Self-Supervised Transfer

Projection Head

Additional neural network applied after the main encoder to map representations to the space where the contrastive loss is calculated. This head is typically removed when transferring to downstream tasks.

← Back