewhat random because the algorithm has no experience!
As training continues, the network will change the way the neurons transmit information to each other to arrive at the correct image categorization. The images pass through the neural network one by one. The network makes adjustments as needed to annotate them correctly.
What Are Convolutional Neural Networks?
We have finished training our neural network. Good news: it’s very good at sorting our portrait photos!
We need to improve our AI tool so that it can tag more than one person in the same image.
To do this, we will use a special neural network called a convolutional neural network (CNN).
We will move sequentially from one part of the image to the other, making passes.
Deep Learning: Artificial Neural Networks and Convolution
Deep learning came about when scientists combined artificial neural networks with convolution.
What distinguishes deep learning and makes it so popular?
Its impressive results and future potential! One of the key advantages of deep learning is its ability to process huge volumes of data. The more data these neural networks receive, the more they can improve. Some image recognition systems are trained on datasets containing tens of millions of images.
Let's Recap!
Deep learning is an especially powerful sub-discipline of machine learning.
It involves the construction of brain-inspired systems of artificial neural networks.
It also uses the notion of convolution, which involves analyzing images step by step using a filter to make it even more powerful.
You now understand the fundamentals of Machine Learning and Deep Learning and the different stages involved in an artificial intelligence project. The latest applications of AI that everyone is talking about, such as general-purpose and generative AI, are based on these techniques. That's the subject of the next chapter.
