Glossario IA

📖

termini

Prototypical Network

Alternative to Relation Networks using prototypes (centroids) to represent each class rather than direct pairwise comparisons. Computes distances between queries and prototypes for classification.

📖

termini

Comparison Function

Neural module learning to compare two embeddings by producing a relation score, core of Relation Networks. Typically implemented via convolution or attention layers on the concatenation of embeddings.

📖

termini

Feature Comparison Module

Specific component of Relation Networks responsible for computing relations between pairs of feature embeddings. Typically combines concatenation, convolution and activation to produce relation scores.