or: Towards Domain-Agnostic Contrastive Learning

Input A is partially noised using Mixup with Input B and fed into a deep network. Contrastive loss is used to discriminate the closeness of Inputs A and B relative to an unrelated Input C.