Normalisation d'Instance
La Normalisation d'Instance est une technique utilisé en apprentissage automatique, particularly in the field of apprentissage profond for vision par ordinateur tasks. It is designed to normalize the features of individual instances (or images) in a batch, rather than normalizing across the entire batch like Batch Normalization.
In Instance Normalization, each feature map is adjusted independently for each input sample. This means that for a given feature map, the mean and variance are calculated only for that specific instance, allowing the model to maintain style information while enhancing the content representation. This is especially useful in tasks like transfert de style, where the goal is to apply the artistic style of one image to the content of another.
Mathématiquement, la normalisation d'instance peut être exprimée comme :
y = (x - E[x]) / sqrt(Var[x] + ε)where y is the normalized output, x is the input feature map, E[x] is the mean of the feature map, Var[x] is the variance, and ε is a small constant added for stabilité numérique.
Instance Normalization has gained popularity in various applications, particularly those involving generative models, such as Generative Adversarial Networks (GANs) and transfert de style neuronal. By normalizing features per instance, it allows models to better capture variations in styles, leading to more aesthetically pleasing outputs.
In summary, Instance Normalization is a powerful technique that enhances the performance of deep learning models on tasks that require a nuanced understanding of style and content, making it a valuable tool in the modern boîte à outils AI.