Le changement de distribution est un phénomène en apprentissage automatique and intelligence artificielle where the statistical properties of the input data change between the training phase and the phase d'inférence. This can occur due to various factors, such as changes in the environment, user behavior, or other external influences that alter the distribution of data.
For example, a model trained on historical sales data may perform well when making predictions in a stable economic environment. However, if a sudden economic downturn occurs, the new data may not reflect the same patterns as the training data, leading to a decline in performance du modèle. This shift can happen in various forms, including décalage de covariables, where the input features change, and décalage d'étiquettes, where the distribution of output labels changes.
Le changement de distribution pose des défis importants pour maintenir la robustesse et fiabilité of AI systems. To mitigate its effects, practitioners often employ techniques such as adaptation de domaine, where the model is retrained on new data, or généralisation de domaine, where the model is designed to perform well across various data distributions without needing retraining.
Comprendre et traiter le changement de distribution est crucial pour garantir que modèles d'IA remain effective and accurate when deployed in real-world scenarios, where data conditions can frequently change.