Meta presents Learning and Leveraging World Models in Visual Representation Learning.

Joint-Embedding Predictive Architecture (JEPA) has emerged as a promising self-supervised approach that learns by leveraging a world model.