A Theory for Representation Learning via Contrastive Objectives
32-155Abstract: Representation learning seeks to represent complicated data (images, text etc.) using a vector embedding, which can then be used to solve complicated new classification tasks using simple methods like…