Artificial neural networks mimic the functions of neurons in the human brain in order to solve machine learning problems. The fields of application are numerous and are most often characterized by an input-output relationship of the information datum.
Table of Contents
What Is A Neural Network?
Designated in English by “ artificial neural network ”, network of artificial neurons in French, it is a hardware or software computer system (or both at the same time), proceeding by imitation of the functioning of the neurons of the human brain. It is a branch of deep learning technology, a subcategory of artificial intelligence or machine learning.
Birth Of Neural Networks
It was in 1943 that the concept of artificial neural networks was invented at the University of Chicago by two researchers. The latter put forward their theory in an article in the journal Brain Theory, which indicates that the activation of neurons is the basic unit of brain activity.
Operation Of An Artificial Neural Network
Neural Network: How Does It Work?
Generally, a neural network is based on a large number of processors that operate in parallel and organized in layers. One of these layers receives raw information inputs, much like the human optic nerve when processing visual signals. Then, each layer receives the information outputs of the previous layer, a process found in humans. This is, in fact, what happens during the transmission of information between neurons via the synapses, particularly near the optic nerve. The results of the system are produced by the last layer of neurons.
Operation Of An Artificial Neural Network
Neural Network: How Does It Work?
Generally, a neural network is based on a large number of processors that operate in parallel and organized in layers. One of these layers receives raw information inputs, much like the human optic nerve when processing visual signals. Then, each layer receives the information outputs of the previous layer, a process found in humans. This is, in fact, what happens during the transmission of information between neurons via the synapses, particularly near the optic nerve. The results of the system are produced by the last layer of neurons.
Neural Network And Learning
At first, the artificial neural network will have a very low prediction performance. The neural network will therefore have to go through a learning phase based on observed data. This learning is done using the backpropagation algorithm. From the predictions of the model and comparing them to reality, the neural network will learn to better predict the outputs. For example, a neural network can learn to recognize objects. For this, it observes a large number of objects belonging to the same category knowing the type of object. Then, the neural network will be able to predict the object present on new images, which is possible thanks to the identification of recurring patterns within the images. This makes it possible, for example, to learn to recognize a cat in any photo by analyzing thousands of photos of cats.
What Are Neural Networks Used For?
The use cases of artificial neural networks are numerous and continue to multiply over time. It can be used in particular for handwriting recognition, financial market forecasting or even the conversion of audio or voice to text (speech-to-text). The use of neural networks is also possible for facial recognition, cancer detection on medical imaging and weather prediction. It is also used to train chatbots, develop natural language processing algorithms or discover new drugs.
There Are Different Types Of Neural Networks
They are most often categorized according to the architecture of the intermediate layers. We can also differentiate neural networks by the number of thicknesses separating the input of data from the production of the result, by taking into account the number of hidden nodes of the model or by considering the number of inputs and outputs of each node.
Feed Forward Neural Network
There is a variation in the propagation of information between the different layers of neurons. In the so-called “ feed forward ” neural network , information passes directly from the input to the processing nodes and then to the outputs.
Recurrent Neural Network
Recurrent neural networks, on the other hand, save the results produced by the processing nodes and use these results to feed the model.
Convolutional Neural Network
There are also convolutional neural networks which are commonly used in a multitude of fields, such as facial recognition, natural language processing and text scanning.
Also Read: Digitize Business Processes With AI
