LeAVJEPA proposes a shared encoder for sound and vision

A post introduces LeAVJEPA as a self-supervised approach to learning from sound and vision. It describes one shared encoder trained without class labels and says dropping a modality makes audio and video learn a common representation.