LeVJEPA: Efficient & Scalable Video Pretraining without the Heuristics 29 August 2026·12 mins Machine Learning Paper Notes Machine-Learning Arxiv Self-Supervised-Learning Video Representation-Learning Jepa