Embedding

and provide high-quality "sensors" to 'strong intelligence systems'.

and the backbone of the AI progress.

The goal of contrastive representation learning is to learn such an embedding space in which similar sample pairs stay close to each other while dissimilar ones are far apart.

a lower-dimensional space that captures the essential features of the input data ELI5: you can think of latent space as a map where similar pieces of data lives together Navigating this is pretty hard, given that dimensionality of the feature space are usually a lot smaller comparing to the origina...