Representation Learning

What to align in multimodal contrastive learning?