La Croisement Entropie Objectif is a widely used fonction de perte in apprentissage automatique, particularly in the context of classification tasks. It quantifies the difference between two distributions de probabilité: the true distribution of labels and the predicted distribution output by a model. The objective is to minimize this difference, which represents how well the model’s predictions align with the actual labels.
Mathématiquement, l'entropie croisée est définie comme :
H(p, q) = -Σ p(x) log(q(x))
où :
- H(p, q) is the cross entropy between the true distribution p and the predicted distribution q.
- p(x) is the true probability of class labels (usually represented as one-hot encoded vectors).
- q(x) est la probabilité prédite des étiquettes de classe fournie par le modèle.
In practical terms, when using cross entropy as the objective function, the model is penalized more heavily for confident but incorrect predictions. This characteristic makes it especially effective for tasks where accurate probability estimation is critical, such as in classification multi-classes problèmes.
Cross entropy is commonly employed in various machine learning frameworks and is particularly effective with neural networks when combined with softmax fonctions d'activation in the output layer. The optimization process adjusts the model parameters to minimize the cross entropy loss, thereby improving the model’s accuracy and reliability in predicting outcomes.