A échantillon négatif refers to a data point in apprentissage automatique that represents an instance of the class that the model is not trying to predict. In contrast to positive samples, which are instances of the target class, negative samples help in training models to distinguish between different classes effectively.
In classification tasks, especially classification binaire, the model is trained to recognize the positive class (also known as the target or positive sample) and differentiate it from negative samples. For example, if a model is being trained to identify cats in images, images that do not contain cats would be considered negative samples.
Utilizing negative samples is essential for creating a robust and accurate model, as they help to minimize false positives—situations where the model incorrectly identifies an instance as belonging to the positive class. Including a diverse set of negative samples during training can enhance the model’s ability to generalize and perform well on unseen data.
Dans certains contextes, échantillonnage négatif techniques might be employed, where random negative samples are selected from a larger dataset to improve training efficiency. This approach is particularly useful in scenarios with jeux de données déséquilibrés, where the number of positive samples is significantly lower than the number of negative samples.
Dans l'ensemble, les échantillons négatifs jouent un rôle crucial dans l'orientation du processus d'apprentissage des modèles d'apprentissage automatique, en veillant à ce qu'ils puissent efficacement différencier entre des instances pertinentes et non pertinentes.