Test-Time Training

A technique that updates model weights during inference via reinforcement learning based on self-scored outputs, aiming to impro