A graphe factoriel is a type of graphe bipartite used in modélisation statistique and inference, particularly within the fields of intelligence artificielle and apprentissage automatique. In a factor graph, the nodes represent variables and factors, where variables are the unknowns of the function being analyzed, and factors are the functions that relate these variables. This structure allows for a clear visual representation of how a complex function can be decomposed into simpler, manageable components.
Factor graphs are particularly useful in applications such as graphical models, which include Bayesian networks and Markov random fields. They facilitate efficient computation of marginal distributions and are often leveraged in algorithms like belief propagation, which is utilized for inference in modèles probabilistes. The ability to express a function as a product of smaller factors enables the efficient use of algorithms that can exploit the independence properties of the involved variables.
In essence, factor graphs provide a framework that simplifies the complexities associated with multi-dimensional probability distributions by breaking them down into smaller, more tractable parts. This decomposition is crucial for various AI applications, including machine learning, computer vision, and traitement du langage naturel, where understanding the relationships among variables is key to improving model performance.