top of page

The use of Generative AI / ML for Media

presentation Glossary

Agent

Physical or virtual entity capable of perceiving its environment and acting on it. The agent can be, for example, software or a robot. It can also evolve with other agents, especially in a multi-agent system, where it interacts with them to solve a problem.

Agent autonome

An agent that determines its own actions and state, without the intervention of a third party.

Agent cognitif

Autonomous agent that symbolically represents its environment, from which it reasons, and whose behavior is voluntary, with an explicit goal and plan.

Agent intelligent

An agent that uses the resources of artificial intelligence, including learning mechanisms, and adapts its actions according to experience and acquired skills. Intelligent agents are used for the representation of a user, the search for information, the routing of information in networks, the sorting of emails, etc. In addition, they can intervene in the processes of simulation and decision support.

Agent réactif

An agent whose actions are in immediate response to external stimuli, not having the ability to reason or represent its environment. Reactive agents cannot take into account past events or plan their future actions; they depend on their real-time perception of their environment.

AI

Artificial intelligence - A field of research aimed at creating machines capable of imitating human abilities, such as thinking, reasoning, and learning.

Algorithme

An ordered set of instructions for performing a process or solving a problem.

Algorithme d'apprentissage

An algorithm whose purpose is to modify the behavior of a robot, an agent or an artificial neural network from a large amount of learning data that belongs to the set of big data. The combined use of learning algorithms and big data has made it possible, for example, to solve certain problems related to the operation of the autonomous car.

Algorithme génétique

Method where we study a set of possible solutions and where the least efficient solutions are eliminated. Sequence of operating rules executed on data and which allow a result to be obtained.

Algorithme prédictif

An algorithm that associates heterogeneous data from past events in order to predict future behaviors. For example, by associating data from various sources, the predictive algorithm produces personalized search results, based on the presumed interests of Internet users.

Analyse des mégadonnées

A research technique that consists of analyzing large volumes of data using algorithms, specialized computer tools, or artificial intelligence systems, in order to obtain information useful for action or decision-making. Big data analysis makes it possible to highlight correlations and underlying structures that are difficult to detect among a mass of raw data. It provides elements of understanding.

Apprentissage

The process by which a computer, by processing and using data using algorithms, acquires new skills to perform new tasks. The study of the mechanisms by which a machine is likely to learn is at the heart of artificial intelligence. The development of knowledge about learning, and in particular about deep learning, enables the manufacture of computers capable of better adapting to various situations and responding to them more effectively.

Apprentissage par renforcement

A machine learning mode that consists of sending a signal to a computer indicating whether the answer it proposes is correct or not in a given context, in order to maximize its performance by seeking, at each step, the best possible action. Reinforcement learning is opposed to supervised learning and unsupervised learning. This learning method requires many trials and errors.Reinforcement learning is used for example to train machines to play games.

Apprentissage profond

A machine learning mode generally performed by an artificial neural network composed of several layers of neurons which, by interacting with each other, allow computers to learn progressively and effectively from big data. Deep learning is inspired by knowledge in neuroscience. In particular, it facilitates the acquisition of complex rules and the perception of signals (images, videos, sounds, speech, etc.). Research in deep learning applies, among other things, to speech recognition and pattern recognition as well as to robotics, computer vision and the automatic processing of languages.

Apprentissage supervisé

Machine learning mode for creating rules from a predetermined classification model and labeled examples. Supervised learning is opposed to unsupervised learning and reinforcement learning.

Approche multiagent

An approach to describing, modeling, or analyzing a complex system in which agents interact to solve one or more problems. The multi-agent approach falls under distributed artificial intelligence. It is also the main approach used in multi-agent systems. The main interactions between agents are communication, collaboration, and negotiation.

Assistant virtuel

Software designed to answer questions that are sent to it or to perform tasks using natural language. When virtual assistants are third-party software integrated into an instant messaging service, we speak more specifically of a conversational agent.When the virtual assistant integrates a voice recognition engine as well as a speech synthesis module, and therefore the interaction between the program and the user goes through the use of the voice, we use more specifically the term personal voice assistant.

Base de connaissances

Database containing all the information integrated into an artificial intelligence system. The knowledge base is generally part of a knowledge-based system.

Biais

Numerical value, added to or subtracted from the weighted sum of the input signals of an artificial neuron, which enters into the calculation of the activation function.

Big Data

A set of data produced in real time and continuously, structured or not, and whose growth is exponential. Big data, because of its size, becomes impossible to manage with traditional database management tools. They come in particular from social media, smartphones, electronic transaction statements, public data put online, photos and digital videos transmitted online, signals from GPS location systems, etc.

Blockchain

Secure, transparent, and decentralized information storage and transmission technology.

Chatbot

Computer program capable of simulating a conversation with a human.

Coefficient synaptique

Number which, by multiplying the different values of the signals received at the input of an artificial neuron, serves to calculate the value of the signal emitted at the output.

Cognition artificielle

A field of research aimed at reproducing human cognitive abilities, such as thinking, reasoning, and learning.

Connaissance

Set of information integrated into an artificial intelligence system. Knowledge can be either predetermined or acquired through learning.

Connexion synaptique

Link between two neurons in an artificial neural network.

Connexionnisme

Field of study that is interested in the creation of systems, such as artificial neural networks, in which, inspired by the functioning of the brain, potentially complex mechanisms are modeled by having many simple interconnected processing units interact.

Couche cachée

Layer of neurons grouping those that have no direct link with the outside of the network of which they are a part. In artificial neural networks, there can be several intermediate layers between the input layers and the output layers. These layers are said to be hidden because they remain invisible from outside the network.

Couche d'entrée

Layer of neurons grouping those that receive signals from outside the network of which they are a part.

Couche de neurones

Grouping of artificial neurons within a network. In artificial neural networks, signals can successively pass through several layers of neurons. We generally oppose the input layers, which receive signals from outside the network, to the output layers, which transmit them to the outside.

Couche de sortie

Layer of neurons grouping those that send the signals they process outside the network of which they are a part.

Data Mining

Data mining allows you to analyze a large volume of data and highlight patterns, correlations, and trends. Even if you can do data mining without machine learning or deep learning, the most advanced software generally integrates these features today.

Deep learning

A type of machine learning that uses complex artificial neural networks to learn from data.

Données

Set of raw or processed information.

Données d'apprentissage

Data used to train a learning algorithm. In general, the better the quality of the training data, the better the algorithm for making predictions will be.

Entrainement

The process by which a machine learning model learns from data.

Exploration de données

A process of data research and analysis that allows for the discovery of trends or correlations often hidden within big data, the detection of strategic information, or the discovery of new knowledge using statistical methods. Data mining can be used, for example, to determine the criteria that drive product purchases, the source of defects detected during manufacturing, or the demographic to target for a mailing.

Fonction d'activation

Mathematical function that is used to calculate the value of the signal emitted at the output of an artificial neuron from the weighted sum of the different values ​​of the signals at the input.

Informatique Cognitive

Cognitive computing is a more nuanced way of talking about Artificial Intelligence. IBM uses this expression to express itself around its Watson platform. Cognitive computing is mainly used by marketing services who find that the expression "artificial intelligence" has a negative connotation.

Intelligence artificielle (IA)

A field of study whose object is the artificial reproduction of the cognitive faculties of human intelligence with the aim of creating systems or machines capable of performing functions normally belonging to it. Artificial intelligence touches on many fields, such as cognitive sciences and mathematics, and on various applications, including pattern recognition, problem solving, robotics, video games, and expert systems.

Intelligence artificielle distribuée

A field of study focusing on the design of autonomous agents, the distribution of knowledge between them for the collective accomplishment of one or more tasks, and the development of multi-agent systems.

Intelligence artificielle faible

An artificial intelligence system designed to imitate a specific portion of the functioning of human intelligence, allowing it to reproduce certain human behaviors in order to accomplish one or more particular tasks.

Intelligence artificielle forte

An artificial intelligence system designed to imitate the functioning of human intelligence as a whole, and having the ability to question, analyze, and understand its reasoning.

Internet des objets

Computing concept whereby everyday objects or places in the physical world can be connected to the Internet and recognized by other objects. A connected object collects data (temperature, speed, humidity, etc.) thanks to sensors and sends it via the Internet so that it can be analyzed by computers.

Langage de représentation des connaissances

Formal language for encoding knowledge by means of numbers, signs, and symbols, so that they can be used by an artificial intelligence system. A knowledge representation language is essentially characterized by its syntax, i.e. the signs and rules that structure the assertions, and by its semantics, i.e. the way in which the assertions are interpreted.

Langage naturel

Natural Language: Language used by humans to communicate with each other.

Machine Learning

Machine learning is a sub-category of artificial intelligence. It is a process that allows computers to improve through learning. Data analyst Arthur Samuel believes that machine learning allows computers to "learn without being explicitly programmed".

Modèle

Mathematical representation of a system or phenomenon.

Neurone artificiel

Basic unit of an artificial neural network whose role is to convert the information-carrying signals it receives into a single signal that it transmits to other units in the network or that it directs to the output. Originally, the inventors of the artificial neuron were inspired by the biological neuron by trying to give it a mathematical model.

Perceptron

Artificial neuron model in which the received signals are first weighted, then added and finally collectively transformed, using a mathematical formula, into a single signal emitted at the output. The perceptron was originally designed as a binary classifier to determine whether or not an element belongs to a class of objects. It served as a model for the design of the artificial neural networks that followed. It has an error correction mechanism.

Perceptron monocouche

Perceptron consisting of an artificial neuron in which the inputs are directly linked to the output to form only one layer.

Perceptron multicouche

Artificial neural network, built by drawing inspiration from the functioning of the perceptron, in which the neurons are grouped into an input layer, one or more hidden layers and an output layer. In a multilayer perceptron, the signals propagate in only one direction.

Raisonnement

A process by which a computer system performs a logical sequence, starting from initial propositions and a knowledge base, in order to arrive at a conclusion. We find in particular deductive reasoning, inductive reasoning and abductive reasoning.

Raisonnement abductif

Reasoning that makes it possible to increase the likelihood of a hypothesis by adding new facts.

Raisonnement déductif

Reasoning which consists of relating several initial propositions to arrive at a logical conclusion.

Raisonnement inductif

Reasoning by which general rules are drawn from particular facts.

Reconnaissance d'images

Image recognition - Technique that uses the methods applied in pattern recognition and which allows a computer system to automatically recognize the content of an image submitted to it. Image recognition makes it possible, for example, to identify a face or an object, to determine the number of people in a group, the breed of a dog in a photo. To do this, the system first analyzes thousands of images in order to build a database.

Reconnaissance des formes

Pattern recognition - The pattern recognition system is developed by machine learning techniques, including supervised learning and unsupervised learning. The fields of application of pattern recognition are very varied, ranging from the optical reading of printed or handwritten texts to the programming of robots that operate in changing environments.

Représentation des connaissances

Knowledge representation - Process which consists of encoding and storing knowledge, so that it can be used by an artificial intelligence system.

Réseau de neurones à propagation avant

Feedforward neural network - Artificial neural network in which signals can only propagate in one direction, from an input layer to an output layer, with no possibility of returning. Feedforward neural networks are opposed to recurrent neural networks, in which signals can go back and feed neurons from a previous layer or the same layer. Multilayer perceptrons are feedforward neural networks.

Réseau de neurones artificiels

Organized set of interconnected artificial neurons, created for the purpose of being able to perform complex operations or solve difficult problems thanks to a learning mechanism enabling it to acquire a form of intelligence. Originally, the creators of artificial neural networks were inspired by the functioning of the nervous system, which is organized according to the connections that are established between biological neurons. Other terms can also be used: neural network, connectionist network or neuromimetic network. Their frequency of use, however, remains low.

Réseau de neurones convolutif

Artificial neural network that makes it possible to detect the presence of simple patterns at different scales of an image and to progressively identify its content by association and cross-referencing. The functioning of convolutional neural networks is inspired by that of the visual cortex of vertebrates.

Réseau de neurones profond

Artificial neural network with many hidden layers that allows it, by multiplying the possibilities of processing, to increase its ability to learn, to improve its efficiency in performing certain complex operations and to increase its means of solving certain difficult problems. The depth of artificial neural networks is measured by the number of hidden layers they have.

Réseau de neurones récurrent

Artificial neural network in which signals can go back and feed neurons from a previous layer or the same layer. Recurrent neural networks are opposed to feedforward neural networks, in which signals can only go in one direction, from an input layer to an output layer.

Réseaux sémantiques

Graphs modeling the representation of knowledge.

Rétropropagation d'erreurs

Mechanism by which interpretation errors, calculated at the output of one or more layers of neurons in an artificial neural network, produce signals that are transmitted to the neurons that previously contributed to creating discrepancies, so that corrections are made by adjusting the synaptic coefficients or the biases responsible.

Robotique

Engineering field that designs, builds, operates, and applies robots.

Seuil d'activation

Numerical value that a signal at the output of an artificial neuron must reach or exceed in order for it to be activated.

Surapprentissage

Phenomenon responsible for errors, which occurs when a system such as an artificial neural network is forced to analyze new cases based almost exclusively on the specific characteristics of those it has already encountered in the learning phase, to the detriment of more general characteristics that would allow it to better understand what it has not already been presented.

Système d'intelligence artificielle

A system designed to simulate the functioning of human intelligence in order to perform functions normally belonging to it.

Système expert

A knowledge-based system designed to replace the expertise of specialists in a given field. Expert systems consist of a knowledge base containing a formalized representation of the knowledge of a domain. It is up to the cognitivist to collect these from experts and formalize them in a representation language. Expert systems are notably used in the fields of medicine, finance, insurance, and equipment repair.

Test de Turing

A test that consists of putting a human and a computer in communication, blindly, to verify whether they are capable of achieving the same levels of performance. If the human operator is unable to distinguish which of his interlocutors is the computer, it is considered that the machine has passed the Turing test and is thus endowed with artificial intelligence. The Turing test was imagined by the British mathematician Alan Turing.

Traitement du langage naturel (NLP)

A field of research aimed at developing computer interfaces capable of understanding and generating human language.

Vision artificielle

A field of research aimed at developing computer systems capable of analyzing and understanding images and videos.

Vision par ordinateur

A field in which techniques are studied and developed to enable a computer system or an artificial intelligence system to analyze and understand visual data obtained using cameras or other electronic devices.

bottom of page