KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Anchor Point
Reference point in the embedding space used as a comparison point to evaluate similarity with other examples in triplet loss.
Positive Pair
Pair of samples sharing the same class label, used in training siamese networks to learn similarity.
Negative Pair
Pair of samples belonging to different classes, essential for learning to discriminate between dissimilar representations.
Twin Network Architecture
Symmetrical network structure where two identical sub-networks share the same weights to ensure equivalent transformations.
Margin-based Loss
Loss function incorporating a separation margin to ensure a minimum distance between different classes in the latent space.
Verification Task
Binary classification task determining if two inputs belong to the same class, typical in siamese network applications.
Relation Network
Variant of siamese networks that explicitly learns a metric function to compare representations of input pairs.