NeurIPS 2020

Contrastive Learning with Adversarial Examples

Meta Review

It is an intuitive idea to combine adversarial learning with noise-contrastive learning. It would be an interesting contribution if this was not done before, and the reviewers mostly agree on this issue. While there have been concerns on the experimental setup being not mostly significant (regarding dataset size and model size), the reviewers tend to think the novelty of the proposed method outweighs the shortcomings.