About contrastive unsupervised representation learning for classification and its convergence
This work provides theoretical underpinnings for contrastive learning, which is important for researchers and practitioners developing and applying self-supervised methods, though the immediate practical impact is incremental.
This paper extends theoretical guarantees for contrastive learning to include training with multiple negative samples and multiway classification. It also provides convergence guarantees for minimizing the contrastive training error using gradient descent with an overparametrized deep neural encoder.
Contrastive representation learning has been recently proved to be very efficient for self-supervised training. These methods have been successfully used to train encoders which perform comparably to supervised training on downstream classification tasks. A few works have started to build a theoretical framework around contrastive learning in which guarantees for its performance can be proven. We provide extensions of these results to training with multiple negative samples and for multiway classification. Furthermore, we provide convergence guarantees for the minimization of the contrastive training error with gradient descent of an overparametrized deep neural encoder, and provide some numerical experiments that complement our theoretical findings