World Model Self-Distillation

New framework trains video models for tasks using self-distillation

Researchers have developed a new framework for training video diffusion models to solve general tasks by combining self-distillation and reinforcement learning. This method allows the models to learn task-solving abilit…