Joint Embedding Architectures

Definition

A class of neural network architectures where two or more views of the same input (original, corrupted, transformed, or from different modalities) are independently passed through encoder networks that output embedding vectors. The architecture trains these encoders to output similar embeddings for semantically related inputs and dissimilar embeddings for different inputs, without requiring the network to reconstruct or generate pixel-level data.

Cast within

No topic-region cast yet — this appears once Joint Embedding Architectures's compiled claims are aligned into a topic region's argument tree.

Claims mentioning Joint Embedding Architectures (4)

Your notes

Loading notes...