Văn bản này: Solving inefficiency of self-supervised representation learning