Seol mar théacs é seo: Solving inefficiency of self-supervised representation learning